Whamcloud - gitweb
- land b_ver_recov
authortappro <tappro>
Thu, 24 Jul 2008 11:38:27 +0000 (11:38 +0000)
committertappro <tappro>
Thu, 24 Jul 2008 11:38:27 +0000 (11:38 +0000)
lustre/ptlrpc/service.c

index 93c1eaf..3c14d25 100644 (file)
@@ -210,7 +210,7 @@ ptlrpc_schedule_difficult_reply (struct ptlrpc_reply_state *rs)
 }
 
 void
-ptlrpc_commit_replies (struct obd_device *obd)
+ptlrpc_commit_replies (struct obd_export *exp)
 {
         struct list_head   *tmp;
         struct list_head   *nxt;
@@ -219,15 +219,16 @@ ptlrpc_commit_replies (struct obd_device *obd)
          * to attend to complete them. */
 
         /* CAVEAT EMPTOR: spinlock ordering!!! */
-        spin_lock(&obd->obd_uncommitted_replies_lock);
+        spin_lock(&exp->exp_uncommitted_replies_lock);
 
-        list_for_each_safe (tmp, nxt, &obd->obd_uncommitted_replies) {
+        list_for_each_safe(tmp, nxt, &exp->exp_uncommitted_replies) {
                 struct ptlrpc_reply_state *rs =
                         list_entry(tmp, struct ptlrpc_reply_state, rs_obd_list);
 
-                LASSERT (rs->rs_difficult);
-
-                if (rs->rs_transno <= obd->obd_last_committed) {
+                LASSERT(rs->rs_difficult);
+                /* VBR: per-export last_committed */
+                LASSERT(rs->rs_export);
+                if (rs->rs_transno <= rs->rs_export->exp_last_committed) {
                         struct ptlrpc_service *svc = rs->rs_service;
 
                         spin_lock (&svc->srv_lock);
@@ -237,7 +238,7 @@ ptlrpc_commit_replies (struct obd_device *obd)
                 }
         }
 
-        spin_unlock(&obd->obd_uncommitted_replies_lock);
+        spin_unlock(&exp->exp_uncommitted_replies_lock);
 }
 
 static int
@@ -554,7 +555,7 @@ static void ptlrpc_update_export_timer(struct obd_export *exp, long extra_delay)
                                oldest_time);
                 }
         } else {
-                if (cfs_time_current_sec() > 
+                if (cfs_time_current_sec() >
                     (exp->exp_obd->obd_eviction_timer + extra_delay)) {
                         /* The evictor won't evict anyone who we've heard from
                          * recently, so we don't have to check before we start
@@ -569,7 +570,7 @@ static void ptlrpc_update_export_timer(struct obd_export *exp, long extra_delay)
 
 static int ptlrpc_check_req(struct ptlrpc_request *req)
 {
-        if (lustre_msg_get_conn_cnt(req->rq_reqmsg) < 
+        if (lustre_msg_get_conn_cnt(req->rq_reqmsg) <
             req->rq_export->exp_conn_cnt) {
                 DEBUG_REQ(D_ERROR, req,
                           "DROPPING req from old connection %d < %d",
@@ -602,11 +603,11 @@ static void ptlrpc_at_set_timer(struct ptlrpc_service *svc)
         }
 
         /* Set timer for closest deadline */
-        rq = list_entry(svc->srv_at_list.next, struct ptlrpc_request, 
+        rq = list_entry(svc->srv_at_list.next, struct ptlrpc_request,
                         rq_timed_list);
         next = (__s32)(rq->rq_deadline - cfs_time_current_sec() -
                        at_early_margin);
-        if (next <= 0) 
+        if (next <= 0)
                 ptlrpc_at_timer((unsigned long)svc);
         else
                 cfs_timer_arm(&svc->srv_at_timer, cfs_time_shift(next));
@@ -621,15 +622,15 @@ static int ptlrpc_at_add_timed(struct ptlrpc_request *req)
         struct ptlrpc_request *rq;
         int found = 0;
 
