NFSv4.1: LAYOUTGET EDELAY loops timeout to the MDS
The client will currently try LAYOUTGETs forever if a server is returning
NFS4ERR_LAYOUTTRYLATER or NFS4ERR_RECALLCONFLICT - even if the client no
longer needs the layout (ie process killed, unmounted).
This patch uses the DS timeout value (module parameter 'dataserver_timeo'
via rpc layer) to set an upper limit of how long the client tries LATOUTGETs
in this situation. Once the timeout is reached, IO is redirected to the MDS.
This also changes how the client checks if a layout is on the clp list
to avoid a double list_add.
Signed-off-by: Weston Andros Adamson <dros@netapp.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 37eb385..b2671cb 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -93,6 +93,8 @@
return err;
switch (err) {
case -NFS4ERR_RESOURCE:
+ case -NFS4ERR_LAYOUTTRYLATER:
+ case -NFS4ERR_RECALLCONFLICT:
return -EREMOTEIO;
case -NFS4ERR_WRONGSEC:
return -EPERM;
@@ -6046,6 +6048,7 @@
struct nfs_server *server = NFS_SERVER(inode);
struct pnfs_layout_hdr *lo;
struct nfs4_state *state = NULL;
+ unsigned long timeo, giveup;
dprintk("--> %s\n", __func__);
@@ -6057,7 +6060,10 @@
goto out;
case -NFS4ERR_LAYOUTTRYLATER:
case -NFS4ERR_RECALLCONFLICT:
- task->tk_status = -NFS4ERR_DELAY;
+ timeo = rpc_get_timeout(task->tk_client);
+ giveup = lgp->args.timestamp + timeo;
+ if (time_after(giveup, jiffies))
+ task->tk_status = -NFS4ERR_DELAY;
break;
case -NFS4ERR_EXPIRED:
case -NFS4ERR_BAD_STATEID:
@@ -6178,6 +6184,7 @@
return ERR_PTR(-ENOMEM);
}
lgp->args.layout.pglen = max_pages * PAGE_SIZE;
+ lgp->args.timestamp = jiffies;
lgp->res.layoutp = &lgp->args.layout;
lgp->res.seq_res.sr_slot = NULL;