1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2002, 2003 Cluster File Systems, Inc.
6 * This file is part of Lustre, http://www.lustre.org.
8 * Lustre is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Lustre is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Lustre; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
22 #define DEBUG_SUBSYSTEM S_LDLM
25 #include <liblustre.h>
28 #include <linux/lustre_dlm.h>
29 #include <linux/obd_class.h>
30 #include <linux/obd.h>
32 #include "ldlm_internal.h"
34 static void interrupted_completion_wait(void *data)
38 struct lock_wait_data {
39 struct ldlm_lock *lwd_lock;
43 int ldlm_expired_completion_wait(void *data)
45 struct lock_wait_data *lwd = data;
46 struct ldlm_lock *lock = lwd->lwd_lock;
47 struct obd_import *imp;
48 struct obd_device *obd;
50 if (lock->l_conn_export == NULL) {
51 static unsigned long next_dump = 0;
53 LDLM_ERROR(lock, "lock timed out; not entering recovery in "
54 "server code, just going back to sleep");
55 if (time_after(jiffies, next_dump)) {
56 ldlm_namespace_dump(lock->l_resource->lr_namespace);
58 portals_debug_dumplog();
59 next_dump = jiffies + 300 * HZ;
64 obd = lock->l_conn_export->exp_obd;
65 imp = obd->u.cli.cl_import;
66 ptlrpc_fail_import(imp, lwd->lwd_generation);
67 LDLM_ERROR(lock, "lock timed out, entering recovery for %s@%s",
68 imp->imp_target_uuid.uuid,
69 imp->imp_connection->c_remote_uuid.uuid);
74 int ldlm_completion_ast(struct ldlm_lock *lock, int flags, void *data)
76 /* XXX ALLOCATE - 160 bytes */
77 struct lock_wait_data lwd;
78 unsigned long irqflags;
79 struct obd_device *obd;
80 struct obd_import *imp = NULL;
81 struct l_wait_info lwi;
85 if (flags == LDLM_FL_WAIT_NOREPROC)
88 if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
89 LDLM_FL_BLOCK_CONV))) {
90 wake_up(&lock->l_waitq);
94 LDLM_DEBUG(lock, "client-side enqueue returned a blocked lock, "
96 ldlm_lock_dump(D_OTHER, lock, 0);
97 ldlm_reprocess_all(lock->l_resource);
101 obd = class_exp2obd(lock->l_conn_export);
103 /* if this is a local lock, then there is no import */
105 imp = obd->u.cli.cl_import;
109 if (flags & LDLM_FL_NO_TIMEOUT) {
110 LDLM_DEBUG(lock, "waiting indefinitely for group lock\n");
111 lwi = LWI_INTR(interrupted_completion_wait, &lwd);
113 lwi = LWI_TIMEOUT_INTR(obd_timeout * HZ,
114 ldlm_expired_completion_wait,
115 interrupted_completion_wait, &lwd);
119 spin_lock_irqsave(&imp->imp_lock, irqflags);
120 lwd.lwd_generation = imp->imp_generation;
121 spin_unlock_irqrestore(&imp->imp_lock, irqflags);
124 /* Go to sleep until the lock is granted or cancelled. */
125 rc = l_wait_event(lock->l_waitq,
126 ((lock->l_req_mode == lock->l_granted_mode) ||
127 (lock->l_flags & LDLM_FL_FAILED)), &lwi);
129 if (lock->l_destroyed || lock->l_flags & LDLM_FL_FAILED) {
130 LDLM_DEBUG(lock, "client-side enqueue waking up: destroyed");
135 LDLM_DEBUG(lock, "client-side enqueue waking up: failed (%d)",
140 LDLM_DEBUG(lock, "client-side enqueue waking up: granted");
144 static int ldlm_cli_enqueue_local(struct ldlm_namespace *ns,
145 struct ldlm_res_id res_id,
147 ldlm_policy_data_t *policy,
150 ldlm_blocking_callback blocking,
151 ldlm_completion_callback completion,
152 ldlm_glimpse_callback glimpse,
153 void *data, __u32 lvb_len,
155 struct lustre_handle *lockh)
