diff options
Diffstat (limited to 'pkg/sentry/syscalls')
67 files changed, 835 insertions, 825 deletions
diff --git a/pkg/sentry/syscalls/epoll.go b/pkg/sentry/syscalls/epoll.go index d9fb808c0..d23a0068a 100644 --- a/pkg/sentry/syscalls/epoll.go +++ b/pkg/sentry/syscalls/epoll.go @@ -28,7 +28,7 @@ import ( // CreateEpoll implements the epoll_create(2) linux syscall. func CreateEpoll(t *kernel.Task, closeOnExec bool) (int32, error) { file := epoll.NewEventPoll(t) - defer file.DecRef() + defer file.DecRef(t) fd, err := t.NewFDFrom(0, file, kernel.FDFlags{ CloseOnExec: closeOnExec, @@ -47,14 +47,14 @@ func AddEpoll(t *kernel.Task, epfd int32, fd int32, flags epoll.EntryFlags, mask if epollfile == nil { return syserror.EBADF } - defer epollfile.DecRef() + defer epollfile.DecRef(t) // Get the target file id. file := t.GetFile(fd) if file == nil { return syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the epollPoll operations. e, ok := epollfile.FileOperations.(*epoll.EventPoll) @@ -73,14 +73,14 @@ func UpdateEpoll(t *kernel.Task, epfd int32, fd int32, flags epoll.EntryFlags, m if epollfile == nil { return syserror.EBADF } - defer epollfile.DecRef() + defer epollfile.DecRef(t) // Get the target file id. file := t.GetFile(fd) if file == nil { return syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the epollPoll operations. e, ok := epollfile.FileOperations.(*epoll.EventPoll) @@ -99,14 +99,14 @@ func RemoveEpoll(t *kernel.Task, epfd int32, fd int32) error { if epollfile == nil { return syserror.EBADF } - defer epollfile.DecRef() + defer epollfile.DecRef(t) // Get the target file id. file := t.GetFile(fd) if file == nil { return syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the epollPoll operations. e, ok := epollfile.FileOperations.(*epoll.EventPoll) @@ -115,7 +115,7 @@ func RemoveEpoll(t *kernel.Task, epfd int32, fd int32) error { } // Try to remove the entry. - return e.RemoveEntry(epoll.FileIdentifier{file, fd}) + return e.RemoveEntry(t, epoll.FileIdentifier{file, fd}) } // WaitEpoll implements the epoll_wait(2) linux syscall. @@ -125,7 +125,7 @@ func WaitEpoll(t *kernel.Task, fd int32, max int, timeout int) ([]linux.EpollEve if epollfile == nil { return nil, syserror.EBADF } - defer epollfile.DecRef() + defer epollfile.DecRef(t) // Extract the epollPoll operations. e, ok := epollfile.FileOperations.(*epoll.EventPoll) diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD index 4a9b04fd0..75752b2e6 100644 --- a/pkg/sentry/syscalls/linux/BUILD +++ b/pkg/sentry/syscalls/linux/BUILD @@ -56,6 +56,7 @@ go_library( "sys_xattr.go", "timespec.go", ], + marshal = True, visibility = ["//:sandbox"], deps = [ "//pkg/abi", @@ -64,6 +65,8 @@ go_library( "//pkg/bpf", "//pkg/context", "//pkg/log", + "//pkg/marshal", + "//pkg/marshal/primitive", "//pkg/metric", "//pkg/rand", "//pkg/safemem", @@ -99,7 +102,5 @@ go_library( "//pkg/syserror", "//pkg/usermem", "//pkg/waiter", - "//tools/go_marshal/marshal", - "//tools/go_marshal/primitive", ], ) diff --git a/pkg/sentry/syscalls/linux/error.go b/pkg/sentry/syscalls/linux/error.go index 64de56ac5..dab6207c0 100644 --- a/pkg/sentry/syscalls/linux/error.go +++ b/pkg/sentry/syscalls/linux/error.go @@ -36,8 +36,8 @@ var ( // errors, we may consume the error and return only the partial read/write. // // op and f are used only for panics. -func HandleIOErrorVFS2(t *kernel.Task, partialResult bool, err, intr error, op string, f *vfs.FileDescription) error { - known, err := handleIOErrorImpl(t, partialResult, err, intr, op) +func HandleIOErrorVFS2(t *kernel.Task, partialResult bool, ioerr, intr error, op string, f *vfs.FileDescription) error { + known, err := handleIOErrorImpl(t, partialResult, ioerr, intr, op) if err != nil { return err } @@ -46,7 +46,7 @@ func HandleIOErrorVFS2(t *kernel.Task, partialResult bool, err, intr error, op s fs := f.Mount().Filesystem().VirtualFilesystem() root := vfs.RootFromContext(t) name, _ := fs.PathnameWithDeleted(t, root, f.VirtualDentry()) - log.Traceback("Invalid request partialResult %v and err (type %T) %v for %s operation on %q", partialResult, err, err, op, name) + log.Traceback("Invalid request partialResult %v and err (type %T) %v for %s operation on %q", partialResult, ioerr, ioerr, op, name) partialResultOnce.Do(partialResultMetric.Increment) } return nil @@ -56,15 +56,15 @@ func HandleIOErrorVFS2(t *kernel.Task, partialResult bool, err, intr error, op s // errors, we may consume the error and return only the partial read/write. // // op and f are used only for panics. -func handleIOError(t *kernel.Task, partialResult bool, err, intr error, op string, f *fs.File) error { - known, err := handleIOErrorImpl(t, partialResult, err, intr, op) +func handleIOError(t *kernel.Task, partialResult bool, ioerr, intr error, op string, f *fs.File) error { + known, err := handleIOErrorImpl(t, partialResult, ioerr, intr, op) if err != nil { return err } if !known { // An unknown error is encountered with a partial read/write. name, _ := f.Dirent.FullName(nil /* ignore chroot */) - log.Traceback("Invalid request partialResult %v and err (type %T) %v for %s operation on %q, %T", partialResult, err, err, op, name, f.FileOperations) + log.Traceback("Invalid request partialResult %v and err (type %T) %v for %s operation on %q, %T", partialResult, ioerr, ioerr, op, name, f.FileOperations) partialResultOnce.Do(partialResultMetric.Increment) } return nil @@ -147,7 +147,7 @@ func handleIOErrorImpl(t *kernel.Task, partialResult bool, err, intr error, op s } switch err.(type) { - case kernel.SyscallRestartErrno: + case syserror.SyscallRestartErrno: // Identical to the EINTR case. return true, nil } diff --git a/pkg/sentry/syscalls/linux/linux64.go b/pkg/sentry/syscalls/linux/linux64.go index 80c65164a..5f26697d2 100644 --- a/pkg/sentry/syscalls/linux/linux64.go +++ b/pkg/sentry/syscalls/linux/linux64.go @@ -138,7 +138,7 @@ var AMD64 = &kernel.SyscallTable{ 83: syscalls.Supported("mkdir", Mkdir), 84: syscalls.Supported("rmdir", Rmdir), 85: syscalls.Supported("creat", Creat), - 86: syscalls.Supported("link", Link), + 86: syscalls.PartiallySupported("link", Link, "Limited support with Gofer. Link count and linked files may get out of sync because gVisor is not aware of external hardlinks.", nil), 87: syscalls.Supported("unlink", Unlink), 88: syscalls.Supported("symlink", Symlink), 89: syscalls.Supported("readlink", Readlink), @@ -305,9 +305,9 @@ var AMD64 = &kernel.SyscallTable{ 250: syscalls.Error("keyctl", syserror.EACCES, "Not available to user.", nil), 251: syscalls.CapError("ioprio_set", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_nice or cap_sys_admin (depending) 252: syscalls.CapError("ioprio_get", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_nice or cap_sys_admin (depending) - 253: syscalls.PartiallySupported("inotify_init", InotifyInit, "inotify events are only available inside the sandbox.", nil), - 254: syscalls.PartiallySupported("inotify_add_watch", InotifyAddWatch, "inotify events are only available inside the sandbox.", nil), - 255: syscalls.PartiallySupported("inotify_rm_watch", InotifyRmWatch, "inotify events are only available inside the sandbox.", nil), + 253: syscalls.PartiallySupported("inotify_init", InotifyInit, "Inotify events are only available inside the sandbox. Hard links are treated as different watch targets in gofer fs.", nil), + 254: syscalls.PartiallySupported("inotify_add_watch", InotifyAddWatch, "Inotify events are only available inside the sandbox. Hard links are treated as different watch targets in gofer fs.", nil), + 255: syscalls.PartiallySupported("inotify_rm_watch", InotifyRmWatch, "Inotify events are only available inside the sandbox. Hard links are treated as different watch targets in gofer fs.", nil), 256: syscalls.CapError("migrate_pages", linux.CAP_SYS_NICE, "", nil), 257: syscalls.Supported("openat", Openat), 258: syscalls.Supported("mkdirat", Mkdirat), @@ -317,7 +317,7 @@ var AMD64 = &kernel.SyscallTable{ 262: syscalls.Supported("fstatat", Fstatat), 263: syscalls.Supported("unlinkat", Unlinkat), 264: syscalls.Supported("renameat", Renameat), - 265: syscalls.Supported("linkat", Linkat), + 265: syscalls.PartiallySupported("linkat", Linkat, "See link(2).", nil), 266: syscalls.Supported("symlinkat", Symlinkat), 267: syscalls.Supported("readlinkat", Readlinkat), 268: syscalls.Supported("fchmodat", Fchmodat), @@ -346,7 +346,7 @@ var AMD64 = &kernel.SyscallTable{ 291: syscalls.Supported("epoll_create1", EpollCreate1), 292: syscalls.Supported("dup3", Dup3), 293: syscalls.Supported("pipe2", Pipe2), - 294: syscalls.Supported("inotify_init1", InotifyInit1), + 294: syscalls.PartiallySupported("inotify_init1", InotifyInit1, "Inotify events are only available inside the sandbox. Hard links are treated as different watch targets in gofer fs.", nil), 295: syscalls.Supported("preadv", Preadv), 296: syscalls.Supported("pwritev", Pwritev), 297: syscalls.Supported("rt_tgsigqueueinfo", RtTgsigqueueinfo), @@ -454,9 +454,9 @@ var ARM64 = &kernel.SyscallTable{ 23: syscalls.Supported("dup", Dup), 24: syscalls.Supported("dup3", Dup3), 25: syscalls.PartiallySupported("fcntl", Fcntl, "Not all options are supported.", nil), - 26: syscalls.Supported("inotify_init1", InotifyInit1), - 27: syscalls.PartiallySupported("inotify_add_watch", InotifyAddWatch, "inotify events are only available inside the sandbox.", nil), - 28: syscalls.PartiallySupported("inotify_rm_watch", InotifyRmWatch, "inotify events are only available inside the sandbox.", nil), + 26: syscalls.PartiallySupported("inotify_init1", InotifyInit1, "Inotify events are only available inside the sandbox. Hard links are treated as different watch targets in gofer fs.", nil), + 27: syscalls.PartiallySupported("inotify_add_watch", InotifyAddWatch, "Inotify events are only available inside the sandbox. Hard links are treated as different watch targets in gofer fs.", nil), + 28: syscalls.PartiallySupported("inotify_rm_watch", InotifyRmWatch, "Inotify events are only available inside the sandbox. Hard links are treated as different watch targets in gofer fs.", nil), 29: syscalls.PartiallySupported("ioctl", Ioctl, "Only a few ioctls are implemented for backing devices and file systems.", nil), 30: syscalls.CapError("ioprio_set", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_nice or cap_sys_admin (depending) 31: syscalls.CapError("ioprio_get", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_nice or cap_sys_admin (depending) diff --git a/pkg/sentry/syscalls/linux/sys_aio.go b/pkg/sentry/syscalls/linux/sys_aio.go index ba2557c52..0bf313a13 100644 --- a/pkg/sentry/syscalls/linux/sys_aio.go +++ b/pkg/sentry/syscalls/linux/sys_aio.go @@ -17,6 +17,7 @@ package linux import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -36,7 +37,7 @@ func IoSetup(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca // // The context pointer _must_ be zero initially. var idIn uint64 - if _, err := t.CopyIn(idAddr, &idIn); err != nil { + if _, err := primitive.CopyUint64In(t, idAddr, &idIn); err != nil { return 0, nil, err } if idIn != 0 { @@ -49,7 +50,7 @@ func IoSetup(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca } // Copy out the new ID. - if _, err := t.CopyOut(idAddr, &id); err != nil { + if _, err := primitive.CopyUint64Out(t, idAddr, id); err != nil { t.MemoryManager().DestroyAIOContext(t, id) return 0, nil, err } @@ -142,7 +143,7 @@ func IoGetevents(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S ev := v.(*linux.IOEvent) // Copy out the result. - if _, err := t.CopyOut(eventsAddr, ev); err != nil { + if _, err := ev.CopyOut(t, eventsAddr); err != nil { if count > 0 { return uintptr(count), nil, nil } @@ -247,7 +248,7 @@ func getAIOCallback(t *kernel.Task, file *fs.File, cbAddr usermem.Addr, cb *linu ev.Result = -int64(kernel.ExtractErrno(err, 0)) } - file.DecRef() + file.DecRef(ctx) // Queue the result for delivery. actx.FinishRequest(ev) @@ -257,7 +258,7 @@ func getAIOCallback(t *kernel.Task, file *fs.File, cbAddr usermem.Addr, cb *linu // wake up. if eventFile != nil { eventFile.FileOperations.(*eventfd.EventOperations).Signal(1) - eventFile.DecRef() + eventFile.DecRef(ctx) } } } @@ -269,7 +270,7 @@ func submitCallback(t *kernel.Task, id uint64, cb *linux.IOCallback, cbAddr user // File not found. return syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Was there an eventFD? Extract it. var eventFile *fs.File @@ -279,7 +280,7 @@ func submitCallback(t *kernel.Task, id uint64, cb *linux.IOCallback, cbAddr user // Bad FD. return syserror.EBADF } - defer eventFile.DecRef() + defer eventFile.DecRef(t) // Check that it is an eventfd. if _, ok := eventFile.FileOperations.(*eventfd.EventOperations); !ok { @@ -338,21 +339,27 @@ func IoSubmit(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc } for i := int32(0); i < nrEvents; i++ { - // Copy in the address. - cbAddrNative := t.Arch().Native(0) - if _, err := t.CopyIn(addr, cbAddrNative); err != nil { - if i > 0 { - // Some successful. - return uintptr(i), nil, nil + // Copy in the callback address. + var cbAddr usermem.Addr + switch t.Arch().Width() { + case 8: + var cbAddrP primitive.Uint64 + if _, err := cbAddrP.CopyIn(t, addr); err != nil { + if i > 0 { + // Some successful. + return uintptr(i), nil, nil + } + // Nothing done. + return 0, nil, err } - // Nothing done. - return 0, nil, err + cbAddr = usermem.Addr(cbAddrP) + default: + return 0, nil, syserror.ENOSYS } // Copy in this callback. var cb linux.IOCallback - cbAddr := usermem.Addr(t.Arch().Value(cbAddrNative)) - if _, err := t.CopyIn(cbAddr, &cb); err != nil { + if _, err := cb.CopyIn(t, cbAddr); err != nil { if i > 0 { // Some have been successful. diff --git a/pkg/sentry/syscalls/linux/sys_capability.go b/pkg/sentry/syscalls/linux/sys_capability.go index adf5ea5f2..d3b85e11b 100644 --- a/pkg/sentry/syscalls/linux/sys_capability.go +++ b/pkg/sentry/syscalls/linux/sys_capability.go @@ -45,7 +45,7 @@ func Capget(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal dataAddr := args[1].Pointer() var hdr linux.CapUserHeader - if _, err := t.CopyIn(hdrAddr, &hdr); err != nil { + if _, err := hdr.CopyIn(t, hdrAddr); err != nil { return 0, nil, err } // hdr.Pid doesn't need to be valid if this capget() is a "version probe" @@ -65,7 +65,7 @@ func Capget(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal Permitted: uint32(p), Inheritable: uint32(i), } - _, err = t.CopyOut(dataAddr, &data) + _, err = data.CopyOut(t, dataAddr) return 0, nil, err case linux.LINUX_CAPABILITY_VERSION_2, linux.LINUX_CAPABILITY_VERSION_3: @@ -88,12 +88,12 @@ func Capget(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal Inheritable: uint32(i >> 32), }, } - _, err = t.CopyOut(dataAddr, &data) + _, err = linux.CopyCapUserDataSliceOut(t, dataAddr, data[:]) return 0, nil, err default: hdr.Version = linux.HighestCapabilityVersion - if _, err := t.CopyOut(hdrAddr, &hdr); err != nil { + if _, err := hdr.CopyOut(t, hdrAddr); err != nil { return 0, nil, err } if dataAddr != 0 { @@ -109,7 +109,7 @@ func Capset(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal dataAddr := args[1].Pointer() var hdr linux.CapUserHeader - if _, err := t.CopyIn(hdrAddr, &hdr); err != nil { + if _, err := hdr.CopyIn(t, hdrAddr); err != nil { return 0, nil, err } switch hdr.Version { @@ -118,7 +118,7 @@ func Capset(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal return 0, nil, syserror.EPERM } var data linux.CapUserData - if _, err := t.CopyIn(dataAddr, &data); err != nil { + if _, err := data.CopyIn(t, dataAddr); err != nil { return 0, nil, err } p := auth.CapabilitySet(data.Permitted) & auth.AllCapabilities @@ -131,7 +131,7 @@ func Capset(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal return 0, nil, syserror.EPERM } var data [2]linux.CapUserData - if _, err := t.CopyIn(dataAddr, &data); err != nil { + if _, err := linux.CopyCapUserDataSliceIn(t, dataAddr, data[:]); err != nil { return 0, nil, err } p := (auth.CapabilitySet(data[0].Permitted) | (auth.CapabilitySet(data[1].Permitted) << 32)) & auth.AllCapabilities @@ -141,7 +141,7 @@ func Capset(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal default: hdr.Version = linux.HighestCapabilityVersion - if _, err := t.CopyOut(hdrAddr, &hdr); err != nil { + if _, err := hdr.CopyOut(t, hdrAddr); err != nil { return 0, nil, err } return 0, nil, syserror.EINVAL diff --git a/pkg/sentry/syscalls/linux/sys_eventfd.go b/pkg/sentry/syscalls/linux/sys_eventfd.go index ed3413ca6..3b4f879e4 100644 --- a/pkg/sentry/syscalls/linux/sys_eventfd.go +++ b/pkg/sentry/syscalls/linux/sys_eventfd.go @@ -37,7 +37,7 @@ func Eventfd2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc event.SetFlags(fs.SettableFileFlags{ NonBlocking: flags&linux.EFD_NONBLOCK != 0, }) - defer event.DecRef() + defer event.DecRef(t) fd, err := t.NewFDFrom(0, event, kernel.FDFlags{ CloseOnExec: flags&linux.EFD_CLOEXEC != 0, diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go index 8cf6401e7..98331eb3c 100644 --- a/pkg/sentry/syscalls/linux/sys_file.go +++ b/pkg/sentry/syscalls/linux/sys_file.go @@ -19,6 +19,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/lock" @@ -40,7 +41,7 @@ func fileOpAt(t *kernel.Task, dirFD int32, path string, fn func(root *fs.Dirent, // Common case: we are accessing a file in the root. root := t.FSContext().RootDirectory() err := fn(root, root, name, linux.MaxSymlinkTraversals) - root.DecRef() + root.DecRef(t) return err } else if dir == "." && dirFD == linux.AT_FDCWD { // Common case: we are accessing a file relative to the current @@ -48,8 +49,8 @@ func fileOpAt(t *kernel.Task, dirFD int32, path string, fn func(root *fs.Dirent, wd := t.FSContext().WorkingDirectory() root := t.FSContext().RootDirectory() err := fn(root, wd, name, linux.MaxSymlinkTraversals) - wd.DecRef() - root.DecRef() + wd.DecRef(t) + root.DecRef(t) return err } @@ -97,19 +98,19 @@ func fileOpOn(t *kernel.Task, dirFD int32, path string, resolve bool, fn func(ro } else { d, err = t.MountNamespace().FindLink(t, root, rel, path, &remainingTraversals) } - root.DecRef() + root.DecRef(t) if wd != nil { - wd.DecRef() + wd.DecRef(t) } if f != nil { - f.DecRef() + f.DecRef(t) } if err != nil { return err } err = fn(root, d, remainingTraversals) - d.DecRef() + d.DecRef(t) return err } @@ -184,9 +185,9 @@ func openAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint) (fd uint file, err := d.Inode.GetFile(t, d, fileFlags) if err != nil { - return syserror.ConvertIntr(err, kernel.ERESTARTSYS) + return syserror.ConvertIntr(err, syserror.ERESTARTSYS) } - defer file.DecRef() + defer file.DecRef(t) // Success. newFD, err := t.NewFDFrom(0, file, kernel.FDFlags{ @@ -242,7 +243,7 @@ func mknodAt(t *kernel.Task, dirFD int32, addr usermem.Addr, mode linux.FileMode if err != nil { return err } - file.DecRef() + file.DecRef(t) return nil case linux.ModeNamedPipe: @@ -332,7 +333,7 @@ func createAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint, mode l if err != nil { break } - defer found.DecRef() + defer found.DecRef(t) // We found something (possibly a symlink). If the // O_EXCL flag was passed, then we can immediately @@ -357,7 +358,7 @@ func createAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint, mode l resolved, err = found.Inode.Getlink(t) if err == nil { // No more resolution necessary. - defer resolved.DecRef() + defer resolved.DecRef(t) break } if err != fs.ErrResolveViaReadlink { @@ -384,7 +385,7 @@ func createAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint, mode l if err != nil { break } - defer newParent.DecRef() + defer newParent.DecRef(t) // Repeat the process with the parent and name of the // symlink target. @@ -414,9 +415,9 @@ func createAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint, mode l // Create a new fs.File. newFile, err = found.Inode.GetFile(t, found, fileFlags) if err != nil { - return syserror.ConvertIntr(err, kernel.ERESTARTSYS) + return syserror.ConvertIntr(err, syserror.ERESTARTSYS) } - defer newFile.DecRef() + defer newFile.DecRef(t) case syserror.ENOENT: // File does not exist. Proceed with creation. @@ -432,7 +433,7 @@ func createAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint, mode l // No luck, bail. return err } - defer newFile.DecRef() + defer newFile.DecRef(t) found = newFile.Dirent default: return err @@ -596,24 +597,24 @@ func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Shared flags between file and socket. switch request { case linux.FIONCLEX: - t.FDTable().SetFlags(fd, kernel.FDFlags{ + t.FDTable().SetFlags(t, fd, kernel.FDFlags{ CloseOnExec: false, }) return 0, nil, nil case linux.FIOCLEX: - t.FDTable().SetFlags(fd, kernel.FDFlags{ + t.FDTable().SetFlags(t, fd, kernel.FDFlags{ CloseOnExec: true, }) return 0, nil, nil case linux.FIONBIO: var set int32 - if _, err := t.CopyIn(args[2].Pointer(), &set); err != nil { + if _, err := primitive.CopyInt32In(t, args[2].Pointer(), &set); err != nil { return 0, nil, err } flags := file.Flags() @@ -627,7 +628,7 @@ func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.FIOASYNC: var set int32 - if _, err := t.CopyIn(args[2].Pointer(), &set); err != nil { + if _, err := primitive.CopyInt32In(t, args[2].Pointer(), &set); err != nil { return 0, nil, err } flags := file.Flags() @@ -641,15 +642,14 @@ func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.FIOSETOWN, linux.SIOCSPGRP: var set int32 - if _, err := t.CopyIn(args[2].Pointer(), &set); err != nil { + if _, err := primitive.CopyInt32In(t, args[2].Pointer(), &set); err != nil { return 0, nil, err } fSetOwn(t, file, set) return 0, nil, nil case linux.FIOGETOWN, linux.SIOCGPGRP: - who := fGetOwn(t, file) - _, err := t.CopyOut(args[2].Pointer(), &who) + _, err := primitive.CopyInt32Out(t, args[2].Pointer(), fGetOwn(t, file)) return 0, nil, err default: @@ -671,9 +671,9 @@ func Getcwd(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal addr := args[0].Pointer() size := args[1].SizeT() cwd := t.FSContext().WorkingDirectory() - defer cwd.DecRef() + defer cwd.DecRef(t) root := t.FSContext().RootDirectory() - defer root.DecRef() + defer root.DecRef(t) // Get our fullname from the root and preprend unreachable if the root was // unreachable from our current dirent this is the same behavior as on linux. @@ -694,7 +694,7 @@ func Getcwd(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal } // Top it off with a terminator. - _, err = t.CopyOut(addr+usermem.Addr(bytes), []byte("\x00")) + _, err = t.CopyOutBytes(addr+usermem.Addr(bytes), []byte("\x00")) return uintptr(bytes + 1), nil, err } @@ -722,7 +722,7 @@ func