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);
313 /* A little strange - this function will be called while holding
314 * dlm_domain_lock and is expected to be holding it on the way out. We
315 * will however drop and reacquire it multiple times */
316 static void dlm_ctxt_release(struct kref *kref)
318 struct dlm_ctxt *dlm;
320 dlm = container_of(kref, struct dlm_ctxt, dlm_refs);
322 BUG_ON(dlm->num_joins);
323 BUG_ON(dlm->dlm_state == DLM_CTXT_JOINED);
325 /* we may still be in the list if we hit an error during join. */
326 list_del_init(&dlm->list);
328 spin_unlock(&dlm_domain_lock);
330 mlog(0, "freeing memory from domain %s\n", dlm->name);
332 wake_up(&dlm_domain_events);
334 dlm_free_ctxt_mem(dlm);
336 spin_lock(&dlm_domain_lock);
339 void dlm_put(struct dlm_ctxt *dlm)
341 spin_lock(&dlm_domain_lock);
342 kref_put(&dlm->dlm_refs, dlm_ctxt_release);
343 spin_unlock(&dlm_domain_lock);
346 static void __dlm_get(struct dlm_ctxt *dlm)
348 kref_get(&dlm->dlm_refs);
351 /* given a questionable reference to a dlm object, gets a reference if
352 * it can find it in the list, otherwise returns NULL in which case
353 * you shouldn't trust your pointer. */
354 struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm)
356 struct list_head *iter;
357 struct dlm_ctxt *target = NULL;
359 spin_lock(&dlm_domain_lock);
361 list_for_each(iter, &dlm_domains) {
362 target = list_entry (iter, struct dlm_ctxt, list);
372 spin_unlock(&dlm_domain_lock);
377 int dlm_domain_fully_joined(struct dlm_ctxt *dlm)
381 spin_lock(&dlm_domain_lock);
382 ret = (dlm->dlm_state == DLM_CTXT_JOINED) ||
383 (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN);
384 spin_unlock(&dlm_domain_lock);
389 static void dlm_destroy_dlm_worker(struct dlm_ctxt *dlm)
391 if (dlm->dlm_worker) {
392 flush_workqueue(dlm->dlm_worker);
393 destroy_workqueue(dlm->dlm_worker);
394 dlm->dlm_worker = NULL;
398 static void dlm_complete_dlm_shutdown(struct dlm_ctxt *dlm)
400 dlm_unregister_domain_handlers(dlm);
401 dlm_complete_thread(dlm);
402 dlm_complete_recovery_thread(dlm);
403 dlm_destroy_dlm_worker(dlm);
405 /* We've left the domain. Now we can take ourselves out of the
406 * list and allow the kref stuff to help us free the
408 spin_lock(&dlm_domain_lock);
409 list_del_init(&dlm->list);
410 spin_unlock(&dlm_domain_lock);
412 /* Wake up anyone waiting for us to remove this domain */
413 wake_up(&dlm_domain_events);
416 static int dlm_migrate_all_locks(struct dlm_ctxt *dlm)
418 int i, num, n, ret = 0;
419 struct dlm_lock_resource *res;
420 struct hlist_node *iter;
421 struct hlist_head *bucket;
424 mlog(0, "Migrating locks from domain %s\n", dlm->name);
427 spin_lock(&dlm->spinlock);
428 for (i = 0; i < DLM_HASH_BUCKETS; i++) {
431 bucket = dlm_lockres_hash(dlm, i);
432 iter = bucket->first;
435 res = hlist_entry(iter, struct dlm_lock_resource,
437 dlm_lockres_get(res);
438 /* migrate, if necessary. this will drop the dlm
439 * spinlock and retake it if it does migration. */
440 dropped = dlm_empty_lockres(dlm, res);
442 spin_lock(&res->spinlock);
443 __dlm_lockres_calc_usage(dlm, res);
444 iter = res->hash_node.next;
445 spin_unlock(&res->spinlock);
447 dlm_lockres_put(res);
452 cond_resched_lock(&dlm->spinlock);
454 mlog(0, "%s: touched %d lockreses in bucket %d "
455 "(tot=%d)\n", dlm->name, n, i, num);
457 spin_unlock(&dlm->spinlock);
458 wake_up(&dlm->dlm_thread_wq);
460 /* let the dlm thread take care of purging, keep scanning until
461 * nothing remains in the hash */
463 mlog(0, "%s: %d lock resources in hash last pass\n",
467 mlog(0, "DONE Migrating locks from domain %s\n", dlm->name);
471 static int dlm_no_joining_node(struct dlm_ctxt *dlm)
475 spin_lock(&dlm->spinlock);
476 ret = dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN;
477 spin_unlock(&dlm->spinlock);
482 static void dlm_mark_domain_leaving(struct dlm_ctxt *dlm)
484 /* Yikes, a double spinlock! I need domain_lock for the dlm
485 * state and the dlm spinlock for join state... Sorry! */
487 spin_lock(&dlm_domain_lock);
488 spin_lock(&dlm->spinlock);
490 if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
491 mlog(0, "Node %d is joining, we wait on it.\n",
493 spin_unlock(&dlm->spinlock);
494 spin_unlock(&dlm_domain_lock);
496 wait_event(dlm->dlm_join_events, dlm_no_joining_node(dlm));
500 dlm->dlm_state = DLM_CTXT_LEAVING;
501 spin_unlock(&dlm->spinlock);
502 spin_unlock(&dlm_domain_lock);
505 static void __dlm_print_nodes(struct dlm_ctxt *dlm)
509 assert_spin_locked(&dlm->spinlock);
511 printk(KERN_INFO "ocfs2_dlm: Nodes in domain (\"%s\"): ", dlm->name);
513 while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
514 node + 1)) < O2NM_MAX_NODES) {
520 static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
523 struct dlm_ctxt *dlm = data;
