NFS: Fix an ABBA spinlock issue in pnfs_update_layout()
authorTrond Myklebust <trond.myklebust@hammerspace.com>
Mon, 13 Apr 2020 19:55:21 +0000 (15:55 -0400)
committerTrond Myklebust <trond.myklebust@hammerspace.com>
Mon, 13 Apr 2020 19:55:21 +0000 (15:55 -0400)
We need to drop the inode spinlock while calling nfs4_select_rw_stateid(),
since nfs4_copy_delegation_stateid() could take the delegation lock.
Note that it is safe to do this, since all other calls to
pnfs_update_layout() for that inode will find themselves blocked by
the lock we hold on NFS_LAYOUT_FIRST_LAYOUTGET.

Fixes: fc51b1cf391d ("NFS: Beware when dereferencing the delegation cred")
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
fs/nfs/pnfs.c

index f2dc35c22964c3da42da19c2ff954e7cdd7274f3..b8d78f3933651a2c042bf92ff3bebed471d6c69e 100644 (file)
@@ -2023,6 +2023,7 @@ lookup_again:
                        goto lookup_again;
                }
 
+               spin_unlock(&ino->i_lock);
                first = true;
                status = nfs4_select_rw_stateid(ctx->state,
                                        iomode == IOMODE_RW ? FMODE_WRITE : FMODE_READ,
@@ -2032,12 +2033,12 @@ lookup_again:
                        trace_pnfs_update_layout(ino, pos, count,
                                        iomode, lo, lseg,
                                        PNFS_UPDATE_LAYOUT_INVALID_OPEN);
-                       spin_unlock(&ino->i_lock);
                        nfs4_schedule_stateid_recovery(server, ctx->state);
                        pnfs_clear_first_layoutget(lo);
                        pnfs_put_layout_hdr(lo);
                        goto lookup_again;
                }
+               spin_lock(&ino->i_lock);
        } else {
                nfs4_stateid_copy(&stateid, &lo->plh_stateid);
        }