NFS: avoid infinite loop in pnfs_update_layout.
authorNeilBrown <neilb@suse.de>
Wed, 28 Feb 2024 00:24:53 +0000 (11:24 +1100)
committerTrond Myklebust <trond.myklebust@hammerspace.com>
Sat, 9 Mar 2024 14:14:51 +0000 (09:14 -0500)
If pnfsd_update_layout() is called on a file for which recovery has
failed it will enter a tight infinite loop.

NFS_LAYOUT_INVALID_STID will be set, nfs4_select_rw_stateid() will
return -EIO, and nfs4_schedule_stateid_recovery() will do nothing, so
nfs4_client_recover_expired_lease() will not wait.  So the code will
loop indefinitely.

Break the loop by testing the validity of the open stateid at the top of
the loop.

Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
fs/nfs/pnfs.c

index 0c0fed1ecd0bf0d0938dec7b7c8cdca30e65954c..a5cc6199127f5876faae1c27e36c128d9eab9996 100644 (file)
@@ -1999,6 +1999,14 @@ pnfs_update_layout(struct inode *ino,
        }
 
 lookup_again:
+       if (!nfs4_valid_open_stateid(ctx->state)) {
+               trace_pnfs_update_layout(ino, pos, count,
+                                        iomode, lo, lseg,
+                                        PNFS_UPDATE_LAYOUT_INVALID_OPEN);
+               lseg = ERR_PTR(-EIO);
+               goto out;
+       }
+
        lseg = ERR_PTR(nfs4_client_recover_expired_lease(clp));
        if (IS_ERR(lseg))
                goto out;