ksys_umount was refactored to into split into another function
(path_umount) to enable sharing code. This changed the order that flags and
permissions are validated in, and made it so that user_path_at was called
before validating flags and capabilities.
Unfortunately, libfuse2[1] and libmount[2] rely on the old flag validation
behaviour to determine whether or not the kernel supports UMOUNT_NOFOLLOW.
The other path that this validation is being checked on is
init_umount->path_umount->can_umount. That's all internal to the kernel.
[1]: https://github.com/libfuse/libfuse/blob/9bfbeb576c5901b62a171d35510f0d1a922020b7/util/fusermount.c#L403
[2]: https://github.com/karelzak/util-linux/blob/7ed579523b556b1270f28dbdb7ee07dee310f157/libmount/src/context_umount.c#L813
Signed-off-by: Sargun Dhillon <redacted>
Cc: Christoph Hellwig <hch@lst.de>
Fixes: 41525f56e256 ("fs: refactor ksys_umount")
---
fs/namespace.c | 10 ++++++----
1 file changed, 6 insertions(+), 4 deletions(-)
diff --git a/fs/namespace.c b/fs/namespace.c
index cebaa3e81794..dc76f1cb89f4 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1710,10 +1710,6 @@ static int can_umount(const struct path *path, int flags)
{
struct mount *mnt = real_mount(path->mnt);
- if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
- return -EINVAL;
- if (!may_mount())
- return -EPERM;
if (path->dentry != path->mnt->mnt_root)
return -EINVAL;
if (!check_mnt(mnt))@@ -1746,6 +1742,12 @@ static int ksys_umount(char __user *name, int flags)
struct path path;
int ret;
+ if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
+ return -EINVAL;
+
+ if (!may_mount())
+ return -EPERM;
+
if (!(flags & UMOUNT_NOFOLLOW))
lookup_flags |= LOOKUP_FOLLOW;
ret = user_path_at(AT_FDCWD, name, lookup_flags, &path);
--
2.25.1
On Wed, Dec 23, 2020 at 02:26:04AM -0800, Sargun Dhillon wrote:
ksys_umount was refactored to into split into another function
(path_umount) to enable sharing code. This changed the order that flags and
permissions are validated in, and made it so that user_path_at was called
before validating flags and capabilities.
Unfortunately, libfuse2[1] and libmount[2] rely on the old flag validation
behaviour to determine whether or not the kernel supports UMOUNT_NOFOLLOW.
The other path that this validation is being checked on is
init_umount->path_umount->can_umount. That's all internal to the kernel.
[1]: https://github.com/libfuse/libfuse/blob/9bfbeb576c5901b62a171d35510f0d1a922020b7/util/fusermount.c#L403
[2]: https://github.com/karelzak/util-linux/blob/7ed579523b556b1270f28dbdb7ee07dee310f157/libmount/src/context_umount.c#L813
Sorry, I don't like that solution. If nothing else, it turns path_umount() into
a landmine for the future. Yes, we have a regression, yes, we need to do something
about it, but that's not a good way to do that.
FWIW, I would rather separate the check of flags validity from can_umount()
and lift _that_ into ksys_umount(), with "path_umount() relies upon the
flags being minimally sane" comment slapped at path_umount() definition.
The rest of can_umount() really shouldn't be taken out of there.
On Mon, Jan 04, 2021 at 07:51:27PM +0000, Al Viro wrote:
On Wed, Dec 23, 2020 at 02:26:04AM -0800, Sargun Dhillon wrote:
quoted
ksys_umount was refactored to into split into another function
(path_umount) to enable sharing code. This changed the order that flags and
permissions are validated in, and made it so that user_path_at was called
before validating flags and capabilities.
Unfortunately, libfuse2[1] and libmount[2] rely on the old flag validation
behaviour to determine whether or not the kernel supports UMOUNT_NOFOLLOW.
The other path that this validation is being checked on is
init_umount->path_umount->can_umount. That's all internal to the kernel.
[1]: https://github.com/libfuse/libfuse/blob/9bfbeb576c5901b62a171d35510f0d1a922020b7/util/fusermount.c#L403
[2]: https://github.com/karelzak/util-linux/blob/7ed579523b556b1270f28dbdb7ee07dee310f157/libmount/src/context_umount.c#L813
Sorry, I don't like that solution. If nothing else, it turns path_umount() into
a landmine for the future. Yes, we have a regression, yes, we need to do something
about it, but that's not a good way to do that.
FWIW, I would rather separate the check of flags validity from can_umount()
and lift _that_ into ksys_umount(), with "path_umount() relies upon the
flags being minimally sane" comment slapped at path_umount() definition.
The rest of can_umount() really shouldn't be taken out of there.
I mean something like the following; unlike your variant, may_mount() is left
where it is.
commit a0a6df9afcaf439a6b4c88a3b522e3d05fdef46f
Author: Al Viro [off-list ref]
Date: Mon Jan 4 15:25:34 2021 -0500
umount(2): move the flag validity checks first
Unfortunately, there's userland code that used to rely upon these
checks being done before anything else to check for UMOUNT_NOFOLLOW
support. That broke in 41525f56e256 ("fs: refactor ksys_umount").
Separate those from the rest of checks and move them to ksys_umount();
unlike everything else in there, this can be sanely done there.
Reported-by: Sargun Dhillon [off-list ref]
Fixes: 41525f56e256 ("fs: refactor ksys_umount")
Signed-off-by: Al Viro [off-list ref]
diff --git a/fs/namespace.c b/fs/namespace.c
index d2db7dfe232b..9d33909d0f9e 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1713,8 +1713,6 @@ static int can_umount(const struct path *path, int flags)
{
struct mount *mnt = real_mount(path->mnt);
- if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
- return -EINVAL;
if (!may_mount())
return -EPERM;
if (path->dentry != path->mnt->mnt_root)@@ -1728,6 +1726,7 @@ static int can_umount(const struct path *path, int flags)
return 0;
}
+// caller is responsible for flags being sane
int path_umount(struct path *path, int flags)
{
struct mount *mnt = real_mount(path->mnt);@@ -1749,6 +1748,10 @@ static int ksys_umount(char __user *name, int flags)
struct path path;
int ret;
+ // basic validity checks done first
+ if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
+ return -EINVAL;
+
if (!(flags & UMOUNT_NOFOLLOW))
lookup_flags |= LOOKUP_FOLLOW;
ret = user_path_at(AT_FDCWD, name, lookup_flags, &path);