Port fallocate to VFS2.

PiperOrigin-RevId: 319283715

15 files changed, 205 insertions, 2 deletions

diff --git a/pkg/p9/p9.go b/pkg/p9/p9.go
index 28d851ff5..122c457d2 100644
--- a/pkg/p9/p9.go
+++ b/pkg/p9/p9.go
@@ -1091,6 +1091,19 @@ type AllocateMode struct {
 	Unshare       bool
 }
 
+// ToAllocateMode returns an AllocateMode from a fallocate(2) mode.
+func ToAllocateMode(mode uint64) AllocateMode {
+	return AllocateMode{
+		KeepSize:      mode&unix.FALLOC_FL_KEEP_SIZE != 0,
+		PunchHole:     mode&unix.FALLOC_FL_PUNCH_HOLE != 0,
+		NoHideStale:   mode&unix.FALLOC_FL_NO_HIDE_STALE != 0,
+		CollapseRange: mode&unix.FALLOC_FL_COLLAPSE_RANGE != 0,
+		ZeroRange:     mode&unix.FALLOC_FL_ZERO_RANGE != 0,
+		InsertRange:   mode&unix.FALLOC_FL_INSERT_RANGE != 0,
+		Unshare:       mode&unix.FALLOC_FL_UNSHARE_RANGE != 0,
+	}
+}
+
 // ToLinux converts to a value compatible with fallocate(2)'s mode.
 func (a *AllocateMode) ToLinux() uint32 {
 	rv := uint32(0)
diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go
index 404a452c4..faba73af2 100644
--- a/pkg/sentry/fsimpl/gofer/regular_file.go
+++ b/pkg/sentry/fsimpl/gofer/regular_file.go
@@ -24,6 +24,7 @@ import (
 	"gvisor.dev/gvisor/pkg/abi/linux"
 	"gvisor.dev/gvisor/pkg/context"
 	"gvisor.dev/gvisor/pkg/log"
+	"gvisor.dev/gvisor/pkg/p9"
 	"gvisor.dev/gvisor/pkg/safemem"
 	"gvisor.dev/gvisor/pkg/sentry/fs/fsutil"
 	"gvisor.dev/gvisor/pkg/sentry/memmap"
@@ -67,6 +68,34 @@ func (fd *regularFileFD) OnClose(ctx context.Context) error {
 	return d.handle.file.flush(ctx)
 }
 
+// Allocate implements vfs.FileDescriptionImpl.Allocate.
+func (fd *regularFileFD) Allocate(ctx context.Context, mode, offset, length uint64) error {
+
+	d := fd.dentry()
+	d.metadataMu.Lock()
+	defer d.metadataMu.Unlock()
+
+	size := offset + length
+
+	// Allocating a smaller size is a noop.
+	if size <= d.size {
+		return nil
+	}
+
+	d.handleMu.Lock()
+	defer d.handleMu.Unlock()
+
+	err := d.handle.file.allocate(ctx, p9.ToAllocateMode(mode), offset, length)
+	if err != nil {
+		return err
+	}
+	d.size = size
+	if !d.cachedMetadataAuthoritative() {
+		d.touchCMtimeLocked()
+	}
+	return nil
+}
+
 // PRead implements vfs.FileDescriptionImpl.PRead.
 func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) {
 	if offset < 0 {
diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go
index d5e73ddae..007b3332e 100644
--- a/pkg/sentry/fsimpl/host/host.go
+++ b/pkg/sentry/fsimpl/host/host.go
@@ -543,6 +543,16 @@ func (f *fileDescription) Release() {
 	// noop
 }
 