157 struct ldlm_lock *lock;
162 CERROR("Trying to enqueue local lock in a shadow namespace\n");
166 lock = ldlm_lock_create(ns, NULL, res_id, type, mode, blocking,
167 completion, glimpse, data, lvb_len);
169 GOTO(out_nolock, err = -ENOMEM);
170 LDLM_DEBUG(lock, "client-side local enqueue handler, new lock created");
172 ldlm_lock_addref_internal(lock, mode);
173 ldlm_lock2handle(lock, lockh);
174 lock->l_flags |= LDLM_FL_LOCAL;
175 lock->l_lvb_swabber = lvb_swabber;
177 memcpy(&lock->l_policy_data, policy, sizeof(*policy));
178 if (type == LDLM_EXTENT)
179 memcpy(&lock->l_req_extent, &policy->l_extent,
180 sizeof(policy->l_extent));
182 err = ldlm_lock_enqueue(ns, &lock, policy, flags);
187 memcpy(policy, &lock->l_policy_data, sizeof(*policy));
188 if ((*flags) & LDLM_FL_LOCK_CHANGED)
189 memcpy(&res_id, &lock->l_resource->lr_name, sizeof(res_id));
191 LDLM_DEBUG_NOLOCK("client-side local enqueue handler END (lock %p)",
194 if (lock->l_completion_ast)
195 lock->l_completion_ast(lock, *flags, NULL);
197 LDLM_DEBUG(lock, "client-side local enqueue END");
205 static void failed_lock_cleanup(struct ldlm_namespace *ns,
206 struct ldlm_lock *lock,
207 struct lustre_handle *lockh, int mode)
209 /* Set a flag to prevent us from sending a CANCEL (bug 407) */
210 l_lock(&ns->ns_lock);
211 lock->l_flags |= LDLM_FL_LOCAL_ONLY;
212 LDLM_DEBUG(lock, "setting FL_LOCAL_ONLY");
213 l_unlock(&ns->ns_lock);
215 ldlm_lock_decref_and_cancel(lockh, mode);
218 int ldlm_cli_enqueue(struct obd_export *exp,
219 struct ptlrpc_request *req,
220 struct ldlm_namespace *ns,
221 struct ldlm_res_id res_id,
223 ldlm_policy_data_t *policy,
226 ldlm_blocking_callback blocking,
227 ldlm_completion_callback completion,
228 ldlm_glimpse_callback glimpse,
233 struct lustre_handle *lockh)
235 struct ldlm_lock *lock;
236 struct ldlm_request *body;
237 struct ldlm_reply *reply;
238 int rc, size[2] = {sizeof(*body), lvb_len}, req_passed_in = 1;
239 int is_replay = *flags & LDLM_FL_REPLAY;
244 rc = ldlm_cli_enqueue_local(ns, res_id, type, policy, mode,
245 flags, blocking, completion,
246 glimpse, data, lvb_len, lvb_swabber,
251 /* If we're replaying this lock, just check some invariants.
252 * If we're creating a new lock, get everything all setup nice. */
254 lock = ldlm_handle2lock(lockh);
255 LDLM_DEBUG(lock, "client-side enqueue START");
256 LASSERT(exp == lock->l_conn_export);
258 lock = ldlm_lock_create(ns, NULL, res_id, type, mode, blocking,
259 completion, glimpse, data, lvb_len);
261 GOTO(out_nolock, rc = -ENOMEM);
262 /* for the local lock, add the reference */
263 ldlm_lock_addref_internal(lock, mode);
264 ldlm_lock2handle(lock, lockh);
265 lock->l_lvb_swabber = lvb_swabber;
267 memcpy(&lock->l_policy_data, policy, sizeof(*policy));
268 if (type == LDLM_EXTENT)
269 memcpy(&lock->l_req_extent, &policy->l_extent,
270 sizeof(policy->l_extent));
271 LDLM_DEBUG(lock, "client-side enqueue START");
275 req = ptlrpc_prep_req(class_exp2cliimp(exp), LDLM_ENQUEUE, 1,
278 GOTO(out_lock, rc = -ENOMEM);
280 } else if (req->rq_reqmsg->buflens[0] != sizeof(*body))
283 /* Dump lock data into the request buffer */
284 body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
285 ldlm_lock2desc(lock, &body->lock_desc);
286 body->lock_flags = *flags;
288 memcpy(&body->lock_handle1, lockh, sizeof(*lockh));
