2022-05-22 06:40:00

by David Howells

[permalink] [raw]
Subject: [PATCH] afs: Fix afs_getattr() to refetch file status if callback break occurred

If a callback break occurs (change notification), afs_getattr() needs to
issue an FS.FetchStatus RPC operation to update the status of the file
being examined by the stat-family of system calls.

Fix afs_getattr() to do this if AFS_VNODE_CB_PROMISED has been cleared on a
vnode by a callback break. Skip this if AT_STATX_DONT_SYNC is set.

This can be tested by appending to a file on one AFS client and then using
"stat -L" to examine its length on a machine running kafs. This can also
be watched through tracing on the kafs machine. The callback break is
seen:

kworker/1:1-46 [001] ..... 978.910812: afs_cb_call: c=0000005f YFSCB.CallBack
kworker/1:1-46 [001] ...1. 978.910829: afs_cb_break: 100058:23b4c:242d2c2 b=2 s=1 break-cb
kworker/1:1-46 [001] ..... 978.911062: afs_call_done: c=0000005f ret=0 ab=0 [0000000082994ead]

And then the stat command generated no traffic if unpatched, but with this
change a call to fetch the status can be observed:

stat-4471 [000] ..... 986.744122: afs_make_fs_call: c=000000ab 100058:023b4c:242d2c2 YFS.FetchStatus
stat-4471 [000] ..... 986.745578: afs_call_done: c=000000ab ret=0 ab=0 [0000000087fc8c84]

Fixes: 08e0e7c82eea ("[AF_RXRPC]: Make the in-kernel AFS filesystem use AF_RXRPC.")
Reported-by: Markus Suvanto <[email protected]>
Signed-off-by: David Howells <[email protected]>
cc: Marc Dionne <[email protected]>
cc: [email protected]
Link: https://bugzilla.kernel.org/show_bug.cgi?id=216010
---

fs/afs/inode.c | 14 +++++++++++++-
1 file changed, 13 insertions(+), 1 deletion(-)

diff --git a/fs/afs/inode.c b/fs/afs/inode.c
index 2fe402483ad5..30b066299d39 100644
--- a/fs/afs/inode.c
+++ b/fs/afs/inode.c
@@ -740,10 +740,22 @@ int afs_getattr(struct user_namespace *mnt_userns, const struct path *path,
{
struct inode *inode = d_inode(path->dentry);
struct afs_vnode *vnode = AFS_FS_I(inode);
- int seq = 0;
+ struct key *key;
+ int ret, seq = 0;

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

+ if (!(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))
+ return PTR_ERR(key);
+ ret = afs_validate(vnode, key);
+ key_put(key);
+ if (ret < 0)
+ return ret;
+ }
+
do {
read_seqbegin_or_lock(&vnode->cb_lock, &seq);
generic_fillattr(&init_user_ns, inode, stat);




2022-05-22 12:24:05

by markus.suvanto

[permalink] [raw]
Subject: Re: [PATCH] afs: Fix afs_getattr() to refetch file status if callback break occurred

pe, 2022-05-20 kello 22:53 +0100, David Howells kirjoitti:
> If a callback break occurs (change notification), afs_getattr() needs
> to
> issue an FS.FetchStatus RPC operation to update the status of the
> file
> being examined by the stat-family of system calls.
>
> Fix afs_getattr() to do this if AFS_VNODE_CB_PROMISED has been
> cleared on a
> vnode by a callback break.  Skip this if AT_STATX_DONT_SYNC is set.
>
> This can be tested by appending to a file on one AFS client and then
> using
> "stat -L" to examine its length on a machine running kafs.  This can
> also
> be watched through tracing on the kafs machine.  The callback break
> is
> seen:
>
>      kworker/1:1-46      [001] .....   978.910812: afs_cb_call:
> c=0000005f YFSCB.CallBack
>      kworker/1:1-46      [001] ...1.   978.910829: afs_cb_break:
> 100058:23b4c:242d2c2 b=2 s=1 break-cb
>      kworker/1:1-46      [001] .....   978.911062: afs_call_done:   
> c=0000005f ret=0 ab=0 [0000000082994ead]
>
> And then the stat command generated no traffic if unpatched, but with
> this
> change a call to fetch the status can be observed:
>
>             stat-4471    [000] .....   986.744122: afs_make_fs_call:
> c=000000ab 100058:023b4c:242d2c2 YFS.FetchStatus
>             stat-4471    [000] .....   986.745578: afs_call_done:   
> c=000000ab ret=0 ab=0 [0000000087fc8c84]
>
> Fixes: 08e0e7c82eea ("[AF_RXRPC]: Make the in-kernel AFS filesystem
> use AF_RXRPC.")
> Reported-by: Markus Suvanto <[email protected]>
> Signed-off-by: David Howells <[email protected]>
> cc: Marc Dionne <[email protected]>
> cc: [email protected]
> Link: https://bugzilla.kernel.org/show_bug.cgi?id=216010
> ---
>
>  fs/afs/inode.c |   14 +++++++++++++-
>  1 file changed, 13 insertions(+), 1 deletion(-)
>
> diff --git a/fs/afs/inode.c b/fs/afs/inode.c
> index 2fe402483ad5..30b066299d39 100644
> --- a/fs/afs/inode.c
> +++ b/fs/afs/inode.c
> @@ -740,10 +740,22 @@ int afs_getattr(struct user_namespace
> *mnt_userns, const struct path *path,
>  {
>         struct inode *inode = d_inode(path->dentry);
>         struct afs_vnode *vnode = AFS_FS_I(inode);
> -       int seq = 0;
> +       struct key *key;
> +       int ret, seq = 0;
>  
>         _enter("{ ino=%lu v=%u }", inode->i_ino, inode-
> >i_generation);
>  
> +       if (!(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))
> +                       return PTR_ERR(key);
> +               ret = afs_validate(vnode, key);
> +               key_put(key);
> +               if (ret < 0)
> +                       return ret;
> +       }
> +
>         do {
>                 read_seqbegin_or_lock(&vnode->cb_lock, &seq);
>                 generic_fillattr(&init_user_ns, inode, stat);
>
>


Tested-by: Markus Suvanto <[email protected]>

-Markus