1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * defines domain join / leave apis
8 * Copyright (C) 2004 Oracle. All rights reserved.
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
27 #include <linux/module.h>
28 #include <linux/types.h>
29 #include <linux/slab.h>
30 #include <linux/highmem.h>
31 #include <linux/utsname.h>
32 #include <linux/init.h>
33 #include <linux/spinlock.h>
34 #include <linux/delay.h>
35 #include <linux/err.h>
36 #include <linux/debugfs.h>
38 #include "cluster/heartbeat.h"
39 #include "cluster/nodemanager.h"
40 #include "cluster/tcp.h"
43 #include "dlmcommon.h"
44 #include "dlmdomain.h"
49 #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_DOMAIN)
50 #include "cluster/masklog.h"
53 * ocfs2 node maps are array of long int, which limits to send them freely
54 * across the wire due to endianness issues. To workaround this, we convert
55 * long ints to byte arrays. Following 3 routines are helper functions to
56 * set/test/copy bits within those array of bytes
58 static inline void byte_set_bit(u8 nr, u8 map[])
60 map[nr >> 3] |= (1UL << (nr & 7));
63 static inline int byte_test_bit(u8 nr, u8 map[])
65 return ((1UL << (nr & 7)) & (map[nr >> 3])) != 0;
68 static inline void byte_copymap(u8 dmap[], unsigned long smap[],
76 memset(dmap, 0, ((sz + 7) >> 3));
77 for (nn = 0 ; nn < sz; nn++)
78 if (test_bit(nn, smap))
79 byte_set_bit(nn, dmap);
82 static void dlm_free_pagevec(void **vec, int pages)
85 free_page((unsigned long)vec[pages]);
89 static void **dlm_alloc_pagevec(int pages)
91 void **vec = kmalloc(pages * sizeof(void *), GFP_KERNEL);
97 for (i = 0; i < pages; i++)
98 if (!(vec[i] = (void *)__get_free_page(GFP_KERNEL)))
101 mlog(0, "Allocated DLM hash pagevec; %d pages (%lu expected), %lu buckets per page\n",
102 pages, (unsigned long)DLM_HASH_PAGES,
103 (unsigned long)DLM_BUCKETS_PER_PAGE);
106 dlm_free_pagevec(vec, i);
112 * spinlock lock ordering: if multiple locks are needed, obey this ordering:
114 * struct dlm_ctxt->spinlock
115 * struct dlm_lock_resource->spinlock
116 * struct dlm_ctxt->master_lock
117 * struct dlm_ctxt->ast_lock
118 * dlm_master_list_entry->spinlock
123 DEFINE_SPINLOCK(dlm_domain_lock);
124 LIST_HEAD(dlm_domains);
125 static DECLARE_WAIT_QUEUE_HEAD(dlm_domain_events);
128 * The supported protocol version for DLM communication. Running domains
129 * will have a negotiated version with the same major number and a minor
130 * number equal or smaller. The dlm_ctxt->dlm_locking_proto field should
131 * be used to determine what a running domain is actually using.
133 static const struct dlm_protocol_version dlm_protocol = {
138 #define DLM_DOMAIN_BACKOFF_MS 200
140 static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
142 static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
144 static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
146 static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
148 static int dlm_protocol_compare(struct dlm_protocol_version *existing,
149 struct dlm_protocol_version *request);
151 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm);
153 void __dlm_unhash_lockres(struct dlm_lock_resource *lockres)
155 if (!hlist_unhashed(&lockres->hash_node)) {
156 hlist_del_init(&lockres->hash_node);
157 dlm_lockres_put(lockres);
161 void __dlm_insert_lockres(struct dlm_ctxt *dlm,
162 struct dlm_lock_resource *res)
164 struct hlist_head *bucket;
167 assert_spin_locked(&dlm->spinlock);
170 bucket = dlm_lockres_hash(dlm, q->hash);
172 /* get a reference for our hashtable */
173 dlm_lockres_get(res);
175 hlist_add_head(&res->hash_node, bucket);
178 struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm,
183 struct hlist_head *bucket;
184 struct hlist_node *list;
186 mlog_entry("%.*s\n", len, name);
188 assert_spin_locked(&dlm->spinlock);
190 bucket = dlm_lockres_hash(dlm, hash);
192 hlist_for_each(list, bucket) {
193 struct dlm_lock_resource *res = hlist_entry(list,
194 struct dlm_lock_resource, hash_node);
195 if (res->lockname.name[0] != name[0])
197 if (unlikely(res->lockname.len != len))
199 if (memcmp(res->lockname.name + 1, name + 1, len - 1))
201 dlm_lockres_get(res);
207 /* intended to be called by functions which do not care about lock
208 * resources which are being purged (most net _handler functions).
209 * this will return NULL for any lock resource which is found but
210 * currently in the process of dropping its mastery reference.
