[PATCH v2 0/2] ovl: allow O_PATH file descriptor when specifying layers

[PATCH v2 0/2] ovl: allow O_PATH file descriptor when specifying layers

[Christian Brauner]

Allow overlayfs to use O_PATH file descriptors when specifying layers.
Userspace must currently use non-O_PATH file desriptors which is often
pointless especially if the file descriptors have been created via
open_tree(OPEN_TREE_CLONE). This has been a frequent request and came up
again in [1].

Link: https://lore.kernel.org/r/fd8f6574-f737-4743-b220-79c815ee1554@mbaynton.com [1]

Signed-off-by: Christian Brauner <brauner@kernel.org>
---
Changes in v2:
- Unconditionally enable O_PATH file descriptors.
- Link to v1: https://lore.kernel.org/r/20250207-work-overlayfs-v1-0-611976e73373@kernel.org

---
Christian Brauner (2):
      fs: support O_PATH fds with FSCONFIG_SET_FD
      selftests/overlayfs: test specifying layers as O_PATH file descriptors

 fs/autofs/autofs_i.h                               |  2 +
 fs/fsopen.c                                        |  2 +-
 .../filesystems/overlayfs/set_layers_via_fds.c     | 65 ++++++++++++++++++++++
 3 files changed, 68 insertions(+), 1 deletion(-)
---
base-commit: 2014c95afecee3e76ca4a56956a936e23283f05b
change-id: 20250207-work-overlayfs-38fb9156d4c4

[PATCH v2 1/2] fs: support O_PATH fds with FSCONFIG_SET_FD

[Christian Brauner]

Let FSCONFIG_SET_FD handle O_PATH file descriptors. This is particularly
useful in the context of overlayfs where layers can be specified via
file descriptors instead of paths. But userspace must currently use
non-O_PATH file desriptors which is often pointless especially if
the file descriptors have been created via open_tree(OPEN_TREE_CLONE).

Fixes: a08557d19ef41 ("ovl: specify layers via file descriptors")
Signed-off-by: Christian Brauner <brauner@kernel.org>
---
 fs/autofs/autofs_i.h | 2 ++
 fs/fsopen.c          | 2 +-
 2 files changed, 3 insertions(+), 1 deletion(-)

diff --git a/fs/autofs/autofs_i.h b/fs/autofs/autofs_i.h
index 77c7991d89aa..23cea74f9933 100644
--- a/fs/autofs/autofs_i.h
+++ b/fs/autofs/autofs_i.h
@@ -218,6 +218,8 @@ void autofs_clean_ino(struct autofs_info *);
 
 static inline int autofs_check_pipe(struct file *pipe)
 {
+	if (pipe->f_mode & FMODE_PATH)
+		return -EINVAL;
 	if (!(pipe->f_mode & FMODE_CAN_WRITE))
 		return -EINVAL;
 	if (!S_ISFIFO(file_inode(pipe)->i_mode))
diff --git a/fs/fsopen.c b/fs/fsopen.c
index 094a7f510edf..1aaf4cb2afb2 100644
--- a/fs/fsopen.c
+++ b/fs/fsopen.c
@@ -453,7 +453,7 @@ SYSCALL_DEFINE5(fsconfig,
 	case FSCONFIG_SET_FD:
 		param.type = fs_value_is_file;
 		ret = -EBADF;
-		param.file = fget(aux);
+		param.file = fget_raw(aux);
 		if (!param.file)
 			goto out_key;
 		param.dirfd = aux;

-- 
2.47.2

[PATCH v2 2/2] selftests/overlayfs: test specifying layers as O_PATH file descriptors

[Christian Brauner]

Verify that userspace can specify layers via O_PATH file descriptors.

Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
---
 .../filesystems/overlayfs/set_layers_via_fds.c     | 65 ++++++++++++++++++++++
 1 file changed, 65 insertions(+)

diff --git a/tools/testing/selftests/filesystems/overlayfs/set_layers_via_fds.c b/tools/testing/selftests/filesystems/overlayfs/set_layers_via_fds.c
index 1d0ae785a667..e693e4102d22 100644
--- a/tools/testing/selftests/filesystems/overlayfs/set_layers_via_fds.c
+++ b/tools/testing/selftests/filesystems/overlayfs/set_layers_via_fds.c
@@ -214,4 +214,69 @@ TEST_F(set_layers_via_fds, set_500_layers_via_fds)
 	ASSERT_EQ(close(fd_overlay), 0);
 }
 