290 /* Continue as normal. */
291 if (!req_passed_in) {
295 size[0] = sizeof(*reply);
296 req->rq_replen = lustre_msg_size(buffers, size);
298 lock->l_conn_export = exp;
299 lock->l_export = NULL;
300 lock->l_blocking_ast = blocking;
302 LDLM_DEBUG(lock, "sending request");
303 rc = ptlrpc_queue_wait(req);
305 if (rc != ELDLM_OK) {
307 LDLM_DEBUG(lock, "client-side enqueue END (%s)",
308 rc == ELDLM_LOCK_ABORTED ? "ABORTED" : "FAILED");
309 if (rc == ELDLM_LOCK_ABORTED) {
310 /* Before we return, swab the reply */
311 reply = lustre_swab_repbuf(req, 0, sizeof(*reply),
312 lustre_swab_ldlm_reply);
314 CERROR("Can't unpack ldlm_reply\n");
319 tmplvb = lustre_swab_repbuf(req, 1, lvb_len,
322 GOTO(out_lock, rc = -EPROTO);
324 memcpy(lvb, tmplvb, lvb_len);
330 reply = lustre_swab_repbuf(req, 0, sizeof(*reply),
331 lustre_swab_ldlm_reply);
333 CERROR("Can't unpack ldlm_reply\n");
334 GOTO(out_lock, rc = -EPROTO);
337 memcpy(&lock->l_remote_handle, &reply->lock_handle,
338 sizeof(lock->l_remote_handle));
339 *flags = reply->lock_flags;
341 CDEBUG(D_INFO, "local: %p, remote cookie: "LPX64", flags: 0x%x\n",
342 lock, reply->lock_handle.cookie, *flags);
344 /* If enqueue returned a blocked lock but the completion handler has
345 * already run, then it fixed up the resource and we don't need to do it
347 if ((*flags) & LDLM_FL_LOCK_CHANGED) {
348 int newmode = reply->lock_desc.l_req_mode;
350 if (newmode && newmode != lock->l_req_mode) {
351 LDLM_DEBUG(lock, "server returned different mode %s",
352 ldlm_lockname[newmode]);
353 lock->l_req_mode = newmode;
356 if (reply->lock_desc.l_resource.lr_name.name[0] !=
357 lock->l_resource->lr_name.name[0]) {
358 CDEBUG(D_INFO, "remote intent success, locking %ld "
360 (long)reply->lock_desc.l_resource.lr_name.name[0],
361 (long)lock->l_resource->lr_name.name[0]);
363 ldlm_lock_change_resource(ns, lock,
364 reply->lock_desc.l_resource.lr_name);
365 if (lock->l_resource == NULL) {
367 GOTO(out_lock, rc = -ENOMEM);
369 LDLM_DEBUG(lock, "client-side enqueue, new resource");
372 memcpy(&lock->l_policy_data,
373 &reply->lock_desc.l_policy_data,
374 sizeof(reply->lock_desc.l_policy_data));
375 if (type != LDLM_PLAIN)
376 LDLM_DEBUG(lock,"client-side enqueue, new policy data");
379 if ((*flags) & LDLM_FL_AST_SENT) {
380 l_lock(&ns->ns_lock);
381 lock->l_flags |= LDLM_FL_CBPENDING;
382 l_unlock(&ns->ns_lock);
383 LDLM_DEBUG(lock, "enqueue reply includes blocking AST");
386 /* If the lock has already been granted by a completion AST, don't
387 * clobber the LVB with an older one. */
388 if (lvb_len && (lock->l_req_mode != lock->l_granted_mode)) {
390 tmplvb = lustre_swab_repbuf(req, 1, lvb_len, lvb_swabber);
392 GOTO(out_lock, rc = -EPROTO);
393 memcpy(lock->l_lvb_data, tmplvb, lvb_len);
397 rc = ldlm_lock_enqueue(ns, &lock, NULL, flags);
398 if (lock->l_completion_ast != NULL) {
399 int err = lock->l_completion_ast(lock, *flags, NULL);
405 if (lvb_len && lvb != NULL) {
406 /* Copy the LVB here, and not earlier, because the completion
407 * AST (if any) can override what we got in the reply */
408 memcpy(lvb, lock->l_lvb_data, lvb_len);
411 LDLM_DEBUG(lock, "client-side enqueue END");
415 failed_lock_cleanup(ns, lock, lockh, mode);
416 if (!req_passed_in && req != NULL)
417 ptlrpc_req_finished(req);
