fs: use lockdep mutexes

PiperOrigin-RevId: 452826155

pkg/sentry/fs/BUILD

@@ -1,5 +1,6 @@
 load("//tools:defs.bzl", "go_library", "go_test")
 load("//tools/go_generics:defs.bzl", "go_template_instance")
+load("//pkg/sync/locking:locking.bzl", "declare_mutex")
 
 package(licenses = ["notice"])
 
@@ -29,11 +30,14 @@ go_library(
         "inode_overlay.go",
         "inotify.go",
         "inotify_event.go",
+        "inotify_event_mutex.go",
+        "inotify_mutex.go",
         "inotify_watch.go",
         "mock.go",
         "mount.go",
         "mount_overlay.go",
         "mounts.go",
+        "namespace_mutex.go",
         "offset.go",
         "overlay.go",
         "path.go",
@@ -68,6 +72,7 @@ go_library(
         "//pkg/sentry/usage",
         "//pkg/state",
         "//pkg/sync",
+        "//pkg/sync/locking",
         "//pkg/usermem",
         "//pkg/waiter",
         "@org_golang_x_sys//unix:go_default_library",
@@ -98,6 +103,27 @@ go_template_instance(
     },
 )
 
+declare_mutex(
+    name = "namespace_mutex",
+    out = "namespace_mutex.go",
+    package = "fs",
+    prefix = "namespace",
+)
+
+declare_mutex(
+    name = "inotify_event_mutex",
+    out = "inotify_event_mutex.go",
+    package = "fs",
+    prefix = "inotifyEvent",
+)
+
+declare_mutex(
+    name = "inotify_mutex",
+    out = "inotify_mutex.go",
+    package = "fs",
+    prefix = "inotify",
+)
+
 go_test(
     name = "fs_x_test",
     size = "small",

pkg/sentry/fs/inotify.go

@@ -25,7 +25,6 @@ import (
 	"gvisor.dev/gvisor/pkg/sentry/arch"
 	"gvisor.dev/gvisor/pkg/sentry/memmap"
 	"gvisor.dev/gvisor/pkg/sentry/uniqueid"
-	"gvisor.dev/gvisor/pkg/sync"
 	"gvisor.dev/gvisor/pkg/usermem"
 	"gvisor.dev/gvisor/pkg/waiter"
 )
@@ -50,7 +49,7 @@ type Inotify struct {
 	// while queuing events, a watch needs to lock the event queue, and using mu
 	// for that would violate lock ordering since at that point the calling
 	// goroutine already holds Watch.target.Watches.mu.
-	evMu sync.Mutex `state:"nosave"`
+	evMu inotifyEventMutex `state:"nosave"`
 
 	// A list of pending events for this inotify instance. Protected by evMu.
 	events eventList
@@ -60,7 +59,7 @@ type Inotify struct {
 	scratch []byte
 
 	// mu protects the fields below.
-	mu sync.Mutex `state:"nosave"`
+	mu inotifyMutex `state:"nosave"`
 
 	// The next watch descriptor number to use for this inotify instance. Note
 	// that Linux starts numbering watch descriptors from 1.

pkg/sentry/fs/mounts.go

@@ -23,7 +23,6 @@ import (
 	"gvisor.dev/gvisor/pkg/errors/linuxerr"
 	"gvisor.dev/gvisor/pkg/refs"
 	"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
-	"gvisor.dev/gvisor/pkg/sync"
 )
 
 // DefaultTraversalLimit provides a sensible default traversal limit that may
@@ -151,7 +150,7 @@ type MountNamespace struct {
 	root *Dirent
 
 	// mu protects mounts and mountID counter.
-	mu sync.Mutex `state:"nosave"`
+	mu namespaceMutex `state:"nosave"`
 
 	// mounts is a map of mounted Dirent -> Mount object. There are three
 	// possible cases:

pkg/sentry/fsimpl/cgroupfs/BUILD

@@ -1,8 +1,23 @@
 load("//tools:defs.bzl", "go_library", "go_test")
 load("//tools/go_generics:defs.bzl", "go_template_instance")
+load("//pkg/sync/locking:locking.bzl", "declare_mutex", "declare_rwmutex")
 
 licenses(["notice"])
 