525 struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf;
527 mlog_entry("%p %u %p", msg, len, data);
532 node = exit_msg->node_idx;
534 printk(KERN_INFO "ocfs2_dlm: Node %u leaves domain %s\n", node, dlm->name);
536 spin_lock(&dlm->spinlock);
537 clear_bit(node, dlm->domain_map);
538 __dlm_print_nodes(dlm);
540 /* notify anything attached to the heartbeat events */
541 dlm_hb_event_notify_attached(dlm, node, 0);
543 spin_unlock(&dlm->spinlock);
550 static int dlm_send_one_domain_exit(struct dlm_ctxt *dlm,
554 struct dlm_exit_domain leave_msg;
556 mlog(0, "Asking node %u if we can leave the domain %s me = %u\n",
557 node, dlm->name, dlm->node_num);
559 memset(&leave_msg, 0, sizeof(leave_msg));
560 leave_msg.node_idx = dlm->node_num;
562 status = o2net_send_message(DLM_EXIT_DOMAIN_MSG, dlm->key,
563 &leave_msg, sizeof(leave_msg), node,
566 mlog(0, "status return %d from o2net_send_message\n", status);
572 static void dlm_leave_domain(struct dlm_ctxt *dlm)
574 int node, clear_node, status;
576 /* At this point we've migrated away all our locks and won't
577 * accept mastership of new ones. The dlm is responsible for
578 * almost nothing now. We make sure not to confuse any joining
579 * nodes and then commence shutdown procedure. */
581 spin_lock(&dlm->spinlock);
582 /* Clear ourselves from the domain map */
583 clear_bit(dlm->node_num, dlm->domain_map);
584 while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
585 0)) < O2NM_MAX_NODES) {
586 /* Drop the dlm spinlock. This is safe wrt the domain_map.
587 * -nodes cannot be added now as the
588 * query_join_handlers knows to respond with OK_NO_MAP
589 * -we catch the right network errors if a node is
590 * removed from the map while we're sending him the
592 spin_unlock(&dlm->spinlock);
596 status = dlm_send_one_domain_exit(dlm, node);
598 status != -ENOPROTOOPT &&
599 status != -ENOTCONN) {
600 mlog(ML_NOTICE, "Error %d sending domain exit message "
601 "to node %d\n", status, node);
603 /* Not sure what to do here but lets sleep for
604 * a bit in case this was a transient
606 msleep(DLM_DOMAIN_BACKOFF_MS);
610 spin_lock(&dlm->spinlock);
611 /* If we're not clearing the node bit then we intend
612 * to loop back around to try again. */
614 clear_bit(node, dlm->domain_map);
616 spin_unlock(&dlm->spinlock);
619 int dlm_joined(struct dlm_ctxt *dlm)
623 spin_lock(&dlm_domain_lock);
625 if (dlm->dlm_state == DLM_CTXT_JOINED)
628 spin_unlock(&dlm_domain_lock);
633 int dlm_shutting_down(struct dlm_ctxt *dlm)
637 spin_lock(&dlm_domain_lock);
639 if (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN)
642 spin_unlock(&dlm_domain_lock);
647 void dlm_unregister_domain(struct dlm_ctxt *dlm)
650 struct dlm_lock_resource *res;
652 spin_lock(&dlm_domain_lock);
653 BUG_ON(dlm->dlm_state != DLM_CTXT_JOINED);
654 BUG_ON(!dlm->num_joins);
657 if (!dlm->num_joins) {
658 /* We mark it "in shutdown" now so new register
659 * requests wait until we've completely left the
660 * domain. Don't use DLM_CTXT_LEAVING yet as we still
661 * want new domain joins to communicate with us at
662 * least until we've completed migration of our
664 dlm->dlm_state = DLM_CTXT_IN_SHUTDOWN;
667 spin_unlock(&dlm_domain_lock);
670 mlog(0, "shutting down domain %s\n", dlm->name);
672 /* We changed dlm state, notify the thread */
673 dlm_kick_thread(dlm, NULL);
675 while (dlm_migrate_all_locks(dlm)) {
676 /* Give dlm_thread time to purge the lockres' */
678 mlog(0, "%s: more migration to do\n", dlm->name);
681 /* This list should be empty. If not, print remaining lockres */
682 if (!list_empty(&dlm->tracking_list)) {
683 mlog(ML_ERROR, "Following lockres' are still on the "
685 list_for_each_entry(res, &dlm->tracking_list, tracking)
686 dlm_print_one_lock_resource(res);
689 dlm_mark_domain_leaving(dlm);
690 dlm_leave_domain(dlm);
691 dlm_complete_dlm_shutdown(dlm);
695 EXPORT_SYMBOL_GPL(dlm_unregister_domain);
697 static int dlm_query_join_proto_check(char *proto_type, int node,
698 struct dlm_protocol_version *ours,
699 struct dlm_protocol_version *request)
702 struct dlm_protocol_version proto = *request;
704 if (!dlm_protocol_compare(ours, &proto)) {
706 "node %u wanted to join with %s locking protocol "
707 "%u.%u, we respond with %u.%u\n",
711 proto.pv_major, proto.pv_minor);
712 request->pv_minor = proto.pv_minor;
716 "Node %u wanted to join with %s locking "
717 "protocol %u.%u, but we have %u.%u, disallowing\n",
730 * struct dlm_query_join_packet is made up of four one-byte fields. They
731 * are effectively in big-endian order already. However, little-endian
732 * machines swap them before putting the packet on the wire (because
733 * query_join's response is a status, and that status is treated as a u32
734 * on the wire). Thus, a big-endian and little-endian machines will treat
735 * this structure differently.