211 * use __dlm_lookup_lockres_full when you need the lock resource
212 * regardless (e.g. dlm_get_lock_resource) */
213 struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm,
218 struct dlm_lock_resource *res = NULL;
220 mlog_entry("%.*s\n", len, name);
222 assert_spin_locked(&dlm->spinlock);
224 res = __dlm_lookup_lockres_full(dlm, name, len, hash);
226 spin_lock(&res->spinlock);
227 if (res->state & DLM_LOCK_RES_DROPPING_REF) {
228 spin_unlock(&res->spinlock);
229 dlm_lockres_put(res);
232 spin_unlock(&res->spinlock);
238 struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm,
242 struct dlm_lock_resource *res;
243 unsigned int hash = dlm_lockid_hash(name, len);
245 spin_lock(&dlm->spinlock);
246 res = __dlm_lookup_lockres(dlm, name, len, hash);
247 spin_unlock(&dlm->spinlock);
251 static struct dlm_ctxt * __dlm_lookup_domain_full(const char *domain, int len)
253 struct dlm_ctxt *tmp = NULL;
254 struct list_head *iter;
256 assert_spin_locked(&dlm_domain_lock);
258 /* tmp->name here is always NULL terminated,
259 * but domain may not be! */
260 list_for_each(iter, &dlm_domains) {
261 tmp = list_entry (iter, struct dlm_ctxt, list);
262 if (strlen(tmp->name) == len &&
263 memcmp(tmp->name, domain, len)==0)
271 /* For null terminated domain strings ONLY */
272 static struct dlm_ctxt * __dlm_lookup_domain(const char *domain)
274 assert_spin_locked(&dlm_domain_lock);
276 return __dlm_lookup_domain_full(domain, strlen(domain));
280 /* returns true on one of two conditions:
281 * 1) the domain does not exist
282 * 2) the domain exists and it's state is "joined" */
283 static int dlm_wait_on_domain_helper(const char *domain)
286 struct dlm_ctxt *tmp = NULL;
288 spin_lock(&dlm_domain_lock);
290 tmp = __dlm_lookup_domain(domain);
293 else if (tmp->dlm_state == DLM_CTXT_JOINED)
296 spin_unlock(&dlm_domain_lock);
300 static void dlm_free_ctxt_mem(struct dlm_ctxt *dlm)
302 dlm_destroy_debugfs_subroot(dlm);
304 if (dlm->lockres_hash)
305 dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
307 if (dlm->master_hash)
308 dlm_free_pagevec((void **)dlm->master_hash, DLM_HASH_PAGES);
316 /* A little strange - this function will be called while holding
317 * dlm_domain_lock and is expected to be holding it on the way out. We
318 * will however drop and reacquire it multiple times */
319 static void dlm_ctxt_release(struct kref *kref)
321 struct dlm_ctxt *dlm;
323 dlm = container_of(kref, struct dlm_ctxt, dlm_refs);
325 BUG_ON(dlm->num_joins);
326 BUG_ON(dlm->dlm_state == DLM_CTXT_JOINED);
328 /* we may still be in the list if we hit an error during join. */
329 list_del_init(&dlm->list);
331 spin_unlock(&dlm_domain_lock);
333 mlog(0, "freeing memory from domain %s\n", dlm->name);
335 wake_up(&dlm_domain_events);
337 dlm_free_ctxt_mem(dlm);
339 spin_lock(&dlm_domain_lock);
342 void dlm_put(struct dlm_ctxt *dlm)
344 spin_lock(&dlm_domain_lock);
345 kref_put(&dlm->dlm_refs, dlm_ctxt_release);
346 spin_unlock(&dlm_domain_lock);
349 static void __dlm_get(struct dlm_ctxt *dlm)
351 kref_get(&dlm->dlm_refs);
354 /* given a questionable reference to a dlm object, gets a reference if
355 * it can find it in the list, otherwise returns NULL in which case
356 * you shouldn't trust your pointer. */
357 struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm)
359 struct list_head *iter;
360 struct dlm_ctxt *target = NULL;
362 spin_lock(&dlm_domain_lock);
364 list_for_each(iter, &dlm_domains) {
365 target = list_entry (iter, struct dlm_ctxt, list);
375 spin_unlock(&dlm_domain_lock);
380 int dlm_domain_fully_joined(struct dlm_ctxt *dlm)
384 spin_lock(&dlm_domain_lock);
385 ret = (dlm->dlm_state == DLM_CTXT_JOINED) ||
386 (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN);
387 spin_unlock(&dlm_domain_lock);
392 static void dlm_destroy_dlm_worker(struct dlm_ctxt *dlm)
394 if (dlm->dlm_worker) {
395 flush_workqueue(dlm->dlm_worker);
396 destroy_workqueue(dlm->dlm_worker);
397 dlm->dlm_worker = NULL;
401 static void dlm_complete_dlm_shutdown(struct dlm_ctxt *dlm)
403 dlm_unregister_domain_handlers(dlm);
404 dlm_debug_shutdown(dlm);
405 dlm_complete_thread(dlm);
406 dlm_complete_recovery_thread(dlm);
407 dlm_destroy_dlm_worker(dlm);
409 /* We've left the domain. Now we can take ourselves out of the
410 * list and allow the kref stuff to help us free the
412 spin_lock(&dlm_domain_lock);
413 list_del_init(&dlm->list);
414 spin_unlock(&dlm_domain_lock);
416 /* Wake up anyone waiting for us to remove this domain */
417 wake_up(&dlm_domain_events);
420 static int dlm_migrate_all_locks(struct dlm_ctxt *dlm)
422 int i, num, n, ret = 0;
423 struct dlm_lock_resource *res;
424 struct hlist_node *iter;
425 struct hlist_head *bucket;
428 mlog(0, "Migrating locks from domain %s\n", dlm->name);
431 spin_lock(&dlm->spinlock);
432 for (i = 0; i < DLM_HASH_BUCKETS; i++) {
435 bucket = dlm_lockres_hash(dlm, i);
436 iter = bucket->first;
439 res = hlist_entry(iter, struct dlm_lock_resource,
441 dlm_lockres_get(res);
442 /* migrate, if necessary. this will drop the dlm
443 * spinlock and retake it if it does migration. */
444 dropped = dlm_empty_lockres(dlm, res);
446 spin_lock(&res->spinlock);
447 __dlm_lockres_calc_usage(dlm, res);
448 iter = res->hash_node.next;
449 spin_unlock(&res->spinlock);
451 dlm_lockres_put(res);
456 cond_resched_lock(&dlm->spinlock);
458 mlog(0, "%s: touched %d lockreses in bucket %d "
459 "(tot=%d)\n", dlm->name, n, i, num);
461 spin_unlock(&dlm->spinlock);
462 wake_up(&dlm->dlm_thread_wq);
464 /* let the dlm thread take care of purging, keep scanning until
465 * nothing remains in the hash */
467 mlog(0, "%s: %d lock resources in hash last pass\n",
471 mlog(0, "DONE Migrating locks from domain %s\n", dlm->name);
475 static int dlm_no_joining_node(struct dlm_ctxt *dlm)
479 spin_lock(&dlm->spinlock);
480 ret = dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN;
481 spin_unlock(&dlm->spinlock);
486 static void dlm_mark_domain_leaving(struct dlm_ctxt *dlm)
488 /* Yikes, a double spinlock! I need domain_lock for the dlm
489 * state and the dlm spinlock for join state... Sorry! */
491 spin_lock(&dlm_domain_lock);
492 spin_lock(&dlm->spinlock);
494 if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
495 mlog(0, "Node %d is joining, we wait on it.\n",
497 spin_unlock(&dlm->spinlock);
498 spin_unlock(&dlm_domain_lock);
500 wait_event(dlm->dlm_join_events, dlm_no_joining_node(dlm));
504 dlm->dlm_state = DLM_CTXT_LEAVING;
505 spin_unlock(&dlm->spinlock);
506 spin_unlock(&dlm_domain_lock);
509 static void __dlm_print_nodes(struct dlm_ctxt *dlm)
513 assert_spin_locked(&dlm->spinlock);
515 printk(KERN_INFO "ocfs2_dlm: Nodes in domain (\"%s\"): ", dlm->name);
517 while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
518 node + 1)) < O2NM_MAX_NODES) {
524 static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
527 struct dlm_ctxt *dlm = data;
529 struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf;
531 mlog_entry("%p %u %p", msg, len, data);
536 node = exit_msg->node_idx;
538 printk(KERN_INFO "ocfs2_dlm: Node %u leaves domain %s\n", node, dlm->name);
540 spin_lock(&dlm->spinlock);
541 clear_bit(node, dlm->domain_map);
542 __dlm_print_nodes(dlm);
544 /* notify anything attached to the heartbeat events */
545 dlm_hb_event_notify_attached(dlm, node, 0);
547 spin_unlock(&dlm->spinlock);
554 static int dlm_send_one_domain_exit(struct dlm_ctxt *dlm,
558 struct dlm_exit_domain leave_msg;
560 mlog(0, "Asking node %u if we can leave the domain %s me = %u\n",
561 node, dlm->name, dlm->node_num);
563 memset(&leave_msg, 0, sizeof(leave_msg));
564 leave_msg.node_idx = dlm->node_num;
566 status = o2net_send_message(DLM_EXIT_DOMAIN_MSG, dlm->key,
567 &leave_msg, sizeof(leave_msg), node,
570 mlog(0, "status return %d from o2net_send_message\n", status);
576 static void dlm_leave_domain(struct dlm_ctxt *dlm)
578 int node, clear_node, status;
580 /* At this point we've migrated away all our locks and won't
581 * accept mastership of new ones. The dlm is responsible for
582 * almost nothing now. We make sure not to confuse any joining
583 * nodes and then commence shutdown procedure. */
585 spin_lock(&dlm->spinlock);
586 /* Clear ourselves from the domain map */
587 clear_bit(dlm->node_num, dlm->domain_map);
588 while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
589 0)) < O2NM_MAX_NODES) {
590 /* Drop the dlm spinlock. This is safe wrt the domain_map.
