Plumb VFS2 inside the Sentry

- Added fsbridge package with interface that can be used to open
  and read from VFS1 and VFS2 files.
- Converted ELF loader to use fsbridge
- Added VFS2 types to FSContext
- Added vfs.MountNamespace to ThreadGroup

Updates #1623

PiperOrigin-RevId: 295183950

4 files changed, 121 insertions, 6 deletions

diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD
index be16ee686..0d24fd3c4 100644
--- a/pkg/sentry/syscalls/linux/BUILD
+++ b/pkg/sentry/syscalls/linux/BUILD
@@ -74,6 +74,7 @@ go_library(
         "//pkg/sentry/fs/lock",
         "//pkg/sentry/fs/timerfd",
         "//pkg/sentry/fs/tmpfs",
+        "//pkg/sentry/fsbridge",
         "//pkg/sentry/kernel",
         "//pkg/sentry/kernel/auth",
         "//pkg/sentry/kernel/epoll",
diff --git a/pkg/sentry/syscalls/linux/sys_prctl.go b/pkg/sentry/syscalls/linux/sys_prctl.go
index 98db32d77..9c6728530 100644
--- a/pkg/sentry/syscalls/linux/sys_prctl.go
+++ b/pkg/sentry/syscalls/linux/sys_prctl.go
@@ -20,6 +20,7 @@ import (
 	"gvisor.dev/gvisor/pkg/abi/linux"
 	"gvisor.dev/gvisor/pkg/sentry/arch"
 	"gvisor.dev/gvisor/pkg/sentry/fs"
+	"gvisor.dev/gvisor/pkg/sentry/fsbridge"
 	"gvisor.dev/gvisor/pkg/sentry/kernel"
 	"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
 	"gvisor.dev/gvisor/pkg/sentry/mm"
@@ -135,7 +136,7 @@ func Prctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
 			}
 
 			// Set the underlying executable.
-			t.MemoryManager().SetExecutable(file.Dirent)
+			t.MemoryManager().SetExecutable(fsbridge.NewFSFile(file))
 
 		case linux.PR_SET_MM_AUXV,
 			linux.PR_SET_MM_START_CODE,
diff --git a/pkg/sentry/syscalls/linux/sys_thread.go b/pkg/sentry/syscalls/linux/sys_thread.go
index 0c9e2255d..00915fdde 100644
--- a/pkg/sentry/syscalls/linux/sys_thread.go
+++ b/pkg/sentry/syscalls/linux/sys_thread.go
@@ -21,6 +21,7 @@ import (
 	"gvisor.dev/gvisor/pkg/abi/linux"
 	"gvisor.dev/gvisor/pkg/sentry/arch"
 	"gvisor.dev/gvisor/pkg/sentry/fs"
+	"gvisor.dev/gvisor/pkg/sentry/fsbridge"
 	"gvisor.dev/gvisor/pkg/sentry/kernel"
 	"gvisor.dev/gvisor/pkg/sentry/kernel/sched"
 	"gvisor.dev/gvisor/pkg/sentry/loader"
@@ -119,7 +120,7 @@ func execveat(t *kernel.Task, dirFD int32, pathnameAddr, argvAddr, envvAddr user
 	defer root.DecRef()
 
 	var wd *fs.Dirent
-	var executable *fs.File
+	var executable fsbridge.File
 	var closeOnExec bool
 	if dirFD == linux.AT_FDCWD || path.IsAbs(pathname) {
 		// Even if the pathname is absolute, we may still need the wd
@@ -136,7 +137,15 @@ func execveat(t *kernel.Task, dirFD int32, pathnameAddr, argvAddr, envvAddr user
 		closeOnExec = fdFlags.CloseOnExec
 
