Whamcloud - gitweb
b=16909 Console message cleanup
[fs/lustre-release.git] / lustre / ptlrpc / events.c
index d2693fe..69716b3 100644 (file)
@@ -1,26 +1,37 @@
 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
  * vim:expandtab:shiftwidth=8:tabstop=8:
  *
- *  Copyright (c) 2002, 2003 Cluster File Systems, Inc.
+ * GPL HEADER START
  *
- *   This file is part of the Lustre file system, http://www.lustre.org
- *   Lustre is a trademark of Cluster File Systems, Inc.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
- *   You may have signed or agreed to another license before downloading
- *   this software.  If so, you are bound by the terms and conditions
- *   of that agreement, and the following does not apply to you.  See the
- *   LICENSE file included with this distribution for more information.
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
  *
- *   If you did not agree to a different license, then this copy of Lustre
- *   is open source software; you can redistribute it and/or modify it
- *   under the terms of version 2 of the GNU General Public License as
- *   published by the Free Software Foundation.
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
  *
- *   In either case, Lustre is distributed in the hope that it will be
- *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
- *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- *   license text for more details.
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see
+ * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  *
+ * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
+ * CA 95054 USA or visit www.sun.com if you need additional information or
+ * have any questions.
+ *
+ * GPL HEADER END
+ */
+/*
+ * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Use is subject to license terms.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
  */
 
 #define DEBUG_SUBSYSTEM S_RPC
 #ifndef __KERNEL__
 # include <liblustre.h>
 #else
+# include <libcfs/libcfs.h>
 # ifdef __mips64__
 #  include <linux/kernel.h>
 # endif
 #endif
+
 #include <obd_class.h>
 #include <lustre_net.h>
 #include <lustre_sec.h>
@@ -52,25 +65,23 @@ void request_out_callback(lnet_event_t *ev)
                  ev->type == LNET_EVENT_UNLINK);
         LASSERT (ev->unlinked);
 
-        DEBUG_REQ((ev->status == 0) ? D_NET : D_ERROR, req,
-                  "type %d, status %d", ev->type, ev->status);
+        DEBUG_REQ(D_NET, req, "type %d, status %d", ev->type, ev->status);
 
         sptlrpc_request_out_callback(req);
+        req->rq_real_sent = cfs_time_current_sec();
 
         if (ev->type == LNET_EVENT_UNLINK || ev->status != 0) {
 
                 /* Failed send: make it seem like the reply timed out, just
                  * like failing sends in client.c does currently...  */
 
-                spin_lock(&req->rq_lock);
+                cfs_spin_lock(&req->rq_lock);
                 req->rq_net_err = 1;
-                spin_unlock(&req->rq_lock);
+                cfs_spin_unlock(&req->rq_lock);
 
-                ptlrpc_wake_client_req(req);
+                ptlrpc_client_wake_req(req);
         }
 
-        /* these balance the references in ptl_send_rpc() */
-        atomic_dec(&req->rq_import->imp_inflight);
         ptlrpc_req_finished(req);
 
         EXIT;
@@ -85,31 +96,80 @@ void reply_in_callback(lnet_event_t *ev)
         struct ptlrpc_request *req = cbid->cbid_arg;
         ENTRY;
 
-        LASSERT (ev->type == LNET_EVENT_PUT ||
-                 ev->type == LNET_EVENT_UNLINK);
-        LASSERT (ev->unlinked);
-        LASSERT (ev->md.start == req->rq_repbuf);
-        LASSERT (ev->offset == 0);
-        LASSERT (ev->mlength <= req->rq_repbuf_len);
+        DEBUG_REQ(D_NET, req, "type %d, status %d", ev->type, ev->status);
 
-        DEBUG_REQ((ev->status == 0) ? D_NET : D_ERROR, req,
-                  "type %d, status %d", ev->type, ev->status);
+        LASSERT (ev->type == LNET_EVENT_PUT || ev->type == LNET_EVENT_UNLINK);
+        LASSERT (ev->md.start == req->rq_repbuf);
+        LASSERT (ev->offset + ev->mlength <= req->rq_repbuf_len);
+        /* We've set LNET_MD_MANAGE_REMOTE for all outgoing requests
+           for adaptive timeouts' early reply. */
+        LASSERT((ev->md.options & LNET_MD_MANAGE_REMOTE) != 0);
 
-        spin_lock(&req->rq_lock);
+        cfs_spin_lock(&req->rq_lock);
 
-        LASSERT (req->rq_receiving_reply);
         req->rq_receiving_reply = 0;
+        req->rq_early = 0;
+        if (ev->unlinked)
+                req->rq_must_unlink = 0;
+
+        if (ev->status)
+                goto out_wake;
 