737 * The solution is to have little-endian machines swap the structure when
738 * converting from the structure to the u32 representation. This will
739 * result in the structure having the correct format on the wire no matter
740 * the host endian format.
742 static void dlm_query_join_packet_to_wire(struct dlm_query_join_packet *packet,
745 union dlm_query_join_response response;
747 response.packet = *packet;
748 *wire = cpu_to_be32(response.intval);
751 static void dlm_query_join_wire_to_packet(u32 wire,
752 struct dlm_query_join_packet *packet)
754 union dlm_query_join_response response;
756 response.intval = cpu_to_be32(wire);
757 *packet = response.packet;
760 static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
763 struct dlm_query_join_request *query;
764 struct dlm_query_join_packet packet = {
765 .code = JOIN_DISALLOW,
767 struct dlm_ctxt *dlm = NULL;
771 query = (struct dlm_query_join_request *) msg->buf;
773 mlog(0, "node %u wants to join domain %s\n", query->node_idx,
777 * If heartbeat doesn't consider the node live, tell it
778 * to back off and try again. This gives heartbeat a chance
781 if (!o2hb_check_node_heartbeating(query->node_idx)) {
782 mlog(0, "node %u is not in our live map yet\n",
785 packet.code = JOIN_DISALLOW;
789 packet.code = JOIN_OK_NO_MAP;
791 spin_lock(&dlm_domain_lock);
792 dlm = __dlm_lookup_domain_full(query->domain, query->name_len);
797 * There is a small window where the joining node may not see the
798 * node(s) that just left but still part of the cluster. DISALLOW
799 * join request if joining node has different node map.
802 while (nodenum < O2NM_MAX_NODES) {
803 if (test_bit(nodenum, dlm->domain_map)) {
804 if (!byte_test_bit(nodenum, query->node_map)) {
805 mlog(0, "disallow join as node %u does not "
806 "have node %u in its nodemap\n",
807 query->node_idx, nodenum);
808 packet.code = JOIN_DISALLOW;
815 /* Once the dlm ctxt is marked as leaving then we don't want
816 * to be put in someone's domain map.
817 * Also, explicitly disallow joining at certain troublesome
818 * times (ie. during recovery). */
819 if (dlm && dlm->dlm_state != DLM_CTXT_LEAVING) {
820 int bit = query->node_idx;
821 spin_lock(&dlm->spinlock);
823 if (dlm->dlm_state == DLM_CTXT_NEW &&
824 dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN) {
825 /*If this is a brand new context and we
826 * haven't started our join process yet, then
827 * the other node won the race. */
828 packet.code = JOIN_OK_NO_MAP;
829 } else if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
830 /* Disallow parallel joins. */
831 packet.code = JOIN_DISALLOW;
832 } else if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
833 mlog(0, "node %u trying to join, but recovery "
834 "is ongoing.\n", bit);
835 packet.code = JOIN_DISALLOW;
836 } else if (test_bit(bit, dlm->recovery_map)) {
837 mlog(0, "node %u trying to join, but it "
838 "still needs recovery.\n", bit);
839 packet.code = JOIN_DISALLOW;
840 } else if (test_bit(bit, dlm->domain_map)) {
841 mlog(0, "node %u trying to join, but it "
842 "is still in the domain! needs recovery?\n",
844 packet.code = JOIN_DISALLOW;
846 /* Alright we're fully a part of this domain
847 * so we keep some state as to who's joining
848 * and indicate to him that needs to be fixed
851 /* Make sure we speak compatible locking protocols. */
852 if (dlm_query_join_proto_check("DLM", bit,
853 &dlm->dlm_locking_proto,
854 &query->dlm_proto)) {
855 packet.code = JOIN_PROTOCOL_MISMATCH;
856 } else if (dlm_query_join_proto_check("fs", bit,
857 &dlm->fs_locking_proto,
859 packet.code = JOIN_PROTOCOL_MISMATCH;
861 packet.dlm_minor = query->dlm_proto.pv_minor;
862 packet.fs_minor = query->fs_proto.pv_minor;
863 packet.code = JOIN_OK;
864 __dlm_set_joining_node(dlm, query->node_idx);