+TEST_F(set_layers_via_fds, set_500_layers_via_opath_fds)
+{
+	int fd_context, fd_tmpfs, fd_overlay, fd_work, fd_upper, fd_lower;
+	int layer_fds[500] = { [0 ... 499] = -EBADF };
+
+	ASSERT_EQ(unshare(CLONE_NEWNS), 0);
+	ASSERT_EQ(sys_mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL), 0);
+
+	fd_context = sys_fsopen("tmpfs", 0);
+	ASSERT_GE(fd_context, 0);
+
+	ASSERT_EQ(sys_fsconfig(fd_context, FSCONFIG_CMD_CREATE, NULL, NULL, 0), 0);
+	fd_tmpfs = sys_fsmount(fd_context, 0, 0);
+	ASSERT_GE(fd_tmpfs, 0);
+	ASSERT_EQ(close(fd_context), 0);
+
+	for (int i = 0; i < ARRAY_SIZE(layer_fds); i++) {
+		char path[100];
+
+		sprintf(path, "l%d", i);
+		ASSERT_EQ(mkdirat(fd_tmpfs, path, 0755), 0);
+		layer_fds[i] = openat(fd_tmpfs, path, O_DIRECTORY | O_PATH);
+		ASSERT_GE(layer_fds[i], 0);
+	}
+
+	ASSERT_EQ(mkdirat(fd_tmpfs, "w", 0755), 0);
+	fd_work = openat(fd_tmpfs, "w", O_DIRECTORY | O_PATH);
+	ASSERT_GE(fd_work, 0);
+
+	ASSERT_EQ(mkdirat(fd_tmpfs, "u", 0755), 0);
+	fd_upper = openat(fd_tmpfs, "u", O_DIRECTORY | O_PATH);
+	ASSERT_GE(fd_upper, 0);
+
+	ASSERT_EQ(mkdirat(fd_tmpfs, "l501", 0755), 0);
+	fd_lower = openat(fd_tmpfs, "l501", O_DIRECTORY | O_PATH);
+	ASSERT_GE(fd_lower, 0);
+
+	ASSERT_EQ(sys_move_mount(fd_tmpfs, "", -EBADF, "/tmp", MOVE_MOUNT_F_EMPTY_PATH), 0);
+	ASSERT_EQ(close(fd_tmpfs), 0);
+
+	fd_context = sys_fsopen("overlay", 0);
+	ASSERT_GE(fd_context, 0);
+
+	ASSERT_EQ(sys_fsconfig(fd_context, FSCONFIG_SET_FD, "workdir",   NULL, fd_work), 0);
+	ASSERT_EQ(close(fd_work), 0);
+
+	ASSERT_EQ(sys_fsconfig(fd_context, FSCONFIG_SET_FD, "upperdir",  NULL, fd_upper), 0);
+	ASSERT_EQ(close(fd_upper), 0);
+
+	for (int i = 0; i < ARRAY_SIZE(layer_fds); i++) {
+		ASSERT_EQ(sys_fsconfig(fd_context, FSCONFIG_SET_FD, "lowerdir+", NULL, layer_fds[i]), 0);
+		ASSERT_EQ(close(layer_fds[i]), 0);
+	}
+
+	ASSERT_NE(sys_fsconfig(fd_context, FSCONFIG_SET_FD, "lowerdir+", NULL, fd_lower), 0);
+	ASSERT_EQ(close(fd_lower), 0);
+
+	ASSERT_EQ(sys_fsconfig(fd_context, FSCONFIG_CMD_CREATE, NULL, NULL, 0), 0);
+
+	fd_overlay = sys_fsmount(fd_context, 0, 0);
+	ASSERT_GE(fd_overlay, 0);
+	ASSERT_EQ(close(fd_context), 0);
+	ASSERT_EQ(close(fd_overlay), 0);
+}
+
 TEST_HARNESS_MAIN

-- 
2.47.2

Re: [PATCH v2 1/2] fs: support O_PATH fds with FSCONFIG_SET_FD

[Amir Goldstein]