591 * -nodes cannot be added now as the
592 * query_join_handlers knows to respond with OK_NO_MAP
593 * -we catch the right network errors if a node is
594 * removed from the map while we're sending him the
596 spin_unlock(&dlm->spinlock);
600 status = dlm_send_one_domain_exit(dlm, node);
602 status != -ENOPROTOOPT &&
603 status != -ENOTCONN) {
604 mlog(ML_NOTICE, "Error %d sending domain exit message "
605 "to node %d\n", status, node);
607 /* Not sure what to do here but lets sleep for
608 * a bit in case this was a transient
610 msleep(DLM_DOMAIN_BACKOFF_MS);
614 spin_lock(&dlm->spinlock);
615 /* If we're not clearing the node bit then we intend
616 * to loop back around to try again. */
618 clear_bit(node, dlm->domain_map);
620 spin_unlock(&dlm->spinlock);
623 int dlm_joined(struct dlm_ctxt *dlm)
627 spin_lock(&dlm_domain_lock);
629 if (dlm->dlm_state == DLM_CTXT_JOINED)
632 spin_unlock(&dlm_domain_lock);
637 int dlm_shutting_down(struct dlm_ctxt *dlm)
641 spin_lock(&dlm_domain_lock);
643 if (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN)
646 spin_unlock(&dlm_domain_lock);
651 void dlm_unregister_domain(struct dlm_ctxt *dlm)
654 struct dlm_lock_resource *res;
656 spin_lock(&dlm_domain_lock);
657 BUG_ON(dlm->dlm_state != DLM_CTXT_JOINED);
658 BUG_ON(!dlm->num_joins);
661 if (!dlm->num_joins) {
662 /* We mark it "in shutdown" now so new register
663 * requests wait until we've completely left the
664 * domain. Don't use DLM_CTXT_LEAVING yet as we still
665 * want new domain joins to communicate with us at
666 * least until we've completed migration of our
668 dlm->dlm_state = DLM_CTXT_IN_SHUTDOWN;
671 spin_unlock(&dlm_domain_lock);
674 mlog(0, "shutting down domain %s\n", dlm->name);
676 /* We changed dlm state, notify the thread */
677 dlm_kick_thread(dlm, NULL);
679 while (dlm_migrate_all_locks(dlm)) {
680 /* Give dlm_thread time to purge the lockres' */
682 mlog(0, "%s: more migration to do\n", dlm->name);
685 /* This list should be empty. If not, print remaining lockres */
686 if (!list_empty(&dlm->tracking_list)) {
687 mlog(ML_ERROR, "Following lockres' are still on the "
689 list_for_each_entry(res, &dlm->tracking_list, tracking)
690 dlm_print_one_lock_resource(res);
693 dlm_mark_domain_leaving(dlm);
694 dlm_leave_domain(dlm);
695 dlm_complete_dlm_shutdown(dlm);
699 EXPORT_SYMBOL_GPL(dlm_unregister_domain);
701 static int dlm_query_join_proto_check(char *proto_type, int node,
702 struct dlm_protocol_version *ours,
703 struct dlm_protocol_version *request)
706 struct dlm_protocol_version proto = *request;
708 if (!dlm_protocol_compare(ours, &proto)) {
710 "node %u wanted to join with %s locking protocol "
711 "%u.%u, we respond with %u.%u\n",
715 proto.pv_major, proto.pv_minor);
716 request->pv_minor = proto.pv_minor;
720 "Node %u wanted to join with %s locking "
721 "protocol %u.%u, but we have %u.%u, disallowing\n",
734 * struct dlm_query_join_packet is made up of four one-byte fields. They
735 * are effectively in big-endian order already. However, little-endian
736 * machines swap them before putting the packet on the wire (because
737 * query_join's response is a status, and that status is treated as a u32
738 * on the wire). Thus, a big-endian and little-endian machines will treat
739 * this structure differently.
741 * The solution is to have little-endian machines swap the structure when
742 * converting from the structure to the u32 representation. This will
743 * result in the structure having the correct format on the wire no matter
744 * the host endian format.
746 static void dlm_query_join_packet_to_wire(struct dlm_query_join_packet *packet,
749 union dlm_query_join_response response;
751 response.packet = *packet;
752 *wire = cpu_to_be32(response.intval);
755 static void dlm_query_join_wire_to_packet(u32 wire,
756 struct dlm_query_join_packet *packet)
758 union dlm_query_join_response response;
760 response.intval = cpu_to_be32(wire);
761 *packet = response.packet;
764 static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
767 struct dlm_query_join_request *query;
768 struct dlm_query_join_packet packet = {
769 .code = JOIN_DISALLOW,
771 struct dlm_ctxt *dlm = NULL;
775 query = (struct dlm_query_join_request *) msg->buf;
777 mlog(0, "node %u wants to join domain %s\n", query->node_idx,
781 * If heartbeat doesn't consider the node live, tell it
782 * to back off and try again. This gives heartbeat a chance
785 if (!o2hb_check_node_heartbeating(query->node_idx)) {
786 mlog(0, "node %u is not in our live map yet\n",
789 packet.code = JOIN_DISALLOW;
793 packet.code = JOIN_OK_NO_MAP;
795 spin_lock(&dlm_domain_lock);
796 dlm = __dlm_lookup_domain_full(query->domain, query->name_len);
801 * There is a small window where the joining node may not see the
802 * node(s) that just left but still part of the cluster. DISALLOW
803 * join request if joining node has different node map.