Chroot(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal return err } - t.FSContext().SetRootDirectory(d) + t.FSContext().SetRootDirectory(t, d) return nil }) } @@ -747,7 +747,7 @@ func Chdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return err } - t.FSContext().SetWorkingDirectory(d) + t.FSContext().SetWorkingDirectory(t, d) return nil }) } @@ -760,7 +760,7 @@ func Fchdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Is it a directory? if !fs.IsDir(file.Dirent.Inode.StableAttr) { @@ -772,7 +772,7 @@ func Fchdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal return 0, nil, err } - t.FSContext().SetWorkingDirectory(file.Dirent) + t.FSContext().SetWorkingDirectory(t, file.Dirent) return 0, nil, nil } @@ -787,11 +787,11 @@ func Close(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // Note that Remove provides a reference on the file that we may use to // flush. It is still active until we drop the final reference below // (and other reference-holding operations complete). - file, _ := t.FDTable().Remove(fd) + file, _ := t.FDTable().Remove(t, fd) if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) err := file.Flush(t) return 0, nil, handleIOError(t, false /* partial */, err, syserror.EINTR, "close", file) @@ -805,7 +805,7 @@ func Dup(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) newFD, err := t.NewFDFrom(0, file, kernel.FDFlags{}) if err != nil { @@ -826,7 +826,7 @@ func Dup2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if oldFile == nil { return 0, nil, syserror.EBADF } - defer oldFile.DecRef() + defer oldFile.DecRef(t) return uintptr(newfd), nil, nil } @@ -850,7 +850,7 @@ func Dup3(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if oldFile == nil { return 0, nil, syserror.EBADF } - defer oldFile.DecRef() + defer oldFile.DecRef(t) err := t.NewFDAt(newfd, oldFile, kernel.FDFlags{CloseOnExec: flags&linux.O_CLOEXEC != 0}) if err != nil { @@ -925,7 +925,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) switch cmd { case linux.F_DUPFD, linux.F_DUPFD_CLOEXEC: @@ -941,7 +941,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return uintptr(flags.ToLinuxFDFlags()), nil, nil case linux.F_SETFD: flags := args[2].Uint() - err := t.FDTable().SetFlags(fd, kernel.FDFlags{ + err := t.FDTable().SetFlags(t, fd, kernel.FDFlags{ CloseOnExec: flags&linux.FD_CLOEXEC != 0, }) return 0, nil, err @@ -962,7 +962,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // Copy in the lock request. flockAddr := args[2].Pointer() var flock linux.Flock - if _, err := t.CopyIn(flockAddr, &flock); err != nil { + if _, err := flock.CopyIn(t, flockAddr); err != nil { return 0, nil, err } @@ -1052,12 +1052,12 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.F_GETOWN_EX: addr := args[2].Pointer() owner := fGetOwnEx(t, file) - _, err := t.CopyOut(addr, &owner) + _, err := owner.CopyOut(t, addr) return 0, nil, err case linux.F_SETOWN_EX: addr := args[2].Pointer() var owner linux.FOwnerEx - _, err := t.CopyIn(addr, &owner) + _, err := owner.CopyIn(t, addr) if err != nil { return 0, nil, err } @@ -1132,7 +1132,7 @@ func Fadvise64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // If the FD refers to a pipe or FIFO, return error. if fs.IsPipe(file.Dirent.Inode.StableAttr) { @@ -1154,6 +1154,10 @@ func Fadvise64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys return 0, nil, nil } +// LINT.ThenChange(vfs2/fd.go) + +// LINT.IfChange + func mkdirAt(t *kernel.Task, dirFD int32, addr usermem.Addr, mode linux.FileMode) error { path, _, err := copyInPath(t, addr, false /* allowEmpty */) if err != nil { @@ -1171,7 +1175,7 @@ func mkdirAt(t *kernel.Task, dirFD int32, addr usermem.Addr, mode linux.FileMode switch err { case nil: // The directory existed. - defer f.DecRef() + defer f.DecRef(t) return syserror.EEXIST case syserror.EACCES: // Permission denied while walking to the directory. @@ -1349,7 +1353,7 @@ func linkAt(t *kernel.Task, oldDirFD int32, oldAddr usermem.Addr, newDirFD int32 if target == nil { return syserror.EBADF } - defer target.DecRef() + defer target.DecRef(t) if err := mayLinkAt(t, target.Dirent.Inode); err != nil { return err } @@ -1602,7 +1606,7 @@ func Ftruncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Reject truncation if the file flags do not permit this operation. // This is different from truncate(2) above. @@ -1730,7 +1734,7 @@ func chownAt(t *kernel.Task, fd int32, addr usermem.Addr, resolve, allowEmpty bo if file == nil { return syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return chown(t, file.Dirent, uid, gid) } @@ -1768,7 +1772,7 @@ func Fchown(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return 0, nil, chown(t, file.Dirent, uid, gid) } @@ -1833,7 +1837,7 @@ func Fchmod(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return 0, nil, chmod(t, file.Dirent, mode) } @@ -1893,10 +1897,10 @@ func utimes(t *kernel.Task, dirFD int32, addr usermem.Addr, ts fs.TimeSpec, reso if f == nil { return syserror.EBADF } - defer f.DecRef() + defer f.DecRef(t) root := t.FSContext().RootDirectory() - defer root.DecRef() + defer root.DecRef(t) return setTimestamp(root, f.Dirent, linux.MaxSymlinkTraversals) } @@ -1918,7 +1922,7 @@ func Utime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall ts := defaultSetToSystemTimeSpec() if timesAddr != 0 { var times linux.Utime - if _, err := t.CopyIn(timesAddr, ×); err != nil { + if _, err := times.CopyIn(t, timesAddr); err != nil { return 0, nil, err } ts = fs.TimeSpec{ @@ -1938,7 +1942,7 @@ func Utimes(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal ts := defaultSetToSystemTimeSpec() if timesAddr != 0 { var times [2]linux.Timeval - if _, err := t.CopyIn(timesAddr, ×); err != nil { + if _, err := linux.CopyTimevalSliceIn(t, timesAddr, times[:]); err != nil { return 0, nil, err } ts = fs.TimeSpec{ @@ -1966,7 +1970,7 @@ func Utimensat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys ts := defaultSetToSystemTimeSpec() if timesAddr != 0 { var times [2]linux.Timespec - if _, err := t.CopyIn(timesAddr, ×); err != nil { + if _, err := linux.CopyTimespecSliceIn(t, timesAddr, times[:]); err != nil { return 0, nil, err } if !timespecIsValid(times[0]) || !timespecIsValid(times[1]) { @@ -2000,7 +2004,7 @@ func Futimesat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys ts := defaultSetToSystemTimeSpec() if timesAddr != 0 { var times [2]linux.Timeval - if _, err := t.CopyIn(timesAddr, ×); err != nil { + if _, err := linux.CopyTimevalSliceIn(t, timesAddr, times[:]); err != nil { return 0, nil, err } if times[0].Usec >= 1e6 || times[0].Usec < 0 || @@ -2088,7 +2092,7 @@ func Fallocate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) if offset < 0 || length <= 0 { return 0, nil, syserror.EINVAL @@ -2141,7 +2145,7 @@ func Flock(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // flock(2): EBADF fd is not an open file descriptor. return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) nonblocking := operation&linux.LOCK_NB != 0 operation &^= linux.LOCK_NB @@ -2224,8 +2228,8 @@ func MemfdCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S return 0, nil, err } - defer dirent.DecRef() - defer file.DecRef() + defer dirent.DecRef(t) + defer file.DecRef(t) newFD, err := t.NewFDFrom(0, file, kernel.FDFlags{ CloseOnExec: cloExec, diff --git a/pkg/sentry/syscalls/linux/sys_futex.go b/pkg/sentry/syscalls/linux/sys_futex.go index f04d78856..f39ce0639 100644 --- a/pkg/sentry/syscalls/linux/sys_futex.go +++ b/pkg/sentry/syscalls/linux/sys_futex.go @@ -73,8 +73,8 @@ func futexWaitAbsolute(t *kernel.Task, clockRealtime bool, ts linux.Timespec, fo err = t.BlockWithDeadline(w.C, true, ktime.FromTimespec(ts)) } - t.Futex().WaitComplete(w) - return 0, syserror.ConvertIntr(err, kernel.ERESTARTSYS) + t.Futex().WaitComplete(w, t) + return 0, syserror.ConvertIntr(err, syserror.ERESTARTSYS) } // futexWaitDuration performs a FUTEX_WAIT, blocking until the wait is @@ -95,7 +95,7 @@ func futexWaitDuration(t *kernel.Task, duration time.Duration, forever bool, add } remaining, err := t.BlockWithTimeout(w.C, !forever, duration) - t.Futex().WaitComplete(w) + t.Futex().WaitComplete(w, t) if err == nil { return 0, nil } @@ -110,7 +110,7 @@ func futexWaitDuration(t *kernel.Task, duration time.Duration, forever bool, add // The wait duration was absolute, restart with the original arguments. if forever { - return 0, kernel.ERESTARTSYS + return 0, syserror.ERESTARTSYS } // The wait duration was relative, restart with the remaining duration. @@ -121,7 +121,7 @@ func futexWaitDuration(t *kernel.Task, duration time.Duration, forever bool, add val: val, mask: mask, }) - return 0, kernel.ERESTART_RESTARTBLOCK + return 0, syserror.ERESTART_RESTARTBLOCK } func futexLockPI(t *kernel.Task, ts linux.Timespec, forever bool, addr usermem.Addr, private bool) error { @@ -148,8 +148,8 @@ func futexLockPI(t *kernel.Task, ts linux.Timespec, forever bool, addr usermem.A timer.Destroy() } - t.Futex().WaitComplete(w) - return syserror.ConvertIntr(err, kernel.ERESTARTSYS) + t.Futex().WaitComplete(w, t) + return syserror.ConvertIntr(err, syserror.ERESTARTSYS) } func tryLockPI(t *kernel.Task, addr usermem.Addr, private bool) error { @@ -306,8 +306,8 @@ func GetRobustList(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel // Despite the syscall using the name 'pid' for this variable, it is // very much a tid. tid := args[0].Int() - head := args[1].Pointer() - size := args[2].Pointer() + headAddr := args[1].Pointer() + sizeAddr := args[2].Pointer() if tid < 0 { return 0, nil, syserror.EINVAL @@ -321,12 +321,16 @@ func GetRobustList(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel } // Copy out head pointer. - if _, err := t.CopyOut(head, uint64(ot.GetRobustList())); err != nil { + head := t.Arch().Native(uintptr(ot.GetRobustList())) + if _, err := head.CopyOut(t, headAddr); err != nil { return 0, nil, err } - // Copy out size, which is a constant. - if _, err := t.CopyOut(size, uint64(linux.SizeOfRobustListHead)); err != nil { + // Copy out size, which is a constant. Note that while size isn't + // an address, it is defined as the arch-dependent size_t, so it + // needs to be converted to a native-sized int. + size := t.Arch().Native(uintptr(linux.SizeOfRobustListHead)) + if _, err := size.CopyOut(t, sizeAddr); err != nil { return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/sys_getdents.go b/pkg/sentry/syscalls/linux/sys_getdents.go index b126fecc0..b25f7d881 100644 --- a/pkg/sentry/syscalls/linux/sys_getdents.go +++ b/pkg/sentry/syscalls/linux/sys_getdents.go @@ -19,7 +19,6 @@ import ( "io" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -68,7 +67,7 @@ func getdents(t *kernel.Task, fd int32, addr usermem.Addr, size int, f func(*dir if dir == nil { return 0, syserror.EBADF } - defer dir.DecRef() + defer dir.DecRef(t) w := &usermem.IOReadWriter{ Ctx: t, @@ -82,7 +81,7 @@ func getdents(t *kernel.Task, fd int32, addr usermem.Addr, size int, f func(*dir ds := newDirentSerializer(f, w, t.Arch(), size) rerr := dir.Readdir(t, ds) - switch err := handleIOError(t, ds.Written() > 0, rerr, kernel.ERESTARTSYS, "getdents", dir); err { + switch err := handleIOError(t, ds.Written() > 0, rerr, syserror.ERESTARTSYS, "getdents", dir); err { case nil: dir.Dirent.InotifyEvent(linux.IN_ACCESS, 0) return uintptr(ds.Written()), nil @@ -93,19 +92,23 @@ func getdents(t *kernel.Task, fd int32, addr usermem.Addr, size int, f func(*dir } } -// oldDirentHdr is a fixed sized header matching the fixed size -// fields found in the old linux dirent struct. +// oldDirentHdr is a fixed sized header matching the fixed size fields found in +// the old linux dirent struct. +// +// +marshal type oldDirentHdr struct { Ino uint64 Off uint64 - Reclen uint16 + Reclen uint16 `marshal:"unaligned"` // Struct ends mid-word. } -// direntHdr is a fixed sized header matching the fixed size -// fields found in the new linux dirent struct. +// direntHdr is a fixed sized header matching the fixed size fields found in the +// new linux dirent struct. +// +// +marshal type direntHdr struct { OldHdr oldDirentHdr - Typ uint8 + Typ uint8 `marshal:"unaligned"` // Struct ends mid-word. } // dirent contains the data pointed to by a new linux dirent struct. @@ -134,20 +137,20 @@ func newDirent(width uint, name string, attr fs.DentAttr, offset uint64) *dirent // the old linux dirent format. func smallestDirent(a arch.Context) uint { d := dirent{} - return uint(binary.Size(d.Hdr.OldHdr)) + a.Width() + 1 + return uint(d.Hdr.OldHdr.SizeBytes()) + a.Width() + 1 } // smallestDirent64 returns the size of the smallest possible dirent using // the new linux dirent format. func smallestDirent64(a arch.Context) uint { d := dirent{} - return uint(binary.Size(d.Hdr)) + a.Width() + return uint(d.Hdr.SizeBytes()) + a.Width() } // padRec pads the name field until the rec length is a multiple of the width, // which must be a power of 2. It returns the padded rec length. func (d *dirent) padRec(width int) uint16 { - a := int(binary.Size(d.Hdr)) + len(d.Name) + a := d.Hdr.SizeBytes() + len(d.Name) r := (a + width) &^ (width - 1) padding := r - a d.Name = append(d.Name, make([]byte, padding)...) @@ -157,7 +160,7 @@ func (d *dirent) padRec(width int) uint16 { // Serialize64 serializes a Dirent struct to a byte slice, keeping the new // linux dirent format. Returns the number of bytes serialized or an error. func (d *dirent) Serialize64(w io.Writer) (int, error) { - n1, err := w.Write(binary.Marshal(nil, usermem.ByteOrder, d.Hdr)) + n1, err := d.Hdr.WriteTo(w) if err != nil { return 0, err } @@ -165,14 +168,14 @@ func (d *dirent) Serialize64(w io.Writer) (int, error) { if err != nil { return 0, err } - return n1 + n2, nil + return int(n1) + n2, nil } // Serialize serializes a Dirent struct to a byte slice, using the old linux // dirent format. // Returns the number of bytes serialized or an error. func (d *dirent) Serialize(w io.Writer) (int, error) { - n1, err := w.Write(binary.Marshal(nil, usermem.ByteOrder, d.Hdr.OldHdr)) + n1, err := d.Hdr.OldHdr.WriteTo(w) if err != nil { return 0, err } @@ -184,7 +187,7 @@ func (d *dirent) Serialize(w io.Writer) (int, error) { if err != nil { return 0, err } - return n1 + n2 + n3, nil + return int(n1) + n2 + n3, nil } // direntSerializer implements fs.InodeOperationsInfoSerializer, serializing dirents to an diff --git a/pkg/sentry/syscalls/linux/sys_identity.go b/pkg/sentry/syscalls/linux/sys_identity.go index 715ac45e6..a29d307e5 100644 --- a/pkg/sentry/syscalls/linux/sys_identity.go +++ b/pkg/sentry/syscalls/linux/sys_identity.go @@ -49,13 +49,13 @@ func Getresuid(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys ruid := c.RealKUID.In(c.UserNamespace).OrOverflow() euid := c.EffectiveKUID.In(c.UserNamespace).OrOverflow() suid := c.SavedKUID.In(c.UserNamespace).OrOverflow() - if _, err := t.CopyOut(ruidAddr, ruid); err != nil { + if _, err := ruid.CopyOut(t, ruidAddr); err != nil { return 0, nil, err } - if _, err := t.CopyOut(euidAddr, euid); err != nil { + if _, err := euid.CopyOut(t, euidAddr); err != nil { return 0, nil, err } - if _, err := t.CopyOut(suidAddr, suid); err != nil { + if _, err := suid.CopyOut(t, suidAddr); err != nil { return 0, nil, err } return 0, nil, nil @@ -84,13 +84,13 @@ func Getresgid(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys rgid := c.RealKGID.In(c.UserNamespace).OrOverflow() egid := c.EffectiveKGID.In(c.UserNamespace).OrOverflow() sgid := c.SavedKGID.In(c.UserNamespace).OrOverflow() - if _, err := t.CopyOut(rgidAddr, rgid); err != nil { + if _, err := rgid.CopyOut(t, rgidAddr); err != nil { return 0, nil, err } - if _, err := t.CopyOut(egidAddr, egid); err != nil { + if _, err := egid.CopyOut(t, egidAddr); err != nil { return 0, nil, err } - if _, err := t.CopyOut(sgidAddr, sgid); err != nil { + if _, err := sgid.CopyOut(t, sgidAddr); err != nil { return 0, nil, err } return 0, nil, nil @@ -157,7 +157,7 @@ func Getgroups(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys for i, kgid := range kgids { gids[i] = kgid.In(t.UserNamespace()).OrOverflow() } - if _, err := t.CopyOut(args[1].Pointer(), gids); err != nil { + if _, err := auth.CopyGIDSliceOut(t, args[1].Pointer(), gids); err != nil { return 0, nil, err } return uintptr(len(gids)), nil, nil @@ -173,7 +173,7 @@ func Setgroups(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys return 0, nil, t.SetExtraGIDs(nil) } gids := make([]auth.GID, size) - if _, err := t.CopyIn(args[1].Pointer(), &gids); err != nil { + if _, err := auth.CopyGIDSliceIn(t, args[1].Pointer(), gids); err != nil { return 0, nil, err } return 0, nil, t.SetExtraGIDs(gids) diff --git a/pkg/sentry/syscalls/linux/sys_inotify.go b/pkg/sentry/syscalls/linux/sys_inotify.go index b2c7b3444..cf47bb9dd 100644 --- a/pkg/sentry/syscalls/linux/sys_inotify.go +++ b/pkg/sentry/syscalls/linux/sys_inotify.go @@ -40,7 +40,7 @@ func InotifyInit1(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. NonBlocking: flags&linux.IN_NONBLOCK != 0, } n := fs.NewFile(t, dirent, fileFlags, fs.NewInotify(t)) - defer n.DecRef() + defer n.DecRef(t) fd, err := t.NewFDFrom(0, n, kernel.FDFlags{ CloseOnExec: flags&linux.IN_CLOEXEC != 0, @@ -71,7 +71,7 @@ func fdToInotify(t *kernel.Task, fd int32) (*fs.Inotify, *fs.File, error) { ino, ok := file.FileOperations.(*fs.Inotify) if !ok { // Not an inotify fd. - file.DecRef() + file.DecRef(t) return nil, nil, syserror.EINVAL } @@ -98,7 +98,7 @@ func InotifyAddWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kern if err != nil { return 0, nil, err } - defer file.DecRef() + defer file.DecRef(t) path, _, err := copyInPath(t, addr, false /* allowEmpty */) if err != nil { @@ -128,6 +128,6 @@ func InotifyRmWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne if err != nil { return 0, nil, err } - defer file.DecRef() - return 0, nil, ino.RmWatch(wd) + defer file.DecRef(t) + return 0, nil, ino.RmWatch(t, wd) } diff --git a/pkg/sentry/syscalls/linux/sys_lseek.go b/pkg/sentry/syscalls/linux/sys_lseek.go index 3f7691eae..0046347cb 100644 --- a/pkg/sentry/syscalls/linux/sys_lseek.go +++ b/pkg/sentry/syscalls/linux/sys_lseek.go @@ -33,7 +33,7 @@ func Lseek(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) var sw fs.SeekWhence switch whence { @@ -48,7 +48,7 @@ func Lseek(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } offset, serr := file.Seek(t, sw, offset) - err := handleIOError(t, false /* partialResult */, serr, kernel.ERESTARTSYS, "lseek", file) + err := handleIOError(t, false /* partialResult */, serr, syserror.ERESTARTSYS, "lseek", file) if err != nil { return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/sys_mmap.go b/pkg/sentry/syscalls/linux/sys_mmap.go index 91694d374..cd8dfdfa4 100644 --- a/pkg/sentry/syscalls/linux/sys_mmap.go +++ b/pkg/sentry/syscalls/linux/sys_mmap.go @@ -75,7 +75,7 @@ func Mmap(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC } defer func() { if opts.MappingIdentity != nil { - opts.MappingIdentity.DecRef() + opts.MappingIdentity.DecRef(t) } }() @@ -85,7 +85,7 @@ func Mmap(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) flags := file.Flags() // mmap unconditionally requires that the FD is readable. @@ -100,6 +100,15 @@ func Mmap(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if err := file.ConfigureMMap(t, &opts); err != nil { return 0, nil, err } + } else if shared { + // Back shared anonymous mappings with a special mappable. + opts.Offset = 0 + m, err := mm.NewSharedAnonMappable(opts.Length, t.Kernel()) + if err != nil { + return 0, nil, err + } + opts.MappingIdentity = m // transfers ownership of m to opts + opts.Mappable = m } rv, err := t.MemoryManager().MMap(t, opts) @@ -239,7 +248,7 @@ func Mincore(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca return 0, nil, syserror.ENOMEM } resident := bytes.Repeat([]byte{1}, int(mapped/usermem.PageSize)) - _, err := t.CopyOut(vec, resident) + _, err := t.CopyOutBytes(vec, resident) return 0, nil, err } @@ -267,7 +276,7 @@ func Msync(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall }) // MSync calls fsync, the same interrupt conversion rules apply, see // mm/msync.c, fsync POSIX.1-2008. - return 0, nil, syserror.ConvertIntr(err, kernel.ERESTARTSYS) + return 0, nil, syserror.ConvertIntr(err, syserror.ERESTARTSYS) } // Mlock implements linux syscall mlock(2). diff --git a/pkg/sentry/syscalls/linux/sys_mount.go b/pkg/sentry/syscalls/linux/sys_mount.go index eb5ff48f5..bd0633564 100644 --- a/pkg/sentry/syscalls/linux/sys_mount.go +++ b/pkg/sentry/syscalls/linux/sys_mount.go @@ -115,7 +115,7 @@ func Mount(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall }); err != nil { // Something went wrong. Drop our ref on rootInode before // returning the error. - rootInode.DecRef() + rootInode.DecRef(t) return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/sys_pipe.go b/pkg/sentry/syscalls/linux/sys_pipe.go index 43c510930..849a47476 100644 --- a/pkg/sentry/syscalls/linux/sys_pipe.go +++ b/pkg/sentry/syscalls/linux/sys_pipe.go @@ -16,6 +16,7 @@ package linux import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -34,10 +35,10 @@ func pipe2(t *kernel.Task, addr usermem.Addr, flags uint) (uintptr, error) { r, w := pipe.NewConnectedPipe(t, pipe.DefaultPipeSize, usermem.PageSize) r.SetFlags(linuxToFlags(flags).Settable()) - defer r.DecRef() + defer r.DecRef(t) w.SetFlags(linuxToFlags(flags).Settable()) - defer w.DecRef() + defer w.DecRef(t) fds, err := t.NewFDs(0, []*fs.File{r, w}, kernel.FDFlags{ CloseOnExec: flags&linux.O_CLOEXEC != 0, @@ -46,10 +47,10 @@ func pipe2(t *kernel.Task, addr usermem.Addr, flags uint) (uintptr, error) { return 0, err } - if _, err := t.CopyOut(addr, fds); err != nil { + if _, err := primitive.CopyInt32SliceOut(t, addr, fds); err != nil { for _, fd := range fds { - if file, _ := t.FDTable().Remove(fd); file != nil { - file.DecRef() + if file, _ := t.FDTable().Remove(t, fd); file != nil { + file.DecRef(t) } } return 0, err diff --git a/pkg/sentry/syscalls/linux/sys_poll.go b/pkg/sentry/syscalls/linux/sys_poll.go index f0198141c..254f4c9f9 100644 --- a/pkg/sentry/syscalls/linux/sys_poll.go +++ b/pkg/sentry/syscalls/linux/sys_poll.go @@ -70,7 +70,7 @@ func initReadiness(t *kernel.Task, pfd *linux.PollFD, state *pollState, ch chan } if ch == nil { - defer file.DecRef() + defer file.DecRef(t) } else { state.file = file state.waiter, _ = waiter.NewChannelEntry(ch) @@ -82,11 +82,11 @@ func initReadiness(t *kernel.Task, pfd *linux.PollFD, state *pollState, ch chan } // releaseState releases all the pollState in "state". -func releaseState(state []pollState) { +func