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>

diff --git a/drivers/staging/lustre/lustre/ptlrpc/client.c b/drivers/staging/lustre/lustre/ptlrpc/client.c
index 7246e8ce9c1965d0e4daba8a753aca761fc20eab..d806257f31ee0f06ffda8c17fc7b1cf3c64fe0af 100644 (file)
--- 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 aa85239f6cd5a7dffa422114ff574bd2c69e2222..9f9b8d1f835eeb6b459c4134bfd2b13be5fa9cef 100644 (file)
--- 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 ef18639036c5887dfc4c8e3007570cc800c2cc31..f760504e0696acc4c3398e01cb4f8e999f7636ff 100644 (file)
--- 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",