From 82ae857877fdf3492f40bca87657a07892c3f59b Mon Sep 17 00:00:00 2001 From: Haibo Xu Date: Fri, 6 Dec 2019 06:29:24 +0000 Subject: [PATCH 01/29] Enable build of test/syscall tests on arm64. Signed-off-by: Haibo Xu Change-Id: I277d6c708bbf5c3edd7c3568941cfd01dc122e17 --- test/syscalls/linux/BUILD | 55 +++++++++++++++++++++++++++------ test/syscalls/linux/bad.cc | 3 +- test/syscalls/linux/chroot.cc | 2 +- test/syscalls/linux/fork.cc | 3 ++ test/syscalls/linux/getdents.cc | 10 +++++- test/syscalls/linux/preadv2.cc | 2 ++ test/syscalls/linux/proc.cc | 2 +- test/syscalls/linux/pwritev2.cc | 2 ++ test/syscalls/linux/seccomp.cc | 5 +++ test/syscalls/linux/stat.cc | 2 ++ test/util/signal_util.h | 14 +++++++++ test/util/test_util.cc | 2 +- 12 files changed, 88 insertions(+), 14 deletions(-) diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 064ce8429..68dcc598b 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -19,6 +19,16 @@ exports_files( visibility = ["//:sandbox"], ) +config_setting( + name = "x86_64", + constraint_values = ["@bazel_tools//platforms:x86_64"], +) + +config_setting( + name = "aarch64", + constraint_values = ["@bazel_tools//platforms:aarch64"], +) + cc_binary( name = "sigaltstack_check", testonly = 1, @@ -197,7 +207,10 @@ cc_binary( cc_binary( name = "32bit_test", testonly = 1, - srcs = ["32bit.cc"], + srcs = select({ + ":x86_64": ["32bit.cc"], + ":aarch64": [], + }), linkstatic = 1, deps = [ "//test/util:memory_util", @@ -584,7 +597,10 @@ cc_binary( cc_binary( name = "exceptions_test", testonly = 1, - srcs = ["exceptions.cc"], + srcs = select({ + ":x86_64": ["exceptions.cc"], + ":aarch64": [], + }), linkstatic = 1, deps = [ "//test/util:logging", @@ -640,7 +656,10 @@ cc_binary( cc_binary( name = "exec_binary_test", testonly = 1, - srcs = ["exec_binary.cc"], + srcs = select({ + ":x86_64": ["exec_binary.cc"], + ":aarch64": [], + }), linkstatic = 1, deps = [ "//test/util:cleanup", @@ -811,7 +830,10 @@ cc_binary( cc_binary( name = "fpsig_fork_test", testonly = 1, - srcs = ["fpsig_fork.cc"], + srcs = select({ + ":x86_64": ["fpsig_fork.cc"], + ":aarch64": [], + }), linkstatic = 1, deps = [ "//test/util:logging", @@ -825,7 +847,10 @@ cc_binary( cc_binary( name = "fpsig_nested_test", testonly = 1, - srcs = ["fpsig_nested.cc"], + srcs = select({ + ":x86_64": ["fpsig_nested.cc"], + ":aarch64": [], + }), linkstatic = 1, deps = [ "//test/util:test_main", @@ -1440,7 +1465,10 @@ cc_binary( cc_binary( name = "arch_prctl_test", testonly = 1, - srcs = ["arch_prctl.cc"], + srcs = select({ + ":x86_64": ["arch_prctl.cc"], + ":aarch64": [], + }), linkstatic = 1, deps = [ "//test/util:test_main", @@ -2035,7 +2063,10 @@ cc_binary( cc_binary( name = "sigiret_test", testonly = 1, - srcs = ["sigiret.cc"], + srcs = select({ + ":x86_64": ["sigiret.cc"], + ":aarch64": [], + }), linkstatic = 1, deps = [ "//test/util:logging", @@ -2043,7 +2074,10 @@ cc_binary( "//test/util:test_util", "//test/util:timer_util", "@com_google_googletest//:gtest", - ], + ] + select({ + ":x86_64": [], + ":aarch64": ["//test/util:test_main"], + }), ) cc_binary( @@ -3260,7 +3294,10 @@ cc_binary( cc_binary( name = "sysret_test", testonly = 1, - srcs = ["sysret.cc"], + srcs = select({ + ":x86_64": ["sysret.cc"], + ":aarch64": [], + }), linkstatic = 1, deps = [ "//test/util:logging", diff --git a/test/syscalls/linux/bad.cc b/test/syscalls/linux/bad.cc index f246a799e..9e4d8ea57 100644 --- a/test/syscalls/linux/bad.cc +++ b/test/syscalls/linux/bad.cc @@ -22,12 +22,13 @@ namespace gvisor { namespace testing { namespace { - +#if defined(__x86_64__) TEST(BadSyscallTest, NotImplemented) { // get_kernel_syms is not supported in Linux > 2.6, and not implemented in // gVisor. EXPECT_THAT(syscall(SYS_get_kernel_syms), SyscallFailsWithErrno(ENOSYS)); } +#endif // defined(__x86_64__) TEST(BadSyscallTest, NegativeOne) { EXPECT_THAT(syscall(-1), SyscallFailsWithErrno(ENOSYS)); diff --git a/test/syscalls/linux/chroot.cc b/test/syscalls/linux/chroot.cc index 04bc2d7b9..0a2d44a2c 100644 --- a/test/syscalls/linux/chroot.cc +++ b/test/syscalls/linux/chroot.cc @@ -162,7 +162,7 @@ TEST(ChrootTest, DotDotFromOpenFD) { // getdents on fd should not error. char buf[1024]; - ASSERT_THAT(syscall(SYS_getdents, fd.get(), buf, sizeof(buf)), + ASSERT_THAT(syscall(SYS_getdents64, fd.get(), buf, sizeof(buf)), SyscallSucceeds()); } diff --git a/test/syscalls/linux/fork.cc b/test/syscalls/linux/fork.cc index 371890110..906f3358d 100644 --- a/test/syscalls/linux/fork.cc +++ b/test/syscalls/linux/fork.cc @@ -215,6 +215,8 @@ TEST_F(ForkTest, PrivateMapping) { EXPECT_THAT(Wait(child), SyscallSucceedsWithValue(0)); } +// CPUID is x86 specific. +#ifdef __x86_64__ // Test that cpuid works after a fork. TEST_F(ForkTest, Cpuid) { pid_t child = Fork(); @@ -227,6 +229,7 @@ TEST_F(ForkTest, Cpuid) { } EXPECT_THAT(Wait(child), SyscallSucceedsWithValue(0)); } +#endif TEST_F(ForkTest, Mmap) { pid_t child = Fork(); diff --git a/test/syscalls/linux/getdents.cc b/test/syscalls/linux/getdents.cc index ad2dbacb8..bfd18d4ff 100644 --- a/test/syscalls/linux/getdents.cc +++ b/test/syscalls/linux/getdents.cc @@ -228,19 +228,27 @@ class GetdentsTest : public ::testing::Test { // Multiple template parameters are not allowed, so we must use explicit // template specialization to set the syscall number. +#ifdef __x86_64__ template <> int GetdentsTest::SyscallNum() { return SYS_getdents; } +#endif template <> int GetdentsTest::SyscallNum() { return SYS_getdents64; } -// Test both legacy getdents and getdents64. +#ifdef __x86_64__ +// Test both legacy getdents and getdents64 on x86_64. typedef ::testing::Types GetdentsTypes; +#elif __aarch64__ +// Test only getdents64 on arm64. +typedef ::testing::Types + GetdentsTypes; +#endif TYPED_TEST_SUITE(GetdentsTest, GetdentsTypes); // N.B. TYPED_TESTs require explicitly using this-> to access members of diff --git a/test/syscalls/linux/preadv2.cc b/test/syscalls/linux/preadv2.cc index c9246367d..3eeaf6ad8 100644 --- a/test/syscalls/linux/preadv2.cc +++ b/test/syscalls/linux/preadv2.cc @@ -35,6 +35,8 @@ namespace { #ifndef SYS_preadv2 #if defined(__x86_64__) #define SYS_preadv2 327 +#elif defined(__aarch64__) +#define SYS_preadv2 286 #else #error "Unknown architecture" #endif diff --git a/test/syscalls/linux/proc.cc b/test/syscalls/linux/proc.cc index 8cf08991b..5b4f29cd9 100644 --- a/test/syscalls/linux/proc.cc +++ b/test/syscalls/linux/proc.cc @@ -1986,7 +1986,7 @@ TEST(Proc, GetdentsEnoent) { }, nullptr, nullptr)); char buf[1024]; - ASSERT_THAT(syscall(SYS_getdents, fd.get(), buf, sizeof(buf)), + ASSERT_THAT(syscall(SYS_getdents64, fd.get(), buf, sizeof(buf)), SyscallFailsWithErrno(ENOENT)); } diff --git a/test/syscalls/linux/pwritev2.cc b/test/syscalls/linux/pwritev2.cc index 1dbc0d6df..3fe5a600f 100644 --- a/test/syscalls/linux/pwritev2.cc +++ b/test/syscalls/linux/pwritev2.cc @@ -34,6 +34,8 @@ namespace { #ifndef SYS_pwritev2 #if defined(__x86_64__) #define SYS_pwritev2 328 +#elif defined(__aarch64__) +#define SYS_pwritev2 287 #else #error "Unknown architecture" #endif diff --git a/test/syscalls/linux/seccomp.cc b/test/syscalls/linux/seccomp.cc index 7e41fe7d8..6d7e543b9 100644 --- a/test/syscalls/linux/seccomp.cc +++ b/test/syscalls/linux/seccomp.cc @@ -49,7 +49,12 @@ namespace testing { namespace { // A syscall not implemented by Linux that we don't expect to be called. +#ifdef __x86_64__ constexpr uint32_t kFilteredSyscall = SYS_vserver; +#elif __aarch64__ +// Using arch_specific_syscalls which are not implemented on arm64. +constexpr uint32_t kFilteredSyscall = SYS_arch_specific_syscall+15; +#endif // Applies a seccomp-bpf filter that returns `filtered_result` for // `sysno` and allows all other syscalls. Async-signal-safe. diff --git a/test/syscalls/linux/stat.cc b/test/syscalls/linux/stat.cc index 30de2f8ff..7a99f2636 100644 --- a/test/syscalls/linux/stat.cc +++ b/test/syscalls/linux/stat.cc @@ -557,6 +557,8 @@ TEST(SimpleStatTest, AnonDeviceAllocatesUniqueInodesAcrossSaveRestore) { #ifndef SYS_statx #if defined(__x86_64__) #define SYS_statx 332 +#elif defined(__aarch64__) +#define SYS_statx 291 #else #error "Unknown architecture" #endif diff --git a/test/util/signal_util.h b/test/util/signal_util.h index bcf85c337..e7b66aa51 100644 --- a/test/util/signal_util.h +++ b/test/util/signal_util.h @@ -85,6 +85,20 @@ inline void FixupFault(ucontext_t* ctx) { // The encoding is 0x48 0xab 0x00. ctx->uc_mcontext.gregs[REG_RIP] += 3; } +#elif __aarch64__ +inline void Fault() { + // Zero and dereference x0. + asm("mov xzr, x0\r\n" + "str xzr, [x0]\r\n" + : + : + : "x0"); +} + +inline void FixupFault(ucontext_t* ctx) { + // Skip the bad instruction above. + ctx->uc_mcontext.pc += 4; +} #endif } // namespace testing diff --git a/test/util/test_util.cc b/test/util/test_util.cc index 848504c88..a4f78eec2 100644 --- a/test/util/test_util.cc +++ b/test/util/test_util.cc @@ -76,7 +76,6 @@ bool IsRunningWithHostinet() { "xchg %%rdi, %%rbx\n" \ : "=a"(a), "=D"(b), "=c"(c), "=d"(d) \ : "a"(a_inp), "2"(c_inp)) -#endif // defined(__x86_64__) CPUVendor GetCPUVendor() { uint32_t eax, ebx, ecx, edx; @@ -93,6 +92,7 @@ CPUVendor GetCPUVendor() { } return CPUVendor::kUnknownVendor; } +#endif // defined(__x86_64__) bool operator==(const KernelVersion& first, const KernelVersion& second) { return first.major == second.major && first.minor == second.minor && From 14959250feb71df74dea13f3cb15dcbe8ce6b3f3 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Thu, 30 Jan 2020 17:37:17 -0800 Subject: [PATCH 02/29] Simplify testing link rules. PiperOrigin-RevId: 292458933 --- test/syscalls/linux/BUILD | 688 +++++++++++++++++++------------------- test/util/BUILD | 30 +- tools/build/defs.bzl | 1 + tools/defs.bzl | 3 +- tools/images/BUILD | 4 +- 5 files changed, 363 insertions(+), 363 deletions(-) diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index ee7a8a673..e4ca5b6db 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -1,4 +1,4 @@ -load("//tools:defs.bzl", "cc_binary", "cc_library", "default_net_util", "select_arch", "select_system") +load("//tools:defs.bzl", "cc_binary", "cc_library", "default_net_util", "gtest", "select_arch", "select_system") package( default_visibility = ["//:sandbox"], @@ -82,14 +82,14 @@ cc_library( srcs = ["base_poll_test.cc"], hdrs = ["base_poll_test.h"], deps = [ + "@com_google_absl//absl/memory", + "@com_google_absl//absl/synchronization", + "@com_google_absl//absl/time", + gtest, "//test/util:logging", "//test/util:signal_util", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/memory", - "@com_google_absl//absl/synchronization", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -99,11 +99,11 @@ cc_library( hdrs = ["file_base.h"], deps = [ "//test/util:file_descriptor", + "@com_google_absl//absl/strings", + gtest, "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_util", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) @@ -130,7 +130,7 @@ cc_library( hdrs = ["socket_test_util.h"], defines = select_system(), deps = default_net_util() + [ - "@com_google_googletest//:gtest", + gtest, "@com_google_absl//absl/memory", "@com_google_absl//absl/strings", "@com_google_absl//absl/strings:str_format", @@ -155,9 +155,9 @@ cc_library( hdrs = ["unix_domain_socket_test_util.h"], deps = [ ":socket_test_util", - "//test/util:test_util", "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", + gtest, + "//test/util:test_util", ], ) @@ -179,14 +179,14 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:cleanup", + "@com_google_absl//absl/time", + gtest, "//test/util:posix_error", "//test/util:signal_util", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", "//test/util:timer_util", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -199,13 +199,13 @@ cc_binary( ), linkstatic = 1, deps = [ + "@com_google_absl//absl/base:core_headers", + gtest, "//test/util:memory_util", "//test/util:platform_util", "//test/util:posix_error", "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/base:core_headers", - "@com_google_googletest//:gtest", ], ) @@ -218,9 +218,9 @@ cc_binary( ":socket_test_util", ":unix_domain_socket_test_util", "//test/util:file_descriptor", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -233,9 +233,9 @@ cc_binary( ":socket_test_util", ":unix_domain_socket_test_util", "//test/util:file_descriptor", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -247,10 +247,10 @@ cc_binary( deps = [ "//test/util:capability_util", "//test/util:fs_util", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -262,12 +262,12 @@ cc_binary( deps = [ "//test/util:cleanup", "//test/util:fs_util", + "@com_google_absl//absl/strings", + gtest, "//test/util:posix_error", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) @@ -280,12 +280,11 @@ cc_binary( ], linkstatic = 1, deps = [ - # The heapchecker doesn't recognize that io_destroy munmaps. - "@com_google_googletest//:gtest", - "@com_google_absl//absl/strings", "//test/util:cleanup", "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/strings", + gtest, "//test/util:memory_util", "//test/util:posix_error", "//test/util:proc_util", @@ -302,12 +301,12 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + "@com_google_absl//absl/time", + gtest, "//test/util:logging", "//test/util:signal_util", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -320,9 +319,9 @@ cc_binary( "//:sandbox", ], deps = [ + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -334,9 +333,9 @@ cc_binary( deps = [ ":socket_test_util", ":unix_domain_socket_test_util", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -348,9 +347,9 @@ cc_binary( deps = [ ":socket_test_util", "//test/util:file_descriptor", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -372,10 +371,10 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:capability_util", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -388,10 +387,10 @@ cc_binary( "//test/util:capability_util", "//test/util:file_descriptor", "//test/util:fs_util", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -404,14 +403,14 @@ cc_binary( "//test/util:capability_util", "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/flags:flag", + "@com_google_absl//absl/synchronization", + gtest, "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/flags:flag", - "@com_google_absl//absl/synchronization", - "@com_google_googletest//:gtest", ], ) @@ -424,12 +423,12 @@ cc_binary( "//test/util:capability_util", "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/flags:flag", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/flags:flag", - "@com_google_googletest//:gtest", ], ) @@ -443,12 +442,12 @@ cc_binary( "//test/util:cleanup", "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/strings", + gtest, "//test/util:mount_util", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) @@ -458,9 +457,9 @@ cc_binary( srcs = ["clock_getres.cc"], linkstatic = 1, deps = [ + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -470,11 +469,11 @@ cc_binary( srcs = ["clock_gettime.cc"], linkstatic = 1, deps = [ + "@com_google_absl//absl/time", + gtest, "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -484,13 +483,13 @@ cc_binary( srcs = ["concurrency.cc"], linkstatic = 1, deps = [ + "@com_google_absl//absl/strings", + "@com_google_absl//absl/time", + gtest, "//test/util:platform_util", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/strings", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -503,9 +502,9 @@ cc_binary( ":socket_test_util", "//test/util:file_descriptor", "//test/util:fs_util", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -516,10 +515,10 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:fs_util", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -530,9 +529,9 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -544,11 +543,11 @@ cc_binary( deps = [ "//test/util:eventfd_util", "//test/util:file_descriptor", + gtest, "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -561,10 +560,10 @@ cc_binary( "//test/util:epoll_util", "//test/util:eventfd_util", "//test/util:file_descriptor", + gtest, "//test/util:posix_error", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -576,10 +575,10 @@ cc_binary( deps = [ "//test/util:epoll_util", "//test/util:eventfd_util", + gtest, "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_googletest//:gtest", ], ) @@ -589,12 +588,12 @@ cc_binary( srcs = ["exceptions.cc"], linkstatic = 1, deps = [ + gtest, "//test/util:logging", "//test/util:platform_util", "//test/util:signal_util", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -604,10 +603,10 @@ cc_binary( srcs = ["getcpu.cc"], linkstatic = 1, deps = [ + "@com_google_absl//absl/time", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -617,10 +616,10 @@ cc_binary( srcs = ["getcpu.cc"], linkstatic = 1, deps = [ + "@com_google_absl//absl/time", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -630,13 +629,13 @@ cc_binary( srcs = ["getrusage.cc"], linkstatic = 1, deps = [ + "@com_google_absl//absl/time", + gtest, "//test/util:logging", "//test/util:memory_util", "//test/util:signal_util", "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -652,14 +651,14 @@ cc_binary( "//test/util:cleanup", "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/strings", + gtest, "//test/util:multiprocess_util", "//test/util:posix_error", "//test/util:proc_util", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) @@ -682,15 +681,15 @@ cc_binary( deps = [ "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/strings", + "@com_google_absl//absl/synchronization", + "@com_google_absl//absl/types:optional", + gtest, "//test/util:multiprocess_util", "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/strings", - "@com_google_absl//absl/synchronization", - "@com_google_absl//absl/types:optional", - "@com_google_googletest//:gtest", ], ) @@ -701,11 +700,11 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + "@com_google_absl//absl/time", + gtest, "//test/util:test_main", "//test/util:test_util", "//test/util:time_util", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -718,10 +717,10 @@ cc_binary( ":file_base", "//test/util:cleanup", "//test/util:file_descriptor", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -731,9 +730,9 @@ cc_binary( srcs = ["fault.cc"], linkstatic = 1, deps = [ + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -744,10 +743,10 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:capability_util", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -761,18 +760,18 @@ cc_binary( "//test/util:cleanup", "//test/util:eventfd_util", "//test/util:fs_util", + "@com_google_absl//absl/base:core_headers", + "@com_google_absl//absl/flags:flag", + "@com_google_absl//absl/memory", + "@com_google_absl//absl/strings", + "@com_google_absl//absl/time", + gtest, "//test/util:multiprocess_util", "//test/util:posix_error", "//test/util:save_util", "//test/util:temp_path", "//test/util:test_util", "//test/util:timer_util", - "@com_google_absl//absl/base:core_headers", - "@com_google_absl//absl/flags:flag", - "@com_google_absl//absl/memory", - "@com_google_absl//absl/strings", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -786,15 +785,15 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + "@com_google_absl//absl/strings", + "@com_google_absl//absl/time", + gtest, "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", "//test/util:timer_util", - "@com_google_absl//absl/strings", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -805,13 +804,13 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:capability_util", + "@com_google_absl//absl/time", + gtest, "//test/util:logging", "//test/util:memory_util", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -824,11 +823,11 @@ cc_binary( ), linkstatic = 1, deps = [ + gtest, "//test/util:logging", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_googletest//:gtest", ], ) @@ -841,10 +840,10 @@ cc_binary( ), linkstatic = 1, deps = [ + gtest, "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_googletest//:gtest", ], ) @@ -855,10 +854,10 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -869,10 +868,10 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -884,6 +883,9 @@ cc_binary( deps = [ "//test/util:cleanup", "//test/util:file_descriptor", + "@com_google_absl//absl/memory", + "@com_google_absl//absl/time", + gtest, "//test/util:memory_util", "//test/util:save_util", "//test/util:temp_path", @@ -892,9 +894,6 @@ cc_binary( "//test/util:thread_util", "//test/util:time_util", "//test/util:timer_util", - "@com_google_absl//absl/memory", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -907,12 +906,12 @@ cc_binary( "//test/util:eventfd_util", "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/strings", + gtest, "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) @@ -922,9 +921,9 @@ cc_binary( srcs = ["getrandom.cc"], linkstatic = 1, deps = [ + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -957,10 +956,10 @@ cc_binary( ":socket_test_util", ":unix_domain_socket_test_util", "//test/util:file_descriptor", + gtest, "//test/util:signal_util", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -984,9 +983,9 @@ cc_binary( ":socket_test_util", "//test/util:capability_util", "//test/util:file_descriptor", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -997,6 +996,9 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + "@com_google_absl//absl/strings", + "@com_google_absl//absl/time", + gtest, "//test/util:logging", "//test/util:multiprocess_util", "//test/util:posix_error", @@ -1004,9 +1006,6 @@ cc_binary( "//test/util:test_util", "//test/util:thread_util", "//test/util:timer_util", - "@com_google_absl//absl/strings", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -1018,15 +1017,15 @@ cc_binary( deps = [ "//test/util:capability_util", "//test/util:file_descriptor", + "@com_google_absl//absl/flags:flag", + "@com_google_absl//absl/synchronization", + "@com_google_absl//absl/time", + gtest, "//test/util:logging", "//test/util:signal_util", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/flags:flag", - "@com_google_absl//absl/synchronization", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -1039,14 +1038,14 @@ cc_binary( "//test/util:capability_util", "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/flags:flag", + "@com_google_absl//absl/strings", + gtest, "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/flags:flag", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) @@ -1057,10 +1056,10 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -1071,6 +1070,7 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + gtest, "//test/util:logging", "//test/util:memory_util", "//test/util:multiprocess_util", @@ -1078,7 +1078,6 @@ cc_binary( "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -1089,12 +1088,12 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:cleanup", + "@com_google_absl//absl/memory", + gtest, "//test/util:memory_util", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/memory", - "@com_google_googletest//:gtest", ], ) @@ -1104,11 +1103,11 @@ cc_binary( srcs = ["mincore.cc"], linkstatic = 1, deps = [ + gtest, "//test/util:memory_util", "//test/util:posix_error", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -1121,10 +1120,10 @@ cc_binary( ":temp_umask", "//test/util:capability_util", "//test/util:fs_util", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -1135,11 +1134,11 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_googletest//:gtest", ], ) @@ -1151,12 +1150,12 @@ cc_binary( deps = [ "//test/util:capability_util", "//test/util:cleanup", + gtest, "//test/util:memory_util", "//test/util:multiprocess_util", "//test/util:rlimit_util", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -1169,13 +1168,13 @@ cc_binary( "//test/util:cleanup", "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/strings", + gtest, "//test/util:memory_util", "//test/util:multiprocess_util", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) @@ -1188,6 +1187,9 @@ cc_binary( "//test/util:capability_util", "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/strings", + "@com_google_absl//absl/time", + gtest, "//test/util:mount_util", "//test/util:multiprocess_util", "//test/util:posix_error", @@ -1195,9 +1197,6 @@ cc_binary( "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/strings", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -1207,10 +1206,9 @@ cc_binary( srcs = ["mremap.cc"], linkstatic = 1, deps = [ - # The heap check fails due to MremapDeathTest - "@com_google_googletest//:gtest", - "@com_google_absl//absl/strings", "//test/util:file_descriptor", + "@com_google_absl//absl/strings", + gtest, "//test/util:logging", "//test/util:memory_util", "//test/util:multiprocess_util", @@ -1242,9 +1240,9 @@ cc_binary( srcs = ["munmap.cc"], linkstatic = 1, deps = [ + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -1261,14 +1259,14 @@ cc_binary( "//test/util:cleanup", "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/memory", + "@com_google_absl//absl/strings", + gtest, "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/memory", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) @@ -1282,10 +1280,10 @@ cc_binary( "//test/util:capability_util", "//test/util:file_descriptor", "//test/util:fs_util", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -1299,11 +1297,11 @@ cc_binary( ":unix_domain_socket_test_util", "//test/util:capability_util", "//test/util:file_descriptor", - "//test/util:test_main", - "//test/util:test_util", "@com_google_absl//absl/base:core_headers", "@com_google_absl//absl/base:endian", - "@com_google_googletest//:gtest", + gtest, + "//test/util:test_main", + "//test/util:test_util", ], ) @@ -1317,11 +1315,11 @@ cc_binary( ":unix_domain_socket_test_util", "//test/util:capability_util", "//test/util:file_descriptor", - "//test/util:test_main", - "//test/util:test_util", "@com_google_absl//absl/base:core_headers", "@com_google_absl//absl/base:endian", - "@com_google_googletest//:gtest", + gtest, + "//test/util:test_main", + "//test/util:test_util", ], ) @@ -1333,16 +1331,16 @@ cc_binary( deps = [ "//test/util:capability_util", "//test/util:file_descriptor", + "@com_google_absl//absl/base:core_headers", + "@com_google_absl//absl/strings", + "@com_google_absl//absl/synchronization", + "@com_google_absl//absl/time", + gtest, "//test/util:posix_error", "//test/util:pty_util", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/base:core_headers", - "@com_google_absl//absl/strings", - "@com_google_absl//absl/synchronization", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -1354,12 +1352,12 @@ cc_binary( deps = [ "//test/util:capability_util", "//test/util:file_descriptor", + "@com_google_absl//absl/base:core_headers", + gtest, "//test/util:posix_error", "//test/util:pty_util", "//test/util:test_main", "//test/util:thread_util", - "@com_google_absl//absl/base:core_headers", - "@com_google_googletest//:gtest", ], ) @@ -1372,12 +1370,12 @@ cc_binary( "//test/syscalls/linux:socket_test_util", "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/time", + gtest, "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -1387,13 +1385,13 @@ cc_binary( srcs = ["pause.cc"], linkstatic = 1, deps = [ + "@com_google_absl//absl/synchronization", + "@com_google_absl//absl/time", + gtest, "//test/util:signal_util", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/synchronization", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -1405,15 +1403,15 @@ cc_binary( deps = [ "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/strings", + "@com_google_absl//absl/synchronization", + "@com_google_absl//absl/time", + gtest, "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/strings", - "@com_google_absl//absl/synchronization", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -1426,13 +1424,13 @@ cc_binary( ":base_poll_test", "//test/util:eventfd_util", "//test/util:file_descriptor", + "@com_google_absl//absl/synchronization", + "@com_google_absl//absl/time", + gtest, "//test/util:logging", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/synchronization", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -1443,11 +1441,11 @@ cc_binary( linkstatic = 1, deps = [ ":base_poll_test", + "@com_google_absl//absl/time", + gtest, "//test/util:signal_util", "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -1458,9 +1456,9 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -1472,12 +1470,12 @@ cc_binary( deps = [ "//test/util:capability_util", "//test/util:cleanup", + "@com_google_absl//absl/flags:flag", + gtest, "//test/util:multiprocess_util", "//test/util:posix_error", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/flags:flag", - "@com_google_googletest//:gtest", ], ) @@ -1488,13 +1486,13 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:capability_util", + "@com_google_absl//absl/flags:flag", + gtest, "//test/util:logging", "//test/util:multiprocess_util", "//test/util:posix_error", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/flags:flag", - "@com_google_googletest//:gtest", ], ) @@ -1505,10 +1503,10 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -1519,6 +1517,8 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + "@com_google_absl//absl/time", + gtest, "//test/util:logging", "//test/util:memory_util", "//test/util:temp_path", @@ -1526,8 +1526,6 @@ cc_binary( "//test/util:test_util", "//test/util:thread_util", "//test/util:timer_util", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -1541,13 +1539,13 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + "@com_google_absl//absl/memory", + "@com_google_absl//absl/strings", + gtest, "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/memory", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) @@ -1559,11 +1557,11 @@ cc_binary( deps = [ "//test/util:capability_util", "//test/util:fs_util", + "@com_google_absl//absl/strings", + gtest, "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) @@ -1577,6 +1575,10 @@ cc_binary( "//test/util:cleanup", "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/strings", + "@com_google_absl//absl/synchronization", + "@com_google_absl//absl/time", + gtest, "//test/util:memory_util", "//test/util:posix_error", "//test/util:temp_path", @@ -1584,10 +1586,6 @@ cc_binary( "//test/util:thread_util", "//test/util:time_util", "//test/util:timer_util", - "@com_google_absl//absl/strings", - "@com_google_absl//absl/synchronization", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -1601,11 +1599,11 @@ cc_binary( "//test/util:capability_util", "//test/util:file_descriptor", "//test/util:fs_util", - "//test/util:test_main", - "//test/util:test_util", "@com_google_absl//absl/strings", "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", + gtest, + "//test/util:test_main", + "//test/util:test_util", ], ) @@ -1617,17 +1615,17 @@ cc_binary( deps = [ "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/container:flat_hash_set", + "@com_google_absl//absl/strings", + "@com_google_absl//absl/strings:str_format", + "@com_google_absl//absl/types:optional", + gtest, "//test/util:memory_util", "//test/util:posix_error", "//test/util:proc_util", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/container:flat_hash_set", - "@com_google_absl//absl/strings", - "@com_google_absl//absl/strings:str_format", - "@com_google_absl//absl/types:optional", - "@com_google_googletest//:gtest", ], ) @@ -1641,6 +1639,8 @@ cc_binary( "//test/util:cleanup", "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/strings", + gtest, "//test/util:logging", "//test/util:multiprocess_util", "//test/util:posix_error", @@ -1648,8 +1648,6 @@ cc_binary( "//test/util:test_main", "//test/util:test_util", "//test/util:time_util", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) @@ -1660,11 +1658,11 @@ cc_binary( linkstatic = 1, deps = [ ":base_poll_test", + "@com_google_absl//absl/time", + gtest, "//test/util:signal_util", "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -1674,6 +1672,9 @@ cc_binary( srcs = ["ptrace.cc"], linkstatic = 1, deps = [ + "@com_google_absl//absl/flags:flag", + "@com_google_absl//absl/time", + gtest, "//test/util:logging", "//test/util:multiprocess_util", "//test/util:platform_util", @@ -1681,9 +1682,6 @@ cc_binary( "//test/util:test_util", "//test/util:thread_util", "//test/util:time_util", - "@com_google_absl//absl/flags:flag", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -1693,10 +1691,10 @@ cc_binary( srcs = ["pwrite64.cc"], linkstatic = 1, deps = [ + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -1710,12 +1708,12 @@ cc_binary( deps = [ ":file_base", "//test/util:file_descriptor", + "@com_google_absl//absl/strings", + gtest, "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) @@ -1729,11 +1727,11 @@ cc_binary( ":unix_domain_socket_test_util", "//test/util:capability_util", "//test/util:file_descriptor", - "//test/util:test_main", - "//test/util:test_util", "@com_google_absl//absl/base:core_headers", "@com_google_absl//absl/base:endian", - "@com_google_googletest//:gtest", + gtest, + "//test/util:test_main", + "//test/util:test_util", ], ) @@ -1747,10 +1745,10 @@ cc_binary( ":unix_domain_socket_test_util", "//test/util:capability_util", "//test/util:file_descriptor", + "@com_google_absl//absl/base:core_headers", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/base:core_headers", - "@com_google_googletest//:gtest", ], ) @@ -1764,10 +1762,10 @@ cc_binary( ":unix_domain_socket_test_util", "//test/util:capability_util", "//test/util:file_descriptor", + "@com_google_absl//absl/base:core_headers", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/base:core_headers", - "@com_google_googletest//:gtest", ], ) @@ -1778,10 +1776,10 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -1792,10 +1790,10 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -1811,13 +1809,13 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + "@com_google_absl//absl/strings", + gtest, "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", "//test/util:timer_util", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) @@ -1832,12 +1830,12 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + "@com_google_absl//absl/strings", + gtest, "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) @@ -1851,11 +1849,11 @@ cc_binary( "//test/util:cleanup", "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/strings", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) @@ -1879,11 +1877,11 @@ cc_binary( linkstatic = 1, deps = [ "//test/syscalls/linux/rseq:lib", + gtest, "//test/util:logging", "//test/util:multiprocess_util", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -1894,11 +1892,11 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:cleanup", + gtest, "//test/util:logging", "//test/util:posix_error", "//test/util:signal_util", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -1908,9 +1906,9 @@ cc_binary( srcs = ["sched.cc"], linkstatic = 1, deps = [ + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -1920,9 +1918,9 @@ cc_binary( srcs = ["sched_yield.cc"], linkstatic = 1, deps = [ + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -1932,6 +1930,8 @@ cc_binary( srcs = ["seccomp.cc"], linkstatic = 1, deps = [ + "@com_google_absl//absl/base:core_headers", + gtest, "//test/util:logging", "//test/util:memory_util", "//test/util:multiprocess_util", @@ -1939,8 +1939,6 @@ cc_binary( "//test/util:proc_util", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/base:core_headers", - "@com_google_googletest//:gtest", ], ) @@ -1952,14 +1950,14 @@ cc_binary( deps = [ ":base_poll_test", "//test/util:file_descriptor", + "@com_google_absl//absl/time", + gtest, "//test/util:multiprocess_util", "//test/util:posix_error", "//test/util:rlimit_util", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -1971,13 +1969,13 @@ cc_binary( deps = [ "//test/util:eventfd_util", "//test/util:file_descriptor", + "@com_google_absl//absl/strings", + "@com_google_absl//absl/time", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/strings", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -1989,12 +1987,12 @@ cc_binary( deps = [ ":socket_test_util", "//test/util:file_descriptor", + "@com_google_absl//absl/strings", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) @@ -2005,13 +2003,13 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + "@com_google_absl//absl/strings", + "@com_google_absl//absl/time", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/strings", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -2021,9 +2019,9 @@ cc_binary( srcs = ["sigaction.cc"], linkstatic = 1, deps = [ + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -2038,13 +2036,13 @@ cc_binary( deps = [ "//test/util:cleanup", "//test/util:fs_util", + gtest, "//test/util:multiprocess_util", "//test/util:posix_error", "//test/util:signal_util", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_googletest//:gtest", ], ) @@ -2057,7 +2055,7 @@ cc_binary( ), linkstatic = 1, deps = [ - "@com_google_googletest//:gtest", + gtest, "//test/util:logging", "//test/util:signal_util", "//test/util:test_util", @@ -2075,14 +2073,14 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + "@com_google_absl//absl/synchronization", + gtest, "//test/util:logging", "//test/util:posix_error", "//test/util:signal_util", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/synchronization", - "@com_google_googletest//:gtest", ], ) @@ -2092,10 +2090,10 @@ cc_binary( srcs = ["sigprocmask.cc"], linkstatic = 1, deps = [ + gtest, "//test/util:signal_util", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -2105,13 +2103,13 @@ cc_binary( srcs = ["sigstop.cc"], linkstatic = 1, deps = [ + "@com_google_absl//absl/flags:flag", + "@com_google_absl//absl/time", + gtest, "//test/util:multiprocess_util", "//test/util:posix_error", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/flags:flag", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -2122,13 +2120,13 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + "@com_google_absl//absl/time", + gtest, "//test/util:logging", "//test/util:signal_util", "//test/util:test_util", "//test/util:thread_util", "//test/util:timer_util", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -2144,10 +2142,10 @@ cc_library( deps = [ ":socket_test_util", ":unix_domain_socket_test_util", - "//test/util:test_util", "@com_google_absl//absl/strings", "@com_google_absl//absl/strings:str_format", - "@com_google_googletest//:gtest", + gtest, + "//test/util:test_util", ], alwayslink = 1, ) @@ -2160,8 +2158,8 @@ cc_library( deps = [ ":socket_test_util", ":unix_domain_socket_test_util", + gtest, "//test/util:test_util", - "@com_google_googletest//:gtest", ], alwayslink = 1, ) @@ -2174,8 +2172,8 @@ cc_library( deps = [ ":socket_test_util", ":unix_domain_socket_test_util", + gtest, "//test/util:test_util", - "@com_google_googletest//:gtest", ], alwayslink = 1, ) @@ -2191,11 +2189,11 @@ cc_library( ], deps = [ ":socket_test_util", - "//test/util:test_util", - "//test/util:thread_util", "@com_google_absl//absl/memory", "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", + gtest, + "//test/util:test_util", + "//test/util:thread_util", ], alwayslink = 1, ) @@ -2212,8 +2210,8 @@ cc_library( deps = [ ":socket_test_util", ":unix_domain_socket_test_util", + gtest, "//test/util:test_util", - "@com_google_googletest//:gtest", ], alwayslink = 1, ) @@ -2230,9 +2228,9 @@ cc_library( deps = [ ":socket_test_util", ":unix_domain_socket_test_util", + gtest, "//test/util:memory_util", "//test/util:test_util", - "@com_google_googletest//:gtest", ], alwayslink = 1, ) @@ -2250,8 +2248,8 @@ cc_library( ":ip_socket_test_util", ":socket_test_util", ":unix_domain_socket_test_util", + gtest, "//test/util:test_util", - "@com_google_googletest//:gtest", ], alwayslink = 1, ) @@ -2268,8 +2266,8 @@ cc_library( deps = [ ":ip_socket_test_util", ":socket_test_util", + gtest, "//test/util:test_util", - "@com_google_googletest//:gtest", ], alwayslink = 1, ) @@ -2286,9 +2284,9 @@ cc_library( deps = [ ":ip_socket_test_util", ":socket_test_util", - "//test/util:test_util", "@com_google_absl//absl/memory", - "@com_google_googletest//:gtest", + gtest, + "//test/util:test_util", ], alwayslink = 1, ) @@ -2305,8 +2303,8 @@ cc_library( deps = [ ":ip_socket_test_util", ":socket_test_util", + gtest, "//test/util:test_util", - "@com_google_googletest//:gtest", ], alwayslink = 1, ) @@ -2323,8 +2321,8 @@ cc_library( deps = [ ":ip_socket_test_util", ":socket_test_util", + gtest, "//test/util:test_util", - "@com_google_googletest//:gtest", ], alwayslink = 1, ) @@ -2387,9 +2385,9 @@ cc_binary( deps = [ ":socket_test_util", ":unix_domain_socket_test_util", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -2419,9 +2417,9 @@ cc_binary( deps = [ ":socket_test_util", ":unix_domain_socket_test_util", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -2451,9 +2449,9 @@ cc_binary( deps = [ ":ip_socket_test_util", ":socket_test_util", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -2551,10 +2549,10 @@ cc_binary( ":socket_bind_to_device_util", ":socket_test_util", "//test/util:capability_util", + gtest, "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_googletest//:gtest", ], ) @@ -2570,10 +2568,10 @@ cc_binary( ":socket_bind_to_device_util", ":socket_test_util", "//test/util:capability_util", + gtest, "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_googletest//:gtest", ], ) @@ -2589,10 +2587,10 @@ cc_binary( ":socket_bind_to_device_util", ":socket_test_util", "//test/util:capability_util", + gtest, "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_googletest//:gtest", ], ) @@ -2638,9 +2636,9 @@ cc_binary( deps = [ ":ip_socket_test_util", ":socket_test_util", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -2719,15 +2717,15 @@ cc_binary( ":ip_socket_test_util", ":socket_test_util", "//test/util:file_descriptor", + "@com_google_absl//absl/memory", + "@com_google_absl//absl/strings", + "@com_google_absl//absl/time", + gtest, "//test/util:posix_error", "//test/util:save_util", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/memory", - "@com_google_absl//absl/strings", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -2739,9 +2737,9 @@ cc_binary( deps = [ ":socket_test_util", "//test/util:file_descriptor", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -2755,10 +2753,10 @@ cc_binary( ":socket_test_util", "//test/util:cleanup", "//test/util:file_descriptor", + "@com_google_absl//absl/strings:str_format", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/strings:str_format", - "@com_google_googletest//:gtest", ], ) @@ -2771,9 +2769,9 @@ cc_binary( ":socket_netlink_util", ":socket_test_util", "//test/util:file_descriptor", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -2791,9 +2789,9 @@ cc_library( deps = [ ":socket_test_util", ":unix_domain_socket_test_util", - "//test/util:test_util", "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", + gtest, + "//test/util:test_util", ], alwayslink = 1, ) @@ -2810,11 +2808,11 @@ cc_library( deps = [ ":socket_test_util", ":unix_domain_socket_test_util", + "@com_google_absl//absl/time", + gtest, "//test/util:test_util", "//test/util:thread_util", "//test/util:timer_util", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], alwayslink = 1, ) @@ -2831,10 +2829,10 @@ cc_library( deps = [ ":socket_test_util", ":unix_domain_socket_test_util", + "@com_google_absl//absl/strings", + gtest, "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], alwayslink = 1, ) @@ -2851,10 +2849,10 @@ cc_library( deps = [ ":socket_test_util", ":unix_domain_socket_test_util", + "@com_google_absl//absl/strings", + gtest, "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], alwayslink = 1, ) @@ -2871,11 +2869,11 @@ cc_library( deps = [ ":socket_test_util", ":unix_domain_socket_test_util", + "@com_google_absl//absl/time", + gtest, "//test/util:test_util", "//test/util:thread_util", "//test/util:timer_util", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], alwayslink = 1, ) @@ -2892,8 +2890,8 @@ cc_library( deps = [ ":socket_test_util", ":unix_domain_socket_test_util", + gtest, "//test/util:test_util", - "@com_google_googletest//:gtest", ], alwayslink = 1, ) @@ -2910,10 +2908,10 @@ cc_library( deps = [ ":socket_test_util", ":unix_domain_socket_test_util", + "@com_google_absl//absl/time", + gtest, "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], alwayslink = 1, ) @@ -3007,9 +3005,9 @@ cc_binary( deps = [ ":socket_test_util", ":unix_domain_socket_test_util", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -3021,9 +3019,9 @@ cc_binary( deps = [ ":socket_test_util", ":unix_domain_socket_test_util", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -3035,9 +3033,9 @@ cc_binary( deps = [ ":socket_test_util", ":unix_domain_socket_test_util", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -3052,9 +3050,9 @@ cc_binary( ":socket_blocking_test_cases", ":socket_test_util", ":unix_domain_socket_test_util", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -3069,9 +3067,9 @@ cc_binary( ":ip_socket_test_util", ":socket_blocking_test_cases", ":socket_test_util", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -3086,9 +3084,9 @@ cc_binary( ":socket_non_stream_blocking_test_cases", ":socket_test_util", ":unix_domain_socket_test_util", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -3103,9 +3101,9 @@ cc_binary( ":ip_socket_test_util", ":socket_non_stream_blocking_test_cases", ":socket_test_util", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -3121,9 +3119,9 @@ cc_binary( ":socket_unix_cmsg_test_cases", ":socket_unix_test_cases", ":unix_domain_socket_test_util", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -3135,9 +3133,9 @@ cc_binary( deps = [ ":socket_test_util", ":unix_domain_socket_test_util", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -3149,9 +3147,9 @@ cc_binary( deps = [ ":socket_test_util", ":unix_domain_socket_test_util", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -3164,10 +3162,10 @@ cc_binary( ":socket_netlink_util", ":socket_test_util", "//test/util:file_descriptor", + "@com_google_absl//absl/base:endian", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/base:endian", - "@com_google_googletest//:gtest", ], ) @@ -3183,12 +3181,12 @@ cc_binary( "//test/util:cleanup", "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/strings", + gtest, "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) @@ -3199,11 +3197,11 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + "@com_google_absl//absl/time", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -3217,12 +3215,12 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + "@com_google_absl//absl/strings", + gtest, "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) @@ -3235,10 +3233,10 @@ cc_binary( "//test/util:capability_util", "//test/util:file_descriptor", "//test/util:fs_util", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -3248,10 +3246,10 @@ cc_binary( srcs = ["sync.cc"], linkstatic = 1, deps = [ + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -3261,10 +3259,10 @@ cc_binary( srcs = ["sysinfo.cc"], linkstatic = 1, deps = [ + "@com_google_absl//absl/time", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -3274,9 +3272,9 @@ cc_binary( srcs = ["syslog.cc"], linkstatic = 1, deps = [ + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -3286,10 +3284,10 @@ cc_binary( srcs = ["sysret.cc"], linkstatic = 1, deps = [ + gtest, "//test/util:logging", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -3301,12 +3299,12 @@ cc_binary( deps = [ ":socket_test_util", "//test/util:file_descriptor", + "@com_google_absl//absl/time", + gtest, "//test/util:posix_error", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -3316,11 +3314,11 @@ cc_binary( srcs = ["tgkill.cc"], linkstatic = 1, deps = [ + gtest, "//test/util:signal_util", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_googletest//:gtest", ], ) @@ -3330,10 +3328,10 @@ cc_binary( srcs = ["time.cc"], linkstatic = 1, deps = [ + gtest, "//test/util:proc_util", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -3358,15 +3356,15 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:cleanup", + "@com_google_absl//absl/flags:flag", + "@com_google_absl//absl/time", + gtest, "//test/util:logging", "//test/util:multiprocess_util", "//test/util:posix_error", "//test/util:signal_util", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/flags:flag", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -3376,11 +3374,11 @@ cc_binary( srcs = ["tkill.cc"], linkstatic = 1, deps = [ + gtest, "//test/util:logging", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_googletest//:gtest", ], ) @@ -3394,11 +3392,11 @@ cc_binary( "//test/util:capability_util", "//test/util:cleanup", "//test/util:file_descriptor", + "@com_google_absl//absl/strings", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) @@ -3414,12 +3412,12 @@ cc_library( deps = [ ":socket_test_util", ":unix_domain_socket_test_util", + "@com_google_absl//absl/base:core_headers", + "@com_google_absl//absl/time", + gtest, "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/base:core_headers", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], alwayslink = 1, ) @@ -3442,9 +3440,9 @@ cc_binary( deps = [ ":socket_test_util", "//test/util:file_descriptor", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -3455,14 +3453,14 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:capability_util", + "@com_google_absl//absl/flags:flag", + "@com_google_absl//absl/strings", + gtest, "//test/util:posix_error", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", "//test/util:uid_util", - "@com_google_absl//absl/flags:flag", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) @@ -3473,11 +3471,11 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:capability_util", + "@com_google_absl//absl/strings", + gtest, "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) @@ -3490,11 +3488,11 @@ cc_binary( "//test/util:capability_util", "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/strings", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) @@ -3504,11 +3502,11 @@ cc_binary( srcs = ["unshare.cc"], linkstatic = 1, deps = [ + "@com_google_absl//absl/synchronization", + gtest, "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", - "@com_google_absl//absl/synchronization", - "@com_google_googletest//:gtest", ], ) @@ -3534,11 +3532,11 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:fs_util", + gtest, "//test/util:posix_error", "//test/util:proc_util", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -3548,13 +3546,13 @@ cc_binary( srcs = ["vfork.cc"], linkstatic = 1, deps = [ + "@com_google_absl//absl/flags:flag", + "@com_google_absl//absl/time", + gtest, "//test/util:logging", "//test/util:multiprocess_util", "//test/util:test_util", "//test/util:time_util", - "@com_google_absl//absl/flags:flag", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -3566,6 +3564,10 @@ cc_binary( deps = [ "//test/util:cleanup", "//test/util:file_descriptor", + "@com_google_absl//absl/strings", + "@com_google_absl//absl/synchronization", + "@com_google_absl//absl/time", + gtest, "//test/util:logging", "//test/util:multiprocess_util", "//test/util:posix_error", @@ -3574,10 +3576,6 @@ cc_binary( "//test/util:test_util", "//test/util:thread_util", "//test/util:time_util", - "@com_google_absl//absl/strings", - "@com_google_absl//absl/synchronization", - "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", ], ) @@ -3588,10 +3586,10 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:cleanup", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -3602,12 +3600,12 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:fs_util", + "@com_google_absl//absl/strings", + "@com_google_absl//absl/strings:str_format", + gtest, "//test/util:posix_error", "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/strings", - "@com_google_absl//absl/strings:str_format", - "@com_google_googletest//:gtest", ], ) @@ -3618,14 +3616,14 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:capability_util", - "//test/util:test_main", - "//test/util:test_util", - "//test/util:thread_util", "@com_google_absl//absl/base:core_headers", "@com_google_absl//absl/memory", "@com_google_absl//absl/synchronization", "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", + gtest, + "//test/util:test_main", + "//test/util:test_util", + "//test/util:thread_util", ], ) @@ -3651,10 +3649,10 @@ cc_binary( linkstatic = 1, deps = [ "//test/util:file_descriptor", + gtest, "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -3664,11 +3662,11 @@ cc_binary( srcs = ["vdso_clock_gettime.cc"], linkstatic = 1, deps = [ - "//test/util:test_main", - "//test/util:test_util", "@com_google_absl//absl/strings", "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", + gtest, + "//test/util:test_main", + "//test/util:test_util", ], ) @@ -3678,10 +3676,10 @@ cc_binary( srcs = ["vsyscall.cc"], linkstatic = 1, deps = [ + gtest, "//test/util:proc_util", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -3694,11 +3692,11 @@ cc_binary( ":unix_domain_socket_test_util", "//test/util:file_descriptor", "//test/util:fs_util", - "//test/util:test_main", - "//test/util:test_util", "@com_google_absl//absl/strings", "@com_google_absl//absl/strings:str_format", - "@com_google_googletest//:gtest", + gtest, + "//test/util:test_main", + "//test/util:test_util", ], ) @@ -3710,12 +3708,12 @@ cc_binary( deps = [ "//test/util:file_descriptor", "//test/util:fs_util", + gtest, "//test/util:memory_util", "//test/util:multiprocess_util", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_googletest//:gtest", ], ) @@ -3727,10 +3725,10 @@ cc_binary( deps = [ ":ip_socket_test_util", "//test/util:file_descriptor", + "@com_google_absl//absl/strings", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) @@ -3742,10 +3740,10 @@ cc_binary( deps = [ ":ip_socket_test_util", "//test/util:file_descriptor", + "@com_google_absl//absl/strings", + gtest, "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) @@ -3761,11 +3759,11 @@ cc_binary( "//test/util:capability_util", "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/strings", + gtest, "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", - "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", ], ) diff --git a/test/util/BUILD b/test/util/BUILD index 1ac8b3fd6..1f22ebe29 100644 --- a/test/util/BUILD +++ b/test/util/BUILD @@ -1,4 +1,4 @@ -load("//tools:defs.bzl", "cc_library", "cc_test", "select_system") +load("//tools:defs.bzl", "cc_library", "cc_test", "gtest", "select_system") package( default_visibility = ["//:sandbox"], @@ -41,7 +41,7 @@ cc_library( ":save_util", "@com_google_absl//absl/strings", "@com_google_absl//absl/strings:str_format", - "@com_google_googletest//:gtest", + gtest, ], ) @@ -55,7 +55,7 @@ cc_library( ":posix_error", ":test_util", "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", + gtest, ], ) @@ -67,7 +67,7 @@ cc_test( ":proc_util", ":test_main", ":test_util", - "@com_google_googletest//:gtest", + gtest, ], ) @@ -87,7 +87,7 @@ cc_library( ":file_descriptor", ":posix_error", "@com_google_absl//absl/strings", - "@com_google_googletest//:gtest", + gtest, ], ) @@ -101,7 +101,7 @@ cc_test( ":temp_path", ":test_main", ":test_util", - "@com_google_googletest//:gtest", + gtest, ], ) @@ -134,7 +134,7 @@ cc_library( ":cleanup", ":posix_error", ":test_util", - "@com_google_googletest//:gtest", + gtest, ], ) @@ -183,7 +183,7 @@ cc_library( "@com_google_absl//absl/base:core_headers", "@com_google_absl//absl/strings", "@com_google_absl//absl/types:variant", - "@com_google_googletest//:gtest", + gtest, ], ) @@ -194,7 +194,7 @@ cc_test( deps = [ ":posix_error", ":test_main", - "@com_google_googletest//:gtest", + gtest, ], ) @@ -218,7 +218,7 @@ cc_library( ":cleanup", ":posix_error", ":test_util", - "@com_google_googletest//:gtest", + gtest, ], ) @@ -233,7 +233,7 @@ cc_library( ":test_util", "@com_google_absl//absl/strings", "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", + gtest, ], ) @@ -259,7 +259,7 @@ cc_library( "@com_google_absl//absl/strings", "@com_google_absl//absl/strings:str_format", "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", + gtest, ], ) @@ -291,7 +291,7 @@ cc_library( ":posix_error", ":test_util", "@com_google_absl//absl/time", - "@com_google_googletest//:gtest", + gtest, ], ) @@ -302,7 +302,7 @@ cc_test( deps = [ ":test_main", ":test_util", - "@com_google_googletest//:gtest", + gtest, ], ) @@ -322,7 +322,7 @@ cc_library( ":file_descriptor", ":posix_error", ":save_util", - "@com_google_googletest//:gtest", + gtest, ], ) diff --git a/tools/build/defs.bzl b/tools/build/defs.bzl index d0556abd1..967c1f900 100644 --- a/tools/build/defs.bzl +++ b/tools/build/defs.bzl @@ -18,6 +18,7 @@ cc_test = _cc_test cc_toolchain = "@bazel_tools//tools/cpp:current_cc_toolchain" go_image = _go_image go_embed_data = _go_embed_data +gtest = "@com_google_googletest//:gtest" loopback = "//tools/build:loopback" proto_library = native.proto_library pkg_deb = _pkg_deb diff --git a/tools/defs.bzl b/tools/defs.bzl index 819f12b0d..ce677cbbf 100644 --- a/tools/defs.bzl +++ b/tools/defs.bzl @@ -7,7 +7,7 @@ change for Google-internal and bazel-compatible rules. load("//tools/go_stateify:defs.bzl", "go_stateify") load("//tools/go_marshal:defs.bzl", "go_marshal", "marshal_deps", "marshal_test_deps") -load("//tools/build:defs.bzl", _cc_binary = "cc_binary", _cc_flags_supplier = "cc_flags_supplier", _cc_library = "cc_library", _cc_proto_library = "cc_proto_library", _cc_test = "cc_test", _cc_toolchain = "cc_toolchain", _container_image = "container_image", _default_installer = "default_installer", _default_net_util = "default_net_util", _go_binary = "go_binary", _go_embed_data = "go_embed_data", _go_image = "go_image", _go_library = "go_library", _go_proto_library = "go_proto_library", _go_test = "go_test", _go_tool_library = "go_tool_library", _loopback = "loopback", _pkg_deb = "pkg_deb", _pkg_tar = "pkg_tar", _proto_library = "proto_library", _py_binary = "py_binary", _py_library = "py_library", _py_requirement = "py_requirement", _py_test = "py_test", _select_arch = "select_arch", _select_system = "select_system") +load("//tools/build:defs.bzl", _cc_binary = "cc_binary", _cc_flags_supplier = "cc_flags_supplier", _cc_library = "cc_library", _cc_proto_library = "cc_proto_library", _cc_test = "cc_test", _cc_toolchain = "cc_toolchain", _container_image = "container_image", _default_installer = "default_installer", _default_net_util = "default_net_util", _go_binary = "go_binary", _go_embed_data = "go_embed_data", _go_image = "go_image", _go_library = "go_library", _go_proto_library = "go_proto_library", _go_test = "go_test", _go_tool_library = "go_tool_library", _gtest = "gtest", _loopback = "loopback", _pkg_deb = "pkg_deb", _pkg_tar = "pkg_tar", _proto_library = "proto_library", _py_binary = "py_binary", _py_library = "py_library", _py_requirement = "py_requirement", _py_test = "py_test", _select_arch = "select_arch", _select_system = "select_system") # Delegate directly. cc_binary = _cc_binary @@ -20,6 +20,7 @@ go_embed_data = _go_embed_data go_image = _go_image go_test = _go_test go_tool_library = _go_tool_library +gtest = _gtest pkg_deb = _pkg_deb pkg_tar = _pkg_tar py_library = _py_library diff --git a/tools/images/BUILD b/tools/images/BUILD index f1699b184..fe11f08a3 100644 --- a/tools/images/BUILD +++ b/tools/images/BUILD @@ -1,4 +1,4 @@ -load("//tools:defs.bzl", "cc_binary") +load("//tools:defs.bzl", "cc_binary", "gtest") load("//tools/images:defs.bzl", "vm_image", "vm_test") package( @@ -32,8 +32,8 @@ cc_binary( srcs = ["test.cc"], linkstatic = 1, deps = [ + gtest, "//test/util:test_main", - "@com_google_googletest//:gtest", ], ) From 7c118f7e192d403e716807c0f75f3f6d077a31ba Mon Sep 17 00:00:00 2001 From: Ting-Yu Wang Date: Fri, 31 Jan 2020 09:55:51 -0800 Subject: [PATCH 03/29] KVM platform does not support 32bit. Fixes: //test/syscalls:32bit_test_runsc_kvm Ref change: 5d569408ef94c753b7aae9392b5e4ebf7e5ea50d PiperOrigin-RevId: 292563926 --- test/util/platform_util.cc | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) diff --git a/test/util/platform_util.cc b/test/util/platform_util.cc index 2724e63f3..c9200d381 100644 --- a/test/util/platform_util.cc +++ b/test/util/platform_util.cc @@ -20,10 +20,9 @@ namespace gvisor { namespace testing { PlatformSupport PlatformSupport32Bit() { - if (GvisorPlatform() == Platform::kPtrace) { + if (GvisorPlatform() == Platform::kPtrace || + GvisorPlatform() == Platform::kKVM) { return PlatformSupport::NotSupported; - } else if (GvisorPlatform() == Platform::kKVM) { - return PlatformSupport::Segfault; } else { return PlatformSupport::Allowed; } From bc3a24d62788bd6b881afeda230f1a5550a5709a Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Fri, 31 Jan 2020 11:49:37 -0800 Subject: [PATCH 04/29] Internal change. PiperOrigin-RevId: 292587459 --- kokoro/kythe/generate_xrefs.sh | 2 +- pkg/sentry/fs/g3doc/inotify.md | 16 ++++++++-------- pkg/sentry/mm/README.md | 8 ++++---- 3 files changed, 13 insertions(+), 13 deletions(-) diff --git a/kokoro/kythe/generate_xrefs.sh b/kokoro/kythe/generate_xrefs.sh index 7a0fbb3cd..323b0f77b 100644 --- a/kokoro/kythe/generate_xrefs.sh +++ b/kokoro/kythe/generate_xrefs.sh @@ -23,7 +23,7 @@ bazel version python3 -V -readonly KYTHE_VERSION='v0.0.39' +readonly KYTHE_VERSION='v0.0.41' readonly WORKDIR="$(mktemp -d)" readonly KYTHE_DIR="${WORKDIR}/kythe-${KYTHE_VERSION}" if [[ -n "$KOKORO_GIT_COMMIT" ]]; then diff --git a/pkg/sentry/fs/g3doc/inotify.md b/pkg/sentry/fs/g3doc/inotify.md index 71a577d9d..85063d4e6 100644 --- a/pkg/sentry/fs/g3doc/inotify.md +++ b/pkg/sentry/fs/g3doc/inotify.md @@ -112,11 +112,11 @@ attempts to queue a new event, it is already holding `fs.Watches.mu`. If we used `Inotify.mu` to also protect the event queue, this would violate the above lock ordering. -[dirent]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/fs/dirent.go -[event]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/fs/inotify_event.go -[fd_table]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/kernel/fd_table.go -[inode]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/fs/inode.go -[inode_watches]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/fs/inode_inotify.go -[inotify]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/fs/inotify.go -[syscall_dir]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/syscalls/linux/ -[watch]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/fs/inotify_watch.go +[dirent]: https://github.com/google/gvisor/blob/master/pkg/sentry/fs/dirent.go +[event]: https://github.com/google/gvisor/blob/master/pkg/sentry/fs/inotify_event.go +[fd_table]: https://github.com/google/gvisor/blob/master/pkg/sentry/kernel/fd_table.go +[inode]: https://github.com/google/gvisor/blob/master/pkg/sentry/fs/inode.go +[inode_watches]: https://github.com/google/gvisor/blob/master/pkg/sentry/fs/inode_inotify.go +[inotify]: https://github.com/google/gvisor/blob/master/pkg/sentry/fs/inotify.go +[syscall_dir]: https://github.com/google/gvisor/blob/master/pkg/sentry/syscalls/linux/ +[watch]: https://github.com/google/gvisor/blob/master/pkg/sentry/fs/inotify_watch.go diff --git a/pkg/sentry/mm/README.md b/pkg/sentry/mm/README.md index e1322e373..f4d43d927 100644 --- a/pkg/sentry/mm/README.md +++ b/pkg/sentry/mm/README.md @@ -274,7 +274,7 @@ In the sentry: methods [`platform.AddressSpace.MapFile` and `platform.AddressSpace.Unmap`][platform]. -[memmap]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/memmap/memmap.go -[mm]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/mm/mm.go -[pgalloc]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/pgalloc/pgalloc.go -[platform]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/platform/platform.go +[memmap]: https://github.com/google/gvisor/blob/master/pkg/sentry/memmap/memmap.go +[mm]: https://github.com/google/gvisor/blob/master/pkg/sentry/mm/mm.go +[pgalloc]: https://github.com/google/gvisor/blob/master/pkg/sentry/pgalloc/pgalloc.go +[platform]: https://github.com/google/gvisor/blob/master/pkg/sentry/platform/platform.go From 528dd1ec72fee1dd63c734fe92d1b972b5735b8f Mon Sep 17 00:00:00 2001 From: Ghanan Gowripalan Date: Fri, 31 Jan 2020 13:24:48 -0800 Subject: [PATCH 05/29] Extract multicast IP to Ethernet address mapping Test: header.TestEthernetAddressFromMulticastIPAddress PiperOrigin-RevId: 292604649 --- pkg/tcpip/header/eth.go | 41 ++++++++++++++++++++++++++++++++++ pkg/tcpip/header/eth_test.go | 34 ++++++++++++++++++++++++++++ pkg/tcpip/network/arp/arp.go | 19 ++-------------- pkg/tcpip/network/ipv6/icmp.go | 20 ++--------------- 4 files changed, 79 insertions(+), 35 deletions(-) diff --git a/pkg/tcpip/header/eth.go b/pkg/tcpip/header/eth.go index f5d2c127f..b1e92d2d7 100644 --- a/pkg/tcpip/header/eth.go +++ b/pkg/tcpip/header/eth.go @@ -134,3 +134,44 @@ func IsValidUnicastEthernetAddress(addr tcpip.LinkAddress) bool { // addr is a valid unicast ethernet address. return true } + +// EthernetAddressFromMulticastIPv4Address returns a multicast Ethernet address +// for a multicast IPv4 address. +// +// addr MUST be a multicast IPv4 address. +func EthernetAddressFromMulticastIPv4Address(addr tcpip.Address) tcpip.LinkAddress { + var linkAddrBytes [EthernetAddressSize]byte + // RFC 1112 Host Extensions for IP Multicasting + // + // 6.4. Extensions to an Ethernet Local Network Module: + // + // An IP host group address is mapped to an Ethernet multicast + // address by placing the low-order 23-bits of the IP address + // into the low-order 23 bits of the Ethernet multicast address + // 01-00-5E-00-00-00 (hex). + linkAddrBytes[0] = 0x1 + linkAddrBytes[2] = 0x5e + linkAddrBytes[3] = addr[1] & 0x7F + copy(linkAddrBytes[4:], addr[IPv4AddressSize-2:]) + return tcpip.LinkAddress(linkAddrBytes[:]) +} + +// EthernetAddressFromMulticastIPv6Address returns a multicast Ethernet address +// for a multicast IPv6 address. +// +// addr MUST be a multicast IPv6 address. +func EthernetAddressFromMulticastIPv6Address(addr tcpip.Address) tcpip.LinkAddress { + // RFC 2464 Transmission of IPv6 Packets over Ethernet Networks + // + // 7. Address Mapping -- Multicast + // + // An IPv6 packet with a multicast destination address DST, + // consisting of the sixteen octets DST[1] through DST[16], is + // transmitted to the Ethernet multicast address whose first + // two octets are the value 3333 hexadecimal and whose last + // four octets are the last four octets of DST. + linkAddrBytes := []byte(addr[IPv6AddressSize-EthernetAddressSize:]) + linkAddrBytes[0] = 0x33 + linkAddrBytes[1] = 0x33 + return tcpip.LinkAddress(linkAddrBytes[:]) +} diff --git a/pkg/tcpip/header/eth_test.go b/pkg/tcpip/header/eth_test.go index 6634c90f5..7a0014ad9 100644 --- a/pkg/tcpip/header/eth_test.go +++ b/pkg/tcpip/header/eth_test.go @@ -66,3 +66,37 @@ func TestIsValidUnicastEthernetAddress(t *testing.T) { }) } } + +func TestEthernetAddressFromMulticastIPv4Address(t *testing.T) { + tests := []struct { + name string + addr tcpip.Address + expectedLinkAddr tcpip.LinkAddress + }{ + { + name: "IPv4 Multicast without 24th bit set", + addr: "\xe0\x7e\xdc\xba", + expectedLinkAddr: "\x01\x00\x5e\x7e\xdc\xba", + }, + { + name: "IPv4 Multicast with 24th bit set", + addr: "\xe0\xfe\xdc\xba", + expectedLinkAddr: "\x01\x00\x5e\x7e\xdc\xba", + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + if got := EthernetAddressFromMulticastIPv4Address(test.addr); got != test.expectedLinkAddr { + t.Fatalf("got EthernetAddressFromMulticastIPv4Address(%s) = %s, want = %s", got, test.expectedLinkAddr) + } + }) + } +} + +func TestEthernetAddressFromMulticastIPv6Address(t *testing.T) { + addr := tcpip.Address("\xff\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x0f\x1a") + if got, want := EthernetAddressFromMulticastIPv6Address(addr), tcpip.LinkAddress("\x33\x33\x0d\x0e\x0f\x1a"); got != want { + t.Fatalf("got EthernetAddressFromMulticastIPv6Address(%s) = %s, want = %s", addr, got, want) + } +} diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index 1ceaebfbd..4da13c5df 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -178,24 +178,9 @@ func (*protocol) ResolveStaticAddress(addr tcpip.Address) (tcpip.LinkAddress, bo return broadcastMAC, true } if header.IsV4MulticastAddress(addr) { - // RFC 1112 Host Extensions for IP Multicasting - // - // 6.4. Extensions to an Ethernet Local Network Module: - // - // An IP host group address is mapped to an Ethernet multicast - // address by placing the low-order 23-bits of the IP address - // into the low-order 23 bits of the Ethernet multicast address - // 01-00-5E-00-00-00 (hex). - return tcpip.LinkAddress([]byte{ - 0x01, - 0x00, - 0x5e, - addr[header.IPv4AddressSize-3] & 0x7f, - addr[header.IPv4AddressSize-2], - addr[header.IPv4AddressSize-1], - }), true + return header.EthernetAddressFromMulticastIPv4Address(addr), true } - return "", false + return tcpip.LinkAddress([]byte(nil)), false } // SetOption implements NetworkProtocol. diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index dc20c0fd7..7491cfc41 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -441,23 +441,7 @@ func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, linkEP stack. // ResolveStaticAddress implements stack.LinkAddressResolver. func (*protocol) ResolveStaticAddress(addr tcpip.Address) (tcpip.LinkAddress, bool) { if header.IsV6MulticastAddress(addr) { - // RFC 2464 Transmission of IPv6 Packets over Ethernet Networks - // - // 7. Address Mapping -- Multicast - // - // An IPv6 packet with a multicast destination address DST, - // consisting of the sixteen octets DST[1] through DST[16], is - // transmitted to the Ethernet multicast address whose first - // two octets are the value 3333 hexadecimal and whose last - // four octets are the last four octets of DST. - return tcpip.LinkAddress([]byte{ - 0x33, - 0x33, - addr[header.IPv6AddressSize-4], - addr[header.IPv6AddressSize-3], - addr[header.IPv6AddressSize-2], - addr[header.IPv6AddressSize-1], - }), true + return header.EthernetAddressFromMulticastIPv6Address(addr), true } - return "", false + return tcpip.LinkAddress([]byte(nil)), false } From 77bf586db75b3dbd9dcb14c349bde8372d26425c Mon Sep 17 00:00:00 2001 From: Ghanan Gowripalan Date: Fri, 31 Jan 2020 13:54:57 -0800 Subject: [PATCH 06/29] Use multicast Ethernet address for multicast NDP As per RFC 2464 section 7, an IPv6 packet with a multicast destination address is transmitted to the mapped Ethernet multicast address. Test: - ipv6.TestLinkResolution - stack_test.TestDADResolve - stack_test.TestRouterSolicitation PiperOrigin-RevId: 292610529 --- pkg/tcpip/header/ipv6_test.go | 29 +++++++++++++++++++ pkg/tcpip/link/channel/channel.go | 29 ++++++++++++------- pkg/tcpip/network/ipv6/icmp.go | 6 +++- pkg/tcpip/network/ipv6/icmp_test.go | 12 ++++++-- pkg/tcpip/stack/ndp.go | 17 +++++++++++ pkg/tcpip/stack/ndp_test.go | 16 +++++++++- pkg/tcpip/stack/route.go | 4 ++- .../transport/tcp/testing/context/context.go | 6 +++- 8 files changed, 101 insertions(+), 18 deletions(-) diff --git a/pkg/tcpip/header/ipv6_test.go b/pkg/tcpip/header/ipv6_test.go index 29f54bc57..c3ad503aa 100644 --- a/pkg/tcpip/header/ipv6_test.go +++ b/pkg/tcpip/header/ipv6_test.go @@ -17,6 +17,7 @@ package header_test import ( "bytes" "crypto/sha256" + "fmt" "testing" "github.com/google/go-cmp/cmp" @@ -300,3 +301,31 @@ func TestScopeForIPv6Address(t *testing.T) { }) } } + +func TestSolicitedNodeAddr(t *testing.T) { + tests := []struct { + addr tcpip.Address + want tcpip.Address + }{ + { + addr: "\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x0f\xa0", + want: "\xff\x02\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\xff\x0e\x0f\xa0", + }, + { + addr: "\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\xdd\x0e\x0f\xa0", + want: "\xff\x02\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\xff\x0e\x0f\xa0", + }, + { + addr: "\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\xdd\x01\x02\x03", + want: "\xff\x02\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\xff\x01\x02\x03", + }, + } + + for _, test := range tests { + t.Run(fmt.Sprintf("%s", test.addr), func(t *testing.T) { + if got := header.SolicitedNodeAddr(test.addr); got != test.want { + t.Fatalf("got header.SolicitedNodeAddr(%s) = %s, want = %s", test.addr, got, test.want) + } + }) + } +} diff --git a/pkg/tcpip/link/channel/channel.go b/pkg/tcpip/link/channel/channel.go index 71b9da797..78d447acd 100644 --- a/pkg/tcpip/link/channel/channel.go +++ b/pkg/tcpip/link/channel/channel.go @@ -30,15 +30,16 @@ type PacketInfo struct { Pkt tcpip.PacketBuffer Proto tcpip.NetworkProtocolNumber GSO *stack.GSO + Route stack.Route } // Endpoint is link layer endpoint that stores outbound packets in a channel // and allows injection of inbound packets. type Endpoint struct { - dispatcher stack.NetworkDispatcher - mtu uint32 - linkAddr tcpip.LinkAddress - GSO bool + dispatcher stack.NetworkDispatcher + mtu uint32 + linkAddr tcpip.LinkAddress + LinkEPCapabilities stack.LinkEndpointCapabilities // c is where outbound packets are queued. c chan PacketInfo @@ -122,11 +123,7 @@ func (e *Endpoint) MTU() uint32 { // Capabilities implements stack.LinkEndpoint.Capabilities. func (e *Endpoint) Capabilities() stack.LinkEndpointCapabilities { - caps := stack.LinkEndpointCapabilities(0) - if e.GSO { - caps |= stack.CapabilityHardwareGSO - } - return caps + return e.LinkEPCapabilities } // GSOMaxSize returns the maximum GSO packet size. @@ -146,11 +143,16 @@ func (e *Endpoint) LinkAddress() tcpip.LinkAddress { } // WritePacket stores outbound packets into the channel. -func (e *Endpoint) WritePacket(_ *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { +func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { + // Clone r then release its resource so we only get the relevant fields from + // stack.Route without holding a reference to a NIC's endpoint. + route := r.Clone() + route.Release() p := PacketInfo{ Pkt: pkt, Proto: protocol, GSO: gso, + Route: route, } select { @@ -162,7 +164,11 @@ func (e *Endpoint) WritePacket(_ *stack.Route, gso *stack.GSO, protocol tcpip.Ne } // WritePackets stores outbound packets into the channel. -func (e *Endpoint) WritePackets(_ *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + // Clone r then release its resource so we only get the relevant fields from + // stack.Route without holding a reference to a NIC's endpoint. + route := r.Clone() + route.Release() payloadView := pkts[0].Data.ToView() n := 0 packetLoop: @@ -176,6 +182,7 @@ packetLoop: }, Proto: protocol, GSO: gso, + Route: route, } select { diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index 7491cfc41..60817d36d 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -408,10 +408,14 @@ func (*protocol) LinkAddressProtocol() tcpip.NetworkProtocolNumber { // LinkAddressRequest implements stack.LinkAddressResolver. func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, linkEP stack.LinkEndpoint) *tcpip.Error { snaddr := header.SolicitedNodeAddr(addr) + + // TODO(b/148672031): Use stack.FindRoute instead of manually creating the + // route here. Note, we would need the nicID to do this properly so the right + // NIC (associated to linkEP) is used to send the NDP NS message. r := &stack.Route{ LocalAddress: localAddr, RemoteAddress: snaddr, - RemoteLinkAddress: broadcastMAC, + RemoteLinkAddress: header.EthernetAddressFromMulticastIPv6Address(snaddr), } hdr := buffer.NewPrependable(int(linkEP.MaxHeaderLength()) + header.IPv6MinimumSize + header.ICMPv6NeighborAdvertSize) pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborAdvertSize)) diff --git a/pkg/tcpip/network/ipv6/icmp_test.go b/pkg/tcpip/network/ipv6/icmp_test.go index 7a6820643..d0e930e20 100644 --- a/pkg/tcpip/network/ipv6/icmp_test.go +++ b/pkg/tcpip/network/ipv6/icmp_test.go @@ -270,8 +270,9 @@ func (c *testContext) cleanup() { } type routeArgs struct { - src, dst *channel.Endpoint - typ header.ICMPv6Type + src, dst *channel.Endpoint + typ header.ICMPv6Type + remoteLinkAddr tcpip.LinkAddress } func routeICMPv6Packet(t *testing.T, args routeArgs, fn func(*testing.T, header.ICMPv6)) { @@ -292,6 +293,11 @@ func routeICMPv6Packet(t *testing.T, args routeArgs, fn func(*testing.T, header. t.Errorf("unexpected protocol number %d", pi.Proto) return } + + if len(args.remoteLinkAddr) != 0 && args.remoteLinkAddr != pi.Route.RemoteLinkAddress { + t.Errorf("got remote link address = %s, want = %s", pi.Route.RemoteLinkAddress, args.remoteLinkAddr) + } + ipv6 := header.IPv6(pi.Pkt.Header.View()) transProto := tcpip.TransportProtocolNumber(ipv6.NextHeader()) if transProto != header.ICMPv6ProtocolNumber { @@ -339,7 +345,7 @@ func TestLinkResolution(t *testing.T) { t.Fatalf("ep.Write(_) = _, , %s, want = _, , tcpip.ErrNoLinkAddress", err) } for _, args := range []routeArgs{ - {src: c.linkEP0, dst: c.linkEP1, typ: header.ICMPv6NeighborSolicit}, + {src: c.linkEP0, dst: c.linkEP1, typ: header.ICMPv6NeighborSolicit, remoteLinkAddr: header.EthernetAddressFromMulticastIPv6Address(header.SolicitedNodeAddr(lladdr1))}, {src: c.linkEP1, dst: c.linkEP0, typ: header.ICMPv6NeighborAdvert}, } { routeICMPv6Packet(t, args, func(t *testing.T, icmpv6 header.ICMPv6) { diff --git a/pkg/tcpip/stack/ndp.go b/pkg/tcpip/stack/ndp.go index 31294345d..6123fda33 100644 --- a/pkg/tcpip/stack/ndp.go +++ b/pkg/tcpip/stack/ndp.go @@ -538,6 +538,14 @@ func (ndp *ndpState) sendDADPacket(addr tcpip.Address) *tcpip.Error { r := makeRoute(header.IPv6ProtocolNumber, header.IPv6Any, snmc, ndp.nic.linkEP.LinkAddress(), ref, false, false) defer r.Release() + // Route should resolve immediately since snmc is a multicast address so a + // remote link address can be calculated without a resolution process. + if c, err := r.Resolve(nil); err != nil { + log.Fatalf("ndp: error when resolving route to send NDP NS for DAD (%s -> %s on NIC(%d)): %s", header.IPv6Any, snmc, ndp.nic.ID(), err) + } else if c != nil { + log.Fatalf("ndp: route resolution not immediate for route to send NDP NS for DAD (%s -> %s on NIC(%d))", header.IPv6Any, snmc, ndp.nic.ID()) + } + hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv6NeighborSolicitMinimumSize) pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborSolicitMinimumSize)) pkt.SetType(header.ICMPv6NeighborSolicit) @@ -1197,6 +1205,15 @@ func (ndp *ndpState) startSolicitingRouters() { r := makeRoute(header.IPv6ProtocolNumber, header.IPv6Any, header.IPv6AllRoutersMulticastAddress, ndp.nic.linkEP.LinkAddress(), ref, false, false) defer r.Release() + // Route should resolve immediately since + // header.IPv6AllRoutersMulticastAddress is a multicast address so a + // remote link address can be calculated without a resolution process. + if c, err := r.Resolve(nil); err != nil { + log.Fatalf("ndp: error when resolving route to send NDP RS (%s -> %s on NIC(%d)): %s", header.IPv6Any, header.IPv6AllRoutersMulticastAddress, ndp.nic.ID(), err) + } else if c != nil { + log.Fatalf("ndp: route resolution not immediate for route to send NDP RS (%s -> %s on NIC(%d))", header.IPv6Any, header.IPv6AllRoutersMulticastAddress, ndp.nic.ID()) + } + payloadSize := header.ICMPv6HeaderSize + header.NDPRSMinimumSize hdr := buffer.NewPrependable(header.IPv6MinimumSize + payloadSize) pkt := header.ICMPv6(hdr.Prepend(payloadSize)) diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go index bc7cfbcb4..8af8565f7 100644 --- a/pkg/tcpip/stack/ndp_test.go +++ b/pkg/tcpip/stack/ndp_test.go @@ -336,6 +336,7 @@ func TestDADResolve(t *testing.T) { opts.NDPConfigs.DupAddrDetectTransmits = test.dupAddrDetectTransmits e := channel.New(int(test.dupAddrDetectTransmits), 1280, linkAddr1) + e.LinkEPCapabilities |= stack.CapabilityResolutionRequired s := stack.New(opts) if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) @@ -413,6 +414,12 @@ func TestDADResolve(t *testing.T) { t.Fatalf("got Proto = %d, want = %d", p.Proto, header.IPv6ProtocolNumber) } + // Make sure the right remote link address is used. + snmc := header.SolicitedNodeAddr(addr1) + if want := header.EthernetAddressFromMulticastIPv6Address(snmc); p.Route.RemoteLinkAddress != want { + t.Errorf("got remote link address = %s, want = %s", p.Route.RemoteLinkAddress, want) + } + // Check NDP NS packet. // // As per RFC 4861 section 4.3, a possible option is the Source Link @@ -420,7 +427,7 @@ func TestDADResolve(t *testing.T) { // address of the packet is the unspecified address. checker.IPv6(t, p.Pkt.Header.View().ToVectorisedView().First(), checker.SrcAddr(header.IPv6Any), - checker.DstAddr(header.SolicitedNodeAddr(addr1)), + checker.DstAddr(snmc), checker.TTL(header.NDPHopLimit), checker.NDPNS( checker.NDPNSTargetAddress(addr1), @@ -3292,6 +3299,7 @@ func TestRouterSolicitation(t *testing.T) { t.Run(test.name, func(t *testing.T) { t.Parallel() e := channel.New(int(test.maxRtrSolicit), 1280, linkAddr1) + e.LinkEPCapabilities |= stack.CapabilityResolutionRequired waitForPkt := func(timeout time.Duration) { t.Helper() ctx, _ := context.WithTimeout(context.Background(), timeout) @@ -3304,6 +3312,12 @@ func TestRouterSolicitation(t *testing.T) { if p.Proto != header.IPv6ProtocolNumber { t.Fatalf("got Proto = %d, want = %d", p.Proto, header.IPv6ProtocolNumber) } + + // Make sure the right remote link address is used. + if want := header.EthernetAddressFromMulticastIPv6Address(header.IPv6AllRoutersMulticastAddress); p.Route.RemoteLinkAddress != want { + t.Errorf("got remote link address = %s, want = %s", p.Route.RemoteLinkAddress, want) + } + checker.IPv6(t, p.Pkt.Header.View(), checker.SrcAddr(header.IPv6Any), diff --git a/pkg/tcpip/stack/route.go b/pkg/tcpip/stack/route.go index 517f4b941..f565aafb2 100644 --- a/pkg/tcpip/stack/route.go +++ b/pkg/tcpip/stack/route.go @@ -225,7 +225,9 @@ func (r *Route) Release() { // Clone Clone a route such that the original one can be released and the new // one will remain valid. func (r *Route) Clone() Route { - r.ref.incRef() + if r.ref != nil { + r.ref.incRef() + } return *r } diff --git a/pkg/tcpip/transport/tcp/testing/context/context.go b/pkg/tcpip/transport/tcp/testing/context/context.go index 730ac4292..1e9a0dea3 100644 --- a/pkg/tcpip/transport/tcp/testing/context/context.go +++ b/pkg/tcpip/transport/tcp/testing/context/context.go @@ -1082,7 +1082,11 @@ func (c *Context) SACKEnabled() bool { // SetGSOEnabled enables or disables generic segmentation offload. func (c *Context) SetGSOEnabled(enable bool) { - c.linkEP.GSO = enable + if enable { + c.linkEP.LinkEPCapabilities |= stack.CapabilityHardwareGSO + } else { + c.linkEP.LinkEPCapabilities &^= stack.CapabilityHardwareGSO + } } // MSSWithoutOptions returns the value for the MSS used by the stack when no From 6c3072243dfbf70062de5f610e14fd6ed2ce5f32 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 31 Jan 2020 14:14:52 -0800 Subject: [PATCH 07/29] Implement file locks for regular tmpfs files in VFSv2. Add a file lock implementation that can be embedded into various filesystem implementations. Updates #1480 PiperOrigin-RevId: 292614758 --- pkg/sentry/fsimpl/tmpfs/BUILD | 3 + pkg/sentry/fsimpl/tmpfs/regular_file.go | 23 +++++++ pkg/sentry/fsimpl/tmpfs/regular_file_test.go | 56 +++++++++++++++ pkg/sentry/fsimpl/tmpfs/tmpfs.go | 43 ++++++++++++ pkg/sentry/vfs/lock/BUILD | 13 ++++ pkg/sentry/vfs/lock/lock.go | 72 ++++++++++++++++++++ 6 files changed, 210 insertions(+) create mode 100644 pkg/sentry/vfs/lock/BUILD create mode 100644 pkg/sentry/vfs/lock/lock.go diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD index c61366224..57abd5583 100644 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ b/pkg/sentry/fsimpl/tmpfs/BUILD @@ -38,6 +38,7 @@ go_library( "//pkg/sentry/arch", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/fs/lock", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/pipe", @@ -47,6 +48,7 @@ go_library( "//pkg/sentry/platform", "//pkg/sentry/usage", "//pkg/sentry/vfs", + "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserror", "//pkg/usermem", @@ -86,6 +88,7 @@ go_test( "//pkg/context", "//pkg/fspath", "//pkg/sentry/contexttest", + "//pkg/sentry/fs/lock", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/contexttest", "//pkg/sentry/vfs", diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index e9e6faf67..dab346a41 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -23,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" + "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" @@ -192,6 +193,28 @@ func (fd *regularFileFD) Sync(ctx context.Context) error { return nil } +// LockBSD implements vfs.FileDescriptionImpl.LockBSD. +func (fd *regularFileFD) LockBSD(ctx context.Context, uid lock.UniqueID, t lock.LockType, block lock.Blocker) error { + return fd.inode().lockBSD(uid, t, block) +} + +// UnlockBSD implements vfs.FileDescriptionImpl.UnlockBSD. +func (fd *regularFileFD) UnlockBSD(ctx context.Context, uid lock.UniqueID) error { + fd.inode().unlockBSD(uid) + return nil +} + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (fd *regularFileFD) LockPOSIX(ctx context.Context, uid lock.UniqueID, t lock.LockType, rng lock.LockRange, block lock.Blocker) error { + return fd.inode().lockPOSIX(uid, t, rng, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (fd *regularFileFD) UnlockPOSIX(ctx context.Context, uid lock.UniqueID, rng lock.LockRange) error { + fd.inode().unlockPOSIX(uid, rng) + return nil +} + // regularFileReadWriter implements safemem.Reader and Safemem.Writer. type regularFileReadWriter struct { file *regularFile diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go index 32552e261..2b52992ea 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go @@ -24,9 +24,11 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" ) @@ -260,6 +262,60 @@ func TestPWrite(t *testing.T) { } } +func TestLocks(t *testing.T) { + ctx := contexttest.Context(t) + fd, cleanup, err := newFileFD(ctx, 0644) + if err != nil { + t.Fatal(err) + } + defer cleanup() + + var ( + uid1 lock.UniqueID + uid2 lock.UniqueID + // Non-blocking. + block lock.Blocker + ) + + uid1 = 123 + uid2 = 456 + + if err := fd.Impl().LockBSD(ctx, uid1, lock.ReadLock, block); err != nil { + t.Fatalf("fd.Impl().LockBSD failed: err = %v", err) + } + if err := fd.Impl().LockBSD(ctx, uid2, lock.ReadLock, block); err != nil { + t.Fatalf("fd.Impl().LockBSD failed: err = %v", err) + } + if got, want := fd.Impl().LockBSD(ctx, uid2, lock.WriteLock, block), syserror.ErrWouldBlock; got != want { + t.Fatalf("fd.Impl().LockBSD failed: got = %v, want = %v", got, want) + } + if err := fd.Impl().UnlockBSD(ctx, uid1); err != nil { + t.Fatalf("fd.Impl().UnlockBSD failed: err = %v", err) + } + if err := fd.Impl().LockBSD(ctx, uid2, lock.WriteLock, block); err != nil { + t.Fatalf("fd.Impl().LockBSD failed: err = %v", err) + } + + rng1 := lock.LockRange{0, 1} + rng2 := lock.LockRange{1, 2} + + if err := fd.Impl().LockPOSIX(ctx, uid1, lock.ReadLock, rng1, block); err != nil { + t.Fatalf("fd.Impl().LockPOSIX failed: err = %v", err) + } + if err := fd.Impl().LockPOSIX(ctx, uid2, lock.ReadLock, rng2, block); err != nil { + t.Fatalf("fd.Impl().LockPOSIX failed: err = %v", err) + } + if err := fd.Impl().LockPOSIX(ctx, uid1, lock.WriteLock, rng1, block); err != nil { + t.Fatalf("fd.Impl().LockPOSIX failed: err = %v", err) + } + if got, want := fd.Impl().LockPOSIX(ctx, uid2, lock.ReadLock, rng1, block), syserror.ErrWouldBlock; got != want { + t.Fatalf("fd.Impl().LockPOSIX failed: got = %v, want = %v", got, want) + } + if err := fd.Impl().UnlockPOSIX(ctx, uid1, rng1); err != nil { + t.Fatalf("fd.Impl().UnlockPOSIX failed: err = %v", err) + } +} + func TestPRead(t *testing.T) { ctx := contexttest.Context(t) fd, cleanup, err := newFileFD(ctx, 0644) diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 88dbd6e35..2108d0f4d 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -30,10 +30,12 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) @@ -153,6 +155,9 @@ type inode struct { rdevMajor uint32 rdevMinor uint32 + // Advisory file locks, which lock at the inode level. + locks lock.FileLocks + impl interface{} // immutable } @@ -352,6 +357,44 @@ func (i *inode) setStat(stat linux.Statx) error { return nil } +// TODO(gvisor.dev/issue/1480): support file locking for file types other than regular. +func (i *inode) lockBSD(uid fslock.UniqueID, t fslock.LockType, block fslock.Blocker) error { + switch i.impl.(type) { + case *regularFile: + return i.locks.LockBSD(uid, t, block) + } + return syserror.EBADF +} + +// TODO(gvisor.dev/issue/1480): support file locking for file types other than regular. +func (i *inode) unlockBSD(uid fslock.UniqueID) error { + switch i.impl.(type) { + case *regularFile: + i.locks.UnlockBSD(uid) + return nil + } + return syserror.EBADF +} + +// TODO(gvisor.dev/issue/1480): support file locking for file types other than regular. +func (i *inode) lockPOSIX(uid fslock.UniqueID, t fslock.LockType, rng fslock.LockRange, block fslock.Blocker) error { + switch i.impl.(type) { + case *regularFile: + return i.locks.LockPOSIX(uid, t, rng, block) + } + return syserror.EBADF +} + +// TODO(gvisor.dev/issue/1480): support file locking for file types other than regular. +func (i *inode) unlockPOSIX(uid fslock.UniqueID, rng fslock.LockRange) error { + switch i.impl.(type) { + case *regularFile: + i.locks.UnlockPOSIX(uid, rng) + return nil + } + return syserror.EBADF +} + // allocatedBlocksForSize returns the number of 512B blocks needed to // accommodate the given size in bytes, as appropriate for struct // stat::st_blocks and struct statx::stx_blocks. (Note that this 512B block diff --git a/pkg/sentry/vfs/lock/BUILD b/pkg/sentry/vfs/lock/BUILD new file mode 100644 index 000000000..d9ab063b7 --- /dev/null +++ b/pkg/sentry/vfs/lock/BUILD @@ -0,0 +1,13 @@ +load("//tools:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "lock", + srcs = ["lock.go"], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/sentry/fs/lock", + "//pkg/syserror", + ], +) diff --git a/pkg/sentry/vfs/lock/lock.go b/pkg/sentry/vfs/lock/lock.go new file mode 100644 index 000000000..724dfe743 --- /dev/null +++ b/pkg/sentry/vfs/lock/lock.go @@ -0,0 +1,72 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package lock provides POSIX and BSD style file locking for VFS2 file +// implementations. +// +// The actual implementations can be found in the lock package under +// sentry/fs/lock. +package lock + +import ( + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" + "gvisor.dev/gvisor/pkg/syserror" +) + +// FileLocks supports POSIX and BSD style locks, which correspond to fcntl(2) +// and flock(2) respectively in Linux. It can be embedded into various file +// implementations for VFS2 that support locking. +// +// Note that in Linux these two types of locks are _not_ cooperative, because +// race and deadlock conditions make merging them prohibitive. We do the same +// and keep them oblivious to each other. +type FileLocks struct { + // bsd is a set of BSD-style advisory file wide locks, see flock(2). + bsd fslock.Locks + + // posix is a set of POSIX-style regional advisory locks, see fcntl(2). + posix fslock.Locks +} + +// LockBSD tries to acquire a BSD-style lock on the entire file. +func (fl *FileLocks) LockBSD(uid fslock.UniqueID, t fslock.LockType, block fslock.Blocker) error { + if fl.bsd.LockRegion(uid, t, fslock.LockRange{0, fslock.LockEOF}, block) { + return nil + } + return syserror.ErrWouldBlock +} + +// UnlockBSD releases a BSD-style lock on the entire file. +// +// This operation is always successful, even if there did not exist a lock on +// the requested region held by uid in the first place. +func (fl *FileLocks) UnlockBSD(uid fslock.UniqueID) { + fl.bsd.UnlockRegion(uid, fslock.LockRange{0, fslock.LockEOF}) +} + +// LockPOSIX tries to acquire a POSIX-style lock on a file region. +func (fl *FileLocks) LockPOSIX(uid fslock.UniqueID, t fslock.LockType, rng fslock.LockRange, block fslock.Blocker) error { + if fl.posix.LockRegion(uid, t, rng, block) { + return nil + } + return syserror.ErrWouldBlock +} + +// UnlockPOSIX releases a POSIX-style lock on a file region. +// +// This operation is always successful, even if there did not exist a lock on +// the requested region held by uid in the first place. +func (fl *FileLocks) UnlockPOSIX(uid fslock.UniqueID, rng fslock.LockRange) { + fl.posix.UnlockRegion(uid, rng) +} From 04cccaaeeed22a28a42fc4c1406b43a966a5d886 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Fri, 31 Jan 2020 14:44:50 -0800 Subject: [PATCH 08/29] Fix logic around AMD/Intel cases. If the support is Ignored, then the call is still executed. We simply rely on it to fall through to the int3. Therefore, we must also bail on the vendor check. PiperOrigin-RevId: 292620558 --- test/syscalls/linux/32bit.cc | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/test/syscalls/linux/32bit.cc b/test/syscalls/linux/32bit.cc index 2751fb4e7..9883aef61 100644 --- a/test/syscalls/linux/32bit.cc +++ b/test/syscalls/linux/32bit.cc @@ -102,7 +102,8 @@ TEST(Syscall32Bit, Int80) { } TEST(Syscall32Bit, Sysenter) { - if (PlatformSupport32Bit() == PlatformSupport::Allowed && + if ((PlatformSupport32Bit() == PlatformSupport::Allowed || + PlatformSupport32Bit() == PlatformSupport::Ignored) && GetCPUVendor() == CPUVendor::kAMD) { // SYSENTER is an illegal instruction in compatibility mode on AMD. EXPECT_EXIT(ExitGroup32(kSysenter, kExitCode), @@ -133,7 +134,8 @@ TEST(Syscall32Bit, Sysenter) { } TEST(Syscall32Bit, Syscall) { - if (PlatformSupport32Bit() == PlatformSupport::Allowed && + if ((PlatformSupport32Bit() == PlatformSupport::Allowed || + PlatformSupport32Bit() == PlatformSupport::Ignored) && GetCPUVendor() == CPUVendor::kIntel) { // SYSCALL is an illegal instruction in compatibility mode on Intel. EXPECT_EXIT(ExitGroup32(kSyscall, kExitCode), From 02997af5abd62d778fca4d01b047a6bdebab2090 Mon Sep 17 00:00:00 2001 From: Ian Gudger Date: Fri, 31 Jan 2020 15:08:17 -0800 Subject: [PATCH 09/29] Fix method comment to match method name. PiperOrigin-RevId: 292624867 --- pkg/tcpip/transport/tcp/accept.go | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/pkg/tcpip/transport/tcp/accept.go b/pkg/tcpip/transport/tcp/accept.go index 6101f2945..08afb7c17 100644 --- a/pkg/tcpip/transport/tcp/accept.go +++ b/pkg/tcpip/transport/tcp/accept.go @@ -271,8 +271,8 @@ func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, i return n, nil } -// createEndpoint creates a new endpoint in connected state and then performs -// the TCP 3-way handshake. +// createEndpointAndPerformHandshake creates a new endpoint in connected state +// and then performs the TCP 3-way handshake. func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *header.TCPSynOptions, queue *waiter.Queue) (*endpoint, *tcpip.Error) { // Create new endpoint. irs := s.sequenceNumber From 4d1a648c7c5db8a51416bff647260a1be3b5c12e Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Mon, 3 Feb 2020 11:39:01 -0800 Subject: [PATCH 10/29] Allow mlock in system call filters Go 1.14 has a workaround for a Linux 5.2-5.4 bug which requires mlock'ing the g stack to prevent register corruption. We need to allow this syscall until it is removed from Go. PiperOrigin-RevId: 292967478 --- runsc/boot/filter/config.go | 12 ++++++++++++ 1 file changed, 12 insertions(+) diff --git a/runsc/boot/filter/config.go b/runsc/boot/filter/config.go index 4fb9adca6..f8d351c7b 100644 --- a/runsc/boot/filter/config.go +++ b/runsc/boot/filter/config.go @@ -174,6 +174,18 @@ var allowedSyscalls = seccomp.SyscallRules{ syscall.SYS_LSEEK: {}, syscall.SYS_MADVISE: {}, syscall.SYS_MINCORE: {}, + // Used by the Go runtime as a temporarily workaround for a Linux + // 5.2-5.4 bug. + // + // See src/runtime/os_linux_x86.go. + // + // TODO(b/148688965): Remove once this is gone from Go. + syscall.SYS_MLOCK: []seccomp.Rule{ + { + seccomp.AllowAny{}, + seccomp.AllowValue(4096), + }, + }, syscall.SYS_MMAP: []seccomp.Rule{ { seccomp.AllowAny{}, From 9742daf3c201771d257c0d043347f4eebf3088e0 Mon Sep 17 00:00:00 2001 From: Eyal Soha Date: Mon, 3 Feb 2020 12:03:38 -0800 Subject: [PATCH 11/29] Add packetdrill tests that use docker. PiperOrigin-RevId: 292973224 --- test/packetdrill/BUILD | 8 + test/packetdrill/Dockerfile | 9 ++ test/packetdrill/defs.bzl | 85 ++++++++++ test/packetdrill/fin_wait2_timeout.pkt | 23 +++ test/packetdrill/packetdrill_setup.sh | 26 +++ test/packetdrill/packetdrill_test.sh | 213 +++++++++++++++++++++++++ 6 files changed, 364 insertions(+) create mode 100644 test/packetdrill/BUILD create mode 100644 test/packetdrill/Dockerfile create mode 100644 test/packetdrill/defs.bzl create mode 100644 test/packetdrill/fin_wait2_timeout.pkt create mode 100755 test/packetdrill/packetdrill_setup.sh create mode 100755 test/packetdrill/packetdrill_test.sh diff --git a/test/packetdrill/BUILD b/test/packetdrill/BUILD new file mode 100644 index 000000000..d113555b1 --- /dev/null +++ b/test/packetdrill/BUILD @@ -0,0 +1,8 @@ +load("defs.bzl", "packetdrill_test") + +package(licenses = ["notice"]) + +packetdrill_test( + name = "fin_wait2_timeout", + scripts = ["fin_wait2_timeout.pkt"], +) diff --git a/test/packetdrill/Dockerfile b/test/packetdrill/Dockerfile new file mode 100644 index 000000000..bd4451355 --- /dev/null +++ b/test/packetdrill/Dockerfile @@ -0,0 +1,9 @@ +FROM ubuntu:bionic + +RUN apt-get update +RUN apt-get install -y net-tools git iptables iputils-ping netcat tcpdump jq tar +RUN hash -r +RUN git clone --branch packetdrill-v2.0 \ + https://github.com/google/packetdrill.git +RUN cd packetdrill/gtests/net/packetdrill && ./configure && \ + apt-get install -y bison flex make && make diff --git a/test/packetdrill/defs.bzl b/test/packetdrill/defs.bzl new file mode 100644 index 000000000..582f97e0c --- /dev/null +++ b/test/packetdrill/defs.bzl @@ -0,0 +1,85 @@ +"""Defines a rule for packetdrill test targets.""" + +def _packetdrill_test_impl(ctx): + test_runner = ctx.executable._test_runner + runner = ctx.actions.declare_file("%s-runner" % ctx.label.name) + + script_paths = [] + for script in ctx.files.scripts: + script_paths.append(script.short_path) + runner_content = "\n".join([ + "#!/bin/bash", + # This test will run part in a distinct user namespace. This can cause + # permission problems, because all runfiles may not be owned by the + # current user, and no other users will be mapped in that namespace. + # Make sure that everything is readable here. + "find . -type f -exec chmod a+rx {} \\;", + "find . -type d -exec chmod a+rx {} \\;", + "%s %s --init_script %s -- %s\n" % ( + test_runner.short_path, + " ".join(ctx.attr.flags), + ctx.files._init_script[0].short_path, + " ".join(script_paths), + ), + ]) + ctx.actions.write(runner, runner_content, is_executable = True) + + transitive_files = depset() + if hasattr(ctx.attr._test_runner, "data_runfiles"): + transitive_files = depset(ctx.attr._test_runner.data_runfiles.files) + runfiles = ctx.runfiles( + files = [test_runner] + ctx.files._init_script + ctx.files.scripts, + transitive_files = transitive_files, + collect_default = True, + collect_data = True, + ) + return [DefaultInfo(executable = runner, runfiles = runfiles)] + +_packetdrill_test = rule( + attrs = { + "_test_runner": attr.label( + executable = True, + cfg = "host", + allow_files = True, + default = "packetdrill_test.sh", + ), + "_init_script": attr.label( + allow_single_file = True, + default = "packetdrill_setup.sh", + ), + "flags": attr.string_list( + mandatory = False, + default = [], + ), + "scripts": attr.label_list( + mandatory = True, + allow_files = True, + ), + }, + test = True, + implementation = _packetdrill_test_impl, +) + +_PACKETDRILL_TAGS = ["local", "manual"] + +def packetdrill_linux_test(name, **kwargs): + if "tags" not in kwargs: + kwargs["tags"] = _PACKETDRILL_TAGS + _packetdrill_test( + name = name + "_linux_test", + flags = ["--dut_platform", "linux"], + **kwargs + ) + +def packetdrill_netstack_test(name, **kwargs): + if "tags" not in kwargs: + kwargs["tags"] = _PACKETDRILL_TAGS + _packetdrill_test( + name = name + "_netstack_test", + flags = ["--dut_platform", "netstack"], + **kwargs + ) + +def packetdrill_test(**kwargs): + packetdrill_linux_test(**kwargs) + packetdrill_netstack_test(**kwargs) diff --git a/test/packetdrill/fin_wait2_timeout.pkt b/test/packetdrill/fin_wait2_timeout.pkt new file mode 100644 index 000000000..613f0bec9 --- /dev/null +++ b/test/packetdrill/fin_wait2_timeout.pkt @@ -0,0 +1,23 @@ +// Test that a socket in FIN_WAIT_2 eventually times out and a subsequent +// packet generates a RST. + +0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3 ++0 bind(3, ..., ...) = 0 + ++0 listen(3, 1) = 0 + +// Establish a connection without timestamps. ++0 < S 0:0(0) win 32792 ++0 > S. 0:0(0) ack 1 <...> ++0 < P. 1:1(0) ack 1 win 257 + ++0.100 accept(3, ..., ...) = 4 +// set FIN_WAIT2 timeout to 1 seconds. ++0.100 setsockopt(4, SOL_TCP, TCP_LINGER2, [1], 4) = 0 ++0 close(4) = 0 + ++0 > F. 1:1(0) ack 1 <...> ++0 < . 1:1(0) ack 2 win 257 + ++1.1 < . 1:1(0) ack 2 win 257 ++0 > R 2:2(0) win 0 diff --git a/test/packetdrill/packetdrill_setup.sh b/test/packetdrill/packetdrill_setup.sh new file mode 100755 index 000000000..b858072f0 --- /dev/null +++ b/test/packetdrill/packetdrill_setup.sh @@ -0,0 +1,26 @@ +#!/bin/bash + +# Copyright 2018 The gVisor Authors. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +# This script runs both within the sentry context and natively. It should tweak +# TCP parameters to match expectations found in the script files. +sysctl -q net.ipv4.tcp_sack=1 +sysctl -q net.ipv4.tcp_rmem="4096 2097152 $((8*1024*1024))" +sysctl -q net.ipv4.tcp_wmem="4096 2097152 $((8*1024*1024))" + +# There may be errors from the above, but they will show up in the test logs and +# we always want to proceed from this point. It's possible that values were +# already set correctly and the nodes were not available in the namespace. +exit 0 diff --git a/test/packetdrill/packetdrill_test.sh b/test/packetdrill/packetdrill_test.sh new file mode 100755 index 000000000..614d94d74 --- /dev/null +++ b/test/packetdrill/packetdrill_test.sh @@ -0,0 +1,213 @@ +#!/bin/bash + +# Copyright 2020 The gVisor Authors. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +# Run a packetdrill test. Two docker containers are made, one for the +# Device-Under-Test (DUT) and one for the test runner. Each is attached with +# two networks, one for control packets that aid the test and one for test +# packets which are sent as part of the test and observed for correctness. + +set -euxo pipefail + +function failure() { + local lineno=$1 + local msg=$2 + local filename="$0" + echo "FAIL: $filename:$lineno: $msg" +} +trap 'failure ${LINENO} "$BASH_COMMAND"' ERR + +declare -r LONGOPTS="dut_platform:,init_script:" + +# Don't use declare below so that the error from getopt will end the script. +PARSED=$(getopt --options "" --longoptions=$LONGOPTS --name "$0" -- "$@") + +eval set -- "$PARSED" + +while true; do + case "$1" in + --dut_platform) + declare -r DUT_PLATFORM="$2" + shift 2 + ;; + --init_script) + declare -r INIT_SCRIPT="$2" + shift 2 + ;; + --) + shift + break + ;; + *) + echo "Programming error" + exit 3 + esac +done + +# All the other arguments are scripts. +declare -r scripts="$@" + +# Check that the required flags are defined in a way that is safe for "set -u". +if [[ "${DUT_PLATFORM-}" == "netstack" ]]; then + declare -r RUNTIME="--runtime runsc-d" +elif [[ "${DUT_PLATFORM-}" == "linux" ]]; then + declare -r RUNTIME="" +else + echo "FAIL: Bad or missing --dut_platform argument: ${DUT_PLATFORM-}" + exit 2 +fi +if [[ ! -x "${INIT_SCRIPT-}" ]]; then + echo "FAIL: Bad or missing --init_script: ${INIT_SCRIPT-}" + exit 2 +fi + +# Variables specific to the control network and interface start with CTRL_. +# Variables specific to the test network and interface start with TEST_. +# Variables specific to the DUT start with DUT_. +# Variables specific to the test runner start with TEST_RUNNER_. +declare -r PACKETDRILL="/packetdrill/gtests/net/packetdrill/packetdrill" +# Use random numbers so that test networks don't collide. +declare -r CTRL_NET="ctrl_net-${RANDOM}${RANDOM}" +declare -r TEST_NET="test_net-${RANDOM}${RANDOM}" +declare -r tolerance_usecs=100000 +# On both DUT and test runner, testing packets are on the eth2 interface. +declare -r TEST_DEVICE="eth2" +# Number of bits in the *_NET_PREFIX variables. +declare -r NET_MASK="24" +function new_net_prefix() { + # Class C, 192.0.0.0 to 223.255.255.255, transitionally has mask 24. + echo "$(shuf -i 192-223 -n 1).$(shuf -i 0-255 -n 1).$(shuf -i 0-255 -n 1)" +} +# Last bits of the DUT's IP address. +declare -r DUT_NET_SUFFIX=".10" +# Control port. +declare -r CTRL_PORT="40000" +# Last bits of the test runner's IP address. +declare -r TEST_RUNNER_NET_SUFFIX=".20" +declare -r TIMEOUT="60" +declare -r IMAGE_TAG="gcr.io/gvisor-presubmit/packetdrill" + +# Make sure that docker is installed. +docker --version + +function finish { + local cleanup_success=1 + for net in "${CTRL_NET}" "${TEST_NET}"; do + # Kill all processes attached to ${net}. + for docker_command in "kill" "rm"; do + (docker network inspect "${net}" \ + --format '{{range $key, $value := .Containers}}{{$key}} {{end}}' \ + | xargs -r docker "${docker_command}") || \ + cleanup_success=0 + done + # Remove the network. + docker network rm "${net}" || \ + cleanup_success=0 + done + + if ((!$cleanup_success)); then + echo "FAIL: Cleanup command failed" + exit 4 + fi +} +trap finish EXIT + +# Subnet for control packets between test runner and DUT. +declare CTRL_NET_PREFIX=$(new_net_prefix) +while ! docker network create \ + "--subnet=${CTRL_NET_PREFIX}.0/${NET_MASK}" "${CTRL_NET}"; do + sleep 0.1 + declare CTRL_NET_PREFIX=$(new_net_prefix) +done + +# Subnet for the packets that are part of the test. +declare TEST_NET_PREFIX=$(new_net_prefix) +while ! docker network create \ + "--subnet=${TEST_NET_PREFIX}.0/${NET_MASK}" "${TEST_NET}"; do + sleep 0.1 + declare TEST_NET_PREFIX=$(new_net_prefix) +done + +docker pull "${IMAGE_TAG}" + +# Create the DUT container and connect to network. +DUT=$(docker create ${RUNTIME} --privileged --rm \ + --stop-timeout ${TIMEOUT} -it ${IMAGE_TAG}) +docker network connect "${CTRL_NET}" \ + --ip "${CTRL_NET_PREFIX}${DUT_NET_SUFFIX}" "${DUT}" \ + || (docker kill ${DUT}; docker rm ${DUT}; false) +docker network connect "${TEST_NET}" \ + --ip "${TEST_NET_PREFIX}${DUT_NET_SUFFIX}" "${DUT}" \ + || (docker kill ${DUT}; docker rm ${DUT}; false) +docker start "${DUT}" + +# Create the test runner container and connect to network. +TEST_RUNNER=$(docker create --privileged --rm \ + --stop-timeout ${TIMEOUT} -it ${IMAGE_TAG}) +docker network connect "${CTRL_NET}" \ + --ip "${CTRL_NET_PREFIX}${TEST_RUNNER_NET_SUFFIX}" "${TEST_RUNNER}" \ + || (docker kill ${TEST_RUNNER}; docker rm ${REST_RUNNER}; false) +docker network connect "${TEST_NET}" \ + --ip "${TEST_NET_PREFIX}${TEST_RUNNER_NET_SUFFIX}" "${TEST_RUNNER}" \ + || (docker kill ${TEST_RUNNER}; docker rm ${REST_RUNNER}; false) +docker start "${TEST_RUNNER}" + +# Run tcpdump in the test runner unbuffered, without dns resolution, just on the +# interface with the test packets. +docker exec -t ${TEST_RUNNER} tcpdump -U -n -i "${TEST_DEVICE}" & + +# Start a packetdrill server on the test_runner. The packetdrill server sends +# packets and asserts that they are received. +docker exec -d "${TEST_RUNNER}" \ + ${PACKETDRILL} --wire_server --wire_server_dev="${TEST_DEVICE}" \ + --wire_server_ip="${CTRL_NET_PREFIX}${TEST_RUNNER_NET_SUFFIX}" \ + --wire_server_port="${CTRL_PORT}" \ + --local_ip="${TEST_NET_PREFIX}${TEST_RUNNER_NET_SUFFIX}" \ + --remote_ip="${TEST_NET_PREFIX}${DUT_NET_SUFFIX}" + +# Because the Linux kernel receives the SYN-ACK but didn't send the SYN it will +# issue a RST. To prevent this IPtables can be used to filter those out. +docker exec "${TEST_RUNNER}" \ + iptables -A OUTPUT -p tcp --tcp-flags RST RST -j DROP + +# Wait for the packetdrill server on the test runner to come. Attempt to +# connect to it from the DUT every 100 milliseconds until success. +while ! docker exec "${DUT}" \ + nc -zv "${CTRL_NET_PREFIX}${TEST_RUNNER_NET_SUFFIX}" "${CTRL_PORT}"; do + sleep 0.1 +done + +# Copy the packetdrill setup script to the DUT. +docker cp -L "${INIT_SCRIPT}" "${DUT}:packetdrill_setup.sh" + +# Copy the packetdrill scripts to the DUT. +declare -a dut_scripts +for script in $scripts; do + docker cp -L "${script}" "${DUT}:$(basename ${script})" + dut_scripts+=("/$(basename ${script})") +done + +# Start a packetdrill client on the DUT. The packetdrill client runs POSIX +# socket commands and also sends instructions to the server. +docker exec -t "${DUT}" \ + ${PACKETDRILL} --wire_client --wire_client_dev="${TEST_DEVICE}" \ + --wire_server_ip="${CTRL_NET_PREFIX}${TEST_RUNNER_NET_SUFFIX}" \ + --wire_server_port="${CTRL_PORT}" \ + --local_ip="${TEST_NET_PREFIX}${DUT_NET_SUFFIX}" \ + --remote_ip="${TEST_NET_PREFIX}${TEST_RUNNER_NET_SUFFIX}" \ + --init_scripts=/packetdrill_setup.sh \ + --tolerance_usecs="${tolerance_usecs}" "${dut_scripts[@]}" + +echo PASS: No errors. From 80ce7f253783feac041ef197b144d920e57955ec Mon Sep 17 00:00:00 2001 From: Brad Burlage Date: Mon, 3 Feb 2020 12:08:46 -0800 Subject: [PATCH 12/29] Tag version_test as noguitar. PiperOrigin-RevId: 292974323 --- runsc/BUILD | 1 + 1 file changed, 1 insertion(+) diff --git a/runsc/BUILD b/runsc/BUILD index b35b41d81..375241921 100644 --- a/runsc/BUILD +++ b/runsc/BUILD @@ -117,4 +117,5 @@ sh_test( srcs = ["version_test.sh"], args = ["$(location :runsc)"], data = [":runsc"], + tags = ["noguitar"], ) From e7846e50f2df070a15dd33235b334e2223f715f3 Mon Sep 17 00:00:00 2001 From: Ting-Yu Wang Date: Mon, 3 Feb 2020 15:30:28 -0800 Subject: [PATCH 13/29] Reduce run time for //test/syscalls:socket_inet_loopback_test_runsc_ptrace. * Tests are picked for a shard differently. It now picks one test from each block, instead of picking the whole block. This makes the same kind of tests spreads across different shards. * Reduce the number of connect() calls in TCPListenClose. PiperOrigin-RevId: 293019281 --- runsc/testutil/testutil.go | 61 ++++++++++----------- test/runtimes/runner.go | 9 +-- test/syscalls/linux/socket_inet_loopback.cc | 13 +++-- test/syscalls/syscall_test_runner.go | 7 +-- 4 files changed, 43 insertions(+), 47 deletions(-) diff --git a/runsc/testutil/testutil.go b/runsc/testutil/testutil.go index fb22eae39..5d0b0ae54 100644 --- a/runsc/testutil/testutil.go +++ b/runsc/testutil/testutil.go @@ -434,43 +434,40 @@ func IsStatic(filename string) (bool, error) { return true, nil } -// TestBoundsForShard calculates the beginning and end indices for the test -// based on the TEST_SHARD_INDEX and TEST_TOTAL_SHARDS environment vars. The -// returned ints are the beginning (inclusive) and end (exclusive) of the -// subslice corresponding to the shard. If either of the env vars are not -// present, then the function will return bounds that include all tests. If -// there are more shards than there are tests, then the returned list may be -// empty. -func TestBoundsForShard(numTests int) (int, int, error) { +// TestIndicesForShard returns indices for this test shard based on the +// TEST_SHARD_INDEX and TEST_TOTAL_SHARDS environment vars. +// +// If either of the env vars are not present, then the function will return all +// tests. If there are more shards than there are tests, then the returned list +// may be empty. +func TestIndicesForShard(numTests int) ([]int, error) { var ( - begin = 0 - end = numTests + shardIndex = 0 + shardTotal = 1 ) - indexStr, totalStr := os.Getenv("TEST_SHARD_INDEX"), os.Getenv("TEST_TOTAL_SHARDS") - if indexStr == "" || totalStr == "" { - return begin, end, nil - } - // Parse index and total to ints. - shardIndex, err := strconv.Atoi(indexStr) - if err != nil { - return 0, 0, fmt.Errorf("invalid TEST_SHARD_INDEX %q: %v", indexStr, err) - } - shardTotal, err := strconv.Atoi(totalStr) - if err != nil { - return 0, 0, fmt.Errorf("invalid TEST_TOTAL_SHARDS %q: %v", totalStr, err) + indexStr, totalStr := os.Getenv("TEST_SHARD_INDEX"), os.Getenv("TEST_TOTAL_SHARDS") + if indexStr != "" && totalStr != "" { + // Parse index and total to ints. + var err error + shardIndex, err = strconv.Atoi(indexStr) + if err != nil { + return nil, fmt.Errorf("invalid TEST_SHARD_INDEX %q: %v", indexStr, err) + } + shardTotal, err = strconv.Atoi(totalStr) + if err != nil { + return nil, fmt.Errorf("invalid TEST_TOTAL_SHARDS %q: %v", totalStr, err) + } } // Calculate! - shardSize := int(math.Ceil(float64(numTests) / float64(shardTotal))) - begin = shardIndex * shardSize - end = ((shardIndex + 1) * shardSize) - if begin > numTests { - // Nothing to run. - return 0, 0, nil + var indices []int + numBlocks := int(math.Ceil(float64(numTests) / float64(shardTotal))) + for i := 0; i < numBlocks; i++ { + pick := i*shardTotal + shardIndex + if pick < numTests { + indices = append(indices, pick) + } } - if end > numTests { - end = numTests - } - return begin, end, nil + return indices, nil } diff --git a/test/runtimes/runner.go b/test/runtimes/runner.go index bec37c69d..ddb890dbc 100644 --- a/test/runtimes/runner.go +++ b/test/runtimes/runner.go @@ -20,7 +20,6 @@ import ( "flag" "fmt" "io" - "log" "os" "sort" "strings" @@ -101,17 +100,15 @@ func getTests(d dockerutil.Docker, blacklist map[string]struct{}) ([]testing.Int // shard. tests := strings.Fields(list) sort.Strings(tests) - begin, end, err := testutil.TestBoundsForShard(len(tests)) + indices, err := testutil.TestIndicesForShard(len(tests)) if err != nil { return nil, fmt.Errorf("TestsForShard() failed: %v", err) } - log.Printf("Got bounds [%d:%d) for shard out of %d total tests", begin, end, len(tests)) - tests = tests[begin:end] var itests []testing.InternalTest - for _, tc := range tests { + for _, tci := range indices { // Capture tc in this scope. - tc := tc + tc := tests[tci] itests = append(itests, testing.InternalTest{ Name: tc, F: func(t *testing.T) { diff --git a/test/syscalls/linux/socket_inet_loopback.cc b/test/syscalls/linux/socket_inet_loopback.cc index 3bf7081b9..b24618a88 100644 --- a/test/syscalls/linux/socket_inet_loopback.cc +++ b/test/syscalls/linux/socket_inet_loopback.cc @@ -325,6 +325,12 @@ TEST_P(SocketInetLoopbackTest, TCPListenClose) { TestAddress const& listener = param.listener; TestAddress const& connector = param.connector; + constexpr int kAcceptCount = 32; + constexpr int kBacklog = kAcceptCount * 2; + constexpr int kFDs = 128; + constexpr int kThreadCount = 4; + constexpr int kFDsPerThread = kFDs / kThreadCount; + // Create the listening socket. FileDescriptor listen_fd = ASSERT_NO_ERRNO_AND_VALUE( Socket(listener.family(), SOCK_STREAM, IPPROTO_TCP)); @@ -332,7 +338,7 @@ TEST_P(SocketInetLoopbackTest, TCPListenClose) { ASSERT_THAT(bind(listen_fd.get(), reinterpret_cast(&listen_addr), listener.addr_len), SyscallSucceeds()); - ASSERT_THAT(listen(listen_fd.get(), 1001), SyscallSucceeds()); + ASSERT_THAT(listen(listen_fd.get(), kBacklog), SyscallSucceeds()); // Get the port bound by the listening socket. socklen_t addrlen = listener.addr_len; @@ -345,9 +351,6 @@ TEST_P(SocketInetLoopbackTest, TCPListenClose) { DisableSave ds; // Too many system calls. sockaddr_storage conn_addr = connector.addr; ASSERT_NO_ERRNO(SetAddrPort(connector.family(), &conn_addr, port)); - constexpr int kFDs = 2048; - constexpr int kThreadCount = 4; - constexpr int kFDsPerThread = kFDs / kThreadCount; FileDescriptor clients[kFDs]; std::unique_ptr threads[kThreadCount]; for (int i = 0; i < kFDs; i++) { @@ -371,7 +374,7 @@ TEST_P(SocketInetLoopbackTest, TCPListenClose) { for (int i = 0; i < kThreadCount; i++) { threads[i]->Join(); } - for (int i = 0; i < 32; i++) { + for (int i = 0; i < kAcceptCount; i++) { auto accepted = ASSERT_NO_ERRNO_AND_VALUE(Accept(listen_fd.get(), nullptr, nullptr)); } diff --git a/test/syscalls/syscall_test_runner.go b/test/syscalls/syscall_test_runner.go index b9fd885ff..ae342b68c 100644 --- a/test/syscalls/syscall_test_runner.go +++ b/test/syscalls/syscall_test_runner.go @@ -450,17 +450,16 @@ func main() { } // Get subset of tests corresponding to shard. - begin, end, err := testutil.TestBoundsForShard(len(testCases)) + indices, err := testutil.TestIndicesForShard(len(testCases)) if err != nil { fatalf("TestsForShard() failed: %v", err) } - testCases = testCases[begin:end] // Run the tests. var tests []testing.InternalTest - for _, tc := range testCases { + for _, tci := range indices { // Capture tc. - tc := tc + tc := testCases[tci] testName := fmt.Sprintf("%s_%s", tc.Suite, tc.Name) tests = append(tests, testing.InternalTest{ Name: testName, From 6cd7901d7d5f9639e95fff3d8927ba8856a83f91 Mon Sep 17 00:00:00 2001 From: Brad Burlage Date: Mon, 3 Feb 2020 15:30:42 -0800 Subject: [PATCH 14/29] Add 1 Kokoro job per runtime test. PiperOrigin-RevId: 293019326 --- kokoro/runtime_tests/go1.12.cfg | 6 ++++++ kokoro/runtime_tests/java11.cfg | 6 ++++++ kokoro/runtime_tests/nodejs12.4.0.cfg | 6 ++++++ kokoro/runtime_tests/php7.3.6.cfg | 6 ++++++ kokoro/runtime_tests/python3.7.3.cfg | 6 ++++++ {scripts => kokoro/runtime_tests}/runtime_tests.sh | 0 6 files changed, 30 insertions(+) create mode 100644 kokoro/runtime_tests/go1.12.cfg create mode 100644 kokoro/runtime_tests/java11.cfg create mode 100644 kokoro/runtime_tests/nodejs12.4.0.cfg create mode 100644 kokoro/runtime_tests/php7.3.6.cfg create mode 100644 kokoro/runtime_tests/python3.7.3.cfg rename {scripts => kokoro/runtime_tests}/runtime_tests.sh (100%) diff --git a/kokoro/runtime_tests/go1.12.cfg b/kokoro/runtime_tests/go1.12.cfg new file mode 100644 index 000000000..024740ab2 --- /dev/null +++ b/kokoro/runtime_tests/go1.12.cfg @@ -0,0 +1,6 @@ +build_file: "github/kokoro/runtime_tests/runtime_tests.sh" + +env_vars { + key: "RUNTIME_TEST_NAME" + value: "go1.12" +} \ No newline at end of file diff --git a/kokoro/runtime_tests/java11.cfg b/kokoro/runtime_tests/java11.cfg new file mode 100644 index 000000000..f01d26153 --- /dev/null +++ b/kokoro/runtime_tests/java11.cfg @@ -0,0 +1,6 @@ +build_file: "github/kokoro/runtime_tests/runtime_tests.sh" + +env_vars { + key: "RUNTIME_TEST_NAME" + value: "java11" +} \ No newline at end of file diff --git a/kokoro/runtime_tests/nodejs12.4.0.cfg b/kokoro/runtime_tests/nodejs12.4.0.cfg new file mode 100644 index 000000000..d4861fb07 --- /dev/null +++ b/kokoro/runtime_tests/nodejs12.4.0.cfg @@ -0,0 +1,6 @@ +build_file: "github/kokoro/runtime_tests/runtime_tests.sh" + +env_vars { + key: "RUNTIME_TEST_NAME" + value: "nodejs12.4.0" +} \ No newline at end of file diff --git a/kokoro/runtime_tests/php7.3.6.cfg b/kokoro/runtime_tests/php7.3.6.cfg new file mode 100644 index 000000000..b737ed9cb --- /dev/null +++ b/kokoro/runtime_tests/php7.3.6.cfg @@ -0,0 +1,6 @@ +build_file: "github/kokoro/runtime_tests/runtime_tests.sh" + +env_vars { + key: "RUNTIME_TEST_NAME" + value: "php7.3.6" +} \ No newline at end of file diff --git a/kokoro/runtime_tests/python3.7.3.cfg b/kokoro/runtime_tests/python3.7.3.cfg new file mode 100644 index 000000000..971fcba05 --- /dev/null +++ b/kokoro/runtime_tests/python3.7.3.cfg @@ -0,0 +1,6 @@ +build_file: "github/kokoro/runtime_tests/runtime_tests.sh" + +env_vars { + key: "RUNTIME_TEST_NAME" + value: "python3.7.3" +} \ No newline at end of file diff --git a/scripts/runtime_tests.sh b/kokoro/runtime_tests/runtime_tests.sh similarity index 100% rename from scripts/runtime_tests.sh rename to kokoro/runtime_tests/runtime_tests.sh From f37e913a358820ea98013772dd2880cc8a3c9218 Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Mon, 3 Feb 2020 16:15:16 -0800 Subject: [PATCH 15/29] seccomp: allow to filter syscalls by instruction pointer PiperOrigin-RevId: 293029446 --- pkg/seccomp/seccomp.go | 20 ++++++++++++++++---- pkg/seccomp/seccomp_rules.go | 6 +++++- pkg/seccomp/seccomp_test.go | 27 +++++++++++++++++++++++++++ 3 files changed, 48 insertions(+), 5 deletions(-) diff --git a/pkg/seccomp/seccomp.go b/pkg/seccomp/seccomp.go index fc36efa23..55fd6967e 100644 --- a/pkg/seccomp/seccomp.go +++ b/pkg/seccomp/seccomp.go @@ -219,24 +219,36 @@ func addSyscallArgsCheck(p *bpf.ProgramBuilder, rules []Rule, action linux.BPFAc switch a := arg.(type) { case AllowAny: case AllowValue: + dataOffsetLow := seccompDataOffsetArgLow(i) + dataOffsetHigh := seccompDataOffsetArgHigh(i) + if i == RuleIP { + dataOffsetLow = seccompDataOffsetIPLow + dataOffsetHigh = seccompDataOffsetIPHigh + } high, low := uint32(a>>32), uint32(a) // assert arg_low == low - p.AddStmt(bpf.Ld|bpf.Abs|bpf.W, seccompDataOffsetArgLow(i)) + p.AddStmt(bpf.Ld|bpf.Abs|bpf.W, dataOffsetLow) p.AddJumpFalseLabel(bpf.Jmp|bpf.Jeq|bpf.K, low, 0, ruleViolationLabel(ruleSetIdx, sysno, ruleidx)) // assert arg_high == high - p.AddStmt(bpf.Ld|bpf.Abs|bpf.W, seccompDataOffsetArgHigh(i)) + p.AddStmt(bpf.Ld|bpf.Abs|bpf.W, dataOffsetHigh) p.AddJumpFalseLabel(bpf.Jmp|bpf.Jeq|bpf.K, high, 0, ruleViolationLabel(ruleSetIdx, sysno, ruleidx)) labelled = true case GreaterThan: + dataOffsetLow := seccompDataOffsetArgLow(i) + dataOffsetHigh := seccompDataOffsetArgHigh(i) + if i == RuleIP { + dataOffsetLow = seccompDataOffsetIPLow + dataOffsetHigh = seccompDataOffsetIPHigh + } labelGood := fmt.Sprintf("gt%v", i) high, low := uint32(a>>32), uint32(a) // assert arg_high < high - p.AddStmt(bpf.Ld|bpf.Abs|bpf.W, seccompDataOffsetArgHigh(i)) + p.AddStmt(bpf.Ld|bpf.Abs|bpf.W, dataOffsetHigh) p.AddJumpFalseLabel(bpf.Jmp|bpf.Jge|bpf.K, high, 0, ruleViolationLabel(ruleSetIdx, sysno, ruleidx)) // arg_high > high p.AddJumpFalseLabel(bpf.Jmp|bpf.Jeq|bpf.K, high, 0, ruleLabel(ruleSetIdx, sysno, ruleidx, labelGood)) // arg_low < low - p.AddStmt(bpf.Ld|bpf.Abs|bpf.W, seccompDataOffsetArgLow(i)) + p.AddStmt(bpf.Ld|bpf.Abs|bpf.W, dataOffsetLow) p.AddJumpFalseLabel(bpf.Jmp|bpf.Jgt|bpf.K, low, 0, ruleViolationLabel(ruleSetIdx, sysno, ruleidx)) p.AddLabel(ruleLabel(ruleSetIdx, sysno, ruleidx, labelGood)) labelled = true diff --git a/pkg/seccomp/seccomp_rules.go b/pkg/seccomp/seccomp_rules.go index 84c841d7f..06308cd29 100644 --- a/pkg/seccomp/seccomp_rules.go +++ b/pkg/seccomp/seccomp_rules.go @@ -62,7 +62,11 @@ func (a AllowValue) String() (s string) { // rule := Rule { // AllowValue(linux.ARCH_GET_FS | linux.ARCH_SET_FS), // arg0 // } -type Rule [6]interface{} +type Rule [7]interface{} // 6 arguments + RIP + +// RuleIP indicates what rules in the Rule array have to be applied to +// instruction pointer. +const RuleIP = 6 func (r Rule) String() (s string) { if len(r) == 0 { diff --git a/pkg/seccomp/seccomp_test.go b/pkg/seccomp/seccomp_test.go index abbee7051..da5a5e4b2 100644 --- a/pkg/seccomp/seccomp_test.go +++ b/pkg/seccomp/seccomp_test.go @@ -388,6 +388,33 @@ func TestBasic(t *testing.T) { }, }, }, + { + ruleSets: []RuleSet{ + { + Rules: SyscallRules{ + 1: []Rule{ + { + RuleIP: AllowValue(0x7aabbccdd), + }, + }, + }, + Action: linux.SECCOMP_RET_ALLOW, + }, + }, + defaultAction: linux.SECCOMP_RET_TRAP, + specs: []spec{ + { + desc: "IP: Syscall instruction pointer allowed", + data: seccompData{nr: 1, arch: linux.AUDIT_ARCH_X86_64, args: [6]uint64{}, instructionPointer: 0x7aabbccdd}, + want: linux.SECCOMP_RET_ALLOW, + }, + { + desc: "IP: Syscall instruction pointer disallowed", + data: seccompData{nr: 1, arch: linux.AUDIT_ARCH_X86_64, args: [6]uint64{}, instructionPointer: 0x711223344}, + want: linux.SECCOMP_RET_TRAP, + }, + }, + }, } { instrs, err := BuildProgram(test.ruleSets, test.defaultAction) if err != nil { From d7cd484091543827678f1548b8e5668a7a86e13f Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Tue, 4 Feb 2020 08:20:10 -0800 Subject: [PATCH 16/29] Add support for sentry internal pipe for gofer mounts Internal pipes are supported similarly to how internal UDS is done. It is also controlled by the same flag. Fixes #1102 PiperOrigin-RevId: 293150045 --- pkg/sentry/fs/gofer/BUILD | 2 + pkg/sentry/fs/gofer/cache_policy.go | 3 + pkg/sentry/fs/gofer/fifo.go | 40 ++++++ pkg/sentry/fs/gofer/gofer_test.go | 2 +- pkg/sentry/fs/gofer/path.go | 185 +++++++++++++++++--------- pkg/sentry/fs/gofer/session.go | 191 +++++++++++++++++---------- pkg/sentry/fs/gofer/session_state.go | 12 +- pkg/sentry/fs/gofer/socket.go | 8 +- test/syscalls/BUILD | 1 - 9 files changed, 298 insertions(+), 146 deletions(-) create mode 100644 pkg/sentry/fs/gofer/fifo.go diff --git a/pkg/sentry/fs/gofer/BUILD b/pkg/sentry/fs/gofer/BUILD index 971d3718e..fea135eea 100644 --- a/pkg/sentry/fs/gofer/BUILD +++ b/pkg/sentry/fs/gofer/BUILD @@ -9,6 +9,7 @@ go_library( "cache_policy.go", "context_file.go", "device.go", + "fifo.go", "file.go", "file_state.go", "fs.go", @@ -38,6 +39,7 @@ go_library( "//pkg/sentry/fs/fsutil", "//pkg/sentry/fs/host", "//pkg/sentry/kernel/auth", + "//pkg/sentry/kernel/pipe", "//pkg/sentry/kernel/time", "//pkg/sentry/memmap", "//pkg/sentry/socket/unix/transport", diff --git a/pkg/sentry/fs/gofer/cache_policy.go b/pkg/sentry/fs/gofer/cache_policy.go index ebea03c42..07a564e92 100644 --- a/pkg/sentry/fs/gofer/cache_policy.go +++ b/pkg/sentry/fs/gofer/cache_policy.go @@ -127,6 +127,9 @@ func (cp cachePolicy) revalidate(ctx context.Context, name string, parent, child childIops, ok := child.InodeOperations.(*inodeOperations) if !ok { + if _, ok := child.InodeOperations.(*fifo); ok { + return false + } panic(fmt.Sprintf("revalidating inode operations of unknown type %T", child.InodeOperations)) } parentIops, ok := parent.InodeOperations.(*inodeOperations) diff --git a/pkg/sentry/fs/gofer/fifo.go b/pkg/sentry/fs/gofer/fifo.go new file mode 100644 index 000000000..456557058 --- /dev/null +++ b/pkg/sentry/fs/gofer/fifo.go @@ -0,0 +1,40 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/fs" +) + +// +stateify savable +type fifo struct { + fs.InodeOperations + fileIops *inodeOperations +} + +var _ fs.InodeOperations = (*fifo)(nil) + +// Rename implements fs.InodeOperations. It forwards the call to the underlying +// file inode to handle the file rename. Note that file key remains the same +// after the rename to keep the endpoint mapping. +func (i *fifo) Rename(ctx context.Context, inode *fs.Inode, oldParent *fs.Inode, oldName string, newParent *fs.Inode, newName string, replacement bool) error { + return i.fileIops.Rename(ctx, inode, oldParent, oldName, newParent, newName, replacement) +} + +// StatFS implements fs.InodeOperations. +func (i *fifo) StatFS(ctx context.Context) (fs.Info, error) { + return i.fileIops.StatFS(ctx) +} diff --git a/pkg/sentry/fs/gofer/gofer_test.go b/pkg/sentry/fs/gofer/gofer_test.go index 0c2f89ae8..2df2fe889 100644 --- a/pkg/sentry/fs/gofer/gofer_test.go +++ b/pkg/sentry/fs/gofer/gofer_test.go @@ -61,7 +61,7 @@ func rootTest(t *testing.T, name string, cp cachePolicy, fn func(context.Context ctx := contexttest.Context(t) sattr, rootInodeOperations := newInodeOperations(ctx, s, contextFile{ file: rootFile, - }, root.QID, p9.AttrMaskAll(), root.Attr, false /* socket */) + }, root.QID, p9.AttrMaskAll(), root.Attr) m := fs.NewMountSource(ctx, s, &filesystem{}, fs.MountSourceFlags{}) rootInode := fs.NewInode(ctx, rootInodeOperations, m, sattr) diff --git a/pkg/sentry/fs/gofer/path.go b/pkg/sentry/fs/gofer/path.go index 0c1be05ef..a35c3a23d 100644 --- a/pkg/sentry/fs/gofer/path.go +++ b/pkg/sentry/fs/gofer/path.go @@ -23,14 +23,24 @@ import ( "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/sentry/device" "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // maxFilenameLen is the maximum length of a filename. This is dictated by 9P's // encoding of strings, which uses 2 bytes for the length prefix. const maxFilenameLen = (1 << 16) - 1 +func changeType(mode p9.FileMode, newType p9.FileMode) p9.FileMode { + if newType&^p9.FileModeMask != 0 { + panic(fmt.Sprintf("newType contained more bits than just file mode: %x", newType)) + } + clear := mode &^ p9.FileModeMask + return clear | newType +} + // Lookup loads an Inode at name into a Dirent based on the session's cache // policy. func (i *inodeOperations) Lookup(ctx context.Context, dir *fs.Inode, name string) (*fs.Dirent, error) { @@ -69,8 +79,25 @@ func (i *inodeOperations) Lookup(ctx context.Context, dir *fs.Inode, name string return nil, err } + if i.session().overrides != nil { + // Check if file belongs to a internal named pipe. Note that it doesn't need + // to check for sockets because it's done in newInodeOperations below. + deviceKey := device.MultiDeviceKey{ + Device: p9attr.RDev, + SecondaryDevice: i.session().connID, + Inode: qids[0].Path, + } + unlock := i.session().overrides.lock() + if pipeInode := i.session().overrides.getPipe(deviceKey); pipeInode != nil { + unlock() + pipeInode.IncRef() + return fs.NewDirent(ctx, pipeInode, name), nil + } + unlock() + } + // Construct the Inode operations. - sattr, node := newInodeOperations(ctx, i.fileState.s, newFile, qids[0], mask, p9attr, false) + sattr, node := newInodeOperations(ctx, i.fileState.s, newFile, qids[0], mask, p9attr) // Construct a positive Dirent. return fs.NewDirent(ctx, fs.NewInode(ctx, node, dir.MountSource, sattr), name), nil @@ -138,7 +165,7 @@ func (i *inodeOperations) Create(ctx context.Context, dir *fs.Inode, name string qid := qids[0] // Construct the InodeOperations. - sattr, iops := newInodeOperations(ctx, i.fileState.s, unopened, qid, mask, p9attr, false) + sattr, iops := newInodeOperations(ctx, i.fileState.s, unopened, qid, mask, p9attr) // Construct the positive Dirent. d := fs.NewDirent(ctx, fs.NewInode(ctx, iops, dir.MountSource, sattr), name) @@ -223,63 +250,22 @@ func (i *inodeOperations) Bind(ctx context.Context, dir *fs.Inode, name string, return nil, syserror.ENAMETOOLONG } - if i.session().endpoints == nil { + if i.session().overrides == nil { return nil, syscall.EOPNOTSUPP } - // Create replaces the directory fid with the newly created/opened - // file, so clone this directory so it doesn't change out from under - // this node. - _, newFile, err := i.fileState.file.walk(ctx, nil) - if err != nil { - return nil, err - } - // We're not going to use newFile after return. - defer newFile.close(ctx) - - // Stabilize the endpoint map while creation is in progress. - unlock := i.session().endpoints.lock() + // Stabilize the override map while creation is in progress. + unlock := i.session().overrides.lock() defer unlock() - // Create a regular file in the gofer and then mark it as a socket by - // adding this inode key in the 'endpoints' map. - owner := fs.FileOwnerFromContext(ctx) - hostFile, err := newFile.create(ctx, name, p9.ReadWrite, p9.FileMode(perm.LinuxMode()), p9.UID(owner.UID), p9.GID(owner.GID)) + sattr, iops, err := i.createEndpointFile(ctx, dir, name, perm, p9.ModeSocket) if err != nil { return nil, err } - // We're not going to use this file. - hostFile.Close() - - i.touchModificationAndStatusChangeTime(ctx, dir) - - // Get the attributes of the file to create inode key. - qid, mask, attr, err := getattr(ctx, newFile) - if err != nil { - return nil, err - } - - key := device.MultiDeviceKey{ - Device: attr.RDev, - SecondaryDevice: i.session().connID, - Inode: qid.Path, - } - - // Create child dirent. - - // Get an unopened p9.File for the file we created so that it can be - // cloned and re-opened multiple times after creation. - _, unopened, err := i.fileState.file.walk(ctx, []string{name}) - if err != nil { - return nil, err - } - - // Construct the InodeOperations. - sattr, iops := newInodeOperations(ctx, i.fileState.s, unopened, qid, mask, attr, true) // Construct the positive Dirent. childDir := fs.NewDirent(ctx, fs.NewInode(ctx, iops, dir.MountSource, sattr), name) - i.session().endpoints.add(key, childDir, ep) + i.session().overrides.addBoundEndpoint(iops.fileState.key, childDir, ep) return childDir, nil } @@ -294,33 +280,110 @@ func (i *inodeOperations) CreateFifo(ctx context.Context, dir *fs.Inode, name st // N.B. FIFOs use major/minor numbers 0. if _, err := i.fileState.file.mknod(ctx, name, mode, 0, 0, p9.UID(owner.UID), p9.GID(owner.GID)); err != nil { - return err + if i.session().overrides == nil || err != syscall.EPERM { + return err + } + // If gofer doesn't support mknod, check if we can create an internal fifo. + return i.createInternalFifo(ctx, dir, name, owner, perm) } i.touchModificationAndStatusChangeTime(ctx, dir) return nil } +func (i *inodeOperations) createInternalFifo(ctx context.Context, dir *fs.Inode, name string, owner fs.FileOwner, perm fs.FilePermissions) error { + if i.session().overrides == nil { + return syserror.EPERM + } + + // Stabilize the override map while creation is in progress. + unlock := i.session().overrides.lock() + defer unlock() + + sattr, fileOps, err := i.createEndpointFile(ctx, dir, name, perm, p9.ModeNamedPipe) + if err != nil { + return err + } + + // First create a pipe. + p := pipe.NewPipe(true /* isNamed */, pipe.DefaultPipeSize, usermem.PageSize) + + // Wrap the fileOps with our Fifo. + iops := &fifo{ + InodeOperations: pipe.NewInodeOperations(ctx, perm, p), + fileIops: fileOps, + } + inode := fs.NewInode(ctx, iops, dir.MountSource, sattr) + + // Construct the positive Dirent. + childDir := fs.NewDirent(ctx, fs.NewInode(ctx, iops, dir.MountSource, sattr), name) + i.session().overrides.addPipe(fileOps.fileState.key, childDir, inode) + return nil +} + +// Caller must hold Session.endpoint lock. +func (i *inodeOperations) createEndpointFile(ctx context.Context, dir *fs.Inode, name string, perm fs.FilePermissions, fileType p9.FileMode) (fs.StableAttr, *inodeOperations, error) { + _, dirClone, err := i.fileState.file.walk(ctx, nil) + if err != nil { + return fs.StableAttr{}, nil, err + } + // We're not going to use dirClone after return. + defer dirClone.close(ctx) + + // Create a regular file in the gofer and then mark it as a socket by + // adding this inode key in the 'overrides' map. + owner := fs.FileOwnerFromContext(ctx) + hostFile, err := dirClone.create(ctx, name, p9.ReadWrite, p9.FileMode(perm.LinuxMode()), p9.UID(owner.UID), p9.GID(owner.GID)) + if err != nil { + return fs.StableAttr{}, nil, err + } + // We're not going to use this file. + hostFile.Close() + + i.touchModificationAndStatusChangeTime(ctx, dir) + + // Get the attributes of the file to create inode key. + qid, mask, attr, err := getattr(ctx, dirClone) + if err != nil { + return fs.StableAttr{}, nil, err + } + + // Get an unopened p9.File for the file we created so that it can be + // cloned and re-opened multiple times after creation. + _, unopened, err := i.fileState.file.walk(ctx, []string{name}) + if err != nil { + return fs.StableAttr{}, nil, err + } + + // Construct new inode with file type overridden. + attr.Mode = changeType(attr.Mode, fileType) + sattr, iops := newInodeOperations(ctx, i.fileState.s, unopened, qid, mask, attr) + return sattr, iops, nil +} + // Remove implements InodeOperations.Remove. func (i *inodeOperations) Remove(ctx context.Context, dir *fs.Inode, name string) error { if len(name) > maxFilenameLen { return syserror.ENAMETOOLONG } - var key device.MultiDeviceKey - removeSocket := false - if i.session().endpoints != nil { - // Find out if file being deleted is a socket that needs to be + var key *device.MultiDeviceKey + if i.session().overrides != nil { + // Find out if file being deleted is a socket or pipe that needs to be // removed from endpoint map. if d, err := i.Lookup(ctx, dir, name); err == nil { defer d.DecRef() - if fs.IsSocket(d.Inode.StableAttr) { - child := d.Inode.InodeOperations.(*inodeOperations) - key = child.fileState.key - removeSocket = true - // Stabilize the endpoint map while deletion is in progress. - unlock := i.session().endpoints.lock() + if fs.IsSocket(d.Inode.StableAttr) || fs.IsPipe(d.Inode.StableAttr) { + switch iops := d.Inode.InodeOperations.(type) { + case *inodeOperations: + key = &iops.fileState.key + case *fifo: + key = &iops.fileIops.fileState.key + } + + // Stabilize the override map while deletion is in progress. + unlock := i.session().overrides.lock() defer unlock() } } @@ -329,8 +392,8 @@ func (i *inodeOperations) Remove(ctx context.Context, dir *fs.Inode, name string if err := i.fileState.file.unlinkAt(ctx, name, 0); err != nil { return err } - if removeSocket { - i.session().endpoints.remove(key) + if key != nil { + i.session().overrides.remove(*key) } i.touchModificationAndStatusChangeTime(ctx, dir) diff --git a/pkg/sentry/fs/gofer/session.go b/pkg/sentry/fs/gofer/session.go index 498c4645a..f6b3ef178 100644 --- a/pkg/sentry/fs/gofer/session.go +++ b/pkg/sentry/fs/gofer/session.go @@ -33,60 +33,107 @@ import ( var DefaultDirentCacheSize uint64 = fs.DefaultDirentCacheSize // +stateify savable -type endpointMaps struct { - // mu protexts the direntMap, the keyMap, and the pathMap below. - mu sync.RWMutex `state:"nosave"` +type overrideInfo struct { + dirent *fs.Dirent - // direntMap links sockets to their dirents. - // It is filled concurrently with the keyMap and is stored upon save. - // Before saving, this map is used to populate the pathMap. - direntMap map[transport.BoundEndpoint]*fs.Dirent + // endpoint is set when dirent points to a socket. inode must not be set. + endpoint transport.BoundEndpoint - // keyMap links MultiDeviceKeys (containing inode IDs) to their sockets. - // It is not stored during save because the inode ID may change upon restore. - keyMap map[device.MultiDeviceKey]transport.BoundEndpoint `state:"nosave"` - - // pathMap links the sockets to their paths. - // It is filled before saving from the direntMap and is stored upon save. - // Upon restore, this map is used to re-populate the keyMap. - pathMap map[transport.BoundEndpoint]string + // inode is set when dirent points to a pipe. endpoint must not be set. + inode *fs.Inode } -// add adds the endpoint to the maps. +func (l *overrideInfo) inodeType() fs.InodeType { + switch { + case l.endpoint != nil: + return fs.Socket + case l.inode != nil: + return fs.Pipe + } + panic("endpoint or node must be set") +} + +// +stateify savable +type overrideMaps struct { + // mu protexts the keyMap, and the pathMap below. + mu sync.RWMutex `state:"nosave"` + + // keyMap links MultiDeviceKeys (containing inode IDs) to their sockets/pipes. + // It is not stored during save because the inode ID may change upon restore. + keyMap map[device.MultiDeviceKey]*overrideInfo `state:"nosave"` + + // pathMap links the sockets/pipes to their paths. + // It is filled before saving from the direntMap and is stored upon save. + // Upon restore, this map is used to re-populate the keyMap. + pathMap map[*overrideInfo]string +} + +// addBoundEndpoint adds the bound endpoint to the map. // A reference is taken on the dirent argument. // // Precondition: maps must have been locked with 'lock'. -func (e *endpointMaps) add(key device.MultiDeviceKey, d *fs.Dirent, ep transport.BoundEndpoint) { - e.keyMap[key] = ep +func (e *overrideMaps) addBoundEndpoint(key device.MultiDeviceKey, d *fs.Dirent, ep transport.BoundEndpoint) { d.IncRef() - e.direntMap[ep] = d + e.keyMap[key] = &overrideInfo{dirent: d, endpoint: ep} +} + +// addPipe adds the pipe inode to the map. +// A reference is taken on the dirent argument. +// +// Precondition: maps must have been locked with 'lock'. +func (e *overrideMaps) addPipe(key device.MultiDeviceKey, d *fs.Dirent, inode *fs.Inode) { + d.IncRef() + e.keyMap[key] = &overrideInfo{dirent: d, inode: inode} } // remove deletes the key from the maps. // // Precondition: maps must have been locked with 'lock'. -func (e *endpointMaps) remove(key device.MultiDeviceKey) { - endpoint := e.get(key) +func (e *overrideMaps) remove(key device.MultiDeviceKey) { + endpoint := e.keyMap[key] delete(e.keyMap, key) - - d := e.direntMap[endpoint] - d.DecRef() - delete(e.direntMap, endpoint) + endpoint.dirent.DecRef() } // lock blocks other addition and removal operations from happening while // the backing file is being created or deleted. Returns a function that unlocks // the endpoint map. -func (e *endpointMaps) lock() func() { +func (e *overrideMaps) lock() func() { e.mu.Lock() return func() { e.mu.Unlock() } } -// get returns the endpoint mapped to the given key. +// getBoundEndpoint returns the bound endpoint mapped to the given key. // -// Precondition: maps must have been locked for reading. -func (e *endpointMaps) get(key device.MultiDeviceKey) transport.BoundEndpoint { - return e.keyMap[key] +// Precondition: maps must have been locked. +func (e *overrideMaps) getBoundEndpoint(key device.MultiDeviceKey) transport.BoundEndpoint { + if v := e.keyMap[key]; v != nil { + return v.endpoint + } + return nil +} + +// getPipe returns the pipe inode mapped to the given key. +// +// Precondition: maps must have been locked. +func (e *overrideMaps) getPipe(key device.MultiDeviceKey) *fs.Inode { + if v := e.keyMap[key]; v != nil { + return v.inode + } + return nil +} + +// getType returns the inode type if there is a corresponding endpoint for the +// given key. Returns false otherwise. +func (e *overrideMaps) getType(key device.MultiDeviceKey) (fs.InodeType, bool) { + e.mu.Lock() + v := e.keyMap[key] + e.mu.Unlock() + + if v != nil { + return v.inodeType(), true + } + return 0, false } // session holds state for each 9p session established during sys_mount. @@ -137,16 +184,16 @@ type session struct { // mounter is the EUID/EGID that mounted this file system. mounter fs.FileOwner `state:"wait"` - // endpoints is used to map inodes that represent socket files to their - // corresponding endpoint. Socket files are created as regular files in the - // gofer and their presence in this map indicate that they should indeed be - // socket files. This allows unix domain sockets to be used with paths that - // belong to a gofer. + // overrides is used to map inodes that represent socket/pipes files to their + // corresponding endpoint/iops. These files are created as regular files in + // the gofer and their presence in this map indicate that they should indeed + // be socket/pipe files. This allows unix domain sockets and named pipes to + // be used with paths that belong to a gofer. // // TODO(gvisor.dev/issue/1200): there are few possible races with someone // stat'ing the file and another deleting it concurrently, where the file // will not be reported as socket file. - endpoints *endpointMaps `state:"wait"` + overrides *overrideMaps `state:"wait"` } // Destroy tears down the session. @@ -179,15 +226,21 @@ func (s *session) SaveInodeMapping(inode *fs.Inode, path string) { // This is very unintuitive. We *CANNOT* trust the inode's StableAttrs, // because overlay copyUp may have changed them out from under us. // So much for "immutable". - sattr := inode.InodeOperations.(*inodeOperations).fileState.sattr - s.inodeMappings[sattr.InodeID] = path + switch iops := inode.InodeOperations.(type) { + case *inodeOperations: + s.inodeMappings[iops.fileState.sattr.InodeID] = path + case *fifo: + s.inodeMappings[iops.fileIops.fileState.sattr.InodeID] = path + default: + panic(fmt.Sprintf("Invalid type: %T", iops)) + } } -// newInodeOperations creates a new 9p fs.InodeOperations backed by a p9.File and attributes -// (p9.QID, p9.AttrMask, p9.Attr). +// newInodeOperations creates a new 9p fs.InodeOperations backed by a p9.File +// and attributes (p9.QID, p9.AttrMask, p9.Attr). // // Endpoints lock must not be held if socket == false. -func newInodeOperations(ctx context.Context, s *session, file contextFile, qid p9.QID, valid p9.AttrMask, attr p9.Attr, socket bool) (fs.StableAttr, *inodeOperations) { +func newInodeOperations(ctx context.Context, s *session, file contextFile, qid p9.QID, valid p9.AttrMask, attr p9.Attr) (fs.StableAttr, *inodeOperations) { deviceKey := device.MultiDeviceKey{ Device: attr.RDev, SecondaryDevice: s.connID, @@ -201,17 +254,11 @@ func newInodeOperations(ctx context.Context, s *session, file contextFile, qid p BlockSize: bsize(attr), } - if s.endpoints != nil { - if socket { - sattr.Type = fs.Socket - } else { - // If unix sockets are allowed on this filesystem, check if this file is - // supposed to be a socket file. - unlock := s.endpoints.lock() - if s.endpoints.get(deviceKey) != nil { - sattr.Type = fs.Socket - } - unlock() + if s.overrides != nil && sattr.Type == fs.RegularFile { + // If overrides are allowed on this filesystem, check if this file is + // supposed to be of a different type, e.g. socket. + if t, ok := s.overrides.getType(deviceKey); ok { + sattr.Type = t } } @@ -267,7 +314,7 @@ func Root(ctx context.Context, dev string, filesystem fs.Filesystem, superBlockF s.EnableLeakCheck("gofer.session") if o.privateunixsocket { - s.endpoints = newEndpointMaps() + s.overrides = newOverrideMaps() } // Construct the MountSource with the session and superBlockFlags. @@ -305,26 +352,24 @@ func Root(ctx context.Context, dev string, filesystem fs.Filesystem, superBlockF return nil, err } - sattr, iops := newInodeOperations(ctx, &s, s.attach, qid, valid, attr, false) + sattr, iops := newInodeOperations(ctx, &s, s.attach, qid, valid, attr) return fs.NewInode(ctx, iops, m, sattr), nil } -// newEndpointMaps creates a new endpointMaps. -func newEndpointMaps() *endpointMaps { - return &endpointMaps{ - direntMap: make(map[transport.BoundEndpoint]*fs.Dirent), - keyMap: make(map[device.MultiDeviceKey]transport.BoundEndpoint), - pathMap: make(map[transport.BoundEndpoint]string), +// newOverrideMaps creates a new overrideMaps. +func newOverrideMaps() *overrideMaps { + return &overrideMaps{ + keyMap: make(map[device.MultiDeviceKey]*overrideInfo), + pathMap: make(map[*overrideInfo]string), } } -// fillKeyMap populates key and dirent maps upon restore from saved -// pathmap. +// fillKeyMap populates key and dirent maps upon restore from saved pathmap. func (s *session) fillKeyMap(ctx context.Context) error { - unlock := s.endpoints.lock() + unlock := s.overrides.lock() defer unlock() - for ep, dirPath := range s.endpoints.pathMap { + for ep, dirPath := range s.overrides.pathMap { _, file, err := s.attach.walk(ctx, splitAbsolutePath(dirPath)) if err != nil { return fmt.Errorf("error filling endpointmaps, failed to walk to %q: %v", dirPath, err) @@ -341,25 +386,25 @@ func (s *session) fillKeyMap(ctx context.Context) error { Inode: qid.Path, } - s.endpoints.keyMap[key] = ep + s.overrides.keyMap[key] = ep } return nil } -// fillPathMap populates paths for endpoints from dirents in direntMap +// fillPathMap populates paths for overrides from dirents in direntMap // before save. func (s *session) fillPathMap() error { - unlock := s.endpoints.lock() + unlock := s.overrides.lock() defer unlock() - for ep, dir := range s.endpoints.direntMap { - mountRoot := dir.MountRoot() + for _, endpoint := range s.overrides.keyMap { + mountRoot := endpoint.dirent.MountRoot() defer mountRoot.DecRef() - dirPath, _ := dir.FullName(mountRoot) + dirPath, _ := endpoint.dirent.FullName(mountRoot) if dirPath == "" { return fmt.Errorf("error getting path from dirent") } - s.endpoints.pathMap[ep] = dirPath + s.overrides.pathMap[endpoint] = dirPath } return nil } @@ -368,7 +413,7 @@ func (s *session) fillPathMap() error { func (s *session) restoreEndpointMaps(ctx context.Context) error { // When restoring, only need to create the keyMap because the dirent and path // maps got stored through the save. - s.endpoints.keyMap = make(map[device.MultiDeviceKey]transport.BoundEndpoint) + s.overrides.keyMap = make(map[device.MultiDeviceKey]*overrideInfo) if err := s.fillKeyMap(ctx); err != nil { return fmt.Errorf("failed to insert sockets into endpoint map: %v", err) } @@ -376,6 +421,6 @@ func (s *session) restoreEndpointMaps(ctx context.Context) error { // Re-create pathMap because it can no longer be trusted as socket paths can // change while process continues to run. Empty pathMap will be re-filled upon // next save. - s.endpoints.pathMap = make(map[transport.BoundEndpoint]string) + s.overrides.pathMap = make(map[*overrideInfo]string) return nil } diff --git a/pkg/sentry/fs/gofer/session_state.go b/pkg/sentry/fs/gofer/session_state.go index 0285c5361..111da59f9 100644 --- a/pkg/sentry/fs/gofer/session_state.go +++ b/pkg/sentry/fs/gofer/session_state.go @@ -25,9 +25,9 @@ import ( // beforeSave is invoked by stateify. func (s *session) beforeSave() { - if s.endpoints != nil { + if s.overrides != nil { if err := s.fillPathMap(); err != nil { - panic("failed to save paths to endpoint map before saving" + err.Error()) + panic("failed to save paths to override map before saving" + err.Error()) } } } @@ -74,10 +74,10 @@ func (s *session) afterLoad() { panic(fmt.Sprintf("new attach name %v, want %v", opts.aname, s.aname)) } - // Check if endpointMaps exist when uds sockets are enabled - // (only pathmap will actualy have been saved). - if opts.privateunixsocket != (s.endpoints != nil) { - panic(fmt.Sprintf("new privateunixsocket option %v, want %v", opts.privateunixsocket, s.endpoints != nil)) + // Check if overrideMaps exist when uds sockets are enabled (only pathmaps + // will actually have been saved). + if opts.privateunixsocket != (s.overrides != nil) { + panic(fmt.Sprintf("new privateunixsocket option %v, want %v", opts.privateunixsocket, s.overrides != nil)) } if args.Flags != s.superBlockFlags { panic(fmt.Sprintf("new mount flags %v, want %v", args.Flags, s.superBlockFlags)) diff --git a/pkg/sentry/fs/gofer/socket.go b/pkg/sentry/fs/gofer/socket.go index 376cfce2c..10ba2f5f0 100644 --- a/pkg/sentry/fs/gofer/socket.go +++ b/pkg/sentry/fs/gofer/socket.go @@ -32,15 +32,15 @@ func (i *inodeOperations) BoundEndpoint(inode *fs.Inode, path string) transport. return nil } - if i.session().endpoints != nil { - unlock := i.session().endpoints.lock() + if i.session().overrides != nil { + unlock := i.session().overrides.lock() defer unlock() - ep := i.session().endpoints.get(i.fileState.key) + ep := i.session().overrides.getBoundEndpoint(i.fileState.key) if ep != nil { return ep } - // Not found in endpoints map, it may be a gofer backed unix socket... + // Not found in overrides map, it may be a gofer backed unix socket... } inode.IncRef() diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 40e974314..8f2b75a1c 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -225,7 +225,6 @@ syscall_test( syscall_test( add_overlay = True, test = "//test/syscalls/linux:mknod_test", - use_tmpfs = True, # mknod is not supported over gofer. ) syscall_test( From 492229d0176c1af2ab4ea4cf91bf211e940b5b12 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Tue, 4 Feb 2020 11:28:36 -0800 Subject: [PATCH 17/29] VFS2 gofer client Updates #1198 Opening host pipes (by spinning in fdpipe) and host sockets is not yet complete, and will be done in a future CL. Major differences from VFS1 gofer client (sentry/fs/gofer), with varying levels of backportability: - "Cache policies" are replaced by InteropMode, which control the behavior of timestamps in addition to caching. Under InteropModeExclusive (analogous to cacheAll) and InteropModeWritethrough (analogous to cacheAllWritethrough), client timestamps are *not* written back to the server (it is not possible in 9P or Linux for clients to set ctime, so writing back client-authoritative timestamps results in incoherence between atime/mtime and ctime). Under InteropModeShared (analogous to cacheRemoteRevalidating), client timestamps are not used at all (remote filesystem clocks are authoritative). cacheNone is translated to InteropModeShared + new option filesystemOptions.specialRegularFiles. - Under InteropModeShared, "unstable attribute" reloading for permission checks, lookup, and revalidation are fused, which is feasible in VFS2 since gofer.filesystem controls path resolution. This results in a ~33% reduction in RPCs for filesystem operations compared to cacheRemoteRevalidating. For example, consider stat("/foo/bar/baz") where "/foo/bar/baz" fails revalidation, resulting in the instantiation of a new dentry: VFS1 RPCs: getattr("/") // fs.MountNamespace.FindLink() => fs.Inode.CheckPermission() => gofer.inodeOperations.check() => gofer.inodeOperations.UnstableAttr() walkgetattr("/", "foo") = fid1 // fs.Dirent.walk() => gofer.session.Revalidate() => gofer.cachePolicy.Revalidate() clunk(fid1) getattr("/foo") // CheckPermission walkgetattr("/foo", "bar") = fid2 // Revalidate clunk(fid2) getattr("/foo/bar") // CheckPermission walkgetattr("/foo/bar", "baz") = fid3 // Revalidate clunk(fid3) walkgetattr("/foo/bar", "baz") = fid4 // fs.Dirent.walk() => gofer.inodeOperations.Lookup getattr("/foo/bar/baz") // linux.stat() => gofer.inodeOperations.UnstableAttr() VFS2 RPCs: getattr("/") // gofer.filesystem.walkExistingLocked() walkgetattr("/", "foo") = fid1 // gofer.filesystem.stepExistingLocked() clunk(fid1) // No getattr: walkgetattr already updated metadata for permission check walkgetattr("/foo", "bar") = fid2 clunk(fid2) walkgetattr("/foo/bar", "baz") = fid3 // No clunk: fid3 used for new gofer.dentry // No getattr: walkgetattr already updated metadata for stat() - gofer.filesystem.unlinkAt() does not require instantiation of a dentry that represents the file to be deleted. Updates #898. - gofer.regularFileFD.OnClose() skips Tflushf for regular files under InteropModeExclusive, as it's nonsensical to request a remote file flush without flushing locally-buffered writes to that remote file first. - Symlink targets are cached when InteropModeShared is not in effect. - p9.QID.Path (which is already required to be unique for each file within a server, and is accordingly already synthesized from device/inode numbers in all known gofers) is used as-is for inode numbers, rather than being mapped along with attr.RDev in the client to yet another synthetic inode number. - Relevant parts of fsutil.CachingInodeOperations are inlined directly into gofer package code. This avoids having to duplicate part