-        if (ev->type == LNET_EVENT_PUT && ev->status == 0) {
+        if (ev->type == LNET_EVENT_UNLINK) {
+                LASSERT(ev->unlinked);
+                DEBUG_REQ(D_RPCTRACE, req, "unlink");
+                goto out_wake;
+        }
+
+        if (ev->mlength < ev->rlength ) {
+                CDEBUG(D_RPCTRACE, "truncate req %p rpc %d - %d+%d\n", req,
+                       req->rq_replen, ev->rlength, ev->offset);
+                req->rq_reply_truncate = 1;
                 req->rq_replied = 1;
+                req->rq_status = -EOVERFLOW;
+                req->rq_nob_received = ev->rlength + ev->offset;
+                goto out_wake;
+        }
+
+        if ((ev->offset == 0) &&
+            ((lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT))) {
+                /* Early reply */
+                DEBUG_REQ(D_ADAPTTO, req,
+                          "Early reply received: mlen=%u offset=%d replen=%d "
+                          "replied=%d unlinked=%d", ev->mlength, ev->offset,
+                          req->rq_replen, req->rq_replied, ev->unlinked);
+
+                req->rq_early_count++; /* number received, client side */
+
+                if (req->rq_replied)   /* already got the real reply */
+                        goto out_wake;
+
+                req->rq_early = 1;
+                req->rq_reply_off = ev->offset;
+                req->rq_nob_received = ev->mlength;
+                /* And we're still receiving */
+                req->rq_receiving_reply = 1;
+        } else {
+                /* Real reply */
+                req->rq_rep_swab_mask = 0;
+                req->rq_replied = 1;
+                req->rq_reply_off = ev->offset;
                 req->rq_nob_received = ev->mlength;
+                /* LNetMDUnlink can't be called under the LNET_LOCK,
+                   so we must unlink in ptlrpc_unregister_reply */
+                DEBUG_REQ(D_INFO, req,
+                          "reply in flags=%x mlen=%u offset=%d replen=%d",
+                          lustre_msg_get_flags(req->rq_reqmsg),
+                          ev->mlength, ev->offset, req->rq_replen);
         }
 
+        req->rq_import->imp_last_reply_time = cfs_time_current_sec();
+
+out_wake:
         /* NB don't unlock till after wakeup; req can disappear under us
          * since we don't have our own ref */
-        ptlrpc_wake_client_req(req);
-
-        spin_unlock(&req->rq_lock);
+        ptlrpc_client_wake_req(req);
+        cfs_spin_unlock(&req->rq_lock);
         EXIT;
 }
 
@@ -133,7 +193,7 @@ void client_bulk_callback (lnet_event_t *ev)
                "event type %d, status %d, desc %p\n",
                ev->type, ev->status, desc);
 
-        spin_lock(&desc->bd_lock);
+        cfs_spin_lock(&desc->bd_lock);
 
         LASSERT(desc->bd_network_rw);
         desc->bd_network_rw = 0;
@@ -144,13 +204,15 @@ void client_bulk_callback (lnet_event_t *ev)
                 desc->bd_sender = ev->sender;
         }
 
-        sptlrpc_enc_pool_put_pages(desc);
+        /* release the encrypted pages for write */
+        if (desc->bd_req->rq_bulk_write)
+                sptlrpc_enc_pool_put_pages(desc);
 
         /* NB don't unlock till after wakeup; desc can disappear under us
          * otherwise */
-        ptlrpc_wake_client_req(desc->bd_req);
+        ptlrpc_client_wake_req(desc->bd_req);
 
-        spin_unlock(&desc->bd_lock);
+        cfs_spin_unlock(&desc->bd_lock);
         EXIT;
 }
 
@@ -206,7 +268,7 @@ void request_in_callback(lnet_event_t *ev)
         req->rq_reqbuf = ev->md.start + ev->offset;
         if (ev->type == LNET_EVENT_PUT && ev->status == 0)
                 req->rq_reqdata_len = ev->mlength;
-        do_gettimeofday(&req->rq_arrival_time);
+        cfs_gettimeofday(&req->rq_arrival_time);
         req->rq_peer = ev->initiator;
         req->rq_self = ev->target.nid;
         req->rq_rqbd = rqbd;
@@ -214,17 +276,23 @@ void request_in_callback(lnet_event_t *ev)
 #ifdef CRAY_XT3
         req->rq_uid = ev->uid;
 #endif
+        cfs_spin_lock_init(&req->rq_lock);
+        CFS_INIT_LIST_HEAD(&req->rq_timed_list);
+        cfs_atomic_set(&req->rq_refcount, 1);
+        if (ev->type == LNET_EVENT_PUT)
+                CDEBUG(D_RPCTRACE, "incoming req@%p x"LPU64" msgsize %u\n",
+                       req, req->rq_xid, ev->mlength);
 