868 spin_unlock(&dlm->spinlock);
871 spin_unlock(&dlm_domain_lock);
874 mlog(0, "We respond with %u\n", packet.code);
876 dlm_query_join_packet_to_wire(&packet, &response);
880 static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
883 struct dlm_assert_joined *assert;
884 struct dlm_ctxt *dlm = NULL;
886 assert = (struct dlm_assert_joined *) msg->buf;
888 mlog(0, "node %u asserts join on domain %s\n", assert->node_idx,
891 spin_lock(&dlm_domain_lock);
892 dlm = __dlm_lookup_domain_full(assert->domain, assert->name_len);
893 /* XXX should we consider no dlm ctxt an error? */
895 spin_lock(&dlm->spinlock);
897 /* Alright, this node has officially joined our
898 * domain. Set him in the map and clean up our
899 * leftover join state. */
900 BUG_ON(dlm->joining_node != assert->node_idx);
901 set_bit(assert->node_idx, dlm->domain_map);
902 __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
904 printk(KERN_INFO "ocfs2_dlm: Node %u joins domain %s\n",
905 assert->node_idx, dlm->name);
906 __dlm_print_nodes(dlm);
908 /* notify anything attached to the heartbeat events */
909 dlm_hb_event_notify_attached(dlm, assert->node_idx, 1);
911 spin_unlock(&dlm->spinlock);
913 spin_unlock(&dlm_domain_lock);
918 static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
921 struct dlm_cancel_join *cancel;
922 struct dlm_ctxt *dlm = NULL;
924 cancel = (struct dlm_cancel_join *) msg->buf;
926 mlog(0, "node %u cancels join on domain %s\n", cancel->node_idx,
929 spin_lock(&dlm_domain_lock);
930 dlm = __dlm_lookup_domain_full(cancel->domain, cancel->name_len);
933 spin_lock(&dlm->spinlock);
935 /* Yikes, this guy wants to cancel his join. No
936 * problem, we simply cleanup our join state. */
937 BUG_ON(dlm->joining_node != cancel->node_idx);
938 __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
940 spin_unlock(&dlm->spinlock);
942 spin_unlock(&dlm_domain_lock);
947 static int dlm_send_one_join_cancel(struct dlm_ctxt *dlm,
951 struct dlm_cancel_join cancel_msg;
953 memset(&cancel_msg, 0, sizeof(cancel_msg));
954 cancel_msg.node_idx = dlm->node_num;
955 cancel_msg.name_len = strlen(dlm->name);
956 memcpy(cancel_msg.domain, dlm->name, cancel_msg.name_len);
958 status = o2net_send_message(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
959 &cancel_msg, sizeof(cancel_msg), node,
970 /* map_size should be in bytes. */
971 static int dlm_send_join_cancels(struct dlm_ctxt *dlm,
972 unsigned long *node_map,
973 unsigned int map_size)
978 if (map_size != (BITS_TO_LONGS(O2NM_MAX_NODES) *
979 sizeof(unsigned long))) {
981 "map_size %u != BITS_TO_LONGS(O2NM_MAX_NODES) %u\n",
982 map_size, (unsigned)BITS_TO_LONGS(O2NM_MAX_NODES));
988 while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
989 node + 1)) < O2NM_MAX_NODES) {
990 if (node == dlm->node_num)
993 tmpstat = dlm_send_one_join_cancel(dlm, node);
995 mlog(ML_ERROR, "Error return %d cancelling join on "
996 "node %d\n", tmpstat, node);
1007 static int dlm_request_join(struct dlm_ctxt *dlm,
1009 enum dlm_query_join_response_code *response)
1012 struct dlm_query_join_request join_msg;
1013 struct dlm_query_join_packet packet;
1016 mlog(0, "querying node %d\n", node);
1018 memset(&join_msg, 0, sizeof(join_msg));
1019 join_msg.node_idx = dlm->node_num;
1020 join_msg.name_len = strlen(dlm->name);
1021 memcpy(join_msg.domain, dlm->name, join_msg.name_len);
1022 join_msg.dlm_proto = dlm->dlm_locking_proto;
1023 join_msg.fs_proto = dlm->fs_locking_proto;
1025 /* copy live node map to join message */
1026 byte_copymap(join_msg.node_map, dlm->live_nodes_map, O2NM_MAX_NODES);
1028 status = o2net_send_message(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, &join_msg,
1029 sizeof(join_msg), node,
1031 if (status < 0 && status != -ENOPROTOOPT) {
1035 dlm_query_join_wire_to_packet(join_resp, &packet);