releaseState(t *kernel.Task, state []pollState) { for i := range state { if state[i].file != nil { state[i].file.EventUnregister(&state[i].waiter) - state[i].file.DecRef() + state[i].file.DecRef(t) } } } @@ -107,7 +107,7 @@ func pollBlock(t *kernel.Task, pfd []linux.PollFD, timeout time.Duration) (time. // result, we stop registering for events but still go through all files // to get their ready masks. state := make([]pollState, len(pfd)) - defer releaseState(state) + defer releaseState(t, state) n := uintptr(0) for i := range pfd { initReadiness(t, &pfd[i], &state[i], ch) @@ -162,7 +162,7 @@ func CopyInPollFDs(t *kernel.Task, addr usermem.Addr, nfds uint) ([]linux.PollFD pfd := make([]linux.PollFD, nfds) if nfds > 0 { - if _, err := t.CopyIn(addr, &pfd); err != nil { + if _, err := linux.CopyPollFDSliceIn(t, addr, pfd); err != nil { return nil, err } } @@ -189,7 +189,7 @@ func doPoll(t *kernel.Task, addr usermem.Addr, nfds uint, timeout time.Duration) // The poll entries are copied out regardless of whether // any are set or not. This aligns with the Linux behavior. if nfds > 0 && err == nil { - if _, err := t.CopyOut(addr, pfd); err != nil { + if _, err := linux.CopyPollFDSliceOut(t, addr, pfd); err != nil { return remainingTimeout, 0, err } } @@ -202,7 +202,7 @@ func CopyInFDSet(t *kernel.Task, addr usermem.Addr, nBytes, nBitsInLastPartialBy set := make([]byte, nBytes) if addr != 0 { - if _, err := t.CopyIn(addr, &set); err != nil { + if _, err := t.CopyInBytes(addr, set); err != nil { return nil, err } // If we only use part of the last byte, mask out the extraneous bits. @@ -266,7 +266,7 @@ func doSelect(t *kernel.Task, nfds int, readFDs, writeFDs, exceptFDs usermem.Add if file == nil { return 0, syserror.EBADF } - file.DecRef() + file.DecRef(t) var mask int16 if (rV & m) != 0 { @@ -329,19 +329,19 @@ func doSelect(t *kernel.Task, nfds int, readFDs, writeFDs, exceptFDs usermem.Add // Copy updated vectors back. if readFDs != 0 { - if _, err := t.CopyOut(readFDs, r); err != nil { + if _, err := t.CopyOutBytes(readFDs, r); err != nil { return 0, err } } if writeFDs != 0 { - if _, err := t.CopyOut(writeFDs, w); err != nil { + if _, err := t.CopyOutBytes(writeFDs, w); err != nil { return 0, err } } if exceptFDs != 0 { - if _, err := t.CopyOut(exceptFDs, e); err != nil { + if _, err := t.CopyOutBytes(exceptFDs, e); err != nil { return 0, err } } @@ -410,7 +410,7 @@ func poll(t *kernel.Task, pfdAddr usermem.Addr, nfds uint, timeout time.Duration nfds: nfds, timeout: remainingTimeout, }) - return 0, kernel.ERESTART_RESTARTBLOCK + return 0, syserror.ERESTART_RESTARTBLOCK } return n, err } @@ -464,7 +464,7 @@ func Ppoll(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // Note that this means that if err is nil but copyErr is not, copyErr is // ignored. This is consistent with Linux. if err == syserror.EINTR && copyErr == nil { - err = kernel.ERESTARTNOHAND + err = syserror.ERESTARTNOHAND } return n, nil, err } @@ -494,7 +494,7 @@ func Select(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal copyErr := copyOutTimevalRemaining(t, startNs, timeout, timevalAddr) // See comment in Ppoll. if err == syserror.EINTR && copyErr == nil { - err = kernel.ERESTARTNOHAND + err = syserror.ERESTARTNOHAND } return n, nil, err } @@ -539,7 +539,7 @@ func Pselect(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca copyErr := copyOutTimespecRemaining(t, startNs, timeout, timespecAddr) // See comment in Ppoll. if err == syserror.EINTR && copyErr == nil { - err = kernel.ERESTARTNOHAND + err = syserror.ERESTARTNOHAND } return n, nil, err } diff --git a/pkg/sentry/syscalls/linux/sys_prctl.go b/pkg/sentry/syscalls/linux/sys_prctl.go index f92bf8096..a892d2c62 100644 --- a/pkg/sentry/syscalls/linux/sys_prctl.go +++ b/pkg/sentry/syscalls/linux/sys_prctl.go @@ -18,6 +18,7 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fsbridge" @@ -43,7 +44,7 @@ func Prctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return 0, nil, nil case linux.PR_GET_PDEATHSIG: - _, err := t.CopyOut(args[1].Pointer(), int32(t.ParentDeathSignal())) + _, err := primitive.CopyInt32Out(t, args[1].Pointer(), int32(t.ParentDeathSignal())) return 0, nil, err case linux.PR_GET_DUMPABLE: @@ -110,7 +111,7 @@ func Prctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall buf[len] = 0 len++ } - _, err := t.CopyOut(addr, buf[:len]) + _, err := t.CopyOutBytes(addr, buf[:len]) if err != nil { return 0, nil, err } @@ -128,7 +129,7 @@ func Prctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // They trying to set exe to a non-file? if !fs.IsFile(file.Dirent.Inode.StableAttr) { @@ -136,7 +137,7 @@ func Prctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } // Set the underlying executable. - t.MemoryManager().SetExecutable(fsbridge.NewFSFile(file)) + t.MemoryManager().SetExecutable(t, fsbridge.NewFSFile(file)) case linux.PR_SET_MM_AUXV, linux.PR_SET_MM_START_CODE, diff --git a/pkg/sentry/syscalls/linux/sys_read.go b/pkg/sentry/syscalls/linux/sys_read.go index 071b4bacc..f655d3db1 100644 --- a/pkg/sentry/syscalls/linux/sys_read.go +++ b/pkg/sentry/syscalls/linux/sys_read.go @@ -48,7 +48,7 @@ func Read(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the file is readable. if !file.Flags().Read { @@ -71,7 +71,7 @@ func Read(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC n, err := readv(t, file, dst) t.IOUsage().AccountReadSyscall(n) - return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "read", file) + return uintptr(n), nil, handleIOError(t, n != 0, err, syserror.ERESTARTSYS, "read", file) } // Readahead implements readahead(2). @@ -84,7 +84,7 @@ func Readahead(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the file is readable. if !file.Flags().Read { @@ -118,7 +118,7 @@ func Pread64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate and does not overflow. if offset < 0 || offset+int64(size) < 0 { @@ -151,7 +151,7 @@ func Pread64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca n, err := preadv(t, file, dst, offset) t.IOUsage().AccountReadSyscall(n) - return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "pread64", file) + return uintptr(n), nil, handleIOError(t, n != 0, err, syserror.ERESTARTSYS, "pread64", file) } // Readv implements linux syscall readv(2). @@ -164,7 +164,7 @@ func Readv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the file is readable. if !file.Flags().Read { @@ -181,7 +181,7 @@ func Readv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall n, err := readv(t, file, dst) t.IOUsage().AccountReadSyscall(n) - return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "readv", file) + return uintptr(n), nil, handleIOError(t, n != 0, err, syserror.ERESTARTSYS, "readv", file) } // Preadv implements linux syscall preadv(2). @@ -195,7 +195,7 @@ func Preadv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate. if offset < 0 { @@ -222,7 +222,7 @@ func Preadv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal n, err := preadv(t, file, dst, offset) t.IOUsage().AccountReadSyscall(n) - return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "preadv", file) + return uintptr(n), nil, handleIOError(t, n != 0, err, syserror.ERESTARTSYS, "preadv", file) } // Preadv2 implements linux syscall preadv2(2). @@ -244,7 +244,7 @@ func Preadv2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate. if offset < -1 { @@ -280,12 +280,12 @@ func Preadv2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if offset == -1 { n, err := readv(t, file, dst) t.IOUsage().AccountReadSyscall(n) - return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "preadv2", file) + return uintptr(n), nil, handleIOError(t, n != 0, err, syserror.ERESTARTSYS, "preadv2", file) } n, err := preadv(t, file, dst, offset) t.IOUsage().AccountReadSyscall(n) - return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "preadv2", file) + return uintptr(n), nil, handleIOError(t, n != 0, err, syserror.ERESTARTSYS, "preadv2", file) } func readv(t *kernel.Task, f *fs.File, dst usermem.IOSequence) (int64, error) { diff --git a/pkg/sentry/syscalls/linux/sys_rlimit.go b/pkg/sentry/syscalls/linux/sys_rlimit.go index d5d5b6959..309c183a3 100644 --- a/pkg/sentry/syscalls/linux/sys_rlimit.go +++ b/pkg/sentry/syscalls/linux/sys_rlimit.go @@ -16,6 +16,7 @@ package linux import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/limits" @@ -26,17 +27,13 @@ import ( // rlimit describes an implementation of 'struct rlimit', which may vary from // system-to-system. type rlimit interface { + marshal.Marshallable + // toLimit converts an rlimit to a limits.Limit. toLimit() *limits.Limit // fromLimit converts a limits.Limit to an rlimit. fromLimit(lim limits.Limit) - - // copyIn copies an rlimit from the untrusted app to the kernel. - copyIn(t *kernel.Task, addr usermem.Addr) error - - // copyOut copies an rlimit from the kernel to the untrusted app. - copyOut(t *kernel.Task, addr usermem.Addr) error } // newRlimit returns the appropriate rlimit type for 'struct rlimit' on this system. @@ -50,6 +47,7 @@ func newRlimit(t *kernel.Task) (rlimit, error) { } } +// +marshal type rlimit64 struct { Cur uint64 Max uint64 @@ -70,12 +68,12 @@ func (r *rlimit64) fromLimit(lim limits.Limit) { } func (r *rlimit64) copyIn(t *kernel.Task, addr usermem.Addr) error { - _, err := t.CopyIn(addr, r) + _, err := r.CopyIn(t, addr) return err } func (r *rlimit64) copyOut(t *kernel.Task, addr usermem.Addr) error { - _, err := t.CopyOut(addr, *r) + _, err := r.CopyOut(t, addr) return err } @@ -140,7 +138,8 @@ func Getrlimit(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys return 0, nil, err } rlim.fromLimit(lim) - return 0, nil, rlim.copyOut(t, addr) + _, err = rlim.CopyOut(t, addr) + return 0, nil, err } // Setrlimit implements linux syscall setrlimit(2). @@ -155,7 +154,7 @@ func Setrlimit(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if err != nil { return 0, nil, err } - if err := rlim.copyIn(t, addr); err != nil { + if _, err := rlim.CopyIn(t, addr); err != nil { return 0, nil, syserror.EFAULT } _, err = prlimit64(t, resource, rlim.toLimit()) diff --git a/pkg/sentry/syscalls/linux/sys_rusage.go b/pkg/sentry/syscalls/linux/sys_rusage.go index 1674c7445..ac5c98a54 100644 --- a/pkg/sentry/syscalls/linux/sys_rusage.go +++ b/pkg/sentry/syscalls/linux/sys_rusage.go @@ -80,7 +80,7 @@ func Getrusage(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys } ru := getrusage(t, which) - _, err := t.CopyOut(addr, &ru) + _, err := ru.CopyOut(t, addr) return 0, nil, err } @@ -104,7 +104,7 @@ func Times(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall CUTime: linux.ClockTFromDuration(cs2.UserTime), CSTime: linux.ClockTFromDuration(cs2.SysTime), } - if _, err := t.CopyOut(addr, &r); err != nil { + if _, err := r.CopyOut(t, addr); err != nil { return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/sys_sched.go b/pkg/sentry/syscalls/linux/sys_sched.go index 99f6993f5..bfcf44b6f 100644 --- a/pkg/sentry/syscalls/linux/sys_sched.go +++ b/pkg/sentry/syscalls/linux/sys_sched.go @@ -27,8 +27,10 @@ const ( ) // SchedParam replicates struct sched_param in sched.h. +// +// +marshal type SchedParam struct { - schedPriority int64 + schedPriority int32 } // SchedGetparam implements linux syscall sched_getparam(2). @@ -45,7 +47,7 @@ func SchedGetparam(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel return 0, nil, syserror.ESRCH } r := SchedParam{schedPriority: onlyPriority} - if _, err := t.CopyOut(param, r); err != nil { + if _, err := r.CopyOut(t, param); err != nil { return 0, nil, err } @@ -79,7 +81,7 @@ func SchedSetscheduler(t *kernel.Task, args arch.SyscallArguments) (uintptr, *ke return 0, nil, syserror.ESRCH } var r SchedParam - if _, err := t.CopyIn(param, &r); err != nil { + if _, err := r.CopyIn(t, param); err != nil { return 0, nil, syserror.EINVAL } if r.schedPriority != onlyPriority { diff --git a/pkg/sentry/syscalls/linux/sys_seccomp.go b/pkg/sentry/syscalls/linux/sys_seccomp.go index 5b7a66f4d..4fdb4463c 100644 --- a/pkg/sentry/syscalls/linux/sys_seccomp.go +++ b/pkg/sentry/syscalls/linux/sys_seccomp.go @@ -24,6 +24,8 @@ import ( ) // userSockFprog is equivalent to Linux's struct sock_fprog on amd64. +// +// +marshal type userSockFprog struct { // Len is the length of the filter in BPF instructions. Len uint16 @@ -33,7 +35,7 @@ type userSockFprog struct { // Filter is a user pointer to the struct sock_filter array that makes up // the filter program. Filter is a uint64 rather than a usermem.Addr // because usermem.Addr is actually uintptr, which is not a fixed-size - // type, and encoding/binary.Read objects to this. + // type. Filter uint64 } @@ -54,11 +56,11 @@ func seccomp(t *kernel.Task, mode, flags uint64, addr usermem.Addr) error { } var fprog userSockFprog - if _, err := t.CopyIn(addr, &fprog); err != nil { + if _, err := fprog.CopyIn(t, addr); err != nil { return err } filter := make([]linux.BPFInstruction, int(fprog.Len)) - if _, err := t.CopyIn(usermem.Addr(fprog.Filter), &filter); err != nil { + if _, err := linux.CopyBPFInstructionSliceIn(t, usermem.Addr(fprog.Filter), filter); err != nil { return err } compiledFilter, err := bpf.Compile(filter) diff --git a/pkg/sentry/syscalls/linux/sys_sem.go b/pkg/sentry/syscalls/linux/sys_sem.go index 5f54f2456..47dadb800 100644 --- a/pkg/sentry/syscalls/linux/sys_sem.go +++ b/pkg/sentry/syscalls/linux/sys_sem.go @@ -18,6 +18,7 @@ import ( "math" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -66,7 +67,7 @@ func Semop(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } ops := make([]linux.Sembuf, nsops) - if _, err := t.CopyIn(sembufAddr, ops); err != nil { + if _, err := linux.CopySembufSliceIn(t, sembufAddr, ops); err != nil { return 0, nil, err } @@ -116,8 +117,8 @@ func Semctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal case linux.IPC_SET: arg := args[3].Pointer() - s := linux.SemidDS{} - if _, err := t.CopyIn(arg, &s); err != nil { + var s linux.SemidDS + if _, err := s.CopyIn(t, arg); err != nil { return 0, nil, err } @@ -188,7 +189,7 @@ func setValAll(t *kernel.Task, id int32, array usermem.Addr) error { return syserror.EINVAL } vals := make([]uint16, set.Size()) - if _, err := t.CopyIn(array, vals); err != nil { + if _, err := primitive.CopyUint16SliceIn(t, array, vals); err != nil { return err } creds := auth.CredentialsFromContext(t) @@ -217,7 +218,7 @@ func getValAll(t *kernel.Task, id int32, array usermem.Addr) error { if err != nil { return err } - _, err = t.CopyOut(array, vals) + _, err = primitive.CopyUint16SliceOut(t, array, vals) return err } diff --git a/pkg/sentry/syscalls/linux/sys_shm.go b/pkg/sentry/syscalls/linux/sys_shm.go index 4a8bc24a2..584064143 100644 --- a/pkg/sentry/syscalls/linux/sys_shm.go +++ b/pkg/sentry/syscalls/linux/sys_shm.go @@ -39,7 +39,7 @@ func Shmget(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, err } - defer segment.DecRef() + defer segment.DecRef(t) return uintptr(segment.ID), nil, nil } @@ -66,7 +66,7 @@ func Shmat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if err != nil { return 0, nil, syserror.EINVAL } - defer segment.DecRef() + defer segment.DecRef(t) opts, err := segment.ConfigureAttach(t, addr, shm.AttachOpts{ Execute: flag&linux.SHM_EXEC == linux.SHM_EXEC, @@ -108,22 +108,22 @@ func Shmctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, syserror.EINVAL } - defer segment.DecRef() + defer segment.DecRef(t) stat, err := segment.IPCStat(t) if err == nil { - _, err = t.CopyOut(buf, stat) + _, err = stat.CopyOut(t, buf) } return 0, nil, err case linux.IPC_INFO: params := r.IPCInfo() - _, err := t.CopyOut(buf, params) + _, err := params.CopyOut(t, buf) return 0, nil, err case linux.SHM_INFO: info := r.ShmInfo() - _, err := t.CopyOut(buf, info) + _, err := info.CopyOut(t, buf) return 0, nil, err } @@ -132,20 +132,19 @@ func Shmctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, syserror.EINVAL } - defer segment.DecRef() + defer segment.DecRef(t) switch cmd { case linux.IPC_SET: var ds linux.ShmidDS - _, err = t.CopyIn(buf, &ds) - if err != nil { + if _, err = ds.CopyIn(t, buf); err != nil { return 0, nil, err } - err = segment.Set(t, &ds) + err := segment.Set(t, &ds) return 0, nil, err case linux.IPC_RMID: - segment.MarkDestroyed() + segment.MarkDestroyed(t) return 0, nil, nil case linux.SHM_LOCK, linux.SHM_UNLOCK: diff --git a/pkg/sentry/syscalls/linux/sys_signal.go b/pkg/sentry/syscalls/linux/sys_signal.go index d2b0012ae..e748d33d8 100644 --- a/pkg/sentry/syscalls/linux/sys_signal.go +++ b/pkg/sentry/syscalls/linux/sys_signal.go @@ -348,7 +348,7 @@ func Sigaltstack(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S // Pause implements linux syscall pause(2). func Pause(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - return 0, nil, syserror.ConvertIntr(t.Block(nil), kernel.ERESTARTNOHAND) + return 0, nil, syserror.ConvertIntr(t.Block(nil), syserror.ERESTARTNOHAND) } // RtSigpending implements linux syscall rt_sigpending(2). @@ -496,7 +496,7 @@ func RtSigsuspend(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. t.SetSavedSignalMask(oldmask) // Perform the wait. - return 0, nil, syserror.ConvertIntr(t.Block(nil), kernel.ERESTARTNOHAND) + return 0, nil, syserror.ConvertIntr(t.Block(nil), syserror.ERESTARTNOHAND) } // RestartSyscall implements the linux syscall restart_syscall(2). @@ -536,7 +536,7 @@ func sharedSignalfd(t *kernel.Task, fd int32, sigset usermem.Addr, sigsetsize ui if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Is this a signalfd? if s, ok := file.FileOperations.(*signalfd.SignalOperations); ok { @@ -553,7 +553,7 @@ func sharedSignalfd(t *kernel.Task, fd int32, sigset usermem.Addr, sigsetsize ui if err != nil { return 0, nil, err } - defer file.DecRef() + defer file.DecRef(t) // Set appropriate flags. file.SetFlags(fs.SettableFileFlags{ diff --git a/pkg/sentry/syscalls/linux/sys_socket.go b/pkg/sentry/syscalls/linux/sys_socket.go index 414fce8e3..9feaca0da 100644 --- a/pkg/sentry/syscalls/linux/sys_socket.go +++ b/pkg/sentry/syscalls/linux/sys_socket.go @@ -19,6 +19,8 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/marshal" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -29,8 +31,6 @@ import ( "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" - "gvisor.dev/gvisor/tools/go_marshal/marshal" - "gvisor.dev/gvisor/tools/go_marshal/primitive" ) // LINT.IfChange @@ -67,10 +67,10 @@ const flagsOffset = 48 const sizeOfInt32 = 4 // messageHeader64Len is the length of a MessageHeader64 struct. -var messageHeader64Len = uint64(binary.Size(MessageHeader64{})) +var messageHeader64Len = uint64((*MessageHeader64)(nil).SizeBytes()) // multipleMessageHeader64Len is the length of a multipeMessageHeader64 struct. -var multipleMessageHeader64Len = uint64(binary.Size(multipleMessageHeader64{})) +var multipleMessageHeader64Len = uint64((*multipleMessageHeader64)(nil).SizeBytes()) // baseRecvFlags are the flags that are accepted across recvmsg(2), // recvmmsg(2), and recvfrom(2). @@ -78,6 +78,8 @@ const baseRecvFlags = linux.MSG_OOB | linux.MSG_DONTROUTE | linux.MSG_DONTWAIT | // MessageHeader64 is the 64-bit representation of the msghdr struct used in // the recvmsg and sendmsg syscalls. +// +// +marshal type MessageHeader64 struct { // Name is the optional pointer to a network address buffer. Name uint64 @@ -106,30 +108,14 @@ type MessageHeader64 struct { // multipleMessageHeader64 is the 64-bit representation of the mmsghdr struct used in // the recvmmsg and sendmmsg syscalls. +// +// +marshal type multipleMessageHeader64 struct { msgHdr MessageHeader64 msgLen uint32 _ int32 } -// CopyInMessageHeader64 copies a message header from user to kernel memory. -func CopyInMessageHeader64(t *kernel.Task, addr usermem.Addr, msg *MessageHeader64) error { - b := t.CopyScratchBuffer(52) - if _, err := t.CopyInBytes(addr, b); err != nil { - return err - } - - msg.Name = usermem.ByteOrder.Uint64(b[0:]) - msg.NameLen = usermem.ByteOrder.Uint32(b[8:]) - msg.Iov = usermem.ByteOrder.Uint64(b[16:]) - msg.IovLen = usermem.ByteOrder.Uint64(b[24:]) - msg.Control = usermem.ByteOrder.Uint64(b[32:]) - msg.ControlLen = usermem.ByteOrder.Uint64(b[40:]) - msg.Flags = int32(usermem.ByteOrder.Uint32(b[48:])) - - return nil -} - // CaptureAddress allocates memory for and copies a socket address structure // from the untrusted address space range. func CaptureAddress(t *kernel.Task, addr usermem.Addr, addrlen uint32) ([]byte, error) { @@ -148,10 +134,10 @@ func CaptureAddress(t *kernel.Task, addr usermem.Addr, addrlen uint32) ([]byte, // writeAddress writes a sockaddr structure and its length to an output buffer // in the unstrusted address space range. If the address is bigger than the // buffer, it is truncated. -func writeAddress(t *kernel.Task, addr interface{}, addrLen uint32, addrPtr usermem.Addr, addrLenPtr usermem.Addr) error { +func writeAddress(t *kernel.Task, addr linux.SockAddr, addrLen uint32, addrPtr usermem.Addr, addrLenPtr usermem.Addr) error { // Get the buffer length. var bufLen uint32 - if _, err := t.CopyIn(addrLenPtr, &bufLen); err != nil { + if _, err := primitive.CopyUint32In(t, addrLenPtr, &bufLen); err != nil { return err } @@ -160,7 +146,7 @@ func writeAddress(t *kernel.Task, addr interface{}, addrLen uint32, addrPtr user } // Write the length unconditionally. - if _, err := t.CopyOut(addrLenPtr, addrLen); err != nil { + if _, err := primitive.CopyUint32Out(t, addrLenPtr, addrLen); err != nil { return err } @@ -173,7 +159,8 @@ func writeAddress(t *kernel.Task, addr interface{}, addrLen uint32, addrPtr user } // Copy as much of the address as will fit in the buffer. - encodedAddr := binary.Marshal(nil, usermem.ByteOrder, addr) + encodedAddr := t.CopyScratchBuffer(addr.SizeBytes()) + addr.MarshalUnsafe(encodedAddr) if bufLen > uint32(len(encodedAddr)) { bufLen = uint32(len(encodedAddr)) } @@ -200,7 +187,7 @@ func Socket(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal s.SetFlags(fs.SettableFileFlags{ NonBlocking: stype&linux.SOCK_NONBLOCK != 0, }) - defer s.DecRef() + defer s.DecRef(t) fd, err := t.NewFDFrom(0, s, kernel.FDFlags{ CloseOnExec: stype&linux.SOCK_CLOEXEC != 0, @@ -235,8 +222,8 @@ func SocketPair(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy } s1.SetFlags(fileFlags) s2.SetFlags(fileFlags) - defer s1.DecRef() - defer s2.DecRef() + defer s1.DecRef(t) + defer s2.DecRef(t) // Create the FDs for the sockets. fds, err := t.NewFDs(0, []*fs.File{s1, s2}, kernel.FDFlags{ @@ -247,10 +234,10 @@ func SocketPair(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy } // Copy the file descriptors out. - if _, err := t.CopyOut(socks, fds); err != nil { + if _, err := primitive.CopyInt32SliceOut(t, socks, fds); err != nil { for _, fd := range fds { - if file, _ := t.FDTable().Remove(fd); file != nil { - file.DecRef() + if file, _ := t.FDTable().Remove(t, fd); file != nil { + file.DecRef(t) } } return 0, nil, err @@ -270,7 +257,7 @@ func Connect(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -285,7 +272,7 @@ func Connect(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca } blocking := !file.Flags().NonBlocking - return 0, nil, syserror.ConvertIntr(s.Connect(t, a, blocking).ToError(), kernel.ERESTARTSYS) + return 0, nil, syserror.ConvertIntr(s.Connect(t, a, blocking).ToError(), syserror.ERESTARTSYS) } // accept is the implementation of the accept syscall. It is called by accept @@ -301,7 +288,7 @@ func accept(t *kernel.Task, fd int32, addr usermem.Addr, addrLen usermem.Addr, f if file == nil { return 0, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -316,7 +303,7 @@ func accept(t *kernel.Task, fd int32, addr usermem.Addr, addrLen usermem.Addr, f peerRequested := addrLen != 0 nfd, peer, peerLen, e := s.Accept(t, peerRequested, flags, blocking) if e != nil { - return 0, syserror.ConvertIntr(e.ToError(), kernel.ERESTARTSYS) + return 0, syserror.ConvertIntr(e.ToError(), syserror.ERESTARTSYS) } if peerRequested { // NOTE(magi): Linux does not give you an error if it can't @@ -360,7 +347,7 @@ func Bind(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -387,7 +374,7 @@ func Listen(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -416,7 +403,7 @@ func Shutdown(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -447,7 +434,7 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -456,8 +443,8 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy } // Read the length. Reject negative values. - optLen := int32(0) - if _, err := t.CopyIn(optLenAddr, &optLen); err != nil { + var optLen int32 + if _, err := primitive.CopyInt32In(t, optLenAddr, &optLen); err != nil { return 0, nil, err } if optLen < 0 { @@ -471,7 +458,7 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy } vLen := int32(binary.Size(v)) - if _, err := t.CopyOut(optLenAddr, vLen); err != nil { + if _, err := primitive.CopyInt32Out(t, optLenAddr, vLen); err != nil { return 0, nil, err } @@ -529,7 +516,7 @@ func SetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -567,7 +554,7 @@ func GetSockName(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -595,7 +582,7 @@ func GetPeerName(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -628,7 +615,7 @@ func RecvMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -681,7 +668,7 @@ func RecvMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -733,7 +720,7 @@ func RecvMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if !ok { return 0, nil, syserror.EFAULT } - if _, err = t.CopyOut(lp, uint32(n)); err != nil { + if _, err = primitive.CopyUint32Out(t, lp, uint32(n)); err != nil { break } count++ @@ -748,7 +735,7 @@ func RecvMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc func recvSingleMsg(t *kernel.Task, s socket.Socket, msgPtr usermem.Addr, flags int32, haveDeadline bool, deadline ktime.Time) (uintptr, error) { // Capture the message header and io vectors. var msg MessageHeader64 - if err := CopyInMessageHeader64(t, msgPtr, &msg); err != nil { + if _, err := msg.CopyIn(t, msgPtr); err != nil { return 0, err } @@ -771,16 +758,16 @@ func recvSingleMsg(t *kernel.Task, s socket.Socket, msgPtr usermem.Addr, flags i if msg.ControlLen == 0 && msg.NameLen == 0 { n, mflags, _, _, cms, err := s.RecvMsg(t, dst, int(flags), haveDeadline, deadline, false, 0) if err != nil { - return 0, syserror.ConvertIntr(err.ToError(), kernel.ERESTARTSYS) + return 0, syserror.ConvertIntr(err.ToError(), syserror.ERESTARTSYS) } if !cms.Unix.Empty() { mflags |= linux.MSG_CTRUNC - cms.Release() + cms.Release(t) } if int(msg.Flags) != mflags { // Copy out the flags to the caller. - if _, err := t.CopyOut(msgPtr+flagsOffset, int32(mflags)); err != nil { + if _, err := primitive.CopyInt32Out(t, msgPtr+flagsOffset, int32(mflags)); err != nil { return 0, err } } @@ -793,9 +780,9 @@ func recvSingleMsg(t *kernel.Task, s socket.Socket, msgPtr usermem.Addr, flags i } n, mflags, sender, senderLen, cms, e := s.RecvMsg(t, dst, int(flags), haveDeadline, deadline, msg.NameLen != 0, msg.ControlLen) if e != nil { - return 0, syserror.ConvertIntr(e.ToError(), kernel.ERESTARTSYS) + return 0, syserror.ConvertIntr(e.ToError(), syserror.ERESTARTSYS) } - defer cms.Release() + defer cms.Release(t) controlData := make([]byte, 0, msg.ControlLen) controlData = control.PackControlMessages(t, cms, controlData) @@ -817,17 +804,17 @@ func recvSingleMsg(t *kernel.Task, s socket.Socket, msgPtr usermem.Addr, flags i } // Copy the control data to the caller. - if _, err := t.CopyOut(msgPtr+controlLenOffset, uint64(len(controlData))); err != nil { + if _, err := primitive.CopyUint64Out(t, msgPtr+controlLenOffset, uint64(len(controlData))); err != nil { return 0, err } if len(controlData) > 0 { - if _, err := t.CopyOut(usermem.Addr(msg.Control), controlData); err != nil { + if _, err := t.CopyOutBytes(usermem.Addr(msg.Control), controlData); err != nil { return 0, err } } // Copy out the flags to the caller. - if _, err := t.CopyOut(msgPtr+flagsOffset, int32(mflags)); err != nil { + if _, err := primitive.CopyInt32Out(t, msgPtr+flagsOffset, int32(mflags)); err != nil { return 0, err } @@ -851,7 +838,7 @@ func recvFrom(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flag if file == nil { return 0, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -880,9 +867,9 @@ func recvFrom(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flag } n, _, sender, senderLen, cm, e := s.RecvMsg(t, dst, int(flags), haveDeadline, deadline, nameLenPtr != 0, 0) - cm.Release() + cm.Release(t) if e != nil { - return 0, syserror.ConvertIntr(e.ToError(), kernel.ERESTARTSYS) + return 0, syserror.ConvertIntr(e.ToError(), syserror.ERESTARTSYS) } // Copy the address to the caller. @@ -924,7 +911,7 @@ func SendMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -962,7 +949,7 @@ func SendMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -996,7 +983,7 @@ func SendMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if !ok { return 0, nil, syserror.EFAULT } - if _, err = t.CopyOut(lp, uint32(n)); err != nil { + if _, err = primitive.CopyUint32Out(t, lp, uint32(n)); err != nil { break } count++ @@ -1011,7 +998,7 @@ func SendMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc func sendSingleMsg(t *kernel.Task, s socket.Socket, file *fs.File, msgPtr usermem.Addr, flags int32) (uintptr, error) { // Capture the message header. var msg MessageHeader64 - if err := CopyInMessageHeader64(t, msgPtr, &msg); err != nil { + if _, err := msg.CopyIn(t, msgPtr); err != nil { return 0, err } @@ -1022,7 +1009,7 @@ func sendSingleMsg(t *kernel.Task, s socket.Socket, file *fs.File, msgPtr userme return 0, syserror.ENOBUFS } controlData = make([]byte, msg.ControlLen) - if _, err := t.CopyIn(usermem.Addr(msg.Control), &controlData); err != nil { + if _, err := t.CopyInBytes(usermem.Addr(msg.Control), controlData); err != nil { return 0, err } } @@ -1064,9 +1051,9 @@ func sendSingleMsg(t *kernel.Task, s socket.Socket, file *fs.File, msgPtr userme // Call the syscall implementation. n, e := s.SendMsg(t, src, to, int(flags), haveDeadline, deadline, controlMessages) - err = handleIOError(t, n != 0, e.ToError(), kernel.ERESTARTSYS, "sendmsg", file) + err = handleIOError(t, n != 0, e.ToError(), syserror.ERESTARTSYS, "sendmsg", file) if err != nil { - controlMessages.Release() + controlMessages.Release(t) } return uintptr(n), err } @@ -1084,7 +1071,7 @@ func sendTo(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flags if file == nil { return 0, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -1124,7 +1111,7 @@ func sendTo(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flags // Call the syscall implementation. n, e := s.SendMsg(t, src, to, int(flags), haveDeadline, deadline, socket.ControlMessages{Unix: control.New(t, s, nil)}) - return uintptr(n), handleIOError(t, n != 0, e.ToError(), kernel.ERESTARTSYS, "sendto", file) + return uintptr(n), handleIOError(t, n != 0, e.ToError(), syserror.ERESTARTSYS, "sendto", file) } // SendTo implements the linux syscall sendto(2). diff --git a/pkg/sentry/syscalls/linux/sys_splice.go b/pkg/sentry/syscalls/linux/sys_splice.go index 77c78889d..46616c961 100644 --- a/pkg/sentry/syscalls/linux/sys_splice.go +++ b/pkg/sentry/syscalls/linux/sys_splice.go @@ -16,6 +16,7 @@ package linux import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -101,7 +102,7 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if inFile == nil { return 0, nil, syserror.EBADF } - defer inFile.DecRef() + defer inFile.DecRef(t) if !inFile.Flags().Read { return 0, nil, syserror.EBADF @@ -111,7 +112,7 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if outFile == nil { return 0, nil, syserror.EBADF } - defer outFile.DecRef() + defer outFile.DecRef(t) if !outFile.Flags().Write { return 0, nil, syserror.EBADF @@ -141,7 +142,7 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc // Copy in the offset. var offset int64 - if _, err := t.CopyIn(offsetAddr, &offset); err != nil { + if _, err := primitive.CopyInt64In(t, offsetAddr, &offset); err != nil { return 0, nil, err } @@ -149,11 +150,11 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc n, err = doSplice(t, outFile, inFile, fs.SpliceOpts{ Length: count, SrcOffset: true, - SrcStart: offset, + SrcStart: int64(offset), }, outFile.Flags().NonBlocking) // Copy out the new offset. - if _, err := t.CopyOut(offsetAddr, n+offset); err != nil { + if _, err := primitive.CopyInt64Out(t, offsetAddr, offset+n); err != nil { return 0, nil, err } } else { @@ -170,7 +171,7 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc // We can only pass a single file to handleIOError, so pick inFile // arbitrarily. This is used only for debugging purposes. - return uintptr(n), nil, handleIOError(t, false, err, kernel.ERESTARTSYS, "sendfile", inFile) + return uintptr(n), nil, handleIOError(t, false, err, syserror.ERESTARTSYS, "sendfile", inFile) } // Splice implements splice(2). @@ -192,13 +193,13 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if outFile == nil { return 0, nil, syserror.EBADF } - defer outFile.DecRef() + defer outFile.DecRef(t) inFile := t.GetFile(inFD) if inFile == nil { return 0, nil, syserror.EBADF } - defer inFile.DecRef() + defer inFile.DecRef(t) // The operation is non-blocking if anything is non-blocking. // @@ -228,7 +229,7 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal } var offset int64 - if _, err := t.CopyIn(outOffset, &offset); err != nil { + if _, err := primitive.CopyInt64In(t, outOffset, &offset); err != nil { return 0, nil, err } @@ -246,7 +247,7 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal } var offset int64 - if _, err := t.CopyIn(inOffset, &offset); err != nil { + if _, err := primitive.CopyInt64In(t, inOffset, &offset); err != nil { return 0, nil, err } @@ -280,7 +281,7 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal } // See above; inFile is chosen arbitrarily here. - return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "splice", inFile) + return uintptr(n), nil, handleIOError(t, n != 0, err, syserror.ERESTARTSYS, "splice", inFile) } // Tee imlements tee(2). @@ -300,13 +301,13 @@ func Tee(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo if outFile == nil { return 0, nil, syserror.EBADF } - defer outFile.DecRef() + defer outFile.DecRef(t) inFile := t.GetFile(inFD) if inFile == nil { return 0, nil, syserror.EBADF } - defer inFile.DecRef() + defer inFile.DecRef(t) // All files must be pipes. if !fs.IsPipe(inFile.Dirent.Inode.StableAttr) || !fs.IsPipe(outFile.Dirent.Inode.StableAttr) { @@ -333,5 +334,5 @@ func Tee(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo } // See above; inFile is chosen arbitrarily here. - return uintptr(n), nil, handleIOError(t, false, err, kernel.ERESTARTSYS, "tee", inFile) + return uintptr(n), nil, handleIOError(t, false, err, syserror.ERESTARTSYS, "tee", inFile) } diff --git a/pkg/sentry/syscalls/linux/sys_stat.go b/pkg/sentry/syscalls/linux/sys_stat.go index 46ebf27a2..cda29a8b5 100644 --- a/pkg/sentry/syscalls/linux/sys_stat.go +++ b/pkg/sentry/syscalls/linux/sys_stat.go @@ -58,7 +58,7 @@ func Fstatat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return 0, nil, fstat(t, file, statAddr) } @@ -100,7 +100,7 @@ func Fstat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return 0, nil, fstat(t, file, statAddr) } @@ -158,7 +158,7 @@ func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) uattr, err := file.UnstableAttr(t) if err != nil { return 0, nil, err @@ -221,7 +221,7 @@ func statx(t *kernel.Task, sattr fs.StableAttr, uattr fs.UnstableAttr, statxAddr DevMajor: uint32(devMajor), DevMinor: devMinor, } - _, err := t.CopyOut(statxAddr, &s) + _, err := s.CopyOut(t, statxAddr) return err } @@ -249,7 +249,7 @@ func Fstatfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return 0, nil, statfsImpl(t, file.Dirent, statfsAddr) } @@ -283,7 +283,7 @@ func statfsImpl(t *kernel.Task, d *fs.Dirent, addr usermem.Addr) error { FragmentSize: d.Inode.StableAttr.BlockSize, // Leave other fields 0 like simple_statfs does. } - _, err = t.CopyOut(addr, &statfs) + _, err = statfs.CopyOut(t, addr) return err } diff --git a/pkg/sentry/syscalls/linux/sys_sync.go b/pkg/sentry/syscalls/linux/sys_sync.go index 5ad465ae3..048a21c6e 100644 --- a/pkg/sentry/syscalls/linux/sys_sync.go +++ b/pkg/sentry/syscalls/linux/sys_sync.go @@ -39,7 +39,7 @@ func Syncfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Use "sync-the-world" for now, it's guaranteed that fd is at least // on the root filesystem. @@ -54,10 +54,10 @@ func Fsync(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) err := file.Fsync(t, 0, fs.FileMaxOffset, fs.SyncAll) - return 0, nil, syserror.ConvertIntr(err, kernel.ERESTARTSYS) + return 0, nil, syserror.ConvertIntr(err, syserror.ERESTARTSYS) } // Fdatasync implements linux syscall fdatasync(2). @@ -70,10 +70,10 @@ func Fdatasync(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) err := file.Fsync(t, 0, fs.FileMaxOffset, fs.SyncData) - return 0, nil, syserror.ConvertIntr(err, kernel.ERESTARTSYS) + return 0, nil, syserror.ConvertIntr(err, syserror.ERESTARTSYS) } // SyncFileRange implements linux syscall sync_file_rage(2) @@ -103,7 +103,7 @@ func SyncFileRange(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // SYNC_FILE_RANGE_WAIT_BEFORE waits upon write-out of all pages in the // specified range that have already been submitted to the device @@ -135,7 +135,7 @@ func SyncFileRange(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel err = file.Fsync(t, offset, fs.FileMaxOffset, fs.SyncData) } - return 0, nil, syserror.ConvertIntr(err, kernel.ERESTARTSYS) + return 0, nil, syserror.ConvertIntr(err, syserror.ERESTARTSYS) } // LINT.ThenChange(vfs2/sync.go) diff --git a/pkg/sentry/syscalls/linux/sys_sysinfo.go b/pkg/sentry/syscalls/linux/sys_sysinfo.go index 297de052a..674d341b6 100644 --- a/pkg/sentry/syscalls/linux/sys_sysinfo.go +++ b/pkg/sentry/syscalls/linux/sys_sysinfo.go @@ -43,6 +43,6 @@ func Sysinfo(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca FreeRAM: memFree, Unit: 1, } - _, err := t.CopyOut(addr, si) + _, err := si.CopyOut(t, addr) return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/sys_thread.go b/pkg/sentry/syscalls/linux/sys_thread.go index 00915fdde..39ca9ea97 100644 --- a/pkg/sentry/syscalls/linux/sys_thread.go +++ b/pkg/sentry/syscalls/linux/sys_thread.go @@ -19,6 +19,7 @@ import ( "syscall" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fsbridge" @@ -117,7 +118,7 @@ func execveat(t *kernel.Task, dirFD int32, pathnameAddr, argvAddr, envvAddr user resolveFinal := flags&linux.AT_SYMLINK_NOFOLLOW == 0 root := t.FSContext().RootDirectory() - defer root.DecRef() + defer root.DecRef(t) var wd *fs.Dirent var executable fsbridge.File @@ -133,7 +134,7 @@ func execveat(t *kernel.Task, dirFD int32, pathnameAddr, argvAddr, envvAddr user if f == nil { return 0, nil, syserror.EBADF } - defer f.DecRef() + defer f.DecRef(t) closeOnExec = fdFlags.CloseOnExec if atEmptyPath && len(pathname) == 0 { @@ -155,7 +156,7 @@ func execveat(t *kernel.Task, dirFD int32, pathnameAddr, argvAddr, envvAddr user } } if wd != nil { - defer wd.DecRef() + defer wd.DecRef(t) } // Load the new TaskContext. @@ -262,7 +263,7 @@ func parseCommonWaitOptions(wopts *kernel.WaitOptions, options int) error { wopts.Events |= kernel.EventGroupContinue } if options&linux.WNOHANG == 0 { - wopts.BlockInterruptErr = kernel.ERESTARTSYS + wopts.BlockInterruptErr = syserror.ERESTARTSYS } if options&linux.WNOTHREAD == 0 { wopts.SiblingChildren = true @@ -311,13 +312,13 @@ func wait4(t *kernel.Task, pid int, statusAddr usermem.Addr, options int, rusage return 0, err } if statusAddr != 0 { - if _, err := t.CopyOut(statusAddr, wr.Status); err != nil { + if _, err := primitive.CopyUint32Out(t, statusAddr, wr.Status); err != nil { return 0, err } } if rusageAddr != 0 { ru := getrusage(wr.Task, linux.RUSAGE_BOTH) - if _, err := t.CopyOut(rusageAddr, &ru); err != nil { + if _, err := ru.CopyOut(t, rusageAddr); err != nil { return 0, err } } @@ -395,14 +396,14 @@ func Waitid(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // as well. if infop != 0 { var si arch.SignalInfo - _, err = t.CopyOut(infop, &si) + _, err = si.CopyOut(t, infop) } } return 0, nil, err } if rusageAddr != 0 { ru := getrusage(wr.Task, linux.RUSAGE_BOTH) - if _, err := t.CopyOut(rusageAddr, &ru); err != nil { + if _, err := ru.CopyOut(t, rusageAddr); err != nil { return 0, nil, err } } @@ -441,7 +442,7 @@ func Waitid(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal default: t.Warningf("waitid got incomprehensible wait status %d", s) } - _, err = t.CopyOut(infop, &si) + _, err = si.CopyOut(t, infop) return 0, nil, err } @@ -558,9 +559,7 @@ func Getcpu(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // third argument to this system call is nowadays unused. if cpu != 0 { - buf := t.CopyScratchBuffer(4) - usermem.ByteOrder.PutUint32(buf, uint32(t.CPU())) - if _, err := t.CopyOutBytes(cpu, buf); err != nil { + if _, err := primitive.CopyInt32Out(t, cpu, t.CPU()); err != nil { return 0, nil, err } } diff --git a/pkg/sentry/syscalls/linux/sys_time.go b/pkg/sentry/syscalls/linux/sys_time.go index 2d2aa0819..c5054d2f1 100644 --- a/pkg/sentry/syscalls/linux/sys_time.go +++ b/pkg/sentry/syscalls/linux/sys_time.go @@ -19,6 +19,7 @@ import ( "time" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" @@ -168,7 +169,7 @@ func Time(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC return uintptr(r), nil, nil } - if _, err := t.CopyOut(addr, r); err != nil { + if _, err := r.CopyOut(t, addr); err != nil { return 0, nil, err } return uintptr(r), nil, nil @@ -213,7 +214,7 @@ func clockNanosleepUntil(t *kernel.Task, c ktime.Clock, ts linux.Timespec) error return nil } - return syserror.ConvertIntr(err, kernel.ERESTARTNOHAND) + return syserror.ConvertIntr(err, syserror.ERESTARTNOHAND) } // clockNanosleepFor blocks for a specified duration. @@ -254,7 +255,7 @@ func clockNanosleepFor(t *kernel.Task, c ktime.Clock, dur time.Duration, rem use duration: remaining, rem: rem, }) - return kernel.ERESTART_RESTARTBLOCK + return syserror.ERESTART_RESTARTBLOCK default: panic(fmt.Sprintf("Impossible BlockWithTimer error %v", err)) } @@ -334,8 +335,8 @@ func Gettimeofday(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. // Ask the time package for the timezone. _, offset := time.Now().Zone() // This int32 array mimics linux's struct timezone. - timezone := [2]int32{-int32(offset) / 60, 0} - _, err := t.CopyOut(tz, timezone) + timezone := []int32{-int32(offset) / 60, 0} + _, err := primitive.CopyInt32SliceOut(t, tz, timezone) return 0, nil, err } return 0, nil, nil diff --git a/pkg/sentry/syscalls/linux/sys_timer.go b/pkg/sentry/syscalls/linux/sys_timer.go index a4c400f87..45eef4feb 100644 --- a/pkg/sentry/syscalls/linux/sys_timer.go +++ b/pkg/sentry/syscalls/linux/sys_timer.go @@ -21,81 +21,63 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" ) const nsecPerSec = int64(time.Second) -// copyItimerValIn copies an ItimerVal from the untrusted app range to the -// kernel. The ItimerVal may be either 32 or 64 bits. -// A NULL address is allowed because because Linux allows -// setitimer(which, NULL, &old_value) which disables the timer. -// There is a KERN_WARN message saying this misfeature will be removed. -// However, that hasn't happened as of 3.19, so we continue to support it. -func copyItimerValIn(t *kernel.Task, addr usermem.Addr) (linux.ItimerVal, error) { - if addr == usermem.Addr(0) { - return linux.ItimerVal{}, nil - } - - switch t.Arch().Width() { - case 8: - // Native size, just copy directly. - var itv linux.ItimerVal - if _, err := t.CopyIn(addr, &itv); err != nil { - return linux.ItimerVal{}, err - } - - return itv, nil - default: - return linux.ItimerVal{}, syserror.ENOSYS - } -} - -// copyItimerValOut copies an ItimerVal to the untrusted app range. -// The ItimerVal may be either 32 or 64 bits. -// A NULL address is allowed, in which case no copy takes place -func copyItimerValOut(t *kernel.Task, addr usermem.Addr, itv *linux.ItimerVal) error { - if addr == usermem.Addr(0) { - return nil - } - - switch t.Arch().Width() { - case 8: - // Native size, just copy directly. - _, err := t.CopyOut(addr, itv) - return err - default: - return syserror.ENOSYS - } -} - // Getitimer implements linux syscall getitimer(2). func Getitimer(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + if t.Arch().Width() != 8 { + // Definition of linux.ItimerVal assumes 64-bit architecture. + return 0, nil, syserror.ENOSYS + } + timerID := args[0].Int() - val := args[1].Pointer() + addr := args[1].Pointer() olditv, err := t.Getitimer(timerID) if err != nil { return 0, nil, err } - return 0, nil, copyItimerValOut(t, val, &olditv) + // A NULL address is allowed, in which case no copy out takes place. + if addr == 0 { + return 0, nil, nil + } + _, err = olditv.CopyOut(t, addr) + return 0, nil, err } // Setitimer implements linux syscall setitimer(2). func Setitimer(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - timerID := args[0].Int() - newVal := args[1].Pointer() - oldVal := args[2].Pointer() + if t.Arch().Width() != 8 { + // Definition of linux.ItimerVal assumes 64-bit architecture. + return 0, nil, syserror.ENOSYS + } - newitv, err := copyItimerValIn(t, newVal) - if err != nil { - return 0, nil, err + timerID := args[0].Int() + newAddr := args[1].Pointer() + oldAddr := args[2].Pointer() + + var newitv linux.ItimerVal + // A NULL address is allowed because because Linux allows + // setitimer(which, NULL, &old_value) which disables the timer. There is a + // KERN_WARN message saying this misfeature will be removed. However, that + // hasn't happened as of 3.19, so we continue to support it. + if newAddr != 0 { + if _, err := newitv.CopyIn(t, newAddr); err != nil { + return 0, nil, err + } } olditv, err := t.Setitimer(timerID, newitv) if err != nil { return 0, nil, err } - return 0, nil, copyItimerValOut(t, oldVal, &olditv) + // A NULL address is allowed, in which case no copy out takes place. + if oldAddr == 0 { + return 0, nil, nil + } + _, err = olditv.CopyOut(t, oldAddr) + return 0, nil, err } // Alarm implements linux syscall alarm(2). @@ -131,7 +113,7 @@ func TimerCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S var sev *linux.Sigevent if sevp != 0 { sev = &linux.Sigevent{} - if _, err = t.CopyIn(sevp, sev); err != nil { + if _, err = sev.CopyIn(t, sevp); err != nil { return 0, nil, err } } @@ -141,7 +123,7 @@ func TimerCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S return 0, nil, err } - if _, err := t.CopyOut(timerIDp, &id); err != nil { + if _, err := id.CopyOut(t, timerIDp); err != nil { t.IntervalTimerDelete(id) return 0, nil, err } @@ -157,7 +139,7 @@ func TimerSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. oldValAddr := args[3].Pointer() var newVal linux.Itimerspec - if _, err := t.CopyIn(newValAddr, &newVal); err != nil { + if _, err := newVal.CopyIn(t, newValAddr); err != nil { return 0, nil, err } oldVal, err := t.IntervalTimerSettime(timerID, newVal, flags&linux.TIMER_ABSTIME != 0) @@ -165,9 +147,8 @@ func TimerSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. return 0, nil, err } if oldValAddr != 0 { - if _, err := t.CopyOut(oldValAddr, &oldVal); err != nil { - return 0, nil, err - } + _, err = oldVal.CopyOut(t, oldValAddr) + return 0, nil, err } return 0, nil, nil } @@ -181,7 +162,7 @@ func TimerGettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. if err != nil { return 0, nil, err } - _, err = t.CopyOut(curValAddr, &curVal) + _, err = curVal.CopyOut(t, curValAddr) return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/sys_timerfd.go b/pkg/sentry/syscalls/linux/sys_timerfd.go index cf49b43db..cadd9d348 100644 --- a/pkg/sentry/syscalls/linux/sys_timerfd.go +++ b/pkg/sentry/syscalls/linux/sys_timerfd.go @@ -43,7 +43,7 @@ func TimerfdCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel return 0, nil, syserror.EINVAL } f := timerfd.NewFile(t, c) - defer f.DecRef() + defer f.DecRef(t) f.SetFlags(fs.SettableFileFlags{ NonBlocking: flags&linux.TFD_NONBLOCK != 0, }) @@ -73,7 +73,7 @@ func TimerfdSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne if f == nil { return 0, nil, syserror.EBADF } - defer f.DecRef() + defer f.DecRef(t) tf, ok := f.FileOperations.