         CDEBUG(D_RPCTRACE, "peer: %s\n", libcfs_id2str(req->rq_peer));
 
-        spin_lock(&service->srv_lock);
+        cfs_spin_lock(&service->srv_lock);
 
         req->rq_history_seq = service->srv_request_seq++;
-        list_add_tail(&req->rq_history_list, &service->srv_request_history);
+        cfs_list_add_tail(&req->rq_history_list, &service->srv_request_history);
 
         if (ev->unlinked) {
                 service->srv_nrqbd_receiving--;
-                CDEBUG(D_RPCTRACE,"Buffer complete: %d buffers still posted\n",
+                CDEBUG(D_INFO, "Buffer complete: %d buffers still posted\n",
                        service->srv_nrqbd_receiving);
 
                 /* Normally, don't complain about 0 buffers posted; LNET won't
@@ -241,14 +309,14 @@ void request_in_callback(lnet_event_t *ev)
                 rqbd->rqbd_refcount++;
         }
 
-        list_add_tail(&req->rq_list, &service->srv_request_queue);
+        cfs_list_add_tail(&req->rq_list, &service->srv_req_in_queue);
         service->srv_n_queued_reqs++;
 
         /* NB everything can disappear under us once the request
          * has been queued and we unlock, so do the wake now... */
         cfs_waitq_signal(&service->srv_waitq);
 
-        spin_unlock(&service->srv_lock);
+        cfs_spin_unlock(&service->srv_lock);
         EXIT;
 }
 
@@ -271,7 +339,7 @@ void reply_out_callback(lnet_event_t *ev)
                  * net's ref on 'rs' */
                 LASSERT (ev->unlinked);
                 ptlrpc_rs_decref(rs);
-                atomic_dec (&svc->srv_outstanding_replies);
+                cfs_atomic_dec (&svc->srv_outstanding_replies);
                 EXIT;
                 return;
         }
@@ -279,12 +347,16 @@ void reply_out_callback(lnet_event_t *ev)
         LASSERT (rs->rs_on_net);
 
         if (ev->unlinked) {
-                /* Last network callback.  The net's ref on 'rs' stays put
-                 * until ptlrpc_server_handle_reply() is done with it */
-                spin_lock(&svc->srv_lock);
+                /* Last network callback. The net's ref on 'rs' stays put
+                 * until ptlrpc_handle_rs() is done with it */
+                cfs_spin_lock(&svc->srv_lock);
+                cfs_spin_lock(&rs->rs_lock);
                 rs->rs_on_net = 0;
-                ptlrpc_schedule_difficult_reply (rs);
-                spin_unlock(&svc->srv_lock);
+                if (!rs->rs_no_ack ||
+                    rs->rs_transno <= rs->rs_export->exp_obd->obd_last_committed)
+                        ptlrpc_schedule_difficult_reply (rs);
+                cfs_spin_unlock(&rs->rs_lock);
+                cfs_spin_unlock(&svc->srv_lock);
         }
 
         EXIT;
@@ -310,7 +382,7 @@ void server_bulk_callback (lnet_event_t *ev)
                "event type %d, status %d, desc %p\n",
                ev->type, ev->status, desc);
 
-        spin_lock(&desc->bd_lock);
+        cfs_spin_lock(&desc->bd_lock);
 
         if ((ev->type == LNET_EVENT_ACK ||
              ev->type == LNET_EVENT_REPLY) &&
@@ -329,7 +401,7 @@ void server_bulk_callback (lnet_event_t *ev)
                 cfs_waitq_signal(&desc->bd_waitq);
         }
 
-        spin_unlock(&desc->bd_lock);
+        cfs_spin_unlock(&desc->bd_lock);
         EXIT;
 }
 
@@ -354,12 +426,12 @@ int ptlrpc_uuid_to_peer (struct obd_uuid *uuid,
                          lnet_process_id_t *peer, lnet_nid_t *self)
 {
         int               best_dist = 0;
-        int               best_order = 0;
+        __u32             best_order = 0;
         int               count = 0;
         int               rc = -ENOENT;
         int               portals_compatibility;
         int               dist;
-        int               order;
+        __u32             order;
         lnet_nid_t        dst_nid;
         lnet_nid_t        src_nid;
 
@@ -379,7 +451,6 @@ int ptlrpc_uuid_to_peer (struct obd_uuid *uuid,
                         break;
                 }
 