of its functionality in fsutil.HostMappable. PiperOrigin-RevId: 293190213 --- pkg/safemem/seq_unsafe.go | 17 + pkg/sentry/fs/fsutil/BUILD | 4 +- pkg/sentry/fs/fsutil/frame_ref_set.go | 13 +- pkg/sentry/fs/fsutil/inode_cached.go | 2 +- pkg/sentry/fsimpl/gofer/BUILD | 55 ++ pkg/sentry/fsimpl/gofer/directory.go | 190 ++++ pkg/sentry/fsimpl/gofer/filesystem.go | 1087 ++++++++++++++++++++ pkg/sentry/fsimpl/gofer/gofer.go | 1147 ++++++++++++++++++++++ pkg/sentry/fsimpl/gofer/handle.go | 135 +++ pkg/sentry/fsimpl/gofer/handle_unsafe.go | 66 ++ pkg/sentry/fsimpl/gofer/p9file.go | 219 +++++ pkg/sentry/fsimpl/gofer/pagemath.go | 31 + pkg/sentry/fsimpl/gofer/regular_file.go | 860 ++++++++++++++++ pkg/sentry/fsimpl/gofer/special_file.go | 159 +++ pkg/sentry/fsimpl/gofer/symlink.go | 47 + pkg/sentry/fsimpl/gofer/time.go | 75 ++ pkg/sentry/fsimpl/tmpfs/filesystem.go | 2 +- pkg/sentry/socket/hostinet/socket.go | 23 +- 18 files changed, 4103 insertions(+), 29 deletions(-) create mode 100644 pkg/sentry/fsimpl/gofer/BUILD create mode 100644 pkg/sentry/fsimpl/gofer/directory.go create mode 100644 pkg/sentry/fsimpl/gofer/filesystem.go create mode 100644 pkg/sentry/fsimpl/gofer/gofer.go create mode 100644 pkg/sentry/fsimpl/gofer/handle.go create mode 100644 pkg/sentry/fsimpl/gofer/handle_unsafe.go create mode 100644 pkg/sentry/fsimpl/gofer/p9file.go create mode 100644 pkg/sentry/fsimpl/gofer/pagemath.go create mode 100644 pkg/sentry/fsimpl/gofer/regular_file.go create mode 100644 pkg/sentry/fsimpl/gofer/special_file.go create mode 100644 pkg/sentry/fsimpl/gofer/symlink.go create mode 100644 pkg/sentry/fsimpl/gofer/time.go diff --git a/pkg/safemem/seq_unsafe.go b/pkg/safemem/seq_unsafe.go index 354a95dde..dcdfc9600 100644 --- a/pkg/safemem/seq_unsafe.go +++ b/pkg/safemem/seq_unsafe.go @@ -18,6 +18,7 @@ import ( "bytes" "fmt" "reflect" + "syscall" "unsafe" ) @@ -297,3 +298,19 @@ func ZeroSeq(dsts BlockSeq) (uint64, error) { } return done, nil } + +// IovecsFromBlockSeq returns a []syscall.Iovec representing seq. +func IovecsFromBlockSeq(bs BlockSeq) []syscall.Iovec { + iovs := make([]syscall.Iovec, 0, bs.NumBlocks()) + for ; !bs.IsEmpty(); bs = bs.Tail() { + b := bs.Head() + iovs = append(iovs, syscall.Iovec{ + Base: &b.ToSlice()[0], + Len: uint64(b.Len()), + }) + // We don't need to care about b.NeedSafecopy(), because the host + // kernel will handle such address ranges just fine (by returning + // EFAULT). + } + return iovs +} diff --git a/pkg/sentry/fs/fsutil/BUILD b/pkg/sentry/fs/fsutil/BUILD index 4ab2a384f..789369220 100644 --- a/pkg/sentry/fs/fsutil/BUILD +++ b/pkg/sentry/fs/fsutil/BUILD @@ -28,13 +28,13 @@ go_template_instance( "platform": "gvisor.dev/gvisor/pkg/sentry/platform", }, package = "fsutil", - prefix = "frameRef", + prefix = "FrameRef", template = "//pkg/segment:generic_set", types = { "Key": "uint64", "Range": "platform.FileRange", "Value": "uint64", - "Functions": "frameRefSetFunctions", + "Functions": "FrameRefSetFunctions", }, ) diff --git a/pkg/sentry/fs/fsutil/frame_ref_set.go b/pkg/sentry/fs/fsutil/frame_ref_set.go index dd63db32b..6564fd0c6 100644 --- a/pkg/sentry/fs/fsutil/frame_ref_set.go +++ b/pkg/sentry/fs/fsutil/frame_ref_set.go @@ -20,24 +20,25 @@ import ( "gvisor.dev/gvisor/pkg/sentry/platform" ) -type frameRefSetFunctions struct{} +// FrameRefSetFunctions implements segment.Functions for FrameRefSet. +type FrameRefSetFunctions struct{} // MinKey implements segment.Functions.MinKey. -func (frameRefSetFunctions) MinKey() uint64 { +func (FrameRefSetFunctions) MinKey() uint64 { return 0 } // MaxKey implements segment.Functions.MaxKey. -func (frameRefSetFunctions) MaxKey() uint64 { +func (FrameRefSetFunctions) MaxKey() uint64 { return math.MaxUint64 } // ClearValue implements segment.Functions.ClearValue. -func (frameRefSetFunctions) ClearValue(val *uint64) { +func (FrameRefSetFunctions) ClearValue(val *uint64) { } // Merge implements segment.Functions.Merge. -func (frameRefSetFunctions) Merge(_ platform.FileRange, val1 uint64, _ platform.FileRange, val2 uint64) (uint64, bool) { +func (FrameRefSetFunctions) Merge(_ platform.FileRange, val1 uint64, _ platform.FileRange, val2 uint64) (uint64, bool) { if val1 != val2 { return 0, false } @@ -45,6 +46,6 @@ func (frameRefSetFunctions) Merge(_ platform.FileRange, val1 uint64, _ platform. } // Split implements segment.Functions.Split. -func (frameRefSetFunctions) Split(_ platform.FileRange, val uint64, _ uint64) (uint64, uint64) { +func (FrameRefSetFunctions) Split(_ platform.FileRange, val uint64, _ uint64) (uint64, uint64) { return val, val } diff --git a/pkg/sentry/fs/fsutil/inode_cached.go b/pkg/sentry/fs/fsutil/inode_cached.go index 573b8586e..800c8b4e1 100644 --- a/pkg/sentry/fs/fsutil/inode_cached.go +++ b/pkg/sentry/fs/fsutil/inode_cached.go @@ -111,7 +111,7 @@ type CachingInodeOperations struct { // refs tracks active references to data in the cache. // // refs is protected by dataMu. - refs frameRefSet + refs FrameRefSet } // CachingInodeOperationsOptions configures a CachingInodeOperations. diff --git a/pkg/sentry/fsimpl/gofer/BUILD b/pkg/sentry/fsimpl/gofer/BUILD new file mode 100644 index 000000000..4ba76a1e8 --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/BUILD @@ -0,0 +1,55 @@ +load("//tools:defs.bzl", "go_library") +load("//tools/go_generics:defs.bzl", "go_template_instance") + +licenses(["notice"]) + +go_template_instance( + name = "dentry_list", + out = "dentry_list.go", + package = "gofer", + prefix = "dentry", + template = "//pkg/ilist:generic_list", + types = { + "Element": "*dentry", + "Linker": "*dentry", + }, +) + +go_library( + name = "gofer", + srcs = [ + "dentry_list.go", + "directory.go", + "filesystem.go", + "gofer.go", + "handle.go", + "handle_unsafe.go", + "p9file.go", + "pagemath.go", + "regular_file.go", + "special_file.go", + "symlink.go", + "time.go", + ], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/abi/linux", + "//pkg/context", + "//pkg/fd", + "//pkg/fspath", + "//pkg/log", + "//pkg/p9", + "//pkg/safemem", + "//pkg/sentry/fs/fsutil", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/kernel/time", + "//pkg/sentry/memmap", + "//pkg/sentry/pgalloc", + "//pkg/sentry/platform", + "//pkg/sentry/usage", + "//pkg/sentry/vfs", + "//pkg/syserror", + "//pkg/unet", + "//pkg/usermem", + ], +) diff --git a/pkg/sentry/fsimpl/gofer/directory.go b/pkg/sentry/fsimpl/gofer/directory.go new file mode 100644 index 000000000..baa2cdd8e --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/directory.go @@ -0,0 +1,190 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "sync" + "sync/atomic" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/p9" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +func (d *dentry) isDir() bool { + return d.fileType() == linux.S_IFDIR +} + +// Preconditions: d.dirMu must be locked. d.isDir(). fs.opts.interop != +// InteropModeShared. +func (d *dentry) cacheNegativeChildLocked(name string) { + if d.negativeChildren == nil { + d.negativeChildren = make(map[string]struct{}) + } + d.negativeChildren[name] = struct{}{} +} + +type directoryFD struct { + fileDescription + vfs.DirectoryFileDescriptionDefaultImpl + + mu sync.Mutex + off int64 + dirents []vfs.Dirent +} + +// Release implements vfs.FileDescriptionImpl.Release. +func (fd *directoryFD) Release() { +} + +// IterDirents implements vfs.FileDescriptionImpl.IterDirents. +func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { + fd.mu.Lock() + defer fd.mu.Unlock() + + if fd.dirents == nil { + ds, err := fd.dentry().getDirents(ctx) + if err != nil { + return err + } + fd.dirents = ds + } + + for fd.off < int64(len(fd.dirents)) { + if !cb.Handle(fd.dirents[fd.off]) { + return nil + } + fd.off++ + } + return nil +} + +// Preconditions: d.isDir(). There exists at least one directoryFD representing d. +func (d *dentry) getDirents(ctx context.Context) ([]vfs.Dirent, error) { + // 9P2000.L's readdir does not specify behavior in the presence of + // concurrent mutation of an iterated directory, so implementations may + // duplicate or omit entries in this case, which violates POSIX semantics. + // Thus we read all directory entries while holding d.dirMu to exclude + // directory mutations. (Note that it is impossible for the client to + // exclude concurrent mutation from other remote filesystem users. Since + // there is no way to detect if the server has incorrectly omitted + // directory entries, we simply assume that the server is well-behaved + // under InteropModeShared.) This is inconsistent with Linux (which appears + // to assume that directory fids have the correct semantics, and translates + // struct file_operations::readdir calls directly to readdir RPCs), but is + // consistent with VFS1. + + d.fs.renameMu.RLock() + defer d.fs.renameMu.RUnlock() + d.dirMu.Lock() + defer d.dirMu.Unlock() + if d.dirents != nil { + return d.dirents, nil + } + + // It's not clear if 9P2000.L's readdir is expected to return "." and "..", + // so we generate them here. + parent := d.vfsd.ParentOrSelf().Impl().(*dentry) + dirents := []vfs.Dirent{ + { + Name: ".", + Type: linux.DT_DIR, + Ino: d.ino, + NextOff: 1, + }, + { + Name: "..", + Type: uint8(atomic.LoadUint32(&parent.mode) >> 12), + Ino: parent.ino, + NextOff: 2, + }, + } + off := uint64(0) + const count = 64 * 1024 // for consistency with the vfs1 client + d.handleMu.RLock() + defer d.handleMu.RUnlock() + if !d.handleReadable { + // This should not be possible because a readable handle should have + // been opened when the calling directoryFD was opened. + panic("gofer.dentry.getDirents called without a readable handle") + } + for { + p9ds, err := d.handle.file.readdir(ctx, off, count) + if err != nil { + return nil, err + } + if len(p9ds) == 0 { + // Cache dirents for future directoryFDs if permitted. + if d.fs.opts.interop != InteropModeShared { + d.dirents = dirents + } + return dirents, nil + } + for _, p9d := range p9ds { + if p9d.Name == "." || p9d.Name == ".." { + continue + } + dirent := vfs.Dirent{ + Name: p9d.Name, + Ino: p9d.QID.Path, + NextOff: int64(len(dirents) + 1), + } + // p9 does not expose 9P2000.U's DMDEVICE, DMNAMEDPIPE, or + // DMSOCKET. + switch p9d.Type { + case p9.TypeSymlink: + dirent.Type = linux.DT_LNK + case p9.TypeDir: + dirent.Type = linux.DT_DIR + default: + dirent.Type = linux.DT_REG + } + dirents = append(dirents, dirent) + } + off = p9ds[len(p9ds)-1].Offset + } +} + +// Seek implements vfs.FileDescriptionImpl.Seek. +func (fd *directoryFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + fd.mu.Lock() + defer fd.mu.Unlock() + + switch whence { + case linux.SEEK_SET: + if offset < 0 { + return 0, syserror.EINVAL + } + if offset == 0 { + // Ensure that the next call to fd.IterDirents() calls + // fd.dentry().getDirents(). + fd.dirents = nil + } + fd.off = offset + return fd.off, nil + case linux.SEEK_CUR: + offset += fd.off + if offset < 0 { + return 0, syserror.EINVAL + } + // Don't clear fd.dirents in this case, even if offset == 0. + fd.off = offset + return fd.off, nil + default: + return 0, syserror.EINVAL + } +} diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go new file mode 100644 index 000000000..8eb61debf --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -0,0 +1,1087 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "sync" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/p9" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// Sync implements vfs.FilesystemImpl.Sync. +func (fs *filesystem) Sync(ctx context.Context) error { + // Snapshot current dentries and special files. + fs.syncMu.Lock() + ds := make([]*dentry, 0, len(fs.dentries)) + for d := range fs.dentries { + ds = append(ds, d) + } + sffds := make([]*specialFileFD, 0, len(fs.specialFileFDs)) + for sffd := range fs.specialFileFDs { + sffds = append(sffds, sffd) + } + fs.syncMu.Unlock() + + // Return the first error we encounter, but sync everything we can + // regardless. + var retErr error + + // Sync regular files. + for _, d := range ds { + if !d.TryIncRef() { + continue + } + err := d.syncSharedHandle(ctx) + d.DecRef() + if err != nil && retErr == nil { + retErr = err + } + } + + // Sync special files, which may be writable but do not use dentry shared + // handles (so they won't be synced by the above). + for _, sffd := range sffds { + if !sffd.vfsfd.TryIncRef() { + continue + } + err := sffd.Sync(ctx) + sffd.vfsfd.DecRef() + if err != nil && retErr == nil { + retErr = err + } + } + + return retErr +} + +// maxFilenameLen is the maximum length of a filename. This is dictated by 9P's +// encoding of strings, which uses 2 bytes for the length prefix. +const maxFilenameLen = (1 << 16) - 1 + +// dentrySlicePool is a pool of *[]*dentry used to store dentries for which +// dentry.checkCachingLocked() must be called. The pool holds pointers to +// slices because Go lacks generics, so sync.Pool operates on interface{}, so +// every call to (what should be) sync.Pool<[]*dentry>.Put() allocates a copy +// of the slice header on the heap. +var dentrySlicePool = sync.Pool{ + New: func() interface{} { + ds := make([]*dentry, 0, 4) // arbitrary non-zero initial capacity + return &ds + }, +} + +func appendDentry(ds *[]*dentry, d *dentry) *[]*dentry { + if ds == nil { + ds = dentrySlicePool.Get().(*[]*dentry) + } + *ds = append(*ds, d) + return ds +} + +// Preconditions: ds != nil. +func putDentrySlice(ds *[]*dentry) { + // Allow dentries to be GC'd. + for i := range *ds { + (*ds)[i] = nil + } + *ds = (*ds)[:0] + dentrySlicePool.Put(ds) +} + +// stepLocked resolves rp.Component() to an existing file, starting from the +// given directory. +// +// Dentries which may become cached as a result of the traversal are appended +// to *ds. +// +// Preconditions: fs.renameMu must be locked. d.dirMu must be locked. +// !rp.Done(). If fs.opts.interop == InteropModeShared, then d's cached +// metadata must be up to date. +func (fs *filesystem) stepLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry, ds **[]*dentry) (*dentry, error) { + if !d.isDir() { + return nil, syserror.ENOTDIR + } + if err := d.checkPermissions(rp.Credentials(), vfs.MayExec, true); err != nil { + return nil, err + } +afterSymlink: + name := rp.Component() + if name == "." { + rp.Advance() + return d, nil + } + if name == ".." { + parentVFSD, err := rp.ResolveParent(&d.vfsd) + if err != nil { + return nil, err + } + parent := parentVFSD.Impl().(*dentry) + if fs.opts.interop == InteropModeShared { + // We must assume that parentVFSD is correct, because if d has been + // moved elsewhere in the remote filesystem so that its parent has + // changed, we have no way of determining its new parent's location + // in the filesystem. Get updated metadata for parentVFSD. + _, attrMask, attr, err := parent.file.getAttr(ctx, dentryAttrMask()) + if err != nil { + return nil, err + } + parent.updateFromP9Attrs(attrMask, &attr) + } + rp.Advance() + return parent, nil + } + childVFSD, err := rp.ResolveChild(&d.vfsd, name) + if err != nil { + return nil, err + } + // FIXME(jamieliu): Linux performs revalidation before mount lookup + // (fs/namei.c:lookup_fast() => __d_lookup_rcu(), d_revalidate(), + // __follow_mount_rcu()). + child, err := fs.revalidateChildLocked(ctx, rp.VirtualFilesystem(), d, name, childVFSD, ds) + if err != nil { + return nil, err + } + if child == nil { + return nil, syserror.ENOENT + } + if child.isSymlink() && rp.ShouldFollowSymlink() { + target, err := child.readlink(ctx, rp.Mount()) + if err != nil { + return nil, err + } + if err := rp.HandleSymlink(target); err != nil { + return nil, err + } + goto afterSymlink // don't check the current directory again + } + rp.Advance() + return child, nil +} + +// revalidateChildLocked must be called after a call to parent.vfsd.Child(name) +// or vfs.ResolvingPath.ResolveChild(name) returns childVFSD (which may be +// nil) to verify that the returned child (or lack thereof) is correct. If no file +// exists at name, revalidateChildLocked returns (nil, nil). +// +// Preconditions: fs.renameMu must be locked. parent.dirMu must be locked. +// parent.isDir(). name is not "." or "..". +// +// Postconditions: If revalidateChildLocked returns a non-nil dentry, its +// cached metadata is up to date. +func (fs *filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.VirtualFilesystem, parent *dentry, name string, childVFSD *vfs.Dentry, ds **[]*dentry) (*dentry, error) { + if childVFSD != nil && fs.opts.interop != InteropModeShared { + // We have a cached dentry that is assumed to be correct. + return childVFSD.Impl().(*dentry), nil + } + // We either don't have a cached dentry or need to verify that it's still + // correct, either of which requires a remote lookup. Check if this name is + // valid before performing the lookup. + if len(name) > maxFilenameLen { + return nil, syserror.ENAMETOOLONG + } + // Check if we've already cached this lookup with a negative result. + if _, ok := parent.negativeChildren[name]; ok { + return nil, nil + } + // Perform the remote lookup. + qid, file, attrMask, attr, err := parent.file.walkGetAttrOne(ctx, name) + if err != nil && err != syserror.ENOENT { + return nil, err + } + if childVFSD != nil { + child := childVFSD.Impl().(*dentry) + if !file.isNil() && qid.Path == child.ino { + // The file at this path hasn't changed. Just update cached + // metadata. + file.close(ctx) + child.updateFromP9Attrs(attrMask, &attr) + return child, nil + } + // The file at this path has changed or no longer exists. Remove + // the stale dentry from the tree, and re-evaluate its caching + // status (i.e. if it has 0 references, drop it). + vfsObj.ForceDeleteDentry(childVFSD) + *ds = appendDentry(*ds, child) + childVFSD = nil + } + if file.isNil() { + // No file exists at this path now. Cache the negative lookup if + // allowed. + if fs.opts.interop != InteropModeShared { + parent.cacheNegativeChildLocked(name) + } + return nil, nil + } + // Create a new dentry representing the file. + child, err := fs.newDentry(ctx, file, qid, attrMask, &attr) + if err != nil { + file.close(ctx) + return nil, err + } + parent.IncRef() // reference held by child on its parent + parent.vfsd.InsertChild(&child.vfsd, name) + // For now, child has 0 references, so our caller should call + // child.checkCachingLocked(). + *ds = appendDentry(*ds, child) + return child, nil +} + +// walkParentDirLocked resolves all but the last path component of rp to an +// existing directory, starting from the given directory (which is usually +// rp.Start().Impl().(*dentry)). It does not check that the returned directory +// is searchable by the provider of rp. +// +// Preconditions: fs.renameMu must be locked. !rp.Done(). If fs.opts.interop == +// InteropModeShared, then d's cached metadata must be up to date. +func (fs *filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry, ds **[]*dentry) (*dentry, error) { + for !rp.Final() { + d.dirMu.Lock() + next, err := fs.stepLocked(ctx, rp, d, ds) + d.dirMu.Unlock() + if err != nil { + return nil, err + } + d = next + } + if !d.isDir() { + return nil, syserror.ENOTDIR + } + return d, nil +} + +// resolveLocked resolves rp to an existing file. +// +// Preconditions: fs.renameMu must be locked. +func (fs *filesystem) resolveLocked(ctx context.Context, rp *vfs.ResolvingPath, ds **[]*dentry) (*dentry, error) { + d := rp.Start().Impl().(*dentry) + if fs.opts.interop == InteropModeShared { + // Get updated metadata for rp.Start() as required by fs.stepLocked(). + if err := d.updateFromGetattr(ctx); err != nil { + return nil, err + } + } + for !rp.Done() { + d.dirMu.Lock() + next, err := fs.stepLocked(ctx, rp, d, ds) + d.dirMu.Unlock() + if err != nil { + return nil, err + } + d = next + } + if rp.MustBeDir() && !d.isDir() { + return nil, syserror.ENOTDIR + } + return d, nil +} + +// doCreateAt checks that creating a file at rp is permitted, then invokes +// create to do so. +// +// Preconditions: !rp.Done(). For the final path component in rp, +// !rp.ShouldFollowSymlink(). +func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool, create func(parent *dentry, name string) error) error { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + start := rp.Start().Impl().(*dentry) + if fs.opts.interop == InteropModeShared { + // Get updated metadata for start as required by + // fs.walkParentDirLocked(). + if err := start.updateFromGetattr(ctx); err != nil { + return err + } + } + parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds) + if err != nil { + return err + } + if err := parent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true); err != nil { + return err + } + if parent.isDeleted() { + return syserror.ENOENT + } + name := rp.Component() + if name == "." || name == ".." { + return syserror.EEXIST + } + if len(name) > maxFilenameLen { + return syserror.ENAMETOOLONG + } + if !dir && rp.MustBeDir() { + return syserror.ENOENT + } + mnt := rp.Mount() + if err := mnt.CheckBeginWrite(); err != nil { + return err + } + defer mnt.EndWrite() + parent.dirMu.Lock() + defer parent.dirMu.Unlock() + if fs.opts.interop == InteropModeShared { + // The existence of a dentry at name would be inconclusive because the + // file it represents may have been deleted from the remote filesystem, + // so we would need to make an RPC to revalidate the dentry. Just + // attempt the file creation RPC instead. If a file does exist, the RPC + // will fail with EEXIST like we would have. If the RPC succeeds, and a + // stale dentry exists, the dentry will fail revalidation next time + // it's used. + return create(parent, name) + } + if parent.vfsd.Child(name) != nil { + return syserror.EEXIST + } + // No cached dentry exists; however, there might still be an existing file + // at name. As above, we attempt the file creation RPC anyway. + if err := create(parent, name); err != nil { + return err + } + parent.touchCMtime(ctx) + delete(parent.negativeChildren, name) + parent.dirents = nil + return nil +} + +// Preconditions: !rp.Done(). +func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool) error { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + start := rp.Start().Impl().(*dentry) + if fs.opts.interop == InteropModeShared { + // Get updated metadata for start as required by + // fs.walkParentDirLocked(). + if err := start.updateFromGetattr(ctx); err != nil { + return err + } + } + parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds) + if err != nil { + return err + } + if err := parent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true); err != nil { + return err + } + if err := rp.Mount().CheckBeginWrite(); err != nil { + return err + } + defer rp.Mount().EndWrite() + + name := rp.Component() + if dir { + if name == "." { + return syserror.EINVAL + } + if name == ".." { + return syserror.ENOTEMPTY + } + } else { + if name == "." || name == ".." { + return syserror.EISDIR + } + } + vfsObj := rp.VirtualFilesystem() + mntns := vfs.MountNamespaceFromContext(ctx) + parent.dirMu.Lock() + defer parent.dirMu.Unlock() + childVFSD := parent.vfsd.Child(name) + var child *dentry + // We only need a dentry representing the file at name if it can be a mount + // point. If childVFSD is nil, then it can't be a mount point. If childVFSD + // is non-nil but stale, the actual file can't be a mount point either; we + // detect this case by just speculatively calling PrepareDeleteDentry and + // only revalidating the dentry if that fails (indicating that the existing + // dentry is a mount point). + if childVFSD != nil { + child = childVFSD.Impl().(*dentry) + if err := vfsObj.PrepareDeleteDentry(mntns, childVFSD); err != nil { + child, err = fs.revalidateChildLocked(ctx, vfsObj, parent, name, childVFSD, &ds) + if err != nil { + return err + } + if child != nil { + childVFSD = &child.vfsd + if err := vfsObj.PrepareDeleteDentry(mntns, childVFSD); err != nil { + return err + } + } else { + childVFSD = nil + } + } + } else if _, ok := parent.negativeChildren[name]; ok { + return syserror.ENOENT + } + flags := uint32(0) + if dir { + if child != nil && !child.isDir() { + return syserror.ENOTDIR + } + flags = linux.AT_REMOVEDIR + } else { + if child != nil && child.isDir() { + return syserror.EISDIR + } + if rp.MustBeDir() { + return syserror.ENOTDIR + } + } + err = parent.file.unlinkAt(ctx, name, flags) + if err != nil { + if childVFSD != nil { + vfsObj.AbortDeleteDentry(childVFSD) + } + return err + } + if fs.opts.interop != InteropModeShared { + parent.touchCMtime(ctx) + parent.cacheNegativeChildLocked(name) + parent.dirents = nil + } + if child != nil { + child.setDeleted() + vfsObj.CommitDeleteDentry(childVFSD) + ds = appendDentry(ds, child) + } + return nil +} + +// renameMuRUnlockAndCheckCaching calls fs.renameMu.RUnlock(), then calls +// dentry.checkCachingLocked on all dentries in *ds with fs.renameMu locked for +// writing. +// +// ds is a pointer-to-pointer since defer evaluates its arguments immediately, +// but dentry slices are allocated lazily, and it's much easier to say "defer +// fs.renameMuRUnlockAndCheckCaching(&ds)" than "defer func() { +// fs.renameMuRUnlockAndCheckCaching(ds) }()" to work around this. +func (fs *filesystem) renameMuRUnlockAndCheckCaching(ds **[]*dentry) { + fs.renameMu.RUnlock() + if *ds == nil { + return + } + if len(**ds) != 0 { + fs.renameMu.Lock() + for _, d := range **ds { + d.checkCachingLocked() + } + fs.renameMu.Unlock() + } + putDentrySlice(*ds) +} + +func (fs *filesystem) renameMuUnlockAndCheckCaching(ds **[]*dentry) { + if *ds == nil { + fs.renameMu.Unlock() + return + } + for _, d := range **ds { + d.checkCachingLocked() + } + fs.renameMu.Unlock() + putDentrySlice(*ds) +} + +// GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. +func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return nil, err + } + if opts.CheckSearchable { + if !d.isDir() { + return nil, syserror.ENOTDIR + } + if err := d.checkPermissions(rp.Credentials(), vfs.MayExec, true); err != nil { + return nil, err + } + } + d.IncRef() + return &d.vfsd, nil +} + +// GetParentDentryAt implements vfs.FilesystemImpl.GetParentDentryAt. +func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + start := rp.Start().Impl().(*dentry) + if fs.opts.interop == InteropModeShared { + // Get updated metadata for start as required by + // fs.walkParentDirLocked(). + if err := start.updateFromGetattr(ctx); err != nil { + return nil, err + } + } + d, err := fs.walkParentDirLocked(ctx, rp, start, &ds) + if err != nil { + return nil, err + } + d.IncRef() + return &d.vfsd, nil +} + +// LinkAt implements vfs.FilesystemImpl.LinkAt. +func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry) error { + return fs.doCreateAt(ctx, rp, false /* dir */, func(parent *dentry, childName string) error { + if rp.Mount() != vd.Mount() { + return syserror.EXDEV + } + // 9P2000.L supports hard links, but we don't. + return syserror.EPERM + }) +} + +// MkdirAt implements vfs.FilesystemImpl.MkdirAt. +func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { + return fs.doCreateAt(ctx, rp, true /* dir */, func(parent *dentry, name string) error { + creds := rp.Credentials() + _, err := parent.file.mkdir(ctx, name, (p9.FileMode)(opts.Mode), (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) + return err + }) +} + +// MknodAt implements vfs.FilesystemImpl.MknodAt. +func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error { + return fs.doCreateAt(ctx, rp, false /* dir */, func(parent *dentry, name string) error { + creds := rp.Credentials() + _, err := parent.file.mknod(ctx, name, (p9.FileMode)(opts.Mode), opts.DevMajor, opts.DevMinor, (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) + return err + }) +} + +// OpenAt implements vfs.FilesystemImpl.OpenAt. +func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + // Reject O_TMPFILE, which is not supported; supporting it correctly in the + // presence of other remote filesystem users requires remote filesystem + // support, and it isn't clear that there's any way to implement this in + // 9P. + if opts.Flags&linux.O_TMPFILE != 0 { + return nil, syserror.EOPNOTSUPP + } + mayCreate := opts.Flags&linux.O_CREAT != 0 + mustCreate := opts.Flags&(linux.O_CREAT|linux.O_EXCL) == (linux.O_CREAT | linux.O_EXCL) + + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + + start := rp.Start().Impl().(*dentry) + if fs.opts.interop == InteropModeShared { + // Get updated metadata for start as required by fs.stepLocked(). + if err := start.updateFromGetattr(ctx); err != nil { + return nil, err + } + } + if rp.Done() { + return start.openLocked(ctx, rp, opts.Flags) + } + +afterTrailingSymlink: + parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds) + if err != nil { + return nil, err + } + // Check for search permission in the parent directory. + if err := parent.checkPermissions(rp.Credentials(), vfs.MayExec, true); err != nil { + return nil, err + } + // Determine whether or not we need to create a file. + parent.dirMu.Lock() + child, err := fs.stepLocked(ctx, rp, parent, &ds) + if err == syserror.ENOENT && mayCreate { + fd, err := parent.createAndOpenChildLocked(ctx, rp, &opts) + parent.dirMu.Unlock() + return fd, err + } + if err != nil { + parent.dirMu.Unlock() + return nil, err + } + // Open existing child or follow symlink. + parent.dirMu.Unlock() + if mustCreate { + return nil, syserror.EEXIST + } + if child.isSymlink() && rp.ShouldFollowSymlink() { + target, err := child.readlink(ctx, rp.Mount()) + if err != nil { + return nil, err + } + if err := rp.HandleSymlink(target); err != nil { + return nil, err + } + start = parent + goto afterTrailingSymlink + } + return child.openLocked(ctx, rp, opts.Flags) +} + +// Preconditions: fs.renameMu must be locked. +func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, flags uint32) (*vfs.FileDescription, error) { + ats := vfs.AccessTypesForOpenFlags(flags) + if err := d.checkPermissions(rp.Credentials(), ats, d.isDir()); err != nil { + return nil, err + } + mnt := rp.Mount() + filetype := d.fileType() + switch { + case filetype == linux.S_IFREG && !d.fs.opts.regularFilesUseSpecialFileFD: + if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, flags&linux.O_TRUNC != 0); err != nil { + return nil, err + } + fd := ®ularFileFD{} + if err := fd.vfsfd.Init(fd, flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{ + AllowDirectIO: true, + }); err != nil { + return nil, err + } + return &fd.vfsfd, nil + case filetype == linux.S_IFDIR: + // Can't open directories with O_CREAT. + if flags&linux.O_CREAT != 0 { + return nil, syserror.EISDIR + } + // Can't open directories writably. + if ats&vfs.MayWrite != 0 { + return nil, syserror.EISDIR + } + if flags&linux.O_DIRECT != 0 { + return nil, syserror.EINVAL + } + if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, false /* write */, false /* trunc */); err != nil { + return nil, err + } + fd := &directoryFD{} + if err := fd.vfsfd.Init(fd, flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { + return nil, err + } + return &fd.vfsfd, nil + case filetype == linux.S_IFLNK: + // Can't open symlinks without O_PATH (which is unimplemented). + return nil, syserror.ELOOP + default: + if flags&linux.O_DIRECT != 0 { + return nil, syserror.EINVAL + } + h, err := openHandle(ctx, d.file, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, flags&linux.O_TRUNC != 0) + if err != nil { + return nil, err + } + fd := &specialFileFD{ + handle: h, + } + if err := fd.vfsfd.Init(fd, flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { + h.close(ctx) + return nil, err + } + return &fd.vfsfd, nil + } +} + +// Preconditions: d.fs.renameMu must be locked. d.dirMu must be locked. +func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.OpenOptions) (*vfs.FileDescription, error) { + if err := d.checkPermissions(rp.Credentials(), vfs.MayWrite, true); err != nil { + return nil, err + } + if d.isDeleted() { + return nil, syserror.ENOENT + } + mnt := rp.Mount() + if err := mnt.CheckBeginWrite(); err != nil { + return nil, err + } + defer mnt.EndWrite() + + // 9P2000.L's lcreate takes a fid representing the parent directory, and + // converts it into an open fid representing the created file, so we need + // to duplicate the directory fid first. + _, dirfile, err := d.file.walk(ctx, nil) + if err != nil { + return nil, err + } + creds := rp.Credentials() + name := rp.Component() + fdobj, openFile, createQID, _, err := dirfile.create(ctx, name, (p9.OpenFlags)(opts.Flags), (p9.FileMode)(opts.Mode), (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) + if err != nil { + dirfile.close(ctx) + return nil, err + } + // Then we need to walk to the file we just created to get a non-open fid + // representing it, and to get its metadata. This must use d.file since, as + // explained above, dirfile was invalidated by dirfile.Create(). + walkQID, nonOpenFile, attrMask, attr, err := d.file.walkGetAttrOne(ctx, name) + if err != nil { + openFile.close(ctx) + if fdobj != nil { + fdobj.Close() + } + return nil, err + } + // Sanity-check that we walked to the file we created. + if createQID.Path != walkQID.Path { + // Probably due to concurrent remote filesystem mutation? + ctx.Warningf("gofer.dentry.createAndOpenChildLocked: created file has QID %v before walk, QID %v after (interop=%v)", createQID, walkQID, d.fs.opts.interop) + nonOpenFile.close(ctx) + openFile.close(ctx) + if fdobj != nil { + fdobj.Close() + } + return nil, syserror.EAGAIN + } + + // Construct the new dentry. + child, err := d.fs.newDentry(ctx, nonOpenFile, createQID, attrMask, &attr) + if err != nil { + nonOpenFile.close(ctx) + openFile.close(ctx) + if fdobj != nil { + fdobj.Close() + } + return nil, err + } + // Incorporate the fid that was opened by lcreate. + useRegularFileFD := child.fileType() == linux.S_IFREG && !d.fs.opts.regularFilesUseSpecialFileFD + if useRegularFileFD { + child.handleMu.Lock() + child.handle.file = openFile + if fdobj != nil { + child.handle.fd = int32(fdobj.Release()) + } + child.handleReadable = vfs.MayReadFileWithOpenFlags(opts.Flags) + child.handleWritable = vfs.MayWriteFileWithOpenFlags(opts.Flags) + child.handleMu.Unlock() + } + // Take a reference on the new dentry to be held by the new file + // description. (This reference also means that the new dentry is not + // eligible for caching yet, so we don't need to append to a dentry slice.) + child.refs = 1 + // Insert the dentry into the tree. + d.IncRef() // reference held by child on its parent d + d.vfsd.InsertChild(&child.vfsd, name) + if d.fs.opts.interop != InteropModeShared { + d.touchCMtime(ctx) + delete(d.negativeChildren, name) + d.dirents = nil + } + + // Finally, construct a file description representing the created file. + var childVFSFD *vfs.FileDescription + mnt.IncRef() + if useRegularFileFD { + fd := ®ularFileFD{} + if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &child.vfsd, &vfs.FileDescriptionOptions{ + AllowDirectIO: true, + }); err != nil { + return nil, err + } + childVFSFD = &fd.vfsfd + } else { + fd := &specialFileFD{ + handle: handle{ + file: openFile, + fd: -1, + }, + } + if fdobj != nil { + fd.handle.fd = int32(fdobj.Release()) + } + if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &child.vfsd, &vfs.FileDescriptionOptions{}); err != nil { + fd.handle.close(ctx) + return nil, err + } + childVFSFD = &fd.vfsfd + } + return childVFSFD, nil +} + +// ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt. +func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return "", err + } + if !d.isSymlink() { + return "", syserror.EINVAL + } + return d.readlink(ctx, rp.Mount()) +} + +// RenameAt implements vfs.FilesystemImpl.RenameAt. +func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldParentVD vfs.VirtualDentry, oldName string, opts vfs.RenameOptions) error { + if opts.Flags != 0 { + // Requires 9P support. + return syserror.EINVAL + } + + var ds *[]*dentry + fs.renameMu.Lock() + defer fs.renameMuUnlockAndCheckCaching(&ds) + newParent, err := fs.walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry), &ds) + if err != nil { + return err + } + newName := rp.Component() + if newName == "." || newName == ".." { + return syserror.EBUSY + } + mnt := rp.Mount() + if mnt != oldParentVD.Mount() { + return syserror.EXDEV + } + if err := mnt.CheckBeginWrite(); err != nil { + return err + } + defer mnt.EndWrite() + + oldParent := oldParentVD.Dentry().Impl().(*dentry) + if fs.opts.interop == InteropModeShared { + if err := oldParent.updateFromGetattr(ctx); err != nil { + return err + } + } + if err := oldParent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true); err != nil { + return err + } + vfsObj := rp.VirtualFilesystem() + // We need a dentry representing the renamed file since, if it's a + // directory, we need to check for write permission on it. + oldParent.dirMu.Lock() + defer oldParent.dirMu.Unlock() + renamed, err := fs.revalidateChildLocked(ctx, vfsObj, oldParent, oldName, oldParent.vfsd.Child(oldName), &ds) + if err != nil { + return err + } + if renamed == nil { + return syserror.ENOENT + } + if renamed.isDir() { + if renamed == newParent || renamed.vfsd.IsAncestorOf(&newParent.vfsd) { + return syserror.EINVAL + } + if oldParent != newParent { + if err := renamed.checkPermissions(rp.Credentials(), vfs.MayWrite, true); err != nil { + return err + } + } + } else { + if opts.MustBeDir || rp.MustBeDir() { + return syserror.ENOTDIR + } + } + + if oldParent != newParent { + if err := newParent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true); err != nil { + return err + } + newParent.dirMu.Lock() + defer newParent.dirMu.Unlock() + } + if newParent.isDeleted() { + return syserror.ENOENT + } + replacedVFSD := newParent.vfsd.Child(newName) + var replaced *dentry + // This is similar to unlinkAt, except: + // + // - We revalidate the replaced dentry unconditionally for simplicity. + // + // - If rp.MustBeDir(), then we need a dentry representing the replaced + // file regardless to confirm that it's a directory. + if replacedVFSD != nil || rp.MustBeDir() { + replaced, err = fs.revalidateChildLocked(ctx, vfsObj, newParent, newName, replacedVFSD, &ds) + if err != nil { + return err + } + if replaced != nil { + if replaced.isDir() { + if !renamed.isDir() { + return syserror.EISDIR + } + } else { + if rp.MustBeDir() || renamed.isDir() { + return syserror.ENOTDIR + } + } + replacedVFSD = &replaced.vfsd + } else { + replacedVFSD = nil + } + } + + if oldParent == newParent && oldName == newName { + return nil + } + if err := vfsObj.PrepareRenameDentry(vfs.MountNamespaceFromContext(ctx), &renamed.vfsd, replacedVFSD); err != nil { + return err + } + if err := renamed.file.rename(ctx, newParent.file, newName); err != nil { + vfsObj.AbortRenameDentry(&renamed.vfsd, replacedVFSD) + return err + } + if fs.opts.interop != InteropModeShared { + oldParent.cacheNegativeChildLocked(oldName) + oldParent.dirents = nil + delete(newParent.negativeChildren, newName) + newParent.dirents = nil + } + vfsObj.CommitRenameReplaceDentry(&renamed.vfsd, &newParent.vfsd, newName, replacedVFSD) + return nil +} + +// RmdirAt implements vfs.FilesystemImpl.RmdirAt. +func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { + return fs.unlinkAt(ctx, rp, true /* dir */) +} + +// SetStatAt implements vfs.FilesystemImpl.SetStatAt. +func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return err + } + return d.setStat(ctx, rp.Credentials(), &opts.Stat, rp.Mount()) +} + +// StatAt implements vfs.FilesystemImpl.StatAt. +func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return linux.Statx{}, err + } + // Since walking updates metadata for all traversed dentries under + // InteropModeShared, including the returned one, we can return cached + // metadata here regardless of fs.opts.interop. + var stat linux.Statx + d.statTo(&stat) + return stat, nil +} + +// StatFSAt implements vfs.FilesystemImpl.StatFSAt. +func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return linux.Statfs{}, err + } + fsstat, err := d.file.statFS(ctx) + if err != nil { + return linux.Statfs{}, err + } + nameLen := uint64(fsstat.NameLength) + if nameLen > maxFilenameLen { + nameLen = maxFilenameLen + } + return linux.Statfs{ + // This is primarily for distinguishing a gofer file system in + // tests. Testing is important, so instead of defining + // something completely random, use a standard value. + Type: linux.V9FS_MAGIC, + BlockSize: int64(fsstat.BlockSize), + Blocks: fsstat.Blocks, + BlocksFree: fsstat.BlocksFree, + BlocksAvailable: fsstat.BlocksAvailable, + Files: fsstat.Files, + FilesFree: fsstat.FilesFree, + NameLength: nameLen, + }, nil +} + +// SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. +func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, target string) error { + return fs.doCreateAt(ctx, rp, false /* dir */, func(parent *dentry, name string) error { + creds := rp.Credentials() + _, err := parent.file.symlink(ctx, target, name, (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) + return err + }) +} + +// UnlinkAt implements vfs.FilesystemImpl.UnlinkAt. +func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { + return fs.unlinkAt(ctx, rp, false /* dir */) +} + +// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. +func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return nil, err + } + return d.listxattr(ctx) +} + +// GetxattrAt implements vfs.FilesystemImpl.GetxattrAt. +func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) (string, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return "", err + } + return d.getxattr(ctx, name) +} + +// SetxattrAt implements vfs.FilesystemImpl.SetxattrAt. +func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return err + } + return d.setxattr(ctx, &opts) +} + +// RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. +func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return err + } + return d.removexattr(ctx, name) +} + +// PrependPath implements vfs.FilesystemImpl.PrependPath. +func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { + fs.renameMu.RLock() + defer fs.renameMu.RUnlock() + return vfs.GenericPrependPath(vfsroot, vd, b) +} diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go new file mode 100644 index 000000000..d0552bd99 --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -0,0 +1,1147 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package gofer provides a filesystem implementation that is backed by a 9p +// server, interchangably referred to as "gofers" throughout this package. +// +// Lock order: +// regularFileFD/directoryFD.mu +// filesystem.renameMu +// dentry.dirMu +// filesystem.syncMu +// dentry.metadataMu +// *** "memmap.Mappable locks" below this point +// dentry.mapsMu +// *** "memmap.Mappable locks taken by Translate" below this point +// dentry.handleMu +// dentry.dataMu +// +// Locking dentry.dirMu in multiple dentries requires holding +// filesystem.renameMu for writing. +package gofer + +import ( + "fmt" + "strconv" + "sync" + "sync/atomic" + "syscall" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/p9" + "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/memmap" + "gvisor.dev/gvisor/pkg/sentry/pgalloc" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/unet" + "gvisor.dev/gvisor/pkg/usermem" +) + +// FilesystemType implements vfs.FilesystemType. +type FilesystemType struct{} + +// filesystem implements vfs.FilesystemImpl. +type filesystem struct { + vfsfs vfs.Filesystem + + // mfp is used to allocate memory that caches regular file contents. mfp is + // immutable. + mfp pgalloc.MemoryFileProvider + + // Immutable options. + opts filesystemOptions + + // client is the client used by this filesystem. client is immutable. + client *p9.Client + + // uid and gid are the effective KUID and KGID of the filesystem's creator, + // and are used as the owner and group for files that don't specify one. + // uid and gid are immutable. + uid auth.KUID + gid auth.KGID + + // renameMu serves two purposes: + // + // - It synchronizes path resolution with renaming initiated by this + // client. + // + // - It is held by path resolution to ensure that reachable dentries remain + // valid. A dentry is reachable by path resolution if it has a non-zero + // reference count (such that it is usable as vfs.ResolvingPath.Start() or + // is reachable from its children), or if it is a child dentry (such that + // it is reachable from its parent). + renameMu sync.RWMutex + + // cachedDentries contains all dentries with 0 references. (Due to race + // conditions, it may also contain dentries with non-zero references.) + // cachedDentriesLen is the number of dentries in cachedDentries. These + // fields are protected by renameMu. + cachedDentries dentryList + cachedDentriesLen uint64 + + // dentries contains all dentries in this filesystem. specialFileFDs + // contains all open specialFileFDs. These fields are protected by syncMu. + syncMu sync.Mutex + dentries map[*dentry]struct{} + specialFileFDs map[*specialFileFD]struct{} +} + +type filesystemOptions struct { + // "Standard" 9P options. + fd int + aname string + interop InteropMode // derived from the "cache" mount option + msize uint32 + version string + + // maxCachedDentries is the maximum number of dentries with 0 references + // retained by the client. + maxCachedDentries uint64 + + // If forcePageCache is true, host FDs may not be used for application + // memory mappings even if available; instead, the client must perform its + // own caching of regular file pages. This is primarily useful for testing. + forcePageCache bool + + // If limitHostFDTranslation is true, apply maxFillRange() constraints to + // host FD mappings returned by dentry.