(*timerfd.TimerOperations) if !ok { @@ -81,7 +81,7 @@ func TimerfdSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne } var newVal linux.Itimerspec - if _, err := t.CopyIn(newValAddr, &newVal); err != nil { + if _, err := newVal.CopyIn(t, newValAddr); err != nil { return 0, nil, err } newS, err := ktime.SettingFromItimerspec(newVal, flags&linux.TFD_TIMER_ABSTIME != 0, tf.Clock()) @@ -91,7 +91,7 @@ func TimerfdSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne tm, oldS := tf.SetTime(newS) if oldValAddr != 0 { oldVal := ktime.ItimerspecFromSetting(tm, oldS) - if _, err := t.CopyOut(oldValAddr, &oldVal); err != nil { + if _, err := oldVal.CopyOut(t, oldValAddr); err != nil { return 0, nil, err } } @@ -107,7 +107,7 @@ func TimerfdGettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne if f == nil { return 0, nil, syserror.EBADF } - defer f.DecRef() + defer f.DecRef(t) tf, ok := f.FileOperations.(*timerfd.TimerOperations) if !ok { @@ -116,6 +116,6 @@ func TimerfdGettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne tm, s := tf.GetTime() curVal := ktime.ItimerspecFromSetting(tm, s) - _, err := t.CopyOut(curValAddr, &curVal) + _, err := curVal.CopyOut(t, curValAddr) return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/sys_tls_amd64.go b/pkg/sentry/syscalls/linux/sys_tls_amd64.go index b3eb96a1c..6ddd30d5c 100644 --- a/pkg/sentry/syscalls/linux/sys_tls_amd64.go +++ b/pkg/sentry/syscalls/linux/sys_tls_amd64.go @@ -18,6 +18,7 @@ package linux import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/syserror" @@ -30,17 +31,19 @@ func ArchPrctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys case linux.ARCH_GET_FS: addr := args[1].Pointer() fsbase := t.Arch().TLS() - _, err := t.CopyOut(addr, uint64(fsbase)) - if err != nil { - return 0, nil, err + switch t.Arch().Width() { + case 8: + if _, err := primitive.CopyUint64Out(t, addr, uint64(fsbase)); err != nil { + return 0, nil, err + } + default: + return 0, nil, syserror.ENOSYS } - case linux.ARCH_SET_FS: fsbase := args[1].Uint64() if !t.Arch().SetTLS(uintptr(fsbase)) { return 0, nil, syserror.EPERM } - case linux.ARCH_GET_GS, linux.ARCH_SET_GS: t.Kernel().EmitUnimplementedEvent(t) fallthrough diff --git a/pkg/sentry/syscalls/linux/sys_utsname.go b/pkg/sentry/syscalls/linux/sys_utsname.go index e9d702e8e..66c5974f5 100644 --- a/pkg/sentry/syscalls/linux/sys_utsname.go +++ b/pkg/sentry/syscalls/linux/sys_utsname.go @@ -46,7 +46,7 @@ func Uname(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // Copy out the result. va := args[0].Pointer() - _, err := t.CopyOut(va, u) + _, err := u.CopyOut(t, va) return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/sys_write.go b/pkg/sentry/syscalls/linux/sys_write.go index 6ec0de96e..95bfe6606 100644 --- a/pkg/sentry/syscalls/linux/sys_write.go +++ b/pkg/sentry/syscalls/linux/sys_write.go @@ -48,7 +48,7 @@ func Write(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the file is writable. if !file.Flags().Write { @@ -71,7 +71,7 @@ func Write(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall n, err := writev(t, file, src) t.IOUsage().AccountWriteSyscall(n) - return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "write", file) + return uintptr(n), nil, handleIOError(t, n != 0, err, syserror.ERESTARTSYS, "write", file) } // Pwrite64 implements linux syscall pwrite64(2). @@ -85,7 +85,7 @@ func Pwrite64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate and does not overflow. if offset < 0 || offset+int64(size) < 0 { @@ -118,7 +118,7 @@ func Pwrite64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc n, err := pwritev(t, file, src, offset) t.IOUsage().AccountWriteSyscall(n) - return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "pwrite64", file) + return uintptr(n), nil, handleIOError(t, n != 0, err, syserror.ERESTARTSYS, "pwrite64", file) } // Writev implements linux syscall writev(2). @@ -131,7 +131,7 @@ func Writev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the file is writable. if !file.Flags().Write { @@ -148,7 +148,7 @@ func Writev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal n, err := writev(t, file, src) t.IOUsage().AccountWriteSyscall(n) - return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "writev", file) + return uintptr(n), nil, handleIOError(t, n != 0, err, syserror.ERESTARTSYS, "writev", file) } // Pwritev implements linux syscall pwritev(2). @@ -162,7 +162,7 @@ func Pwritev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate. if offset < 0 { @@ -189,7 +189,7 @@ func Pwritev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca n, err := pwritev(t, file, src, offset) t.IOUsage().AccountWriteSyscall(n) - return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "pwritev", file) + return uintptr(n), nil, handleIOError(t, n != 0, err, syserror.ERESTARTSYS, "pwritev", file) } // Pwritev2 implements linux syscall pwritev2(2). @@ -215,7 +215,7 @@ func Pwritev2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate. if offset < -1 { @@ -250,12 +250,12 @@ func Pwritev2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if offset == -1 { n, err := writev(t, file, src) t.IOUsage().AccountWriteSyscall(n) - return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "pwritev2", file) + return uintptr(n), nil, handleIOError(t, n != 0, err, syserror.ERESTARTSYS, "pwritev2", file) } n, err := pwritev(t, file, src, offset) t.IOUsage().AccountWriteSyscall(n) - return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "pwritev2", file) + return uintptr(n), nil, handleIOError(t, n != 0, err, syserror.ERESTARTSYS, "pwritev2", file) } func writev(t *kernel.Task, f *fs.File, src usermem.IOSequence) (int64, error) { diff --git a/pkg/sentry/syscalls/linux/sys_xattr.go b/pkg/sentry/syscalls/linux/sys_xattr.go index c24946160..97474fd3c 100644 --- a/pkg/sentry/syscalls/linux/sys_xattr.go +++ b/pkg/sentry/syscalls/linux/sys_xattr.go @@ -49,7 +49,7 @@ func FGetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if f == nil { return 0, nil, syserror.EBADF } - defer f.DecRef() + defer f.DecRef(t) n, err := getXattr(t, f.Dirent, nameAddr, valueAddr, size) if err != nil { @@ -153,7 +153,7 @@ func FSetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if f == nil { return 0, nil, syserror.EBADF } - defer f.DecRef() + defer f.DecRef(t) return 0, nil, setXattr(t, f.Dirent, nameAddr, valueAddr, uint64(size), flags) } @@ -270,7 +270,7 @@ func FListXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy if f == nil { return 0, nil, syserror.EBADF } - defer f.DecRef() + defer f.DecRef(t) n, err := listXattr(t, f.Dirent, listAddr, size) if err != nil { @@ -384,7 +384,7 @@ func FRemoveXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. if f == nil { return 0, nil, syserror.EBADF } - defer f.DecRef() + defer f.DecRef(t) return 0, nil, removeXattr(t, f.Dirent, nameAddr) } diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD index 64696b438..9ee766552 100644 --- a/pkg/sentry/syscalls/linux/vfs2/BUILD +++ b/pkg/sentry/syscalls/linux/vfs2/BUILD @@ -44,6 +44,9 @@ go_library( "//pkg/context", "//pkg/fspath", "//pkg/gohacks", + "//pkg/log", + "//pkg/marshal", + "//pkg/marshal/primitive", "//pkg/sentry/arch", "//pkg/sentry/fs/lock", "//pkg/sentry/fsbridge", @@ -72,7 +75,5 @@ go_library( "//pkg/syserror", "//pkg/usermem", "//pkg/waiter", - "//tools/go_marshal/marshal", - "//tools/go_marshal/primitive", ], ) diff --git a/pkg/sentry/syscalls/linux/vfs2/aio.go b/pkg/sentry/syscalls/linux/vfs2/aio.go index e5cdefc50..6d0a38330 100644 --- a/pkg/sentry/syscalls/linux/vfs2/aio.go +++ b/pkg/sentry/syscalls/linux/vfs2/aio.go @@ -17,6 +17,7 @@ package vfs2 import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fsimpl/eventfd" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -38,21 +39,27 @@ func IoSubmit(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc } for i := int32(0); i < nrEvents; i++ { - // Copy in the address. - cbAddrNative := t.Arch().Native(0) - if _, err := t.CopyIn(addr, cbAddrNative); err != nil { - if i > 0 { - // Some successful. - return uintptr(i), nil, nil + // Copy in the callback address. + var cbAddr usermem.Addr + switch t.Arch().Width() { + case 8: + var cbAddrP primitive.Uint64 + if _, err := cbAddrP.CopyIn(t, addr); err != nil { + if i > 0 { + // Some successful. + return uintptr(i), nil, nil + } + // Nothing done. + return 0, nil, err } - // Nothing done. - return 0, nil, err + cbAddr = usermem.Addr(cbAddrP) + default: + return 0, nil, syserror.ENOSYS } // Copy in this callback. var cb linux.IOCallback - cbAddr := usermem.Addr(t.Arch().Value(cbAddrNative)) - if _, err := t.CopyIn(cbAddr, &cb); err != nil { + if _, err := cb.CopyIn(t, cbAddr); err != nil { if i > 0 { // Some have been successful. return uintptr(i), nil, nil @@ -88,7 +95,7 @@ func submitCallback(t *kernel.Task, id uint64, cb *linux.IOCallback, cbAddr user if fd == nil { return syserror.EBADF } - defer fd.DecRef() + defer fd.DecRef(t) // Was there an eventFD? Extract it. var eventFD *vfs.FileDescription @@ -97,7 +104,7 @@ func submitCallback(t *kernel.Task, id uint64, cb *linux.IOCallback, cbAddr user if eventFD == nil { return syserror.EBADF } - defer eventFD.DecRef() + defer eventFD.DecRef(t) // Check that it is an eventfd. if _, ok := eventFD.Impl().(*eventfd.EventFileDescription); !ok { @@ -144,6 +151,12 @@ func submitCallback(t *kernel.Task, id uint64, cb *linux.IOCallback, cbAddr user func getAIOCallback(t *kernel.Task, fd, eventFD *vfs.FileDescription, cbAddr usermem.Addr, cb *linux.IOCallback, ioseq usermem.IOSequence, aioCtx *mm.AIOContext) kernel.AIOCallback { return func(ctx context.Context) { + // Release references after completing the callback. + defer fd.DecRef(ctx) + if eventFD != nil { + defer eventFD.DecRef(ctx) + } + if aioCtx.Dead() { aioCtx.CancelPendingRequest() return @@ -169,8 +182,6 @@ func getAIOCallback(t *kernel.Task, fd, eventFD *vfs.FileDescription, cbAddr use ev.Result = -int64(kernel.ExtractErrno(err, 0)) } - fd.DecRef() - // Queue the result for delivery. aioCtx.FinishRequest(ev) @@ -179,7 +190,6 @@ func getAIOCallback(t *kernel.Task, fd, eventFD *vfs.FileDescription, cbAddr use // wake up. if eventFD != nil { eventFD.Impl().(*eventfd.EventFileDescription).Signal(1) - eventFD.DecRef() } } } diff --git a/pkg/sentry/syscalls/linux/vfs2/epoll.go b/pkg/sentry/syscalls/linux/vfs2/epoll.go index 34c90ae3e..d0cbb77eb 100644 --- a/pkg/sentry/syscalls/linux/vfs2/epoll.go +++ b/pkg/sentry/syscalls/linux/vfs2/epoll.go @@ -24,7 +24,6 @@ import ( ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) @@ -37,11 +36,11 @@ func EpollCreate1(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. return 0, nil, syserror.EINVAL } - file, err := t.Kernel().VFS().NewEpollInstanceFD() + file, err := t.Kernel().VFS().NewEpollInstanceFD(t) if err != nil { return 0, nil, err } - defer file.DecRef() + defer file.DecRef(t) fd, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{ CloseOnExec: flags&linux.EPOLL_CLOEXEC != 0, @@ -62,11 +61,11 @@ func EpollCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S return 0, nil, syserror.EINVAL } - file, err := t.Kernel().VFS().NewEpollInstanceFD() + file, err := t.Kernel().VFS().NewEpollInstanceFD(t) if err != nil { return 0, nil, err } - defer file.DecRef() + defer file.DecRef(t) fd, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{}) if err != nil { @@ -86,7 +85,7 @@ func EpollCtl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if epfile == nil { return 0, nil, syserror.EBADF } - defer epfile.DecRef() + defer epfile.DecRef(t) ep, ok := epfile.Impl().(*vfs.EpollInstance) if !ok { return 0, nil, syserror.EINVAL @@ -95,7 +94,7 @@ func EpollCtl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) if epfile == file { return 0, nil, syserror.EINVAL } @@ -135,56 +134,32 @@ func EpollWait(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if epfile == nil { return 0, nil, syserror.EBADF } - defer epfile.DecRef() + defer epfile.DecRef(t) ep, ok := epfile.Impl().(*vfs.EpollInstance) if !ok { return 0, nil, syserror.EINVAL } - // Use a fixed-size buffer in a loop, instead of make([]linux.EpollEvent, - // maxEvents), so that the buffer can be allocated on the stack. + // Allocate space for a few events on the stack for the common case in + // which we don't have too many events. var ( - events [16]linux.EpollEvent - total int + eventsArr [16]linux.EpollEvent ch chan struct{} haveDeadline bool deadline ktime.Time ) for { - batchEvents := len(events) - if batchEvents > maxEvents { - batchEvents = maxEvents - } - n := ep.ReadEvents(events[:batchEvents]) - maxEvents -= n - if n != 0 { - // Copy what we read out. - copiedBytes, err := linux.CopyEpollEventSliceOut(t, eventsAddr, events[:n]) + events := ep.ReadEvents(eventsArr[:0], maxEvents) + if len(events) != 0 { + copiedBytes, err := linux.CopyEpollEventSliceOut(t, eventsAddr, events) copiedEvents := copiedBytes / sizeofEpollEvent // rounded down - eventsAddr += usermem.Addr(copiedEvents * sizeofEpollEvent) - total += copiedEvents - if err != nil { - if total != 0 { - return uintptr(total), nil, nil - } - return 0, nil, err - } - // If we've filled the application's event buffer, we're done. - if maxEvents == 0 { - return uintptr(total), nil, nil - } - // Loop if we read a full batch, under the expectation that there - // may be more events to read. - if n == batchEvents { - continue + if copiedEvents != 0 { + return uintptr(copiedEvents), nil, nil } + return 0, nil, err } - // We get here if n != batchEvents. If we read any number of events - // (just now, or in a previous iteration of this loop), or if timeout - // is 0 (such that epoll_wait should be non-blocking), return the - // events we've read so far to the application. - if total != 0 || timeout == 0 { - return uintptr(total), nil, nil + if timeout == 0 { + return 0, nil, nil } // In the first iteration of this loop, register with the epoll // instance for readability events, but then immediately continue the @@ -207,8 +182,6 @@ func EpollWait(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if err == syserror.ETIMEDOUT { err = nil } - // total must be 0 since otherwise we would have returned - // above. return 0, nil, err } } diff --git a/pkg/sentry/syscalls/linux/vfs2/eventfd.go b/pkg/sentry/syscalls/linux/vfs2/eventfd.go index aff1a2070..807f909da 100644 --- a/pkg/sentry/syscalls/linux/vfs2/eventfd.go +++ b/pkg/sentry/syscalls/linux/vfs2/eventfd.go @@ -38,11 +38,11 @@ func Eventfd2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc fileFlags |= linux.O_NONBLOCK } semMode := flags&linux.EFD_SEMAPHORE != 0 - eventfd, err := eventfd.New(vfsObj, initVal, semMode, fileFlags) + eventfd, err := eventfd.New(t, vfsObj, initVal, semMode, fileFlags) if err != nil { return 0, nil, err } - defer eventfd.DecRef() + defer eventfd.DecRef(t) fd, err := t.NewFDFromVFS2(0, eventfd, kernel.FDFlags{ CloseOnExec: flags&linux.EFD_CLOEXEC != 0, diff --git a/pkg/sentry/syscalls/linux/vfs2/execve.go b/pkg/sentry/syscalls/linux/vfs2/execve.go index aef0078a8..066ee0863 100644 --- a/pkg/sentry/syscalls/linux/vfs2/execve.go +++ b/pkg/sentry/syscalls/linux/vfs2/execve.go @@ -71,7 +71,7 @@ func execveat(t *kernel.Task, dirfd int32, pathnameAddr, argvAddr, envvAddr user } root := t.FSContext().RootDirectoryVFS2() - defer root.DecRef() + defer root.DecRef(t) var executable fsbridge.File closeOnExec := false if path := fspath.Parse(pathname); dirfd != linux.AT_FDCWD && !path.Absolute { @@ -90,7 +90,7 @@ func execveat(t *kernel.Task, dirfd int32, pathnameAddr, argvAddr, envvAddr user } start := dirfile.VirtualDentry() start.IncRef() - dirfile.DecRef() + dirfile.DecRef(t) closeOnExec = dirfileFlags.CloseOnExec file, err := t.Kernel().VFS().OpenAt(t, t.Credentials(), &vfs.PathOperation{ Root: root, @@ -101,19 +101,19 @@ func execveat(t *kernel.Task, dirfd int32, pathnameAddr, argvAddr, envvAddr user Flags: linux.O_RDONLY, FileExec: true, }) - start.DecRef() + start.DecRef(t) if err != nil { return 0, nil, err } - defer file.DecRef() + defer file.DecRef(t) executable = fsbridge.NewVFSFile(file) } // Load the new TaskContext. mntns := t.MountNamespaceVFS2() // FIXME(jamieliu): useless refcount change - defer mntns.DecRef() + defer mntns.DecRef(t) wd := t.FSContext().WorkingDirectoryVFS2() - defer wd.DecRef() + defer wd.DecRef(t) remainingTraversals := uint(linux.MaxSymlinkTraversals) loadArgs := loader.LoadArgs{ Opener: fsbridge.NewVFSLookup(mntns, root, wd), diff --git a/pkg/sentry/syscalls/linux/vfs2/fd.go b/pkg/sentry/syscalls/linux/vfs2/fd.go index 67f191551..d8b8d9783 100644 --- a/pkg/sentry/syscalls/linux/vfs2/fd.go +++ b/pkg/sentry/syscalls/linux/vfs2/fd.go @@ -34,11 +34,11 @@ func Close(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // Note that Remove provides a reference on the file that we may use to // flush. It is still active until we drop the final reference below // (and other reference-holding operations complete). - _, file := t.FDTable().Remove(fd) + _, file := t.FDTable().Remove(t, fd) if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) err := file.OnClose(t) return 0, nil, slinux.HandleIOErrorVFS2(t, false /* partial */, err, syserror.EINTR, "close", file) @@ -52,7 +52,7 @@ func Dup(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) newFD, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{}) if err != nil { @@ -72,7 +72,7 @@ func Dup2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if file == nil { return 0, nil, syserror.EBADF } - file.DecRef() + file.DecRef(t) return uintptr(newfd), nil, nil } @@ -101,7 +101,7 @@ func dup3(t *kernel.Task, oldfd, newfd int32, flags uint32) (uintptr, *kernel.Sy if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) err := t.NewFDAtVFS2(newfd, file, kernel.FDFlags{ CloseOnExec: flags&linux.O_CLOEXEC != 0, @@ -121,7 +121,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) switch cmd { case linux.F_DUPFD, linux.F_DUPFD_CLOEXEC: @@ -137,7 +137,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return uintptr(flags.ToLinuxFDFlags()), nil, nil case linux.F_SETFD: flags := args[2].Uint() - err := t.FDTable().SetFlagsVFS2(fd, kernel.FDFlags{ + err := t.FDTable().SetFlagsVFS2(t, fd, kernel.FDFlags{ CloseOnExec: flags&linux.FD_CLOEXEC != 0, }) return 0, nil, err @@ -181,11 +181,11 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if !hasOwner { return 0, nil, nil } - _, err := t.CopyOut(args[2].Pointer(), &owner) + _, err := owner.CopyOut(t, args[2].Pointer()) return 0, nil, err case linux.F_SETOWN_EX: var owner linux.FOwnerEx - _, err := t.CopyIn(args[2].Pointer(), &owner) + _, err := owner.CopyIn(t, args[2].Pointer()) if err != nil { return 0, nil, err } @@ -208,7 +208,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.F_SETLK, linux.F_SETLKW: return 0, nil, posixLock(t, args, file, cmd) default: - // TODO(gvisor.dev/issue/2920): Everything else is not yet supported. + // Everything else is not yet supported. return 0, nil, syserror.EINVAL } } @@ -286,7 +286,7 @@ func posixLock(t *kernel.Task, args arch.SyscallArguments, file *vfs.FileDescrip // Copy in the lock request. flockAddr := args[2].Pointer() var flock linux.Flock - if _, err := t.CopyIn(flockAddr, &flock); err != nil { + if _, err := flock.CopyIn(t, flockAddr); err != nil { return err } @@ -332,7 +332,7 @@ func Fadvise64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // If the FD refers to a pipe or FIFO, return error. if _, isPipe := file.Impl().