+// Allocate implements vfs.FileDescriptionImpl.
+func (f *fileDescription) Allocate(ctx context.Context, mode, offset, length uint64) error {
+	if !f.inode.seekable {
+		return syserror.ESPIPE
+	}
+
+	// TODO(gvisor.dev/issue/2923): Implement Allocate for non-pipe hostfds.
+	return syserror.EOPNOTSUPP
+}
+
 // PRead implements FileDescriptionImpl.
 func (f *fileDescription) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) {
 	i := f.inode
diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go
index 5f7853a2a..ca8b8c63b 100644
--- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go
+++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go
@@ -236,6 +236,11 @@ func (fd *GenericDirectoryFD) SetStat(ctx context.Context, opts vfs.SetStatOptio
 	return inode.SetStat(ctx, fd.filesystem(), creds, opts)
 }
 
+// Allocate implements vfs.FileDescriptionImpl.Allocate.
+func (fd *GenericDirectoryFD) Allocate(ctx context.Context, mode, offset, length uint64) error {
+	return fd.DirectoryFileDescriptionDefaultImpl.Allocate(ctx, mode, offset, length)
+}
+
 // LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX.
 func (fd *GenericDirectoryFD) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error {
 	return fd.Locks().LockPOSIX(ctx, &fd.vfsfd, uid, t, start, length, whence, block)
diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go
index b805aadd0..6691add96 100644
--- a/pkg/sentry/fsimpl/tmpfs/regular_file.go
+++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go
@@ -274,6 +274,21 @@ func (fd *regularFileFD) Release() {
 	// noop
 }
 
+// Allocate implements vfs.FileDescriptionImpl.Allocate.
+func (fd *regularFileFD) Allocate(ctx context.Context, mode, offset, length uint64) error {
+	f := fd.inode().impl.(*regularFile)
+
+	f.inode.mu.Lock()
+	defer f.inode.mu.Unlock()
+	oldSize := f.size
+	size := offset + length
+	if oldSize >= size {
+		return nil
+	}
+	_, err := f.truncateLocked(size)
+	return err
+}
+
 // PRead implements vfs.FileDescriptionImpl.PRead.
 func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) {
 	if offset < 0 {
diff --git a/pkg/sentry/kernel/pipe/vfs.go b/pkg/sentry/kernel/pipe/vfs.go
index a4519363f..45d4c5fc1 100644
--- a/pkg/sentry/kernel/pipe/vfs.go
+++ b/pkg/sentry/kernel/pipe/vfs.go
@@ -200,6 +200,11 @@ func (fd *VFSPipeFD) Readiness(mask waiter.EventMask) waiter.EventMask {
 	}
 }
 
+// Allocate implements vfs.FileDescriptionImpl.Allocate.
+func (fd *VFSPipeFD) Allocate(ctx context.Context, mode, offset, length uint64) error {
+	return syserror.ESPIPE
+}
+
 // EventRegister implements waiter.Waitable.EventRegister.
 func (fd *VFSPipeFD) EventRegister(e *waiter.Entry, mask waiter.EventMask) {
 	fd.pipe.EventRegister(e, mask)
diff --git a/pkg/sentry/socket/hostinet/socket_vfs2.go b/pkg/sentry/socket/hostinet/socket_vfs2.go
index ad5f64799..8f192c62f 100644
--- a/pkg/sentry/socket/hostinet/socket_vfs2.go
+++ b/pkg/sentry/socket/hostinet/socket_vfs2.go
@@ -96,7 +96,12 @@ func (s *socketVFS2) Ioctl(ctx context.Context, uio usermem.IO, args arch.Syscal
 	return ioctl(ctx, s.fd, uio, args)
 }
 