(memmap.Mappable).Translate(). This + // makes memory accounting behavior more consistent between cases where + // host FDs are / are not available, but may increase the frequency of + // sentry-handled page faults on files for which a host FD is available. + limitHostFDTranslation bool + + // If overlayfsStaleRead is true, O_RDONLY host FDs provided by the remote + // filesystem may not be coherent with writable host FDs opened later, so + // mappings of the former must be replaced by mappings of the latter. This + // is usually only the case when the remote filesystem is an overlayfs + // mount on Linux < 4.19. + overlayfsStaleRead bool + + // If regularFilesUseSpecialFileFD is true, application FDs representing + // regular files will use distinct file handles for each FD, in the same + // way that application FDs representing "special files" such as sockets + // do. Note that this disables client caching and mmap for regular files. + regularFilesUseSpecialFileFD bool +} + +// InteropMode controls the client's interaction with other remote filesystem +// users. +type InteropMode uint32 + +const ( + // InteropModeExclusive is appropriate when the filesystem client is the + // only user of the remote filesystem. + // + // - The client may cache arbitrary filesystem state (file data, metadata, + // filesystem structure, etc.). + // + // - Client changes to filesystem state may be sent to the remote + // filesystem asynchronously, except when server permission checks are + // necessary. + // + // - File timestamps are based on client clocks. This ensures that users of + // the client observe timestamps that are coherent with their own clocks + // and consistent with Linux's semantics. However, since it is not always + // possible for clients to set arbitrary atimes and mtimes, and never + // possible for clients to set arbitrary ctimes, file timestamp changes are + // stored in the client only and never sent to the remote filesystem. + InteropModeExclusive InteropMode = iota + + // InteropModeWritethrough is appropriate when there are read-only users of + // the remote filesystem that expect to observe changes made by the + // filesystem client. + // + // - The client may cache arbitrary filesystem state. + // + // - Client changes to filesystem state must be sent to the remote + // filesystem synchronously. + // + // - File timestamps are based on client clocks. As a corollary, access + // timestamp changes from other remote filesystem users will not be visible + // to the client. + InteropModeWritethrough + + // InteropModeShared is appropriate when there are users of the remote + // filesystem that may mutate its state other than the client. + // + // - The client must verify cached filesystem state before using it. + // + // - Client changes to filesystem state must be sent to the remote + // filesystem synchronously. + // + // - File timestamps are based on server clocks. This is necessary to + // ensure that timestamp changes are synchronized between remote filesystem + // users. + // + // Note that the correctness of InteropModeShared depends on the server + // correctly implementing 9P fids (i.e. each fid immutably represents a + // single filesystem object), even in the presence of remote filesystem + // mutations from other users. If this is violated, the behavior of the + // client is undefined. + InteropModeShared +) + +// GetFilesystem implements vfs.FilesystemType.GetFilesystem. +func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { + mfp := pgalloc.MemoryFileProviderFromContext(ctx) + if mfp == nil { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: context does not provide a pgalloc.MemoryFileProvider") + return nil, nil, syserror.EINVAL + } + + mopts := vfs.GenericParseMountOptions(opts.Data) + var fsopts filesystemOptions + + // Check that the transport is "fd". + trans, ok := mopts["trans"] + if !ok { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: transport must be specified as 'trans=fd'") + return nil, nil, syserror.EINVAL + } + delete(mopts, "trans") + if trans != "fd" { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: unsupported transport: trans=%s", trans) + return nil, nil, syserror.EINVAL + } + + // Check that read and write FDs are provided and identical. + rfdstr, ok := mopts["rfdno"] + if !ok { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: read FD must be specified as 'rfdno=") + return nil, nil, syserror.EINVAL + } + delete(mopts, "rfdno") + rfd, err := strconv.Atoi(rfdstr) + if err != nil { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: invalid read FD: rfdno=%s", rfdstr) + return nil, nil, syserror.EINVAL + } + wfdstr, ok := mopts["wfdno"] + if !ok { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: write FD must be specified as 'wfdno=") + return nil, nil, syserror.EINVAL + } + delete(mopts, "wfdno") + wfd, err := strconv.Atoi(wfdstr) + if err != nil { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: invalid write FD: wfdno=%s", wfdstr) + return nil, nil, syserror.EINVAL + } + if rfd != wfd { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: read FD (%d) and write FD (%d) must be equal", rfd, wfd) + return nil, nil, syserror.EINVAL + } + fsopts.fd = rfd + + // Get the attach name. + fsopts.aname = "/" + if aname, ok := mopts["aname"]; ok { + delete(mopts, "aname") + fsopts.aname = aname + } + + // Parse the cache policy. For historical reasons, this defaults to the + // least generally-applicable option, InteropModeExclusive. + fsopts.interop = InteropModeExclusive + if cache, ok := mopts["cache"]; ok { + delete(mopts, "cache") + switch cache { + case "fscache": + fsopts.interop = InteropModeExclusive + case "fscache_writethrough": + fsopts.interop = InteropModeWritethrough + case "none": + fsopts.regularFilesUseSpecialFileFD = true + fallthrough + case "remote_revalidating": + fsopts.interop = InteropModeShared + default: + ctx.Warningf("gofer.FilesystemType.GetFilesystem: invalid cache policy: cache=%s", cache) + return nil, nil, syserror.EINVAL + } + } + + // Parse the 9P message size. + fsopts.msize = 1024 * 1024 // 1M, tested to give good enough performance up to 64M + if msizestr, ok := mopts["msize"]; ok { + delete(mopts, "msize") + msize, err := strconv.ParseUint(msizestr, 10, 32) + if err != nil { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: invalid message size: msize=%s", msizestr) + return nil, nil, syserror.EINVAL + } + fsopts.msize = uint32(msize) + } + + // Parse the 9P protocol version. + fsopts.version = p9.HighestVersionString() + if version, ok := mopts["version"]; ok { + delete(mopts, "version") + fsopts.version = version + } + + // Parse the dentry cache limit. + fsopts.maxCachedDentries = 1000 + if str, ok := mopts["dentry_cache_limit"]; ok { + delete(mopts, "dentry_cache_limit") + maxCachedDentries, err := strconv.ParseUint(str, 10, 64) + if err != nil { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: invalid dentry cache limit: dentry_cache_limit=%s", str) + return nil, nil, syserror.EINVAL + } + fsopts.maxCachedDentries = maxCachedDentries + } + + // Handle simple flags. + if _, ok := mopts["force_page_cache"]; ok { + delete(mopts, "force_page_cache") + fsopts.forcePageCache = true + } + if _, ok := mopts["limit_host_fd_translation"]; ok { + delete(mopts, "limit_host_fd_translation") + fsopts.limitHostFDTranslation = true + } + if _, ok := mopts["overlayfs_stale_read"]; ok { + delete(mopts, "overlayfs_stale_read") + fsopts.overlayfsStaleRead = true + } + // fsopts.regularFilesUseSpecialFileFD can only be enabled by specifying + // "cache=none". + + // Check for unparsed options. + if len(mopts) != 0 { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: unknown options: %v", mopts) + return nil, nil, syserror.EINVAL + } + + // Establish a connection with the server. + conn, err := unet.NewSocket(fsopts.fd) + if err != nil { + return nil, nil, err + } + + // Perform version negotiation with the server. + ctx.UninterruptibleSleepStart(false) + client, err := p9.NewClient(conn, fsopts.msize, fsopts.version) + ctx.UninterruptibleSleepFinish(false) + if err != nil { + conn.Close() + return nil, nil, err + } + // Ownership of conn has been transferred to client. + + // Perform attach to obtain the filesystem root. + ctx.UninterruptibleSleepStart(false) + attached, err := client.Attach(fsopts.aname) + ctx.UninterruptibleSleepFinish(false) + if err != nil { + client.Close() + return nil, nil, err + } + attachFile := p9file{attached} + qid, attrMask, attr, err := attachFile.getAttr(ctx, dentryAttrMask()) + if err != nil { + attachFile.close(ctx) + client.Close() + return nil, nil, err + } + + // Construct the filesystem object. + fs := &filesystem{ + mfp: mfp, + opts: fsopts, + uid: creds.EffectiveKUID, + gid: creds.EffectiveKGID, + client: client, + dentries: make(map[*dentry]struct{}), + specialFileFDs: make(map[*specialFileFD]struct{}), + } + fs.vfsfs.Init(vfsObj, fs) + + // Construct the root dentry. + root, err := fs.newDentry(ctx, attachFile, qid, attrMask, &attr) + if err != nil { + attachFile.close(ctx) + fs.vfsfs.DecRef() + return nil, nil, err + } + // Set the root's reference count to 2. One reference is returned to the + // caller, and the other is deliberately leaked to prevent the root from + // being "cached" and subsequently evicted. Its resources will still be + // cleaned up by fs.Release(). + root.refs = 2 + + return &fs.vfsfs, &root.vfsd, nil +} + +// Release implements vfs.FilesystemImpl.Release. +func (fs *filesystem) Release() { + ctx := context.Background() + mf := fs.mfp.MemoryFile() + + fs.syncMu.Lock() + for d := range fs.dentries { + d.handleMu.Lock() + d.dataMu.Lock() + if d.handleWritable { + // Write dirty cached data to the remote file. + if err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, fs.mfp.MemoryFile(), d.handle.writeFromBlocksAt); err != nil { + log.Warningf("gofer.filesystem.Release: failed to flush dentry: %v", err) + } + // TODO(jamieliu): Do we need to flushf/fsync d? + } + // Discard cached pages. + d.cache.DropAll(mf) + d.dirty.RemoveAll() + d.dataMu.Unlock() + // Close the host fd if one exists. + if d.handle.fd >= 0 { + syscall.Close(int(d.handle.fd)) + d.handle.fd = -1 + } + d.handleMu.Unlock() + } + // There can't be any specialFileFDs still using fs, since each such + // FileDescription would hold a reference on a Mount holding a reference on + // fs. + fs.syncMu.Unlock() + + // Close the connection to the server. This implicitly clunks all fids. + fs.client.Close() +} + +// dentry implements vfs.DentryImpl. +type dentry struct { + vfsd vfs.Dentry + + // refs is the reference count. Each dentry holds a reference on its + // parent, even if disowned. refs is accessed using atomic memory + // operations. + refs int64 + + // fs is the owning filesystem. fs is immutable. + fs *filesystem + + // We don't support hard links, so each dentry maps 1:1 to an inode. + + // file is the unopened p9.File that backs this dentry. file is immutable. + file p9file + + // If deleted is non-zero, the file represented by this dentry has been + // deleted. deleted is accessed using atomic memory operations. + deleted uint32 + + // If cached is true, dentryEntry links dentry into + // filesystem.cachedDentries. cached and dentryEntry are protected by + // filesystem.renameMu. + cached bool + dentryEntry + + dirMu sync.Mutex + + // If this dentry represents a directory, and InteropModeShared is not in + // effect, negativeChildren is a set of child names in this directory that + // are known not to exist. negativeChildren is protected by dirMu. + negativeChildren map[string]struct{} + + // If this dentry represents a directory, InteropModeShared is not in + // effect, and dirents is not nil, it is a cache of all entries in the + // directory, in the order they were returned by the server. dirents is + // protected by dirMu. + dirents []vfs.Dirent + + // Cached metadata; protected by metadataMu and accessed using atomic + // memory operations unless otherwise specified. + metadataMu sync.Mutex + ino uint64 // immutable + mode uint32 // type is immutable, perms are mutable + uid uint32 // auth.KUID, but stored as raw uint32 for sync/atomic + gid uint32 // auth.KGID, but ... + blockSize uint32 // 0 if unknown + // Timestamps, all nsecs from the Unix epoch. + atime int64 + mtime int64 + ctime int64 + btime int64 + // File size, protected by both metadataMu and dataMu (i.e. both must be + // locked to mutate it). + size uint64 + + mapsMu sync.Mutex + + // If this dentry represents a regular file, mappings tracks mappings of + // the file into memmap.MappingSpaces. mappings is protected by mapsMu. + mappings memmap.MappingSet + + // If this dentry represents a regular file or directory: + // + // - handle is the I/O handle used by all regularFileFDs/directoryFDs + // representing this dentry. + // + // - handleReadable is true if handle is readable. + // + // - handleWritable is true if handle is writable. + // + // Invariants: + // + // - If handleReadable == handleWritable == false, then handle.file == nil + // (i.e. there is no open handle). Conversely, if handleReadable || + // handleWritable == true, then handle.file != nil (i.e. there is an open + // handle). + // + // - handleReadable and handleWritable cannot transition from true to false + // (i.e. handles may not be downgraded). + // + // These fields are protected by handleMu. + handleMu sync.RWMutex + handle handle + handleReadable bool + handleWritable bool + + dataMu sync.RWMutex + + // If this dentry represents a regular file that is client-cached, cache + // maps offsets into the cached file to offsets into + // filesystem.mfp.MemoryFile() that store the file's data. cache is + // protected by dataMu. + cache fsutil.FileRangeSet + + // If this dentry represents a regular file that is client-cached, dirty + // tracks dirty segments in cache. dirty is protected by dataMu. + dirty fsutil.DirtySet + + // pf implements platform.File for mappings of handle.fd. + pf dentryPlatformFile + + // If this dentry represents a symbolic link, InteropModeShared is not in + // effect, and haveTarget is true, target is the symlink target. haveTarget + // and target are protected by dataMu. + haveTarget bool + target string +} + +// dentryAttrMask returns a p9.AttrMask enabling all attributes used by the +// gofer client. +func dentryAttrMask() p9.AttrMask { + return p9.AttrMask{ + Mode: true, + UID: true, + GID: true, + ATime: true, + MTime: true, + CTime: true, + Size: true, + BTime: true, + } +} + +// newDentry creates a new dentry representing the given file. The dentry +// initially has no references, but is not cached; it is the caller's +// responsibility to set the dentry's reference count and/or call +// dentry.checkCachingLocked() as appropriate. +func (fs *filesystem) newDentry(ctx context.Context, file p9file, qid p9.QID, mask p9.AttrMask, attr *p9.Attr) (*dentry, error) { + if !mask.Mode { + ctx.Warningf("can't create gofer.dentry without file type") + return nil, syserror.EIO + } + if attr.Mode.FileType() == p9.ModeRegular && !mask.Size { + ctx.Warningf("can't create regular file gofer.dentry without file size") + return nil, syserror.EIO + } + + d := &dentry{ + fs: fs, + file: file, + ino: qid.Path, + mode: uint32(attr.Mode), + uid: uint32(fs.uid), + gid: uint32(fs.gid), + blockSize: usermem.PageSize, + handle: handle{ + fd: -1, + }, + } + d.pf.dentry = d + if mask.UID { + d.uid = uint32(attr.UID) + } + if mask.GID { + d.gid = uint32(attr.GID) + } + if mask.Size { + d.size = attr.Size + } + if attr.BlockSize != 0 { + d.blockSize = uint32(attr.BlockSize) + } + if mask.ATime { + d.atime = dentryTimestampFromP9(attr.ATimeSeconds, attr.ATimeNanoSeconds) + } + if mask.MTime { + d.mtime = dentryTimestampFromP9(attr.MTimeSeconds, attr.MTimeNanoSeconds) + } + if mask.CTime { + d.ctime = dentryTimestampFromP9(attr.CTimeSeconds, attr.CTimeNanoSeconds) + } + if mask.BTime { + d.btime = dentryTimestampFromP9(attr.BTimeSeconds, attr.BTimeNanoSeconds) + } + d.vfsd.Init(d) + + fs.syncMu.Lock() + fs.dentries[d] = struct{}{} + fs.syncMu.Unlock() + return d, nil +} + +// updateFromP9Attrs is called to update d's metadata after an update from the +// remote filesystem. +func (d *dentry) updateFromP9Attrs(mask p9.AttrMask, attr *p9.Attr) { + d.metadataMu.Lock() + if mask.Mode { + if got, want := uint32(attr.Mode.FileType()), d.fileType(); got != want { + d.metadataMu.Unlock() + panic(fmt.Sprintf("gofer.dentry file type changed from %#o to %#o", want, got)) + } + atomic.StoreUint32(&d.mode, uint32(attr.Mode)) + } + if mask.UID { + atomic.StoreUint32(&d.uid, uint32(attr.UID)) + } + if mask.GID { + atomic.StoreUint32(&d.gid, uint32(attr.GID)) + } + // There is no P9_GETATTR_* bit for I/O block size. + if attr.BlockSize != 0 { + atomic.StoreUint32(&d.blockSize, uint32(attr.BlockSize)) + } + if mask.ATime { + atomic.StoreInt64(&d.atime, dentryTimestampFromP9(attr.ATimeSeconds, attr.ATimeNanoSeconds)) + } + if mask.MTime { + atomic.StoreInt64(&d.mtime, dentryTimestampFromP9(attr.MTimeSeconds, attr.MTimeNanoSeconds)) + } + if mask.CTime { + atomic.StoreInt64(&d.ctime, dentryTimestampFromP9(attr.CTimeSeconds, attr.CTimeNanoSeconds)) + } + if mask.BTime { + atomic.StoreInt64(&d.btime, dentryTimestampFromP9(attr.BTimeSeconds, attr.BTimeNanoSeconds)) + } + if mask.Size { + d.dataMu.Lock() + atomic.StoreUint64(&d.size, attr.Size) + d.dataMu.Unlock() + } + d.metadataMu.Unlock() +} + +func (d *dentry) updateFromGetattr(ctx context.Context) error { + // Use d.handle.file, which represents a 9P fid that has been opened, in + // preference to d.file, which represents a 9P fid that has not. This may + // be significantly more efficient in some implementations. + var ( + file p9file + handleMuRLocked bool + ) + d.handleMu.RLock() + if !d.handle.file.isNil() { + file = d.handle.file + handleMuRLocked = true + } else { + file = d.file + d.handleMu.RUnlock() + } + _, attrMask, attr, err := file.getAttr(ctx, dentryAttrMask()) + if handleMuRLocked { + d.handleMu.RUnlock() + } + if err != nil { + return err + } + d.updateFromP9Attrs(attrMask, &attr) + return nil +} + +func (d *dentry) fileType() uint32 { + return atomic.LoadUint32(&d.mode) & linux.S_IFMT +} + +func (d *dentry) statTo(stat *linux.Statx) { + stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | linux.STATX_UID | linux.STATX_GID | linux.STATX_ATIME | linux.STATX_MTIME | linux.STATX_CTIME | linux.STATX_INO | linux.STATX_SIZE | linux.STATX_BLOCKS | linux.STATX_BTIME + stat.Blksize = atomic.LoadUint32(&d.blockSize) + stat.Nlink = 1 + if d.isDir() { + stat.Nlink = 2 + } + stat.UID = atomic.LoadUint32(&d.uid) + stat.GID = atomic.LoadUint32(&d.gid) + stat.Mode = uint16(atomic.LoadUint32(&d.mode)) + stat.Ino = d.ino + stat.Size = atomic.LoadUint64(&d.size) + // This is consistent with regularFileFD.Seek(), which treats regular files + // as having no holes. + stat.Blocks = (stat.Size + 511) / 512 + stat.Atime = statxTimestampFromDentry(atomic.LoadInt64(&d.atime)) + stat.Btime = statxTimestampFromDentry(atomic.LoadInt64(&d.btime)) + stat.Ctime = statxTimestampFromDentry(atomic.LoadInt64(&d.ctime)) + stat.Mtime = statxTimestampFromDentry(atomic.LoadInt64(&d.mtime)) + // TODO(jamieliu): device number +} + +func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *linux.Statx, mnt *vfs.Mount) error { + if stat.Mask == 0 { + return nil + } + if stat.Mask&^(linux.STATX_MODE|linux.STATX_UID|linux.STATX_GID|linux.STATX_ATIME|linux.STATX_MTIME|linux.STATX_SIZE) != 0 { + return syserror.EPERM + } + if err := vfs.CheckSetStat(creds, stat, uint16(atomic.LoadUint32(&d.mode))&^linux.S_IFMT, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))); err != nil { + return err + } + if err := mnt.CheckBeginWrite(); err != nil { + return err + } + defer mnt.EndWrite() + setLocalAtime := false + setLocalMtime := false + if d.fs.opts.interop != InteropModeShared { + // Timestamp updates will be handled locally. + setLocalAtime = stat.Mask&linux.STATX_ATIME != 0 + setLocalMtime = stat.Mask&linux.STATX_MTIME != 0 + stat.Mask &^= linux.STATX_ATIME | linux.STATX_MTIME + if !setLocalMtime && (stat.Mask&linux.STATX_SIZE != 0) { + // Truncate updates mtime. + setLocalMtime = true + stat.Mtime.Nsec = linux.UTIME_NOW + } + } + d.metadataMu.Lock() + defer d.metadataMu.Unlock() + if stat.Mask != 0 { + if err := d.file.setAttr(ctx, p9.SetAttrMask{ + Permissions: stat.Mask&linux.STATX_MODE != 0, + UID: stat.Mask&linux.STATX_UID != 0, + GID: stat.Mask&linux.STATX_GID != 0, + Size: stat.Mask&linux.STATX_SIZE != 0, + ATime: stat.Mask&linux.STATX_ATIME != 0, + MTime: stat.Mask&linux.STATX_MTIME != 0, + ATimeNotSystemTime: stat.Atime.Nsec != linux.UTIME_NOW, + MTimeNotSystemTime: stat.Mtime.Nsec != linux.UTIME_NOW, + }, p9.SetAttr{ + Permissions: p9.FileMode(stat.Mode), + UID: p9.UID(stat.UID), + GID: p9.GID(stat.GID), + Size: stat.Size, + ATimeSeconds: uint64(stat.Atime.Sec), + ATimeNanoSeconds: uint64(stat.Atime.Nsec), + MTimeSeconds: uint64(stat.Mtime.Sec), + MTimeNanoSeconds: uint64(stat.Mtime.Nsec), + }); err != nil { + return err + } + } + if d.fs.opts.interop == InteropModeShared { + // There's no point to updating d's metadata in this case since it'll + // be overwritten by revalidation before the next time it's used + // anyway. (InteropModeShared inhibits client caching of regular file + // data, so there's no cache to truncate either.) + return nil + } + now, haveNow := nowFromContext(ctx) + if !haveNow { + ctx.Warningf("gofer.dentry.setStat: current time not available") + } + if stat.Mask&linux.STATX_MODE != 0 { + atomic.StoreUint32(&d.mode, d.fileType()|uint32(stat.Mode)) + } + if stat.Mask&linux.STATX_UID != 0 { + atomic.StoreUint32(&d.uid, stat.UID) + } + if stat.Mask&linux.STATX_GID != 0 { + atomic.StoreUint32(&d.gid, stat.GID) + } + if setLocalAtime { + if stat.Atime.Nsec == linux.UTIME_NOW { + if haveNow { + atomic.StoreInt64(&d.atime, now) + } + } else { + atomic.StoreInt64(&d.atime, dentryTimestampFromStatx(stat.Atime)) + } + } + if setLocalMtime { + if stat.Mtime.Nsec == linux.UTIME_NOW { + if haveNow { + atomic.StoreInt64(&d.mtime, now) + } + } else { + atomic.StoreInt64(&d.mtime, dentryTimestampFromStatx(stat.Mtime)) + } + } + if haveNow { + atomic.StoreInt64(&d.ctime, now) + } + if stat.Mask&linux.STATX_SIZE != 0 { + d.dataMu.Lock() + oldSize := d.size + d.size = stat.Size + // d.dataMu must be unlocked to lock d.mapsMu and invalidate mappings + // below. This allows concurrent calls to Read/Translate/etc. These + // functions synchronize with truncation by refusing to use cache + // contents beyond the new d.size. (We are still holding d.metadataMu, + // so we can't race with Write or another truncate.) + d.dataMu.Unlock() + if d.size < oldSize { + oldpgend := pageRoundUp(oldSize) + newpgend := pageRoundUp(d.size) + if oldpgend != newpgend { + d.mapsMu.Lock() + d.mappings.Invalidate(memmap.MappableRange{newpgend, oldpgend}, memmap.InvalidateOpts{ + // Compare Linux's mm/truncate.c:truncate_setsize() => + // truncate_pagecache() => + // mm/memory.c:unmap_mapping_range(evencows=1). + InvalidatePrivate: true, + }) + d.mapsMu.Unlock() + } + // We are now guaranteed that there are no translations of + // truncated pages, and can remove them from the cache. Since + // truncated pages have been removed from the remote file, they + // should be dropped without being written back. + d.dataMu.Lock() + d.cache.Truncate(d.size, d.fs.mfp.MemoryFile()) + d.dirty.KeepClean(memmap.MappableRange{d.size, oldpgend}) + d.dataMu.Unlock() + } + } + return nil +} + +func (d *dentry) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes, isDir bool) error { + return vfs.GenericCheckPermissions(creds, ats, isDir, uint16(atomic.LoadUint32(&d.mode))&0777, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))) +} + +// IncRef implements vfs.DentryImpl.IncRef. +func (d *dentry) IncRef() { + // d.refs may be 0 if d.fs.renameMu is locked, which serializes against + // d.checkCachingLocked(). + atomic.AddInt64(&d.refs, 1) +} + +// TryIncRef implements vfs.DentryImpl.TryIncRef. +func (d *dentry) TryIncRef() bool { + for { + refs := atomic.LoadInt64(&d.refs) + if refs == 0 { + return false + } + if atomic.CompareAndSwapInt64(&d.refs, refs, refs+1) { + return true + } + } +} + +// DecRef implements vfs.DentryImpl.DecRef. +func (d *dentry) DecRef() { + if refs := atomic.AddInt64(&d.refs, -1); refs == 0 { + d.fs.renameMu.Lock() + d.checkCachingLocked() + d.fs.renameMu.Unlock() + } else if refs < 0 { + panic("gofer.dentry.DecRef() called without holding a reference") + } +} + +// checkCachingLocked should be called after d's reference count becomes 0 or it +// becomes disowned. +// +// Preconditions: d.fs.renameMu must be locked for writing. +func (d *dentry) checkCachingLocked() { + // Dentries with a non-zero reference count must be retained. (The only way + // to obtain a reference on a dentry with zero references is via path + // resolution, which requires renameMu, so if d.refs is zero then it will + // remain zero while we hold renameMu for writing.) + if atomic.LoadInt64(&d.refs) != 0 { + if d.cached { + d.fs.cachedDentries.Remove(d) + d.fs.cachedDentriesLen-- + d.cached = false + } + return + } + // Non-child dentries with zero references are no longer reachable by path + // resolution and should be dropped immediately. + if d.vfsd.Parent() == nil || d.vfsd.IsDisowned() { + if d.cached { + d.fs.cachedDentries.Remove(d) + d.fs.cachedDentriesLen-- + d.cached = false + } + d.destroyLocked() + return + } + // If d is already cached, just move it to the front of the LRU. + if d.cached { + d.fs.cachedDentries.Remove(d) + d.fs.cachedDentries.PushFront(d) + return + } + // Cache the dentry, then evict the least recently used cached dentry if + // the cache becomes over-full. + d.fs.cachedDentries.PushFront(d) + d.fs.cachedDentriesLen++ + d.cached = true + if d.fs.cachedDentriesLen > d.fs.opts.maxCachedDentries { + victim := d.fs.cachedDentries.Back() + d.fs.cachedDentries.Remove(victim) + d.fs.cachedDentriesLen-- + victim.cached = false + // victim.refs may have become non-zero from an earlier path + // resolution since it was inserted into fs.cachedDentries; see + // dentry.incRefLocked(). Either way, we brought + // fs.cachedDentriesLen back down to fs.opts.maxCachedDentries, so + // we don't loop. + if atomic.LoadInt64(&victim.refs) == 0 { + if victimParentVFSD := victim.vfsd.Parent(); victimParentVFSD != nil { + victimParent := victimParentVFSD.Impl().(*dentry) + victimParent.dirMu.Lock() + if !victim.vfsd.IsDisowned() { + // victim can't be a mount point (in any mount + // namespace), since VFS holds references on mount + // points. + d.fs.vfsfs.VirtualFilesystem().ForceDeleteDentry(&victim.vfsd) + // We're only deleting the dentry, not the file it + // represents, so we don't need to update + // victimParent.dirents etc. + } + victimParent.dirMu.Unlock() + } + victim.destroyLocked() + } + } +} + +// Preconditions: d.fs.renameMu must be locked for writing. d.refs == 0. d is +// not a child dentry. +func (d *dentry) destroyLocked() { + ctx := context.Background() + d.handleMu.Lock() + if !d.handle.file.isNil() { + mf := d.fs.mfp.MemoryFile() + d.dataMu.Lock() + // Write dirty pages back to the remote filesystem. + if d.handleWritable { + if err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, mf, d.handle.writeFromBlocksAt); err != nil { + log.Warningf("gofer.dentry.DecRef: failed to write dirty data back: %v", err) + } + } + // Discard cached data. + d.cache.DropAll(mf) + d.dirty.RemoveAll() + d.dataMu.Unlock() + // Clunk open fids and close open host FDs. + d.handle.close(ctx) + } + d.handleMu.Unlock() + d.file.close(ctx) + // Remove d from the set of all dentries. + d.fs.syncMu.Lock() + delete(d.fs.dentries, d) + d.fs.syncMu.Unlock() + // Drop the reference held by d on its parent. + if parentVFSD := d.vfsd.Parent(); parentVFSD != nil { + parent := parentVFSD.Impl().(*dentry) + // This is parent.DecRef() without recursive locking of d.fs.renameMu. + if refs := atomic.AddInt64(&parent.refs, -1); refs == 0 { + parent.checkCachingLocked() + } else if refs < 0 { + panic("gofer.dentry.DecRef() called without holding a reference") + } + } +} + +func (d *dentry) isDeleted() bool { + return atomic.LoadUint32(&d.deleted) != 0 +} + +func (d *dentry) setDeleted() { + atomic.StoreUint32(&d.deleted, 1) +} + +func (d *dentry) listxattr(ctx context.Context) ([]string, error) { + return nil, syserror.ENOTSUP +} + +func (d *dentry) getxattr(ctx context.Context, name string) (string, error) { + // TODO(jamieliu): add vfs.GetxattrOptions.Size + return d.file.getXattr(ctx, name, linux.XATTR_SIZE_MAX) +} + +func (d *dentry) setxattr(ctx context.Context, opts *vfs.SetxattrOptions) error { + return d.file.setXattr(ctx, opts.Name, opts.Value, opts.Flags) +} + +func (d *dentry) removexattr(ctx context.Context, name string) error { + return syserror.ENOTSUP +} + +// Preconditions: d.isRegularFile() || d.isDirectory(). +func (d *dentry) ensureSharedHandle(ctx context.Context, read, write, trunc bool) error { + // O_TRUNC unconditionally requires us to obtain a new handle (opened with + // O_TRUNC). + if !trunc { + d.handleMu.RLock() + if (!read || d.handleReadable) && (!write || d.handleWritable) { + // The current handle is sufficient. + d.handleMu.RUnlock() + return nil + } + d.handleMu.RUnlock() + } + + haveOldFD := false + d.handleMu.Lock() + if (read && !d.handleReadable) || (write && !d.handleWritable) || trunc { + // Get a new handle. + wantReadable := d.handleReadable || read + wantWritable := d.handleWritable || write + h, err := openHandle(ctx, d.file, wantReadable, wantWritable, trunc) + if err != nil { + d.handleMu.Unlock() + return err + } + if !d.handle.file.isNil() { + // Check that old and new handles are compatible: If the old handle + // includes a host file descriptor but the new one does not, or + // vice versa, old and new memory mappings may be incoherent. + haveOldFD = d.handle.fd >= 0 + haveNewFD := h.fd >= 0 + if haveOldFD != haveNewFD { + d.handleMu.Unlock() + ctx.Warningf("gofer.dentry.ensureSharedHandle: can't change host FD availability from %v to %v across dentry handle upgrade", haveOldFD, haveNewFD) + h.close(ctx) + return syserror.EIO + } + if haveOldFD { + // We may have raced with callers of d.pf.FD() that are now + // using the old file descriptor, preventing us from safely + // closing it. We could handle this by invalidating existing + // memmap.Translations, but this is expensive. Instead, use + // dup2() to make the old file descriptor refer to the new file + // description, then close the new file descriptor (which is no + // longer needed). Racing callers may use the old or new file + // description, but this doesn't matter since they refer to the + // same file (unless d.fs.opts.overlayfsStaleRead is true, + // which we handle separately). + if err := syscall.Dup2(int(h.fd), int(d.handle.fd)); err != nil { + d.handleMu.Unlock() + ctx.Warningf("gofer.dentry.ensureSharedHandle: failed to dup fd %d to fd %d: %v", h.fd, d.handle.fd, err) + h.close(ctx) + return err + } + syscall.Close(int(h.fd)) + h.fd = d.handle.fd + if d.fs.opts.overlayfsStaleRead { + // Replace sentry mappings of the old FD with mappings of + // the new FD, since the two are not necessarily coherent. + if err := d.pf.hostFileMapper.RegenerateMappings(int(h.fd)); err != nil { + d.handleMu.Unlock() + ctx.Warningf("gofer.dentry.ensureSharedHandle: failed to replace sentry mappings of old FD with mappings of new FD: %v", err) + h.close(ctx) + return err + } + } + // Clunk the old fid before making the new handle visible (by + // unlocking d.handleMu). + d.handle.file.close(ctx) + } + } + // Switch to the new handle. + d.handle = h + d.handleReadable = wantReadable + d.handleWritable = wantWritable + } + d.handleMu.Unlock() + + if d.fs.opts.overlayfsStaleRead && haveOldFD { + // Invalidate application mappings that may be using the old FD; they + // will be replaced with mappings using the new FD after future calls + // to d.Translate(). This requires holding d.mapsMu, which precedes + // d.handleMu in the lock order. + d.mapsMu.Lock() + d.mappings.InvalidateAll(memmap.InvalidateOpts{}) + d.mapsMu.Unlock() + } + + return nil +} + +// fileDescription is embedded by gofer implementations of +// vfs.FileDescriptionImpl. +type fileDescription struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl +} + +func (fd *fileDescription) filesystem() *filesystem { + return fd.vfsfd.Mount().Filesystem().Impl().(*filesystem) +} + +func (fd *fileDescription) dentry() *dentry { + return fd.vfsfd.Dentry().Impl().(*dentry) +} + +// Stat implements vfs.FileDescriptionImpl.Stat. +func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { + d := fd.dentry() + if d.fs.opts.interop == InteropModeShared && opts.Mask&(linux.STATX_MODE|linux.STATX_UID|linux.STATX_GID|linux.STATX_ATIME|linux.STATX_MTIME|linux.STATX_CTIME|linux.STATX_SIZE|linux.STATX_BLOCKS|linux.STATX_BTIME) != 0 && opts.Sync != linux.AT_STATX_DONT_SYNC { + // TODO(jamieliu): Use specialFileFD.handle.file for the getattr if + // available? + if err := d.updateFromGetattr(ctx); err != nil { + return linux.Statx{}, err + } + } + var stat linux.Statx + d.statTo(&stat) + return stat, nil +} + +// SetStat implements vfs.FileDescriptionImpl.SetStat. +func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { + return fd.dentry().setStat(ctx, auth.CredentialsFromContext(ctx), &opts.Stat, fd.vfsfd.Mount()) +} + +// Listxattr implements vfs.FileDescriptionImpl.Listxattr. +func (fd *fileDescription) Listxattr(ctx context.Context) ([]string, error) { + return fd.dentry().listxattr(ctx) +} + +// Getxattr implements vfs.FileDescriptionImpl.Getxattr. +func (fd *fileDescription) Getxattr(ctx context.Context, name string) (string, error) { + return fd.dentry().getxattr(ctx, name) +} + +// Setxattr implements vfs.FileDescriptionImpl.Setxattr. +func (fd *fileDescription) Setxattr(ctx context.Context, opts vfs.SetxattrOptions) error { + return fd.dentry().setxattr(ctx, &opts) +} + +// Removexattr implements vfs.FileDescriptionImpl.Removexattr. +func (fd *fileDescription) Removexattr(ctx context.Context, name string) error { + return fd.dentry().removexattr(ctx, name) +} diff --git a/pkg/sentry/fsimpl/gofer/handle.go b/pkg/sentry/fsimpl/gofer/handle.go new file mode 100644 index 000000000..cfe66f797 --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/handle.go @@ -0,0 +1,135 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "syscall" + + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/p9" + "gvisor.dev/gvisor/pkg/safemem" +) + +// handle represents a remote "open file descriptor", consisting of an opened +// fid (p9.File) and optionally a host file descriptor. +type handle struct { + file p9file + fd int32 // -1 if unavailable +} + +// Preconditions: read || write. +func openHandle(ctx context.Context, file p9file, read, write, trunc bool) (handle, error) { + _, newfile, err := file.walk(ctx, nil) + if err != nil { + return handle{fd: -1}, err + } + var flags p9.OpenFlags + switch { + case read && !write: + flags = p9.ReadOnly + case !read && write: + flags = p9.WriteOnly + case read && write: + flags = p9.ReadWrite + } + if trunc { + flags |= p9.OpenTruncate + } + fdobj, _, _, err := newfile.open(ctx, flags) + if err != nil { + newfile.close(ctx) + return handle{fd: -1}, err + } + fd := int32(-1) + if fdobj != nil { + fd = int32(fdobj.Release()) + } + return handle{ + file: newfile, + fd: fd, + }, nil +} + +func (h *handle) close(ctx context.Context) { + h.file.close(ctx) + h.file = p9file{} + if h.fd >= 0 { + syscall.Close(int(h.fd)) + h.fd = -1 + } +} + +func (h *handle) readToBlocksAt(ctx context.Context, dsts safemem.BlockSeq, offset uint64) (uint64, error) { + if dsts.IsEmpty() { + return 0, nil + } + if h.fd >= 0 { + ctx.UninterruptibleSleepStart(false) + n, err := hostPreadv(h.fd, dsts, int64(offset)) + ctx.UninterruptibleSleepFinish(false) + return n, err + } + if dsts.NumBlocks() == 1 && !dsts.Head().NeedSafecopy() { + n, err := h.file.readAt(ctx, dsts.Head().ToSlice(), offset) + return uint64(n), err + } + // Buffer the read since p9.File.ReadAt() takes []byte. + buf := make([]byte, dsts.NumBytes()) + n, err := h.file.readAt(ctx, buf, offset) + if n == 0 { + return 0, err + } + if cp, cperr := safemem.CopySeq(dsts, safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf[:n]))); cperr != nil { + return cp, cperr + } + return uint64(n), err +} + +func (h *handle) writeFromBlocksAt(ctx context.Context, srcs safemem.BlockSeq, offset uint64) (uint64, error) { + if srcs.IsEmpty() { + return 0, nil + } + if h.fd >= 0 { + ctx.UninterruptibleSleepStart(false) + n, err := hostPwritev(h.fd, srcs, int64(offset)) + ctx.UninterruptibleSleepFinish(false) + return n, err + } + if srcs.NumBlocks() == 1 && !srcs.Head().NeedSafecopy() { + n, err := h.file.writeAt(ctx, srcs.Head().ToSlice(), offset) + return uint64(n), err + } + // Buffer the write since p9.File.WriteAt() takes []byte. + buf := make([]byte, srcs.NumBytes()) + cp, cperr := safemem.CopySeq(safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf)), srcs) + if cp == 0 { + return 0, cperr + } + n, err := h.file.writeAt(ctx, buf[:cp], offset) + if err != nil { + return uint64(n), err + } + return cp, cperr +} + +func (h *handle) sync(ctx context.Context) error { + if h.fd >= 0 { + ctx.UninterruptibleSleepStart(false) + err := syscall.Fsync(int(h.fd)) + ctx.UninterruptibleSleepFinish(false) + return err + } + return h.file.fsync(ctx) +} diff --git a/pkg/sentry/fsimpl/gofer/handle_unsafe.go b/pkg/sentry/fsimpl/gofer/handle_unsafe.go new file mode 100644 index 000000000..19560ab26 --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/handle_unsafe.go @@ -0,0 +1,66 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "syscall" + "unsafe" + + "gvisor.dev/gvisor/pkg/safemem" +) + +// Preconditions: !dsts.IsEmpty(). +func hostPreadv(fd int32, dsts safemem.BlockSeq, off int64) (uint64, error) { + // No buffering is necessary regardless of safecopy; host syscalls will + // return EFAULT if appropriate, instead of raising SIGBUS. + if dsts.NumBlocks() == 1 { + // Use pread() instead of preadv() to avoid iovec allocation and + // copying. + dst := dsts.Head() + n, _, e := syscall.Syscall6(syscall.SYS_PREAD64, uintptr(fd), dst.Addr(), uintptr(dst.Len()), uintptr(off), 0, 0) + if e != 0 { + return 0, e + } + return uint64(n), nil + } + iovs := safemem.IovecsFromBlockSeq(dsts) + n, _, e := syscall.Syscall6(syscall.SYS_PREADV, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs)), uintptr(off), 0, 0) + if e != 0 { + return 0, e + } + return uint64(n), nil +} + +// Preconditions: !srcs.IsEmpty(). +func hostPwritev(fd int32, srcs safemem.BlockSeq, off int64) (uint64, error) { + // No buffering is necessary regardless of safecopy; host syscalls will + // return EFAULT if appropriate, instead of raising SIGBUS. + if srcs.NumBlocks() == 1 { + // Use pwrite() instead of pwritev() to avoid iovec allocation and + // copying. + src := srcs.Head() + n, _, e := syscall.Syscall6(syscall.SYS_PWRITE64, uintptr(fd), src.Addr(), uintptr(src.Len()), uintptr(off), 0, 0) + if e != 0 { + return 0, e + } + return uint64(n), nil + } + iovs := safemem.IovecsFromBlockSeq(srcs) + n, _, e := syscall.Syscall6(syscall.SYS_PWRITEV, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs)), uintptr(off), 0, 0) + if e != 0 { + return 0, e + } + return uint64(n), nil +} diff --git a/pkg/sentry/fsimpl/gofer/p9file.go b/pkg/sentry/fsimpl/gofer/p9file.go new file mode 100644 index 000000000..755ac2985 --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/p9file.go @@ -0,0 +1,219 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fd" + "gvisor.dev/gvisor/pkg/p9" + "gvisor.dev/gvisor/pkg/syserror" +) + +// p9file is a wrapper around p9.File that provides methods that are +// Context-aware. +type p9file struct { + file p9.File +} + +func (f p9file) isNil() bool { + return f.file == nil +} + +func (f p9file) walk(ctx context.Context, names []string) ([]p9.QID, p9file, error) { + ctx.UninterruptibleSleepStart(false) + qids, newfile, err := f.file.Walk(names) + ctx.UninterruptibleSleepFinish(false) + return qids, p9file{newfile}, err +} + +func (f p9file) walkGetAttr(ctx context.Context, names []string) ([]p9.QID, p9file, p9.AttrMask, p9.Attr, error) { + ctx.UninterruptibleSleepStart(false) + qids, newfile, attrMask, attr, err := f.file.WalkGetAttr(names) + ctx.UninterruptibleSleepFinish(false) + return qids, p9file{newfile}, attrMask, attr, err +} + +// walkGetAttrOne is a wrapper around p9.File.WalkGetAttr that takes a single +// path component and returns a single qid. +func (f p9file) walkGetAttrOne(ctx context.Context, name string) (p9.QID, p9file, p9.AttrMask, p9.Attr, error) { + ctx.UninterruptibleSleepStart(false) + qids, newfile, attrMask, attr, err := f.file.WalkGetAttr([]string{name}) + ctx.UninterruptibleSleepFinish(false) + if err != nil { + return p9.QID{}, p9file{}, p9.AttrMask{}, p9.Attr{}, err + } + if len(qids) != 1 { + ctx.Warningf("p9.File.WalkGetAttr returned %d qids (%v), wanted 1", len(qids), qids) + if newfile != nil { + p9file{newfile}.close(ctx) + } + return p9.QID{}, p9file{}, p9.AttrMask{}, p9.Attr{}, syserror.EIO + } + return qids[0], p9file{newfile}, attrMask, attr, nil +} + +func (f p9file) statFS(ctx context.Context) (p9.FSStat, error) { + ctx.UninterruptibleSleepStart(false) + fsstat, err := f.file.StatFS() + ctx.UninterruptibleSleepFinish(false) + return fsstat, err +} + +func (f p9file) getAttr(ctx context.Context, req p9.AttrMask) (p9.QID, p9.AttrMask, p9.Attr, error) { + ctx.UninterruptibleSleepStart(false) + qid, attrMask, attr, err := f.file.GetAttr(req) + ctx.UninterruptibleSleepFinish(false) + return qid, attrMask, attr, err +} + +func (f p9file) setAttr(ctx context.Context, valid p9.SetAttrMask, attr p9.SetAttr) error { + ctx.UninterruptibleSleepStart(false) + err := f.file.SetAttr(valid, attr) + ctx.UninterruptibleSleepFinish(false) + return err +} + +func (f p9file) getXattr(ctx context.Context, name string, size uint64) (string, error) { + ctx.UninterruptibleSleepStart(false) + val, err := f.file.GetXattr(name, size) + ctx.UninterruptibleSleepFinish(false) + return val, err +} + +func (f p9file) setXattr(ctx context.Context, name, value string, flags uint32) error { + ctx.UninterruptibleSleepStart(false) + err := f.file.SetXattr(name, value, flags) + ctx.UninterruptibleSleepFinish(false) + return err +} + +func (f p9file) allocate(ctx context.Context, mode p9.AllocateMode, offset, length uint64) error { + ctx.UninterruptibleSleepStart(false) + err := f.file.Allocate(mode, offset, length) + ctx.UninterruptibleSleepFinish(false) + return err +} + +func (f p9file) close(ctx context.Context) error { + ctx.UninterruptibleSleepStart(false) + err := f.file.Close() + ctx.UninterruptibleSleepFinish(false) + return err +} + +func (f p9file) open(ctx context.Context, flags p9.OpenFlags) (*fd.FD, p9.QID, uint32, error) { + ctx.UninterruptibleSleepStart(false) + fdobj, qid, iounit, err := f.file.Open(flags) + ctx.UninterruptibleSleepFinish(false) + return fdobj, qid, iounit, err +} + +func (f p9file) readAt(ctx context.Context, p []byte, offset uint64) (int, error) { + ctx.UninterruptibleSleepStart(false) + n, err := f.file.ReadAt(p, offset) + ctx.UninterruptibleSleepFinish(false) + return n, err +} + +func (f p9file) writeAt(ctx context.Context, p []byte, offset uint64) (int, error) { + ctx.UninterruptibleSleepStart(false) + n, err := f.file.WriteAt(p, offset) + ctx.UninterruptibleSleepFinish(false) + return n, err +} + +func (f p9file) fsync(ctx context.Context) error { + ctx.UninterruptibleSleepStart(false) + err := f.file.FSync() + ctx.UninterruptibleSleepFinish(false) + return err +} + +func (f p9file) create(ctx context.Context, name string, flags p9.OpenFlags, permissions p9.FileMode, uid p9.UID, gid p9.GID) (*fd.FD, p9file, p9.QID, uint32, error) { + ctx.UninterruptibleSleepStart(false) + fdobj, newfile, qid, iounit, err := f.file.Create(name, flags, permissions, uid, gid) + ctx.UninterruptibleSleepFinish(false) + return fdobj, p9file{newfile}, qid, iounit, err +} + +func (f p9file) mkdir(ctx context.Context, name string, permissions p9.FileMode, uid p9.UID, gid p9.GID) (p9.QID, error) { + ctx.UninterruptibleSleepStart(false) + qid, err := f.file.Mkdir(name, permissions, uid, gid) + ctx.UninterruptibleSleepFinish(false) + return qid, err +} + +func (f p9file) symlink(ctx context.Context, oldName string, newName string, uid p9.UID, gid p9.GID) (p9.QID, error) { + ctx.UninterruptibleSleepStart(false) + qid, err := f.file.Symlink(oldName, newName, uid, gid) + ctx.UninterruptibleSleepFinish(false) + return qid, err +} + +func (f p9file) link(ctx context.Context, target p9file, newName string) error { + ctx.UninterruptibleSleepStart(false) + err := f.file.Link(target.file, newName) + ctx.UninterruptibleSleepFinish(false) + return err +} + +func (f p9file) mknod(ctx context.Context, name string, mode p9.FileMode, major uint32, minor uint32, uid p9.UID, gid p9.GID) (p9.QID, error) { + ctx.UninterruptibleSleepStart(false) + qid, err := f.file.Mknod(name, mode, major, minor, uid, gid) + ctx.UninterruptibleSleepFinish(false) + return qid, err +} + +func (f p9file) rename(ctx context.Context, newDir p9file, newName string) error { + ctx.UninterruptibleSleepStart(false) + err := f.file.Rename(newDir.file, newName) + ctx.UninterruptibleSleepFinish(false) + return err +} + +func (f p9file) unlinkAt(ctx context.Context, name string, flags uint32) error { + ctx.UninterruptibleSleepStart(false) + err := f.file.UnlinkAt(name, flags) + ctx.UninterruptibleSleepFinish(false) + return err +} + +func (f p9file) readdir(ctx context.Context, offset uint64, count uint32) ([]p9.Dirent, error) { + ctx.UninterruptibleSleepStart(false) + dirents, err := f.file.Readdir(offset, count) + ctx.UninterruptibleSleepFinish(false) + return dirents, err +} + +func (f p9file) readlink(ctx context.Context) (string, error) { + ctx.UninterruptibleSleepStart(false) + target, err := f.file.Readlink() + ctx.UninterruptibleSleepFinish(false) + return target, err +} + +func (f p9file) flush(ctx context.Context) error { + ctx.UninterruptibleSleepStart(false) + err := f.file.Flush() + ctx.UninterruptibleSleepFinish(false) + return err +} + +func (f p9file) connect(ctx context.Context, flags p9.ConnectFlags) (*fd.FD, error) { + ctx.UninterruptibleSleepStart(false) + fdobj, err := f.file.Connect(flags) + ctx.UninterruptibleSleepFinish(false) + return fdobj, err +} diff --git a/pkg/sentry/fsimpl/gofer/pagemath.go b/pkg/sentry/fsimpl/gofer/pagemath.go new file mode 100644 index 000000000..847cb0784 --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/pagemath.go @@ -0,0 +1,31 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "gvisor.dev/gvisor/pkg/usermem" +) + +// This are equivalent to usermem.Addr.RoundDown/Up, but without the +// potentially truncating conversion to usermem.Addr. This is necessary because +// there is no way to define generic "PageRoundDown/Up" functions in Go. + +func pageRoundDown(x uint64) uint64 { + return x &^ (usermem.PageSize - 1) +} + +func pageRoundUp(x uint64) uint64 { + return pageRoundDown(x + usermem.PageSize - 1) +} diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go new file mode 100644 index 000000000..8e11e06b3 --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/regular_file.go @@ -0,0 +1,860 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "fmt" + "io" + "math" + "sync" + "sync/atomic" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/safemem" + "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" + "gvisor.dev/gvisor/pkg/sentry/memmap" + "gvisor.dev/gvisor/pkg/sentry/pgalloc" + "gvisor.dev/gvisor/pkg/sentry/platform" + "gvisor.dev/gvisor/pkg/sentry/usage" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +func (d *dentry) isRegularFile() bool { + return d.fileType() == linux.S_IFREG +} + +type regularFileFD struct { + fileDescription + + // off is the file offset. off is protected by mu. + mu sync.Mutex + off int64 +} + +// Release implements vfs.FileDescriptionImpl.Release. +func (fd *regularFileFD) Release() { +} + +// OnClose implements vfs.FileDescriptionImpl.OnClose. +func (fd *regularFileFD) OnClose(ctx context.Context) error { + if !fd.vfsfd.IsWritable() { + return nil + } + // Skip flushing if writes may be buffered by the client, since (as with + // the VFS1 client) we don't flush buffered writes on close anyway. + d := fd.dentry() + if d.fs.opts.interop == InteropModeExclusive { + return nil + } + d.handleMu.RLock() + defer d.handleMu.RUnlock() + return d.handle.file.flush(ctx) +} + +// PRead implements vfs.FileDescriptionImpl.PRead. +func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + if offset < 0 { + return 0, syserror.EINVAL + } + if opts.Flags != 0 { + return 0, syserror.EOPNOTSUPP + } + + // Check for reading at EOF before calling into MM (but not under + // InteropModeShared, which makes d.size unreliable). + d := fd.dentry() + if d.fs.opts.interop != InteropModeShared && uint64(offset) >= atomic.LoadUint64(&d.size) { + return 0, io.EOF + } + + if fd.vfsfd.StatusFlags()&linux.O_DIRECT != 0 { + // Lock d.metadataMu for the rest of the read to prevent d.size from + // changing. + d.metadataMu.Lock() + defer d.metadataMu.Unlock() + // Write dirty cached pages that will be touched by the read back to + // the remote file. + if err := d.writeback(ctx, offset, dst.NumBytes()); err != nil { + return 0, err + } + } + + rw := getDentryReadWriter(ctx, d, offset) + if fd.vfsfd.StatusFlags()&linux.O_DIRECT != 0 { + // Require the read to go to the remote file. + rw.direct = true + } + n, err := dst.CopyOutFrom(ctx, rw) + putDentryReadWriter(rw) + if d.fs.opts.interop != InteropModeShared { + // Compare Linux's mm/filemap.c:do_generic_file_read() => file_accessed(). + d.touchAtime(ctx, fd.vfsfd.Mount()) + } + return n, err +} + +// Read implements vfs.FileDescriptionImpl.Read. +func (fd *regularFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + fd.mu.Lock() + n, err := fd.PRead(ctx, dst, fd.off, opts) + fd.off += n + fd.mu.Unlock() + return n, err +} + +// PWrite implements vfs.FileDescriptionImpl.PWrite. +func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + if offset < 0 { + return 0, syserror.EINVAL + } + if opts.Flags != 0 { + return 0, syserror.EOPNOTSUPP + } + + d := fd.dentry() + d.metadataMu.Lock() + defer d.metadataMu.Unlock() + if d.fs.opts.interop != InteropModeShared { + // Compare Linux's mm/filemap.c:__generic_file_write_iter() => + // file_update_time(). This is d.touchCMtime(), but without locking + // d.metadataMu (recursively). + if now, ok := nowFromContext(ctx); ok { + atomic.StoreInt64(&d.mtime, now) + atomic.StoreInt64(&d.ctime, now) + } + } + if fd.vfsfd.StatusFlags()&linux.O_DIRECT != 0 { + // Write dirty cached pages that will be touched by the write back to + // the remote file. + if err := d.writeback(ctx, offset, src.NumBytes()); err != nil { + return 0, err + } + // Remove touched pages from the cache. + pgstart := pageRoundDown(uint64(offset)) + pgend := pageRoundUp(uint64(offset + src.NumBytes())) + if pgend < pgstart { + return 0, syserror.EINVAL + } + mr := memmap.MappableRange{pgstart, pgend} + var freed []platform.FileRange + d.dataMu.Lock() + cseg := d.cache.LowerBoundSegment(mr.Start) + for cseg.Ok() && cseg.Start() < mr.End { + cseg = d.cache.Isolate(cseg, mr) + freed = append(freed, platform.FileRange{cseg.Value(), cseg.Value() + cseg.Range().Length()}) + cseg = d.cache.Remove(cseg).NextSegment() + } + d.dataMu.Unlock() + // Invalidate mappings of removed pages. + d.mapsMu.Lock() + d.mappings.Invalidate(mr, memmap.InvalidateOpts{}) + d.mapsMu.Unlock() + // Finally free pages removed from the cache. + mf := d.fs.mfp.MemoryFile() + for _, freedFR := range freed { + mf.DecRef(freedFR) + } + } + rw := getDentryReadWriter(ctx, d, offset) + if fd.vfsfd.StatusFlags()&linux.O_DIRECT != 0 { + // Require the write to go to the remote file. + rw.direct = true + } + n, err := src.CopyInTo(ctx, rw) + putDentryReadWriter(rw) + if n != 0 && fd.vfsfd.StatusFlags()&(linux.O_DSYNC|linux.O_SYNC) != 0 { + // Write dirty cached pages touched by the write back to the remote + // file. + if err := d.writeback(ctx, offset, src.NumBytes()); err != nil { + return 0, err + } + // Request the remote filesystem to sync the remote file. + if err := d.handle.file.fsync(ctx); err != nil { + return 0, err + } + } + return n, err +} + +// Write implements vfs.FileDescriptionImpl.Write. +func (fd *regularFileFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + fd.mu.Lock() + n, err := fd.PWrite(ctx, src, fd.off, opts) + fd.off += n + fd.mu.Unlock() + return n, err +} + +type dentryReadWriter struct { + ctx context.Context + d *dentry + off uint64 + direct bool +} + +var dentryReadWriterPool = sync.Pool{ + New: func() interface{} { + return &dentryReadWriter{} + }, +} + +func getDentryReadWriter(ctx context.Context, d *dentry, offset int64) *dentryReadWriter { + rw := dentryReadWriterPool.Get().