1037 /* -ENOPROTOOPT from the net code means the other side isn't
1038 listening for our message type -- that's fine, it means
1039 his dlm isn't up, so we can consider him a 'yes' but not
1040 joined into the domain. */
1041 if (status == -ENOPROTOOPT) {
1043 *response = JOIN_OK_NO_MAP;
1044 } else if (packet.code == JOIN_DISALLOW ||
1045 packet.code == JOIN_OK_NO_MAP) {
1046 *response = packet.code;
1047 } else if (packet.code == JOIN_PROTOCOL_MISMATCH) {
1049 "This node requested DLM locking protocol %u.%u and "
1050 "filesystem locking protocol %u.%u. At least one of "
1051 "the protocol versions on node %d is not compatible, "
1053 dlm->dlm_locking_proto.pv_major,
1054 dlm->dlm_locking_proto.pv_minor,
1055 dlm->fs_locking_proto.pv_major,
1056 dlm->fs_locking_proto.pv_minor,
1059 *response = packet.code;
1060 } else if (packet.code == JOIN_OK) {
1061 *response = packet.code;
1062 /* Use the same locking protocol as the remote node */
1063 dlm->dlm_locking_proto.pv_minor = packet.dlm_minor;
1064 dlm->fs_locking_proto.pv_minor = packet.fs_minor;
1066 "Node %d responds JOIN_OK with DLM locking protocol "
1067 "%u.%u and fs locking protocol %u.%u\n",
1069 dlm->dlm_locking_proto.pv_major,
1070 dlm->dlm_locking_proto.pv_minor,
1071 dlm->fs_locking_proto.pv_major,
1072 dlm->fs_locking_proto.pv_minor);
1075 mlog(ML_ERROR, "invalid response %d from node %u\n",
1079 mlog(0, "status %d, node %d response is %d\n", status, node,
1086 static int dlm_send_one_join_assert(struct dlm_ctxt *dlm,
1090 struct dlm_assert_joined assert_msg;
1092 mlog(0, "Sending join assert to node %u\n", node);
1094 memset(&assert_msg, 0, sizeof(assert_msg));
1095 assert_msg.node_idx = dlm->node_num;
1096 assert_msg.name_len = strlen(dlm->name);
1097 memcpy(assert_msg.domain, dlm->name, assert_msg.name_len);
1099 status = o2net_send_message(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1100 &assert_msg, sizeof(assert_msg), node,
1108 static void dlm_send_join_asserts(struct dlm_ctxt *dlm,
1109 unsigned long *node_map)
1111 int status, node, live;
1115 while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
1116 node + 1)) < O2NM_MAX_NODES) {
1117 if (node == dlm->node_num)
1121 /* It is very important that this message be
1122 * received so we spin until either the node
1123 * has died or it gets the message. */
1124 status = dlm_send_one_join_assert(dlm, node);
1126 spin_lock(&dlm->spinlock);
1127 live = test_bit(node, dlm->live_nodes_map);
1128 spin_unlock(&dlm->spinlock);
1131 mlog(ML_ERROR, "Error return %d asserting "
1132 "join on node %d\n", status, node);
1134 /* give us some time between errors... */
1136 msleep(DLM_DOMAIN_BACKOFF_MS);
1138 } while (status && live);
1142 struct domain_join_ctxt {
1143 unsigned long live_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1144 unsigned long yes_resp_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1147 static int dlm_should_restart_join(struct dlm_ctxt *dlm,
1148 struct domain_join_ctxt *ctxt,
1149 enum dlm_query_join_response_code response)
1153 if (response == JOIN_DISALLOW) {
1154 mlog(0, "Latest response of disallow -- should restart\n");
1158 spin_lock(&dlm->spinlock);
1159 /* For now, we restart the process if the node maps have
1161 ret = memcmp(ctxt->live_map, dlm->live_nodes_map,
1162 sizeof(dlm->live_nodes_map));
1163 spin_unlock(&dlm->spinlock);
1166 mlog(0, "Node maps changed -- should restart\n");
1171 static int dlm_try_to_join_domain(struct dlm_ctxt *dlm)
1173 int status = 0, tmpstat, node;
1174 struct domain_join_ctxt *ctxt;
1175 enum dlm_query_join_response_code response = JOIN_DISALLOW;
1177 mlog_entry("%p", dlm);
1179 ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
1186 /* group sem locking should work for us here -- we're already