 		if atEmptyPath && len(pathname) == 0 {
-			executable = f
+			// TODO(gvisor.dev/issue/160): Linux requires only execute permission,
+			// not read. However, our backing filesystems may prevent us from reading
+			// the file without read permission. Additionally, a task with a
+			// non-readable executable has additional constraints on access via
+			// ptrace and procfs.
+			if err := f.Dirent.Inode.CheckPermission(t, fs.PermMask{Read: true, Execute: true}); err != nil {
+				return 0, nil, err
+			}
+			executable = fsbridge.NewFSFile(f)
 		} else {
 			wd = f.Dirent
 			wd.IncRef()
@@ -152,9 +161,7 @@ func execveat(t *kernel.Task, dirFD int32, pathnameAddr, argvAddr, envvAddr user
 	// Load the new TaskContext.
 	remainingTraversals := uint(linux.MaxSymlinkTraversals)
 	loadArgs := loader.LoadArgs{
-		Mounts:              t.MountNamespace(),
-		Root:                root,
-		WorkingDirectory:    wd,
+		Opener:              fsbridge.NewFSLookup(t.MountNamespace(), root, wd),
 		RemainingTraversals: &remainingTraversals,
 		ResolveFinal:        resolveFinal,
 		Filename:            pathname,
diff --git a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go
index c134714ee..e0ac32b33 100644
--- a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go
+++ b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go
@@ -22,4 +22,110 @@ import (
 // Override syscall table to add syscalls implementations from this package.
 func Override(table map[uintptr]kernel.Syscall) {
 	table[0] = syscalls.Supported("read", Read)
+
+	// Remove syscalls that haven't been converted yet. It's better to get ENOSYS
+	// rather than a SIGSEGV deep in the stack.
+	delete(table, 1)   // write
+	delete(table, 2)   // open
+	delete(table, 3)   // close
+	delete(table, 4)   // stat
+	delete(table, 5)   // fstat
+	delete(table, 6)   // lstat
+	delete(table, 7)   // poll
+	delete(table, 8)   // lseek
+	delete(table, 9)   // mmap
+	delete(table, 16)  // ioctl
+	delete(table, 17)  // pread64
+	delete(table, 18)  // pwrite64
+	delete(table, 19)  // readv
+	delete(table, 20)  // writev
+	delete(table, 21)  // access
+	delete(table, 22)  // pipe
+	delete(table, 32)  // dup
+	delete(table, 33)  // dup2
+	delete(table, 40)  // sendfile
+	delete(table, 59)  // execve
+	delete(table, 72)  // fcntl
+	delete(table, 73)  // flock
+	delete(table, 74)  // fsync
+	delete(table, 75)  // fdatasync
+	delete(table, 76)  // truncate
+	delete(table, 77)  // ftruncate
+	delete(table, 78)  // getdents
+	delete(table, 79)  // getcwd
+	delete(table, 80)  // chdir
+	delete(table, 81)  // fchdir
+	delete(table, 82)  // rename
+	delete(table, 83)  // mkdir
+	delete(table, 84)  // rmdir
+	delete(table, 85)  // creat
+	delete(table, 86)  // link
+	delete(table, 87)  // unlink
+	delete(table, 88)  // symlink
+	delete(table, 89)  // readlink
+	delete(table, 90)  // chmod
+	delete(table, 91)  // fchmod
+	delete(table, 92)  // chown
+	delete(table, 93)  // fchown
+	delete(table, 94)  // lchown
+	delete(table, 133) // mknod
+	delete(table, 137) // statfs
+	delete(table, 138) // fstatfs
+	delete(table, 161) // chroot
+	delete(table, 162) // sync
+	delete(table, 165) // mount
+	delete(table, 166) // umount2
+	delete(table, 172) // iopl
+	delete(table, 173) // ioperm
+	delete(table, 187) // readahead
+	delete(table, 188) // setxattr
+	delete(table, 189) // lsetxattr
+	delete(table, 190) // fsetxattr
+	delete(table, 191) // getxattr
+	delete(table, 192) // lgetxattr
+	delete(table, 193) // fgetxattr
+	delete(table, 206) // io_setup
+	delete(table, 207) // io_destroy
+	delete(table, 208) // io_getevents
+	delete(table, 209) // io_submit
+	delete(table, 210) // io_cancel
+	delete(table, 213) // epoll_create
+	delete(table, 214) // epoll_ctl_old
+	delete(table, 215) // epoll_wait_old
+	delete(table, 216) // remap_file_pages
+	delete(table, 217) // getdents64
+	delete(table, 232) // epoll_wait
+	delete(table, 233) // epoll_ctl
+	delete(table, 253) // inotify_init
+	delete(table, 254) // inotify_add_watch
+	delete(table, 255) // inotify_rm_watch
+	delete(table, 257) // openat
+	delete(table, 258) // mkdirat
+	delete(table, 259) // mknodat
+	delete(table, 260) // fchownat
+	delete(table, 261) // futimesat
+	delete(table, 262) // fstatat
+	delete(table, 263) // unlinkat
+	delete(table, 264) // renameat
+	delete(table, 265) // linkat
+	delete(table, 266) // symlinkat
+	delete(table, 267) // readlinkat
+	delete(table, 268) // fchmodat
+	delete(table, 269) // faccessat
+	delete(table, 270) // pselect
+	delete(table, 271) // ppoll
+	delete(table, 285) // fallocate
+	delete(table, 291) // epoll_create1
+	delete(table, 292) // dup3
+	delete(table, 293) // pipe2
+	delete(table, 294) // inotify_init1
+	delete(table, 295) // preadv
+	delete(table, 296) // pwritev
+	delete(table, 306) // syncfs
+	delete(table, 316) // renameat2
+	delete(table, 319) // memfd_create
+	delete(table, 322) // execveat
+	delete(table, 327) // preadv2
+	delete(table, 328) // pwritev2
+	delete(table, 332) // statx
 }