-        if (AT_OFF) 
+        if (AT_OFF)
                 return(0);
 
         if ((lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT) == 0)
                 return(-ENOSYS);
-        
-        DEBUG_REQ(D_ADAPTTO, req, "add timed %lds", 
+
+        DEBUG_REQ(D_ADAPTTO, req, "add timed %lds",
                   req->rq_deadline - cfs_time_current_sec());
-        
+
         spin_lock(&svc->srv_at_lock);
 
         if (unlikely(req->rq_sent_final)) {
@@ -660,7 +661,7 @@ static int ptlrpc_at_add_timed(struct ptlrpc_request *req)
                 ptlrpc_at_set_timer(svc);
 
         return 0;
-}            
+}
 
 static int ptlrpc_at_send_early_reply(struct ptlrpc_request *req, 
                                       int extra_time)
@@ -672,18 +673,18 @@ static int ptlrpc_at_send_early_reply(struct ptlrpc_request *req,
         time_t newdl;
         int rc;
         ENTRY;
-                            
-        /* deadline is when the client expects us to reply, margin is the 
+
+        /* deadline is when the client expects us to reply, margin is the
            difference between clients' and servers' expectations */
-        DEBUG_REQ(D_ADAPTTO, req, 
+        DEBUG_REQ(D_ADAPTTO, req,
                   "%ssending early reply (deadline %+lds, margin %+lds) for "
                   "%d+%d", AT_OFF ? "AT off - not " : "",
                   olddl, olddl - at_get(&svc->srv_at_estimate),
                   at_get(&svc->srv_at_estimate), extra_time);
 
-        if (AT_OFF) 
+        if (AT_OFF)
                 RETURN(0);
-        
+
         if (olddl < 0) {
                 CDEBUG(D_WARNING, "x"LPU64": Already past deadline (%+lds), not"
                        " sending early reply. Increase at_early_margin (%d)?\n",
@@ -724,7 +725,7 @@ static int ptlrpc_at_send_early_reply(struct ptlrpc_request *req,
                 OBD_FREE(reqcopy, sizeof *reqcopy);
                 RETURN(-ENOMEM);
         }
-        
+
         *reqcopy = *req;
         reqcopy->rq_reply_state = NULL;
         reqcopy->rq_rep_swab_mask = 0;
@@ -746,7 +747,7 @@ static int ptlrpc_at_send_early_reply(struct ptlrpc_request *req,
 
         /* RPC ref */
         class_export_rpc_get(reqcopy->rq_export);
-        if (reqcopy->rq_export->exp_obd && 
+        if (reqcopy->rq_export->exp_obd &&
             reqcopy->rq_export->exp_obd->obd_fail)
                 GOTO(out_put, rc = -ENODEV);
 
@@ -764,7 +765,7 @@ static int ptlrpc_at_send_early_reply(struct ptlrpc_request *req,
                 DEBUG_REQ(D_ERROR, req, "Early reply send failed %d", rc);
         }
 
-        /* Free the (early) reply state from lustre_pack_reply. 
+        /* Free the (early) reply state from lustre_pack_reply.
            (ptlrpc_send_reply takes it's own rs ref, so this is safe here) */
         ptlrpc_req_drop_rs(reqcopy);
 
@@ -795,10 +796,10 @@ static int ptlrpc_at_check_timed(struct ptlrpc_service *svc)
         }
         delay = cfs_time_sub(cfs_time_current(), svc->srv_at_checktime);
         svc->srv_at_check = 0;
-        
+
         if (list_empty(&svc->srv_at_list)) {
                 spin_unlock(&svc->srv_at_lock);
-                RETURN(0);      
+                RETURN(0);
         }
 
         /* The timer went off, but maybe the nearest rpc already completed. */
@@ -809,7 +810,7 @@ static int ptlrpc_at_check_timed(struct ptlrpc_service *svc)
                 /* We've still got plenty of time.  Reset the timer. */
                 spin_unlock(&svc->srv_at_lock);
                 ptlrpc_at_set_timer(svc);
-                RETURN(0);      
+                RETURN(0);
         }
 
         /* We're close to a timeout, and we don't know how much longer the 
@@ -831,7 +832,7 @@ static int ptlrpc_at_check_timed(struct ptlrpc_service *svc)
 
         CDEBUG(D_ADAPTTO, "timeout in %+ds, asking for %d secs on %d early "
                "replies\n", first, at_extra, counter);
-        
+
         if (first < 0) {
                 /* We're already past request deadlines before we even get a 
                    chance to send early replies */
@@ -929,18 +930,18 @@ ptlrpc_server_handle_req_in(struct ptlrpc_service *svc)
                 rc = ptlrpc_check_req(req);
                 class_export_put(req->rq_export);
                 req->rq_export = NULL;
-                if (rc) 
+                if (rc)
                         goto err_req;
         }
 