806 while (nodenum < O2NM_MAX_NODES) {
807 if (test_bit(nodenum, dlm->domain_map)) {
808 if (!byte_test_bit(nodenum, query->node_map)) {
809 mlog(0, "disallow join as node %u does not "
810 "have node %u in its nodemap\n",
811 query->node_idx, nodenum);
812 packet.code = JOIN_DISALLOW;
819 /* Once the dlm ctxt is marked as leaving then we don't want
820 * to be put in someone's domain map.
821 * Also, explicitly disallow joining at certain troublesome
822 * times (ie. during recovery). */
823 if (dlm && dlm->dlm_state != DLM_CTXT_LEAVING) {
824 int bit = query->node_idx;
825 spin_lock(&dlm->spinlock);
827 if (dlm->dlm_state == DLM_CTXT_NEW &&
828 dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN) {
829 /*If this is a brand new context and we
830 * haven't started our join process yet, then
831 * the other node won the race. */
832 packet.code = JOIN_OK_NO_MAP;
833 } else if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
834 /* Disallow parallel joins. */
835 packet.code = JOIN_DISALLOW;
836 } else if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
837 mlog(0, "node %u trying to join, but recovery "
838 "is ongoing.\n", bit);
839 packet.code = JOIN_DISALLOW;
840 } else if (test_bit(bit, dlm->recovery_map)) {
841 mlog(0, "node %u trying to join, but it "
842 "still needs recovery.\n", bit);
843 packet.code = JOIN_DISALLOW;
844 } else if (test_bit(bit, dlm->domain_map)) {
845 mlog(0, "node %u trying to join, but it "
846 "is still in the domain! needs recovery?\n",
848 packet.code = JOIN_DISALLOW;
850 /* Alright we're fully a part of this domain
851 * so we keep some state as to who's joining
852 * and indicate to him that needs to be fixed
855 /* Make sure we speak compatible locking protocols. */
856 if (dlm_query_join_proto_check("DLM", bit,
857 &dlm->dlm_locking_proto,
858 &query->dlm_proto)) {
859 packet.code = JOIN_PROTOCOL_MISMATCH;
860 } else if (dlm_query_join_proto_check("fs", bit,
861 &dlm->fs_locking_proto,
863 packet.code = JOIN_PROTOCOL_MISMATCH;
865 packet.dlm_minor = query->dlm_proto.pv_minor;
866 packet.fs_minor = query->fs_proto.pv_minor;
867 packet.code = JOIN_OK;
868 __dlm_set_joining_node(dlm, query->node_idx);
872 spin_unlock(&dlm->spinlock);
875 spin_unlock(&dlm_domain_lock);
878 mlog(0, "We respond with %u\n", packet.code);
880 dlm_query_join_packet_to_wire(&packet, &response);
884 static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
887 struct dlm_assert_joined *assert;
888 struct dlm_ctxt *dlm = NULL;
890 assert = (struct dlm_assert_joined *) msg->buf;
892 mlog(0, "node %u asserts join on domain %s\n", assert->node_idx,
895 spin_lock(&dlm_domain_lock);
896 dlm = __dlm_lookup_domain_full(assert->domain, assert->name_len);
897 /* XXX should we consider no dlm ctxt an error? */
899 spin_lock(&dlm->spinlock);
901 /* Alright, this node has officially joined our
902 * domain. Set him in the map and clean up our
903 * leftover join state. */
904 BUG_ON(dlm->joining_node != assert->node_idx);
905 set_bit(assert->node_idx, dlm->domain_map);
906 __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
908 printk(KERN_INFO "ocfs2_dlm: Node %u joins domain %s\n",
909 assert->node_idx, dlm->name);
910 __dlm_print_nodes(dlm);
912 /* notify anything attached to the heartbeat events */
913 dlm_hb_event_notify_attached(dlm, assert->node_idx, 1);
915 spin_unlock(&dlm->spinlock);
917 spin_unlock(&dlm_domain_lock);
922 static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
925 struct dlm_cancel_join *cancel;
926 struct dlm_ctxt *dlm = NULL;
928 cancel = (struct dlm_cancel_join *) msg->buf;
930 mlog(0, "node %u cancels join on domain %s\n", cancel->node_idx,
933 spin_lock(&dlm_domain_lock);
934 dlm = __dlm_lookup_domain_full(cancel->domain, cancel->name_len);
937 spin_lock(&dlm->spinlock);
939 /* Yikes, this guy wants to cancel his join. No
940 * problem, we simply cleanup our join state. */
941 BUG_ON(dlm->joining_node != cancel->node_idx);
942 __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
944 spin_unlock(&dlm->spinlock);
946 spin_unlock(&dlm_domain_lock);
951 static int dlm_send_one_join_cancel(struct dlm_ctxt *dlm,
955 struct dlm_cancel_join cancel_msg;
957 memset(&cancel_msg, 0, sizeof(cancel_msg));
958 cancel_msg.node_idx = dlm->node_num;
959 cancel_msg.name_len = strlen(dlm->name);
960 memcpy(cancel_msg.domain, dlm->name, cancel_msg.name_len);
962 status = o2net_send_message(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
963 &cancel_msg, sizeof(cancel_msg), node,
974 /* map_size should be in bytes. */
975 static int dlm_send_join_cancels(struct dlm_ctxt *dlm,
976 unsigned long *node_map,
977 unsigned int map_size)
982 if (map_size != (BITS_TO_LONGS(O2NM_MAX_NODES) *
983 sizeof(unsigned long))) {
985 "map_size %u != BITS_TO_LONGS(O2NM_MAX_NODES) %u\n",
986 map_size, (unsigned)BITS_TO_LONGS(O2NM_MAX_NODES));
992 while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
993 node + 1)) < O2NM_MAX_NODES) {
994 if (node == dlm->node_num)
997 tmpstat = dlm_send_one_join_cancel(dlm, node);
999 mlog(ML_ERROR, "Error return %d cancelling join on "
1000 "node %d\n", tmpstat, node);
1011 static int dlm_request_join(struct dlm_ctxt *dlm,
1013 enum dlm_query_join_response_code *response)
1016 struct dlm_query_join_request join_msg;
1017 struct dlm_query_join_packet packet;
1020 mlog(0, "querying node %d\n", node);
1022 memset(&join_msg, 0, sizeof(join_msg));
