Port most syscalls to VFS2.

pipe and pipe2 aren't ported, pending a slight rework of pipe FDs for VFS2.
mount and umount2 aren't ported out of temporary laziness. access and faccessat
need additional FSImpl methods to implement properly, but are stubbed to
prevent googletest from CHECK-failing. Other syscalls require additional
plumbing.

Updates #1623

PiperOrigin-RevId: 297188448

3 files changed, 86 insertions, 3 deletions

diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go
index 23b88f7a6..58001d56c 100644
--- a/pkg/sentry/kernel/fd_table.go
+++ b/pkg/sentry/kernel/fd_table.go
@@ -296,6 +296,50 @@ func (f *FDTable) NewFDs(ctx context.Context, fd int32, files []*fs.File, flags
 	return fds, nil
 }
 
+// NewFDVFS2 allocates a file descriptor greater than or equal to minfd for
+// the given file description. If it succeeds, it takes a reference on file.
+func (f *FDTable) NewFDVFS2(ctx context.Context, minfd int32, file *vfs.FileDescription, flags FDFlags) (int32, error) {
+	if minfd < 0 {
+		// Don't accept negative FDs.
+		return -1, syscall.EINVAL
+	}
+
+	// Default limit.
+	end := int32(math.MaxInt32)
+
+	// Ensure we don't get past the provided limit.
+	if limitSet := limits.FromContext(ctx); limitSet != nil {
+		lim := limitSet.Get(limits.NumberOfFiles)
+		if lim.Cur != limits.Infinity {
+			end = int32(lim.Cur)
+		}
+		if minfd >= end {
+			return -1, syscall.EMFILE
+		}
+	}
+
+	f.mu.Lock()
+	defer f.mu.Unlock()
+
+	// From f.next to find available fd.
+	fd := minfd
+	if fd < f.next {
+		fd = f.next
+	}
+	for fd < end {
+		if d, _, _ := f.get(fd); d == nil {
+			f.setVFS2(fd, file, flags)
+			if fd == f.next {
+				// Update next search start position.
+				f.next = fd + 1
+			}
+			return fd, nil
+		}
+		fd++
+	}
+	return -1, syscall.EMFILE
+}
+
 // NewFDAt sets the file reference for the given FD. If there is an active
 // reference for that FD, the ref count for that existing reference is
 // decremented.
@@ -316,9 +360,6 @@ func (f *FDTable) newFDAt(ctx context.Context, fd int32, file *fs.File, fileVFS2
 		return syscall.EBADF
 	}
 
-	f.mu.Lock()
-	defer f.mu.Unlock()
-
 	// Check the limit for the provided file.
 	if limitSet := limits.FromContext(ctx); limitSet != nil {
 		if lim := limitSet.Get(limits.NumberOfFiles); lim.Cur != limits.Infinity && uint64(fd) >= lim.Cur {
@@ -327,6 +368,8 @@ func (f *FDTable) newFDAt(ctx context.Context, fd int32, file *fs.File, fileVFS2
 	}
 
 	// Install the entry.
+	f.mu.Lock()
+	defer f.mu.Unlock()
 	f.setAll(fd, file, fileVFS2, flags)
 	return nil
 }
diff --git a/pkg/sentry/kernel/fs_context.go b/pkg/sentry/kernel/fs_context.go
index 7218aa24e..47f78df9a 100644
--- a/pkg/sentry/kernel/fs_context.go
+++ b/pkg/sentry/kernel/fs_context.go
@@ -244,6 +244,28 @@ func (f *FSContext) SetRootDirectory(d *fs.Dirent) {
 	old.DecRef()
 }
 
+// SetRootDirectoryVFS2 sets the root directory. It takes a reference on vd.
+//
+// This is not a valid call after free.
+func (f *FSContext) SetRootDirectoryVFS2(vd vfs.VirtualDentry) {
+	if !vd.Ok() {
+		panic("FSContext.SetRootDirectoryVFS2 called with zero-value VirtualDentry")
+	}
+
+	f.mu.Lock()
+
+	if !f.rootVFS2.Ok() {
+		f.mu.Unlock()
+		panic(fmt.Sprintf("FSContext.SetRootDirectoryVFS2(%v)) called after destroy", vd))
+	}
+
+	old := f.rootVFS2
+	vd.IncRef()
+	f.rootVFS2 = vd
+	f.mu.Unlock()
+	old.DecRef()
+}
+
 // Umask returns the current umask.
 func (f *FSContext) Umask() uint {
 	f.mu.Lock()
diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go
index e37e23231..2cee2e6ed 100644
--- a/pkg/sentry/kernel/task.go
+++ b/pkg/sentry/kernel/task.go
@@ -789,6 +789,15 @@ func (t *Task) NewFDFrom(fd int32, file *fs.File, flags FDFlags) (int32, error)
 	return fds[0], nil
 }
 
+// NewFDFromVFS2 is a convenience wrapper for t.FDTable().NewFDVFS2.
+//
+// This automatically passes the task as the context.
+//
+// Precondition: same as FDTable.Get.
+func (t *Task) NewFDFromVFS2(fd int32, file *vfs.FileDescription, flags FDFlags) (int32, error) {
+	return t.fdTable.NewFDVFS2(t, fd, file, flags)
+}
+
 // NewFDAt is a convenience wrapper for t.FDTable().NewFDAt.
 //
 // This automatically passes the task as the context.
@@ -798,6 +807,15 @@ func (t *Task) NewFDAt(fd int32, file *fs.File, flags FDFlags) error {
 	return t.fdTable.NewFDAt(t, fd, file, flags)
 }
 
+// NewFDAtVFS2 is a convenience wrapper for t.FDTable().NewFDAtVFS2.
+//
+// This automatically passes the task as the context.
+//
+// Precondition: same as FDTable.
+func (t *Task) NewFDAtVFS2(fd int32, file *vfs.FileDescription, flags FDFlags) error {
+	return t.fdTable.NewFDAtVFS2(t, fd, file, flags)
+}
+
 // WithMuLocked executes f with t.mu locked.
 func (t *Task) WithMuLocked(f func(*Task)) {
 	t.mu.Lock()