-// PRead implements vfs.FileDescriptionImpl.
+// Allocate implements vfs.FileDescriptionImpl.Allocate.
+func (s *socketVFS2) Allocate(ctx context.Context, mode, offset, length uint64) error {
+	return syserror.ENODEV
+}
+
+// PRead implements vfs.FileDescriptionImpl.PRead.
 func (s *socketVFS2) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) {
 	return 0, syserror.ESPIPE
 }
diff --git a/pkg/sentry/syscalls/linux/vfs2/filesystem.go b/pkg/sentry/syscalls/linux/vfs2/filesystem.go
index 5dac77e4d..b12b5967b 100644
--- a/pkg/sentry/syscalls/linux/vfs2/filesystem.go
+++ b/pkg/sentry/syscalls/linux/vfs2/filesystem.go
@@ -18,6 +18,7 @@ import (
 	"gvisor.dev/gvisor/pkg/abi/linux"
 	"gvisor.dev/gvisor/pkg/sentry/arch"
 	"gvisor.dev/gvisor/pkg/sentry/kernel"
+	"gvisor.dev/gvisor/pkg/sentry/limits"
 	"gvisor.dev/gvisor/pkg/sentry/vfs"
 	"gvisor.dev/gvisor/pkg/syserror"
 	"gvisor.dev/gvisor/pkg/usermem"
@@ -239,6 +240,55 @@ func renameat(t *kernel.Task, olddirfd int32, oldpathAddr usermem.Addr, newdirfd
 	})
 }
 
+// Fallocate implements linux system call fallocate(2).
+func Fallocate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+	fd := args[0].Int()
+	mode := args[1].Uint64()
+	offset := args[2].Int64()
+	length := args[3].Int64()
+
+	file := t.GetFileVFS2(fd)
+
+	if file == nil {
+		return 0, nil, syserror.EBADF
+	}
+	defer file.DecRef()
+
+	if !file.IsWritable() {
+		return 0, nil, syserror.EBADF
+	}
+
+	if mode != 0 {
+		return 0, nil, syserror.ENOTSUP
+	}
+
+	if offset < 0 || length <= 0 {
+		return 0, nil, syserror.EINVAL
+	}
+
+	size := offset + length
+
+	if size < 0 {
+		return 0, nil, syserror.EFBIG
+	}
+
+	limit := limits.FromContext(t).Get(limits.FileSize).Cur
+
+	if uint64(size) >= limit {
+		t.SendSignal(&arch.SignalInfo{
+			Signo: int32(linux.SIGXFSZ),
+			Code:  arch.SignalInfoUser,
+		})
+		return 0, nil, syserror.EFBIG
+	}
+
+	return 0, nil, file.Impl().Allocate(t, mode, uint64(offset), uint64(length))
+
+	// File length modified, generate notification.
+	// TODO(gvisor.dev/issue/1479): Reenable when Inotify is ported.
+	// file.Dirent.InotifyEvent(linux.IN_MODIFY, 0)
+}
+
 // Rmdir implements Linux syscall rmdir(2).
 func Rmdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
 	pathAddr := args[0].Pointer()
diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2.go b/pkg/sentry/syscalls/linux/vfs2/vfs2.go
index 9e60c4a1c..8f497ecc7 100644
--- a/pkg/sentry/syscalls/linux/vfs2/vfs2.go
+++ b/pkg/sentry/syscalls/linux/vfs2/vfs2.go
@@ -138,7 +138,7 @@ func Override() {
 	s.Table[282] = syscalls.Supported("signalfd", Signalfd)
 	s.Table[283] = syscalls.Supported("timerfd_create", TimerfdCreate)
 	s.Table[284] = syscalls.Supported("eventfd", Eventfd)
-	delete(s.Table, 285) // fallocate
+	s.Table[285] = syscalls.PartiallySupported("fallocate", Fallocate, "Not all options are supported.", nil)
 	s.Table[286] = syscalls.Supported("timerfd_settime", TimerfdSettime)
 	s.Table[287] = syscalls.Supported("timerfd_gettime", TimerfdGettime)
 	s.Table[288] = syscalls.Supported("accept4", Accept4)
diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go
index eb5dfd7e2..0c42574db 100644
--- a/pkg/sentry/vfs/file_description.go
+++ b/pkg/sentry/vfs/file_description.go
@@ -354,6 +354,10 @@ type FileDescriptionImpl interface {
 	// represented by the FileDescription.
 	StatFS(ctx context.Context) (linux.Statfs, error)
 