1023 join_msg.node_idx = dlm->node_num;
1024 join_msg.name_len = strlen(dlm->name);
1025 memcpy(join_msg.domain, dlm->name, join_msg.name_len);
1026 join_msg.dlm_proto = dlm->dlm_locking_proto;
1027 join_msg.fs_proto = dlm->fs_locking_proto;
1029 /* copy live node map to join message */
1030 byte_copymap(join_msg.node_map, dlm->live_nodes_map, O2NM_MAX_NODES);
1032 status = o2net_send_message(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, &join_msg,
1033 sizeof(join_msg), node,
1035 if (status < 0 && status != -ENOPROTOOPT) {
1039 dlm_query_join_wire_to_packet(join_resp, &packet);
1041 /* -ENOPROTOOPT from the net code means the other side isn't
1042 listening for our message type -- that's fine, it means
1043 his dlm isn't up, so we can consider him a 'yes' but not
1044 joined into the domain. */
1045 if (status == -ENOPROTOOPT) {
1047 *response = JOIN_OK_NO_MAP;
1048 } else if (packet.code == JOIN_DISALLOW ||
1049 packet.code == JOIN_OK_NO_MAP) {
1050 *response = packet.code;
1051 } else if (packet.code == JOIN_PROTOCOL_MISMATCH) {
1053 "This node requested DLM locking protocol %u.%u and "
1054 "filesystem locking protocol %u.%u. At least one of "
1055 "the protocol versions on node %d is not compatible, "
1057 dlm->dlm_locking_proto.pv_major,
1058 dlm->dlm_locking_proto.pv_minor,
1059 dlm->fs_locking_proto.pv_major,
1060 dlm->fs_locking_proto.pv_minor,
1063 *response = packet.code;
1064 } else if (packet.code == JOIN_OK) {
1065 *response = packet.code;
1066 /* Use the same locking protocol as the remote node */
1067 dlm->dlm_locking_proto.pv_minor = packet.dlm_minor;
1068 dlm->fs_locking_proto.pv_minor = packet.fs_minor;
1070 "Node %d responds JOIN_OK with DLM locking protocol "
1071 "%u.%u and fs locking protocol %u.%u\n",
1073 dlm->dlm_locking_proto.pv_major,
1074 dlm->dlm_locking_proto.pv_minor,
1075 dlm->fs_locking_proto.pv_major,
1076 dlm->fs_locking_proto.pv_minor);
1079 mlog(ML_ERROR, "invalid response %d from node %u\n",
1083 mlog(0, "status %d, node %d response is %d\n", status, node,
1090 static int dlm_send_one_join_assert(struct dlm_ctxt *dlm,
1094 struct dlm_assert_joined assert_msg;
1096 mlog(0, "Sending join assert to node %u\n", node);
1098 memset(&assert_msg, 0, sizeof(assert_msg));
1099 assert_msg.node_idx = dlm->node_num;
1100 assert_msg.name_len = strlen(dlm->name);
1101 memcpy(assert_msg.domain, dlm->name, assert_msg.name_len);
1103 status = o2net_send_message(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1104 &assert_msg, sizeof(assert_msg), node,
1112 static void dlm_send_join_asserts(struct dlm_ctxt *dlm,
1113 unsigned long *node_map)
1115 int status, node, live;
1119 while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
1120 node + 1)) < O2NM_MAX_NODES) {
1121 if (node == dlm->node_num)
1125 /* It is very important that this message be
1126 * received so we spin until either the node
1127 * has died or it gets the message. */
1128 status = dlm_send_one_join_assert(dlm, node);
1130 spin_lock(&dlm->spinlock);
1131 live = test_bit(node, dlm->live_nodes_map);
1132 spin_unlock(&dlm->spinlock);
1135 mlog(ML_ERROR, "Error return %d asserting "
1136 "join on node %d\n", status, node);
1138 /* give us some time between errors... */
1140 msleep(DLM_DOMAIN_BACKOFF_MS);
1142 } while (status && live);
1146 struct domain_join_ctxt {
1147 unsigned long live_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1148 unsigned long yes_resp_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1151 static int dlm_should_restart_join(struct dlm_ctxt *dlm,
1152 struct domain_join_ctxt *ctxt,
1153 enum dlm_query_join_response_code response)
1157 if (response == JOIN_DISALLOW) {
1158 mlog(0, "Latest response of disallow -- should restart\n");
1162 spin_lock(&dlm->spinlock);
1163 /* For now, we restart the process if the node maps have
1165 ret = memcmp(ctxt->live_map, dlm->live_nodes_map,
1166 sizeof(dlm->live_nodes_map));
1167 spin_unlock(&dlm->spinlock);
1170 mlog(0, "Node maps changed -- should restart\n");
1175 static int dlm_try_to_join_domain(struct dlm_ctxt *dlm)
1177 int status = 0, tmpstat, node;
1178 struct domain_join_ctxt *ctxt;
1179 enum dlm_query_join_response_code response = JOIN_DISALLOW;
1181 mlog_entry("%p", dlm);
1183 ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
1190 /* group sem locking should work for us here -- we're already
1191 * registered for heartbeat events so filling this should be
1192 * atomic wrt getting those handlers called. */
1193 o2hb_fill_node_map(dlm->live_nodes_map, sizeof(dlm->live_nodes_map));
1195 spin_lock(&dlm->spinlock);
1196 memcpy(ctxt->live_map, dlm->live_nodes_map, sizeof(ctxt->live_map));
1198 __dlm_set_joining_node(dlm, dlm->node_num);
1200 spin_unlock(&dlm->spinlock);
1203 while ((node = find_next_bit(ctxt->live_map, O2NM_MAX_NODES,
1204 node + 1)) < O2NM_MAX_NODES) {
1205 if (node == dlm->node_num)
1208 status = dlm_request_join(dlm, node, &response);
1214 /* Ok, either we got a response or the node doesn't have a
1216 if (response == JOIN_OK)
1217 set_bit(node, ctxt->yes_resp_map);
1219 if (dlm_should_restart_join(dlm, ctxt, response)) {
1225 mlog(0, "Yay, done querying nodes!\n");
1227 /* Yay, everyone agree's we can join the domain. My domain is
1228 * comprised of all nodes who were put in the
1229 * yes_resp_map. Copy that into our domain map and send a join
1230 * assert message to clean up everyone elses state. */
