From: Alexander.Boyko Date: Mon, 23 Jun 2014 01:32:12 +0000 (-0400) Subject: staging/lustre/ptlrpc: race at req processing X-Git-Tag: firefly_0821_release~176^2~3491^2~1355 X-Git-Url: http://demsky.eecs.uci.edu/git/?a=commitdiff_plain;h=5c689e689baa5a8a239a8b5d29c3fb77aa62f134;p=firefly-linux-kernel-4.4.55.git 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 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 Reviewed-by: Mike Pershin Reviewed-by: Chris Horn Signed-off-by: Oleg Drokin Signed-off-by: Greg Kroah-Hartman --- diff --git a/drivers/staging/lustre/lustre/ptlrpc/client.c b/drivers/staging/lustre/lustre/ptlrpc/client.c index 7246e8ce9c19..d806257f31ee 100644 --- a/drivers/staging/lustre/lustre/ptlrpc/client.c +++ b/drivers/staging/lustre/lustre/ptlrpc/client.c @@ -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; diff --git a/drivers/staging/lustre/lustre/ptlrpc/events.c b/drivers/staging/lustre/lustre/ptlrpc/events.c index aa85239f6cd5..9f9b8d1f835e 100644 --- a/drivers/staging/lustre/lustre/ptlrpc/events.c +++ b/drivers/staging/lustre/lustre/ptlrpc/events.c @@ -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, diff --git a/drivers/staging/lustre/lustre/ptlrpc/niobuf.c b/drivers/staging/lustre/lustre/ptlrpc/niobuf.c index ef18639036c5..f760504e0696 100644 --- a/drivers/staging/lustre/lustre/ptlrpc/niobuf.c +++ b/drivers/staging/lustre/lustre/ptlrpc/niobuf.c @@ -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",