2022-06-01 19:22:41

by David Howells

[permalink] [raw]
Subject: [PATCH] afs: Fix dynamic root getattr

The recent patch to make afs_getattr consult the server didn't account for
the pseudo-inodes employed by the dynamic root-type afs superblock not
having a volume or a server to access, and thus an oops occurs if such a
directory is stat'd.

Fix this by checking to see if the vnode->volume pointer actually points
anywhere before following it in afs_getattr().

This can be tested by stat'ing a directory in /afs. It may be sufficient
just to do "ls /afs" and the oops looks something like:

BUG: kernel NULL pointer dereference, address: 0000000000000020
...
RIP: 0010:afs_getattr+0x8b/0x14b
...
Call Trace:
<TASK>
vfs_statx+0x79/0xf5
vfs_fstatat+0x49/0x62

Fixes: 2aeb8c86d499 ("afs: Fix afs_getattr() to refetch file status if callback break occurred")
Reported-by: Marc Dionne <[email protected]>
Signed-off-by: David Howells <[email protected]>
cc: [email protected]
---

fs/afs/inode.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/fs/afs/inode.c b/fs/afs/inode.c
index 30b066299d39..33ecbfea0199 100644
--- a/fs/afs/inode.c
+++ b/fs/afs/inode.c
@@ -745,7 +745,8 @@ int afs_getattr(struct user_namespace *mnt_userns, const struct path *path,

_enter("{ ino=%lu v=%u }", inode->i_ino, inode->i_generation);

- if (!(query_flags & AT_STATX_DONT_SYNC) &&
+ if (vnode->volume &&
+ !(query_flags & AT_STATX_DONT_SYNC) &&
!test_bit(AFS_VNODE_CB_PROMISED, &vnode->flags)) {
key = afs_request_key(vnode->volume->cell);
if (IS_ERR(key))




2022-06-16 18:26:01

by Marc Dionne

[permalink] [raw]
Subject: Re: [PATCH] afs: Fix dynamic root getattr

On Wed, Jun 1, 2022 at 8:55 AM David Howells <[email protected]> wrote:
>
> The recent patch to make afs_getattr consult the server didn't account for
> the pseudo-inodes employed by the dynamic root-type afs superblock not
> having a volume or a server to access, and thus an oops occurs if such a
> directory is stat'd.
>
> Fix this by checking to see if the vnode->volume pointer actually points
> anywhere before following it in afs_getattr().
>
> This can be tested by stat'ing a directory in /afs. It may be sufficient
> just to do "ls /afs" and the oops looks something like:
>
> BUG: kernel NULL pointer dereference, address: 0000000000000020
> ...
> RIP: 0010:afs_getattr+0x8b/0x14b
> ...
> Call Trace:
> <TASK>
> vfs_statx+0x79/0xf5
> vfs_fstatat+0x49/0x62
>
> Fixes: 2aeb8c86d499 ("afs: Fix afs_getattr() to refetch file status if callback break occurred")
> Reported-by: Marc Dionne <[email protected]>
> Signed-off-by: David Howells <[email protected]>
> cc: [email protected]
> ---
>
> fs/afs/inode.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/fs/afs/inode.c b/fs/afs/inode.c
> index 30b066299d39..33ecbfea0199 100644
> --- a/fs/afs/inode.c
> +++ b/fs/afs/inode.c
> @@ -745,7 +745,8 @@ int afs_getattr(struct user_namespace *mnt_userns, const struct path *path,
>
> _enter("{ ino=%lu v=%u }", inode->i_ino, inode->i_generation);
>
> - if (!(query_flags & AT_STATX_DONT_SYNC) &&
> + if (vnode->volume &&
> + !(query_flags & AT_STATX_DONT_SYNC) &&
> !test_bit(AFS_VNODE_CB_PROMISED, &vnode->flags)) {
> key = afs_request_key(vnode->volume->cell);
> if (IS_ERR(key))

Reviewed-by: Marc Dionne <[email protected]>
Tested-by: Marc Dionne <[email protected]>

Marc