On Mon, Feb 10, 2025 at 1:39 PM Christian Brauner <brauner@kernel.org> wrote:
>
> Let FSCONFIG_SET_FD handle O_PATH file descriptors. This is particularly
> useful in the context of overlayfs where layers can be specified via
> file descriptors instead of paths. But userspace must currently use
> non-O_PATH file desriptors which is often pointless especially if
> the file descriptors have been created via open_tree(OPEN_TREE_CLONE).
>
> Fixes: a08557d19ef41 ("ovl: specify layers via file descriptors")
> Signed-off-by: Christian Brauner <brauner@kernel.org>
> ---
>  fs/autofs/autofs_i.h | 2 ++
>  fs/fsopen.c          | 2 +-
>  2 files changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/fs/autofs/autofs_i.h b/fs/autofs/autofs_i.h
> index 77c7991d89aa..23cea74f9933 100644
> --- a/fs/autofs/autofs_i.h
> +++ b/fs/autofs/autofs_i.h
> @@ -218,6 +218,8 @@ void autofs_clean_ino(struct autofs_info *);
>
>  static inline int autofs_check_pipe(struct file *pipe)
>  {
> +       if (pipe->f_mode & FMODE_PATH)
> +               return -EINVAL;
>         if (!(pipe->f_mode & FMODE_CAN_WRITE))
>                 return -EINVAL;

I thought you said the above check is redundant due to the lower check.

In any case feel free to add

Reviewed-by: Amir Goldstein <amir73il@gmail.com>

>         if (!S_ISFIFO(file_inode(pipe)->i_mode))
> diff --git a/fs/fsopen.c b/fs/fsopen.c
> index 094a7f510edf..1aaf4cb2afb2 100644
> --- a/fs/fsopen.c
> +++ b/fs/fsopen.c
> @@ -453,7 +453,7 @@ SYSCALL_DEFINE5(fsconfig,
>         case FSCONFIG_SET_FD:
>                 param.type = fs_value_is_file;
>                 ret = -EBADF;
> -               param.file = fget(aux);
> +               param.file = fget_raw(aux);
>                 if (!param.file)
>                         goto out_key;
>                 param.dirfd = aux;
>
> --
> 2.47.2
>

Re: [PATCH v2 1/2] fs: support O_PATH fds with FSCONFIG_SET_FD

[Christian Brauner]

On Mon, Feb 10, 2025 at 02:26:56PM +0100, Amir Goldstein wrote:
> On Mon, Feb 10, 2025 at 1:39 PM Christian Brauner <brauner@kernel.org> wrote:
> >
> > Let FSCONFIG_SET_FD handle O_PATH file descriptors. This is particularly
> > useful in the context of overlayfs where layers can be specified via
> > file descriptors instead of paths. But userspace must currently use
> > non-O_PATH file desriptors which is often pointless especially if
> > the file descriptors have been created via open_tree(OPEN_TREE_CLONE).
> >
> > Fixes: a08557d19ef41 ("ovl: specify layers via file descriptors")
> > Signed-off-by: Christian Brauner <brauner@kernel.org>
> > ---
> >  fs/autofs/autofs_i.h | 2 ++
> >  fs/fsopen.c          | 2 +-
> >  2 files changed, 3 insertions(+), 1 deletion(-)
> >
> > diff --git a/fs/autofs/autofs_i.h b/fs/autofs/autofs_i.h
> > index 77c7991d89aa..23cea74f9933 100644
> > --- a/fs/autofs/autofs_i.h
> > +++ b/fs/autofs/autofs_i.h
> > @@ -218,6 +218,8 @@ void autofs_clean_ino(struct autofs_info *);
> >
> >  static inline int autofs_check_pipe(struct file *pipe)
> >  {
> > +       if (pipe->f_mode & FMODE_PATH)
> > +               return -EINVAL;
> >         if (!(pipe->f_mode & FMODE_CAN_WRITE))
> >                 return -EINVAL;
> 
> I thought you said the above check is redundant due to the lower check.

It is but that's only obvious to people quite familiar with VFS code.
So I like the explicitly check here.

> 
> In any case feel free to add
> 
> Reviewed-by: Amir Goldstein <amir73il@gmail.com>
> 
> >         if (!S_ISFIFO(file_inode(pipe)->i_mode))
> > diff --git a/fs/fsopen.c b/fs/fsopen.c
> > index 094a7f510edf..1aaf4cb2afb2 100644
> > --- a/fs/fsopen.c
> > +++ b/fs/fsopen.c
> > @@ -453,7 +453,7 @@ SYSCALL_DEFINE5(fsconfig,
> >         case FSCONFIG_SET_FD:
> >                 param.type = fs_value_is_file;
> >                 ret = -EBADF;
> > -               param.file = fget(aux);
> > +               param.file = fget_raw(aux);
> >                 if (!param.file)
> >                         goto out_key;
> >                 param.dirfd = aux;
> >
> > --
> > 2.47.2
> >