diff options
| author | Trond Myklebust <trond.myklebust@hammerspace.com> | 2022-05-31 11:03:07 -0400 |
|---|---|---|
| committer | Anna Schumaker <Anna.Schumaker@Netapp.com> | 2022-06-06 11:53:55 -0400 |
| commit | 880265c77ac415090090d1fe72a188fee71cb458 (patch) | |
| tree | 1a3b589cd95c81472c781b27f4804641571ef787 /fs/nfs/pnfs.h | |
| parent | fe44fb23d6ccde4c914c44ef74ab8d9d9ba02bea (diff) | |
pNFS: Avoid a live lock condition in pnfs_update_layout()
If we're about to send the first layoutget for an empty layout, we want
to make sure that we drain out the existing pending layoutget calls
first. The reason is that these layouts may have been already implicitly
returned to the server by a recall to which the client gave a
NFS4ERR_NOMATCHING_LAYOUT response.
The problem is that wait_var_event_killable() could in principle see the
plh_outstanding count go back to '1' when the first process to wake up
starts sending a new layoutget. If it fails to get a layout, then this
loop can continue ad infinitum...
Fixes: 0b77f97a7e42 ("NFSv4/pnfs: Fix layoutget behaviour after invalidation")
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Diffstat (limited to 'fs/nfs/pnfs.h')
| -rw-r--r-- | fs/nfs/pnfs.h | 1 |
1 files changed, 1 insertions, 0 deletions
diff --git a/fs/nfs/pnfs.h b/fs/nfs/pnfs.h index 07f11489e4e9..f331f067691b 100644 --- a/fs/nfs/pnfs.h +++ b/fs/nfs/pnfs.h @@ -105,6 +105,7 @@ enum { NFS_LAYOUT_FIRST_LAYOUTGET, /* Serialize first layoutget */ NFS_LAYOUT_INODE_FREEING, /* The inode is being freed */ NFS_LAYOUT_HASHED, /* The layout visible */ + NFS_LAYOUT_DRAIN, }; enum layoutdriver_policy_flags { |