423 static int ldlm_cli_convert_local(struct ldlm_lock *lock, int new_mode,
427 if (lock->l_resource->lr_namespace->ns_client) {
428 CERROR("Trying to cancel local lock\n");
431 LDLM_DEBUG(lock, "client-side local convert");
433 ldlm_lock_convert(lock, new_mode, flags);
434 ldlm_reprocess_all(lock->l_resource);
436 LDLM_DEBUG(lock, "client-side local convert handler END");
441 /* FIXME: one of ldlm_cli_convert or the server side should reject attempted
442 * conversion of locks which are on the waiting or converting queue */
443 int ldlm_cli_convert(struct lustre_handle *lockh, int new_mode, int *flags)
445 struct ldlm_request *body;
446 struct ldlm_reply *reply;
447 struct ldlm_lock *lock;
448 struct ldlm_resource *res;
449 struct ptlrpc_request *req;
450 int rc, size = sizeof(*body);
453 lock = ldlm_handle2lock(lockh);
460 if (lock->l_conn_export == NULL)
461 RETURN(ldlm_cli_convert_local(lock, new_mode, flags));
463 LDLM_DEBUG(lock, "client-side convert");
465 req = ptlrpc_prep_req(class_exp2cliimp(lock->l_conn_export),
466 LDLM_CONVERT, 1, &size, NULL);
468 GOTO(out, rc = -ENOMEM);
470 body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
471 memcpy(&body->lock_handle1, &lock->l_remote_handle,
472 sizeof(body->lock_handle1));
474 body->lock_desc.l_req_mode = new_mode;
475 body->lock_flags = *flags;
477 size = sizeof(*reply);
478 req->rq_replen = lustre_msg_size(1, &size);
480 rc = ptlrpc_queue_wait(req);
484 reply = lustre_swab_repbuf(req, 0, sizeof (*reply),
485 lustre_swab_ldlm_reply);
487 CERROR ("Can't unpack ldlm_reply\n");
488 GOTO (out, rc = -EPROTO);
491 res = ldlm_lock_convert(lock, new_mode, &reply->lock_flags);
493 ldlm_reprocess_all(res);
494 /* Go to sleep until the lock is granted. */
495 /* FIXME: or cancelled. */
496 if (lock->l_completion_ast)
497 lock->l_completion_ast(lock, LDLM_FL_WAIT_NOREPROC, NULL);
501 ptlrpc_req_finished(req);
505 int ldlm_cli_cancel(struct lustre_handle *lockh)
507 struct ptlrpc_request *req;
508 struct ldlm_lock *lock;
509 struct ldlm_request *body;
510 int rc = 0, size = sizeof(*body);
513 /* concurrent cancels on the same handle can happen */
514 lock = __ldlm_handle2lock(lockh, LDLM_FL_CANCELING);
518 if (lock->l_conn_export) {
520 struct obd_import *imp;
522 LDLM_DEBUG(lock, "client-side cancel");
523 /* Set this flag to prevent others from getting new references*/
524 l_lock(&lock->l_resource->lr_namespace->ns_lock);
525 lock->l_flags |= LDLM_FL_CBPENDING;
526 local_only = (lock->l_flags & LDLM_FL_LOCAL_ONLY);
527 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
528 ldlm_cancel_callback(lock);
531 CDEBUG(D_INFO, "not sending request (at caller's "
537 imp = class_exp2cliimp(lock->l_conn_export);
538 if (imp == NULL || imp->imp_invalid) {
539 CDEBUG(D_HA, "skipping cancel on invalid import %p\n",
544 req = ptlrpc_prep_req(imp, LDLM_CANCEL, 1, &size, NULL);
546 GOTO(out, rc = -ENOMEM);
547 req->rq_no_resend = 1;
549 /* XXX FIXME bug 249 */
550 req->rq_request_portal = LDLM_CANCEL_REQUEST_PORTAL;
551 req->rq_reply_portal = LDLM_CANCEL_REPLY_PORTAL;
553 body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
554 memcpy(&body->lock_handle1, &lock->l_remote_handle,
555 sizeof(body->lock_handle1));
557 req->rq_replen = lustre_msg_size(0, NULL);
559 rc = ptlrpc_queue_wait(req);
