openkylin
/
linux
mirror of https://gitee.com/openkylin/linux.git
9
0
Fork 0
Code Issues Projects Releases Wiki Activity

staging/lustre/ptlrpc: race at req processing 

Race between ptlrpc_resend_req() and ptlrpc_check_set().
1 thread do ptlrpc_check_set()->after_reply()
2 thread do ptlrpc_resend_req()
The result is request with rq_resend = 1 and MSG_REPLY flag.
When this request will came to server it will cause client eviction.
The patch skip ptlrpc_resend_req logic if rq_replied is set,
and clear rq_resend flag at reply_in_callback() when client got
reply.

Signed-off-by: Alexander Boyko <alexander_boyko@xyratex.com>
Xyratex-bug-id: MRP-1888
Reviewed-on: http://review.whamcloud.com/10471
Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-5116
Reviewed-by: Andreas Dilger <andreas.dilger@intel.com>
Reviewed-by: Mike Pershin <mike.pershin@intel.com>
Reviewed-by: Chris Horn <hornc@cray.com>
Signed-off-by: Oleg Drokin <oleg.drokin@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This commit is contained in:
Alexander.Boyko 2014-06-22 21:32:12 -04:00 committed by Greg Kroah-Hartman
parent a2ff0f973c
commit 5c689e689b
3 changed files with 14 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
drivers/staging/lustre/lustre/ptlrpc/client.c
View File

@ -2530,10 +2530,19 @@ EXPORT_SYMBOL(ptlrpc_cleanup_client);
void ptlrpc_resend_req(struct ptlrpc_request *req)
{
DEBUG_REQ(D_HA, req, "going to resend");
spin_lock(&req->rq_lock);
/* Request got reply but linked to the import list still.
Let ptlrpc_check_set() to process it. */
if (ptlrpc_client_replied(req)) {
spin_unlock(&req->rq_lock);
DEBUG_REQ(D_HA, req, "it has reply, so skip it");
return;
}
lustre_msg_set_handle(req->rq_reqmsg, &(struct lustre_handle){ 0 });
req->rq_status = -EAGAIN;
spin_lock(&req->rq_lock);
req->rq_resend = 1;
req->rq_net_err = 0;
req->rq_timedout = 0;

2
drivers/staging/lustre/lustre/ptlrpc/events.c
View File

@ -145,6 +145,8 @@ void reply_in_callback(lnet_event_t *ev)
/* Real reply */
req->rq_rep_swab_mask = 0;
req->rq_replied = 1;
/* Got reply, no resend required */
req->rq_resend = 0;
req->rq_reply_off = ev->offset;
req->rq_nob_received = ev->mlength;
/* LNetMDUnlink can't be called under the LNET_LOCK,

2
drivers/staging/lustre/lustre/ptlrpc/niobuf.c
View File

@ -505,6 +505,8 @@ int ptl_send_rpc(struct ptlrpc_request *request, int noreply)
/* If this is a re-transmit, we're required to have disengaged
* cleanly from the previous attempt */
LASSERT(!request->rq_receiving_reply);
LASSERT(!((lustre_msg_get_flags(request->rq_reqmsg) & MSG_REPLAY) &&
(request->rq_import->imp_state == LUSTRE_IMP_FULL)));
if (unlikely(obd != NULL && obd->obd_fail)) {
CDEBUG(D_HA, "muting rpc for failed imp obd %s\n",