+declare_mutex(
+    name = "pids_controller_mutex",
+    out = "pids_controller_mutex.go",
+    package = "cgroupfs",
+    prefix = "pidsController",
+)
+
+declare_rwmutex(
+    name = "task_mutex",
+    out = "task_mutex.go",
+    package = "cgroupfs",
+    prefix = "task",
+)
+
 go_template_instance(
     name = "dir_refs",
     out = "dir_refs.go",
@@ -27,6 +42,8 @@ go_library(
         "job.go",
         "memory.go",
         "pids.go",
+        "pids_controller_mutex.go",
+        "task_mutex.go",
     ],
     visibility = ["//pkg/sentry:internal"],
     deps = [
@@ -49,6 +66,7 @@ go_library(
         "//pkg/sentry/usage",
         "//pkg/sentry/vfs",
         "//pkg/sync",
+        "//pkg/sync/locking",
         "//pkg/usermem",
     ],
 )

pkg/sentry/fsimpl/cgroupfs/cgroupfs.go

@@ -72,7 +72,6 @@ import (
 	"gvisor.dev/gvisor/pkg/sentry/kernel"
 	"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
 	"gvisor.dev/gvisor/pkg/sentry/vfs"
-	"gvisor.dev/gvisor/pkg/sync"
 	"gvisor.dev/gvisor/pkg/usermem"
 )
 
@@ -173,7 +172,7 @@ type filesystem struct {
 
 	// tasksMu serializes task membership changes across all cgroups within a
 	// filesystem.
-	tasksMu sync.RWMutex `state:"nosave"`
+	tasksMu taskRWMutex `state:"nosave"`
 }
 
 // InitializeHierarchyID implements kernel.cgroupFS.InitializeHierarchyID.

pkg/sentry/fsimpl/cgroupfs/pids.go

@@ -27,7 +27,6 @@ import (
 	"gvisor.dev/gvisor/pkg/sentry/kernel"
 	"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
 	"gvisor.dev/gvisor/pkg/sentry/vfs"
-	"gvisor.dev/gvisor/pkg/sync"
 	"gvisor.dev/gvisor/pkg/usermem"
 )
 
@@ -61,7 +60,7 @@ type pidsController struct {
 	isRoot bool
 
 	// mu protects the fields below.
-	mu sync.Mutex `state:"nosave"`
+	mu pidsControllerMutex `state:"nosave"`
 
 	// pendingTotal and pendingPool tracks the charge for processes starting
 	// up. During startup, we check if PIDs are available by charging the

pkg/sentry/fsimpl/kernfs/BUILD

@@ -1,5 +1,6 @@
 load("//tools:defs.bzl", "go_library", "go_test")
 load("//tools/go_generics:defs.bzl", "go_template_instance")
+load("//pkg/sync/locking:locking.bzl", "declare_mutex", "declare_rwmutex")
 
 licenses(["notice"])
 
@@ -82,13 +83,29 @@ go_template_instance(
     },
 )
 
+declare_rwmutex(
+    name = "filesystem_mutex",
+    out = "filesystem_mutex.go",
+    package = "kernfs",
+    prefix = "filesystem",
+)
+
+declare_mutex(
+    name = "deferred_dec_refs_mutex",
+    out = "deferred_dec_refs_mutex.go",
+    package = "kernfs",
+    prefix = "deferredDecRefs",
+)
+
 go_library(
     name = "kernfs",
     srcs = [
+        "deferred_dec_refs_mutex.go",
         "dentry_list.go",
         "dynamic_bytes_file.go",
         "fd_impl_util.go",
         "filesystem.go",
+        "filesystem_mutex.go",
         "fstree.go",
         "inode_impl_util.go",
         "kernfs.go",
@@ -120,6 +137,7 @@ go_library(
         "//pkg/sentry/socket/unix/transport",
         "//pkg/sentry/vfs",
         "//pkg/sync",
+        "//pkg/sync/locking",
         "//pkg/usermem",
     ],
 )

pkg/sentry/fsimpl/kernfs/kernfs.go

@@ -80,7 +80,7 @@ import (
 type Filesystem struct {
 	vfsfs vfs.Filesystem
 
-	deferredDecRefsMu sync.Mutex `state:"nosave"`
+	deferredDecRefsMu deferredDecRefsMutex `state:"nosave"`
 
 	// deferredDecRefs is a list of dentries waiting to be DecRef()ed. This is
 	// used to defer dentry destruction until mu can be acquired for
@@ -108,7 +108,7 @@ type Filesystem struct {
 	//   defer fs.mu.RUnlock()
 	//   ...
 	//   fs.deferDecRef(dentry)
-	mu sync.RWMutex `state:"nosave"`
+	mu filesystemRWMutex `state:"nosave"`
 
 	// nextInoMinusOne is used to to allocate inode numbers on this
 	// filesystem. Must be accessed by atomic operations.

pkg/sentry/vfs/BUILD

@@ -1,8 +1,44 @@
 load("//tools:defs.bzl", "go_library", "go_test")
 load("//tools/go_generics:defs.bzl", "go_template_instance")
+load("//pkg/sync/locking:locking.bzl", "declare_mutex")
 
 licenses(["notice"])
 