1187 * registered for heartbeat events so filling this should be
1188 * atomic wrt getting those handlers called. */
1189 o2hb_fill_node_map(dlm->live_nodes_map, sizeof(dlm->live_nodes_map));
1191 spin_lock(&dlm->spinlock);
1192 memcpy(ctxt->live_map, dlm->live_nodes_map, sizeof(ctxt->live_map));
1194 __dlm_set_joining_node(dlm, dlm->node_num);
1196 spin_unlock(&dlm->spinlock);
1199 while ((node = find_next_bit(ctxt->live_map, O2NM_MAX_NODES,
1200 node + 1)) < O2NM_MAX_NODES) {
1201 if (node == dlm->node_num)
1204 status = dlm_request_join(dlm, node, &response);
1210 /* Ok, either we got a response or the node doesn't have a
1212 if (response == JOIN_OK)
1213 set_bit(node, ctxt->yes_resp_map);
1215 if (dlm_should_restart_join(dlm, ctxt, response)) {
1221 mlog(0, "Yay, done querying nodes!\n");
1223 /* Yay, everyone agree's we can join the domain. My domain is
1224 * comprised of all nodes who were put in the
1225 * yes_resp_map. Copy that into our domain map and send a join
1226 * assert message to clean up everyone elses state. */
1227 spin_lock(&dlm->spinlock);
1228 memcpy(dlm->domain_map, ctxt->yes_resp_map,
1229 sizeof(ctxt->yes_resp_map));
1230 set_bit(dlm->node_num, dlm->domain_map);
1231 spin_unlock(&dlm->spinlock);
1233 dlm_send_join_asserts(dlm, ctxt->yes_resp_map);
1235 /* Joined state *must* be set before the joining node
1236 * information, otherwise the query_join handler may read no
1237 * current joiner but a state of NEW and tell joining nodes
1238 * we're not in the domain. */
1239 spin_lock(&dlm_domain_lock);
1240 dlm->dlm_state = DLM_CTXT_JOINED;
1242 spin_unlock(&dlm_domain_lock);
1245 spin_lock(&dlm->spinlock);
1246 __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
1248 __dlm_print_nodes(dlm);
1249 spin_unlock(&dlm->spinlock);
1252 /* Do we need to send a cancel message to any nodes? */
1254 tmpstat = dlm_send_join_cancels(dlm,
1256 sizeof(ctxt->yes_resp_map));
1258 mlog_errno(tmpstat);
1263 mlog(0, "returning %d\n", status);
1267 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm)
1269 o2hb_unregister_callback(NULL, &dlm->dlm_hb_up);
1270 o2hb_unregister_callback(NULL, &dlm->dlm_hb_down);
1271 o2net_unregister_handler_list(&dlm->dlm_domain_handlers);
1274 static int dlm_register_domain_handlers(struct dlm_ctxt *dlm)
1278 mlog(0, "registering handlers.\n");
1280 o2hb_setup_callback(&dlm->dlm_hb_down, O2HB_NODE_DOWN_CB,
1281 dlm_hb_node_down_cb, dlm, DLM_HB_NODE_DOWN_PRI);
1282 status = o2hb_register_callback(NULL, &dlm->dlm_hb_down);
1286 o2hb_setup_callback(&dlm->dlm_hb_up, O2HB_NODE_UP_CB,
1287 dlm_hb_node_up_cb, dlm, DLM_HB_NODE_UP_PRI);
1288 status = o2hb_register_callback(NULL, &dlm->dlm_hb_up);
1292 status = o2net_register_handler(DLM_MASTER_REQUEST_MSG, dlm->key,
1293 sizeof(struct dlm_master_request),
1294 dlm_master_request_handler,
1295 dlm, NULL, &dlm->dlm_domain_handlers);
1299 status = o2net_register_handler(DLM_ASSERT_MASTER_MSG, dlm->key,
1300 sizeof(struct dlm_assert_master),
1301 dlm_assert_master_handler,
1302 dlm, dlm_assert_master_post_handler,
1303 &dlm->dlm_domain_handlers);
1307 status = o2net_register_handler(DLM_CREATE_LOCK_MSG, dlm->key,
1308 sizeof(struct dlm_create_lock),
1309 dlm_create_lock_handler,
1310 dlm, NULL, &dlm->dlm_domain_handlers);
1314 status = o2net_register_handler(DLM_CONVERT_LOCK_MSG, dlm->key,
1315 DLM_CONVERT_LOCK_MAX_LEN,
1316 dlm_convert_lock_handler,
1317 dlm, NULL, &dlm->dlm_domain_handlers);
1321 status = o2net_register_handler(DLM_UNLOCK_LOCK_MSG, dlm->key,
1322 DLM_UNLOCK_LOCK_MAX_LEN,
1323 dlm_unlock_lock_handler,
1324 dlm, NULL, &dlm->dlm_domain_handlers);
1328 status = o2net_register_handler(DLM_PROXY_AST_MSG, dlm->key,
1329 DLM_PROXY_AST_MAX_LEN,
1330 dlm_proxy_ast_handler,
1331 dlm, NULL, &dlm->dlm_domain_handlers);