+	// Allocate grows file represented by FileDescription to offset + length bytes.
+	// Only mode == 0 is supported currently.
+	Allocate(ctx context.Context, mode, offset, length uint64) error
+
 	// waiter.Waitable methods may be used to poll for I/O events.
 	waiter.Waitable
 
diff --git a/pkg/sentry/vfs/file_description_impl_util.go b/pkg/sentry/vfs/file_description_impl_util.go
index 3fec0d6d6..6b8b4ad49 100644
--- a/pkg/sentry/vfs/file_description_impl_util.go
+++ b/pkg/sentry/vfs/file_description_impl_util.go
@@ -56,6 +56,12 @@ func (FileDescriptionDefaultImpl) StatFS(ctx context.Context) (linux.Statfs, err
 	return linux.Statfs{}, syserror.ENOSYS
 }
 
+// Allocate implements FileDescriptionImpl.Allocate analogously to
+// fallocate called on regular file, directory or FIFO in Linux.
+func (FileDescriptionDefaultImpl) Allocate(ctx context.Context, mode, offset, length uint64) error {
+	return syserror.ENODEV
+}
+
 // Readiness implements waiter.Waitable.Readiness analogously to
 // file_operations::poll == NULL in Linux.
 func (FileDescriptionDefaultImpl) Readiness(mask waiter.EventMask) waiter.EventMask {
@@ -158,6 +164,11 @@ func (FileDescriptionDefaultImpl) Removexattr(ctx context.Context, name string)
 // implementations of non-directory I/O methods that return EISDIR.
 type DirectoryFileDescriptionDefaultImpl struct{}
 
+// Allocate implements DirectoryFileDescriptionDefaultImpl.Allocate.
+func (DirectoryFileDescriptionDefaultImpl) Allocate(ctx context.Context, mode, offset, length uint64) error {
+	return syserror.EISDIR
+}
+
 // PRead implements FileDescriptionImpl.PRead.
 func (DirectoryFileDescriptionDefaultImpl) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts ReadOptions) (int64, error) {
 	return 0, syserror.EISDIR
diff --git a/pkg/sentry/vfs/inotify.go b/pkg/sentry/vfs/inotify.go
index 509034531..c2e21ac5f 100644
--- a/pkg/sentry/vfs/inotify.go
+++ b/pkg/sentry/vfs/inotify.go
@@ -148,6 +148,11 @@ func (i *Inotify) Release() {
 	}
 }
 
+// Allocate implements FileDescription.Allocate.
+func (i *Inotify) Allocate(ctx context.Context, mode, offset, length uint64) error {
+	panic("Allocate should not be called on read-only inotify fds")
+}
+
 // EventRegister implements waiter.Waitable.
 func (i *Inotify) EventRegister(e *waiter.Entry, mask waiter.EventMask) {
 	i.queue.EventRegister(e, mask)
diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD
index 88ed36b69..67645cc83 100644
--- a/test/syscalls/BUILD
+++ b/test/syscalls/BUILD
@@ -199,6 +199,7 @@ syscall_test(
 syscall_test(
     add_overlay = True,
     test = "//test/syscalls/linux:fallocate_test",
+    vfs2 = "True",
 )
 
 syscall_test(
diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD
index 8c7d54b21..9e097c888 100644
--- a/test/syscalls/linux/BUILD
+++ b/test/syscalls/linux/BUILD
@@ -748,9 +748,14 @@ cc_binary(
     linkstatic = 1,
     deps = [
         ":file_base",
+        ":socket_test_util",
         "//test/util:cleanup",
+        "//test/util:eventfd_util",
         "//test/util:file_descriptor",
+        "@com_google_absl//absl/strings",
+        "@com_google_absl//absl/time",
         gtest,
+        "//test/util:posix_error",
         "//test/util:temp_path",
         "//test/util:test_main",
         "//test/util:test_util",
diff --git a/test/syscalls/linux/fallocate.cc b/test/syscalls/linux/fallocate.cc
index 7819f4ac3..cabc2b751 100644
--- a/test/syscalls/linux/fallocate.cc
+++ b/test/syscalls/linux/fallocate.cc
@@ -15,16 +15,27 @@
 #include <errno.h>
 #include <fcntl.h>
 #include <signal.h>
+#include <sys/eventfd.h>
 #include <sys/resource.h>
+#include <sys/signalfd.h>
+#include <sys/socket.h>
 #include <sys/stat.h>
+#include <sys/timerfd.h>
 #include <syscall.h>
 #include <time.h>
 #include <unistd.h>
 
