4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 * lustre/ptlrpc/recover.c
33 * Author: Mike Shaver <shaver@clusterfs.com>
36 #define DEBUG_SUBSYSTEM S_RPC
37 #include <linux/list.h>
38 #include <libcfs/libcfs.h>
39 #include <obd_support.h>
40 #include <lustre_ha.h>
41 #include <lustre_net.h>
42 #include <lustre_import.h>
43 #include <lustre_export.h>
45 #include <obd_class.h>
47 #include "ptlrpc_internal.h"
50 * Start recovery on disconnected import.
51 * This is done by just attempting a connect
53 void ptlrpc_initiate_recovery(struct obd_import *imp)
57 CDEBUG(D_HA, "%s: starting recovery\n", obd2cli_tgt(imp->imp_obd));
58 ptlrpc_connect_import(imp);
64 * Identify what request from replay list needs to be replayed next
65 * (based on what we have already replayed) and send it to server.
67 int ptlrpc_replay_next(struct obd_import *imp, int *inflight)
70 struct ptlrpc_request *req = NULL;
76 /* It might have committed some after we last spoke, so make sure we
77 * get rid of them now.
79 spin_lock(&imp->imp_lock);
80 imp->imp_last_transno_checked = 0;
81 ptlrpc_free_committed(imp);
82 last_transno = imp->imp_last_replay_transno;
84 CDEBUG(D_HA, "import %p from %s committed %llu last %llu\n",
85 imp, obd2cli_tgt(imp->imp_obd),
86 imp->imp_peer_committed_transno, last_transno);
88 /* Replay all the committed open requests on committed_list first */
89 if (!list_empty(&imp->imp_committed_list)) {
90 req = list_last_entry(&imp->imp_committed_list,
91 struct ptlrpc_request, rq_replay_list);
93 /* The last request on committed_list hasn't been replayed */
94 if (req->rq_transno > last_transno) {
95 if (!imp->imp_resend_replay ||
96 imp->imp_replay_cursor == &imp->imp_committed_list)
97 imp->imp_replay_cursor =
98 imp->imp_replay_cursor->next;
100 while (imp->imp_replay_cursor !=
101 &imp->imp_committed_list) {
102 req = list_entry(imp->imp_replay_cursor,
103 struct ptlrpc_request,
105 if (req->rq_transno > last_transno)
109 LASSERT(!list_empty(imp->imp_replay_cursor));
110 imp->imp_replay_cursor =
111 imp->imp_replay_cursor->next;
114 /* All requests on committed_list have been replayed */
115 imp->imp_replay_cursor = &imp->imp_committed_list;
120 /* All the requests in committed list have been replayed, let's replay
121 * the imp_replay_list */
123 struct ptlrpc_request *tmp;
125 list_for_each_entry(tmp, &imp->imp_replay_list,
127 if (tmp->rq_transno > last_transno) {
134 /* If need to resend the last sent transno (because a reconnect
135 * has occurred), then stop on the matching req and send it again.
136 * If, however, the last sent transno has been committed then we
137 * continue replay from the next request. */
138 if (req != NULL && imp->imp_resend_replay)
139 lustre_msg_add_flags(req->rq_reqmsg, MSG_RESENT);
141 /* ptlrpc_prepare_replay() may fail to add the reqeust into unreplied
142 * list if the request hasn't been added to replay list then. Another
143 * exception is that resend replay could have been removed from the
145 if (req != NULL && list_empty(&req->rq_unreplied_list)) {
146 DEBUG_REQ(D_HA, req, "resend_replay=%d, last_transno=%llu",
147 imp->imp_resend_replay, last_transno);
148 ptlrpc_add_unreplied(req);
149 imp->imp_known_replied_xid = ptlrpc_known_replied_xid(imp);
152 imp->imp_resend_replay = 0;
153 spin_unlock(&imp->imp_lock);
156 LASSERT(!list_empty(&req->rq_unreplied_list));
158 rc = ptlrpc_replay_req(req);
160 CERROR("recovery replay error %d for req %llu\n",
170 * Schedule resending of request on sending_list. This is done after
171 * we completed replaying of requests and locks.
173 int ptlrpc_resend(struct obd_import *imp)
175 struct ptlrpc_request *req;
179 /* As long as we're in recovery, nothing should be added to the sending
180 * list, so we don't need to hold the lock during this iteration and
183 /* Well... what if lctl recover is called twice at the same time?
