netbird/gvisor
0
0
Fork 0
mirror of https://github.com/netbirdio/gvisor.git synced 2026-05-22 17:12:49 -07:00
Code Issues Packages Projects Releases Wiki Activity

io_uring_enter: adding support for READV operation.

Browse Source 
Handling NOP operations helped us to introduce facilities to process submission
requests. Now we can proceed to adding more meaningful operations. The logical
choice for the first one is the READV operation as it is a next operation after
NOP and it is also used for the cat-like application example based of IO_URING.

PiperOrigin-RevId: 487615618
This commit is contained in:
Sergey Madaminov
2022-11-10 12:54:50 -08:00
committed by gVisor bot
parent ae136df849
commit 60f83d99ea
7 changed files with 630 additions and 36 deletions
Download Patch File Download Diff File Expand all files Collapse all files
pkg/abi/linux/iouring.go
+17 -6
View File
@@ -26,6 +26,11 @@ const (
IORING_SETUP_SUBMIT_ALL = (1 << 7)
)
// Constants for io_uring_enter(2). See include/uapi/linux/io_uring.h.
const (
IORING_ENTER_GETEVENTS = (1 << 0)
)
// Constants for IoUringParams.Features. See include/uapi/linux/io_uring.h.
const (
IORING_FEAT_SINGLE_MMAP = (1 << 0)
@@ -53,6 +58,12 @@ const (
IORING_OFF_SQES = 0x10000000
)
// Constants for the IO_URING opcodes. See include/uapi/linux/io_uring.h.
const (
IORING_OP_NOP = 0
IORING_OP_READV = 1
)
// IORingIndex represents SQE array indexes.
//
// +marshal
@@ -161,14 +172,14 @@ type IORings struct {
type IOUringSqe struct {
Opcode uint8
Flags uint8
ioPrio uint16
fd int32
offOrAddrOrCmdOp uint64
addrOrSpliceOff uint64
len uint32
IoPrio uint16
Fd int32
OffOrAddrOrCmdOp uint64
AddrOrSpliceOff uint64
Len uint32
specialFlags uint32
UserData uint64
bufIndexOrGroup uint16
BufIndexOrGroup uint16
personality uint16
spliceFDOrFileIndex int32
addr3 uint64
pkg/sentry/fsimpl/iouringfs/BUILD
+2
View File
@@ -13,10 +13,12 @@ go_library(
"//pkg/errors/linuxerr",
"//pkg/hostarch",
"//pkg/safemem",
"//pkg/sentry/kernel",
"//pkg/sentry/memmap",
"//pkg/sentry/pgalloc",
"//pkg/sentry/usage",
"//pkg/sentry/vfs",
"//pkg/usermem",
],
)
pkg/sentry/fsimpl/iouringfs/iouringfs.go
+70 -17
View File
@@ -32,10 +32,12 @@ import (
"gvisor.dev/gvisor/pkg/errors/linuxerr"
"gvisor.dev/gvisor/pkg/hostarch"
"gvisor.dev/gvisor/pkg/safemem"
"gvisor.dev/gvisor/pkg/sentry/kernel"
"gvisor.dev/gvisor/pkg/sentry/memmap"
"gvisor.dev/gvisor/pkg/sentry/pgalloc"
"gvisor.dev/gvisor/pkg/sentry/usage"
"gvisor.dev/gvisor/pkg/sentry/vfs"
"gvisor.dev/gvisor/pkg/usermem"
)
// FileDescription implements vfs.FileDescriptionImpl for file-based IO_URING.
@@ -313,7 +315,7 @@ func (fd *FileDescription) ConfigureMMap(ctx context.Context, opts *memmap.MMapO
}
// ProcessSubmissions processes submission requests.
func (fd *FileDescription) ProcessSubmissions(toSubmit uint32, minComplete uint32, flags uint32) (int, error) {
func (fd *FileDescription) ProcessSubmissions(t *kernel.Task, toSubmit uint32, minComplete uint32, flags uint32) (int, error) {
fd.mu.Lock()
defer fd.mu.Unlock()
@@ -345,10 +347,7 @@ func (fd *FileDescription) ProcessSubmissions(toSubmit uint32, minComplete uint3
return -1, err
}
cqe, err := fd.ProcessSubmission(&sqe, flags)
if err != nil {
return -1, err
}
cqe := fd.ProcessSubmission(t, &sqe, flags)
sqHead.Add(1)
if (cqTail.Load()-cqHead.Load())/ioRings.CqRingEntries == 1 {
ioRings.CqOverflow++
@@ -371,21 +370,75 @@ func (fd *FileDescription) ProcessSubmissions(toSubmit uint32, minComplete uint3
}
// ProcessSubmission processes a single submission request.
func (fd *FileDescription) ProcessSubmission(sqe *linux.IOUringSqe, flags uint32) (*linux.IOUringCqe, error) {
func (fd *FileDescription) ProcessSubmission(t *kernel.Task, sqe *linux.IOUringSqe, flags uint32) *linux.IOUringCqe {