1335 status = o2net_register_handler(DLM_EXIT_DOMAIN_MSG, dlm->key,
1336 sizeof(struct dlm_exit_domain),
1337 dlm_exit_domain_handler,
1338 dlm, NULL, &dlm->dlm_domain_handlers);
1342 status = o2net_register_handler(DLM_DEREF_LOCKRES_MSG, dlm->key,
1343 sizeof(struct dlm_deref_lockres),
1344 dlm_deref_lockres_handler,
1345 dlm, NULL, &dlm->dlm_domain_handlers);
1349 status = o2net_register_handler(DLM_MIGRATE_REQUEST_MSG, dlm->key,
1350 sizeof(struct dlm_migrate_request),
1351 dlm_migrate_request_handler,
1352 dlm, NULL, &dlm->dlm_domain_handlers);
1356 status = o2net_register_handler(DLM_MIG_LOCKRES_MSG, dlm->key,
1357 DLM_MIG_LOCKRES_MAX_LEN,
1358 dlm_mig_lockres_handler,
1359 dlm, NULL, &dlm->dlm_domain_handlers);
1363 status = o2net_register_handler(DLM_MASTER_REQUERY_MSG, dlm->key,
1364 sizeof(struct dlm_master_requery),
1365 dlm_master_requery_handler,
1366 dlm, NULL, &dlm->dlm_domain_handlers);
1370 status = o2net_register_handler(DLM_LOCK_REQUEST_MSG, dlm->key,
1371 sizeof(struct dlm_lock_request),
1372 dlm_request_all_locks_handler,
1373 dlm, NULL, &dlm->dlm_domain_handlers);
1377 status = o2net_register_handler(DLM_RECO_DATA_DONE_MSG, dlm->key,
1378 sizeof(struct dlm_reco_data_done),
1379 dlm_reco_data_done_handler,
1380 dlm, NULL, &dlm->dlm_domain_handlers);
1384 status = o2net_register_handler(DLM_BEGIN_RECO_MSG, dlm->key,
1385 sizeof(struct dlm_begin_reco),
1386 dlm_begin_reco_handler,
1387 dlm, NULL, &dlm->dlm_domain_handlers);
1391 status = o2net_register_handler(DLM_FINALIZE_RECO_MSG, dlm->key,
1392 sizeof(struct dlm_finalize_reco),
1393 dlm_finalize_reco_handler,
1394 dlm, NULL, &dlm->dlm_domain_handlers);
1400 dlm_unregister_domain_handlers(dlm);
1405 static int dlm_join_domain(struct dlm_ctxt *dlm)
1408 unsigned int backoff;
1409 unsigned int total_backoff = 0;
1413 mlog(0, "Join domain %s\n", dlm->name);
1415 status = dlm_register_domain_handlers(dlm);
1421 status = dlm_launch_thread(dlm);
1427 status = dlm_launch_recovery_thread(dlm);
1433 dlm->dlm_worker = create_singlethread_workqueue("dlm_wq");
1434 if (!dlm->dlm_worker) {
1441 status = dlm_try_to_join_domain(dlm);
1443 /* If we're racing another node to the join, then we
1444 * need to back off temporarily and let them
1446 #define DLM_JOIN_TIMEOUT_MSECS 90000
1447 if (status == -EAGAIN) {
1448 if (signal_pending(current)) {
1449 status = -ERESTARTSYS;
1454 msecs_to_jiffies(DLM_JOIN_TIMEOUT_MSECS)) {
1455 status = -ERESTARTSYS;
1456 mlog(ML_NOTICE, "Timed out joining dlm domain "
1457 "%s after %u msecs\n", dlm->name,
1458 jiffies_to_msecs(total_backoff));
1464 * <dale> No, after you!
1466 * <dale> But you first!
1469 backoff = (unsigned int)(jiffies & 0x3);
1470 backoff *= DLM_DOMAIN_BACKOFF_MS;
1471 total_backoff += backoff;
1472 mlog(0, "backoff %d\n", backoff);
1475 } while (status == -EAGAIN);
1484 wake_up(&dlm_domain_events);
1487 dlm_unregister_domain_handlers(dlm);
1488 dlm_complete_thread(dlm);
1489 dlm_complete_recovery_thread(dlm);
1490 dlm_destroy_dlm_worker(dlm);
1496 static struct dlm_ctxt *dlm_alloc_ctxt(const char *domain,
1501 struct dlm_ctxt *dlm = NULL;
1503 dlm = kzalloc(sizeof(*dlm), GFP_KERNEL);
1505 mlog_errno(-ENOMEM);
1509 dlm->name = kmalloc(strlen(domain) + 1, GFP_KERNEL);
1510 if (dlm->name == NULL) {
1511 mlog_errno(-ENOMEM);
1517 dlm->lockres_hash = (struct hlist_head **)dlm_alloc_pagevec(DLM_HASH_PAGES);
1518 if (!dlm->lockres_hash) {
1519 mlog_errno(-ENOMEM);
1526 for (i = 0; i < DLM_HASH_BUCKETS; i++)
1527 INIT_HLIST_HEAD(dlm_lockres_hash(dlm, i));
1529 strcpy(dlm->name, domain);
1531 dlm->node_num = o2nm_this_node();
1533 ret = dlm_create_debugfs_subroot(dlm);
1535 dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
1542 spin_lock_init(&dlm->spinlock);
1543 spin_lock_init(&dlm->master_lock);
1544 spin_lock_init(&dlm->ast_lock);