185 spin_lock(&imp->imp_lock);
186 if (imp->imp_state != LUSTRE_IMP_RECOVER) {
187 spin_unlock(&imp->imp_lock);
191 list_for_each_entry(req, &imp->imp_sending_list, rq_list) {
192 LASSERTF((long)req > PAGE_SIZE && req != LP_POISON,
193 "req %p bad\n", req);
194 LASSERTF(req->rq_type != LI_POISON, "req %p freed\n", req);
196 /* If the request is allowed to be sent during replay and it
197 * is not timeout yet, then it does not need to be resent. */
198 if (!ptlrpc_no_resend(req) &&
199 (req->rq_timedout || !req->rq_allow_replay))
200 ptlrpc_resend_req(req);
202 spin_unlock(&imp->imp_lock);
204 OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_ENQUEUE_OLD_EXPORT, 2);
209 * Go through all requests in delayed list and wake their threads
212 void ptlrpc_wake_delayed(struct obd_import *imp)
214 struct ptlrpc_request *req;
216 spin_lock(&imp->imp_lock);
217 list_for_each_entry(req, &imp->imp_delayed_list, rq_list) {
218 DEBUG_REQ(D_HA, req, "waking (set %p):", req->rq_set);
219 ptlrpc_client_wake_req(req);
221 spin_unlock(&imp->imp_lock);
224 void ptlrpc_request_handle_notconn(struct ptlrpc_request *failed_req)
226 struct obd_import *imp = failed_req->rq_import;
227 int conn = lustre_msg_get_conn_cnt(failed_req->rq_reqmsg);
230 CDEBUG(D_HA, "import %s of %s@%s abruptly disconnected: reconnecting\n",
231 imp->imp_obd->obd_name, obd2cli_tgt(imp->imp_obd),
232 imp->imp_connection->c_remote_uuid.uuid);
234 if (ptlrpc_set_import_discon(imp, conn, true)) {
235 /* to control recovery via lctl {disable|enable}_recovery */
236 if (imp->imp_deactive == 0)
237 ptlrpc_connect_import(imp);
240 /* Wait for recovery to complete and resend. If evicted, then
241 this request will be errored out later.*/
242 spin_lock(&failed_req->rq_lock);
243 if (!failed_req->rq_no_resend)
244 failed_req->rq_resend = 1;
245 spin_unlock(&failed_req->rq_lock);
251 * Administratively active/deactive a client.
252 * This should only be called by the ioctl interface, currently
253 * - the lctl deactivate and activate commands
254 * - echo 0/1 >> /proc/osc/XXX/active
255 * - client umount -f (ll_umount_begin)
257 int ptlrpc_set_import_active(struct obd_import *imp, int active)
259 struct obd_device *obd = imp->imp_obd;
265 /* When deactivating, mark import invalid, and abort in-flight
268 LCONSOLE_WARN("setting import %s INACTIVE by administrator "
269 "request\n", obd2cli_tgt(imp->imp_obd));
271 /* set before invalidate to avoid messages about imp_inval
272 * set without imp_deactive in ptlrpc_import_delay_req */
273 spin_lock(&imp->imp_lock);
274 imp->imp_deactive = 1;
275 spin_unlock(&imp->imp_lock);
277 obd_import_event(imp->imp_obd, imp, IMP_EVENT_DEACTIVATE);
279 ptlrpc_invalidate_import(imp);
282 /* When activating, mark import valid, and attempt recovery */
284 CDEBUG(D_HA, "setting import %s VALID\n",
285 obd2cli_tgt(imp->imp_obd));
287 spin_lock(&imp->imp_lock);
288 imp->imp_deactive = 0;
289 spin_unlock(&imp->imp_lock);
290 obd_import_event(imp->imp_obd, imp, IMP_EVENT_ACTIVATE);
292 rc = ptlrpc_recover_import(imp, NULL, 0);
297 EXPORT_SYMBOL(ptlrpc_set_import_active);
299 /* Attempt to reconnect an import */
300 int ptlrpc_recover_import(struct obd_import *imp, char *new_uuid, int async)
305 spin_lock(&imp->imp_lock);
306 if (imp->imp_state == LUSTRE_IMP_NEW || imp->imp_deactive ||
307 atomic_read(&imp->imp_inval_count))
309 spin_unlock(&imp->imp_lock);
313 /* force import to be disconnected. */
314 ptlrpc_set_import_discon(imp, 0, false);
317 struct obd_uuid uuid;
319 /* intruct import to use new uuid */
320 obd_str2uuid(&uuid, new_uuid);
321 rc = import_set_conn_priority(imp, &uuid);
326 /* Check if reconnect is already in progress */
327 spin_lock(&imp->imp_lock);
328 if (imp->imp_state != LUSTRE_IMP_DISCON) {
329 imp->imp_force_verify = 1;
332 spin_unlock(&imp->imp_lock);
336 OBD_RACE(OBD_FAIL_PTLRPC_CONNECT_RACE);
338 rc = ptlrpc_connect_import(imp);
343 long timeout = cfs_time_seconds(obd_timeout);
345 CDEBUG(D_HA, "%s: recovery started, waiting %u jiffies\n",
346 obd2cli_tgt(imp->imp_obd), obd_timeout);
348 rc = wait_event_idle_timeout(imp->imp_recovery_waitq,
349 !ptlrpc_import_in_recovery(imp),
355 CDEBUG(D_HA, "%s: recovery finished\n",
356 obd2cli_tgt(imp->imp_obd));
363 EXPORT_SYMBOL(ptlrpc_recover_import);
365 int ptlrpc_import_in_recovery(struct obd_import *imp)
369 spin_lock(&imp->imp_lock);
370 if (imp->imp_state <= LUSTRE_IMP_DISCON ||
371 imp->imp_state >= LUSTRE_IMP_FULL ||
372 imp->imp_obd->obd_no_recov)
374 spin_unlock(&imp->imp_lock);