-                LASSERT (order >= 0);
                 if (rc < 0 ||
                     dist < best_dist ||
                     (dist == best_dist && order < best_order)) {
@@ -401,8 +472,6 @@ int ptlrpc_uuid_to_peer (struct obd_uuid *uuid,
         }
 
         CDEBUG(D_NET,"%s->%s\n", uuid->uuid, libcfs_id2str(*peer));
-        if (rc != 0)
-                CERROR("No NID found for %s\n", uuid->uuid);
         return rc;
 }
 
@@ -497,7 +566,7 @@ CFS_LIST_HEAD(liblustre_idle_callbacks);
 void *liblustre_services_callback;
 
 void *
-liblustre_register_waitidle_callback (struct list_head *callback_list,
+liblustre_register_waitidle_callback (cfs_list_t *callback_list,
                                       const char *name,
                                       int (*fn)(void *arg), void *arg)
 {
@@ -509,7 +578,7 @@ liblustre_register_waitidle_callback (struct list_head *callback_list,
         llwc->llwc_name = name;
         llwc->llwc_fn = fn;
         llwc->llwc_arg = arg;
-        list_add_tail(&llwc->llwc_list, callback_list);
+        cfs_list_add_tail(&llwc->llwc_list, callback_list);
 
         return (llwc);
 }
@@ -519,7 +588,7 @@ liblustre_deregister_waitidle_callback (void *opaque)
 {
         struct liblustre_wait_callback *llwc = opaque;
 
-        list_del(&llwc->llwc_list);
+        cfs_list_del(&llwc->llwc_list);
         OBD_FREE(llwc, sizeof(*llwc));
 }
 
@@ -565,7 +634,7 @@ liblustre_check_events (int timeout)
 
         LASSERT (rc == -EOVERFLOW || rc == 1);
 
-        /* liblustre: no asynch callback so we can't affort to miss any
+        /* liblustre: no asynch callback so we can't afford to miss any
          * events... */
         if (rc == -EOVERFLOW) {
                 CERROR ("Dropped an event!!!\n");
@@ -581,7 +650,7 @@ int liblustre_waiting = 0;
 int
 liblustre_wait_event (int timeout)
 {
-        struct list_head               *tmp;
+        cfs_list_t                     *tmp;
         struct liblustre_wait_callback *llwc;
         int                             found_something = 0;
 
@@ -594,9 +663,10 @@ liblustre_wait_event (int timeout)
                         found_something = 1;
 
                 /* Give all registered callbacks a bite at the cherry */
-                list_for_each(tmp, &liblustre_wait_callbacks) {
-                        llwc = list_entry(tmp, struct liblustre_wait_callback,
-                                          llwc_list);
+                cfs_list_for_each(tmp, &liblustre_wait_callbacks) {
+                        llwc = cfs_list_entry(tmp,
+                                              struct liblustre_wait_callback,
+                                              llwc_list);
 
                         if (llwc->llwc_fn(llwc->llwc_arg))
                                 found_something = 1;
@@ -620,29 +690,30 @@ void
 liblustre_wait_idle(void)
 {
         static int recursed = 0;
-        
-        struct list_head               *tmp;
+
+        cfs_list_t                     *tmp;
         struct liblustre_wait_callback *llwc;
         int                             idle = 0;
 
         LASSERT(!recursed);
         recursed = 1;
-        
+
         do {
                 liblustre_wait_event(0);
 
                 idle = 1;
 
-                list_for_each(tmp, &liblustre_idle_callbacks) {
-                        llwc = list_entry(tmp, struct liblustre_wait_callback,
-                                          llwc_list);
-                        
+                cfs_list_for_each(tmp, &liblustre_idle_callbacks) {
+                        llwc = cfs_list_entry(tmp,
+                                              struct liblustre_wait_callback,
+                                              llwc_list);
+
                         if (!llwc->llwc_fn(llwc->llwc_arg)) {
                                 idle = 0;
                                 break;
                         }
                 }
-                        
+
         } while (!idle);
 
         recursed = 0;
@@ -663,11 +734,12 @@ int ptlrpc_init_portals(void)
                 liblustre_register_wait_callback("liblustre_check_services",
                                                  &liblustre_check_services,
                                                  NULL);
+        cfs_init_completion_module(liblustre_wait_event);
 #endif
         rc = ptlrpcd_addref();
         if (rc == 0)
                 return 0;
-        
+
         CERROR("rpcd initialisation failed\n");
 #ifndef __KERNEL__
         liblustre_deregister_wait_callback(liblustre_services_callback);