NFSv4/pnfs: Fix an infinite layoutget loop

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>

1 files changed, 1 insertions, 2 deletions

diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 6c61e2b99635..f90090e8c959 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -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;