NFSv4/pnfs: Fix an infinite layoutget loop
authorTrond Myklebust <trond.myklebust@primarydata.com>
Wed, 4 Oct 2017 17:49:12 +0000 (13:49 -0400)
committerTrond Myklebust <trond.myklebust@primarydata.com>
Wed, 4 Oct 2017 18:06:54 +0000 (14:06 -0400)
Since we can now use a lock stateid or a delegation stateid, that
differs from the context stateid, we need to change the test in
nfs4_layoutget_handle_exception() to take this into account.

This fixes an infinite layoutget loop in the NFS client whereby
it keeps retrying the initial layoutget using the same broken
stateid.

Fixes: 70d2f7b1ea19b ("pNFS: Use the standard I/O stateid when...")
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
fs/nfs/nfs4proc.c

index 6c61e2b996351cde05b1c922674cd3a2670913ea..f90090e8c959b14346baf76ac7eaefed7ae30f82 100644 (file)
@@ -8399,8 +8399,7 @@ nfs4_layoutget_handle_exception(struct rpc_task *task,
                lo = NFS_I(inode)->layout;
                /* If the open stateid was bad, then recover it. */
                if (!lo || test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags) ||
-                   nfs4_stateid_match_other(&lgp->args.stateid,
-                                       &lgp->args.ctx->state->stateid)) {
+                   !nfs4_stateid_match_other(&lgp->args.stateid, &lo->plh_stateid)) {
                        spin_unlock(&inode->i_lock);
                        exception->state = lgp->args.ctx->state;
                        exception->stateid = &lgp->args.stateid;