1231 spin_lock(&dlm->spinlock);
1232 memcpy(dlm->domain_map, ctxt->yes_resp_map,
1233 sizeof(ctxt->yes_resp_map));
1234 set_bit(dlm->node_num, dlm->domain_map);
1235 spin_unlock(&dlm->spinlock);
1237 dlm_send_join_asserts(dlm, ctxt->yes_resp_map);
1239 /* Joined state *must* be set before the joining node
1240 * information, otherwise the query_join handler may read no
1241 * current joiner but a state of NEW and tell joining nodes
1242 * we're not in the domain. */
1243 spin_lock(&dlm_domain_lock);
1244 dlm->dlm_state = DLM_CTXT_JOINED;
1246 spin_unlock(&dlm_domain_lock);
1249 spin_lock(&dlm->spinlock);
1250 __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
1252 __dlm_print_nodes(dlm);
1253 spin_unlock(&dlm->spinlock);
1256 /* Do we need to send a cancel message to any nodes? */
1258 tmpstat = dlm_send_join_cancels(dlm,
1260 sizeof(ctxt->yes_resp_map));
1262 mlog_errno(tmpstat);
1267 mlog(0, "returning %d\n", status);
1271 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm)
1273 o2hb_unregister_callback(NULL, &dlm->dlm_hb_up);
1274 o2hb_unregister_callback(NULL, &dlm->dlm_hb_down);
1275 o2net_unregister_handler_list(&dlm->dlm_domain_handlers);
1278 static int dlm_register_domain_handlers(struct dlm_ctxt *dlm)
1282 mlog(0, "registering handlers.\n");
1284 o2hb_setup_callback(&dlm->dlm_hb_down, O2HB_NODE_DOWN_CB,
1285 dlm_hb_node_down_cb, dlm, DLM_HB_NODE_DOWN_PRI);
1286 status = o2hb_register_callback(NULL, &dlm->dlm_hb_down);
1290 o2hb_setup_callback(&dlm->dlm_hb_up, O2HB_NODE_UP_CB,
1291 dlm_hb_node_up_cb, dlm, DLM_HB_NODE_UP_PRI);
1292 status = o2hb_register_callback(NULL, &dlm->dlm_hb_up);
1296 status = o2net_register_handler(DLM_MASTER_REQUEST_MSG, dlm->key,
1297 sizeof(struct dlm_master_request),
1298 dlm_master_request_handler,
1299 dlm, NULL, &dlm->dlm_domain_handlers);
1303 status = o2net_register_handler(DLM_ASSERT_MASTER_MSG, dlm->key,
1304 sizeof(struct dlm_assert_master),
1305 dlm_assert_master_handler,
1306 dlm, dlm_assert_master_post_handler,
1307 &dlm->dlm_domain_handlers);
1311 status = o2net_register_handler(DLM_CREATE_LOCK_MSG, dlm->key,
1312 sizeof(struct dlm_create_lock),
1313 dlm_create_lock_handler,
1314 dlm, NULL, &dlm->dlm_domain_handlers);
1318 status = o2net_register_handler(DLM_CONVERT_LOCK_MSG, dlm->key,
1319 DLM_CONVERT_LOCK_MAX_LEN,
1320 dlm_convert_lock_handler,
1321 dlm, NULL, &dlm->dlm_domain_handlers);
1325 status = o2net_register_handler(DLM_UNLOCK_LOCK_MSG, dlm->key,
1326 DLM_UNLOCK_LOCK_MAX_LEN,
1327 dlm_unlock_lock_handler,
1328 dlm, NULL, &dlm->dlm_domain_handlers);
1332 status = o2net_register_handler(DLM_PROXY_AST_MSG, dlm->key,
1333 DLM_PROXY_AST_MAX_LEN,
1334 dlm_proxy_ast_handler,
1335 dlm, NULL, &dlm->dlm_domain_handlers);
1339 status = o2net_register_handler(DLM_EXIT_DOMAIN_MSG, dlm->key,
1340 sizeof(struct dlm_exit_domain),
1341 dlm_exit_domain_handler,
1342 dlm, NULL, &dlm->dlm_domain_handlers);
1346 status = o2net_register_handler(DLM_DEREF_LOCKRES_MSG, dlm->key,
1347 sizeof(struct dlm_deref_lockres),
1348 dlm_deref_lockres_handler,
1349 dlm, NULL, &dlm->dlm_domain_handlers);
1353 status = o2net_register_handler(DLM_MIGRATE_REQUEST_MSG, dlm->key,
1354 sizeof(struct dlm_migrate_request),
1355 dlm_migrate_request_handler,
1356 dlm, NULL, &dlm->dlm_domain_handlers);
1360 status = o2net_register_handler(DLM_MIG_LOCKRES_MSG, dlm->key,
1361 DLM_MIG_LOCKRES_MAX_LEN,
1362 dlm_mig_lockres_handler,
1363 dlm, NULL, &dlm->dlm_domain_handlers);
1367 status = o2net_register_handler(DLM_MASTER_REQUERY_MSG, dlm->key,
1368 sizeof(struct dlm_master_requery),
1369 dlm_master_requery_handler,
1370 dlm, NULL, &dlm->dlm_domain_handlers);
1374 status = o2net_register_handler(DLM_LOCK_REQUEST_MSG, dlm->key,
1375 sizeof(struct dlm_lock_request),
1376 dlm_request_all_locks_handler,
1377 dlm, NULL, &dlm->dlm_domain_handlers);
1381 status = o2net_register_handler(DLM_RECO_DATA_DONE_MSG, dlm->key,
1382 sizeof(struct dlm_reco_data_done),
1383 dlm_reco_data_done_handler,
1384 dlm, NULL, &dlm->dlm_domain_handlers);
1388 status = o2net_register_handler(DLM_BEGIN_RECO_MSG, dlm->key,
1389 sizeof(struct dlm_begin_reco),
1390 dlm_begin_reco_handler,
1391 dlm, NULL, &dlm->dlm_domain_handlers);
1395 status = o2net_register_handler(DLM_FINALIZE_RECO_MSG, dlm->key,
1396 sizeof(struct dlm_finalize_reco),
1397 dlm_finalize_reco_handler,
1398 dlm, NULL, &dlm->dlm_domain_handlers);
1404 dlm_unregister_domain_handlers(dlm);
1409 static int dlm_join_domain(struct dlm_ctxt *dlm)
1412 unsigned int backoff;
1413 unsigned int total_backoff = 0;
1417 mlog(0, "Join domain %s\n", dlm->name);
1419 status = dlm_register_domain_handlers(dlm);
1425 status = dlm_debug_init(dlm);
1431 status = dlm_launch_thread(dlm);
1437 status = dlm_launch_recovery_thread(dlm);
1443 dlm->dlm_worker = create_singlethread_workqueue("dlm_wq");
1444 if (!dlm->dlm_worker) {
1451 status = dlm_try_to_join_domain(dlm);
1453 /* If we're racing another node to the join, then we
1454 * need to back off temporarily and let them
1456 #define DLM_JOIN_TIMEOUT_MSECS 90000
1457 if (status == -EAGAIN) {
1458 if (signal_pending(current)) {
1459 status = -ERESTARTSYS;
1464 msecs_to_jiffies(DLM_JOIN_TIMEOUT_MSECS)) {
1465 status = -ERESTARTSYS;
1466 mlog(ML_NOTICE, "Timed out joining dlm domain "
1467 "%s after %u msecs\n", dlm->name,
1468 jiffies_to_msecs(total_backoff));
1474 * <dale> No, after you!