562 CERROR("client/server (nid "LPU64") out of sync--not "
564 req->rq_import->imp_connection->c_peer.peer_nid);
565 } else if (rc == -ETIMEDOUT) {
566 ptlrpc_req_finished(req);
568 } else if (rc != ELDLM_OK) {
569 CERROR("Got rc %d from cancel RPC: canceling "
573 ptlrpc_req_finished(req);
575 ldlm_lock_cancel(lock);
577 if (lock->l_resource->lr_namespace->ns_client) {
578 LDLM_ERROR(lock, "Trying to cancel local lock\n");
581 LDLM_DEBUG(lock, "client-side local cancel");
582 ldlm_lock_cancel(lock);
583 ldlm_reprocess_all(lock->l_resource);
584 LDLM_DEBUG(lock, "client-side local cancel handler END");
593 int ldlm_cancel_lru(struct ldlm_namespace *ns)
595 struct list_head *tmp, *next;
602 l_lock(&ns->ns_lock);
603 count = ns->ns_nr_unused - ns->ns_max_unused;
606 l_unlock(&ns->ns_lock);
610 list_for_each_safe(tmp, next, &ns->ns_unused_list) {
611 struct ldlm_lock *lock;
612 lock = list_entry(tmp, struct ldlm_lock, l_lru);
614 LASSERT(!lock->l_readers && !lock->l_writers);
616 /* Setting the CBPENDING flag is a little misleading, but
617 * prevents an important race; namely, once CBPENDING is set,
618 * the lock can accumulate no more readers/writers. Since
619 * readers and writers are already zero here, ldlm_lock_decref
620 * won't see this flag and call l_blocking_ast */
621 lock->l_flags |= LDLM_FL_CBPENDING;
623 LDLM_LOCK_GET(lock); /* dropped by bl thread */
624 ldlm_lock_remove_from_lru(lock);
626 ldlm_bl_to_thread(ns, NULL, lock);
628 list_add(&lock->l_lru, &cblist);
634 l_unlock(&ns->ns_lock);
636 while (!list_empty(&cblist)) {
637 struct ldlm_lock *lock;
639 lock = list_entry(cblist.next, struct ldlm_lock, l_lru);
640 list_del_init(&lock->l_lru);
641 liblustre_ldlm_handle_bl_callback(ns, NULL, lock);
647 static int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
648 struct ldlm_res_id res_id, int flags,
651 struct ldlm_resource *res;
652 struct list_head *tmp, *next, list = LIST_HEAD_INIT(list);
653 struct ldlm_ast_work *w;
656 res = ldlm_resource_get(ns, NULL, res_id, 0, 0);
658 /* This is not a problem. */
659 CDEBUG(D_INFO, "No resource "LPU64"\n", res_id.name[0]);
663 l_lock(&ns->ns_lock);
664 list_for_each(tmp, &res->lr_granted) {
665 struct ldlm_lock *lock;
666 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
668 if (opaque != NULL && lock->l_ast_data != opaque) {
669 LDLM_ERROR(lock, "data %p doesn't match opaque %p",
670 lock->l_ast_data, opaque);
675 if (lock->l_readers || lock->l_writers) {
676 if (flags & LDLM_FL_WARN) {
677 LDLM_ERROR(lock, "lock in use");
683 /* See CBPENDING comment in ldlm_cancel_lru */
684 lock->l_flags |= LDLM_FL_CBPENDING;
686 OBD_ALLOC(w, sizeof(*w));
689 w->w_lock = LDLM_LOCK_GET(lock);
691 list_add(&w->w_list, &list);
693 l_unlock(&ns->ns_lock);
695 list_for_each_safe(tmp, next, &list) {
696 struct lustre_handle lockh;
698 w = list_entry(tmp, struct ldlm_ast_work, w_list);
700 if (flags & LDLM_FL_LOCAL_ONLY) {
701 ldlm_lock_cancel(w->w_lock);
703 ldlm_lock2handle(w->w_lock, &lockh);
704 rc = ldlm_cli_cancel(&lockh);
706 CERROR("ldlm_cli_cancel: %d\n", rc);
708 list_del(&w->w_list);
709 LDLM_LOCK_PUT(w->w_lock);
710 OBD_FREE(w, sizeof(*w));
713 ldlm_resource_putref(res);
718 /* Cancel all locks on a namespace (or a specific resource, if given)
719 * that have 0 readers/writers.