(*pipe.VFSPipeFD); isPipe { diff --git a/pkg/sentry/syscalls/linux/vfs2/filesystem.go b/pkg/sentry/syscalls/linux/vfs2/filesystem.go index b6d2ddd65..01e0f9010 100644 --- a/pkg/sentry/syscalls/linux/vfs2/filesystem.go +++ b/pkg/sentry/syscalls/linux/vfs2/filesystem.go @@ -56,7 +56,7 @@ func linkat(t *kernel.Task, olddirfd int32, oldpathAddr usermem.Addr, newdirfd i if err != nil { return err } - defer oldtpop.Release() + defer oldtpop.Release(t) newpath, err := copyInPath(t, newpathAddr) if err != nil { @@ -66,7 +66,7 @@ func linkat(t *kernel.Task, olddirfd int32, oldpathAddr usermem.Addr, newdirfd i if err != nil { return err } - defer newtpop.Release() + defer newtpop.Release(t) return t.Kernel().VFS().LinkAt(t, t.Credentials(), &oldtpop.pop, &newtpop.pop) } @@ -95,7 +95,7 @@ func mkdirat(t *kernel.Task, dirfd int32, addr usermem.Addr, mode uint) error { if err != nil { return err } - defer tpop.Release() + defer tpop.Release(t) return t.Kernel().VFS().MkdirAt(t, t.Credentials(), &tpop.pop, &vfs.MkdirOptions{ Mode: linux.FileMode(mode & (0777 | linux.S_ISVTX) &^ t.FSContext().Umask()), }) @@ -127,7 +127,7 @@ func mknodat(t *kernel.Task, dirfd int32, addr usermem.Addr, mode linux.FileMode if err != nil { return err } - defer tpop.Release() + defer tpop.Release(t) // "Zero file type is equivalent to type S_IFREG." - mknod(2) if mode.FileType() == 0 { @@ -174,7 +174,7 @@ func openat(t *kernel.Task, dirfd int32, pathAddr usermem.Addr, flags uint32, mo if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) file, err := t.Kernel().VFS().OpenAt(t, t.Credentials(), &tpop.pop, &vfs.OpenOptions{ Flags: flags | linux.O_LARGEFILE, @@ -183,7 +183,7 @@ func openat(t *kernel.Task, dirfd int32, pathAddr usermem.Addr, flags uint32, mo if err != nil { return 0, nil, err } - defer file.DecRef() + defer file.DecRef(t) fd, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{ CloseOnExec: flags&linux.O_CLOEXEC != 0, @@ -227,7 +227,7 @@ func renameat(t *kernel.Task, olddirfd int32, oldpathAddr usermem.Addr, newdirfd if err != nil { return err } - defer oldtpop.Release() + defer oldtpop.Release(t) newpath, err := copyInPath(t, newpathAddr) if err != nil { @@ -237,7 +237,7 @@ func renameat(t *kernel.Task, olddirfd int32, oldpathAddr usermem.Addr, newdirfd if err != nil { return err } - defer newtpop.Release() + defer newtpop.Release(t) return t.Kernel().VFS().RenameAt(t, t.Credentials(), &oldtpop.pop, &newtpop.pop, &vfs.RenameOptions{ Flags: flags, @@ -259,7 +259,7 @@ func rmdirat(t *kernel.Task, dirfd int32, pathAddr usermem.Addr) error { if err != nil { return err } - defer tpop.Release() + defer tpop.Release(t) return t.Kernel().VFS().RmdirAt(t, t.Credentials(), &tpop.pop) } @@ -278,7 +278,7 @@ func unlinkat(t *kernel.Task, dirfd int32, pathAddr usermem.Addr) error { if err != nil { return err } - defer tpop.Release() + defer tpop.Release(t) return t.Kernel().VFS().UnlinkAt(t, t.Credentials(), &tpop.pop) } @@ -329,6 +329,6 @@ func symlinkat(t *kernel.Task, targetAddr usermem.Addr, newdirfd int32, linkpath if err != nil { return err } - defer tpop.Release() + defer tpop.Release(t) return t.Kernel().VFS().SymlinkAt(t, t.Credentials(), &tpop.pop, target) } diff --git a/pkg/sentry/syscalls/linux/vfs2/fscontext.go b/pkg/sentry/syscalls/linux/vfs2/fscontext.go index 317409a18..a7d4d2a36 100644 --- a/pkg/sentry/syscalls/linux/vfs2/fscontext.go +++ b/pkg/sentry/syscalls/linux/vfs2/fscontext.go @@ -31,8 +31,8 @@ func Getcwd(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal root := t.FSContext().RootDirectoryVFS2() wd := t.FSContext().WorkingDirectoryVFS2() s, err := t.Kernel().VFS().PathnameForGetcwd(t, root, wd) - root.DecRef() - wd.DecRef() + root.DecRef(t) + wd.DecRef(t) if err != nil { return 0, nil, err } @@ -67,7 +67,7 @@ func Chdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) vd, err := t.Kernel().VFS().GetDentryAt(t, t.Credentials(), &tpop.pop, &vfs.GetDentryOptions{ CheckSearchable: true, @@ -75,8 +75,8 @@ func Chdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if err != nil { return 0, nil, err } - t.FSContext().SetWorkingDirectoryVFS2(vd) - vd.DecRef() + t.FSContext().SetWorkingDirectoryVFS2(t, vd) + vd.DecRef(t) return 0, nil, nil } @@ -88,7 +88,7 @@ func Fchdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) vd, err := t.Kernel().VFS().GetDentryAt(t, t.Credentials(), &tpop.pop, &vfs.GetDentryOptions{ CheckSearchable: true, @@ -96,8 +96,8 @@ func Fchdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, err } - t.FSContext().SetWorkingDirectoryVFS2(vd) - vd.DecRef() + t.FSContext().SetWorkingDirectoryVFS2(t, vd) + vd.DecRef(t) return 0, nil, nil } @@ -117,7 +117,7 @@ func Chroot(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) vd, err := t.Kernel().VFS().GetDentryAt(t, t.Credentials(), &tpop.pop, &vfs.GetDentryOptions{ CheckSearchable: true, @@ -125,7 +125,7 @@ func Chroot(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, err } - t.FSContext().SetRootDirectoryVFS2(vd) - vd.DecRef() + t.FSContext().SetRootDirectoryVFS2(t, vd) + vd.DecRef(t) return 0, nil, nil } diff --git a/pkg/sentry/syscalls/linux/vfs2/getdents.go b/pkg/sentry/syscalls/linux/vfs2/getdents.go index c7c7bf7ce..5517595b5 100644 --- a/pkg/sentry/syscalls/linux/vfs2/getdents.go +++ b/pkg/sentry/syscalls/linux/vfs2/getdents.go @@ -44,7 +44,7 @@ func getdents(t *kernel.Task, args arch.SyscallArguments, isGetdents64 bool) (ui if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) cb := getGetdentsCallback(t, addr, size, isGetdents64) err := file.IterDirents(t, cb) diff --git a/pkg/sentry/syscalls/linux/vfs2/inotify.go b/pkg/sentry/syscalls/linux/vfs2/inotify.go index 5d98134a5..11753d8e5 100644 --- a/pkg/sentry/syscalls/linux/vfs2/inotify.go +++ b/pkg/sentry/syscalls/linux/vfs2/inotify.go @@ -35,7 +35,7 @@ func InotifyInit1(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. if err != nil { return 0, nil, err } - defer ino.DecRef() + defer ino.DecRef(t) fd, err := t.NewFDFromVFS2(0, ino, kernel.FDFlags{ CloseOnExec: flags&linux.IN_CLOEXEC != 0, @@ -66,7 +66,7 @@ func fdToInotify(t *kernel.Task, fd int32) (*vfs.Inotify, *vfs.FileDescription, ino, ok := f.Impl().(*vfs.Inotify) if !ok { // Not an inotify fd. - f.DecRef() + f.DecRef(t) return nil, nil, syserror.EINVAL } @@ -96,7 +96,7 @@ func InotifyAddWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kern if err != nil { return 0, nil, err } - defer f.DecRef() + defer f.DecRef(t) path, err := copyInPath(t, addr) if err != nil { @@ -109,12 +109,12 @@ func InotifyAddWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kern if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) d, err := t.Kernel().VFS().GetDentryAt(t, t.Credentials(), &tpop.pop, &vfs.GetDentryOptions{}) if err != nil { return 0, nil, err } - defer d.DecRef() + defer d.DecRef(t) fd, err = ino.AddWatch(d.Dentry(), mask) if err != nil { @@ -132,6 +132,6 @@ func InotifyRmWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne if err != nil { return 0, nil, err } - defer f.DecRef() - return 0, nil, ino.RmWatch(wd) + defer f.DecRef(t) + return 0, nil, ino.RmWatch(t, wd) } diff --git a/pkg/sentry/syscalls/linux/vfs2/ioctl.go b/pkg/sentry/syscalls/linux/vfs2/ioctl.go index fd6ab94b2..2806c3f6f 100644 --- a/pkg/sentry/syscalls/linux/vfs2/ioctl.go +++ b/pkg/sentry/syscalls/linux/vfs2/ioctl.go @@ -16,6 +16,7 @@ package vfs2 import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/syserror" @@ -29,25 +30,25 @@ func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Handle ioctls that apply to all FDs. switch args[1].Int() { case linux.FIONCLEX: - t.FDTable().SetFlagsVFS2(fd, kernel.FDFlags{ + t.FDTable().SetFlagsVFS2(t, fd, kernel.FDFlags{ CloseOnExec: false, }) return 0, nil, nil case linux.FIOCLEX: - t.FDTable().SetFlagsVFS2(fd, kernel.FDFlags{ + t.FDTable().SetFlagsVFS2(t, fd, kernel.FDFlags{ CloseOnExec: true, }) return 0, nil, nil case linux.FIONBIO: var set int32 - if _, err := t.CopyIn(args[2].Pointer(), &set); err != nil { + if _, err := primitive.CopyInt32In(t, args[2].Pointer(), &set); err != nil { return 0, nil, err } flags := file.StatusFlags() @@ -60,7 +61,7 @@ func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.FIOASYNC: var set int32 - if _, err := t.CopyIn(args[2].Pointer(), &set); err != nil { + if _, err := primitive.CopyInt32In(t, args[2].Pointer(), &set); err != nil { return 0, nil, err } flags := file.StatusFlags() @@ -82,12 +83,12 @@ func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall who = owner.PID } } - _, err := t.CopyOut(args[2].Pointer(), &who) + _, err := primitive.CopyInt32Out(t, args[2].Pointer(), who) return 0, nil, err case linux.FIOSETOWN, linux.SIOCSPGRP: var who int32 - if _, err := t.CopyIn(args[2].Pointer(), &who); err != nil { + if _, err := primitive.CopyInt32In(t, args[2].Pointer(), &who); err != nil { return 0, nil, err } ownerType := int32(linux.F_OWNER_PID) diff --git a/pkg/sentry/syscalls/linux/vfs2/lock.go b/pkg/sentry/syscalls/linux/vfs2/lock.go index bf19028c4..b910b5a74 100644 --- a/pkg/sentry/syscalls/linux/vfs2/lock.go +++ b/pkg/sentry/syscalls/linux/vfs2/lock.go @@ -32,7 +32,7 @@ func Flock(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // flock(2): EBADF fd is not an open file descriptor. return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) nonblocking := operation&linux.LOCK_NB != 0 operation &^= linux.LOCK_NB diff --git a/pkg/sentry/syscalls/linux/vfs2/memfd.go b/pkg/sentry/syscalls/linux/vfs2/memfd.go index bbe248d17..c4c0f9e0a 100644 --- a/pkg/sentry/syscalls/linux/vfs2/memfd.go +++ b/pkg/sentry/syscalls/linux/vfs2/memfd.go @@ -47,10 +47,11 @@ func MemfdCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S } shmMount := t.Kernel().ShmMount() - file, err := tmpfs.NewMemfd(shmMount, t.Credentials(), allowSeals, memfdPrefix+name) + file, err := tmpfs.NewMemfd(t, t.Credentials(), shmMount, allowSeals, memfdPrefix+name) if err != nil { return 0, nil, err } + defer file.DecRef(t) fd, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{ CloseOnExec: cloExec, diff --git a/pkg/sentry/syscalls/linux/vfs2/mmap.go b/pkg/sentry/syscalls/linux/vfs2/mmap.go index 60a43f0a0..9d9dbf775 100644 --- a/pkg/sentry/syscalls/linux/vfs2/mmap.go +++ b/pkg/sentry/syscalls/linux/vfs2/mmap.go @@ -17,6 +17,7 @@ package vfs2 import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/syserror" @@ -61,7 +62,7 @@ func Mmap(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC } defer func() { if opts.MappingIdentity != nil { - opts.MappingIdentity.DecRef() + opts.MappingIdentity.DecRef(t) } }() @@ -71,7 +72,7 @@ func Mmap(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // mmap unconditionally requires that the FD is readable. if !file.IsReadable() { @@ -85,6 +86,17 @@ func Mmap(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if err := file.ConfigureMMap(t, &opts); err != nil { return 0, nil, err } + } else if shared { + // Back shared anonymous mappings with an anonymous tmpfs file. + opts.Offset = 0 + file, err := tmpfs.NewZeroFile(t, t.Credentials(), t.Kernel().ShmMount(), opts.Length) + if err != nil { + return 0, nil, err + } + defer file.DecRef(t) + if err := file.ConfigureMMap(t, &opts); err != nil { + return 0, nil, err + } } rv, err := t.MemoryManager().MMap(t, opts) diff --git a/pkg/sentry/syscalls/linux/vfs2/mount.go b/pkg/sentry/syscalls/linux/vfs2/mount.go index ea337de7c..769c9b92f 100644 --- a/pkg/sentry/syscalls/linux/vfs2/mount.go +++ b/pkg/sentry/syscalls/linux/vfs2/mount.go @@ -108,9 +108,9 @@ func Mount(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if err != nil { return 0, nil, err } - defer target.Release() - - return 0, nil, t.Kernel().VFS().MountAt(t, creds, source, &target.pop, fsType, &opts) + defer target.Release(t) + _, err = t.Kernel().VFS().MountAt(t, creds, source, &target.pop, fsType, &opts) + return 0, nil, err } // Umount2 implements Linux syscall umount2(2). @@ -140,7 +140,7 @@ func Umount2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) opts := vfs.UmountOptions{ Flags: uint32(flags), diff --git a/pkg/sentry/syscalls/linux/vfs2/path.go b/pkg/sentry/syscalls/linux/vfs2/path.go index 97da6c647..90a511d9a 100644 --- a/pkg/sentry/syscalls/linux/vfs2/path.go +++ b/pkg/sentry/syscalls/linux/vfs2/path.go @@ -42,7 +42,7 @@ func getTaskPathOperation(t *kernel.Task, dirfd int32, path fspath.Path, shouldA haveStartRef := false if !path.Absolute { if !path.HasComponents() && !bool(shouldAllowEmptyPath) { - root.DecRef() + root.DecRef(t) return taskPathOperation{}, syserror.ENOENT } if dirfd == linux.AT_FDCWD { @@ -51,13 +51,13 @@ func getTaskPathOperation(t *kernel.Task, dirfd int32, path fspath.Path, shouldA } else { dirfile := t.GetFileVFS2(dirfd) if dirfile == nil { - root.DecRef() + root.DecRef(t) return taskPathOperation{}, syserror.EBADF } start = dirfile.VirtualDentry() start.IncRef() haveStartRef = true - dirfile.DecRef() + dirfile.DecRef(t) } } return taskPathOperation{ @@ -71,10 +71,10 @@ func getTaskPathOperation(t *kernel.Task, dirfd int32, path fspath.Path, shouldA }, nil } -func (tpop *taskPathOperation) Release() { - tpop.pop.Root.DecRef() +func (tpop *taskPathOperation) Release(t *kernel.Task) { + tpop.pop.Root.DecRef(t) if tpop.haveStartRef { - tpop.pop.Start.DecRef() + tpop.pop.Start.DecRef(t) tpop.haveStartRef = false } } diff --git a/pkg/sentry/syscalls/linux/vfs2/pipe.go b/pkg/sentry/syscalls/linux/vfs2/pipe.go index 4a01e4209..ee38fdca0 100644 --- a/pkg/sentry/syscalls/linux/vfs2/pipe.go +++ b/pkg/sentry/syscalls/linux/vfs2/pipe.go @@ -16,6 +16,7 @@ package vfs2 import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fsimpl/pipefs" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -42,8 +43,8 @@ func pipe2(t *kernel.Task, addr usermem.Addr, flags int32) error { return syserror.EINVAL } r, w := pipefs.NewConnectedPipeFDs(t, t.Kernel().PipeMount(), uint32(flags&linux.O_NONBLOCK)) - defer r.DecRef() - defer w.DecRef() + defer r.DecRef(t) + defer w.DecRef(t) fds, err := t.NewFDsVFS2(0, []*vfs.FileDescription{r, w}, kernel.FDFlags{ CloseOnExec: flags&linux.O_CLOEXEC != 0, @@ -51,10 +52,10 @@ func pipe2(t *kernel.Task, addr usermem.Addr, flags int32) error { if err != nil { return err } - if _, err := t.CopyOut(addr, fds); err != nil { + if _, err := primitive.CopyInt32SliceOut(t, addr, fds); err != nil { for _, fd := range fds { - if _, file := t.FDTable().Remove(fd); file != nil { - file.DecRef() + if _, file := t.FDTable().Remove(t, fd); file != nil { + file.DecRef(t) } } return err diff --git a/pkg/sentry/syscalls/linux/vfs2/poll.go b/pkg/sentry/syscalls/linux/vfs2/poll.go index ff1b25d7b..c22e4ce54 100644 --- a/pkg/sentry/syscalls/linux/vfs2/poll.go +++ b/pkg/sentry/syscalls/linux/vfs2/poll.go @@ -73,7 +73,7 @@ func initReadiness(t *kernel.Task, pfd *linux.PollFD, state *pollState, ch chan } if ch == nil { - defer file.DecRef() + defer file.DecRef(t) } else { state.file = file state.waiter, _ = waiter.NewChannelEntry(ch) @@ -85,11 +85,11 @@ func initReadiness(t *kernel.Task, pfd *linux.PollFD, state *pollState, ch chan } // releaseState releases all the pollState in "state". -func releaseState(state []pollState) { +func releaseState(t *kernel.Task, state []pollState) { for i := range state { if state[i].file != nil { state[i].file.EventUnregister(&state[i].waiter) - state[i].file.DecRef() + state[i].file.DecRef(t) } } } @@ -110,7 +110,7 @@ func pollBlock(t *kernel.Task, pfd []linux.PollFD, timeout time.Duration) (time. // result, we stop registering for events but still go through all files // to get their ready masks. state := make([]pollState, len(pfd)) - defer releaseState(state) + defer releaseState(t, state) n := uintptr(0) for i := range pfd { initReadiness(t, &pfd[i], &state[i], ch) @@ -165,7 +165,7 @@ func copyInPollFDs(t *kernel.Task, addr usermem.Addr, nfds uint) ([]linux.PollFD pfd := make([]linux.PollFD, nfds) if nfds > 0 { - if _, err := t.CopyIn(addr, &pfd); err != nil { + if _, err := linux.CopyPollFDSliceIn(t, addr, pfd); err != nil { return nil, err } } @@ -192,7 +192,7 @@ func doPoll(t *kernel.Task, addr usermem.Addr, nfds uint, timeout time.Duration) // The poll entries are copied out regardless of whether // any are set or not. This aligns with the Linux behavior. if nfds > 0 && err == nil { - if _, err := t.CopyOut(addr, pfd); err != nil { + if _, err := linux.CopyPollFDSliceOut(t, addr, pfd); err != nil { return remainingTimeout, 0, err } } @@ -205,7 +205,7 @@ func CopyInFDSet(t *kernel.Task, addr usermem.Addr, nBytes, nBitsInLastPartialBy set := make([]byte, nBytes) if addr != 0 { - if _, err := t.CopyIn(addr, &set); err != nil { + if _, err := t.CopyInBytes(addr, set); err != nil { return nil, err } // If we only use part of the last byte, mask out the extraneous bits. @@ -269,7 +269,7 @@ func doSelect(t *kernel.Task, nfds int, readFDs, writeFDs, exceptFDs usermem.Add if file == nil { return 0, syserror.EBADF } - file.DecRef() + file.DecRef(t) var mask int16 if (rV & m) != 0 { @@ -332,19 +332,19 @@ func doSelect(t *kernel.Task, nfds int, readFDs, writeFDs, exceptFDs usermem.Add // Copy updated vectors back. if readFDs != 0 { - if _, err := t.CopyOut(readFDs, r); err != nil { + if _, err := t.CopyOutBytes(readFDs, r); err != nil { return 0, err } } if writeFDs != 0 { - if _, err := t.CopyOut(writeFDs, w); err != nil { + if _, err := t.CopyOutBytes(writeFDs, w); err != nil { return 0, err } } if exceptFDs != 0 { - if _, err := t.CopyOut(exceptFDs, e); err != nil { + if _, err := t.CopyOutBytes(exceptFDs, e); err != nil { return 0, err } } @@ -415,7 +415,7 @@ func poll(t *kernel.Task, pfdAddr usermem.Addr, nfds uint, timeout time.Duration nfds: nfds, timeout: remainingTimeout, }) - return 0, kernel.ERESTART_RESTARTBLOCK + return 0, syserror.ERESTART_RESTARTBLOCK } return n, err } @@ -462,7 +462,7 @@ func Ppoll(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // Note that this means that if err is nil but copyErr is not, copyErr is // ignored. This is consistent with Linux. if err == syserror.EINTR && copyErr == nil { - err = kernel.ERESTARTNOHAND + err = syserror.ERESTARTNOHAND } return n, nil, err } @@ -492,11 +492,17 @@ func Select(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal copyErr := copyOutTimevalRemaining(t, startNs, timeout, timevalAddr) // See comment in Ppoll. if err == syserror.EINTR && copyErr == nil { - err = kernel.ERESTARTNOHAND + err = syserror.ERESTARTNOHAND } return n, nil, err } +// +marshal +type sigSetWithSize struct { + sigsetAddr uint64 + sizeofSigset uint64 +} + // Pselect implements linux syscall pselect(2). func Pselect(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { nfds := int(args[0].Int()) // select(2) uses an int. @@ -533,17 +539,11 @@ func Pselect(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca copyErr := copyOutTimespecRemaining(t, startNs, timeout, timespecAddr) // See comment in Ppoll. if err == syserror.EINTR && copyErr == nil { - err = kernel.ERESTARTNOHAND + err = syserror.ERESTARTNOHAND } return n, nil, err } -// +marshal -type sigSetWithSize struct { - sigsetAddr uint64 - sizeofSigset uint64 -} - // copyTimespecInToDuration copies a Timespec from the untrusted app range, // validates it and converts it to a Duration. // diff --git a/pkg/sentry/syscalls/linux/vfs2/read_write.go b/pkg/sentry/syscalls/linux/vfs2/read_write.go index cd25597a7..b77b29dcc 100644 --- a/pkg/sentry/syscalls/linux/vfs2/read_write.go +++ b/pkg/sentry/syscalls/linux/vfs2/read_write.go @@ -44,7 +44,7 @@ func Read(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the size is legitimate. si := int(size) @@ -62,7 +62,7 @@ func Read(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC n, err := read(t, file, dst, vfs.ReadOptions{}) t.IOUsage().AccountReadSyscall(n) - return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, kernel.ERESTARTSYS, "read", file) + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, syserror.ERESTARTSYS, "read", file) } // Readv implements Linux syscall readv(2). @@ -75,7 +75,7 @@ func Readv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Get the destination of the read. dst, err := t.IovecsIOSequence(addr, iovcnt, usermem.IOOpts{ @@ -87,14 +87,14 @@ func Readv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall n, err := read(t, file, dst, vfs.ReadOptions{}) t.IOUsage().AccountReadSyscall(n) - return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, kernel.ERESTARTSYS, "readv", file) + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, syserror.ERESTARTSYS, "readv", file) } func read(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { n, err := file.Read(t, dst, opts) if err != syserror.ErrWouldBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) } return n, err } @@ -102,7 +102,7 @@ func read(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, opt allowBlock, deadline, hasDeadline := blockPolicy(t, file) if !allowBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) } return n, err } @@ -135,7 +135,7 @@ func read(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, opt file.EventUnregister(&w) if total > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) } return total, err } @@ -151,7 +151,7 @@ func Pread64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate and does not overflow. if offset < 0 || offset+int64(size) < 0 { @@ -174,7 +174,7 @@ func Pread64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca n, err := pread(t, file, dst, offset, vfs.ReadOptions{}) t.IOUsage().AccountReadSyscall(n) - return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, kernel.ERESTARTSYS, "pread64", file) + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, syserror.ERESTARTSYS, "pread64", file) } // Preadv implements Linux syscall preadv(2). @@ -188,7 +188,7 @@ func Preadv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate. if offset < 0 { @@ -205,7 +205,7 @@ func Preadv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal n, err := pread(t, file, dst, offset, vfs.ReadOptions{}) t.IOUsage().AccountReadSyscall(n) - return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, kernel.ERESTARTSYS, "preadv", file) + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, syserror.ERESTARTSYS, "preadv", file) } // Preadv2 implements Linux syscall preadv2(2). @@ -226,7 +226,7 @@ func Preadv2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate. if offset < -1 { @@ -251,14 +251,14 @@ func Preadv2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca n, err = pread(t, file, dst, offset, opts) } t.IOUsage().AccountReadSyscall(n) - return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, kernel.ERESTARTSYS, "preadv2", file) + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, syserror.ERESTARTSYS, "preadv2", file) } func