+declare_mutex(
+    name = "virtual_filesystem_mutex",
+    out = "virtual_filesystem_mutex.go",
+    package = "vfs",
+    prefix = "virtualFilesystem",
+)
+
+declare_mutex(
+    name = "inotify_event_mutex",
+    out = "inotify_event_mutex.go",
+    package = "vfs",
+    prefix = "inotifyEvent",
+)
+
+declare_mutex(
+    name = "inotify_mutex",
+    out = "inotify_mutex.go",
+    package = "vfs",
+    prefix = "inotify",
+)
+
+declare_mutex(
+    name = "epoll_instance_mutex",
+    out = "epoll_instance_mutex.go",
+    package = "vfs",
+    prefix = "epollReadyInstance",
+)
+
+declare_mutex(
+    name = "epoll_mutex",
+    out = "epoll_mutex.go",
+    package = "vfs",
+    prefix = "epoll",
+)
+
 go_template_instance(
     name = "epoll_interest_list",
     out = "epoll_interest_list.go",
@@ -69,7 +105,9 @@ go_library(
         "dentry.go",
         "device.go",
         "epoll.go",
+        "epoll_instance_mutex.go",
         "epoll_interest_list.go",
+        "epoll_mutex.go",
         "event_list.go",
         "file_description.go",
         "file_description_impl_util.go",
@@ -79,6 +117,8 @@ go_library(
         "filesystem_refs.go",
         "filesystem_type.go",
         "inotify.go",
+        "inotify_event_mutex.go",
+        "inotify_mutex.go",
         "lock.go",
         "mount.go",
         "mount_namespace_refs.go",
@@ -90,6 +130,7 @@ go_library(
         "resolving_path.go",
         "save_restore.go",
         "vfs.go",
+        "virtual_filesystem_mutex.go",
     ],
     visibility = ["//pkg/sentry:internal"],
     deps = [
@@ -117,6 +158,7 @@ go_library(
         "//pkg/sentry/socket/unix/transport",
         "//pkg/sentry/uniqueid",
         "//pkg/sync",
+        "//pkg/sync/locking",
         "//pkg/usermem",
         "//pkg/waiter",
         "@org_golang_x_sys//unix:go_default_library",

pkg/sentry/vfs/epoll.go

@@ -50,7 +50,7 @@ type EpollInstance struct {
 	// readyMu protects ready, readySeq, epollInterest.ready, and
 	// epollInterest.epollInterestEntry. ready is analogous to Linux's struct
 	// eventpoll::lock.
-	readyMu sync.Mutex `state:"nosave"`
+	readyMu epollReadyInstanceMutex `state:"nosave"`
 
 	// ready is the set of file descriptors that may be "ready" for I/O. Note
 	// that this must be an ordered list, not a map: "If more than maxevents

pkg/sentry/vfs/file_description.go

@@ -60,7 +60,7 @@ type FileDescription struct {
 
 	// epolls is the set of epollInterests registered for this FileDescription.
 	// epolls is protected by epollMu.
-	epollMu sync.Mutex `state:"nosave"`
+	epollMu epollMutex `state:"nosave"`
 	epolls  map[*epollInterest]struct{}
 
 	// vd is the filesystem location at which this FileDescription was opened.

pkg/sentry/vfs/inotify.go

@@ -71,7 +71,7 @@ type Inotify struct {
 	// evMu *only* protects the events list. We need a separate lock while
 	// queuing events: using mu may violate lock ordering, since at that point
 	// the calling goroutine may already hold Watches.mu.
-	evMu sync.Mutex `state:"nosave"`
+	evMu inotifyEventMutex `state:"nosave"`
 
 	// A list of pending events for this inotify instance. Protected by evMu.
 	events eventList
@@ -81,7 +81,7 @@ type Inotify struct {
 	scratch []byte
 
 	// mu protects the fields below.
-	mu sync.Mutex `state:"nosave"`
+	mu inotifyMutex `state:"nosave"`
 
 	// nextWatchMinusOne is used to allocate watch descriptors on this Inotify
 	// instance. Note that Linux starts numbering watch descriptors from 1.

pkg/sentry/vfs/vfs.go

@@ -60,7 +60,7 @@ type VirtualFilesystem struct {
 	// mountMu serializes mount mutations.
 	//
 	// mountMu is analogous to Linux's namespace_sem.
-	mountMu sync.Mutex `state:"nosave"`
+	mountMu virtualFilesystemMutex `state:"nosave"`
 
 	// mounts maps (mount parent, mount point) pairs to mounts. (Since mounts
 	// are uniquely namespaced, including mount parent in the key correctly