721 * If flags & LDLM_FL_LOCAL_ONLY, throw the locks away without trying
722 * to notify the server.
723 * If flags & LDLM_FL_WARN, print a warning if some locks are still in use. */
724 int ldlm_cli_cancel_unused(struct ldlm_namespace *ns,
725 struct ldlm_res_id *res_id, int flags, void *opaque)
734 RETURN(ldlm_cli_cancel_unused_resource(ns, *res_id, flags,
737 l_lock(&ns->ns_lock);
738 for (i = 0; i < RES_HASH_SIZE; i++) {
739 struct list_head *tmp, *pos;
740 list_for_each_safe(tmp, pos, &(ns->ns_hash[i])) {
742 struct ldlm_resource *res;
743 res = list_entry(tmp, struct ldlm_resource, lr_hash);
744 ldlm_resource_getref(res);
746 rc = ldlm_cli_cancel_unused_resource(ns, res->lr_name,
750 CERROR("cancel_unused_res ("LPU64"): %d\n",
751 res->lr_name.name[0], rc);
752 ldlm_resource_putref(res);
755 l_unlock(&ns->ns_lock);
760 /* Lock iterators. */
762 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
765 struct list_head *tmp, *next;
766 struct ldlm_lock *lock;
767 int rc = LDLM_ITER_CONTINUE;
768 struct ldlm_namespace *ns = res->lr_namespace;
773 RETURN(LDLM_ITER_CONTINUE);
775 l_lock(&ns->ns_lock);
776 list_for_each_safe(tmp, next, &res->lr_granted) {
777 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
779 if (iter(lock, closure) == LDLM_ITER_STOP)
780 GOTO(out, rc = LDLM_ITER_STOP);
783 list_for_each_safe(tmp, next, &res->lr_converting) {
784 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
786 if (iter(lock, closure) == LDLM_ITER_STOP)
787 GOTO(out, rc = LDLM_ITER_STOP);
790 list_for_each_safe(tmp, next, &res->lr_waiting) {
791 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
793 if (iter(lock, closure) == LDLM_ITER_STOP)
794 GOTO(out, rc = LDLM_ITER_STOP);
797 l_unlock(&ns->ns_lock);
801 struct iter_helper_data {
802 ldlm_iterator_t iter;
806 static int ldlm_iter_helper(struct ldlm_lock *lock, void *closure)
808 struct iter_helper_data *helper = closure;
809 return helper->iter(lock, helper->closure);
812 static int ldlm_res_iter_helper(struct ldlm_resource *res, void *closure)
814 return ldlm_resource_foreach(res, ldlm_iter_helper, closure);
817 int ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
820 struct iter_helper_data helper = { iter: iter, closure: closure };
821 return ldlm_namespace_foreach_res(ns, ldlm_res_iter_helper, &helper);
824 int ldlm_namespace_foreach_res(struct ldlm_namespace *ns,
825 ldlm_res_iterator_t iter, void *closure)
827 int i, rc = LDLM_ITER_CONTINUE;
829 l_lock(&ns->ns_lock);
830 for (i = 0; i < RES_HASH_SIZE; i++) {
831 struct list_head *tmp, *next;
832 list_for_each_safe(tmp, next, &(ns->ns_hash[i])) {
833 struct ldlm_resource *res =
834 list_entry(tmp, struct ldlm_resource, lr_hash);
836 ldlm_resource_getref(res);
837 rc = iter(res, closure);
838 ldlm_resource_putref(res);
839 if (rc == LDLM_ITER_STOP)
844 l_unlock(&ns->ns_lock);
848 /* non-blocking function to manipulate a lock whose cb_data is being put away.*/
849 void ldlm_change_cbdata(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
850 ldlm_iterator_t iter, void *data)
852 struct ldlm_resource *res;
856 CERROR("must pass in namespace");
860 res = ldlm_resource_get(ns, NULL, *res_id, 0, 0);
866 l_lock(&ns->ns_lock);
867 ldlm_resource_foreach(res, iter, data);
868 l_unlock(&ns->ns_lock);
869 ldlm_resource_putref(res);