pread(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { n, err := file.PRead(t, dst, offset, opts) if err != syserror.ErrWouldBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) } return n, err } @@ -266,7 +266,7 @@ func pread(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, of allowBlock, deadline, hasDeadline := blockPolicy(t, file) if !allowBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) } return n, err } @@ -299,7 +299,7 @@ func pread(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, of file.EventUnregister(&w) if total > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) } return total, err } @@ -314,7 +314,7 @@ func Write(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the size is legitimate. si := int(size) @@ -332,7 +332,7 @@ func Write(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall n, err := write(t, file, src, vfs.WriteOptions{}) t.IOUsage().AccountWriteSyscall(n) - return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, kernel.ERESTARTSYS, "write", file) + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, syserror.ERESTARTSYS, "write", file) } // Writev implements Linux syscall writev(2). @@ -345,7 +345,7 @@ func Writev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Get the source of the write. src, err := t.IovecsIOSequence(addr, iovcnt, usermem.IOOpts{ @@ -357,14 +357,14 @@ func Writev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal n, err := write(t, file, src, vfs.WriteOptions{}) t.IOUsage().AccountWriteSyscall(n) - return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, kernel.ERESTARTSYS, "writev", file) + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, syserror.ERESTARTSYS, "writev", file) } func write(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { n, err := file.Write(t, src, opts) if err != syserror.ErrWouldBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) } return n, err } @@ -372,7 +372,7 @@ func write(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, op allowBlock, deadline, hasDeadline := blockPolicy(t, file) if !allowBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) } return n, err } @@ -405,7 +405,7 @@ func write(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, op file.EventUnregister(&w) if total > 0 { - file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) } return total, err } @@ -421,7 +421,7 @@ func Pwrite64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate and does not overflow. if offset < 0 || offset+int64(size) < 0 { @@ -444,7 +444,7 @@ func Pwrite64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc n, err := pwrite(t, file, src, offset, vfs.WriteOptions{}) t.IOUsage().AccountWriteSyscall(n) - return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, kernel.ERESTARTSYS, "pwrite64", file) + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, syserror.ERESTARTSYS, "pwrite64", file) } // Pwritev implements Linux syscall pwritev(2). @@ -458,7 +458,7 @@ func Pwritev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate. if offset < 0 { @@ -475,7 +475,7 @@ func Pwritev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca n, err := pwrite(t, file, src, offset, vfs.WriteOptions{}) t.IOUsage().AccountReadSyscall(n) - return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, kernel.ERESTARTSYS, "pwritev", file) + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, syserror.ERESTARTSYS, "pwritev", file) } // Pwritev2 implements Linux syscall pwritev2(2). @@ -496,7 +496,7 @@ func Pwritev2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate. if offset < -1 { @@ -521,14 +521,14 @@ func Pwritev2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc n, err = pwrite(t, file, src, offset, opts) } t.IOUsage().AccountWriteSyscall(n) - return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, kernel.ERESTARTSYS, "pwritev2", file) + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, syserror.ERESTARTSYS, "pwritev2", file) } func pwrite(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { n, err := file.PWrite(t, src, offset, opts) if err != syserror.ErrWouldBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) } return n, err } @@ -536,7 +536,7 @@ func pwrite(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, o allowBlock, deadline, hasDeadline := blockPolicy(t, file) if !allowBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) } return n, err } @@ -569,7 +569,7 @@ func pwrite(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, o file.EventUnregister(&w) if total > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) } return total, err } @@ -601,7 +601,7 @@ func Lseek(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) newoff, err := file.Seek(t, offset, whence) return uintptr(newoff), nil, err @@ -617,7 +617,7 @@ func Readahead(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the file is readable. if !file.IsReadable() { diff --git a/pkg/sentry/syscalls/linux/vfs2/setstat.go b/pkg/sentry/syscalls/linux/vfs2/setstat.go index 37fa56c19..1ee37e5a8 100644 --- a/pkg/sentry/syscalls/linux/vfs2/setstat.go +++ b/pkg/sentry/syscalls/linux/vfs2/setstat.go @@ -66,7 +66,7 @@ func Fchmod(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return 0, nil, file.SetStat(t, vfs.SetStatOptions{ Stat: linux.Statx{ @@ -151,7 +151,7 @@ func Fchown(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) var opts vfs.SetStatOptions if err := populateSetStatOptionsForChown(t, owner, group, &opts); err != nil { @@ -197,7 +197,7 @@ func Ftruncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) if !file.IsWritable() { return 0, nil, syserror.EINVAL @@ -224,7 +224,7 @@ func Fallocate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) if !file.IsWritable() { return 0, nil, syserror.EBADF @@ -254,11 +254,12 @@ func Fallocate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys return 0, nil, syserror.EFBIG } - return 0, nil, file.Allocate(t, mode, uint64(offset), uint64(length)) + if err := file.Allocate(t, mode, uint64(offset), uint64(length)); err != nil { + return 0, nil, err + } - // File length modified, generate notification. - // TODO(gvisor.dev/issue/1479): Reenable when Inotify is ported. - // file.Dirent.InotifyEvent(linux.IN_MODIFY, 0) + file.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) + return 0, nil, nil } // Utime implements Linux syscall utime(2). @@ -345,7 +346,7 @@ func populateSetStatOptionsForUtimes(t *kernel.Task, timesAddr usermem.Addr, opt return nil } var times [2]linux.Timeval - if _, err := t.CopyIn(timesAddr, ×); err != nil { + if _, err := linux.CopyTimevalSliceIn(t, timesAddr, times[:]); err != nil { return err } if times[0].Usec < 0 || times[0].Usec > 999999 || times[1].Usec < 0 || times[1].Usec > 999999 { @@ -409,7 +410,7 @@ func populateSetStatOptionsForUtimens(t *kernel.Task, timesAddr usermem.Addr, op return nil } var times [2]linux.Timespec - if _, err := t.CopyIn(timesAddr, ×); err != nil { + if _, err := linux.CopyTimespecSliceIn(t, timesAddr, times[:]); err != nil { return err } if times[0].Nsec != linux.UTIME_OMIT { @@ -437,7 +438,7 @@ func populateSetStatOptionsForUtimens(t *kernel.Task, timesAddr usermem.Addr, op func setstatat(t *kernel.Task, dirfd int32, path fspath.Path, shouldAllowEmptyPath shouldAllowEmptyPath, shouldFollowFinalSymlink shouldFollowFinalSymlink, opts *vfs.SetStatOptions) error { root := t.FSContext().RootDirectoryVFS2() - defer root.DecRef() + defer root.DecRef(t) start := root if !path.Absolute { if !path.HasComponents() && !bool(shouldAllowEmptyPath) { @@ -445,7 +446,7 @@ func setstatat(t *kernel.Task, dirfd int32, path fspath.Path, shouldAllowEmptyPa } if dirfd == linux.AT_FDCWD { start = t.FSContext().WorkingDirectoryVFS2() - defer start.DecRef() + defer start.DecRef(t) } else { dirfile := t.GetFileVFS2(dirfd) if dirfile == nil { @@ -456,13 +457,13 @@ func setstatat(t *kernel.Task, dirfd int32, path fspath.Path, shouldAllowEmptyPa // VirtualFilesystem.SetStatAt(), since the former may be able // to use opened file state to expedite the SetStat. err := dirfile.SetStat(t, *opts) - dirfile.DecRef() + dirfile.DecRef(t) return err } start = dirfile.VirtualDentry() start.IncRef() - defer start.DecRef() - dirfile.DecRef() + defer start.DecRef(t) + dirfile.DecRef(t) } } return t.Kernel().VFS().SetStatAt(t, t.Credentials(), &vfs.PathOperation{ diff --git a/pkg/sentry/syscalls/linux/vfs2/signal.go b/pkg/sentry/syscalls/linux/vfs2/signal.go index 623992f6f..b89f34cdb 100644 --- a/pkg/sentry/syscalls/linux/vfs2/signal.go +++ b/pkg/sentry/syscalls/linux/vfs2/signal.go @@ -45,7 +45,7 @@ func sharedSignalfd(t *kernel.Task, fd int32, sigset usermem.Addr, sigsetsize ui if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Is this a signalfd? if sfd, ok := file.Impl().(*signalfd.SignalFileDescription); ok { @@ -68,7 +68,7 @@ func sharedSignalfd(t *kernel.Task, fd int32, sigset usermem.Addr, sigsetsize ui if err != nil { return 0, nil, err } - defer file.DecRef() + defer file.DecRef(t) // Create a new descriptor. fd, err = t.NewFDFromVFS2(0, file, kernel.FDFlags{ diff --git a/pkg/sentry/syscalls/linux/vfs2/socket.go b/pkg/sentry/syscalls/linux/vfs2/socket.go index 8096a8f9c..bfae6b7e9 100644 --- a/pkg/sentry/syscalls/linux/vfs2/socket.go +++ b/pkg/sentry/syscalls/linux/vfs2/socket.go @@ -19,6 +19,8 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/marshal" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" @@ -30,8 +32,6 @@ import ( "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" - "gvisor.dev/gvisor/tools/go_marshal/marshal" - "gvisor.dev/gvisor/tools/go_marshal/primitive" ) // minListenBacklog is the minimum reasonable backlog for listening sockets. @@ -66,10 +66,10 @@ const flagsOffset = 48 const sizeOfInt32 = 4 // messageHeader64Len is the length of a MessageHeader64 struct. -var messageHeader64Len = uint64(binary.Size(MessageHeader64{})) +var messageHeader64Len = uint64((*MessageHeader64)(nil).SizeBytes()) // multipleMessageHeader64Len is the length of a multipeMessageHeader64 struct. -var multipleMessageHeader64Len = uint64(binary.Size(multipleMessageHeader64{})) +var multipleMessageHeader64Len = uint64((*multipleMessageHeader64)(nil).SizeBytes()) // baseRecvFlags are the flags that are accepted across recvmsg(2), // recvmmsg(2), and recvfrom(2). @@ -77,6 +77,8 @@ const baseRecvFlags = linux.MSG_OOB | linux.MSG_DONTROUTE | linux.MSG_DONTWAIT | // MessageHeader64 is the 64-bit representation of the msghdr struct used in // the recvmsg and sendmsg syscalls. +// +// +marshal type MessageHeader64 struct { // Name is the optional pointer to a network address buffer. Name uint64 @@ -105,30 +107,14 @@ type MessageHeader64 struct { // multipleMessageHeader64 is the 64-bit representation of the mmsghdr struct used in // the recvmmsg and sendmmsg syscalls. +// +// +marshal type multipleMessageHeader64 struct { msgHdr MessageHeader64 msgLen uint32 _ int32 } -// CopyInMessageHeader64 copies a message header from user to kernel memory. -func CopyInMessageHeader64(t *kernel.Task, addr usermem.Addr, msg *MessageHeader64) error { - b := t.CopyScratchBuffer(52) - if _, err := t.CopyInBytes(addr, b); err != nil { - return err - } - - msg.Name = usermem.ByteOrder.Uint64(b[0:]) - msg.NameLen = usermem.ByteOrder.Uint32(b[8:]) - msg.Iov = usermem.ByteOrder.Uint64(b[16:]) - msg.IovLen = usermem.ByteOrder.Uint64(b[24:]) - msg.Control = usermem.ByteOrder.Uint64(b[32:]) - msg.ControlLen = usermem.ByteOrder.Uint64(b[40:]) - msg.Flags = int32(usermem.ByteOrder.Uint32(b[48:])) - - return nil -} - // CaptureAddress allocates memory for and copies a socket address structure // from the untrusted address space range. func CaptureAddress(t *kernel.Task, addr usermem.Addr, addrlen uint32) ([]byte, error) { @@ -147,10 +133,10 @@ func CaptureAddress(t *kernel.Task, addr usermem.Addr, addrlen uint32) ([]byte, // writeAddress writes a sockaddr structure and its length to an output buffer // in the unstrusted address space range. If the address is bigger than the // buffer, it is truncated. -func writeAddress(t *kernel.Task, addr interface{}, addrLen uint32, addrPtr usermem.Addr, addrLenPtr usermem.Addr) error { +func writeAddress(t *kernel.Task, addr linux.SockAddr, addrLen uint32, addrPtr usermem.Addr, addrLenPtr usermem.Addr) error { // Get the buffer length. var bufLen uint32 - if _, err := t.CopyIn(addrLenPtr, &bufLen); err != nil { + if _, err := primitive.CopyUint32In(t, addrLenPtr, &bufLen); err != nil { return err } @@ -159,7 +145,7 @@ func writeAddress(t *kernel.Task, addr interface{}, addrLen uint32, addrPtr user } // Write the length unconditionally. - if _, err := t.CopyOut(addrLenPtr, addrLen); err != nil { + if _, err := primitive.CopyUint32Out(t, addrLenPtr, addrLen); err != nil { return err } @@ -172,7 +158,8 @@ func writeAddress(t *kernel.Task, addr interface{}, addrLen uint32, addrPtr user } // Copy as much of the address as will fit in the buffer. - encodedAddr := binary.Marshal(nil, usermem.ByteOrder, addr) + encodedAddr := t.CopyScratchBuffer(addr.SizeBytes()) + addr.MarshalUnsafe(encodedAddr) if bufLen > uint32(len(encodedAddr)) { bufLen = uint32(len(encodedAddr)) } @@ -196,7 +183,7 @@ func Socket(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if e != nil { return 0, nil, e.ToError() } - defer s.DecRef() + defer s.DecRef(t) if err := s.SetStatusFlags(t, t.Credentials(), uint32(stype&linux.SOCK_NONBLOCK)); err != nil { return 0, nil, err @@ -230,8 +217,8 @@ func SocketPair(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy return 0, nil, e.ToError() } // Adding to the FD table will cause an extra reference to be acquired. - defer s1.DecRef() - defer s2.DecRef() + defer s1.DecRef(t) + defer s2.DecRef(t) nonblocking := uint32(stype & linux.SOCK_NONBLOCK) if err := s1.SetStatusFlags(t, t.Credentials(), nonblocking); err != nil { @@ -250,10 +237,10 @@ func SocketPair(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy return 0, nil, err } - if _, err := t.CopyOut(addr, fds); err != nil { + if _, err := primitive.CopyInt32SliceOut(t, addr, fds); err != nil { for _, fd := range fds { - if _, file := t.FDTable().Remove(fd); file != nil { - file.DecRef() + if _, file := t.FDTable().Remove(t, fd); file != nil { + file.DecRef(t) } } return 0, nil, err @@ -273,7 +260,7 @@ func Connect(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -288,7 +275,7 @@ func Connect(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca } blocking := (file.StatusFlags() & linux.SOCK_NONBLOCK) == 0 - return 0, nil, syserror.ConvertIntr(s.Connect(t, a, blocking).ToError(), kernel.ERESTARTSYS) + return 0, nil, syserror.ConvertIntr(s.Connect(t, a, blocking).ToError(), syserror.ERESTARTSYS) } // accept is the implementation of the accept syscall. It is called by accept @@ -304,7 +291,7 @@ func accept(t *kernel.Task, fd int32, addr usermem.Addr, addrLen usermem.Addr, f if file == nil { return 0, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -319,7 +306,7 @@ func accept(t *kernel.Task, fd int32, addr usermem.Addr, addrLen usermem.Addr, f peerRequested := addrLen != 0 nfd, peer, peerLen, e := s.Accept(t, peerRequested, flags, blocking) if e != nil { - return 0, syserror.ConvertIntr(e.ToError(), kernel.ERESTARTSYS) + return 0, syserror.ConvertIntr(e.ToError(), syserror.ERESTARTSYS) } if peerRequested { // NOTE(magi): Linux does not give you an error if it can't @@ -363,7 +350,7 @@ func Bind(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -390,7 +377,7 @@ func Listen(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -419,7 +406,7 @@ func Shutdown(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -450,7 +437,7 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -459,8 +446,8 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy } // Read the length. Reject negative values. - optLen := int32(0) - if _, err := t.CopyIn(optLenAddr, &optLen); err != nil { + var optLen int32 + if _, err := primitive.CopyInt32In(t, optLenAddr, &optLen); err != nil { return 0, nil, err } if optLen < 0 { @@ -474,7 +461,7 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy } vLen := int32(binary.Size(v)) - if _, err := t.CopyOut(optLenAddr, vLen); err != nil { + if _, err := primitive.CopyInt32Out(t, optLenAddr, vLen); err != nil { return 0, nil, err } @@ -532,7 +519,7 @@ func SetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -570,7 +557,7 @@ func GetSockName(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -598,7 +585,7 @@ func GetPeerName(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -631,7 +618,7 @@ func RecvMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -684,7 +671,7 @@ func RecvMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -736,7 +723,7 @@ func RecvMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if !ok { return 0, nil, syserror.EFAULT } - if _, err = t.CopyOut(lp, uint32(n)); err != nil { + if _, err = primitive.CopyUint32Out(t, lp, uint32(n)); err != nil { break } count++ @@ -751,7 +738,7 @@ func RecvMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc func recvSingleMsg(t *kernel.Task, s socket.SocketVFS2, msgPtr usermem.Addr, flags int32, haveDeadline bool, deadline ktime.Time) (uintptr, error) { // Capture the message header and io vectors. var msg MessageHeader64 - if err := CopyInMessageHeader64(t, msgPtr, &msg); err != nil { + if _, err := msg.CopyIn(t, msgPtr); err != nil { return 0, err } @@ -774,16 +761,16 @@ func recvSingleMsg(t *kernel.Task, s socket.SocketVFS2, msgPtr usermem.Addr, fla if msg.ControlLen == 0 && msg.NameLen == 0 { n, mflags, _, _, cms, err := s.RecvMsg(t, dst, int(flags), haveDeadline, deadline, false, 0) if err != nil { - return 0, syserror.ConvertIntr(err.ToError(), kernel.ERESTARTSYS) + return 0, syserror.ConvertIntr(err.ToError(), syserror.ERESTARTSYS) } if !cms.Unix.Empty() { mflags |= linux.MSG_CTRUNC - cms.Release() + cms.Release(t) } if int(msg.Flags) != mflags { // Copy out the flags to the caller. - if _, err := t.CopyOut(msgPtr+flagsOffset, int32(mflags)); err != nil { + if _, err := primitive.CopyInt32Out(t, msgPtr+flagsOffset, int32(mflags)); err != nil { return 0, err } } @@ -796,9 +783,9 @@ func recvSingleMsg(t *kernel.Task, s socket.SocketVFS2, msgPtr usermem.Addr, fla } n, mflags, sender, senderLen, cms, e := s.RecvMsg(t, dst, int(flags), haveDeadline, deadline, msg.NameLen != 0, msg.ControlLen) if e != nil { - return 0, syserror.ConvertIntr(e.ToError(), kernel.ERESTARTSYS) + return 0, syserror.ConvertIntr(e.ToError(), syserror.ERESTARTSYS) } - defer cms.Release() + defer cms.Release(t) controlData := make([]byte, 0, msg.ControlLen) controlData = control.PackControlMessages(t, cms, controlData) @@ -820,17 +807,17 @@ func recvSingleMsg(t *kernel.Task, s socket.SocketVFS2, msgPtr usermem.Addr, fla } // Copy the control data to the caller. - if _, err := t.CopyOut(msgPtr+controlLenOffset, uint64(len(controlData))); err != nil { + if _, err := primitive.CopyUint64Out(t, msgPtr+controlLenOffset, uint64(len(controlData))); err != nil { return 0, err } if len(controlData) > 0 { - if _, err := t.CopyOut(usermem.Addr(msg.Control), controlData); err != nil { + if _, err := t.CopyOutBytes(usermem.Addr(msg.Control), controlData); err != nil { return 0, err } } // Copy out the flags to the caller. - if _, err := t.CopyOut(msgPtr+flagsOffset, int32(mflags)); err != nil { + if _, err := primitive.CopyInt32Out(t, msgPtr+flagsOffset, int32(mflags)); err != nil { return 0, err } @@ -854,7 +841,7 @@ func recvFrom(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flag if file == nil { return 0, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -883,9 +870,9 @@ func recvFrom(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flag } n, _, sender, senderLen, cm, e := s.RecvMsg(t, dst, int(flags), haveDeadline, deadline, nameLenPtr != 0, 0) - cm.Release() + cm.Release(t) if e != nil { - return 0, syserror.ConvertIntr(e.ToError(), kernel.ERESTARTSYS) + return 0, syserror.ConvertIntr(e.ToError(), syserror.ERESTARTSYS) } // Copy the address to the caller. @@ -927,7 +914,7 @@ func SendMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -965,7 +952,7 @@ func SendMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -999,7 +986,7 @@ func SendMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if !ok { return 0, nil, syserror.EFAULT } - if _, err = t.CopyOut(lp, uint32(n)); err != nil { + if _, err = primitive.CopyUint32Out(t, lp, uint32(n)); err != nil { break } count++ @@ -1014,7 +1001,7 @@ func SendMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc func sendSingleMsg(t *kernel.Task, s socket.SocketVFS2, file *vfs.FileDescription, msgPtr usermem.Addr, flags int32) (uintptr, error) { // Capture the message header. var msg MessageHeader64 - if err := CopyInMessageHeader64(t, msgPtr, &msg); err != nil { + if _, err := msg.CopyIn(t, msgPtr); err != nil { return 0, err } @@ -1025,7 +1012,7 @@ func sendSingleMsg(t *kernel.Task, s socket.SocketVFS2, file *vfs.FileDescriptio return 0, syserror.ENOBUFS } controlData = make([]byte, msg.ControlLen) - if _, err := t.CopyIn(usermem.Addr(msg.Control), &controlData); err != nil { + if _, err := t.CopyInBytes(usermem.Addr(msg.Control), controlData); err != nil { return 0, err } } @@ -1067,9 +1054,9 @@ func sendSingleMsg(t *kernel.Task, s socket.SocketVFS2, file *vfs.FileDescriptio // Call the syscall implementation. n, e := s.SendMsg(t, src, to, int(flags), haveDeadline, deadline, controlMessages) - err = slinux.HandleIOErrorVFS2(t, n != 0, e.ToError(), kernel.ERESTARTSYS, "sendmsg", file) + err = slinux.HandleIOErrorVFS2(t, n != 0, e.ToError(), syserror.ERESTARTSYS, "sendmsg", file) if err != nil { - controlMessages.Release() + controlMessages.Release(t) } return uintptr(n), err } @@ -1087,7 +1074,7 @@ func sendTo(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flags if file == nil { return 0, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -1127,7 +1114,7 @@ func sendTo(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flags // Call the syscall implementation. n, e := s.SendMsg(t, src, to, int(flags), haveDeadline, deadline, socket.ControlMessages{Unix: control.New(t, s, nil)}) - return uintptr(n), slinux.HandleIOErrorVFS2(t, n != 0, e.ToError(), kernel.ERESTARTSYS, "sendto", file) + return uintptr(n), slinux.HandleIOErrorVFS2(t, n != 0, e.ToError(), syserror.ERESTARTSYS, "sendto", file) } // SendTo implements the linux syscall sendto(2). diff --git a/pkg/sentry/syscalls/linux/vfs2/splice.go b/pkg/sentry/syscalls/linux/vfs2/splice.go index 63ab11f8c..bf5c1171f 100644 --- a/pkg/sentry/syscalls/linux/vfs2/splice.go +++ b/pkg/sentry/syscalls/linux/vfs2/splice.go @@ -18,9 +18,12 @@ import ( "io" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" + slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -53,12 +56,12 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if inFile == nil { return 0, nil, syserror.EBADF } - defer inFile.DecRef() + defer inFile.DecRef(t) outFile := t.GetFileVFS2(outFD) if outFile == nil { return 0, nil, syserror.EBADF } - defer outFile.DecRef() + defer outFile.DecRef(t) // Check that both files support the required directionality. if !inFile.IsReadable() || !outFile.IsWritable() { @@ -88,7 +91,7 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if inFile.Options().DenyPRead { return 0, nil, syserror.EINVAL } - if _, err := t.CopyIn(inOffsetPtr, &inOffset); err != nil { + if _, err := primitive.CopyInt64In(t, inOffsetPtr, &inOffset); err != nil { return 0, nil, err } if inOffset < 0 { @@ -103,7 +106,7 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if outFile.Options().DenyPWrite { return 0, nil, syserror.EINVAL } - if _, err := t.CopyIn(outOffsetPtr, &outOffset); err != nil { + if _, err := primitive.CopyInt64In(t, outOffsetPtr, &outOffset); err != nil { return 0, nil, err } if outOffset < 0 { @@ -131,21 +134,17 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal case inIsPipe && outIsPipe: n, err = pipe.Splice(t, outPipeFD, inPipeFD, count) case inIsPipe: + n, err = inPipeFD.SpliceToNonPipe(t, outFile, outOffset, count) if outOffset != -1 { - n, err = outFile.PWrite(t, inPipeFD.IOSequence(count), outOffset, vfs.WriteOptions{}) outOffset += n - } else { - n, err = outFile.Write(t, inPipeFD.IOSequence(count), vfs.WriteOptions{}) } case outIsPipe: + n, err = outPipeFD.SpliceFromNonPipe(t, inFile, inOffset, count) if inOffset != -1 { - n, err = inFile.PRead(t, outPipeFD.IOSequence(count), inOffset, vfs.ReadOptions{}) inOffset += n - } else { - n, err = inFile.Read(t, outPipeFD.IOSequence(count), vfs.ReadOptions{}) } default: - panic("not possible") + panic("at least one end of splice must be a pipe") } if n != 0 || err != syserror.ErrWouldBlock || nonBlock { @@ -158,25 +157,26 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // Copy updated offsets out. if inOffsetPtr != 0 { - if _, err := t.CopyOut(inOffsetPtr, &inOffset); err != nil { + if _, err := primitive.CopyInt64Out(t, inOffsetPtr, inOffset); err != nil { return 0, nil, err } } if outOffsetPtr != 0 { - if _, err := t.CopyOut(outOffsetPtr, &outOffset); err != nil { + if _, err := primitive.CopyInt64Out(t, outOffsetPtr, outOffset); err != nil { return 0, nil, err } } - if n == 0 { - return 0, nil, err + if n != 0 { + // On Linux, inotify behavior is not very consistent with splice(2). We try + // our best to emulate Linux for very basic calls to splice, where for some + // reason, events are generated for output files, but not input files. + outFile.