(*dentryReadWriter) + rw.ctx = ctx + rw.d = d + rw.off = uint64(offset) + rw.direct = false + return rw +} + +func putDentryReadWriter(rw *dentryReadWriter) { + rw.ctx = nil + rw.d = nil + dentryReadWriterPool.Put(rw) +} + +// ReadToBlocks implements safemem.Reader.ReadToBlocks. +func (rw *dentryReadWriter) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) { + if dsts.IsEmpty() { + return 0, nil + } + + // If we have a mmappable host FD (which must be used here to ensure + // coherence with memory-mapped I/O), or if InteropModeShared is in effect + // (which prevents us from caching file contents and makes dentry.size + // unreliable), or if the file was opened O_DIRECT, read directly from + // dentry.handle without locking dentry.dataMu. + rw.d.handleMu.RLock() + if (rw.d.handle.fd >= 0 && !rw.d.fs.opts.forcePageCache) || rw.d.fs.opts.interop == InteropModeShared || rw.direct { + n, err := rw.d.handle.readToBlocksAt(rw.ctx, dsts, rw.off) + rw.d.handleMu.RUnlock() + rw.off += n + return n, err + } + + // Otherwise read from/through the cache. + mf := rw.d.fs.mfp.MemoryFile() + fillCache := mf.ShouldCacheEvictable() + var dataMuUnlock func() + if fillCache { + rw.d.dataMu.Lock() + dataMuUnlock = rw.d.dataMu.Unlock + } else { + rw.d.dataMu.RLock() + dataMuUnlock = rw.d.dataMu.RUnlock + } + + // Compute the range to read (limited by file size and overflow-checked). + if rw.off >= rw.d.size { + dataMuUnlock() + rw.d.handleMu.RUnlock() + return 0, io.EOF + } + end := rw.d.size + if rend := rw.off + dsts.NumBytes(); rend > rw.off && rend < end { + end = rend + } + + var done uint64 + seg, gap := rw.d.cache.Find(rw.off) + for rw.off < end { + mr := memmap.MappableRange{rw.off, end} + switch { + case seg.Ok(): + // Get internal mappings from the cache. + ims, err := mf.MapInternal(seg.FileRangeOf(seg.Range().Intersect(mr)), usermem.Read) + if err != nil { + dataMuUnlock() + rw.d.handleMu.RUnlock() + return done, err + } + + // Copy from internal mappings. + n, err := safemem.CopySeq(dsts, ims) + done += n + rw.off += n + dsts = dsts.DropFirst64(n) + if err != nil { + dataMuUnlock() + rw.d.handleMu.RUnlock() + return done, err + } + + // Continue. + seg, gap = seg.NextNonEmpty() + + case gap.Ok(): + gapMR := gap.Range().Intersect(mr) + if fillCache { + // Read into the cache, then re-enter the loop to read from the + // cache. + reqMR := memmap.MappableRange{ + Start: pageRoundDown(gapMR.Start), + End: pageRoundUp(gapMR.End), + } + optMR := gap.Range() + err := rw.d.cache.Fill(rw.ctx, reqMR, maxFillRange(reqMR, optMR), mf, usage.PageCache, rw.d.handle.readToBlocksAt) + mf.MarkEvictable(rw.d, pgalloc.EvictableRange{optMR.Start, optMR.End}) + seg, gap = rw.d.cache.Find(rw.off) + if !seg.Ok() { + dataMuUnlock() + rw.d.handleMu.RUnlock() + return done, err + } + // err might have occurred in part of gap.Range() outside + // gapMR. Forget about it for now; if the error matters and + // persists, we'll run into it again in a later iteration of + // this loop. + } else { + // Read directly from the file. + gapDsts := dsts.TakeFirst64(gapMR.Length()) + n, err := rw.d.handle.readToBlocksAt(rw.ctx, gapDsts, gapMR.Start) + done += n + rw.off += n + dsts = dsts.DropFirst64(n) + // Partial reads are fine. But we must stop reading. + if n != gapDsts.NumBytes() || err != nil { + dataMuUnlock() + rw.d.handleMu.RUnlock() + return done, err + } + + // Continue. + seg, gap = gap.NextSegment(), fsutil.FileRangeGapIterator{} + } + } + } + dataMuUnlock() + rw.d.handleMu.RUnlock() + return done, nil +} + +// WriteFromBlocks implements safemem.Writer.WriteFromBlocks. +// +// Preconditions: rw.d.metadataMu must be locked. +func (rw *dentryReadWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) { + if srcs.IsEmpty() { + return 0, nil + } + + // If we have a mmappable host FD (which must be used here to ensure + // coherence with memory-mapped I/O), or if InteropModeShared is in effect + // (which prevents us from caching file contents), or if the file was + // opened with O_DIRECT, write directly to dentry.handle without locking + // dentry.dataMu. + rw.d.handleMu.RLock() + if (rw.d.handle.fd >= 0 && !rw.d.fs.opts.forcePageCache) || rw.d.fs.opts.interop == InteropModeShared || rw.direct { + n, err := rw.d.handle.writeFromBlocksAt(rw.ctx, srcs, rw.off) + rw.d.handleMu.RUnlock() + rw.off += n + return n, err + } + + // Otherwise write to/through the cache. + mf := rw.d.fs.mfp.MemoryFile() + rw.d.dataMu.Lock() + + // Compute the range to write (overflow-checked). + start := rw.off + end := rw.off + srcs.NumBytes() + if end <= rw.off { + end = math.MaxInt64 + } + + var ( + done uint64 + retErr error + ) + seg, gap := rw.d.cache.Find(rw.off) + for rw.off < end { + mr := memmap.MappableRange{rw.off, end} + switch { + case seg.Ok(): + // Get internal mappings from the cache. + segMR := seg.Range().Intersect(mr) + ims, err := mf.MapInternal(seg.FileRangeOf(segMR), usermem.Write) + if err != nil { + retErr = err + goto exitLoop + } + + // Copy to internal mappings. + n, err := safemem.CopySeq(ims, srcs) + done += n + rw.off += n + srcs = srcs.DropFirst64(n) + rw.d.dirty.MarkDirty(segMR) + if err != nil { + retErr = err + goto exitLoop + } + + // Continue. + seg, gap = seg.NextNonEmpty() + + case gap.Ok(): + // Write directly to the file. At present, we never fill the cache + // when writing, since doing so can convert small writes into + // inefficient read-modify-write cycles, and we have no mechanism + // for detecting or avoiding this. + gapMR := gap.Range().Intersect(mr) + gapSrcs := srcs.TakeFirst64(gapMR.Length()) + n, err := rw.d.handle.writeFromBlocksAt(rw.ctx, gapSrcs, gapMR.Start) + done += n + rw.off += n + srcs = srcs.DropFirst64(n) + // Partial writes are fine. But we must stop writing. + if n != gapSrcs.NumBytes() || err != nil { + retErr = err + goto exitLoop + } + + // Continue. + seg, gap = gap.NextSegment(), fsutil.FileRangeGapIterator{} + } + } +exitLoop: + if rw.off > rw.d.size { + atomic.StoreUint64(&rw.d.size, rw.off) + // The remote file's size will implicitly be extended to the correct + // value when we write back to it. + } + // If InteropModeWritethrough is in effect, flush written data back to the + // remote filesystem. + if rw.d.fs.opts.interop == InteropModeWritethrough && done != 0 { + if err := fsutil.SyncDirty(rw.ctx, memmap.MappableRange{ + Start: start, + End: rw.off, + }, &rw.d.cache, &rw.d.dirty, rw.d.size, mf, rw.d.handle.writeFromBlocksAt); err != nil { + // We have no idea how many bytes were actually flushed. + rw.off = start + done = 0 + retErr = err + } + } + rw.d.dataMu.Unlock() + rw.d.handleMu.RUnlock() + return done, retErr +} + +func (d *dentry) writeback(ctx context.Context, offset, size int64) error { + if size == 0 { + return nil + } + d.handleMu.RLock() + defer d.handleMu.RUnlock() + d.dataMu.Lock() + defer d.dataMu.Unlock() + // Compute the range of valid bytes (overflow-checked). + if uint64(offset) >= d.size { + return nil + } + end := int64(d.size) + if rend := offset + size; rend > offset && rend < end { + end = rend + } + return fsutil.SyncDirty(ctx, memmap.MappableRange{ + Start: uint64(offset), + End: uint64(end), + }, &d.cache, &d.dirty, d.size, d.fs.mfp.MemoryFile(), d.handle.writeFromBlocksAt) +} + +// Seek implements vfs.FileDescriptionImpl.Seek. +func (fd *regularFileFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + fd.mu.Lock() + defer fd.mu.Unlock() + switch whence { + case linux.SEEK_SET: + // Use offset as specified. + case linux.SEEK_CUR: + offset += fd.off + case linux.SEEK_END, linux.SEEK_DATA, linux.SEEK_HOLE: + // Ensure file size is up to date. + d := fd.dentry() + if fd.filesystem().opts.interop == InteropModeShared { + if err := d.updateFromGetattr(ctx); err != nil { + return 0, err + } + } + size := int64(atomic.LoadUint64(&d.size)) + // For SEEK_DATA and SEEK_HOLE, treat the file as a single contiguous + // block of data. + switch whence { + case linux.SEEK_END: + offset += size + case linux.SEEK_DATA: + if offset > size { + return 0, syserror.ENXIO + } + // Use offset as specified. + case linux.SEEK_HOLE: + if offset > size { + return 0, syserror.ENXIO + } + offset = size + } + default: + return 0, syserror.EINVAL + } + if offset < 0 { + return 0, syserror.EINVAL + } + fd.off = offset + return offset, nil +} + +// Sync implements vfs.FileDescriptionImpl.Sync. +func (fd *regularFileFD) Sync(ctx context.Context) error { + return fd.dentry().syncSharedHandle(ctx) +} + +func (d *dentry) syncSharedHandle(ctx context.Context) error { + d.handleMu.RLock() + if !d.handleWritable { + d.handleMu.RUnlock() + return nil + } + d.dataMu.Lock() + // Write dirty cached data to the remote file. + err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, d.fs.mfp.MemoryFile(), d.handle.writeFromBlocksAt) + d.dataMu.Unlock() + if err == nil { + // Sync the remote file. + err = d.handle.sync(ctx) + } + d.handleMu.RUnlock() + return err +} + +// ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. +func (fd *regularFileFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { + d := fd.dentry() + switch d.fs.opts.interop { + case InteropModeExclusive: + // Any mapping is fine. + case InteropModeWritethrough: + // Shared writable mappings require a host FD, since otherwise we can't + // synchronously flush memory-mapped writes to the remote file. + if opts.Private || !opts.MaxPerms.Write { + break + } + fallthrough + case InteropModeShared: + // All mappings require a host FD to be coherent with other filesystem + // users. + if d.fs.opts.forcePageCache { + // Whether or not we have a host FD, we're not allowed to use it. + return syserror.ENODEV + } + d.handleMu.RLock() + haveFD := d.handle.fd >= 0 + d.handleMu.RUnlock() + if !haveFD { + return syserror.ENODEV + } + default: + panic(fmt.Sprintf("unknown InteropMode %v", d.fs.opts.interop)) + } + return vfs.GenericConfigureMMap(&fd.vfsfd, d, opts) +} + +func (d *dentry) mayCachePages() bool { + if d.fs.opts.interop == InteropModeShared { + return false + } + if d.fs.opts.forcePageCache { + return true + } + d.handleMu.RLock() + haveFD := d.handle.fd >= 0 + d.handleMu.RUnlock() + return haveFD +} + +// AddMapping implements memmap.Mappable.AddMapping. +func (d *dentry) AddMapping(ctx context.Context, ms memmap.MappingSpace, ar usermem.AddrRange, offset uint64, writable bool) error { + d.mapsMu.Lock() + mapped := d.mappings.AddMapping(ms, ar, offset, writable) + // Do this unconditionally since whether we have a host FD can change + // across save/restore. + for _, r := range mapped { + d.pf.hostFileMapper.IncRefOn(r) + } + if d.mayCachePages() { + // d.Evict() will refuse to evict memory-mapped pages, so tell the + // MemoryFile to not bother trying. + mf := d.fs.mfp.MemoryFile() + for _, r := range mapped { + mf.MarkUnevictable(d, pgalloc.EvictableRange{r.Start, r.End}) + } + } + d.mapsMu.Unlock() + return nil +} + +// RemoveMapping implements memmap.Mappable.RemoveMapping. +func (d *dentry) RemoveMapping(ctx context.Context, ms memmap.MappingSpace, ar usermem.AddrRange, offset uint64, writable bool) { + d.mapsMu.Lock() + unmapped := d.mappings.RemoveMapping(ms, ar, offset, writable) + for _, r := range unmapped { + d.pf.hostFileMapper.DecRefOn(r) + } + if d.mayCachePages() { + // Pages that are no longer referenced by any application memory + // mappings are now considered unused; allow MemoryFile to evict them + // when necessary. + mf := d.fs.mfp.MemoryFile() + d.dataMu.Lock() + for _, r := range unmapped { + // Since these pages are no longer mapped, they are no longer + // concurrently dirtyable by a writable memory mapping. + d.dirty.AllowClean(r) + mf.MarkEvictable(d, pgalloc.EvictableRange{r.Start, r.End}) + } + d.dataMu.Unlock() + } + d.mapsMu.Unlock() +} + +// CopyMapping implements memmap.Mappable.CopyMapping. +func (d *dentry) CopyMapping(ctx context.Context, ms memmap.MappingSpace, srcAR, dstAR usermem.AddrRange, offset uint64, writable bool) error { + return d.AddMapping(ctx, ms, dstAR, offset, writable) +} + +// Translate implements memmap.Mappable.Translate. +func (d *dentry) Translate(ctx context.Context, required, optional memmap.MappableRange, at usermem.AccessType) ([]memmap.Translation, error) { + d.handleMu.RLock() + if d.handle.fd >= 0 && !d.fs.opts.forcePageCache { + d.handleMu.RUnlock() + mr := optional + if d.fs.opts.limitHostFDTranslation { + mr = maxFillRange(required, optional) + } + return []memmap.Translation{ + { + Source: mr, + File: &d.pf, + Offset: mr.Start, + Perms: usermem.AnyAccess, + }, + }, nil + } + + d.dataMu.Lock() + + // Constrain translations to d.size (rounded up) to prevent translation to + // pages that may be concurrently truncated. + pgend := pageRoundUp(d.size) + var beyondEOF bool + if required.End > pgend { + if required.Start >= pgend { + d.dataMu.Unlock() + d.handleMu.RUnlock() + return nil, &memmap.BusError{io.EOF} + } + beyondEOF = true + required.End = pgend + } + if optional.End > pgend { + optional.End = pgend + } + + mf := d.fs.mfp.MemoryFile() + cerr := d.cache.Fill(ctx, required, maxFillRange(required, optional), mf, usage.PageCache, d.handle.readToBlocksAt) + + var ts []memmap.Translation + var translatedEnd uint64 + for seg := d.cache.FindSegment(required.Start); seg.Ok() && seg.Start() < required.End; seg, _ = seg.NextNonEmpty() { + segMR := seg.Range().Intersect(optional) + // TODO(jamieliu): Make Translations writable even if writability is + // not required if already kept-dirty by another writable translation. + perms := usermem.AccessType{ + Read: true, + Execute: true, + } + if at.Write { + // From this point forward, this memory can be dirtied through the + // mapping at any time. + d.dirty.KeepDirty(segMR) + perms.Write = true + } + ts = append(ts, memmap.Translation{ + Source: segMR, + File: mf, + Offset: seg.FileRangeOf(segMR).Start, + Perms: perms, + }) + translatedEnd = segMR.End + } + + d.dataMu.Unlock() + d.handleMu.RUnlock() + + // Don't return the error returned by c.cache.Fill if it occurred outside + // of required. + if translatedEnd < required.End && cerr != nil { + return ts, &memmap.BusError{cerr} + } + if beyondEOF { + return ts, &memmap.BusError{io.EOF} + } + return ts, nil +} + +func maxFillRange(required, optional memmap.MappableRange) memmap.MappableRange { + const maxReadahead = 64 << 10 // 64 KB, chosen arbitrarily + if required.Length() >= maxReadahead { + return required + } + if optional.Length() <= maxReadahead { + return optional + } + optional.Start = required.Start + if optional.Length() <= maxReadahead { + return optional + } + optional.End = optional.Start + maxReadahead + return optional +} + +// InvalidateUnsavable implements memmap.Mappable.InvalidateUnsavable. +func (d *dentry) InvalidateUnsavable(ctx context.Context) error { + // Whether we have a host fd (and consequently what platform.File is + // mapped) can change across save/restore, so invalidate all translations + // unconditionally. + d.mapsMu.Lock() + defer d.mapsMu.Unlock() + d.mappings.InvalidateAll(memmap.InvalidateOpts{}) + + // Write the cache's contents back to the remote file so that if we have a + // host fd after restore, the remote file's contents are coherent. + mf := d.fs.mfp.MemoryFile() + d.dataMu.Lock() + defer d.dataMu.Unlock() + if err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, mf, d.handle.writeFromBlocksAt); err != nil { + return err + } + + // Discard the cache so that it's not stored in saved state. This is safe + // because per InvalidateUnsavable invariants, no new translations can have + // been returned after we invalidated all existing translations above. + d.cache.DropAll(mf) + d.dirty.RemoveAll() + + return nil +} + +// Evict implements pgalloc.EvictableMemoryUser.Evict. +func (d *dentry) Evict(ctx context.Context, er pgalloc.EvictableRange) { + d.mapsMu.Lock() + defer d.mapsMu.Unlock() + d.dataMu.Lock() + defer d.dataMu.Unlock() + + mr := memmap.MappableRange{er.Start, er.End} + mf := d.fs.mfp.MemoryFile() + // Only allow pages that are no longer memory-mapped to be evicted. + for mgap := d.mappings.LowerBoundGap(mr.Start); mgap.Ok() && mgap.Start() < mr.End; mgap = mgap.NextGap() { + mgapMR := mgap.Range().Intersect(mr) + if mgapMR.Length() == 0 { + continue + } + if err := fsutil.SyncDirty(ctx, mgapMR, &d.cache, &d.dirty, d.size, mf, d.handle.writeFromBlocksAt); err != nil { + log.Warningf("Failed to writeback cached data %v: %v", mgapMR, err) + } + d.cache.Drop(mgapMR, mf) + d.dirty.KeepClean(mgapMR) + } +} + +// dentryPlatformFile implements platform.File. It exists solely because dentry +// cannot implement both vfs.DentryImpl.IncRef and platform.File.IncRef. +// +// dentryPlatformFile is only used when a host FD representing the remote file +// is available (i.e. dentry.handle.fd >= 0), and that FD is used for +// application memory mappings (i.e. !filesystem.opts.forcePageCache). +type dentryPlatformFile struct { + *dentry + + // fdRefs counts references on platform.File offsets. fdRefs is protected + // by dentry.dataMu. + fdRefs fsutil.FrameRefSet + + // If this dentry represents a regular file, and handle.fd >= 0, + // hostFileMapper caches mappings of handle.fd. + hostFileMapper fsutil.HostFileMapper +} + +// IncRef implements platform.File.IncRef. +func (d *dentryPlatformFile) IncRef(fr platform.FileRange) { + d.dataMu.Lock() + seg, gap := d.fdRefs.Find(fr.Start) + for { + switch { + case seg.Ok() && seg.Start() < fr.End: + seg = d.fdRefs.Isolate(seg, fr) + seg.SetValue(seg.Value() + 1) + seg, gap = seg.NextNonEmpty() + case gap.Ok() && gap.Start() < fr.End: + newRange := gap.Range().Intersect(fr) + usage.MemoryAccounting.Inc(newRange.Length(), usage.Mapped) + seg, gap = d.fdRefs.InsertWithoutMerging(gap, newRange, 1).NextNonEmpty() + default: + d.fdRefs.MergeAdjacent(fr) + d.dataMu.Unlock() + return + } + } +} + +// DecRef implements platform.File.DecRef. +func (d *dentryPlatformFile) DecRef(fr platform.FileRange) { + d.dataMu.Lock() + seg := d.fdRefs.FindSegment(fr.Start) + + for seg.Ok() && seg.Start() < fr.End { + seg = d.fdRefs.Isolate(seg, fr) + if old := seg.Value(); old == 1 { + usage.MemoryAccounting.Dec(seg.Range().Length(), usage.Mapped) + seg = d.fdRefs.Remove(seg).NextSegment() + } else { + seg.SetValue(old - 1) + seg = seg.NextSegment() + } + } + d.fdRefs.MergeAdjacent(fr) + d.dataMu.Unlock() + +} + +// MapInternal implements platform.File.MapInternal. +func (d *dentryPlatformFile) MapInternal(fr platform.FileRange, at usermem.AccessType) (safemem.BlockSeq, error) { + d.handleMu.RLock() + bs, err := d.hostFileMapper.MapInternal(fr, int(d.handle.fd), at.Write) + d.handleMu.RUnlock() + return bs, err +} + +// FD implements platform.File.FD. +func (d *dentryPlatformFile) FD() int { + d.handleMu.RLock() + fd := d.handle.fd + d.handleMu.RUnlock() + return int(fd) +} diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go new file mode 100644 index 000000000..08c691c47 --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/special_file.go @@ -0,0 +1,159 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "sync" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/safemem" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +// specialFileFD implements vfs.FileDescriptionImpl for files other than +// regular files, directories, and symlinks: pipes, sockets, etc. It is also +// used for regular files when filesystemOptions.specialRegularFiles is in +// effect. specialFileFD differs from regularFileFD by using per-FD handles +// instead of shared per-dentry handles, and never buffering I/O. +type specialFileFD struct { + fileDescription + + // handle is immutable. + handle handle + + // off is the file offset. off is protected by mu. (POSIX 2.9.7 only + // requires operations using the file offset to be atomic for regular files + // and symlinks; however, since specialFileFD may be used for regular + // files, we apply this atomicity unconditionally.) + mu sync.Mutex + off int64 +} + +// Release implements vfs.FileDescriptionImpl.Release. +func (fd *specialFileFD) Release() { + fd.handle.close(context.Background()) + fs := fd.vfsfd.Mount().Filesystem().Impl().(*filesystem) + fs.syncMu.Lock() + delete(fs.specialFileFDs, fd) + fs.syncMu.Unlock() +} + +// OnClose implements vfs.FileDescriptionImpl.OnClose. +func (fd *specialFileFD) OnClose(ctx context.Context) error { + if !fd.vfsfd.IsWritable() { + return nil + } + return fd.handle.file.flush(ctx) +} + +// PRead implements vfs.FileDescriptionImpl.PRead. +func (fd *specialFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + if offset < 0 { + return 0, syserror.EINVAL + } + if opts.Flags != 0 { + return 0, syserror.EOPNOTSUPP + } + + // Going through dst.CopyOutFrom() holds MM locks around file operations of + // unknown duration. For regularFileFD, doing so is necessary to support + // mmap due to lock ordering; MM locks precede dentry.dataMu. That doesn't + // hold here since specialFileFD doesn't client-cache data. Just buffer the + // read instead. + if d := fd.dentry(); d.fs.opts.interop != InteropModeShared { + d.touchAtime(ctx, fd.vfsfd.Mount()) + } + buf := make([]byte, dst.NumBytes()) + n, err := fd.handle.readToBlocksAt(ctx, safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf)), uint64(offset)) + if n == 0 { + return 0, err + } + if cp, cperr := dst.CopyOut(ctx, buf[:n]); cperr != nil { + return int64(cp), cperr + } + return int64(n), err +} + +// Read implements vfs.FileDescriptionImpl.Read. +func (fd *specialFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + fd.mu.Lock() + n, err := fd.PRead(ctx, dst, fd.off, opts) + fd.off += n + fd.mu.Unlock() + return n, err +} + +// PWrite implements vfs.FileDescriptionImpl.PWrite. +func (fd *specialFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + if offset < 0 { + return 0, syserror.EINVAL + } + if opts.Flags != 0 { + return 0, syserror.EOPNOTSUPP + } + + // Do a buffered write. See rationale in PRead. + if d := fd.dentry(); d.fs.opts.interop != InteropModeShared { + d.touchCMtime(ctx) + } + buf := make([]byte, src.NumBytes()) + // Don't do partial writes if we get a partial read from src. + if _, err := src.CopyIn(ctx, buf); err != nil { + return 0, err + } + n, err := fd.handle.writeFromBlocksAt(ctx, safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf)), uint64(offset)) + return int64(n), err +} + +// Write implements vfs.FileDescriptionImpl.Write. +func (fd *specialFileFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + fd.mu.Lock() + n, err := fd.PWrite(ctx, src, fd.off, opts) + fd.off += n + fd.mu.Unlock() + return n, err +} + +// Seek implements vfs.FileDescriptionImpl.Seek. +func (fd *specialFileFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + fd.mu.Lock() + defer fd.mu.Unlock() + switch whence { + case linux.SEEK_SET: + // Use offset as given. + case linux.SEEK_CUR: + offset += fd.off + default: + // SEEK_END, SEEK_DATA, and SEEK_HOLE aren't supported since it's not + // clear that file size is even meaningful for these files. + return 0, syserror.EINVAL + } + if offset < 0 { + return 0, syserror.EINVAL + } + fd.off = offset + return offset, nil +} + +// Sync implements vfs.FileDescriptionImpl.Sync. +func (fd *specialFileFD) Sync(ctx context.Context) error { + if !fd.vfsfd.IsWritable() { + return nil + } + return fd.handle.sync(ctx) +} diff --git a/pkg/sentry/fsimpl/gofer/symlink.go b/pkg/sentry/fsimpl/gofer/symlink.go new file mode 100644 index 000000000..adf43be60 --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/symlink.go @@ -0,0 +1,47 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +func (d *dentry) isSymlink() bool { + return d.fileType() == linux.S_IFLNK +} + +// Precondition: d.isSymlink(). +func (d *dentry) readlink(ctx context.Context, mnt *vfs.Mount) (string, error) { + if d.fs.opts.interop != InteropModeShared { + d.touchAtime(ctx, mnt) + d.dataMu.Lock() + if d.haveTarget { + target := d.target + d.dataMu.Unlock() + return target, nil + } + } + target, err := d.file.readlink(ctx) + if d.fs.opts.interop != InteropModeShared { + if err == nil { + d.haveTarget = true + d.target = target + } + d.dataMu.Unlock() + } + return target, err +} diff --git a/pkg/sentry/fsimpl/gofer/time.go b/pkg/sentry/fsimpl/gofer/time.go new file mode 100644 index 000000000..7598ec6a8 --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/time.go @@ -0,0 +1,75 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "sync/atomic" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +func dentryTimestampFromP9(s, ns uint64) int64 { + return int64(s*1e9 + ns) +} + +func dentryTimestampFromStatx(ts linux.StatxTimestamp) int64 { + return ts.Sec*1e9 + int64(ts.Nsec) +} + +func statxTimestampFromDentry(ns int64) linux.StatxTimestamp { + return linux.StatxTimestamp{ + Sec: ns / 1e9, + Nsec: uint32(ns % 1e9), + } +} + +func nowFromContext(ctx context.Context) (int64, bool) { + if clock := ktime.RealtimeClockFromContext(ctx); clock != nil { + return clock.Now().Nanoseconds(), true + } + return 0, false +} + +// Preconditions: fs.interop != InteropModeShared. +func (d *dentry) touchAtime(ctx context.Context, mnt *vfs.Mount) { + if err := mnt.CheckBeginWrite(); err != nil { + return + } + now, ok := nowFromContext(ctx) + if !ok { + mnt.EndWrite() + return + } + d.metadataMu.Lock() + atomic.StoreInt64(&d.atime, now) + d.metadataMu.Unlock() + mnt.EndWrite() +} + +// Preconditions: fs.interop != InteropModeShared. The caller has successfully +// called vfs.Mount.CheckBeginWrite(). +func (d *dentry) touchCMtime(ctx context.Context) { + now, ok := nowFromContext(ctx) + if !ok { + return + } + d.metadataMu.Lock() + atomic.StoreInt64(&d.mtime, now) + atomic.StoreInt64(&d.ctime, now) + d.metadataMu.Unlock() +} diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 5ee9cf1e9..72bc15264 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -622,7 +622,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error if child.inode.isDir() { return syserror.EISDIR } - if !rp.MustBeDir() { + if rp.MustBeDir() { return syserror.ENOTDIR } mnt := rp.Mount() diff --git a/pkg/sentry/socket/hostinet/socket.go b/pkg/sentry/socket/hostinet/socket.go index bde4c7a1e..34f63986f 100644 --- a/pkg/sentry/socket/hostinet/socket.go +++ b/pkg/sentry/socket/hostinet/socket.go @@ -126,7 +126,7 @@ func (s *socketOperations) Read(ctx context.Context, _ *fs.File, dst usermem.IOS } return uint64(n), nil } - return readv(s.fd, iovecsFromBlockSeq(dsts)) + return readv(s.fd, safemem.IovecsFromBlockSeq(dsts)) })) return int64(n), err } @@ -149,7 +149,7 @@ func (s *socketOperations) Write(ctx context.Context, _ *fs.File, src usermem.IO } return uint64(n), nil } - return writev(s.fd, iovecsFromBlockSeq(srcs)) + return writev(s.fd, safemem.IovecsFromBlockSeq(srcs)) })) return int64(n), err } @@ -402,7 +402,7 @@ func (s *socketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags // We always do a non-blocking recv*(). sysflags := flags | syscall.MSG_DONTWAIT - iovs := iovecsFromBlockSeq(dsts) + iovs := safemem.IovecsFromBlockSeq(dsts) msg := syscall.Msghdr{ Iov: &iovs[0], Iovlen: uint64(len(iovs)), @@ -522,7 +522,7 @@ func (s *socketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to [] return uint64(n), nil } - iovs := iovecsFromBlockSeq(srcs) + iovs := safemem.IovecsFromBlockSeq(srcs) msg := syscall.Msghdr{ Iov: &iovs[0], Iovlen: uint64(len(iovs)), @@ -567,21 +567,6 @@ func (s *socketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to [] return int(n), syserr.FromError(err) } -func iovecsFromBlockSeq(bs safemem.BlockSeq) []syscall.Iovec { - iovs := make([]syscall.Iovec, 0, bs.NumBlocks()) - for ; !bs.IsEmpty(); bs = bs.Tail() { - b := bs.Head() - iovs = append(iovs, syscall.Iovec{ - Base: &b.ToSlice()[0], - Len: uint64(b.Len()), - }) - // We don't need to care about b.NeedSafecopy(), because the host - // kernel will handle such address ranges just fine (by returning - // EFAULT). - } - return iovs -} - func translateIOSyscallError(err error) error { if err == syscall.EAGAIN || err == syscall.EWOULDBLOCK { return syserror.ErrWouldBlock From dcffddf0cae026411e7e678744a1e39dc2b513cf Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Tue, 4 Feb 2020 11:47:41 -0800 Subject: [PATCH 18/29] Remove argument from vfs.MountNamespace.DecRef() Updates #1035 PiperOrigin-RevId: 293194631 --- pkg/sentry/fsimpl/devtmpfs/devtmpfs.go | 2 +- pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go | 2 +- pkg/sentry/fsimpl/testutil/testutil.go | 2 +- pkg/sentry/fsimpl/tmpfs/benchmark_test.go | 4 ++-- pkg/sentry/fsimpl/tmpfs/regular_file_test.go | 2 +- pkg/sentry/vfs/mount.go | 3 ++- 6 files changed, 8 insertions(+), 7 deletions(-) diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go index d36fa74fb..e03a0c665 100644 --- a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go +++ b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go @@ -86,7 +86,7 @@ func NewAccessor(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth // Release must be called when a is no longer in use. func (a *Accessor) Release() { a.root.DecRef() - a.mntns.DecRef(a.vfsObj) + a.mntns.DecRef() } // accessorContext implements context.Context by extending an existing diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go index 82c58c900..73308a2b5 100644 --- a/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go +++ b/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go @@ -45,7 +45,7 @@ func TestDevtmpfs(t *testing.T) { if err != nil { t.Fatalf("failed to create tmpfs root mount: %v", err) } - defer mntns.DecRef(vfsObj) + defer mntns.DecRef() root := mntns.Root() defer root.DecRef() devpop := vfs.PathOperation{ diff --git a/pkg/sentry/fsimpl/testutil/testutil.go b/pkg/sentry/fsimpl/testutil/testutil.go index 1c98335c1..69fd84ddd 100644 --- a/pkg/sentry/fsimpl/testutil/testutil.go +++ b/pkg/sentry/fsimpl/testutil/testutil.go @@ -98,7 +98,7 @@ func (s *System) WithTemporaryContext(ctx context.Context) *System { // Destroy release resources associated with a test system. func (s *System) Destroy() { s.Root.DecRef() - s.mns.DecRef(s.VFS) // Reference on mns passed to NewSystem. + s.mns.DecRef() // Reference on mns passed to NewSystem. } // ReadToEnd reads the contents of fd until EOF to a string. diff --git a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go index 54241c8e8..9fce5e4b4 100644 --- a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go +++ b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go @@ -183,7 +183,7 @@ func BenchmarkVFS2MemfsStat(b *testing.B) { if err != nil { b.Fatalf("failed to create tmpfs root mount: %v", err) } - defer mntns.DecRef(vfsObj) + defer mntns.DecRef() var filePathBuilder strings.Builder filePathBuilder.WriteByte('/') @@ -374,7 +374,7 @@ func BenchmarkVFS2MemfsMountStat(b *testing.B) { if err != nil { b.Fatalf("failed to create tmpfs root mount: %v", err) } - defer mntns.DecRef(vfsObj) + defer mntns.DecRef() var filePathBuilder strings.Builder filePathBuilder.WriteByte('/') diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go index 2b52992ea..e9f71e334 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go @@ -51,7 +51,7 @@ func newTmpfsRoot(ctx context.Context) (*vfs.VirtualFilesystem, vfs.VirtualDentr root := mntns.Root() return vfsObj, root, func() { root.DecRef() - mntns.DecRef(vfsObj) + mntns.DecRef() }, nil } diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index d39528051..1fbb420f9 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -423,7 +423,8 @@ func (mntns *MountNamespace) IncRef() { } // DecRef decrements mntns' reference count. -func (mntns *MountNamespace) DecRef(vfs *VirtualFilesystem) { +func (mntns *MountNamespace) DecRef() { + vfs := mntns.root.fs.VirtualFilesystem() if refs := atomic.AddInt64(&mntns.refs, -1); refs == 0 { vfs.mountMu.Lock() vfs.mounts.seq.BeginWrite() From f5072caaf85b9f067d737a874804c04e2b9039b8 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Tue, 4 Feb 2020 12:39:58 -0800 Subject: [PATCH 19/29] Fix safecopy test. This is failing in Go1.14 due to new checkptr constraints. This version should avoid hitting this constraints by doing "dangerous" pointer math less dangerously? PiperOrigin-RevId: 293205764 --- pkg/safecopy/safecopy_test.go | 88 ++++++++++++++++------------- pkg/safecopy/safecopy_unsafe.go | 98 +++++++++++++++++++++------------ 2 files changed, 112 insertions(+), 74 deletions(-) diff --git a/pkg/safecopy/safecopy_test.go b/pkg/safecopy/safecopy_test.go index 5818f7f9b..7f7f69d61 100644 --- a/pkg/safecopy/safecopy_test.go +++ b/pkg/safecopy/safecopy_test.go @@ -138,10 +138,14 @@ func TestSwapUint32Success(t *testing.T) { func TestSwapUint32AlignmentError(t *testing.T) { // Test that SwapUint32 returns an AlignmentError when passed an unaligned // address. - data := new(struct{ val uint64 }) - addr := uintptr(unsafe.Pointer(&data.val)) + 1 - want := AlignmentError{Addr: addr, Alignment: 4} - if _, err := SwapUint32(unsafe.Pointer(addr), 1); err != want { + data := make([]byte, 8) // 2 * sizeof(uint32). + alignedIndex := uintptr(0) + if offset := uintptr(unsafe.Pointer(&data[0])) % 4; offset != 0 { + alignedIndex = 4 - offset + } + ptr := unsafe.Pointer(&data[alignedIndex+1]) + want := AlignmentError{Addr: uintptr(ptr), Alignment: 4} + if _, err := SwapUint32(ptr, 1); err != want { t.Errorf("Unexpected error: got %v, want %v", err, want) } } @@ -171,10 +175,14 @@ func TestSwapUint64Success(t *testing.T) { func TestSwapUint64AlignmentError(t *testing.T) { // Test that SwapUint64 returns an AlignmentError when passed an unaligned // address. - data := new(struct{ val1, val2 uint64 }) - addr := uintptr(unsafe.Pointer(&data.val1)) + 1 - want := AlignmentError{Addr: addr, Alignment: 8} - if _, err := SwapUint64(unsafe.Pointer(addr), 1); err != want { + data := make([]byte, 16) // 2 * sizeof(uint64). + alignedIndex := uintptr(0) + if offset := uintptr(unsafe.Pointer(&data[0])) % 8; offset != 0 { + alignedIndex = 8 - offset + } + ptr := unsafe.Pointer(&data[alignedIndex+1]) + want := AlignmentError{Addr: uintptr(ptr), Alignment: 8} + if _, err := SwapUint64(ptr, 1); err != want { t.Errorf("Unexpected error: got %v, want %v", err, want) } } @@ -201,10 +209,14 @@ func TestCompareAndSwapUint32Success(t *testing.T) { func TestCompareAndSwapUint32AlignmentError(t *testing.T) { // Test that CompareAndSwapUint32 returns an AlignmentError when passed an // unaligned address. - data := new(struct{ val uint64 }) - addr := uintptr(unsafe.Pointer(&data.val)) + 1 - want := AlignmentError{Addr: addr, Alignment: 4} - if _, err := CompareAndSwapUint32(unsafe.Pointer(addr), 0, 1); err != want { + data := make([]byte, 8) // 2 * sizeof(uint32). + alignedIndex := uintptr(0) + if offset := uintptr(unsafe.Pointer(&data[0])) % 4; offset != 0 { + alignedIndex = 4 - offset + } + ptr := unsafe.Pointer(&data[alignedIndex+1]) + want := AlignmentError{Addr: uintptr(ptr), Alignment: 4} + if _, err := CompareAndSwapUint32(ptr, 0, 1); err != want { t.Errorf("Unexpected error: got %v, want %v", err, want) } } @@ -252,8 +264,8 @@ func TestCopyInSegvError(t *testing.T) { for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { t.Run(fmt.Sprintf("starting copy %d bytes before SIGSEGV", bytesBeforeFault), func(t *testing.T) { withSegvErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - src := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) + secondPage := uintptr(unsafe.Pointer(&mapping[pageSize])) + src := unsafe.Pointer(&mapping[pageSize-bytesBeforeFault]) dst := randBuf(pageSize) n, err := CopyIn(dst, src) if n != bytesBeforeFault { @@ -276,8 +288,8 @@ func TestCopyInBusError(t *testing.T) { for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { t.Run(fmt.Sprintf("starting copy %d bytes before SIGBUS", bytesBeforeFault), func(t *testing.T) { withBusErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - src := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) + secondPage := uintptr(unsafe.Pointer(&mapping[pageSize])) + src := unsafe.Pointer(&mapping[pageSize-bytesBeforeFault]) dst := randBuf(pageSize) n, err := CopyIn(dst, src) if n != bytesBeforeFault { @@ -300,8 +312,8 @@ func TestCopyOutSegvError(t *testing.T) { for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { t.Run(fmt.Sprintf("starting copy %d bytes before SIGSEGV", bytesBeforeFault), func(t *testing.T) { withSegvErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - dst := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) + secondPage := uintptr(unsafe.Pointer(&mapping[pageSize])) + dst := unsafe.Pointer(&mapping[pageSize-bytesBeforeFault]) src := randBuf(pageSize) n, err := CopyOut(dst, src) if n != bytesBeforeFault { @@ -324,8 +336,8 @@ func TestCopyOutBusError(t *testing.T) { for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { t.Run(fmt.Sprintf("starting copy %d bytes before SIGSEGV", bytesBeforeFault), func(t *testing.T) { withBusErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - dst := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) + secondPage := uintptr(unsafe.Pointer(&mapping[pageSize])) + dst := unsafe.Pointer(&mapping[pageSize-bytesBeforeFault]) src := randBuf(pageSize) n, err := CopyOut(dst, src) if n != bytesBeforeFault { @@ -348,8 +360,8 @@ func TestCopySourceSegvError(t *testing.T) { for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { t.Run(fmt.Sprintf("starting copy %d bytes before SIGSEGV", bytesBeforeFault), func(t *testing.T) { withSegvErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - src := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) + secondPage := uintptr(unsafe.Pointer(&mapping[pageSize])) + src := unsafe.Pointer(&mapping[pageSize-bytesBeforeFault]) dst := randBuf(pageSize) n, err := Copy(unsafe.Pointer(&dst[0]), src, pageSize) if n != uintptr(bytesBeforeFault) { @@ -372,8 +384,8 @@ func TestCopySourceBusError(t *testing.T) { for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { t.Run(fmt.Sprintf("starting copy %d bytes before SIGBUS", bytesBeforeFault), func(t *testing.T) { withBusErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - src := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) + secondPage := uintptr(unsafe.Pointer(&mapping[pageSize])) + src := unsafe.Pointer(&mapping[pageSize-bytesBeforeFault]) dst := randBuf(pageSize) n, err := Copy(unsafe.Pointer(&dst[0]), src, pageSize) if n != uintptr(bytesBeforeFault) { @@ -396,8 +408,8 @@ func TestCopyDestinationSegvError(t *testing.T) { for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { t.Run(fmt.Sprintf("starting copy %d bytes before SIGSEGV", bytesBeforeFault), func(t *testing.T) { withSegvErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - dst := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) + secondPage := uintptr(unsafe.Pointer(&mapping[pageSize])) + dst := unsafe.Pointer(&mapping[pageSize-bytesBeforeFault]) src := randBuf(pageSize) n, err := Copy(dst, unsafe.Pointer(&src[0]), pageSize) if n != uintptr(bytesBeforeFault) { @@ -420,8 +432,8 @@ func TestCopyDestinationBusError(t *testing.T) { for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { t.Run(fmt.Sprintf("starting copy %d bytes before SIGBUS", bytesBeforeFault), func(t *testing.T) { withBusErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - dst := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) + secondPage := uintptr(unsafe.Pointer(&mapping[pageSize])) + dst := unsafe.Pointer(&mapping[pageSize-bytesBeforeFault]) src := randBuf(pageSize) n, err := Copy(dst, unsafe.Pointer(&src[0]), pageSize) if n != uintptr(bytesBeforeFault) { @@ -444,8 +456,8 @@ func TestZeroOutSegvError(t *testing.T) { for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { t.Run(fmt.Sprintf("starting write %d bytes before SIGSEGV", bytesBeforeFault), func(t *testing.T) { withSegvErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - dst := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) + secondPage := uintptr(unsafe.Pointer(&mapping[pageSize])) + dst := unsafe.Pointer(&mapping[pageSize-bytesBeforeFault]) n, err := ZeroOut(dst, pageSize) if n != uintptr(bytesBeforeFault) { t.Errorf("Unexpected write length: got %v, want %v", n, bytesBeforeFault) @@ -467,8 +479,8 @@ func TestZeroOutBusError(t *testing.T) { for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { t.Run(fmt.Sprintf("starting write %d bytes before SIGBUS", bytesBeforeFault), func(t *testing.T) { withBusErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - dst := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) + secondPage := uintptr(unsafe.Pointer(&mapping[pageSize])) + dst := unsafe.Pointer(&mapping[pageSize-bytesBeforeFault]) n, err := ZeroOut(dst, pageSize) if n != uintptr(bytesBeforeFault) { t.Errorf("Unexpected write length: got %v, want %v", n, bytesBeforeFault) @@ -488,7 +500,7 @@ func TestSwapUint32SegvError(t *testing.T) { // Test that SwapUint32 returns a SegvError when reaching a page that // signals SIGSEGV. withSegvErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize + secondPage := uintptr(unsafe.Pointer(&mapping[pageSize])) _, err := SwapUint32(unsafe.Pointer(secondPage), 1) if want := (SegvError{secondPage}); err != want { t.Errorf("Unexpected error: got %v, want %v", err, want) @@ -500,7 +512,7 @@ func TestSwapUint32BusError(t *testing.T) { // Test that SwapUint32 returns a BusError when reaching a page that // signals SIGBUS. withBusErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize + secondPage := uintptr(unsafe.Pointer(&mapping[pageSize])) _, err := SwapUint32(unsafe.Pointer(secondPage), 1) if want := (BusError{secondPage}); err != want { t.Errorf("Unexpected error: got %v, want %v", err, want) @@ -512,7 +524,7 @@ func TestSwapUint64SegvError(t *testing.T) { // Test that SwapUint64 returns a SegvError when reaching a page that // signals SIGSEGV. withSegvErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize + secondPage := uintptr(unsafe.Pointer(&mapping[pageSize])) _, err := SwapUint64(unsafe.Pointer(secondPage), 1) if want := (SegvError{secondPage}); err != want { t.Errorf("Unexpected error: got %v, want %v", err, want) @@ -524,7 +536,7 @@ func TestSwapUint64BusError(t *testing.T) { // Test that SwapUint64 returns a BusError when reaching a page that // signals SIGBUS. withBusErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize + secondPage := uintptr(unsafe.Pointer(&mapping[pageSize])) _, err := SwapUint64(unsafe.Pointer(secondPage), 1) if want := (BusError{secondPage}); err != want { t.Errorf("Unexpected error: got %v, want %v", err, want) @@ -536,7 +548,7 @@ func TestCompareAndSwapUint32SegvError(t *testing.T) { // Test that CompareAndSwapUint32 returns a SegvError when reaching a page // that signals SIGSEGV. withSegvErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize + secondPage := uintptr(unsafe.Pointer(&mapping[pageSize])) _, err := CompareAndSwapUint32(unsafe.Pointer(secondPage), 0, 1) if want := (SegvError{secondPage}); err != want { t.Errorf("Unexpected error: got %v, want %v", err, want) @@ -548,7 +560,7 @@ func TestCompareAndSwapUint32BusError(t *testing.T) { // Test that CompareAndSwapUint32 returns a BusError when reaching a page // that signals SIGBUS. withBusErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize + secondPage := uintptr(unsafe.Pointer(&mapping[pageSize])) _, err := CompareAndSwapUint32(unsafe.Pointer(secondPage), 0, 1) if want := (BusError{secondPage}); err != want { t.Errorf("Unexpected error: got %v, want %v", err, want) diff --git a/pkg/safecopy/safecopy_unsafe.go b/pkg/safecopy/safecopy_unsafe.go index eef028e68..41dd567f3 100644 --- a/pkg/safecopy/safecopy_unsafe.go +++ b/pkg/safecopy/safecopy_unsafe.go @@ -16,6 +16,7 @@ package safecopy import ( "fmt" + "runtime" "syscall" "unsafe" ) @@ -35,7 +36,7 @@ const maxRegisterSize = 16 // successfully copied. // //go:noescape -func memcpy(dst, src unsafe.Pointer, n uintptr) (fault unsafe.Pointer, sig int32) +func memcpy(dst, src uintptr, n uintptr) (fault uintptr, sig int32) // memclr sets the n bytes following ptr to zeroes. If a SIGSEGV or SIGBUS // signal is received during the write, it returns the address that caused the @@ -47,7 +48,7 @@ func memcpy(dst, src unsafe.Pointer, n uintptr) (fault unsafe.Pointer, sig int32 // successfully written. // //go:noescape -func memclr(ptr unsafe.Pointer, n uintptr) (fault unsafe.Pointer, sig int32) +func memclr(ptr uintptr, n uintptr) (fault uintptr, sig int32) // swapUint32 atomically stores new into *ptr and returns (the previous *ptr // value, 0). If a SIGSEGV or SIGBUS signal is received during the swap, the @@ -90,29 +91,35 @@ func loadUint32(ptr unsafe.Pointer) (val uint32, sig int32) // CopyIn copies len(dst) bytes from src to dst. It returns the number of bytes // copied and an error if SIGSEGV or SIGBUS is received while reading from src. func CopyIn(dst []byte, src unsafe.Pointer) (int, error) { + n, err := copyIn(dst, uintptr(src)) + runtime.KeepAlive(src) + return n, err +} + +// copyIn is the underlying definition for CopyIn. +func copyIn(dst []byte, src uintptr) (int, error) { toCopy := uintptr(len(dst)) if len(dst) == 0 { return 0, nil } - fault, sig := memcpy(unsafe.Pointer(&dst[0]), src, toCopy) + fault, sig := memcpy(uintptr(unsafe.Pointer(&dst[0])), src, toCopy) if sig == 0 { return len(dst), nil } - faultN, srcN := uintptr(fault), uintptr(src) - if faultN < srcN || faultN >= srcN+toCopy { - panic(fmt.Sprintf("CopyIn raised signal %d at %#x, which is outside source [%#x, %#x)", sig, faultN, srcN, srcN+toCopy)) + if fault < src || fault >= src+toCopy { + panic(fmt.Sprintf("CopyIn raised signal %d at %#x, which is outside source [%#x, %#x)", sig, fault, src, src+toCopy)) } // memcpy might have ended the copy up to maxRegisterSize bytes before // fault, if an instruction caused a memory access that straddled two // pages, and the second one faulted. Try to copy up to the fault. var done int - if faultN-srcN > maxRegisterSize { - done = int(faultN - srcN - maxRegisterSize) + if fault-src > maxRegisterSize { + done = int(fault - src - maxRegisterSize) } - n, err := CopyIn(dst[done:int(faultN-srcN)], unsafe.Pointer(srcN+uintptr(done))) + n, err := copyIn(dst[done:int(fault-src)], src+uintptr(done)) done += n if err != nil { return done, err @@ -124,29 +131,35 @@ func CopyIn(dst []byte, src unsafe.Pointer) (int, error) { // bytes done and an error if SIGSEGV or SIGBUS is received while writing to // dst. func CopyOut(dst unsafe.Pointer, src []byte) (int, error) { + n, err := copyOut(uintptr(dst), src) + runtime.KeepAlive(dst) + return n, err +} + +// copyOut is the underlying definition for CopyOut. +func copyOut(dst uintptr, src []byte) (int, error) { toCopy := uintptr(len(src)) if toCopy == 0 { return 0, nil } - fault, sig := memcpy(dst, unsafe.Pointer(&src[0]), toCopy) + fault, sig := memcpy(dst, uintptr(unsafe.Pointer(&src[0])), toCopy) if sig == 0 { return len(src), nil } - faultN, dstN := uintptr(fault), uintptr(dst) - if faultN < dstN || faultN >= dstN+toCopy { - panic(fmt.Sprintf("CopyOut raised signal %d at %#x, which is outside destination [%#x, %#x)", sig, faultN, dstN, dstN+toCopy)) + if fault < dst || fault >= dst+toCopy { + panic(fmt.Sprintf("CopyOut raised signal %d at %#x, which is outside destination [%#x, %#x)", sig, fault, dst, dst+toCopy)) } // memcpy might have ended the copy up to maxRegisterSize bytes before // fault, if an instruction caused a memory access that straddled two // pages, and the second one faulted. Try to copy up to the fault. var done int - if faultN-dstN > maxRegisterSize { - done = int(faultN - dstN - maxRegisterSize) + if fault-dst > maxRegisterSize { + done = int(fault - dst - maxRegisterSize) } - n, err := CopyOut(unsafe.Pointer(dstN+uintptr(done)), src[done:int(faultN-dstN)]) + n, err := copyOut(dst+uintptr(done), src[done:int(fault-dst)]) done += n if err != nil { return done, err @@ -161,6 +174,14 @@ func CopyOut(dst unsafe.Pointer, src []byte) (int, error) { // Data is copied in order; if [src, src+toCopy) and [dst, dst+toCopy) overlap, // the resulting contents of dst are unspecified. func Copy(dst, src unsafe.Pointer, toCopy uintptr) (uintptr, error) { + n, err := copyN(uintptr(dst), uintptr(src), toCopy) + runtime.KeepAlive(dst) + runtime.KeepAlive(src) + return n, err +} + +// copyN is the underlying definition for Copy. +func copyN(dst, src uintptr, toCopy uintptr) (uintptr, error) { if toCopy == 0 { return 0, nil } @@ -171,17 +192,16 @@ func Copy(dst, src unsafe.Pointer, toCopy uintptr) (uintptr, error) { } // Did the fault occur while reading from src or writing to dst? - faultN, srcN, dstN := uintptr(fault), uintptr(src), uintptr(dst) faultAfterSrc := ^uintptr(0) - if faultN >= srcN { - faultAfterSrc = faultN - srcN + if fault >= src { + faultAfterSrc = fault - src } faultAfterDst := ^uintptr(0) - if faultN >= dstN { - faultAfterDst = faultN - dstN + if fault >= dst { + faultAfterDst = fault - dst } if faultAfterSrc >= toCopy && faultAfterDst >= toCopy { - panic(fmt.Sprintf("Copy raised signal %d at %#x, which is outside source [%#x, %#x) and destination [%#x, %#x)", sig, faultN, srcN, srcN+toCopy, dstN, dstN+toCopy)) + panic(fmt.Sprintf("Copy raised signal %d at %#x, which is outside source [%#x, %#x) and destination [%#x, %#x)", sig, fault, src, src+toCopy, dst, dst+toCopy)) } faultedAfter := faultAfterSrc if faultedAfter > faultAfterDst { @@ -195,7 +215,7 @@ func Copy(dst, src unsafe.Pointer, toCopy uintptr) (uintptr, error) { if faultedAfter > maxRegisterSize { done = faultedAfter - maxRegisterSize } - n, err := Copy(unsafe.Pointer(dstN+done), unsafe.Pointer(srcN+done), faultedAfter-done) + n, err := copyN(dst+done, src+done, faultedAfter-done) done += n if err != nil { return done, err @@ -206,6 +226,13 @@ func Copy(dst, src unsafe.Pointer, toCopy uintptr) (uintptr, error) { // ZeroOut writes toZero zero bytes to dst. It returns the number of bytes // written and an error if SIGSEGV or SIGBUS is received while writing to dst. func ZeroOut(dst unsafe.Pointer, toZero uintptr) (uintptr, error) { + n, err := zeroOut(uintptr(dst), toZero) + runtime.KeepAlive(dst) + return n, err +} + +// zeroOut is the underlying definition for ZeroOut. +func zeroOut(dst uintptr, toZero uintptr) (uintptr, error) { if toZero == 0 { return 0, nil } @@ -215,19 +242,18 @@ func ZeroOut(dst unsafe.Pointer, toZero uintptr) (uintptr, error) { return toZero, nil } - faultN, dstN := uintptr(fault), uintptr(dst) - if faultN < dstN || faultN >= dstN+toZero { - panic(fmt.Sprintf("ZeroOut raised signal %d at %#x, which is outside destination [%#x, %#x)", sig, faultN, dstN, dstN+toZero)) + if fault < dst || fault >= dst+toZero { + panic(fmt.Sprintf("ZeroOut raised signal %d at %#x, which is outside destination [%#x, %#x)", sig, fault, dst, dst+toZero)) } // memclr might have ended the write up to maxRegisterSize bytes before // fault, if an instruction caused a memory access that straddled two // pages, and the second one faulted. Try to write up to the fault. var done uintptr - if faultN-dstN > maxRegisterSize { - done = faultN - dstN - maxRegisterSize + if fault-dst > maxRegisterSize { + done = fault - dst - maxRegisterSize } - n, err := ZeroOut(unsafe.Pointer(dstN+done), faultN-dstN-done) + n, err := zeroOut(dst+done, fault-dst-done) done += n if err != nil { return done, err @@ -243,7 +269,7 @@ func SwapUint32(ptr unsafe.Pointer, new uint32) (uint32, error) { return 0, AlignmentError{addr, 4} } old, sig := swapUint32(ptr, new) - return old, errorFromFaultSignal(ptr, sig) + return old, errorFromFaultSignal(uintptr(ptr), sig) } // SwapUint64 is equivalent to sync/atomic.SwapUint64, except that it returns @@ -254,7 +280,7 @@ func SwapUint64(ptr unsafe.Pointer, new uint64) (uint64, error) { return 0, AlignmentError{addr, 8} } old, sig := swapUint64(ptr, new) - return old, errorFromFaultSignal(ptr, sig) + return old, errorFromFaultSignal(uintptr(ptr), sig) } // CompareAndSwapUint32 is equivalent to atomicbitops.CompareAndSwapUint32, @@ -265,7 +291,7 @@ func CompareAndSwapUint32(ptr unsafe.Pointer, old, new uint32) (uint32, error) { return 0, AlignmentError{addr, 4} } prev, sig := compareAndSwapUint32(ptr, old, new) - return prev, errorFromFaultSignal(ptr, sig) + return prev, errorFromFaultSignal(uintptr(ptr), sig) } // LoadUint32 is like sync/atomic.LoadUint32, but operates with user memory. It @@ -277,17 +303,17 @@ func LoadUint32(ptr unsafe.Pointer) (uint32, error) { return 0, AlignmentError{addr, 4} } val, sig := loadUint32(ptr) - return val, errorFromFaultSignal(ptr, sig) + return val, errorFromFaultSignal(uintptr(ptr), sig) } -func errorFromFaultSignal(addr unsafe.Pointer, sig int32) error { +func errorFromFaultSignal(addr uintptr, sig int32) error { switch sig { case 0: return nil case int32(syscall.SIGSEGV): - return SegvError{uintptr(addr)} + return SegvError{addr} case int32(syscall.SIGBUS): - return BusError{uintptr(addr)} + return BusError{addr} default: panic(fmt.Sprintf("safecopy got unexpected signal %d at address %#x", sig, addr)) } From c5d4041623ac6405135e966af6d06c178a86870d Mon Sep 17 00:00:00 2001 From: Jay Zhuang Date: Tue, 4 Feb 2020 12:53:10 -0800 Subject: [PATCH 20/29] Include socket_ip_udp_loopback.cc in exportes_files So it can be included in fuchsia's syscall tests PiperOrigin-RevId: 293208306 --- test/syscalls/linux/BUILD | 1 + 1 file changed, 1 insertion(+) diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index e4ca5b6db..737e2329f 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -11,6 +11,7 @@ exports_files( "socket_inet_loopback.cc", "socket_ip_loopback_blocking.cc", "socket_ip_tcp_loopback.cc", + "socket_ip_udp_loopback.cc", "socket_ipv4_udp_unbound_loopback.cc", "tcp_socket.cc", "udp_socket.cc", From 6823b5e244a5748032130574ae3a25a0a36bbbf5 Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Tue, 4 Feb 2020 13:05:30 -0800 Subject: [PATCH 21/29] timer_create(2) should return 0 on success The timer ID is copied out to the argument. Fixes #1738 PiperOrigin-RevId: 293210801 --- pkg/sentry/syscalls/linux/sys_timer.go | 2 +- test/syscalls/linux/timers.cc | 18 +++++++++++++++++- 2 files changed, 18 insertions(+), 2 deletions(-) diff --git a/pkg/sentry/syscalls/linux/sys_timer.go b/pkg/sentry/syscalls/linux/sys_timer.go index 432351917..a4c400f87 100644 --- a/pkg/sentry/syscalls/linux/sys_timer.go +++ b/pkg/sentry/syscalls/linux/sys_timer.go @@ -146,7 +146,7 @@ func TimerCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S return 0, nil, err } - return uintptr(id), nil, nil + return 0, nil, nil } // TimerSettime implements linux syscall timer_settime(2). diff --git a/test/syscalls/linux/timers.cc b/test/syscalls/linux/timers.cc index 3db18d7ac..2f92c27da 100644 --- a/test/syscalls/linux/timers.cc +++ b/test/syscalls/linux/timers.cc @@ -297,9 +297,13 @@ class IntervalTimer { PosixErrorOr TimerCreate(clockid_t clockid, const struct sigevent& sev) { int timerid; - if (syscall(SYS_timer_create, clockid, &sev, &timerid) < 0) { + int ret = syscall(SYS_timer_create, clockid, &sev, &timerid); + if (ret < 0) { return PosixError(errno, "timer_create"); } + if (ret > 0) { + return PosixError(EINVAL, "timer_create should never return positive"); + } MaybeSave(); return IntervalTimer(timerid); } @@ -317,6 +321,18 @@ TEST(IntervalTimerTest, IsInitiallyStopped) { EXPECT_EQ(0, its.it_value.tv_nsec); } +// Kernel can create multiple timers without issue. +// +// Regression test for gvisor.dev/issue/1738. +TEST(IntervalTimerTest, MultipleTimers) { + struct sigevent sev = {}; + sev.sigev_notify = SIGEV_NONE; + const auto timer1 = + ASSERT_NO_ERRNO_AND_VALUE(TimerCreate(CLOCK_MONOTONIC, sev)); + const auto timer2 = + ASSERT_NO_ERRNO_AND_VALUE(TimerCreate(CLOCK_MONOTONIC, sev)); +} + TEST(IntervalTimerTest, SingleShotSilent) { struct sigevent sev = {}; sev.sigev_notify = SIGEV_NONE; From 6d8bf405bc5e887247534172713bf7d2f5252734 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Tue, 4 Feb 2020 13:15:05 -0800 Subject: [PATCH 22/29] Allow mlock in fsgofer system call filters Go 1.14 has a workaround for a Linux 5.2-5.4 bug which requires mlock'ing the g stack to prevent register corruption. We need to allow this syscall until it is removed from Go. PiperOrigin-RevId: 293212935 --- runsc/fsgofer/filter/config.go | 12 ++++++++++++ 1 file changed, 12 insertions(+) diff --git a/runsc/fsgofer/filter/config.go b/runsc/fsgofer/filter/config.go index a1792330f..1dce36965 100644 --- a/runsc/fsgofer/filter/config.go +++ b/runsc/fsgofer/filter/config.go @@ -128,6 +128,18 @@ var allowedSyscalls = seccomp.SyscallRules{ syscall.SYS_MADVISE: {}, unix.SYS_MEMFD_CREATE: {}, /// Used by flipcall.PacketWindowAllocator.Init(). syscall.SYS_MKDIRAT: {}, + // Used by the Go runtime as a temporarily workaround for a Linux + // 5.2-5.4 bug. + // + // See src/runtime/os_linux_x86.go. + // + // TODO(b/148688965): Remove once this is gone from Go. + syscall.SYS_MLOCK: []seccomp.Rule{ + { + seccomp.AllowAny{}, + seccomp.AllowValue(4096), + }, + }, syscall.SYS_MMAP: []seccomp.Rule{ { seccomp.AllowAny{}, From 95ce8bb4c7ecb23e47e68c60b1de0b99ad8a856d Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Tue, 4 Feb 2020 14:36:43 -0800 Subject: [PATCH 23/29] Automatically propagate tags for stateify and marshal. Note that files will need to be appropriately segmented in order for the mechanism to work, in suffixes implying special tags. This only needs to happen for cases where marshal or state structures are defined, which should be rare and mostly architecture specific. PiperOrigin-RevId: 293231579 --- tools/build/defs.bzl | 2 + tools/build/tags.bzl | 36 +++++++ tools/defs.bzl | 105 +++++++++++++------- tools/go_marshal/gomarshal/BUILD | 1 + tools/go_marshal/gomarshal/generator.go | 11 +++ tools/go_stateify/BUILD | 1 + tools/go_stateify/defs.bzl | 10 +- tools/go_stateify/main.go | 122 ++---------------------- tools/tags/BUILD | 11 +++ tools/tags/tags.go | 89 +++++++++++++++++ 10 files changed, 235 insertions(+), 153 deletions(-) create mode 100644 tools/build/tags.bzl create mode 100644 tools/tags/BUILD create mode 100644 tools/tags/tags.go diff --git a/tools/build/defs.bzl b/tools/build/defs.bzl index 967c1f900..1a1a0d825 100644 --- a/tools/build/defs.bzl +++ b/tools/build/defs.bzl @@ -8,6 +8,7 @@ load("@rules_pkg//:pkg.bzl", _pkg_deb = "pkg_deb", _pkg_tar = "pkg_tar") load("@io_bazel_rules_docker//go:image.bzl", _go_image = "go_image") load("@io_bazel_rules_docker//container:container.bzl", _container_image = "container_image") load("@pydeps//:requirements.bzl", _py_requirement = "requirement") +load("//tools/build:tags.bzl", _go_suffixes = "go_suffixes") container_image = _container_image cc_binary = _cc_binary @@ -18,6 +19,7 @@ cc_test = _cc_test cc_toolchain = "@bazel_tools//tools/cpp:current_cc_toolchain" go_image = _go_image go_embed_data = _go_embed_data +go_suffixes = _go_suffixes gtest = "@com_google_googletest//:gtest" loopback = "//tools/build:loopback" proto_library = native.proto_library diff --git a/tools/build/tags.bzl b/tools/build/tags.bzl new file mode 100644 index 000000000..e99c87f81 --- /dev/null +++ b/tools/build/tags.bzl @@ -0,0 +1,36 @@ +"""List of special Go suffixes.""" + +go_suffixes = [ + "_386", + "_386_unsafe", + "_amd64", + "_amd64_unsafe", + "_aarch64", + "_aarch64_unsafe", + "_arm", + "_arm_unsafe", + "_arm64", + "_arm64_unsafe", + "_mips", + "_mips_unsafe", + "_mipsle", + "_mipsle_unsafe", + "_mips64", + "_mips64_unsafe", + "_mips64le", + "_mips64le_unsafe", + "_ppc64", + "_ppc64_unsafe", + "_ppc64le", + "_ppc64le_unsafe", + "_riscv64", + "_riscv64_unsafe", + "_s390x", + "_s390x_unsafe", + "_sparc64", + "_sparc64_unsafe", + "_wasm", + "_wasm_unsafe", + "_linux", + "_linux_unsafe", +] diff --git a/tools/defs.bzl b/tools/defs.bzl index ce677cbbf..5d5fa134a 100644 --- a/tools/defs.bzl +++ b/tools/defs.bzl @@ -7,7 +7,7 @@ change for Google-internal and bazel-compatible rules. load("//tools/go_stateify:defs.bzl", "go_stateify") load("//tools/go_marshal:defs.bzl", "go_marshal", "marshal_deps", "marshal_test_deps") -load("//tools/build:defs.bzl", _cc_binary = "cc_binary", _cc_flags_supplier = "cc_flags_supplier", _cc_library = "cc_library", _cc_proto_library = "cc_proto_library", _cc_test = "cc_test", _cc_toolchain = "cc_toolchain", _container_image = "container_image", _default_installer = "default_installer", _default_net_util = "default_net_util", _go_binary = "go_binary", _go_embed_data = "go_embed_data", _go_image = "go_image", _go_library = "go_library", _go_proto_library = "go_proto_library", _go_test = "go_test", _go_tool_library = "go_tool_library", _gtest = "gtest", _loopback = "loopback", _pkg_deb = "pkg_deb", _pkg_tar = "pkg_tar", _proto_library = "proto_library", _py_binary = "py_binary", _py_library = "py_library", _py_requirement = "py_requirement", _py_test = "py_test", _select_arch = "select_arch", _select_system = "select_system") +load("//tools/build:defs.bzl", "go_suffixes", _cc_binary = "cc_binary", _cc_flags_supplier = "cc_flags_supplier", _cc_library = "cc_library", _cc_proto_library = "cc_proto_library", _cc_test = "cc_test", _cc_toolchain = "cc_toolchain", _container_image = "container_image", _default_installer = "default_installer", _default_net_util = "default_net_util", _go_binary = "go_binary", _go_embed_data = "go_embed_data", _go_image = "go_image", _go_library = "go_library", _go_proto_library = "go_proto_library", _go_test = "go_test", _go_tool_library = "go_tool_library", _gtest = "gtest", _loopback = "loopback", _pkg_deb = "pkg_deb", _pkg_tar = "pkg_tar", _proto_library = "proto_library", _py_binary = "py_binary", _py_library = "py_library", _py_requirement = "py_requirement", _py_test = "py_test", _select_arch = "select_arch", _select_system = "select_system") # Delegate directly. cc_binary = _cc_binary @@ -45,6 +45,34 @@ def go_binary(name, **kwargs): **kwargs ) +def calculate_sets(srcs): + """Calculates special Go sets for templates. + + Args: + srcs: the full set of Go sources. + + Returns: + A dictionary of the form: + + "": [src1.go, src2.go] + "suffix": [src3suffix.go, src4suffix.go] + + Note that suffix will typically start with '_'. + """ + result = dict() + for file in srcs: + if not file.endswith(".go"): + continue + target = "" + for suffix in go_suffixes: + if file.endswith(suffix + ".go"): + target = suffix + if not target in result: + result[target] = [file] + else: + result[target].append(file) + return result + def go_library(name, srcs, deps = [], imports = [], stateify = True, marshal = False, **kwargs): """Wraps the standard go_library and does stateification and marshalling. @@ -70,39 +98,49 @@ def go_library(name, srcs, deps = [], imports = [], stateify = True, marshal = F marshal: whether marshal is enabled (default: false). **kwargs: standard go_library arguments. """ + all_srcs = srcs + all_deps = deps if stateify: # Only do stateification for non-state packages without manual autogen. - go_stateify( - name = name + "_state_autogen", - srcs = [src for src in srcs if src.endswith(".go")], - imports = imports, - package = name, - arch = select_arch(), - out = name + "_state_autogen.go", - ) - all_srcs = srcs + [name + "_state_autogen.go"] - if "//pkg/state" not in deps: - all_deps = deps + ["//pkg/state"] - else: - all_deps = deps - else: - all_deps = deps - all_srcs = srcs + # First, we need to segregate the input files via the special suffixes, + # and calculate the final output set. + state_sets = calculate_sets(srcs) + for (suffix, srcs) in state_sets.items(): + go_stateify( + name = name + suffix + "_state_autogen", + srcs = srcs, + imports = imports, + package = name, + out = name + suffix + "_state_autogen.go", + ) + all_srcs = all_srcs + [ + name + suffix + "_state_autogen.go" + for suffix in state_sets.keys() + ] + if "//pkg/state" not in all_deps: + all_deps = all_deps + ["//pkg/state"] + if marshal: - go_marshal( - name = name + "_abi_autogen", - srcs = [src for src in srcs if src.endswith(".go")], - debug = False, - imports = imports, - package = name, - ) + # See above. + marshal_sets = calculate_sets(srcs) + for (suffix, srcs) in marshal_sets.items(): + go_marshal( + name = name + suffix + "_abi_autogen", + srcs = srcs, + debug = False, + imports = imports, + package = name, + ) extra_deps = [ dep for dep in marshal_deps if not dep in all_deps ] all_deps = all_deps + extra_deps - all_srcs = srcs + [name + "_abi_autogen_unsafe.go"] + all_srcs = all_srcs + [ + name + suffix + "_abi_autogen_unsafe.go" + for suffix in marshal_sets.keys() + ] _go_library( name = name, @@ -115,13 +153,16 @@ def go_library(name, srcs, deps = [], imports = [], stateify = True, marshal = F # Ignore importpath for go_test. kwargs.pop("importpath", None) - _go_test( - name = name + "_abi_autogen_test", - srcs = [name + "_abi_autogen_test.go"], - library = ":" + name, - deps = marshal_test_deps, - **kwargs - ) + # See above. + marshal_sets = calculate_sets(srcs) + for (suffix, srcs) in marshal_sets.items(): + _go_test( + name = name + suffix + "_abi_autogen_test", + srcs = [name + suffix + "_abi_autogen_test.go"], + library = ":" + name + suffix, + deps = marshal_test_deps, + **kwargs + ) def proto_library(name, srcs, **kwargs): """Wraps the standard proto_library. diff --git a/tools/go_marshal/gomarshal/BUILD b/tools/go_marshal/gomarshal/BUILD index c92b59dd6..b5d5a4487 100644 --- a/tools/go_marshal/gomarshal/BUILD +++ b/tools/go_marshal/gomarshal/BUILD @@ -14,4 +14,5 @@ go_library( visibility = [ "//:sandbox", ], + deps = ["//tools/tags"], ) diff --git a/tools/go_marshal/gomarshal/generator.go b/tools/go_marshal/gomarshal/generator.go index af90bdecb..0b3f600fe 100644 --- a/tools/go_marshal/gomarshal/generator.go +++ b/tools/go_marshal/gomarshal/generator.go @@ -23,6 +23,9 @@ import ( "go/token" "os" "sort" + "strings" + + "gvisor.dev/gvisor/tools/tags" ) const ( @@ -104,6 +107,14 @@ func NewGenerator(srcs []string, out, outTest, pkg string, imports []string) (*G func (g *Generator) writeHeader() error { var b sourceBuffer b.emit("// Automatically generated marshal implementation. See tools/go_marshal.\n\n") + + // Emit build tags. + if t := tags.Aggregate(g.inputs); len(t) > 0 { + b.emit(strings.Join(t.Lines(), "\n")) + b.emit("\n") + } + + // Package header. b.emit("package %s\n\n", g.pkg) if err := b.write(g.output); err != nil { return err diff --git a/tools/go_stateify/BUILD b/tools/go_stateify/BUILD index a133d6f8b..6036faf7b 100644 --- a/tools/go_stateify/BUILD +++ b/tools/go_stateify/BUILD @@ -6,4 +6,5 @@ go_binary( name = "stateify", srcs = ["main.go"], visibility = ["//visibility:public"], + deps = ["//tools/tags"], ) diff --git a/tools/go_stateify/defs.bzl b/tools/go_stateify/defs.bzl index 0f261d89f..bdb966362 100644 --- a/tools/go_stateify/defs.bzl +++ b/tools/go_stateify/defs.bzl @@ -7,7 +7,6 @@ def _go_stateify_impl(ctx): # Run the stateify command. args = ["-output=%s" % output.path] args.append("-pkg=%s" % ctx.attr.package) - args.append("-arch=%s" % ctx.attr.arch) if ctx.attr._statepkg: args.append("-statepkg=%s" % ctx.attr._statepkg) if ctx.attr.imports: @@ -47,15 +46,8 @@ for statified types. doc = "The package name for the input sources.", mandatory = True, ), - "arch": attr.string( - doc = "Target platform.", - mandatory = True, - ), "out": attr.output( - doc = """ -The name of the generated file output. This must not conflict with any other -files and must be added to the srcs of the relevant go_library. -""", + doc = "Name of the generator output file.", mandatory = True, ), "_tool": attr.label( diff --git a/tools/go_stateify/main.go b/tools/go_stateify/main.go index 7d5d291e6..aa9d4543e 100644 --- a/tools/go_stateify/main.go +++ b/tools/go_stateify/main.go @@ -22,12 +22,12 @@ import ( "go/ast" "go/parser" "go/token" - "io/ioutil" "os" - "path/filepath" "reflect" "strings" "sync" + + "gvisor.dev/gvisor/tools/tags" ) var ( @@ -35,113 +35,8 @@ var ( imports = flag.String("imports", "", "extra imports for the output file") output = flag.String("output", "", "output file") statePkg = flag.String("statepkg", "", "state import package; defaults to empty") - arch = flag.String("arch", "", "specify the target platform") ) -// The known architectures. -var okgoarch = []string{ - "386", - "amd64", - "arm", - "arm64", - "mips", - "mipsle", - "mips64", - "mips64le", - "ppc64", - "ppc64le", - "riscv64", - "s390x", - "sparc64", - "wasm", -} - -// readfile returns the content of the named file. -func readfile(file string) string { - data, err := ioutil.ReadFile(file) - if err != nil { - panic(fmt.Sprintf("readfile err: %v", err)) - } - return string(data) -} - -// matchfield reports whether the field (x,y,z) matches this build. -// all the elements in the field must be satisfied. -func matchfield(f string, goarch string) bool { - for _, tag := range strings.Split(f, ",") { - if !matchtag(tag, goarch) { - return false - } - } - return true -} - -// matchtag reports whether the tag (x or !x) matches this build. -func matchtag(tag string, goarch string) bool { - if tag == "" { - return false - } - if tag[0] == '!' { - if len(tag) == 1 || tag[1] == '!' { - return false - } - return !matchtag(tag[1:], goarch) - } - return tag == goarch -} - -// canBuild reports whether we can build this file for target platform by -// checking file name and build tags. The code is derived from the Go source -// cmd.dist.build.shouldbuild. -func canBuild(file, goTargetArch string) bool { - name := filepath.Base(file) - excluded := func(list []string, ok string) bool { - for _, x := range list { - if x == ok || (ok == "android" && x == "linux") || (ok == "illumos" && x == "solaris") { - continue - } - i := strings.Index(name, x) - if i <= 0 || name[i-1] != '_' { - continue - } - i += len(x) - if i == len(name) || name[i] == '.' || name[i] == '_' { - return true - } - } - return false - } - if excluded(okgoarch, goTargetArch) { - return false - } - - // Check file contents for // +build lines. - for _, p := range strings.Split(readfile(file), "\n") { - p = strings.TrimSpace(p) - if p == "" { - continue - } - if !strings.HasPrefix(p, "//") { - break - } - if !strings.Contains(p, "+build") { - continue - } - fields := strings.Fields(p[2:]) - if len(fields) < 1 || fields[0] != "+build" { - continue - } - for _, p := range fields[1:] { - if matchfield(p, goTargetArch) { - goto fieldmatch - } - } - return false - fieldmatch: - } - return true -} - // resolveTypeName returns a qualified type name. func resolveTypeName(name string, typ ast.Expr) (field string, qualified string) { for done := false; !done; { @@ -329,8 +224,15 @@ func main() { fmt.Fprintf(outputFile, " m.Save(\"%s\", &x.%s)\n", name, name) } - // Emit the package name. + // Automated warning. fmt.Fprint(outputFile, "// automatically generated by stateify.\n\n") + + // Emit build tags. + if t := tags.Aggregate(flag.Args()); len(t) > 0 { + fmt.Fprintf(outputFile, "%s\n\n", strings.Join(t.Lines(), "\n")) + } + + // Emit the package name. fmt.Fprintf(outputFile, "package %s\n\n", *pkg) // Emit the imports lazily. @@ -364,10 +266,6 @@ func main() { os.Exit(1) } - if !canBuild(filename, *arch) { - continue - } - files = append(files, f) } diff --git a/tools/tags/BUILD b/tools/tags/BUILD new file mode 100644 index 000000000..1c02e2c89 --- /dev/null +++ b/tools/tags/BUILD @@ -0,0 +1,11 @@ +load("//tools:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "tags", + srcs = ["tags.go"], + marshal = False, + stateify = False, + visibility = ["//tools:__subpackages__"], +) diff --git a/tools/tags/tags.go b/tools/tags/tags.go new file mode 100644 index 000000000..f35904e0a --- /dev/null +++ b/tools/tags/tags.go @@ -0,0 +1,89 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package tags is a utility for parsing build tags. +package tags + +import ( + "fmt" + "io/ioutil" + "strings" +) + +// OrSet is a set of tags on a single line. +// +// Note that tags may include ",", and we don't distinguish this case in the +// logic below. Ideally, this constraints can be split into separate top-level +// build tags in order to resolve any issues. +type OrSet []string + +// Line returns the line for this or. +func (or OrSet) Line() string { + return fmt.Sprintf("// +build %s", strings.Join([]string(or), " ")) +} + +// AndSet is the set of all OrSets. +type AndSet []OrSet + +// Lines returns the lines to be printed. +func (and AndSet) Lines() (ls []string) { + for _, or := range and { + ls = append(ls, or.Line()) + } + return +} + +// Join joins this AndSet with another. +func (and AndSet) Join(other AndSet) AndSet { + return append(and, other...) +} + +// Tags returns the unique set of +build tags. +// +// Derived form the runtime's canBuild. +func Tags(file string) (tags AndSet) { + data, err := ioutil.ReadFile(file) + if err != nil { + return nil + } + // Check file contents for // +build lines. + for _, p := range strings.Split(string(data), "\n") { + p = strings.TrimSpace(p) + if p == "" { + continue + } + if !strings.HasPrefix(p, "//") { + break + } + if !strings.Contains(p, "+build") { + continue + } + fields := strings.Fields(p[2:]) + if len(fields) < 1 || fields[0] != "+build" { + continue + } + tags = append(tags, OrSet(fields[1:])) + } + return tags +} + +// Aggregate aggregates all tags from a set of files. +// +// Note that these may be in conflict, in which case the build will fail. +func Aggregate(files []string) (tags AndSet) { + for _, file := range files { + tags = tags.Join(Tags(file)) + } + return tags +} From 3f5642c5afdb8e633287ba10c2cb6b00f1849570 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Tue, 4 Feb 2020 15:15:43 -0800 Subject: [PATCH 24/29] Increase container_test size. container_test was flaking because a small percentage of runs timed out. Tested this fix with --runs_per_test=100. PiperOrigin-RevId: 293240102 --- runsc/container/BUILD | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/runsc/container/BUILD b/runsc/container/BUILD index e21431e4c..0aaeea3a8 100644 --- a/runsc/container/BUILD +++ b/runsc/container/BUILD @@ -30,7 +30,7 @@ go_library( go_test( name = "container_test", - size = "medium", + size = "large", srcs = [ "console_test.go", "container_test.go", From a26a954946ad2e7910d3ad7578960a93b73a1f9b Mon Sep 17 00:00:00 2001 From: Ian Gudger Date: Tue, 4 Feb 2020 15:20:30 -0800 Subject: [PATCH 25/29] Add socket connection stress test. Tests 65k connection attempts on common types of sockets to check for port leaks. Also fixes a bug where dual-stack sockets wouldn't properly re-queue segments received while closing. PiperOrigin-RevId: 293241166 --- pkg/tcpip/transport/tcp/connect.go | 4 + test/syscalls/BUILD | 9 ++ test/syscalls/linux/BUILD | 17 ++++ test/syscalls/linux/ip_socket_test_util.cc | 27 +++++ test/syscalls/linux/ip_socket_test_util.h | 15 +++ test/syscalls/linux/socket_generic_stress.cc | 83 +++++++++++++++ test/syscalls/linux/socket_test_util.cc | 101 +++++++++++++++++-- test/syscalls/linux/socket_test_util.h | 6 ++ 8 files changed, 251 insertions(+), 11 deletions(-) create mode 100644 test/syscalls/linux/socket_generic_stress.cc diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 9ff7ac261..5c5397823 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -989,6 +989,10 @@ func (e *endpoint) transitionToStateCloseLocked() { // to any other listening endpoint. We reply with RST if we cannot find one. func (e *endpoint) tryDeliverSegmentFromClosedEndpoint(s *segment) { ep := e.stack.FindTransportEndpoint(e.NetProto, e.TransProto, e.ID, &s.route) + if ep == nil && e.NetProto == header.IPv6ProtocolNumber && e.EndpointInfo.TransportEndpointInfo.ID.LocalAddress.To4() != "" { + // Dual-stack socket, try IPv4. + ep = e.stack.FindTransportEndpoint(header.IPv4ProtocolNumber, e.TransProto, e.ID, &s.route) + } if ep == nil { replyWithReset(s) s.decRef() diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 8f2b75a1c..31d239c0e 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -45,6 +45,15 @@ syscall_test(test = "//test/syscalls/linux:brk_test") syscall_test(test = "//test/syscalls/linux:socket_test") +syscall_test( + size = "large", + shard_count = 50, + # Takes too long for TSAN. Since this is kind of a stress test that doesn't + # involve much concurrency, TSAN's usefulness here is limited anyway. + tags = ["nogotsan"], + test = "//test/syscalls/linux:socket_stress_test", +) + syscall_test( add_overlay = True, test = "//test/syscalls/linux:chdir_test", diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 737e2329f..273b014d6 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -136,6 +136,7 @@ cc_library( "@com_google_absl//absl/strings", "@com_google_absl//absl/strings:str_format", "@com_google_absl//absl/time", + "@com_google_absl//absl/types:optional", "//test/util:file_descriptor", "//test/util:posix_error", "//test/util:temp_path", @@ -2151,6 +2152,22 @@ cc_library( alwayslink = 1, ) +cc_binary( + name = "socket_stress_test", + testonly = 1, + srcs = [ + "socket_generic_stress.cc", + ], + linkstatic = 1, + deps = [ + ":ip_socket_test_util", + ":socket_test_util", + gtest, + "//test/util:test_main", + "//test/util:test_util", + ], +) + cc_library( name = "socket_unix_dgram_test_cases", testonly = 1, diff --git a/test/syscalls/linux/ip_socket_test_util.cc b/test/syscalls/linux/ip_socket_test_util.cc index 6b472eb2f..bba022a41 100644 --- a/test/syscalls/linux/ip_socket_test_util.cc +++ b/test/syscalls/linux/ip_socket_test_util.cc @@ -79,6 +79,33 @@ SocketPairKind DualStackTCPAcceptBindSocketPair(int type) { /* dual_stack = */ true)}; } +SocketPairKind IPv6TCPAcceptBindPersistentListenerSocketPair(int type) { + std::string description = + absl::StrCat(DescribeSocketType(type), "connected IPv6 TCP socket"); + return SocketPairKind{description, AF_INET6, type | SOCK_STREAM, IPPROTO_TCP, + TCPAcceptBindPersistentListenerSocketPairCreator( + AF_INET6, type | SOCK_STREAM, 0, + /* dual_stack = */ false)}; +} + +SocketPairKind IPv4TCPAcceptBindPersistentListenerSocketPair(int type) { + std::string description = + absl::StrCat(DescribeSocketType(type), "connected IPv4 TCP socket"); + return SocketPairKind{description, AF_INET, type | SOCK_STREAM, IPPROTO_TCP, + TCPAcceptBindPersistentListenerSocketPairCreator( + AF_INET, type | SOCK_STREAM, 0, + /* dual_stack = */ false)}; +} + +SocketPairKind DualStackTCPAcceptBindPersistentListenerSocketPair(int type) { + std::string description = + absl::StrCat(DescribeSocketType(type), "connected dual stack TCP socket"); + return SocketPairKind{description, AF_INET6, type | SOCK_STREAM, IPPROTO_TCP, + TCPAcceptBindPersistentListenerSocketPairCreator( + AF_INET6, type | SOCK_STREAM, 0, + /* dual_stack = */ true)}; +} + SocketPairKind IPv6UDPBidirectionalBindSocketPair(int type) { std::string description = absl::StrCat(DescribeSocketType(type), "connected IPv6 UDP socket"); diff --git a/test/syscalls/linux/ip_socket_test_util.h b/test/syscalls/linux/ip_socket_test_util.h index 0f58e0f77..083ebbcf0 100644 --- a/test/syscalls/linux/ip_socket_test_util.h +++ b/test/syscalls/linux/ip_socket_test_util.h @@ -50,6 +50,21 @@ SocketPairKind IPv4TCPAcceptBindSocketPair(int type); // given type bound to the IPv4 loopback. SocketPairKind DualStackTCPAcceptBindSocketPair(int type); +// IPv6TCPAcceptBindPersistentListenerSocketPair is like +// IPv6TCPAcceptBindSocketPair except it uses a persistent listening socket to +// create all socket pairs. +SocketPairKind IPv6TCPAcceptBindPersistentListenerSocketPair(int type); + +// IPv4TCPAcceptBindPersistentListenerSocketPair is like +// IPv4TCPAcceptBindSocketPair except it uses a persistent listening socket to +// create all socket pairs. +SocketPairKind IPv4TCPAcceptBindPersistentListenerSocketPair(int type); + +// DualStackTCPAcceptBindPersistentListenerSocketPair is like +// DualStackTCPAcceptBindSocketPair except it uses a persistent listening socket +// to create all socket pairs. +SocketPairKind DualStackTCPAcceptBindPersistentListenerSocketPair(int type); + // IPv6UDPBidirectionalBindSocketPair returns a SocketPairKind that represents // SocketPairs created with bind() and connect() syscalls with AF_INET6 and the // given type bound to the IPv6 loopback. diff --git a/test/syscalls/linux/socket_generic_stress.cc b/test/syscalls/linux/socket_generic_stress.cc new file mode 100644 index 000000000..6a232238d --- /dev/null +++ b/test/syscalls/linux/socket_generic_stress.cc @@ -0,0 +1,83 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include +#include +#include +#include + +#include "gtest/gtest.h" +#include "test/syscalls/linux/ip_socket_test_util.h" +#include "test/syscalls/linux/socket_test_util.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +// Test fixture for tests that apply to pairs of connected sockets. +using ConnectStressTest = SocketPairTest; + +TEST_P(ConnectStressTest, Reset65kTimes) { + for (int i = 0; i < 1 << 16; ++i) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + // Send some data to ensure that the connection gets reset and the port gets + // released immediately. This avoids either end entering TIME-WAIT. + char sent_data[100] = {}; + ASSERT_THAT(write(sockets->first_fd(), sent_data, sizeof(sent_data)), + SyscallSucceedsWithValue(sizeof(sent_data))); + } +} + +INSTANTIATE_TEST_SUITE_P( + AllConnectedSockets, ConnectStressTest, + ::testing::Values(IPv6UDPBidirectionalBindSocketPair(0), + IPv4UDPBidirectionalBindSocketPair(0), + DualStackUDPBidirectionalBindSocketPair(0), + + // Without REUSEADDR, we get port exhaustion on Linux. + SetSockOpt(SOL_SOCKET, SO_REUSEADDR, + &kSockOptOn)(IPv6TCPAcceptBindSocketPair(0)), + SetSockOpt(SOL_SOCKET, SO_REUSEADDR, + &kSockOptOn)(IPv4TCPAcceptBindSocketPair(0)), + SetSockOpt(SOL_SOCKET, SO_REUSEADDR, &kSockOptOn)( + DualStackTCPAcceptBindSocketPair(0)))); + +// Test fixture for tests that apply to pairs of connected sockets created with +// a persistent listener (if applicable). +using PersistentListenerConnectStressTest = SocketPairTest; + +TEST_P(PersistentListenerConnectStressTest, 65kTimes) { + for (int i = 0; i < 1 << 16; ++i) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + } +} + +INSTANTIATE_TEST_SUITE_P( + AllConnectedSockets, PersistentListenerConnectStressTest, + ::testing::Values( + IPv6UDPBidirectionalBindSocketPair(0), + IPv4UDPBidirectionalBindSocketPair(0), + DualStackUDPBidirectionalBindSocketPair(0), + + // Without REUSEADDR, we get port exhaustion on Linux. + SetSockOpt(SOL_SOCKET, SO_REUSEADDR, &kSockOptOn)( + IPv6TCPAcceptBindPersistentListenerSocketPair(0)), + SetSockOpt(SOL_SOCKET, SO_REUSEADDR, &kSockOptOn)( + IPv4TCPAcceptBindPersistentListenerSocketPair(0)), + SetSockOpt(SOL_SOCKET, SO_REUSEADDR, &kSockOptOn)( + DualStackTCPAcceptBindPersistentListenerSocketPair(0)))); + +} // namespace testing +} // namespace gvisor diff --git a/test/syscalls/linux/socket_test_util.cc b/test/syscalls/linux/socket_test_util.cc index eff7d577e..c0c5ab3fe 100644 --- a/test/syscalls/linux/socket_test_util.cc +++ b/test/syscalls/linux/socket_test_util.cc @@ -18,10 +18,13 @@ #include #include +#include + #include "gtest/gtest.h" #include "absl/memory/memory.h" #include "absl/strings/str_cat.h" #include "absl/time/clock.h" +#include "absl/types/optional.h" #include "test/util/file_descriptor.h" #include "test/util/posix_error.h" #include "test/util/temp_path.h" @@ -109,7 +112,10 @@ Creator AcceptBindSocketPairCreator(bool abstract, int domain, MaybeSave(); // Unlinked path. } - return absl::make_unique(connected, accepted, bind_addr, + // accepted is before connected to destruct connected before accepted. + // Destructors for nonstatic member objects are called in the reverse order + // in which they appear in the class declaration. + return absl::make_unique(accepted, connected, bind_addr, extra_addr); }; } @@ -311,11 +317,16 @@ PosixErrorOr BindIP(int fd, bool dual_stack) { } template -PosixErrorOr> CreateTCPAcceptBindSocketPair( - int bound, int connected, int type, bool dual_stack) { - ASSIGN_OR_RETURN_ERRNO(T bind_addr, BindIP(bound, dual_stack)); - RETURN_ERROR_IF_SYSCALL_FAIL(listen(bound, /* backlog = */ 5)); +PosixErrorOr TCPBindAndListen(int fd, bool dual_stack) { + ASSIGN_OR_RETURN_ERRNO(T addr, BindIP(fd, dual_stack)); + RETURN_ERROR_IF_SYSCALL_FAIL(listen(fd, /* backlog = */ 5)); + return addr; +} +template +PosixErrorOr> +CreateTCPConnectAcceptSocketPair(int bound, int connected, int type, + bool dual_stack, T bind_addr) { int connect_result = 0; RETURN_ERROR_IF_SYSCALL_FAIL( (connect_result = RetryEINTR(connect)( @@ -358,16 +369,27 @@ PosixErrorOr> CreateTCPAcceptBindSocketPair( absl::SleepFor(absl::Seconds(1)); } - // Cleanup no longer needed resources. - RETURN_ERROR_IF_SYSCALL_FAIL(close(bound)); - MaybeSave(); // Successful close. - T extra_addr = {}; LocalhostAddr(&extra_addr, dual_stack); return absl::make_unique(connected, accepted, bind_addr, extra_addr); } +template +PosixErrorOr> CreateTCPAcceptBindSocketPair( + int bound, int connected, int type, bool dual_stack) { + ASSIGN_OR_RETURN_ERRNO(T bind_addr, TCPBindAndListen(bound, dual_stack)); + + auto result = CreateTCPConnectAcceptSocketPair(bound, connected, type, + dual_stack, bind_addr); + + // Cleanup no longer needed resources. + RETURN_ERROR_IF_SYSCALL_FAIL(close(bound)); + MaybeSave(); // Successful close. + + return result; +} + Creator TCPAcceptBindSocketPairCreator(int domain, int type, int protocol, bool dual_stack) { @@ -389,6 +411,63 @@ Creator TCPAcceptBindSocketPairCreator(int domain, int type, }; } +Creator TCPAcceptBindPersistentListenerSocketPairCreator( + int domain, int type, int protocol, bool dual_stack) { + // These are lazily initialized below, on the first call to the returned + // lambda. These values are private to each returned lambda, but shared across + // invocations of a specific lambda. + // + // The sharing allows pairs created with the same parameters to share a + // listener. This prevents future connects from failing if the connecting + // socket selects a port which had previously been used by a listening socket + // that still has some connections in TIME-WAIT. + // + // The lazy initialization is to avoid creating sockets during parameter + // enumeration. This is important because parameters are enumerated during the + // build process where networking may not be available. + auto listener = std::make_shared>(absl::optional()); + auto addr4 = std::make_shared>( + absl::optional()); + auto addr6 = std::make_shared>( + absl::optional()); + + return [=]() -> PosixErrorOr> { + int connected; + RETURN_ERROR_IF_SYSCALL_FAIL(connected = socket(domain, type, protocol)); + MaybeSave(); // Successful socket creation. + + // Share the listener across invocations. + if (!listener->has_value()) { + int fd = socket(domain, type, protocol); + if (fd < 0) { + return PosixError(errno, absl::StrCat("socket(", domain, ", ", type, + ", ", protocol, ")")); + } + listener->emplace(fd); + MaybeSave(); // Successful socket creation. + } + + // Bind the listener once, but create a new connect/accept pair each + // time. + if (domain == AF_INET) { + if (!addr4->has_value()) { + addr4->emplace( + TCPBindAndListen(listener->value(), dual_stack) + .ValueOrDie()); + } + return CreateTCPConnectAcceptSocketPair(listener->value(), connected, + type, dual_stack, addr4->value()); + } + if (!addr6->has_value()) { + addr6->emplace( + TCPBindAndListen(listener->value(), dual_stack) + .ValueOrDie()); + } + return CreateTCPConnectAcceptSocketPair(listener->value(), connected, type, + dual_stack, addr6->value()); + }; +} + template PosixErrorOr> CreateUDPBoundSocketPair( int sock1, int sock2, int type, bool dual_stack) { @@ -518,8 +597,8 @@ size_t CalculateUnixSockAddrLen(const char* sun_path) { if (sun_path[0] == 0) { return sizeof(sockaddr_un); } - // Filesystem addresses use the address length plus the 2 byte sun_family and - // null terminator. + // Filesystem addresses use the address length plus the 2 byte sun_family + // and null terminator. return strlen(sun_path) + 3; } diff --git a/test/syscalls/linux/socket_test_util.h b/test/syscalls/linux/socket_test_util.h index 2dbb8bed3..bfaa6e397 100644 --- a/test/syscalls/linux/socket_test_util.h +++ b/test/syscalls/linux/socket_test_util.h @@ -273,6 +273,12 @@ Creator TCPAcceptBindSocketPairCreator(int domain, int type, int protocol, bool dual_stack); +// TCPAcceptBindPersistentListenerSocketPairCreator is like +// TCPAcceptBindSocketPairCreator, except it uses the same listening socket to +// create all SocketPairs. +Creator TCPAcceptBindPersistentListenerSocketPairCreator( + int domain, int type, int protocol, bool dual_stack); + // UDPBidirectionalBindSocketPairCreator returns a Creator that // obtains file descriptors by invoking the bind() and connect() syscalls on UDP // sockets. From 665b614e4a6e715bac25bea15c5c29184016e549 Mon Sep 17 00:00:00 2001 From: Ting-Yu Wang Date: Tue, 4 Feb 2020 18:04:26 -0800 Subject: [PATCH 26/29] Support RTM_NEWADDR and RTM_GETLINK in (rt)netlink. PiperOrigin-RevId: 293271055 --- pkg/sentry/inet/inet.go | 4 + pkg/sentry/inet/test_stack.go | 6 + pkg/sentry/socket/hostinet/stack.go | 5 + pkg/sentry/socket/netlink/BUILD | 14 +- pkg/sentry/socket/netlink/message.go | 129 ++++++++ pkg/sentry/socket/netlink/message_test.go | 312 +++++++++++++++++++ pkg/sentry/socket/netlink/provider.go | 2 +- pkg/sentry/socket/netlink/route/BUILD | 2 - pkg/sentry/socket/netlink/route/protocol.go | 248 ++++++++++----- pkg/sentry/socket/netlink/socket.go | 54 ++-- pkg/sentry/socket/netlink/uevent/protocol.go | 2 +- pkg/sentry/socket/netstack/stack.go | 55 ++++ pkg/tcpip/stack/stack.go | 9 + test/syscalls/linux/BUILD | 2 + test/syscalls/linux/socket_netlink_route.cc | 304 ++++++++++++++---- test/syscalls/linux/socket_netlink_util.cc | 45 ++- test/syscalls/linux/socket_netlink_util.h | 9 + 17 files changed, 1031 insertions(+), 171 deletions(-) create mode 100644 pkg/sentry/socket/netlink/message_test.go diff --git a/pkg/sentry/inet/inet.go b/pkg/sentry/inet/inet.go index a7dfb78a7..2916a0644 100644 --- a/pkg/sentry/inet/inet.go +++ b/pkg/sentry/inet/inet.go @@ -28,6 +28,10 @@ type Stack interface { // interface indexes to a slice of associated interface address properties. InterfaceAddrs() map[int32][]InterfaceAddr + // AddInterfaceAddr adds an address to the network interface identified by + // index. + AddInterfaceAddr(idx int32, addr InterfaceAddr) error + // SupportsIPv6 returns true if the stack supports IPv6 connectivity. SupportsIPv6() bool diff --git a/pkg/sentry/inet/test_stack.go b/pkg/sentry/inet/test_stack.go index dcfcbd97e..d8961fc94 100644 --- a/pkg/sentry/inet/test_stack.go +++ b/pkg/sentry/inet/test_stack.go @@ -47,6 +47,12 @@ func (s *TestStack) InterfaceAddrs() map[int32][]InterfaceAddr { return s.InterfaceAddrsMap } +// AddInterfaceAddr implements Stack.AddInterfaceAddr. +func (s *TestStack) AddInterfaceAddr(idx int32, addr InterfaceAddr) error { + s.InterfaceAddrsMap[idx] = append(s.InterfaceAddrsMap[idx], addr) + return nil +} + // SupportsIPv6 implements Stack.SupportsIPv6. func (s *TestStack) SupportsIPv6() bool { return s.SupportsIPv6Flag diff --git a/pkg/sentry/socket/hostinet/stack.go b/pkg/sentry/socket/hostinet/stack.go index 034eca676..a48082631 100644 --- a/pkg/sentry/socket/hostinet/stack.go +++ b/pkg/sentry/socket/hostinet/stack.go @@ -310,6 +310,11 @@ func (s *Stack) InterfaceAddrs() map[int32][]inet.InterfaceAddr { return addrs } +// AddInterfaceAddr implements inet.Stack.AddInterfaceAddr. +func (s *Stack) AddInterfaceAddr(idx int32, addr inet.InterfaceAddr) error { + return syserror.EACCES +} + // SupportsIPv6 implements inet.Stack.SupportsIPv6. func (s *Stack) SupportsIPv6() bool { return s.supportsIPv6 diff --git a/pkg/sentry/socket/netlink/BUILD b/pkg/sentry/socket/netlink/BUILD index f8b8e467d..1911cd9b8 100644 --- a/pkg/sentry/socket/netlink/BUILD +++ b/pkg/sentry/socket/netlink/BUILD @@ -1,4 +1,4 @@ -load("//tools:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -33,3 +33,15 @@ go_library( "//pkg/waiter", ], ) + +go_test( + name = "netlink_test", + size = "small", + srcs = [ + "message_test.go", + ], + deps = [ + ":netlink", + "//pkg/abi/linux", + ], +) diff --git a/pkg/sentry/socket/netlink/message.go b/pkg/sentry/socket/netlink/message.go index b21e0ca4b..4ea252ccb 100644 --- a/pkg/sentry/socket/netlink/message.go +++ b/pkg/sentry/socket/netlink/message.go @@ -30,8 +30,16 @@ func alignUp(length int, align uint) int { return (length + int(align) - 1) &^ (int(align) - 1) } +// alignPad returns the length of padding required for alignment. +// +// Preconditions: align is a power of two. +func alignPad(length int, align uint) int { + return alignUp(length, align) - length +} + // Message contains a complete serialized netlink message. type Message struct { + hdr linux.NetlinkMessageHeader buf []byte } @@ -40,10 +48,86 @@ type Message struct { // The header length will be updated by Finalize. func NewMessage(hdr linux.NetlinkMessageHeader) *Message { return &Message{ + hdr: hdr, buf: binary.Marshal(nil, usermem.ByteOrder, hdr), } } +// ParseMessage parses the first message seen at buf, returning the rest of the +// buffer. If message is malformed, ok of false is returned. For last message, +// padding check is loose, if there isn't enought padding, whole buf is consumed +// and ok is set to true. +func ParseMessage(buf []byte) (msg *Message, rest []byte, ok bool) { + b := BytesView(buf) + + hdrBytes, ok := b.Extract(linux.NetlinkMessageHeaderSize) + if !ok { + return + } + var hdr linux.NetlinkMessageHeader + binary.Unmarshal(hdrBytes, usermem.ByteOrder, &hdr) + + // Msg portion. + totalMsgLen := int(hdr.Length) + _, ok = b.Extract(totalMsgLen - linux.NetlinkMessageHeaderSize) + if !ok { + return + } + + // Padding. + numPad := alignPad(totalMsgLen, linux.NLMSG_ALIGNTO) + // Linux permits the last message not being aligned, just consume all of it. + // Ref: net/netlink/af_netlink.c:netlink_rcv_skb + if numPad > len(b) { + numPad = len(b) + } + _, ok = b.Extract(numPad) + if !ok { + return + } + + return &Message{ + hdr: hdr, + buf: buf[:totalMsgLen], + }, []byte(b), true +} + +// Header returns the header of this message. +func (m *Message) Header() linux.NetlinkMessageHeader { + return m.hdr +} + +// GetData unmarshals the payload message header from this netlink message, and +// returns the attributes portion. +func (m *Message) GetData(msg interface{}) (AttrsView, bool) { + b := BytesView(m.buf) + + _, ok := b.Extract(linux.NetlinkMessageHeaderSize) + if !ok { + return nil, false + } + + size := int(binary.Size(msg)) + msgBytes, ok := b.Extract(size) + if !ok { + return nil, false + } + binary.Unmarshal(msgBytes, usermem.ByteOrder, msg) + + numPad := alignPad(linux.NetlinkMessageHeaderSize+size, linux.NLMSG_ALIGNTO) + // Linux permits the last message not being aligned, just consume all of it. + // Ref: net/netlink/af_netlink.c:netlink_rcv_skb + if numPad > len(b) { + numPad = len(b) + } + _, ok = b.Extract(numPad) + if !ok { + return nil, false + } + + return AttrsView(b), true +} + // Finalize returns the []byte containing the entire message, with the total // length set in the message header. The Message must not be modified after // calling Finalize. @@ -157,3 +241,48 @@ func (ms *MessageSet) AddMessage(hdr linux.NetlinkMessageHeader) *Message { ms.Messages = append(ms.Messages, m) return m } + +// AttrsView is a view into the attributes portion of a netlink message. +type AttrsView []byte + +// Empty returns whether there is no attribute left in v. +func (v AttrsView) Empty() bool { + return len(v) == 0 +} + +// ParseFirst parses first netlink attribute at the beginning of v. +func (v AttrsView) ParseFirst() (hdr linux.NetlinkAttrHeader, value []byte, rest AttrsView, ok bool) { + b := BytesView(v) + + hdrBytes, ok := b.Extract(linux.NetlinkAttrHeaderSize) + if !ok { + return + } + binary.Unmarshal(hdrBytes, usermem.ByteOrder, &hdr) + + value, ok = b.Extract(int(hdr.Length) - linux.NetlinkAttrHeaderSize) + if !ok { + return + } + + _, ok = b.Extract(alignPad(int(hdr.Length), linux.NLA_ALIGNTO)) + if !ok { + return + } + + return hdr, value, AttrsView(b), ok +} + +// BytesView supports extracting data from a byte slice with bounds checking. +type BytesView []byte + +// Extract removes the first n bytes from v and returns it. If n is out of +// bounds, it returns false. +func (v *BytesView) Extract(n int) ([]byte, bool) { + if n < 0 || n > len(*v) { + return nil, false + } + extracted := (*v)[:n] + *v = (*v)[n:] + return extracted, true +} diff --git a/pkg/sentry/socket/netlink/message_test.go b/pkg/sentry/socket/netlink/message_test.go new file mode 100644 index 000000000..ef13d9386 --- /dev/null +++ b/pkg/sentry/socket/netlink/message_test.go @@ -0,0 +1,312 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package message_test + +import ( + "bytes" + "reflect" + "testing" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/socket/netlink" +) + +type dummyNetlinkMsg struct { + Foo uint16 +} + +func TestParseMessage(t *testing.T) { + tests := []struct { + desc string + input []byte + + header linux.NetlinkMessageHeader + dataMsg *dummyNetlinkMsg + restLen int + ok bool + }{ + { + desc: "valid", + input: []byte{ + 0x14, 0x00, 0x00, 0x00, // Length + 0x01, 0x00, // Type + 0x02, 0x00, // Flags + 0x03, 0x00, 0x00, 0x00, // Seq + 0x04, 0x00, 0x00, 0x00, // PortID + 0x30, 0x31, 0x00, 0x00, // Data message with 2 bytes padding + }, + header: linux.NetlinkMessageHeader{ + Length: 20, + Type: 1, + Flags: 2, + Seq: 3, + PortID: 4, + }, + dataMsg: &dummyNetlinkMsg{ + Foo: 0x3130, + }, + restLen: 0, + ok: true, + }, + { + desc: "valid with next message", + input: []byte{ + 0x14, 0x00, 0x00, 0x00, // Length + 0x01, 0x00, // Type + 0x02, 0x00, // Flags + 0x03, 0x00, 0x00, 0x00, // Seq + 0x04, 0x00, 0x00, 0x00, // PortID + 0x30, 0x31, 0x00, 0x00, // Data message with 2 bytes padding + 0xFF, // Next message (rest) + }, + header: linux.NetlinkMessageHeader{ + Length: 20, + Type: 1, + Flags: 2, + Seq: 3, + PortID: 4, + }, + dataMsg: &dummyNetlinkMsg{ + Foo: 0x3130, + }, + restLen: 1, + ok: true, + }, + { + desc: "valid for last message without padding", + input: []byte{ + 0x12, 0x00, 0x00, 0x00, // Length + 0x01, 0x00, // Type + 0x02, 0x00, // Flags + 0x03, 0x00, 0x00, 0x00, // Seq + 0x04, 0x00, 0x00, 0x00, // PortID + 0x30, 0x31, // Data message + }, + header: linux.NetlinkMessageHeader{ + Length: 18, + Type: 1, + Flags: 2, + Seq: 3, + PortID: 4, + }, + dataMsg: &dummyNetlinkMsg{ + Foo: 0x3130, + }, + restLen: 0, + ok: true, + }, + { + desc: "valid for last message not to be aligned", + input: []byte{ + 0x13, 0x00, 0x00, 0x00, // Length + 0x01, 0x00, // Type + 0x02, 0x00, // Flags + 0x03, 0x00, 0x00, 0x00, // Seq + 0x04, 0x00, 0x00, 0x00, // PortID + 0x30, 0x31, // Data message + 0x00, // Excessive 1 byte permitted at end + }, + header: linux.NetlinkMessageHeader{ + Length: 19, + Type: 1, + Flags: 2, + Seq: 3, + PortID: 4, + }, + dataMsg: &dummyNetlinkMsg{ + Foo: 0x3130, + }, + restLen: 0, + ok: true, + }, + { + desc: "header.Length too short", + input: []byte{ + 0x04, 0x00, 0x00, 0x00, // Length + 0x01, 0x00, // Type + 0x02, 0x00, // Flags + 0x03, 0x00, 0x00, 0x00, // Seq + 0x04, 0x00, 0x00, 0x00, // PortID + 0x30, 0x31, 0x00, 0x00, // Data message with 2 bytes padding + }, + ok: false, + }, + { + desc: "header.Length too long", + input: []byte{ + 0xFF, 0xFF, 0x00, 0x00, // Length + 0x01, 0x00, // Type + 0x02, 0x00, // Flags + 0x03, 0x00, 0x00, 0x00, // Seq + 0x04, 0x00, 0x00, 0x00, // PortID + 0x30, 0x31, 0x00, 0x00, // Data message with 2 bytes padding + }, + ok: false, + }, + { + desc: "header incomplete", + input: []byte{ + 0x04, 0x00, 0x00, 0x00, // Length + }, + ok: false, + }, + { + desc: "empty message", + input: []byte{}, + ok: false, + }, + } + for _, test := range tests { + msg, rest, ok := netlink.ParseMessage(test.input) + if ok != test.ok { + t.Errorf("%v: got ok = %v, want = %v", test.desc, ok, test.ok) + continue + } + if !test.ok { + continue + } + if !reflect.DeepEqual(msg.Header(), test.header) { + t.Errorf("%v: got hdr = %+v, want = %+v", test.desc, msg.Header(), test.header) + } + + dataMsg := &dummyNetlinkMsg{} + _, dataOk := msg.GetData(dataMsg) + if !dataOk { + t.Errorf("%v: GetData.ok = %v, want = true", test.desc, dataOk) + } else if !reflect.DeepEqual(dataMsg, test.dataMsg) { + t.Errorf("%v: GetData.msg = %+v, want = %+v", test.desc, dataMsg, test.dataMsg) + } + + if got, want := rest, test.input[len(test.input)-test.restLen:]; !bytes.Equal(got, want) { + t.Errorf("%v: got rest = %v, want = %v", test.desc, got, want) + } + } +} + +func TestAttrView(t *testing.T) { + tests := []struct { + desc string + input []byte + + // Outputs for ParseFirst. + hdr linux.NetlinkAttrHeader + value []byte + restLen int + ok bool + + // Outputs for Empty. + isEmpty bool + }{ + { + desc: "valid", + input: []byte{ + 0x06, 0x00, // Length + 0x01, 0x00, // Type + 0x30, 0x31, 0x00, 0x00, // Data with 2 bytes padding + }, + hdr: linux.NetlinkAttrHeader{ + Length: 6, + Type: 1, + }, + value: []byte{0x30, 0x31}, + restLen: 0, + ok: true, + isEmpty: false, + }, + { + desc: "at alignment", + input: []byte{ + 0x08, 0x00, // Length + 0x01, 0x00, // Type + 0x30, 0x31, 0x32, 0x33, // Data + }, + hdr: linux.NetlinkAttrHeader{ + Length: 8, + Type: 1, + }, + value: []byte{0x30, 0x31, 0x32, 0x33}, + restLen: 0, + ok: true, + isEmpty: false, + }, + { + desc: "at alignment with rest data", + input: []byte{ + 0x08, 0x00, // Length + 0x01, 0x00, // Type + 0x30, 0x31, 0x32, 0x33, // Data + 0xFF, 0xFE, // Rest data + }, + hdr: linux.NetlinkAttrHeader{ + Length: 8, + Type: 1, + }, + value: []byte{0x30, 0x31, 0x32, 0x33}, + restLen: 2, + ok: true, + isEmpty: false, + }, + { + desc: "hdr.Length too long", + input: []byte{ + 0xFF, 0x00, // Length + 0x01, 0x00, // Type + 0x30, 0x31, 0x32, 0x33, // Data + }, + ok: false, + isEmpty: false, + }, + { + desc: "hdr.Length too short", + input: []byte{ + 0x01, 0x00, // Length + 0x01, 0x00, // Type + 0x30, 0x31, 0x32, 0x33, // Data + }, + ok: false, + isEmpty: false, + }, + { + desc: "empty", + input: []byte{}, + ok: false, + isEmpty: true, + }, + } + for _, test := range tests { + attrs := netlink.AttrsView(test.input) + + // Test ParseFirst(). + hdr, value, rest, ok := attrs.ParseFirst() + if ok != test.ok { + t.Errorf("%v: got ok = %v, want = %v", test.desc, ok, test.ok) + } else if test.ok { + if !reflect.DeepEqual(hdr, test.hdr) { + t.Errorf("%v: got hdr = %+v, want = %+v", test.desc, hdr, test.hdr) + } + if !bytes.Equal(value, test.value) { + t.Errorf("%v: got value = %v, want = %v", test.desc, value, test.value) + } + if wantRest := test.input[len(test.input)-test.restLen:]; !bytes.Equal(rest, wantRest) { + t.Errorf("%v: got rest = %v, want = %v", test.desc, rest, wantRest) + } + } + + // Test Empty(). + if got, want := attrs.Empty(), test.isEmpty; got != want { + t.Errorf("%v: got empty = %v, want = %v", test.desc, got, want) + } + } +} diff --git a/pkg/sentry/socket/netlink/provider.go b/pkg/sentry/socket/netlink/provider.go index 07f860a49..b0dc70e5c 100644 --- a/pkg/sentry/socket/netlink/provider.go +++ b/pkg/sentry/socket/netlink/provider.go @@ -42,7 +42,7 @@ type Protocol interface { // If err == nil, any messages added to ms will be sent back to the // other end of the socket. Setting ms.Multi will cause an NLMSG_DONE // message to be sent even if ms contains no messages. - ProcessMessage(ctx context.Context, hdr linux.NetlinkMessageHeader, data []byte, ms *MessageSet) *syserr.Error + ProcessMessage(ctx context.Context, msg *Message, ms *MessageSet) *syserr.Error } // Provider is a function that creates a new Protocol for a specific netlink diff --git a/pkg/sentry/socket/netlink/route/BUILD b/pkg/sentry/socket/netlink/route/BUILD index 622a1eafc..93127398d 100644 --- a/pkg/sentry/socket/netlink/route/BUILD +++ b/pkg/sentry/socket/netlink/route/BUILD @@ -10,13 +10,11 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", - "//pkg/binary", "//pkg/context", "//pkg/sentry/inet", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/socket/netlink", "//pkg/syserr", - "//pkg/usermem", ], ) diff --git a/pkg/sentry/socket/netlink/route/protocol.go b/pkg/sentry/socket/netlink/route/protocol.go index 2b3c7f5b3..c84d8bd7c 100644 --- a/pkg/sentry/socket/netlink/route/protocol.go +++ b/pkg/sentry/socket/netlink/route/protocol.go @@ -17,16 +17,15 @@ package route import ( "bytes" + "syscall" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/socket/netlink" "gvisor.dev/gvisor/pkg/syserr" - "gvisor.dev/gvisor/pkg/usermem" ) // commandKind describes the operational class of a message type. @@ -69,13 +68,7 @@ func (p *Protocol) CanSend() bool { } // dumpLinks handles RTM_GETLINK dump requests. -func (p *Protocol) dumpLinks(ctx context.Context, hdr linux.NetlinkMessageHeader, data []byte, ms *netlink.MessageSet) *syserr.Error { - // TODO(b/68878065): Only the dump variant of the types below are - // supported. - if hdr.Flags&linux.NLM_F_DUMP != linux.NLM_F_DUMP { - return syserr.ErrNotSupported - } - +func (p *Protocol) dumpLinks(ctx context.Context, msg *netlink.Message, ms *netlink.MessageSet) *syserr.Error { // NLM_F_DUMP + RTM_GETLINK messages are supposed to include an // ifinfomsg. However, Linux <3.9 only checked for rtgenmsg, and some // userspace applications (including glibc) still include rtgenmsg. @@ -99,44 +92,105 @@ func (p *Protocol) dumpLinks(ctx context.Context, hdr linux.NetlinkMessageHeader return nil } - for id, i := range stack.Interfaces() { - m := ms.AddMessage(linux.NetlinkMessageHeader{ - Type: linux.RTM_NEWLINK, - }) - - m.Put(linux.InterfaceInfoMessage{ - Family: linux.AF_UNSPEC, - Type: i.DeviceType, - Index: id, - Flags: i.Flags, - }) - - m.PutAttrString(linux.IFLA_IFNAME, i.Name) - m.PutAttr(linux.IFLA_MTU, i.MTU) - - mac := make([]byte, 6) - brd := mac - if len(i.Addr) > 0 { - mac = i.Addr - brd = bytes.Repeat([]byte{0xff}, len(i.Addr)) - } - m.PutAttr(linux.IFLA_ADDRESS, mac) - m.PutAttr(linux.IFLA_BROADCAST, brd) - - // TODO(gvisor.dev/issue/578): There are many more attributes. + for idx, i := range stack.Interfaces() { + addNewLinkMessage(ms, idx, i) } return