1476 * <dale> But you first!
1479 backoff = (unsigned int)(jiffies & 0x3);
1480 backoff *= DLM_DOMAIN_BACKOFF_MS;
1481 total_backoff += backoff;
1482 mlog(0, "backoff %d\n", backoff);
1485 } while (status == -EAGAIN);
1494 wake_up(&dlm_domain_events);
1497 dlm_unregister_domain_handlers(dlm);
1498 dlm_debug_shutdown(dlm);
1499 dlm_complete_thread(dlm);
1500 dlm_complete_recovery_thread(dlm);
1501 dlm_destroy_dlm_worker(dlm);
1507 static struct dlm_ctxt *dlm_alloc_ctxt(const char *domain,
1512 struct dlm_ctxt *dlm = NULL;
1514 dlm = kzalloc(sizeof(*dlm), GFP_KERNEL);
1516 mlog_errno(-ENOMEM);
1520 dlm->name = kmalloc(strlen(domain) + 1, GFP_KERNEL);
1521 if (dlm->name == NULL) {
1522 mlog_errno(-ENOMEM);
1528 dlm->lockres_hash = (struct hlist_head **)dlm_alloc_pagevec(DLM_HASH_PAGES);
1529 if (!dlm->lockres_hash) {
1530 mlog_errno(-ENOMEM);
1537 for (i = 0; i < DLM_HASH_BUCKETS; i++)
1538 INIT_HLIST_HEAD(dlm_lockres_hash(dlm, i));
1540 dlm->master_hash = (struct hlist_head **)
1541 dlm_alloc_pagevec(DLM_HASH_PAGES);
1542 if (!dlm->master_hash) {
1543 mlog_errno(-ENOMEM);
1544 dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
1551 for (i = 0; i < DLM_HASH_BUCKETS; i++)
1552 INIT_HLIST_HEAD(dlm_master_hash(dlm, i));
1554 strcpy(dlm->name, domain);
1556 dlm->node_num = o2nm_this_node();
1558 ret = dlm_create_debugfs_subroot(dlm);
1560 dlm_free_pagevec((void **)dlm->master_hash, DLM_HASH_PAGES);
1561 dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
1568 spin_lock_init(&dlm->spinlock);
1569 spin_lock_init(&dlm->master_lock);
1570 spin_lock_init(&dlm->ast_lock);
1571 spin_lock_init(&dlm->track_lock);
1572 INIT_LIST_HEAD(&dlm->list);
1573 INIT_LIST_HEAD(&dlm->dirty_list);
1574 INIT_LIST_HEAD(&dlm->reco.resources);
1575 INIT_LIST_HEAD(&dlm->reco.received);
1576 INIT_LIST_HEAD(&dlm->reco.node_data);
1577 INIT_LIST_HEAD(&dlm->purge_list);
1578 INIT_LIST_HEAD(&dlm->dlm_domain_handlers);
1579 INIT_LIST_HEAD(&dlm->tracking_list);
1580 dlm->reco.state = 0;
1582 INIT_LIST_HEAD(&dlm->pending_asts);
1583 INIT_LIST_HEAD(&dlm->pending_basts);
1585 mlog(0, "dlm->recovery_map=%p, &(dlm->recovery_map[0])=%p\n",
1586 dlm->recovery_map, &(dlm->recovery_map[0]));
1588 memset(dlm->recovery_map, 0, sizeof(dlm->recovery_map));
1589 memset(dlm->live_nodes_map, 0, sizeof(dlm->live_nodes_map));
1590 memset(dlm->domain_map, 0, sizeof(dlm->domain_map));
1592 dlm->dlm_thread_task = NULL;
1593 dlm->dlm_reco_thread_task = NULL;
1594 dlm->dlm_worker = NULL;
1595 init_waitqueue_head(&dlm->dlm_thread_wq);
1596 init_waitqueue_head(&dlm->dlm_reco_thread_wq);
1597 init_waitqueue_head(&dlm->reco.event);
1598 init_waitqueue_head(&dlm->ast_wq);
1599 init_waitqueue_head(&dlm->migration_wq);
1600 INIT_LIST_HEAD(&dlm->mle_hb_events);
1602 dlm->joining_node = DLM_LOCK_RES_OWNER_UNKNOWN;
1603 init_waitqueue_head(&dlm->dlm_join_events);
1605 dlm->reco.new_master = O2NM_INVALID_NODE_NUM;
1606 dlm->reco.dead_node = O2NM_INVALID_NODE_NUM;
1608 atomic_set(&dlm->res_tot_count, 0);
1609 atomic_set(&dlm->res_cur_count, 0);
1610 for (i = 0; i < DLM_MLE_NUM_TYPES; ++i) {
1611 atomic_set(&dlm->mle_tot_count[i], 0);
1612 atomic_set(&dlm->mle_cur_count[i], 0);
1615 spin_lock_init(&dlm->work_lock);
1616 INIT_LIST_HEAD(&dlm->work_list);
1617 INIT_WORK(&dlm->dispatched_work, dlm_dispatch_work);
1619 kref_init(&dlm->dlm_refs);
1620 dlm->dlm_state = DLM_CTXT_NEW;
1622 INIT_LIST_HEAD(&dlm->dlm_eviction_callbacks);
1624 mlog(0, "context init: refcount %u\n",
1625 atomic_read(&dlm->dlm_refs.refcount));
1632 * Compare a requested locking protocol version against the current one.
1634 * If the major numbers are different, they are incompatible.
1635 * If the current minor is greater than the request, they are incompatible.
1636 * If the current minor is less than or equal to the request, they are
1637 * compatible, and the requester should run at the current minor version.
1639 static int dlm_protocol_compare(struct dlm_protocol_version *existing,
1640 struct dlm_protocol_version *request)
1642 if (existing->pv_major != request->pv_major)
1645 if (existing->pv_minor > request->pv_minor)
1648 if (existing->pv_minor < request->pv_minor)
1649 request->pv_minor = existing->pv_minor;
1655 * dlm_register_domain: one-time setup per "domain".