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) } - // On Linux, inotify behavior is not very consistent with splice(2). We try - // our best to emulate Linux for very basic calls to splice, where for some - // reason, events are generated for output files, but not input files. - outFile.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) - return uintptr(n), nil, nil + // We can only pass a single file to handleIOError, so pick inFile arbitrarily. + // This is used only for debugging purposes. + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, syserror.ERESTARTSYS, "splice", outFile) } // Tee implements Linux syscall tee(2). @@ -203,12 +203,12 @@ func Tee(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo if inFile == nil { return 0, nil, syserror.EBADF } - defer inFile.DecRef() + defer inFile.DecRef(t) outFile := t.GetFileVFS2(outFD) if outFile == nil { return 0, nil, syserror.EBADF } - defer outFile.DecRef() + defer outFile.DecRef(t) // Check that both files support the required directionality. if !inFile.IsReadable() || !outFile.IsWritable() { @@ -248,11 +248,20 @@ func Tee(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo break } } - if n == 0 { - return 0, nil, err + + if n != 0 { + outFile.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) + + // If a partial write is completed, the error is dropped. Log it here. + if err != nil && err != io.EOF && err != syserror.ErrWouldBlock { + log.Debugf("tee completed a partial write with error: %v", err) + err = nil + } } - outFile.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) - return uintptr(n), nil, nil + + // We can only pass a single file to handleIOError, so pick inFile arbitrarily. + // This is used only for debugging purposes. + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, syserror.ERESTARTSYS, "tee", inFile) } // Sendfile implements linux system call sendfile(2). @@ -266,7 +275,7 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if inFile == nil { return 0, nil, syserror.EBADF } - defer inFile.DecRef() + defer inFile.DecRef(t) if !inFile.IsReadable() { return 0, nil, syserror.EBADF } @@ -275,7 +284,7 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if outFile == nil { return 0, nil, syserror.EBADF } - defer outFile.DecRef() + defer outFile.DecRef(t) if !outFile.IsWritable() { return 0, nil, syserror.EBADF } @@ -301,9 +310,12 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if inFile.Options().DenyPRead { return 0, nil, syserror.ESPIPE } - if _, err := t.CopyIn(offsetAddr, &offset); err != nil { + var offsetP primitive.Int64 + if _, err := offsetP.CopyIn(t, offsetAddr); err != nil { return 0, nil, err } + offset = int64(offsetP) + if offset < 0 { return 0, nil, syserror.EINVAL } @@ -341,11 +353,9 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if outIsPipe { for n < count { var spliceN int64 + spliceN, err = outPipeFD.SpliceFromNonPipe(t, inFile, offset, count) if offset != -1 { - spliceN, err = inFile.PRead(t, outPipeFD.IOSequence(count), offset, vfs.ReadOptions{}) offset += spliceN - } else { - spliceN, err = inFile.Read(t, outPipeFD.IOSequence(count), vfs.ReadOptions{}) } n += spliceN if err == syserror.ErrWouldBlock && !nonBlock { @@ -366,20 +376,11 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc } else { readN, err = inFile.Read(t, usermem.BytesIOSequence(buf), vfs.ReadOptions{}) } - if readN == 0 && err == io.EOF { - // We reached the end of the file. Eat the - // error and exit the loop. - err = nil - break - } n += readN - if err != nil { - break - } // Write all of the bytes that we read. This may need // multiple write calls to complete. - wbuf := buf[:n] + wbuf := buf[:readN] for len(wbuf) > 0 { var writeN int64 writeN, err = outFile.Write(t, usermem.BytesIOSequence(wbuf), vfs.WriteOptions{}) @@ -388,12 +389,21 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc err = dw.waitForOut(t) } if err != nil { - // We didn't complete the write. Only - // report the bytes that were actually - // written, and rewind the offset. + // We didn't complete the write. Only report the bytes that were actually + // written, and rewind offsets as needed. notWritten := int64(len(wbuf)) n -= notWritten - if offset != -1 { + if offset == -1 { + // We modified the offset of the input file itself during the read + // operation. Rewind it. + if _, seekErr := inFile.Seek(t, -notWritten, linux.SEEK_CUR); seekErr != nil { + // Log the error but don't return it, since the write has already + // completed successfully. + log.Warningf("failed to roll back input file offset: %v", seekErr) + } + } else { + // The sendfile call was provided an offset parameter that should be + // adjusted to reflect the number of bytes sent. Rewind it. offset -= notWritten } break @@ -410,18 +420,26 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if offsetAddr != 0 { // Copy out the new offset. - if _, err := t.CopyOut(offsetAddr, offset); err != nil { + offsetP := primitive.Uint64(offset) + if _, err := offsetP.CopyOut(t, offsetAddr); err != nil { return 0, nil, err } } - if n == 0 { - return 0, nil, err + if n != 0 { + inFile.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) + outFile.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) + + if err != nil && err != io.EOF && err != syserror.ErrWouldBlock { + // If a partial write is completed, the error is dropped. Log it here. + log.Debugf("sendfile completed a partial write with error: %v", err) + err = nil + } } - inFile.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) - outFile.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) - return uintptr(n), nil, nil + // We can only pass a single file to handleIOError, so pick inFile arbitrarily. + // This is used only for debugging purposes. + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, syserror.ERESTARTSYS, "sendfile", inFile) } // dualWaiter is used to wait on one or both vfs.FileDescriptions. It is not diff --git a/pkg/sentry/syscalls/linux/vfs2/stat.go b/pkg/sentry/syscalls/linux/vfs2/stat.go index bb1d5cac4..0f5d5189c 100644 --- a/pkg/sentry/syscalls/linux/vfs2/stat.go +++ b/pkg/sentry/syscalls/linux/vfs2/stat.go @@ -65,7 +65,7 @@ func fstatat(t *kernel.Task, dirfd int32, pathAddr, statAddr usermem.Addr, flags } root := t.FSContext().RootDirectoryVFS2() - defer root.DecRef() + defer root.DecRef(t) start := root if !path.Absolute { if !path.HasComponents() && flags&linux.AT_EMPTY_PATH == 0 { @@ -73,7 +73,7 @@ func fstatat(t *kernel.Task, dirfd int32, pathAddr, statAddr usermem.Addr, flags } if dirfd == linux.AT_FDCWD { start = t.FSContext().WorkingDirectoryVFS2() - defer start.DecRef() + defer start.DecRef(t) } else { dirfile := t.GetFileVFS2(dirfd) if dirfile == nil { @@ -85,7 +85,7 @@ func fstatat(t *kernel.Task, dirfd int32, pathAddr, statAddr usermem.Addr, flags // former may be able to use opened file state to expedite the // Stat. statx, err := dirfile.Stat(t, opts) - dirfile.DecRef() + dirfile.DecRef(t) if err != nil { return err } @@ -96,8 +96,8 @@ func fstatat(t *kernel.Task, dirfd int32, pathAddr, statAddr usermem.Addr, flags } start = dirfile.VirtualDentry() start.IncRef() - defer start.DecRef() - dirfile.DecRef() + defer start.DecRef(t) + dirfile.DecRef(t) } } @@ -132,7 +132,7 @@ func Fstat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) statx, err := file.Stat(t, vfs.StatOptions{ Mask: linux.STATX_BASIC_STATS, @@ -177,7 +177,7 @@ func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } root := t.FSContext().RootDirectoryVFS2() - defer root.DecRef() + defer root.DecRef(t) start := root if !path.Absolute { if !path.HasComponents() && flags&linux.AT_EMPTY_PATH == 0 { @@ -185,7 +185,7 @@ func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } if dirfd == linux.AT_FDCWD { start = t.FSContext().WorkingDirectoryVFS2() - defer start.DecRef() + defer start.DecRef(t) } else { dirfile := t.GetFileVFS2(dirfd) if dirfile == nil { @@ -197,7 +197,7 @@ func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // former may be able to use opened file state to expedite the // Stat. statx, err := dirfile.Stat(t, opts) - dirfile.DecRef() + dirfile.DecRef(t) if err != nil { return 0, nil, err } @@ -207,8 +207,8 @@ func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } start = dirfile.VirtualDentry() start.IncRef() - defer start.DecRef() - dirfile.DecRef() + defer start.DecRef(t) + dirfile.DecRef(t) } } @@ -282,7 +282,7 @@ func accessAt(t *kernel.Task, dirfd int32, pathAddr usermem.Addr, mode uint) err if err != nil { return err } - defer tpop.Release() + defer tpop.Release(t) // access(2) and faccessat(2) check permissions using real // UID/GID, not effective UID/GID. @@ -328,7 +328,7 @@ func readlinkat(t *kernel.Task, dirfd int32, pathAddr, bufAddr usermem.Addr, siz if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) target, err := t.Kernel().VFS().ReadlinkAt(t, t.Credentials(), &tpop.pop) if err != nil { @@ -358,7 +358,7 @@ func Statfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) statfs, err := t.Kernel().VFS().StatFSAt(t, t.Credentials(), &tpop.pop) if err != nil { @@ -377,7 +377,7 @@ func Fstatfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) statfs, err := t.Kernel().VFS().StatFSAt(t, t.Credentials(), &tpop.pop) if err != nil { diff --git a/pkg/sentry/syscalls/linux/vfs2/sync.go b/pkg/sentry/syscalls/linux/vfs2/sync.go index 0d0ebf46a..6e9b599e2 100644 --- a/pkg/sentry/syscalls/linux/vfs2/sync.go +++ b/pkg/sentry/syscalls/linux/vfs2/sync.go @@ -34,7 +34,7 @@ func Syncfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return 0, nil, file.SyncFS(t) } @@ -47,7 +47,7 @@ func Fsync(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return 0, nil, file.Sync(t) } @@ -77,7 +77,7 @@ func SyncFileRange(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // TODO(gvisor.dev/issue/1897): Currently, the only file syncing we support // is a full-file sync, i.e. fsync(2). As a result, there are severe @@ -108,7 +108,7 @@ func SyncFileRange(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel if flags&linux.SYNC_FILE_RANGE_WAIT_AFTER != 0 { if err := file.Sync(t); err != nil { - return 0, nil, syserror.ConvertIntr(err, kernel.ERESTARTSYS) + return 0, nil, syserror.ConvertIntr(err, syserror.ERESTARTSYS) } } return 0, nil, nil diff --git a/pkg/sentry/syscalls/linux/vfs2/timerfd.go b/pkg/sentry/syscalls/linux/vfs2/timerfd.go index 5ac79bc09..250870c03 100644 --- a/pkg/sentry/syscalls/linux/vfs2/timerfd.go +++ b/pkg/sentry/syscalls/linux/vfs2/timerfd.go @@ -50,11 +50,11 @@ func TimerfdCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel return 0, nil, syserror.EINVAL } vfsObj := t.Kernel().VFS() - file, err := timerfd.New(vfsObj, clock, fileFlags) + file, err := timerfd.New(t, vfsObj, clock, fileFlags) if err != nil { return 0, nil, err } - defer file.DecRef() + defer file.DecRef(t) fd, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{ CloseOnExec: flags&linux.TFD_CLOEXEC != 0, }) @@ -79,7 +79,7 @@ func TimerfdSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) tfd, ok := file.Impl().(*timerfd.TimerFileDescription) if !ok { @@ -87,7 +87,7 @@ func TimerfdSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne } var newVal linux.Itimerspec - if _, err := t.CopyIn(newValAddr, &newVal); err != nil { + if _, err := newVal.CopyIn(t, newValAddr); err != nil { return 0, nil, err } newS, err := ktime.SettingFromItimerspec(newVal, flags&linux.TFD_TIMER_ABSTIME != 0, tfd.Clock()) @@ -97,7 +97,7 @@ func TimerfdSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne tm, oldS := tfd.SetTime(newS) if oldValAddr != 0 { oldVal := ktime.ItimerspecFromSetting(tm, oldS) - if _, err := t.CopyOut(oldValAddr, &oldVal); err != nil { + if _, err := oldVal.CopyOut(t, oldValAddr); err != nil { return 0, nil, err } } @@ -113,7 +113,7 @@ func TimerfdGettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) tfd, ok := file.Impl().(*timerfd.TimerFileDescription) if !ok { @@ -122,6 +122,6 @@ func TimerfdGettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne tm, s := tfd.GetTime() curVal := ktime.ItimerspecFromSetting(tm, s) - _, err := t.CopyOut(curValAddr, &curVal) + _, err := curVal.CopyOut(t, curValAddr) return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2.go b/pkg/sentry/syscalls/linux/vfs2/vfs2.go index c576d9475..0df3bd449 100644 --- a/pkg/sentry/syscalls/linux/vfs2/vfs2.go +++ b/pkg/sentry/syscalls/linux/vfs2/vfs2.go @@ -93,16 +93,16 @@ func Override() { s.Table[165] = syscalls.Supported("mount", Mount) s.Table[166] = syscalls.Supported("umount2", Umount2) s.Table[187] = syscalls.Supported("readahead", Readahead) - s.Table[188] = syscalls.Supported("setxattr", Setxattr) + s.Table[188] = syscalls.Supported("setxattr", SetXattr) s.Table[189] = syscalls.Supported("lsetxattr", Lsetxattr) s.Table[190] = syscalls.Supported("fsetxattr", Fsetxattr) - s.Table[191] = syscalls.Supported("getxattr", Getxattr) + s.Table[191] = syscalls.Supported("getxattr", GetXattr) s.Table[192] = syscalls.Supported("lgetxattr", Lgetxattr) s.Table[193] = syscalls.Supported("fgetxattr", Fgetxattr) - s.Table[194] = syscalls.Supported("listxattr", Listxattr) + s.Table[194] = syscalls.Supported("listxattr", ListXattr) s.Table[195] = syscalls.Supported("llistxattr", Llistxattr) s.Table[196] = syscalls.Supported("flistxattr", Flistxattr) - s.Table[197] = syscalls.Supported("removexattr", Removexattr) + s.Table[197] = syscalls.Supported("removexattr", RemoveXattr) s.Table[198] = syscalls.Supported("lremovexattr", Lremovexattr) s.Table[199] = syscalls.Supported("fremovexattr", Fremovexattr) s.Table[209] = syscalls.PartiallySupported("io_submit", IoSubmit, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}) @@ -163,16 +163,16 @@ func Override() { // Override ARM64. s = linux.ARM64 - s.Table[5] = syscalls.Supported("setxattr", Setxattr) + s.Table[5] = syscalls.Supported("setxattr", SetXattr) s.Table[6] = syscalls.Supported("lsetxattr", Lsetxattr) s.Table[7] = syscalls.Supported("fsetxattr", Fsetxattr) - s.Table[8] = syscalls.Supported("getxattr", Getxattr) + s.Table[8] = syscalls.Supported("getxattr", GetXattr) s.Table[9] = syscalls.Supported("lgetxattr", Lgetxattr) s.Table[10] = syscalls.Supported("fgetxattr", Fgetxattr) - s.Table[11] = syscalls.Supported("listxattr", Listxattr) + s.Table[11] = syscalls.Supported("listxattr", ListXattr) s.Table[12] = syscalls.Supported("llistxattr", Llistxattr) s.Table[13] = syscalls.Supported("flistxattr", Flistxattr) - s.Table[14] = syscalls.Supported("removexattr", Removexattr) + s.Table[14] = syscalls.Supported("removexattr", RemoveXattr) s.Table[15] = syscalls.Supported("lremovexattr", Lremovexattr) s.Table[16] = syscalls.Supported("fremovexattr", Fremovexattr) s.Table[17] = syscalls.Supported("getcwd", Getcwd) diff --git a/pkg/sentry/syscalls/linux/vfs2/xattr.go b/pkg/sentry/syscalls/linux/vfs2/xattr.go index af455d5c1..e05723ef9 100644 --- a/pkg/sentry/syscalls/linux/vfs2/xattr.go +++ b/pkg/sentry/syscalls/linux/vfs2/xattr.go @@ -26,8 +26,8 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) -// Listxattr implements Linux syscall listxattr(2). -func Listxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { +// ListXattr implements Linux syscall listxattr(2). +func ListXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { return listxattr(t, args, followFinalSymlink) } @@ -49,9 +49,9 @@ func listxattr(t *kernel.Task, args arch.SyscallArguments, shouldFollowFinalSyml if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) - names, err := t.Kernel().VFS().ListxattrAt(t, t.Credentials(), &tpop.pop, uint64(size)) + names, err := t.Kernel().VFS().ListXattrAt(t, t.Credentials(), &tpop.pop, uint64(size)) if err != nil { return 0, nil, err } @@ -72,9 +72,9 @@ func Flistxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) - names, err := file.Listxattr(t, uint64(size)) + names, err := file.ListXattr(t, uint64(size)) if err != nil { return 0, nil, err } @@ -85,8 +85,8 @@ func Flistxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy return uintptr(n), nil, nil } -// Getxattr implements Linux syscall getxattr(2). -func Getxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { +// GetXattr implements Linux syscall getxattr(2). +func GetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { return getxattr(t, args, followFinalSymlink) } @@ -109,14 +109,14 @@ func getxattr(t *kernel.Task, args arch.SyscallArguments, shouldFollowFinalSymli if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) name, err := copyInXattrName(t, nameAddr) if err != nil { return 0, nil, err } - value, err := t.Kernel().VFS().GetxattrAt(t, t.Credentials(), &tpop.pop, &vfs.GetxattrOptions{ + value, err := t.Kernel().VFS().GetXattrAt(t, t.Credentials(), &tpop.pop, &vfs.GetXattrOptions{ Name: name, Size: uint64(size), }) @@ -141,14 +141,14 @@ func Fgetxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) name, err := copyInXattrName(t, nameAddr) if err != nil { return 0, nil, err } - value, err := file.Getxattr(t, &vfs.GetxattrOptions{Name: name, Size: uint64(size)}) + value, err := file.GetXattr(t, &vfs.GetXattrOptions{Name: name, Size: uint64(size)}) if err != nil { return 0, nil, err } @@ -159,8 +159,8 @@ func Fgetxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys return uintptr(n), nil, nil } -// Setxattr implements Linux syscall setxattr(2). -func Setxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { +// SetXattr implements Linux syscall setxattr(2). +func SetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { return 0, nil, setxattr(t, args, followFinalSymlink) } @@ -188,7 +188,7 @@ func setxattr(t *kernel.Task, args arch.SyscallArguments, shouldFollowFinalSymli if err != nil { return err } - defer tpop.Release() + defer tpop.Release(t) name, err := copyInXattrName(t, nameAddr) if err != nil { @@ -199,7 +199,7 @@ func setxattr(t *kernel.Task, args arch.SyscallArguments, shouldFollowFinalSymli return err } - return t.Kernel().VFS().SetxattrAt(t, t.Credentials(), &tpop.pop, &vfs.SetxattrOptions{ + return t.Kernel().VFS().SetXattrAt(t, t.Credentials(), &tpop.pop, &vfs.SetXattrOptions{ Name: name, Value: value, Flags: uint32(flags), @@ -222,7 +222,7 @@ func Fsetxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) name, err := copyInXattrName(t, nameAddr) if err != nil { @@ -233,15 +233,15 @@ func Fsetxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys return 0, nil, err } - return 0, nil, file.Setxattr(t, &vfs.SetxattrOptions{ + return 0, nil, file.SetXattr(t, &vfs.SetXattrOptions{ Name: name, Value: value, Flags: uint32(flags), }) } -// Removexattr implements Linux syscall removexattr(2). -func Removexattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { +// RemoveXattr implements Linux syscall removexattr(2). +func RemoveXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { return 0, nil, removexattr(t, args, followFinalSymlink) } @@ -262,14 +262,14 @@ func removexattr(t *kernel.Task, args arch.SyscallArguments, shouldFollowFinalSy if err != nil { return err } - defer tpop.Release() + defer tpop.Release(t) name, err := copyInXattrName(t, nameAddr) if err != nil { return err } - return t.Kernel().VFS().RemovexattrAt(t, t.Credentials(), &tpop.pop, name) + return t.Kernel().VFS().RemoveXattrAt(t, t.Credentials(), &tpop.pop, name) } // Fremovexattr implements Linux syscall fremovexattr(2). @@ -281,14 +281,14 @@ func Fremovexattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) name, err := copyInXattrName(t, nameAddr) if err != nil { return 0, nil, err } - return 0, nil, file.Removexattr(t, name) + return 0, nil, file.RemoveXattr(t, name) } func copyInXattrName(t *kernel.Task, nameAddr usermem.Addr) (string, error) { |