875 static int ldlm_chain_lock_for_replay(struct ldlm_lock *lock, void *closure)
877 struct list_head *list = closure;
879 /* we use l_pending_chain here, because it's unused on clients. */
880 list_add(&lock->l_pending_chain, list);
881 return LDLM_ITER_CONTINUE;
884 static int replay_lock_interpret(struct ptlrpc_request *req,
887 struct ldlm_lock *lock;
888 struct ldlm_reply *reply;
890 atomic_dec(&req->rq_import->imp_replay_inflight);
894 lock = req->rq_async_args.pointer_arg[0];
895 LASSERT(lock != NULL);
897 reply = lustre_swab_repbuf(req, 0, sizeof (*reply),
898 lustre_swab_ldlm_reply);
900 CERROR("Can't unpack ldlm_reply\n");
901 GOTO (out, rc = -EPROTO);
904 memcpy(&lock->l_remote_handle, &reply->lock_handle,
905 sizeof(lock->l_remote_handle));
906 LDLM_DEBUG(lock, "replayed lock:");
907 ptlrpc_import_recovery_state_machine(req->rq_import);
912 static int replay_one_lock(struct obd_import *imp, struct ldlm_lock *lock)
914 struct ptlrpc_request *req;
915 struct ldlm_request *body;
916 struct ldlm_reply *reply;
922 * If granted mode matches the requested mode, this lock is granted.
924 * If they differ, but we have a granted mode, then we were granted
925 * one mode and now want another: ergo, converting.
927 * If we haven't been granted anything and are on a resource list,
928 * then we're blocked/waiting.
930 * If we haven't been granted anything and we're NOT on a resource list,
931 * then we haven't got a reply yet and don't have a known disposition.
932 * This happens whenever a lock enqueue is the request that triggers
935 if (lock->l_granted_mode == lock->l_req_mode)
936 flags = LDLM_FL_REPLAY | LDLM_FL_BLOCK_GRANTED;
937 else if (lock->l_granted_mode)
938 flags = LDLM_FL_REPLAY | LDLM_FL_BLOCK_CONV;
939 else if (!list_empty(&lock->l_res_link))
940 flags = LDLM_FL_REPLAY | LDLM_FL_BLOCK_WAIT;
942 flags = LDLM_FL_REPLAY;
944 size[0] = sizeof(*body);
945 req = ptlrpc_prep_req(imp, LDLM_ENQUEUE, 1, size, NULL);
949 /* We're part of recovery, so don't wait for it. */
950 req->rq_send_state = LUSTRE_IMP_REPLAY_LOCKS;
952 body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
953 ldlm_lock2desc(lock, &body->lock_desc);
954 body->lock_flags = flags;
956 ldlm_lock2handle(lock, &body->lock_handle1);
957 size[0] = sizeof(*reply);
958 if (lock->l_lvb_len != 0) {
960 size[1] = lock->l_lvb_len;
962 req->rq_replen = lustre_msg_size(buffers, size);
964 LDLM_DEBUG(lock, "replaying lock:");
966 atomic_inc(&req->rq_import->imp_replay_inflight);
967 req->rq_async_args.pointer_arg[0] = lock;
968 req->rq_interpret_reply = replay_lock_interpret;
969 ptlrpcd_add_req(req);
974 int ldlm_replay_locks(struct obd_import *imp)
976 struct ldlm_namespace *ns = imp->imp_obd->obd_namespace;
977 struct list_head list, *pos, *next;
978 struct ldlm_lock *lock;
982 INIT_LIST_HEAD(&list);
984 LASSERT(atomic_read(&imp->imp_replay_inflight) == 0);
986 /* ensure this doesn't fall to 0 before all have been queued */
987 atomic_inc(&imp->imp_replay_inflight);
989 l_lock(&ns->ns_lock);
990 (void)ldlm_namespace_foreach(ns, ldlm_chain_lock_for_replay, &list);
992 list_for_each_safe(pos, next, &list) {
993 lock = list_entry(pos, struct ldlm_lock, l_pending_chain);
994 rc = replay_one_lock(imp, lock);
996 break; /* or try to do the rest? */
998 l_unlock(&ns->ns_lock);
1000 atomic_dec(&imp->imp_replay_inflight);