var (
cqeErr error
cqeFlags uint32
retValue int32
)
switch op := sqe.Opcode; op {
case 0: // NOP
return &linux.IOUringCqe{
UserData: sqe.UserData,
Res: 0,
Flags: 0,
}, nil
case linux.IORING_OP_NOP:
// For the NOP operation, we don't do anything special.
case linux.IORING_OP_READV:
retValue, cqeErr = fd.handleReadv(t, sqe, flags)
default: // Unsupported operation
return &linux.IOUringCqe{
UserData: sqe.UserData,
Res: -int32(linuxerr.EINVAL.Errno()),
Flags: 0,
}, nil
retValue = -int32(linuxerr.EINVAL.Errno())
}
if cqeErr != nil {
retValue = -int32(kernel.ExtractErrno(cqeErr, -1))
}
return &linux.IOUringCqe{
UserData: sqe.UserData,
Res: retValue,
Flags: cqeFlags,
}
}
// handleReadv handles IORING_OP_READV.
func (fd *FileDescription) handleReadv(t *kernel.Task, sqe *linux.IOUringSqe, flags uint32) (int32, error) {
// Check that a file descriptor is valid.
if sqe.Fd < 0 {
return 0, linuxerr.EBADF
}
// Currently we don't support any flags for the SQEs.
if sqe.Flags != 0 {
return 0, linuxerr.EINVAL
}
// If the file is not seekable then offset must be zero. And currently, we don't support them.
if sqe.OffOrAddrOrCmdOp != 0 {
return 0, linuxerr.EINVAL
}
// ioprio should not be set for the READV operation.
if sqe.IoPrio != 0 {
return 0, linuxerr.EINVAL
}
// buf_index should not be set for the READV operation.
if sqe.BufIndexOrGroup != 0 {
return 0, linuxerr.EINVAL
}
// AddressSpaceActive is set to true as we are doing this from the task goroutine.And this is a
// case as we currently don't support neither IOPOLL nor SQPOLL modes.
dst, err := t.IovecsIOSequence(hostarch.Addr(sqe.AddrOrSpliceOff), int(sqe.Len), usermem.IOOpts{
AddressSpaceActive: true,
})
if err != nil {
return 0, err
}
file := t.GetFileVFS2(sqe.Fd)
if file == nil {
return 0, linuxerr.EBADF
}
defer file.DecRef(t)
n, err := file.PRead(t, dst, 0, vfs.ReadOptions{})
if err != nil {
return 0, err
}
return int32(n), nil
}
// updateCq updates a completion queue by adding a given completion queue entry.
pkg/sentry/syscalls/linux/vfs2/iouringfs.go
+2 -2
View File
@@ -86,7 +86,7 @@ func IOUringEnter(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.
ret := -1
// List of currently supported flags for io_uring_enter(2).
const supportedFlags = 0 // Currently support none
const supportedFlags = linux.IORING_ENTER_GETEVENTS
// Since we don't implement everything, we fail explicitly on flags that are unimplemented.
if flags|supportedFlags != supportedFlags {
@@ -112,7 +112,7 @@ func IOUringEnter(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.
if !ok {
return uintptr(ret), nil, linuxerr.EBADF
}
ret, err := iouringfd.ProcessSubmissions(toSubmit, minComplete, flags)
ret, err := iouringfd.ProcessSubmissions(t, toSubmit, minComplete, flags)
if err != nil {
return uintptr(ret), nil, err
}
test/syscalls/linux/io_uring.cc
+530 -11
View File
File diff suppressed because it is too large Load Diff
test/util/BUILD
+2
View File
@@ -68,6 +68,8 @@ cc_library(
":file_descriptor",
":posix_error",
":save_util",
"//test/util:temp_path",
"//test/util:test_util",
],
)
test/util/io_uring_util.h
+7
View File
@@ -16,6 +16,7 @@
#define GVISOR_TEST_UTIL_IOURING_UTIL_H_
#include <linux/fs.h>
#include <sys/ioctl.h>
#include <sys/mman.h>
#include <atomic>
@@ -36,6 +37,9 @@ namespace testing {
#define IORING_SETUP_SQPOLL (1U << 1)
#define IORING_SETUP_CQSIZE (1U << 3)
// io_uring_enter(2) flags
#define IORING_ENTER_GETEVENTS (1U << 0)
#define IORING_FEAT_SINGLE_MMAP (1U << 0)
#define IORING_OFF_SQ_RING 0ULL
@@ -44,6 +48,9 @@ namespace testing {
// IO_URING operation codes.
#define IORING_OP_NOP 0
#define IORING_OP_READV 1
#define BLOCK_SZ kPageSize
struct io_sqring_offsets {
uint32_t head;