nil } -// dumpAddrs handles RTM_GETADDR dump requests. -func (p *Protocol) dumpAddrs(ctx context.Context, hdr linux.NetlinkMessageHeader, data []byte, ms *netlink.MessageSet) *syserr.Error { - // TODO(b/68878065): Only the dump variant of the types below are - // supported. - if hdr.Flags&linux.NLM_F_DUMP != linux.NLM_F_DUMP { - return syserr.ErrNotSupported +// getLinks handles RTM_GETLINK requests. +func (p *Protocol) getLink(ctx context.Context, msg *netlink.Message, ms *netlink.MessageSet) *syserr.Error { + stack := inet.StackFromContext(ctx) + if stack == nil { + // No network devices. + return nil } + // Parse message. + var ifi linux.InterfaceInfoMessage + attrs, ok := msg.GetData(&ifi) + if !ok { + return syserr.ErrInvalidArgument + } + + // Parse attributes. + var byName []byte + for !attrs.Empty() { + ahdr, value, rest, ok := attrs.ParseFirst() + if !ok { + return syserr.ErrInvalidArgument + } + attrs = rest + + switch ahdr.Type { + case linux.IFLA_IFNAME: + if len(value) < 1 { + return syserr.ErrInvalidArgument + } + byName = value[:len(value)-1] + + // TODO(gvisor.dev/issue/578): Support IFLA_EXT_MASK. + } + } + + found := false + for idx, i := range stack.Interfaces() { + switch { + case ifi.Index > 0: + if idx != ifi.Index { + continue + } + case byName != nil: + if string(byName) != i.Name { + continue + } + default: + // Criteria not specified. + return syserr.ErrInvalidArgument + } + + addNewLinkMessage(ms, idx, i) + found = true + break + } + if !found { + return syserr.ErrNoDevice + } + return nil +} + +// addNewLinkMessage appends RTM_NEWLINK message for the given interface into +// the message set. +func addNewLinkMessage(ms *netlink.MessageSet, idx int32, i inet.Interface) { + m := ms.AddMessage(linux.NetlinkMessageHeader{ + Type: linux.RTM_NEWLINK, + }) + + m.Put(linux.InterfaceInfoMessage{ + Family: linux.AF_UNSPEC, + Type: i.DeviceType, + Index: idx, + Flags: i.Flags, + }) + + m.PutAttrString(linux.IFLA_IFNAME, i.Name) + m.PutAttr(linux.IFLA_MTU, i.MTU) + + mac := make([]byte, 6) + brd := mac + if len(i.Addr) > 0 { + mac = i.Addr + brd = bytes.Repeat([]byte{0xff}, len(i.Addr)) + } + m.PutAttr(linux.IFLA_ADDRESS, mac) + m.PutAttr(linux.IFLA_BROADCAST, brd) + + // TODO(gvisor.dev/issue/578): There are many more attributes. +} + +// dumpAddrs handles RTM_GETADDR dump requests. +func (p *Protocol) dumpAddrs(ctx context.Context, msg *netlink.Message, ms *netlink.MessageSet) *syserr.Error { // RTM_GETADDR dump requests need not contain anything more than the // netlink header and 1 byte protocol family common to all // NETLINK_ROUTE requests. @@ -168,6 +222,7 @@ func (p *Protocol) dumpAddrs(ctx context.Context, hdr linux.NetlinkMessageHeader Index: uint32(id), }) + m.PutAttr(linux.IFA_LOCAL, []byte(a.Addr)) m.PutAttr(linux.IFA_ADDRESS, []byte(a.Addr)) // TODO(gvisor.dev/issue/578): There are many more attributes. @@ -252,12 +307,12 @@ func fillRoute(routes []inet.Route, addr []byte) (inet.Route, *syserr.Error) { } // parseForDestination parses a message as format of RouteMessage-RtAttr-dst. -func parseForDestination(data []byte) ([]byte, *syserr.Error) { +func parseForDestination(msg *netlink.Message) ([]byte, *syserr.Error) { var rtMsg linux.RouteMessage - if len(data) < linux.SizeOfRouteMessage { + attrs, ok := msg.GetData(&rtMsg) + if !ok { return nil, syserr.ErrInvalidArgument } - binary.Unmarshal(data[:linux.SizeOfRouteMessage], usermem.ByteOrder, &rtMsg) // iproute2 added the RTM_F_LOOKUP_TABLE flag in version v4.4.0. See // commit bc234301af12. Note we don't check this flag for backward // compatibility. @@ -265,26 +320,15 @@ func parseForDestination(data []byte) ([]byte, *syserr.Error) { return nil, syserr.ErrNotSupported } - data = data[linux.SizeOfRouteMessage:] - - // TODO(gvisor.dev/issue/1611): Add generic attribute parsing. - var rtAttr linux.RtAttr - if len(data) < linux.SizeOfRtAttr { - return nil, syserr.ErrInvalidArgument + // Expect first attribute is RTA_DST. + if hdr, value, _, ok := attrs.ParseFirst(); ok && hdr.Type == linux.RTA_DST { + return value, nil } - binary.Unmarshal(data[:linux.SizeOfRtAttr], usermem.ByteOrder, &rtAttr) - if rtAttr.Type != linux.RTA_DST { - return nil, syserr.ErrInvalidArgument - } - - if len(data) < int(rtAttr.Len) { - return nil, syserr.ErrInvalidArgument - } - return data[linux.SizeOfRtAttr:rtAttr.Len], nil + return nil, syserr.ErrInvalidArgument } // dumpRoutes handles RTM_GETROUTE requests. -func (p *Protocol) dumpRoutes(ctx context.Context, hdr linux.NetlinkMessageHeader, data []byte, ms *netlink.MessageSet) *syserr.Error { +func (p *Protocol) dumpRoutes(ctx context.Context, msg *netlink.Message, ms *netlink.MessageSet) *syserr.Error { // RTM_GETROUTE dump requests need not contain anything more than the // netlink header and 1 byte protocol family common to all // NETLINK_ROUTE requests. @@ -295,10 +339,11 @@ func (p *Protocol) dumpRoutes(ctx context.Context, hdr linux.NetlinkMessageHeade return nil } + hdr := msg.Header() routeTables := stack.RouteTable() if hdr.Flags == linux.NLM_F_REQUEST { - dst, err := parseForDestination(data) + dst, err := parseForDestination(msg) if err != nil { return err } @@ -357,10 +402,55 @@ func (p *Protocol) dumpRoutes(ctx context.Context, hdr linux.NetlinkMessageHeade return nil } +// newAddr handles RTM_NEWADDR requests. +func (p *Protocol) newAddr(ctx context.Context, msg *netlink.Message, ms *netlink.MessageSet) *syserr.Error { + stack := inet.StackFromContext(ctx) + if stack == nil { + // No network stack. + return syserr.ErrProtocolNotSupported + } + + var ifa linux.InterfaceAddrMessage + attrs, ok := msg.GetData(&ifa) + if !ok { + return syserr.ErrInvalidArgument + } + + for !attrs.Empty() { + ahdr, value, rest, ok := attrs.ParseFirst() + if !ok { + return syserr.ErrInvalidArgument + } + attrs = rest + + switch ahdr.Type { + case linux.IFA_LOCAL: + err := stack.AddInterfaceAddr(int32(ifa.Index), inet.InterfaceAddr{ + Family: ifa.Family, + PrefixLen: ifa.PrefixLen, + Flags: ifa.Flags, + Addr: value, + }) + if err == syscall.EEXIST { + flags := msg.Header().Flags + if flags&linux.NLM_F_EXCL != 0 { + return syserr.ErrExists + } + } else if err != nil { + return syserr.ErrInvalidArgument + } + } + } + return nil +} + // ProcessMessage implements netlink.Protocol.ProcessMessage. -func (p *Protocol) ProcessMessage(ctx context.Context, hdr linux.NetlinkMessageHeader, data []byte, ms *netlink.MessageSet) *syserr.Error { +func (p *Protocol) ProcessMessage(ctx context.Context, msg *netlink.Message, ms *netlink.MessageSet) *syserr.Error { + hdr := msg.Header() + // All messages start with a 1 byte protocol family. - if len(data) < 1 { + var family uint8 + if _, ok := msg.GetData(&family); !ok { // Linux ignores messages missing the protocol family. See // net/core/rtnetlink.c:rtnetlink_rcv_msg. return nil @@ -374,16 +464,32 @@ func (p *Protocol) ProcessMessage(ctx context.Context, hdr linux.NetlinkMessageH } } - switch hdr.Type { - case linux.RTM_GETLINK: - return p.dumpLinks(ctx, hdr, data, ms) - case linux.RTM_GETADDR: - return p.dumpAddrs(ctx, hdr, data, ms) - case linux.RTM_GETROUTE: - return p.dumpRoutes(ctx, hdr, data, ms) - default: - return syserr.ErrNotSupported + if hdr.Flags&linux.NLM_F_DUMP == linux.NLM_F_DUMP { + // TODO(b/68878065): Only the dump variant of the types below are + // supported. + switch hdr.Type { + case linux.RTM_GETLINK: + return p.dumpLinks(ctx, msg, ms) + case linux.RTM_GETADDR: + return p.dumpAddrs(ctx, msg, ms) + case linux.RTM_GETROUTE: + return p.dumpRoutes(ctx, msg, ms) + default: + return syserr.ErrNotSupported + } + } else if hdr.Flags&linux.NLM_F_REQUEST == linux.NLM_F_REQUEST { + switch hdr.Type { + case linux.RTM_GETLINK: + return p.getLink(ctx, msg, ms) + case linux.RTM_GETROUTE: + return p.dumpRoutes(ctx, msg, ms) + case linux.RTM_NEWADDR: + return p.newAddr(ctx, msg, ms) + default: + return syserr.ErrNotSupported + } } + return syserr.ErrNotSupported } // init registers the NETLINK_ROUTE provider. diff --git a/pkg/sentry/socket/netlink/socket.go b/pkg/sentry/socket/netlink/socket.go index c4b95debb..2ca02567d 100644 --- a/pkg/sentry/socket/netlink/socket.go +++ b/pkg/sentry/socket/netlink/socket.go @@ -644,47 +644,38 @@ func (s *Socket) sendResponse(ctx context.Context, ms *MessageSet) *syserr.Error return nil } -func (s *Socket) dumpErrorMesage(ctx context.Context, hdr linux.NetlinkMessageHeader, ms *MessageSet, err *syserr.Error) *syserr.Error { +func dumpErrorMesage(hdr linux.NetlinkMessageHeader, ms *MessageSet, err *syserr.Error) { m := ms.AddMessage(linux.NetlinkMessageHeader{ Type: linux.NLMSG_ERROR, }) - m.Put(linux.NetlinkErrorMessage{ Error: int32(-err.ToLinux().Number()), Header: hdr, }) - return nil +} +func dumpAckMesage(hdr linux.NetlinkMessageHeader, ms *MessageSet) { + m := ms.AddMessage(linux.NetlinkMessageHeader{ + Type: linux.NLMSG_ERROR, + }) + m.Put(linux.NetlinkErrorMessage{ + Error: 0, + Header: hdr, + }) } // processMessages handles each message in buf, passing it to the protocol // handler for final handling. func (s *Socket) processMessages(ctx context.Context, buf []byte) *syserr.Error { for len(buf) > 0 { - if len(buf) < linux.NetlinkMessageHeaderSize { + msg, rest, ok := ParseMessage(buf) + if !ok { // Linux ignores messages that are too short. See // net/netlink/af_netlink.c:netlink_rcv_skb. break } - - var hdr linux.NetlinkMessageHeader - binary.Unmarshal(buf[:linux.NetlinkMessageHeaderSize], usermem.ByteOrder, &hdr) - - if hdr.Length < linux.NetlinkMessageHeaderSize || uint64(hdr.Length) > uint64(len(buf)) { - // Linux ignores malformed messages. See - // net/netlink/af_netlink.c:netlink_rcv_skb. - break - } - - // Data from this message. - data := buf[linux.NetlinkMessageHeaderSize:hdr.Length] - - // Advance to the next message. - next := alignUp(int(hdr.Length), linux.NLMSG_ALIGNTO) - if next >= len(buf)-1 { - next = len(buf) - 1 - } - buf = buf[next:] + buf = rest + hdr := msg.Header() // Ignore control messages. if hdr.Type < linux.NLMSG_MIN_TYPE { @@ -692,19 +683,10 @@ func (s *Socket) processMessages(ctx context.Context, buf []byte) *syserr.Error } ms := NewMessageSet(s.portID, hdr.Seq) - var err *syserr.Error - // TODO(b/68877377): ACKs not supported yet. - if hdr.Flags&linux.NLM_F_ACK == linux.NLM_F_ACK { - err = syserr.ErrNotSupported - } else { - - err = s.protocol.ProcessMessage(ctx, hdr, data, ms) - } - if err != nil { - ms = NewMessageSet(s.portID, hdr.Seq) - if err := s.dumpErrorMesage(ctx, hdr, ms, err); err != nil { - return err - } + if err := s.protocol.ProcessMessage(ctx, msg, ms); err != nil { + dumpErrorMesage(hdr, ms, err) + } else if hdr.Flags&linux.NLM_F_ACK == linux.NLM_F_ACK { + dumpAckMesage(hdr, ms) } if err := s.sendResponse(ctx, ms); err != nil { diff --git a/pkg/sentry/socket/netlink/uevent/protocol.go b/pkg/sentry/socket/netlink/uevent/protocol.go index 1ee4296bc..029ba21b5 100644 --- a/pkg/sentry/socket/netlink/uevent/protocol.go +++ b/pkg/sentry/socket/netlink/uevent/protocol.go @@ -49,7 +49,7 @@ func (p *Protocol) CanSend() bool { } // ProcessMessage implements netlink.Protocol.ProcessMessage. -func (p *Protocol) ProcessMessage(ctx context.Context, hdr linux.NetlinkMessageHeader, data []byte, ms *netlink.MessageSet) *syserr.Error { +func (p *Protocol) ProcessMessage(ctx context.Context, msg *netlink.Message, ms *netlink.MessageSet) *syserr.Error { // Silently ignore all messages. return nil } diff --git a/pkg/sentry/socket/netstack/stack.go b/pkg/sentry/socket/netstack/stack.go index 31ea66eca..0692482e9 100644 --- a/pkg/sentry/socket/netstack/stack.go +++ b/pkg/sentry/socket/netstack/stack.go @@ -20,6 +20,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/socket/netfilter" "gvisor.dev/gvisor/pkg/syserr" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/iptables" "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" @@ -88,6 +90,59 @@ func (s *Stack) InterfaceAddrs() map[int32][]inet.InterfaceAddr { return nicAddrs } +// AddInterfaceAddr implements inet.Stack.AddInterfaceAddr. +func (s *Stack) AddInterfaceAddr(idx int32, addr inet.InterfaceAddr) error { + var ( + protocol tcpip.NetworkProtocolNumber + address tcpip.Address + ) + switch addr.Family { + case linux.AF_INET: + if len(addr.Addr) < header.IPv4AddressSize { + return syserror.EINVAL + } + if addr.PrefixLen > header.IPv4AddressSize*8 { + return syserror.EINVAL + } + protocol = ipv4.ProtocolNumber + address = tcpip.Address(addr.Addr[:header.IPv4AddressSize]) + + case linux.AF_INET6: + if len(addr.Addr) < header.IPv6AddressSize { + return syserror.EINVAL + } + if addr.PrefixLen > header.IPv6AddressSize*8 { + return syserror.EINVAL + } + protocol = ipv6.ProtocolNumber + address = tcpip.Address(addr.Addr[:header.IPv6AddressSize]) + + default: + return syserror.ENOTSUP + } + + protocolAddress := tcpip.ProtocolAddress{ + Protocol: protocol, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: address, + PrefixLen: int(addr.PrefixLen), + }, + } + + // Attach address to interface. + if err := s.Stack.AddProtocolAddressWithOptions(tcpip.NICID(idx), protocolAddress, stack.CanBePrimaryEndpoint); err != nil { + return syserr.TranslateNetstackError(err).ToError() + } + + // Add route for local network. + s.Stack.AddRoute(tcpip.Route{ + Destination: protocolAddress.AddressWithPrefix.Subnet(), + Gateway: "", // No gateway for local network. + NIC: tcpip.NICID(idx), + }) + return nil +} + // TCPReceiveBufferSize implements inet.Stack.TCPReceiveBufferSize. func (s *Stack) TCPReceiveBufferSize() (inet.TCPBufferSize, error) { var rs tcp.ReceiveBufferSizeOption diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index 7057b110e..b793f1d74 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -795,6 +795,8 @@ func (s *Stack) Forwarding() bool { // SetRouteTable assigns the route table to be used by this stack. It // specifies which NIC to use for given destination address ranges. +// +// This method takes ownership of the table. func (s *Stack) SetRouteTable(table []tcpip.Route) { s.mu.Lock() defer s.mu.Unlock() @@ -809,6 +811,13 @@ func (s *Stack) GetRouteTable() []tcpip.Route { return append([]tcpip.Route(nil), s.routeTable...) } +// AddRoute appends a route to the route table. +func (s *Stack) AddRoute(route tcpip.Route) { + s.mu.Lock() + defer s.mu.Unlock() + s.routeTable = append(s.routeTable, route) +} + // NewEndpoint creates a new transport layer endpoint of the given protocol. func (s *Stack) NewEndpoint(transport tcpip.TransportProtocolNumber, network tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { t, ok := s.transportProtocols[transport] diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 273b014d6..f2e3c7072 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -2769,9 +2769,11 @@ cc_binary( deps = [ ":socket_netlink_util", ":socket_test_util", + "//test/util:capability_util", "//test/util:cleanup", "//test/util:file_descriptor", "@com_google_absl//absl/strings:str_format", + "@com_google_absl//absl/types:optional", gtest, "//test/util:test_main", "//test/util:test_util", diff --git a/test/syscalls/linux/socket_netlink_route.cc b/test/syscalls/linux/socket_netlink_route.cc index 1e28e658d..e5aed1eec 100644 --- a/test/syscalls/linux/socket_netlink_route.cc +++ b/test/syscalls/linux/socket_netlink_route.cc @@ -14,6 +14,7 @@ #include #include +#include #include #include #include @@ -25,8 +26,10 @@ #include "gtest/gtest.h" #include "absl/strings/str_format.h" +#include "absl/types/optional.h" #include "test/syscalls/linux/socket_netlink_util.h" #include "test/syscalls/linux/socket_test_util.h" +#include "test/util/capability_util.h" #include "test/util/cleanup.h" #include "test/util/file_descriptor.h" #include "test/util/test_util.h" @@ -38,6 +41,8 @@ namespace testing { namespace { +constexpr uint32_t kSeq = 12345; + using ::testing::AnyOf; using ::testing::Eq; @@ -113,46 +118,214 @@ void CheckGetLinkResponse(const struct nlmsghdr* hdr, int seq, int port) { // TODO(mpratt): Check ifinfomsg contents and following attrs. } +PosixError DumpLinks( + const FileDescriptor& fd, uint32_t seq, + const std::function& fn) { + struct request { + struct nlmsghdr hdr; + struct ifinfomsg ifm; + }; + + struct request req = {}; + req.hdr.nlmsg_len = sizeof(req); + req.hdr.nlmsg_type = RTM_GETLINK; + req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP; + req.hdr.nlmsg_seq = seq; + req.ifm.ifi_family = AF_UNSPEC; + + return NetlinkRequestResponse(fd, &req, sizeof(req), fn, false); +} + TEST(NetlinkRouteTest, GetLinkDump) { FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); uint32_t port = ASSERT_NO_ERRNO_AND_VALUE(NetlinkPortID(fd.get())); + // Loopback is common among all tests, check that it's found. + bool loopbackFound = false; + ASSERT_NO_ERRNO(DumpLinks(fd, kSeq, [&](const struct nlmsghdr* hdr) { + CheckGetLinkResponse(hdr, kSeq, port); + if (hdr->nlmsg_type != RTM_NEWLINK) { + return; + } + ASSERT_GE(hdr->nlmsg_len, NLMSG_SPACE(sizeof(struct ifinfomsg))); + const struct ifinfomsg* msg = + reinterpret_cast(NLMSG_DATA(hdr)); + std::cout << "Found interface idx=" << msg->ifi_index + << ", type=" << std::hex << msg->ifi_type; + if (msg->ifi_type == ARPHRD_LOOPBACK) { + loopbackFound = true; + EXPECT_NE(msg->ifi_flags & IFF_LOOPBACK, 0); + } + })); + EXPECT_TRUE(loopbackFound); +} + +struct Link { + int index; + std::string name; +}; + +PosixErrorOr> FindLoopbackLink() { + ASSIGN_OR_RETURN_ERRNO(FileDescriptor fd, NetlinkBoundSocket(NETLINK_ROUTE)); + + absl::optional link; + RETURN_IF_ERRNO(DumpLinks(fd, kSeq, [&](const struct nlmsghdr* hdr) { + if (hdr->nlmsg_type != RTM_NEWLINK || + hdr->nlmsg_len < NLMSG_SPACE(sizeof(struct ifinfomsg))) { + return; + } + const struct ifinfomsg* msg = + reinterpret_cast(NLMSG_DATA(hdr)); + if (msg->ifi_type == ARPHRD_LOOPBACK) { + const auto* rta = FindRtAttr(hdr, msg, IFLA_IFNAME); + if (rta == nullptr) { + // Ignore links that do not have a name. + return; + } + + link = Link(); + link->index = msg->ifi_index; + link->name = std::string(reinterpret_cast(RTA_DATA(rta))); + } + })); + return link; +} + +// CheckLinkMsg checks a netlink message against an expected link. +void CheckLinkMsg(const struct nlmsghdr* hdr, const Link& link) { + ASSERT_THAT(hdr->nlmsg_type, Eq(RTM_NEWLINK)); + ASSERT_GE(hdr->nlmsg_len, NLMSG_SPACE(sizeof(struct ifinfomsg))); + const struct ifinfomsg* msg = + reinterpret_cast(NLMSG_DATA(hdr)); + EXPECT_EQ(msg->ifi_index, link.index); + + const struct rtattr* rta = FindRtAttr(hdr, msg, IFLA_IFNAME); + EXPECT_NE(nullptr, rta) << "IFLA_IFNAME not found in message."; + if (rta != nullptr) { + std::string name(reinterpret_cast(RTA_DATA(rta))); + EXPECT_EQ(name, link.name); + } +} + +TEST(NetlinkRouteTest, GetLinkByIndex) { + absl::optional loopback_link = + ASSERT_NO_ERRNO_AND_VALUE(FindLoopbackLink()); + ASSERT_TRUE(loopback_link.has_value()); + + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); + struct request { struct nlmsghdr hdr; struct ifinfomsg ifm; }; - constexpr uint32_t kSeq = 12345; + struct request req = {}; + req.hdr.nlmsg_len = sizeof(req); + req.hdr.nlmsg_type = RTM_GETLINK; + req.hdr.nlmsg_flags = NLM_F_REQUEST; + req.hdr.nlmsg_seq = kSeq; + req.ifm.ifi_family = AF_UNSPEC; + req.ifm.ifi_index = loopback_link->index; + + bool found = false; + ASSERT_NO_ERRNO(NetlinkRequestResponse( + fd, &req, sizeof(req), + [&](const struct nlmsghdr* hdr) { + CheckLinkMsg(hdr, *loopback_link); + found = true; + }, + false)); + EXPECT_TRUE(found) << "Netlink response does not contain any links."; +} + +TEST(NetlinkRouteTest, GetLinkByName) { + absl::optional loopback_link = + ASSERT_NO_ERRNO_AND_VALUE(FindLoopbackLink()); + ASSERT_TRUE(loopback_link.has_value()); + + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); + + struct request { + struct nlmsghdr hdr; + struct ifinfomsg ifm; + struct rtattr rtattr; + char ifname[IFNAMSIZ]; + char pad[NLMSG_ALIGNTO + RTA_ALIGNTO]; + }; + + struct request req = {}; + req.hdr.nlmsg_type = RTM_GETLINK; + req.hdr.nlmsg_flags = NLM_F_REQUEST; + req.hdr.nlmsg_seq = kSeq; + req.ifm.ifi_family = AF_UNSPEC; + req.rtattr.rta_type = IFLA_IFNAME; + req.rtattr.rta_len = RTA_LENGTH(loopback_link->name.size() + 1); + strncpy(req.ifname, loopback_link->name.c_str(), sizeof(req.ifname)); + req.hdr.nlmsg_len = + NLMSG_LENGTH(sizeof(req.ifm)) + NLMSG_ALIGN(req.rtattr.rta_len); + + bool found = false; + ASSERT_NO_ERRNO(NetlinkRequestResponse( + fd, &req, sizeof(req), + [&](const struct nlmsghdr* hdr) { + CheckLinkMsg(hdr, *loopback_link); + found = true; + }, + false)); + EXPECT_TRUE(found) << "Netlink response does not contain any links."; +} + +TEST(NetlinkRouteTest, GetLinkByIndexNotFound) { + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); + + struct request { + struct nlmsghdr hdr; + struct ifinfomsg ifm; + }; struct request req = {}; req.hdr.nlmsg_len = sizeof(req); req.hdr.nlmsg_type = RTM_GETLINK; - req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP; + req.hdr.nlmsg_flags = NLM_F_REQUEST; req.hdr.nlmsg_seq = kSeq; req.ifm.ifi_family = AF_UNSPEC; + req.ifm.ifi_index = 1234590; - // Loopback is common among all tests, check that it's found. - bool loopbackFound = false; - ASSERT_NO_ERRNO(NetlinkRequestResponse( - fd, &req, sizeof(req), - [&](const struct nlmsghdr* hdr) { - CheckGetLinkResponse(hdr, kSeq, port); - if (hdr->nlmsg_type != RTM_NEWLINK) { - return; - } - ASSERT_GE(hdr->nlmsg_len, NLMSG_SPACE(sizeof(struct ifinfomsg))); - const struct ifinfomsg* msg = - reinterpret_cast(NLMSG_DATA(hdr)); - std::cout << "Found interface idx=" << msg->ifi_index - << ", type=" << std::hex << msg->ifi_type; - if (msg->ifi_type == ARPHRD_LOOPBACK) { - loopbackFound = true; - EXPECT_NE(msg->ifi_flags & IFF_LOOPBACK, 0); - } - }, - false)); - EXPECT_TRUE(loopbackFound); + EXPECT_THAT(NetlinkRequestAckOrError(fd, kSeq, &req, sizeof(req)), + PosixErrorIs(ENODEV, ::testing::_)); +} + +TEST(NetlinkRouteTest, GetLinkByNameNotFound) { + const std::string name = "nodevice?!"; + + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); + + struct request { + struct nlmsghdr hdr; + struct ifinfomsg ifm; + struct rtattr rtattr; + char ifname[IFNAMSIZ]; + char pad[NLMSG_ALIGNTO + RTA_ALIGNTO]; + }; + + struct request req = {}; + req.hdr.nlmsg_type = RTM_GETLINK; + req.hdr.nlmsg_flags = NLM_F_REQUEST; + req.hdr.nlmsg_seq = kSeq; + req.ifm.ifi_family = AF_UNSPEC; + req.rtattr.rta_type = IFLA_IFNAME; + req.rtattr.rta_len = RTA_LENGTH(name.size() + 1); + strncpy(req.ifname, name.c_str(), sizeof(req.ifname)); + req.hdr.nlmsg_len = + NLMSG_LENGTH(sizeof(req.ifm)) + NLMSG_ALIGN(req.rtattr.rta_len); + + EXPECT_THAT(NetlinkRequestAckOrError(fd, kSeq, &req, sizeof(req)), + PosixErrorIs(ENODEV, ::testing::_)); } TEST(NetlinkRouteTest, MsgHdrMsgUnsuppType) { @@ -164,8 +337,6 @@ TEST(NetlinkRouteTest, MsgHdrMsgUnsuppType) { struct ifinfomsg ifm; }; - constexpr uint32_t kSeq = 12345; - struct request req = {}; req.hdr.nlmsg_len = sizeof(req); // If type & 0x3 is equal to 0x2, this means a get request @@ -175,18 +346,8 @@ TEST(NetlinkRouteTest, MsgHdrMsgUnsuppType) { req.hdr.nlmsg_seq = kSeq; req.ifm.ifi_family = AF_UNSPEC; - ASSERT_NO_ERRNO(NetlinkRequestResponse( - fd, &req, sizeof(req), - [&](const struct nlmsghdr* hdr) { - EXPECT_THAT(hdr->nlmsg_type, Eq(NLMSG_ERROR)); - EXPECT_EQ(hdr->nlmsg_seq, kSeq); - EXPECT_GE(hdr->nlmsg_len, sizeof(*hdr) + sizeof(struct nlmsgerr)); - - const struct nlmsgerr* msg = - reinterpret_cast(NLMSG_DATA(hdr)); - EXPECT_EQ(msg->error, -EOPNOTSUPP); - }, - true)); + EXPECT_THAT(NetlinkRequestAckOrError(fd, kSeq, &req, sizeof(req)), + PosixErrorIs(EOPNOTSUPP, ::testing::_)); } TEST(NetlinkRouteTest, MsgHdrMsgTrunc) { @@ -198,8 +359,6 @@ TEST(NetlinkRouteTest, MsgHdrMsgTrunc) { struct ifinfomsg ifm; }; - constexpr uint32_t kSeq = 12345; - struct request req = {}; req.hdr.nlmsg_len = sizeof(req); req.hdr.nlmsg_type = RTM_GETLINK; @@ -238,8 +397,6 @@ TEST(NetlinkRouteTest, MsgTruncMsgHdrMsgTrunc) { struct ifinfomsg ifm; }; - constexpr uint32_t kSeq = 12345; - struct request req = {}; req.hdr.nlmsg_len = sizeof(req); req.hdr.nlmsg_type = RTM_GETLINK; @@ -282,8 +439,6 @@ TEST(NetlinkRouteTest, ControlMessageIgnored) { struct ifinfomsg ifm; }; - constexpr uint32_t kSeq = 12345; - struct request req = {}; // This control message is ignored. We still receive a response for the @@ -317,8 +472,6 @@ TEST(NetlinkRouteTest, GetAddrDump) { struct rtgenmsg rgm; }; - constexpr uint32_t kSeq = 12345; - struct request req; req.hdr.nlmsg_len = sizeof(req); req.hdr.nlmsg_type = RTM_GETADDR; @@ -367,6 +520,57 @@ TEST(NetlinkRouteTest, LookupAll) { ASSERT_GT(count, 0); } +TEST(NetlinkRouteTest, AddAddr) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN))); + + absl::optional loopback_link = + ASSERT_NO_ERRNO_AND_VALUE(FindLoopbackLink()); + ASSERT_TRUE(loopback_link.has_value()); + + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); + + struct request { + struct nlmsghdr hdr; + struct ifaddrmsg ifa; + struct rtattr rtattr; + struct in_addr addr; + char pad[NLMSG_ALIGNTO + RTA_ALIGNTO]; + }; + + struct request req = {}; + req.hdr.nlmsg_type = RTM_NEWADDR; + req.hdr.nlmsg_seq = kSeq; + req.ifa.ifa_family = AF_INET; + req.ifa.ifa_prefixlen = 24; + req.ifa.ifa_flags = 0; + req.ifa.ifa_scope = 0; + req.ifa.ifa_index = loopback_link->index; + req.rtattr.rta_type = IFA_LOCAL; + req.rtattr.rta_len = RTA_LENGTH(sizeof(req.addr)); + inet_pton(AF_INET, "10.0.0.1", &req.addr); + req.hdr.nlmsg_len = + NLMSG_LENGTH(sizeof(req.ifa)) + NLMSG_ALIGN(req.rtattr.rta_len); + + // Create should succeed, as no such address in kernel. + req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_CREATE | NLM_F_ACK; + EXPECT_NO_ERRNO( + NetlinkRequestAckOrError(fd, req.hdr.nlmsg_seq, &req, req.hdr.nlmsg_len)); + + // Replace an existing address should succeed. + req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_REPLACE | NLM_F_ACK; + req.hdr.nlmsg_seq++; + EXPECT_NO_ERRNO( + NetlinkRequestAckOrError(fd, req.hdr.nlmsg_seq, &req, req.hdr.nlmsg_len)); + + // Create exclusive should fail, as we created the address above. + req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_CREATE | NLM_F_EXCL | NLM_F_ACK; + req.hdr.nlmsg_seq++; + EXPECT_THAT( + NetlinkRequestAckOrError(fd, req.hdr.nlmsg_seq, &req, req.hdr.nlmsg_len), + PosixErrorIs(EEXIST, ::testing::_)); +} + // GetRouteDump tests a RTM_GETROUTE + NLM_F_DUMP request. TEST(NetlinkRouteTest, GetRouteDump) { FileDescriptor fd = @@ -378,8 +582,6 @@ TEST(NetlinkRouteTest, GetRouteDump) { struct rtmsg rtm; }; - constexpr uint32_t kSeq = 12345; - struct request req = {}; req.hdr.nlmsg_len = sizeof(req); req.hdr.nlmsg_type = RTM_GETROUTE; @@ -538,8 +740,6 @@ TEST(NetlinkRouteTest, RecvmsgTrunc) { struct rtgenmsg rgm; }; - constexpr uint32_t kSeq = 12345; - struct request req; req.hdr.nlmsg_len = sizeof(req); req.hdr.nlmsg_type = RTM_GETADDR; @@ -615,8 +815,6 @@ TEST(NetlinkRouteTest, RecvmsgTruncPeek) { struct rtgenmsg rgm; }; - constexpr uint32_t kSeq = 12345; - struct request req; req.hdr.nlmsg_len = sizeof(req); req.hdr.nlmsg_type = RTM_GETADDR; @@ -695,8 +893,6 @@ TEST(NetlinkRouteTest, NoPasscredNoCreds) { struct rtgenmsg rgm; }; - constexpr uint32_t kSeq = 12345; - struct request req; req.hdr.nlmsg_len = sizeof(req); req.hdr.nlmsg_type = RTM_GETADDR; @@ -743,8 +939,6 @@ TEST(NetlinkRouteTest, PasscredCreds) { struct rtgenmsg rgm; }; - constexpr uint32_t kSeq = 12345; - struct request req; req.hdr.nlmsg_len = sizeof(req); req.hdr.nlmsg_type = RTM_GETADDR; diff --git a/test/syscalls/linux/socket_netlink_util.cc b/test/syscalls/linux/socket_netlink_util.cc index cd2212a1a..952eecfe8 100644 --- a/test/syscalls/linux/socket_netlink_util.cc +++ b/test/syscalls/linux/socket_netlink_util.cc @@ -16,6 +16,7 @@ #include #include +#include #include #include @@ -71,9 +72,10 @@ PosixError NetlinkRequestResponse( iov.iov_base = buf.data(); iov.iov_len = buf.size(); - // Response is a series of NLM_F_MULTI messages, ending with a NLMSG_DONE - // message. + // If NLM_F_MULTI is set, response is a series of messages that ends with a + // NLMSG_DONE message. int type = -1; + int flags = 0; do { int len; RETURN_ERROR_IF_SYSCALL_FAIL(len = RetryEINTR(recvmsg)(fd.get(), &msg, 0)); @@ -89,6 +91,7 @@ PosixError NetlinkRequestResponse( for (struct nlmsghdr* hdr = reinterpret_cast(buf.data()); NLMSG_OK(hdr, len); hdr = NLMSG_NEXT(hdr, len)) { fn(hdr); + flags = hdr->nlmsg_flags; type = hdr->nlmsg_type; // Done should include an integer payload for dump_done_errno. // See net/netlink/af_netlink.c:netlink_dump @@ -98,11 +101,11 @@ PosixError NetlinkRequestResponse( EXPECT_GE(hdr->nlmsg_len, NLMSG_LENGTH(sizeof(int))); } } - } while (type != NLMSG_DONE && type != NLMSG_ERROR); + } while ((flags & NLM_F_MULTI) && type != NLMSG_DONE && type != NLMSG_ERROR); if (expect_nlmsgerr) { EXPECT_EQ(type, NLMSG_ERROR); - } else { + } else if (flags & NLM_F_MULTI) { EXPECT_EQ(type, NLMSG_DONE); } return NoError(); @@ -146,5 +149,39 @@ PosixError NetlinkRequestResponseSingle( return NoError(); } +PosixError NetlinkRequestAckOrError(const FileDescriptor& fd, uint32_t seq, + void* request, size_t len) { + // Dummy negative number for no error message received. + // We won't get a negative error number so there will be no confusion. + int err = -42; + RETURN_IF_ERRNO(NetlinkRequestResponse( + fd, request, len, + [&](const struct nlmsghdr* hdr) { + EXPECT_EQ(NLMSG_ERROR, hdr->nlmsg_type); + EXPECT_EQ(hdr->nlmsg_seq, seq); + EXPECT_GE(hdr->nlmsg_len, sizeof(*hdr) + sizeof(struct nlmsgerr)); + + const struct nlmsgerr* msg = + reinterpret_cast(NLMSG_DATA(hdr)); + err = -msg->error; + }, + true)); + return PosixError(err); +} + +const struct rtattr* FindRtAttr(const struct nlmsghdr* hdr, + const struct ifinfomsg* msg, int16_t attr) { + const int ifi_space = NLMSG_SPACE(sizeof(*msg)); + int attrlen = hdr->nlmsg_len - ifi_space; + const struct rtattr* rta = reinterpret_cast( + reinterpret_cast(hdr) + NLMSG_ALIGN(ifi_space)); + for (; RTA_OK(rta, attrlen); rta = RTA_NEXT(rta, attrlen)) { + if (rta->rta_type == attr) { + return rta; + } + } + return nullptr; +} + } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_netlink_util.h b/test/syscalls/linux/socket_netlink_util.h index 3678c0599..e13ead406 100644 --- a/test/syscalls/linux/socket_netlink_util.h +++ b/test/syscalls/linux/socket_netlink_util.h @@ -19,6 +19,7 @@ // socket.h has to be included before if_arp.h. #include #include +#include #include "test/util/file_descriptor.h" #include "test/util/posix_error.h" @@ -47,6 +48,14 @@ PosixError NetlinkRequestResponseSingle( const FileDescriptor& fd, void* request, size_t len, const std::function& fn); +// Send the passed request then expect and return an ack or error. +PosixError NetlinkRequestAckOrError(const FileDescriptor& fd, uint32_t seq, + void* request, size_t len); + +// Find rtnetlink attribute in message. +const struct rtattr* FindRtAttr(const struct nlmsghdr* hdr, + const struct ifinfomsg* msg, int16_t attr); + } // namespace testing } // namespace gvisor From 37abbbc547d9d78e0bf42f192403c8eca30593d8 Mon Sep 17 00:00:00 2001 From: Eyal Soha Date: Wed, 5 Feb 2020 11:17:14 -0800 Subject: [PATCH 27/29] Add packetdrill tests to presubmit and CI testing PiperOrigin-RevId: 293409718 --- kokoro/packetdrill_tests.cfg | 9 +++++++++ scripts/packetdrill_tests.sh | 20 ++++++++++++++++++++ test/packetdrill/defs.bzl | 6 ++++-- test/packetdrill/packetdrill_test.sh | 20 ++++++++++++++++---- 4 files changed, 49 insertions(+), 6 deletions(-) create mode 100644 kokoro/packetdrill_tests.cfg create mode 100755 scripts/packetdrill_tests.sh diff --git a/kokoro/packetdrill_tests.cfg b/kokoro/packetdrill_tests.cfg new file mode 100644 index 000000000..258d7deb4 --- /dev/null +++ b/kokoro/packetdrill_tests.cfg @@ -0,0 +1,9 @@ +build_file: "repo/scripts/packetdrill_tests.sh" + +action { + define_artifacts { + regex: "**/sponge_log.xml" + regex: "**/sponge_log.log" + regex: "**/outputs.zip" + } +} diff --git a/scripts/packetdrill_tests.sh b/scripts/packetdrill_tests.sh new file mode 100755 index 000000000..fc6bef79c --- /dev/null +++ b/scripts/packetdrill_tests.sh @@ -0,0 +1,20 @@ +#!/bin/bash + +# Copyright 2019 The gVisor Authors. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +source $(dirname $0)/common.sh + +install_runsc_for_test runsc-d +test_runsc $(bazel query "attr(tags, manual, tests(//test/packetdrill/...))") diff --git a/test/packetdrill/defs.bzl b/test/packetdrill/defs.bzl index 582f97e0c..8623ce7b1 100644 --- a/test/packetdrill/defs.bzl +++ b/test/packetdrill/defs.bzl @@ -15,7 +15,7 @@ def _packetdrill_test_impl(ctx): # Make sure that everything is readable here. "find . -type f -exec chmod a+rx {} \\;", "find . -type d -exec chmod a+rx {} \\;", - "%s %s --init_script %s -- %s\n" % ( + "%s %s --init_script %s $@ -- %s\n" % ( test_runner.short_path, " ".join(ctx.attr.flags), ctx.files._init_script[0].short_path, @@ -76,7 +76,9 @@ def packetdrill_netstack_test(name, **kwargs): kwargs["tags"] = _PACKETDRILL_TAGS _packetdrill_test( name = name + "_netstack_test", - flags = ["--dut_platform", "netstack"], + # This is the default runtime unless + # "--test_arg=--runtime=OTHER_RUNTIME" is used to override the value. + flags = ["--dut_platform", "netstack", "--runtime", "runsc-d"], **kwargs ) diff --git a/test/packetdrill/packetdrill_test.sh b/test/packetdrill/packetdrill_test.sh index 614d94d74..0b22dfd5c 100755 --- a/test/packetdrill/packetdrill_test.sh +++ b/test/packetdrill/packetdrill_test.sh @@ -29,7 +29,7 @@ function failure() { } trap 'failure ${LINENO} "$BASH_COMMAND"' ERR -declare -r LONGOPTS="dut_platform:,init_script:" +declare -r LONGOPTS="dut_platform:,init_script:,runtime:" # Don't use declare below so that the error from getopt will end the script. PARSED=$(getopt --options "" --longoptions=$LONGOPTS --name "$0" -- "$@") @@ -39,6 +39,7 @@ eval set -- "$PARSED" while true; do case "$1" in --dut_platform) + # Either "linux" or "netstack". declare -r DUT_PLATFORM="$2" shift 2 ;; @@ -46,6 +47,13 @@ while true; do declare -r INIT_SCRIPT="$2" shift 2 ;; + --runtime) + # Not readonly because there might be multiple --runtime arguments and we + # want to use just the last one. Only used if --dut_platform is + # "netstack". + declare RUNTIME="$2" + shift 2 + ;; --) shift break @@ -61,9 +69,13 @@ declare -r scripts="$@" # Check that the required flags are defined in a way that is safe for "set -u". if [[ "${DUT_PLATFORM-}" == "netstack" ]]; then - declare -r RUNTIME="--runtime runsc-d" + if [[ -z "${RUNTIME-}" ]]; then + echo "FAIL: Missing --runtime argument: ${RUNTIME-}" + exit 2 + fi + declare -r RUNTIME_ARG="--runtime ${RUNTIME}" elif [[ "${DUT_PLATFORM-}" == "linux" ]]; then - declare -r RUNTIME="" + declare -r RUNTIME_ARG="" else echo "FAIL: Bad or missing --dut_platform argument: ${DUT_PLATFORM-}" exit 2 @@ -143,7 +155,7 @@ done docker pull "${IMAGE_TAG}" # Create the DUT container and connect to network. -DUT=$(docker create ${RUNTIME} --privileged --rm \ +DUT=$(docker create ${RUNTIME_ARG} --privileged --rm \ --stop-timeout ${TIMEOUT} -it ${IMAGE_TAG}) docker network connect "${CTRL_NET}" \ --ip "${CTRL_NET_PREFIX}${DUT_NET_SUFFIX}" "${DUT}" \ From eea0eeee933ba8406ae688fce4348271f9513514 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Wed, 5 Feb 2020 11:25:10 -0800 Subject: [PATCH 28/29] Disable get/set xattrs until list/remove exist too. PiperOrigin-RevId: 293411655 --- pkg/sentry/syscalls/linux/linux64_amd64.go | 27 +++-- pkg/sentry/syscalls/linux/linux64_arm64.go | 37 +++--- test/syscalls/linux/xattr.cc | 124 +++++++++++++++++++++ 3 files changed, 159 insertions(+), 29 deletions(-) diff --git a/pkg/sentry/syscalls/linux/linux64_amd64.go b/pkg/sentry/syscalls/linux/linux64_amd64.go index 7435b50bf..588f8b087 100644 --- a/pkg/sentry/syscalls/linux/linux64_amd64.go +++ b/pkg/sentry/syscalls/linux/linux64_amd64.go @@ -228,18 +228,21 @@ var AMD64 = &kernel.SyscallTable{ 185: syscalls.Error("security", syserror.ENOSYS, "Not implemented in Linux.", nil), 186: syscalls.Supported("gettid", Gettid), 187: syscalls.Supported("readahead", Readahead), - 188: syscalls.PartiallySupported("setxattr", SetXattr, "Only supported for tmpfs.", nil), - 189: syscalls.PartiallySupported("lsetxattr", LSetXattr, "Only supported for tmpfs.", nil), - 190: syscalls.PartiallySupported("fsetxattr", FSetXattr, "Only supported for tmpfs.", nil), - 191: syscalls.PartiallySupported("getxattr", GetXattr, "Only supported for tmpfs.", nil), - 192: syscalls.PartiallySupported("lgetxattr", LGetXattr, "Only supported for tmpfs.", nil), - 193: syscalls.PartiallySupported("fgetxattr", FGetXattr, "Only supported for tmpfs.", nil), - 194: syscalls.ErrorWithEvent("listxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 195: syscalls.ErrorWithEvent("llistxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 196: syscalls.ErrorWithEvent("flistxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 197: syscalls.ErrorWithEvent("removexattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 198: syscalls.ErrorWithEvent("lremovexattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 199: syscalls.ErrorWithEvent("fremovexattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + // TODO(b/148303075): Enable set/getxattr (in their various + // forms) once we also have list and removexattr. The JVM + // assumes that if get/set exist, then list and remove do too. + 188: syscalls.ErrorWithEvent("setxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 189: syscalls.ErrorWithEvent("lsetxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 190: syscalls.ErrorWithEvent("fsetxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 191: syscalls.ErrorWithEvent("getxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 192: syscalls.ErrorWithEvent("lgetxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 193: syscalls.ErrorWithEvent("fgetxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 194: syscalls.ErrorWithEvent("listxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 195: syscalls.ErrorWithEvent("llistxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 196: syscalls.ErrorWithEvent("flistxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 197: syscalls.ErrorWithEvent("removexattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 198: syscalls.ErrorWithEvent("lremovexattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 199: syscalls.ErrorWithEvent("fremovexattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), 200: syscalls.Supported("tkill", Tkill), 201: syscalls.Supported("time", Time), 202: syscalls.PartiallySupported("futex", Futex, "Robust futexes not supported.", nil), diff --git a/pkg/sentry/syscalls/linux/linux64_arm64.go b/pkg/sentry/syscalls/linux/linux64_arm64.go index 03a39fe65..06e5ee401 100644 --- a/pkg/sentry/syscalls/linux/linux64_arm64.go +++ b/pkg/sentry/syscalls/linux/linux64_arm64.go @@ -36,23 +36,26 @@ var ARM64 = &kernel.SyscallTable{ }, AuditNumber: linux.AUDIT_ARCH_AARCH64, Table: map[uintptr]kernel.Syscall{ - 0: syscalls.PartiallySupported("io_setup", IoSetup, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 1: syscalls.PartiallySupported("io_destroy", IoDestroy, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 2: syscalls.PartiallySupported("io_submit", IoSubmit, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 3: syscalls.PartiallySupported("io_cancel", IoCancel, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 4: syscalls.PartiallySupported("io_getevents", IoGetevents, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 5: syscalls.PartiallySupported("setxattr", SetXattr, "Only supported for tmpfs.", nil), - 6: syscalls.PartiallySupported("lsetxattr", LSetXattr, "Only supported for tmpfs.", nil), - 7: syscalls.PartiallySupported("fsetxattr", FSetXattr, "Only supported for tmpfs.", nil), - 8: syscalls.PartiallySupported("getxattr", GetXattr, "Only supported for tmpfs.", nil), - 9: syscalls.PartiallySupported("lgetxattr", LGetXattr, "Only supported for tmpfs.", nil), - 10: syscalls.PartiallySupported("fgetxattr", FGetXattr, "Only supported for tmpfs.", nil), - 11: syscalls.ErrorWithEvent("listxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 12: syscalls.ErrorWithEvent("llistxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 13: syscalls.ErrorWithEvent("flistxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 14: syscalls.ErrorWithEvent("removexattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 15: syscalls.ErrorWithEvent("lremovexattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 16: syscalls.ErrorWithEvent("fremovexattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 0: syscalls.PartiallySupported("io_setup", IoSetup, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 1: syscalls.PartiallySupported("io_destroy", IoDestroy, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 2: syscalls.PartiallySupported("io_submit", IoSubmit, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 3: syscalls.PartiallySupported("io_cancel", IoCancel, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 4: syscalls.PartiallySupported("io_getevents", IoGetevents, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + // TODO(b/148303075): Enable set/getxattr (in their various + // forms) once we also have list and removexattr. The JVM + // assumes that if get/set exist, then list and remove do too. + 5: syscalls.ErrorWithEvent("setxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 6: syscalls.ErrorWithEvent("lsetxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 7: syscalls.ErrorWithEvent("fsetxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 8: syscalls.ErrorWithEvent("getxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 9: syscalls.ErrorWithEvent("lgetxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 10: syscalls.ErrorWithEvent("fgetxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 11: syscalls.ErrorWithEvent("listxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 13: syscalls.ErrorWithEvent("llistxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 13: syscalls.ErrorWithEvent("flistxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 14: syscalls.ErrorWithEvent("removexattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 15: syscalls.ErrorWithEvent("lremovexattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 16: syscalls.ErrorWithEvent("fremovexattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), 17: syscalls.Supported("getcwd", Getcwd), 18: syscalls.CapError("lookup_dcookie", linux.CAP_SYS_ADMIN, "", nil), 19: syscalls.Supported("eventfd2", Eventfd2), diff --git a/test/syscalls/linux/xattr.cc b/test/syscalls/linux/xattr.cc index ab21d68c6..85eb31847 100644 --- a/test/syscalls/linux/xattr.cc +++ b/test/syscalls/linux/xattr.cc @@ -39,6 +39,10 @@ namespace { class XattrTest : public FileTest {}; TEST_F(XattrTest, XattrNullName) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); EXPECT_THAT(setxattr(path, nullptr, nullptr, 0, /*flags=*/0), @@ -48,6 +52,10 @@ TEST_F(XattrTest, XattrNullName) { } TEST_F(XattrTest, XattrEmptyName) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); EXPECT_THAT(setxattr(path, "", nullptr, 0, /*flags=*/0), @@ -56,6 +64,10 @@ TEST_F(XattrTest, XattrEmptyName) { } TEST_F(XattrTest, XattrLargeName) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); std::string name = "user."; name += std::string(XATTR_NAME_MAX - name.length(), 'a'); @@ -77,6 +89,10 @@ TEST_F(XattrTest, XattrLargeName) { } TEST_F(XattrTest, XattrInvalidPrefix) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); std::string name(XATTR_NAME_MAX, 'a'); EXPECT_THAT(setxattr(path, name.c_str(), nullptr, 0, /*flags=*/0), @@ -88,6 +104,10 @@ TEST_F(XattrTest, XattrInvalidPrefix) { // Do not allow save/restore cycles after making the test file read-only, as // the restore will fail to open it with r/w permissions. TEST_F(XattrTest, XattrReadOnly_NoRandomSave) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + // Drop capabilities that allow us to override file and directory permissions. ASSERT_NO_ERRNO(SetCapability(CAP_DAC_OVERRIDE, false)); ASSERT_NO_ERRNO(SetCapability(CAP_DAC_READ_SEARCH, false)); @@ -113,6 +133,10 @@ TEST_F(XattrTest, XattrReadOnly_NoRandomSave) { // Do not allow save/restore cycles after making the test file write-only, as // the restore will fail to open it with r/w permissions. TEST_F(XattrTest, XattrWriteOnly_NoRandomSave) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + // Drop capabilities that allow us to override file and directory permissions. ASSERT_NO_ERRNO(SetCapability(CAP_DAC_OVERRIDE, false)); ASSERT_NO_ERRNO(SetCapability(CAP_DAC_READ_SEARCH, false)); @@ -143,6 +167,10 @@ TEST_F(XattrTest, XattrTrustedWithNonadmin) { } TEST_F(XattrTest, XattrOnDirectory) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); const char name[] = "user.test"; EXPECT_THAT(setxattr(dir.path().c_str(), name, NULL, 0, /*flags=*/0), @@ -152,6 +180,10 @@ TEST_F(XattrTest, XattrOnDirectory) { } TEST_F(XattrTest, XattrOnSymlink) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); TempPath link = ASSERT_NO_ERRNO_AND_VALUE( TempPath::CreateSymlinkTo(dir.path(), test_file_name_)); @@ -163,6 +195,10 @@ TEST_F(XattrTest, XattrOnSymlink) { } TEST_F(XattrTest, XattrOnInvalidFileTypes) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char name[] = "user.test"; char char_device[] = "/dev/zero"; @@ -181,6 +217,10 @@ TEST_F(XattrTest, XattrOnInvalidFileTypes) { } TEST_F(XattrTest, SetxattrSizeSmallerThanValue) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; std::vector val = {'a', 'a'}; @@ -196,6 +236,10 @@ TEST_F(XattrTest, SetxattrSizeSmallerThanValue) { } TEST_F(XattrTest, SetxattrZeroSize) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; char val = 'a'; @@ -208,6 +252,10 @@ TEST_F(XattrTest, SetxattrZeroSize) { } TEST_F(XattrTest, SetxattrSizeTooLarge) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; @@ -223,6 +271,10 @@ TEST_F(XattrTest, SetxattrSizeTooLarge) { } TEST_F(XattrTest, SetxattrNullValueAndNonzeroSize) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; EXPECT_THAT(setxattr(path, name, nullptr, 1, /*flags=*/0), @@ -232,6 +284,10 @@ TEST_F(XattrTest, SetxattrNullValueAndNonzeroSize) { } TEST_F(XattrTest, SetxattrNullValueAndZeroSize) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; EXPECT_THAT(setxattr(path, name, nullptr, 0, /*flags=*/0), SyscallSucceeds()); @@ -240,6 +296,10 @@ TEST_F(XattrTest, SetxattrNullValueAndZeroSize) { } TEST_F(XattrTest, SetxattrValueTooLargeButOKSize) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; std::vector val(XATTR_SIZE_MAX + 1); @@ -256,6 +316,10 @@ TEST_F(XattrTest, SetxattrValueTooLargeButOKSize) { } TEST_F(XattrTest, SetxattrReplaceWithSmaller) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; std::vector val = {'a', 'a'}; @@ -271,6 +335,10 @@ TEST_F(XattrTest, SetxattrReplaceWithSmaller) { } TEST_F(XattrTest, SetxattrReplaceWithLarger) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; std::vector val = {'a', 'a'}; @@ -285,6 +353,10 @@ TEST_F(XattrTest, SetxattrReplaceWithLarger) { } TEST_F(XattrTest, SetxattrCreateFlag) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; EXPECT_THAT(setxattr(path, name, nullptr, 0, XATTR_CREATE), @@ -296,6 +368,10 @@ TEST_F(XattrTest, SetxattrCreateFlag) { } TEST_F(XattrTest, SetxattrReplaceFlag) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; EXPECT_THAT(setxattr(path, name, nullptr, 0, XATTR_REPLACE), @@ -308,6 +384,10 @@ TEST_F(XattrTest, SetxattrReplaceFlag) { } TEST_F(XattrTest, SetxattrInvalidFlags) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); int invalid_flags = 0xff; EXPECT_THAT(setxattr(path, nullptr, nullptr, 0, invalid_flags), @@ -315,6 +395,10 @@ TEST_F(XattrTest, SetxattrInvalidFlags) { } TEST_F(XattrTest, Getxattr) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; int val = 1234; @@ -327,6 +411,10 @@ TEST_F(XattrTest, Getxattr) { } TEST_F(XattrTest, GetxattrSizeSmallerThanValue) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; std::vector val = {'a', 'a'}; @@ -339,6 +427,10 @@ TEST_F(XattrTest, GetxattrSizeSmallerThanValue) { } TEST_F(XattrTest, GetxattrSizeLargerThanValue) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; char val = 'a'; @@ -354,6 +446,10 @@ TEST_F(XattrTest, GetxattrSizeLargerThanValue) { } TEST_F(XattrTest, GetxattrZeroSize) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; char val = 'a'; @@ -367,6 +463,10 @@ TEST_F(XattrTest, GetxattrZeroSize) { } TEST_F(XattrTest, GetxattrSizeTooLarge) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; char val = 'a'; @@ -383,6 +483,10 @@ TEST_F(XattrTest, GetxattrSizeTooLarge) { } TEST_F(XattrTest, GetxattrNullValue) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; char val = 'a'; @@ -394,6 +498,10 @@ TEST_F(XattrTest, GetxattrNullValue) { } TEST_F(XattrTest, GetxattrNullValueAndZeroSize) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; char val = 'a'; @@ -410,12 +518,20 @@ TEST_F(XattrTest, GetxattrNullValueAndZeroSize) { } TEST_F(XattrTest, GetxattrNonexistentName) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; EXPECT_THAT(getxattr(path, name, nullptr, 0), SyscallFailsWithErrno(ENODATA)); } TEST_F(XattrTest, LGetSetxattrOnSymlink) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); TempPath link = ASSERT_NO_ERRNO_AND_VALUE( TempPath::CreateSymlinkTo(dir.path(), test_file_name_)); @@ -427,6 +543,10 @@ TEST_F(XattrTest, LGetSetxattrOnSymlink) { } TEST_F(XattrTest, LGetSetxattrOnNonsymlink) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; int val = 1234; @@ -441,6 +561,10 @@ TEST_F(XattrTest, LGetSetxattrOnNonsymlink) { } TEST_F(XattrTest, FGetSetxattr) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(test_file_name_.c_str(), 0)); const char name[] = "user.test"; From f2d3efca1deded31a2929ea77c0eecf476764660 Mon Sep 17 00:00:00 2001 From: Brad Burlage Date: Wed, 5 Feb 2020 11:33:48 -0800 Subject: [PATCH 29/29] Fix undeclared variable error in common_build.sh. PiperOrigin-RevId: 293413711 --- scripts/common_build.sh | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/scripts/common_build.sh b/scripts/common_build.sh index a473a88a4..2c2a826c7 100755 --- a/scripts/common_build.sh +++ b/scripts/common_build.sh @@ -32,21 +32,21 @@ declare -r BAZEL_FLAGS=( "--keep_going" "--verbose_failures=true" ) +BAZEL_RBE_AUTH_FLAGS="" +BAZEL_RBE_FLAGS="" if [[ -v KOKORO_BAZEL_AUTH_CREDENTIAL ]]; then - declare -r BAZEL_RBE_AUTH_FLAGS=( - "--auth_credentials=${KOKORO_BAZEL_AUTH_CREDENTIAL}" - ) - declare -r BAZEL_RBE_FLAGS=("--config=remote") + declare -r BAZEL_RBE_AUTH_FLAGS="--auth_credentials=${KOKORO_BAZEL_AUTH_CREDENTIAL}" + declare -r BAZEL_RBE_FLAGS="--config=remote" fi # Wrap bazel. function build() { - bazel build "${BAZEL_RBE_FLAGS[@]}" "${BAZEL_RBE_AUTH_FLAGS[@]}" "${BAZEL_FLAGS[@]}" "$@" 2>&1 | + bazel build "${BAZEL_RBE_FLAGS}" "${BAZEL_RBE_AUTH_FLAGS}" "${BAZEL_FLAGS[@]}" "$@" 2>&1 | tee /dev/fd/2 | grep -E '^ bazel-bin/' | awk '{ print $1; }' } function test() { - bazel test "${BAZEL_RBE_FLAGS[@]}" "${BAZEL_RBE_AUTH_FLAGS[@]}" "${BAZEL_FLAGS[@]}" "$@" + bazel test "${BAZEL_RBE_FLAGS}" "${BAZEL_RBE_AUTH_FLAGS}" "${BAZEL_FLAGS[@]}" "$@" } function run() {