1657 * The filesystem passes in the requested locking version via proto.
1658 * If registration was successful, proto will contain the negotiated
1661 struct dlm_ctxt * dlm_register_domain(const char *domain,
1663 struct dlm_protocol_version *fs_proto)
1666 struct dlm_ctxt *dlm = NULL;
1667 struct dlm_ctxt *new_ctxt = NULL;
1669 if (strlen(domain) > O2NM_MAX_NAME_LEN) {
1670 ret = -ENAMETOOLONG;
1671 mlog(ML_ERROR, "domain name length too long\n");
1675 if (!o2hb_check_local_node_heartbeating()) {
1676 mlog(ML_ERROR, "the local node has not been configured, or is "
1677 "not heartbeating\n");
1682 mlog(0, "register called for domain \"%s\"\n", domain);
1686 if (signal_pending(current)) {
1692 spin_lock(&dlm_domain_lock);
1694 dlm = __dlm_lookup_domain(domain);
1696 if (dlm->dlm_state != DLM_CTXT_JOINED) {
1697 spin_unlock(&dlm_domain_lock);
1699 mlog(0, "This ctxt is not joined yet!\n");
1700 wait_event_interruptible(dlm_domain_events,
1701 dlm_wait_on_domain_helper(
1706 if (dlm_protocol_compare(&dlm->fs_locking_proto, fs_proto)) {
1708 "Requested locking protocol version is not "
1709 "compatible with already registered domain "
1710 "\"%s\"\n", domain);
1718 spin_unlock(&dlm_domain_lock);
1726 spin_unlock(&dlm_domain_lock);
1728 new_ctxt = dlm_alloc_ctxt(domain, key);
1737 /* a little variable switch-a-roo here... */
1741 /* add the new domain */
1742 list_add_tail(&dlm->list, &dlm_domains);
1743 spin_unlock(&dlm_domain_lock);
1746 * Pass the locking protocol version into the join. If the join
1747 * succeeds, it will have the negotiated protocol set.
1749 dlm->dlm_locking_proto = dlm_protocol;
1750 dlm->fs_locking_proto = *fs_proto;
1752 ret = dlm_join_domain(dlm);
1759 /* Tell the caller what locking protocol we negotiated */
1760 *fs_proto = dlm->fs_locking_proto;
1765 dlm_free_ctxt_mem(new_ctxt);
1772 EXPORT_SYMBOL_GPL(dlm_register_domain);
1774 static LIST_HEAD(dlm_join_handlers);
1776 static void dlm_unregister_net_handlers(void)
1778 o2net_unregister_handler_list(&dlm_join_handlers);
1781 static int dlm_register_net_handlers(void)
1785 status = o2net_register_handler(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
1786 sizeof(struct dlm_query_join_request),
1787 dlm_query_join_handler,
1788 NULL, NULL, &dlm_join_handlers);
1792 status = o2net_register_handler(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1793 sizeof(struct dlm_assert_joined),
1794 dlm_assert_joined_handler,
1795 NULL, NULL, &dlm_join_handlers);
1799 status = o2net_register_handler(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
1800 sizeof(struct dlm_cancel_join),
1801 dlm_cancel_join_handler,
1802 NULL, NULL, &dlm_join_handlers);
1806 dlm_unregister_net_handlers();
1811 /* Domain eviction callback handling.
1813 * The file system requires notification of node death *before* the
1814 * dlm completes it's recovery work, otherwise it may be able to
1815 * acquire locks on resources requiring recovery. Since the dlm can
1816 * evict a node from it's domain *before* heartbeat fires, a similar
1817 * mechanism is required. */
1819 /* Eviction is not expected to happen often, so a per-domain lock is
1820 * not necessary. Eviction callbacks are allowed to sleep for short
1821 * periods of time. */
1822 static DECLARE_RWSEM(dlm_callback_sem);
1824 void dlm_fire_domain_eviction_callbacks(struct dlm_ctxt *dlm,
1827 struct list_head *iter;
1828 struct dlm_eviction_cb *cb;
1830 down_read(&dlm_callback_sem);
1831 list_for_each(iter, &dlm->dlm_eviction_callbacks) {
1832 cb = list_entry(iter, struct dlm_eviction_cb, ec_item);
1834 cb->ec_func(node_num, cb->ec_data);
1836 up_read(&dlm_callback_sem);
1839 void dlm_setup_eviction_cb(struct dlm_eviction_cb *cb,
1840 dlm_eviction_func *f,
1843 INIT_LIST_HEAD(&cb->ec_item);
1847 EXPORT_SYMBOL_GPL(dlm_setup_eviction_cb);
1849 void dlm_register_eviction_cb(struct dlm_ctxt *dlm,
1850 struct dlm_eviction_cb *cb)
1852 down_write(&dlm_callback_sem);
1853 list_add_tail(&cb->ec_item, &dlm->dlm_eviction_callbacks);
1854 up_write(&dlm_callback_sem);
1856 EXPORT_SYMBOL_GPL(dlm_register_eviction_cb);
1858 void dlm_unregister_eviction_cb(struct dlm_eviction_cb *cb)
1860 down_write(&dlm_callback_sem);
1861 list_del_init(&cb->ec_item);
1862 up_write(&dlm_callback_sem);
1864 EXPORT_SYMBOL_GPL(dlm_unregister_eviction_cb);
1866 static int __init dlm_init(void)
1870 dlm_print_version();
1872 status = dlm_init_mle_cache();
1874 mlog(ML_ERROR, "Could not create o2dlm_mle slabcache\n");
1878 status = dlm_init_master_caches();
1880 mlog(ML_ERROR, "Could not create o2dlm_lockres and "
1881 "o2dlm_lockname slabcaches\n");
1885 status = dlm_init_lock_cache();
1887 mlog(ML_ERROR, "Count not create o2dlm_lock slabcache\n");
1891 status = dlm_register_net_handlers();
1893 mlog(ML_ERROR, "Unable to register network handlers\n");
1897 status = dlm_create_debugfs_root();
1903 dlm_unregister_net_handlers();
1904 dlm_destroy_lock_cache();
1905 dlm_destroy_master_caches();
1906 dlm_destroy_mle_cache();
1910 static void __exit dlm_exit (void)
1912 dlm_destroy_debugfs_root();
1913 dlm_unregister_net_handlers();
1914 dlm_destroy_lock_cache();
1915 dlm_destroy_master_caches();
1916 dlm_destroy_mle_cache();
1919 MODULE_AUTHOR("Oracle");
1920 MODULE_LICENSE("GPL");
1922 module_init(dlm_init);
1923 module_exit(dlm_exit);