         /* req_in handling should/must be fast */
-        if (cfs_time_current_sec() - req->rq_arrival_time.tv_sec > 5) 
+        if (cfs_time_current_sec() - req->rq_arrival_time.tv_sec > 5)
                 DEBUG_REQ(D_WARNING, req, "Slow req_in handling %lus",
                           cfs_time_current_sec() - req->rq_arrival_time.tv_sec);
 
         /* Set rpc server deadline and add it to the timed list */
         deadline = (lustre_msghdr_get_flags(req->rq_reqmsg) &
-                    MSGHDR_AT_SUPPORT) ? 
+                    MSGHDR_AT_SUPPORT) ?
                    /* The max time the client expects us to take */
                    lustre_msg_get_timeout(req->rq_reqmsg) : obd_timeout;
         req->rq_deadline = req->rq_arrival_time.tv_sec + deadline;
@@ -948,7 +949,7 @@ ptlrpc_server_handle_req_in(struct ptlrpc_service *svc)
                 DEBUG_REQ(D_ERROR, req, "Dropping request with 0 timeout");
                 goto err_req;
         }
-        
+
         ptlrpc_at_add_timed(req);
 
         /* Move it over to the request processing queue */
@@ -990,7 +991,7 @@ ptlrpc_server_handle_request(struct ptlrpc_service *svc,
              svc->srv_n_difficult_replies != 0 &&
 #endif
              svc->srv_n_active_reqs >= (svc->srv_threads_running - 1))) {
-                /* Don't handle regular requests in the last thread, in order               * remain free to handle any 'difficult' replies (that might
+                /* Don't handle regular requests in the last thread, in order
                  * to handle difficult replies (which might block other threads)
                  * as well as handle any incoming reqs, early replies, etc. 
                  * That means we always need at least 2 service threads. */
@@ -998,8 +999,8 @@ ptlrpc_server_handle_request(struct ptlrpc_service *svc,
                 RETURN(0);
         }
 
-        request = list_entry (svc->srv_request_queue.next,
-                              struct ptlrpc_request, rq_list);
+        request = list_entry(svc->srv_request_queue.next,
+                             struct ptlrpc_request, rq_list);
         list_del_init (&request->rq_list);
         svc->srv_n_queued_reqs--;
         svc->srv_n_active_reqs++;
@@ -1018,9 +1019,9 @@ ptlrpc_server_handle_request(struct ptlrpc_service *svc,
                 lprocfs_counter_add(svc->srv_stats, PTLRPC_TIMEOUT,
                                     at_get(&svc->srv_at_estimate));
         }
-        
+
         CDEBUG(D_NET, "got req "LPD64"\n", request->rq_xid);
-        
+
         request->rq_svc_thread = thread;
         request->rq_export = class_conn2export(
                                      lustre_msg_get_handle(request->rq_reqmsg));
@@ -1032,7 +1033,7 @@ ptlrpc_server_handle_request(struct ptlrpc_service *svc,
                 export = class_export_rpc_get(request->rq_export);
         }
 
-        /* Discard requests queued for longer than the deadline.  
+        /* Discard requests queued for longer than the deadline.
            The deadline is increased if we send an early reply. */
         if (cfs_time_current_sec() > request->rq_deadline) {
                 DEBUG_REQ(D_ERROR, request, "Dropping timed-out request from %s"
@@ -1059,7 +1060,7 @@ ptlrpc_server_handle_request(struct ptlrpc_service *svc,
         OBD_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_PAUSE_REQ, obd_fail_val);
 
         rc = svc->srv_handler(request);
-        
+
         request->rq_phase = RQ_PHASE_COMPLETE;
 
         CDEBUG(D_RPCTRACE, "Handled RPC pname:cluuid+ref:pid:xid:nid:opc "
@@ -1097,7 +1098,7 @@ put_conn:
                cfs_timeval_sub(&work_end, &request->rq_arrival_time, NULL),
                request->rq_repmsg ? lustre_msg_get_transno(request->rq_repmsg) :
                request->rq_transno, request->rq_status,
-               request->rq_repmsg ? lustre_msg_get_status(request->rq_repmsg): 
+               request->rq_repmsg ? lustre_msg_get_status(request->rq_repmsg):
                -999);
         if (svc->srv_stats != NULL) {
                 __u32 op = lustre_msg_get_opc(request->rq_reqmsg);
@@ -1110,7 +1111,7 @@ put_conn:
                 }
         }
         if (request->rq_early_count) {
-                DEBUG_REQ(D_ADAPTTO, request, 
+                DEBUG_REQ(D_ADAPTTO, request,
                           "sent %d early replies before finishing in %lds",
                           request->rq_early_count,
                           work_end.tv_sec - request->rq_arrival_time.tv_sec);
@@ -1151,10 +1152,10 @@ ptlrpc_server_handle_reply (struct ptlrpc_service *svc)
         /* Disengage from notifiers carefully (lock order - irqrestore below!)*/
         spin_unlock(&svc->srv_lock);
 