+#include <ctime>
+
 #include "gtest/gtest.h"
+#include "absl/strings/str_cat.h"
+#include "absl/time/time.h"
 #include "test/syscalls/linux/file_base.h"
+#include "test/syscalls/linux/socket_test_util.h"
 #include "test/util/cleanup.h"
+#include "test/util/eventfd_util.h"
 #include "test/util/file_descriptor.h"
+#include "test/util/posix_error.h"
 #include "test/util/temp_path.h"
 #include "test/util/test_util.h"
 
@@ -70,6 +81,12 @@ TEST_F(AllocateTest, Fallocate) {
   ASSERT_THAT(fallocate(test_file_fd_.get(), 0, 39, 1), SyscallSucceeds());
   ASSERT_THAT(fstat(test_file_fd_.get(), &buf), SyscallSucceeds());
   EXPECT_EQ(buf.st_size, 40);
+
+  // Given length 0 should fail with EINVAL.
+  ASSERT_THAT(fallocate(test_file_fd_.get(), 0, 50, 0),
+              SyscallFailsWithErrno(EINVAL));
+  ASSERT_THAT(fstat(test_file_fd_.get(), &buf), SyscallSucceeds());
+  EXPECT_EQ(buf.st_size, 40);
 }
 
 TEST_F(AllocateTest, FallocateInvalid) {
@@ -136,6 +153,34 @@ TEST_F(AllocateTest, FallocateRlimit) {
   ASSERT_THAT(sigprocmask(SIG_UNBLOCK, &new_mask, nullptr), SyscallSucceeds());
 }
 
+TEST_F(AllocateTest, FallocateOtherFDs) {
+  int fd;
+  ASSERT_THAT(fd = timerfd_create(CLOCK_MONOTONIC, 0), SyscallSucceeds());
+  auto timer_fd = FileDescriptor(fd);
+  EXPECT_THAT(fallocate(timer_fd.get(), 0, 0, 10),
+              SyscallFailsWithErrno(ENODEV));
+
+  sigset_t mask;
+  sigemptyset(&mask);
+  ASSERT_THAT(fd = signalfd(-1, &mask, 0), SyscallSucceeds());
+  auto sfd = FileDescriptor(fd);
+  EXPECT_THAT(fallocate(sfd.get(), 0, 0, 10), SyscallFailsWithErrno(ENODEV));
+
+  auto efd =
+      ASSERT_NO_ERRNO_AND_VALUE(NewEventFD(0, EFD_NONBLOCK | EFD_SEMAPHORE));
+  EXPECT_THAT(fallocate(efd.get(), 0, 0, 10), SyscallFailsWithErrno(ENODEV));
+
+  auto sockfd = ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_DGRAM, 0));
+  EXPECT_THAT(fallocate(sockfd.get(), 0, 0, 10), SyscallFailsWithErrno(ENODEV));
+
+  int socks[2];
+  ASSERT_THAT(socketpair(AF_UNIX, SOCK_STREAM, PF_UNIX, socks),
+              SyscallSucceeds());
+  auto sock0 = FileDescriptor(socks[0]);
+  auto sock1 = FileDescriptor(socks[1]);
+  EXPECT_THAT(fallocate(sock0.get(), 0, 0, 10), SyscallFailsWithErrno(ENODEV));
+}
+
 }  // namespace
 }  // namespace testing
 }  // namespace gvisor