-        spin_lock (&obd->obd_uncommitted_replies_lock);
+        spin_lock (&exp->exp_uncommitted_replies_lock);
         /* Noop if removed already */
         list_del_init (&rs->rs_obd_list);
-        spin_unlock (&obd->obd_uncommitted_replies_lock);
+        spin_unlock (&exp->exp_uncommitted_replies_lock);
 
         spin_lock (&exp->exp_lock);
         /* Noop if removed already */
@@ -1173,9 +1174,7 @@ ptlrpc_server_handle_reply (struct ptlrpc_service *svc)
                 /* If we see this, we should already have seen the warning
                  * in mds_steal_ack_locks()  */
                 CWARN("All locks stolen from rs %p x"LPD64".t"LPD64
-                      " o%d NID %s\n",
-                      rs,
-                      rs->rs_xid, rs->rs_transno,
+                      " o%d NID %s\n", rs, rs->rs_xid, rs->rs_transno,
                       lustre_msg_get_opc(rs->rs_msg),
                       libcfs_nid2str(exp->exp_connection->c_peer.nid));
         }
@@ -1406,8 +1405,8 @@ static int ptlrpc_main(void *arg)
                               &lwi);
 
                 lc_watchdog_touch_ms(watchdog, max_t(int, obd_timeout,
-                                     AT_OFF ? 0 : 
-                                     at_get(&svc->srv_at_estimate)) * 
+                                     AT_OFF ? 0 :
+                                     at_get(&svc->srv_at_estimate)) *
                                      svc->srv_watchdog_factor);
 
                 ptlrpc_check_rqbd_pool(svc);
@@ -1424,13 +1423,13 @@ static int ptlrpc_main(void *arg)
                 if (!list_empty(&svc->srv_req_in_queue)) {
                         /* Process all incoming reqs before handling any */
                         ptlrpc_server_handle_req_in(svc);
-                        /* but limit ourselves in case of flood */ 
+                        /* but limit ourselves in case of flood */
                         if (counter++ < 1000)
                                 continue;
                         counter = 0;
                 }
 
-                if (svc->srv_at_check) 
+                if (svc->srv_at_check)
                         ptlrpc_at_check_timed(svc);
 
                 /* don't handle requests in the last thread */
@@ -1516,6 +1515,7 @@ int ptlrpc_start_threads(struct obd_device *dev, struct ptlrpc_service *svc)
 
         /* We require 2 threads min - see note in
          * ptlrpc_server_handle_request() */
+
         LASSERT(svc->srv_threads_min >= 2);
         for (i = 0; i < svc->srv_threads_min; i++) {
                 rc = ptlrpc_start_thread(dev, svc);
@@ -1571,7 +1571,7 @@ int ptlrpc_start_thread(struct obd_device *dev, struct ptlrpc_service *svc)
         d.thread = thread;
 
         CDEBUG(D_RPCTRACE, "starting thread '%s'\n", name);
-        
+
         /* CLONE_VM and CLONE_FILES just avoid a needless copy, because we
          * just drop the VM and FILES in ptlrpc_daemonize() right away.
          */
@@ -1626,7 +1626,7 @@ int ptlrpc_unregister_service(struct ptlrpc_service *service)
          * its 'unlink' flag set for each posted rqbd */
         list_for_each(tmp, &service->srv_active_rqbds) {
                 struct ptlrpc_request_buffer_desc *rqbd =
-                        list_entry(tmp, struct ptlrpc_request_buffer_desc, 
+                        list_entry(tmp, struct ptlrpc_request_buffer_desc,
                                    rqbd_list);
 
                 rc = LNetMDUnlink(rqbd->rqbd_md_h);
@@ -1754,7 +1754,7 @@ int ptlrpc_service_health_check(struct ptlrpc_service *svc)
                 spin_unlock(&svc->srv_lock);
                 return 0;
         }
-        
+
         /* How long has the next entry been waiting? */
         request = list_entry(svc->srv_request_queue.next,
                              struct ptlrpc_request, rq_list);