1545 INIT_LIST_HEAD(&dlm->list);
1546 INIT_LIST_HEAD(&dlm->dirty_list);
1547 INIT_LIST_HEAD(&dlm->reco.resources);
1548 INIT_LIST_HEAD(&dlm->reco.received);
1549 INIT_LIST_HEAD(&dlm->reco.node_data);
1550 INIT_LIST_HEAD(&dlm->purge_list);
1551 INIT_LIST_HEAD(&dlm->dlm_domain_handlers);
1552 INIT_LIST_HEAD(&dlm->tracking_list);
1553 dlm->reco.state = 0;
1555 INIT_LIST_HEAD(&dlm->pending_asts);
1556 INIT_LIST_HEAD(&dlm->pending_basts);
1558 mlog(0, "dlm->recovery_map=%p, &(dlm->recovery_map[0])=%p\n",
1559 dlm->recovery_map, &(dlm->recovery_map[0]));
1561 memset(dlm->recovery_map, 0, sizeof(dlm->recovery_map));
1562 memset(dlm->live_nodes_map, 0, sizeof(dlm->live_nodes_map));
1563 memset(dlm->domain_map, 0, sizeof(dlm->domain_map));
1565 dlm->dlm_thread_task = NULL;
1566 dlm->dlm_reco_thread_task = NULL;
1567 dlm->dlm_worker = NULL;
1568 init_waitqueue_head(&dlm->dlm_thread_wq);
1569 init_waitqueue_head(&dlm->dlm_reco_thread_wq);
1570 init_waitqueue_head(&dlm->reco.event);
1571 init_waitqueue_head(&dlm->ast_wq);
1572 init_waitqueue_head(&dlm->migration_wq);
1573 INIT_LIST_HEAD(&dlm->master_list);
1574 INIT_LIST_HEAD(&dlm->mle_hb_events);
1576 dlm->joining_node = DLM_LOCK_RES_OWNER_UNKNOWN;
1577 init_waitqueue_head(&dlm->dlm_join_events);
1579 dlm->reco.new_master = O2NM_INVALID_NODE_NUM;
1580 dlm->reco.dead_node = O2NM_INVALID_NODE_NUM;
1581 atomic_set(&dlm->local_resources, 0);
1582 atomic_set(&dlm->remote_resources, 0);
1583 atomic_set(&dlm->unknown_resources, 0);
1585 spin_lock_init(&dlm->work_lock);
1586 INIT_LIST_HEAD(&dlm->work_list);
1587 INIT_WORK(&dlm->dispatched_work, dlm_dispatch_work);
1589 kref_init(&dlm->dlm_refs);
1590 dlm->dlm_state = DLM_CTXT_NEW;
1592 INIT_LIST_HEAD(&dlm->dlm_eviction_callbacks);
1594 mlog(0, "context init: refcount %u\n",
1595 atomic_read(&dlm->dlm_refs.refcount));
1602 * Compare a requested locking protocol version against the current one.
1604 * If the major numbers are different, they are incompatible.
1605 * If the current minor is greater than the request, they are incompatible.
1606 * If the current minor is less than or equal to the request, they are
1607 * compatible, and the requester should run at the current minor version.
1609 static int dlm_protocol_compare(struct dlm_protocol_version *existing,
1610 struct dlm_protocol_version *request)
1612 if (existing->pv_major != request->pv_major)
1615 if (existing->pv_minor > request->pv_minor)
1618 if (existing->pv_minor < request->pv_minor)
1619 request->pv_minor = existing->pv_minor;
1625 * dlm_register_domain: one-time setup per "domain".
1627 * The filesystem passes in the requested locking version via proto.
1628 * If registration was successful, proto will contain the negotiated
1631 struct dlm_ctxt * dlm_register_domain(const char *domain,
1633 struct dlm_protocol_version *fs_proto)
1636 struct dlm_ctxt *dlm = NULL;
1637 struct dlm_ctxt *new_ctxt = NULL;
1639 if (strlen(domain) > O2NM_MAX_NAME_LEN) {
1640 ret = -ENAMETOOLONG;
1641 mlog(ML_ERROR, "domain name length too long\n");
1645 if (!o2hb_check_local_node_heartbeating()) {
1646 mlog(ML_ERROR, "the local node has not been configured, or is "
1647 "not heartbeating\n");
1652 mlog(0, "register called for domain \"%s\"\n", domain);
1656 if (signal_pending(current)) {
1662 spin_lock(&dlm_domain_lock);
1664 dlm = __dlm_lookup_domain(domain);
1666 if (dlm->dlm_state != DLM_CTXT_JOINED) {
1667 spin_unlock(&dlm_domain_lock);
1669 mlog(0, "This ctxt is not joined yet!\n");
1670 wait_event_interruptible(dlm_domain_events,
1671 dlm_wait_on_domain_helper(
1676 if (dlm_protocol_compare(&dlm->fs_locking_proto, fs_proto)) {
1678 "Requested locking protocol version is not "
1679 "compatible with already registered domain "
1680 "\"%s\"\n", domain);
1688 spin_unlock(&dlm_domain_lock);
1696 spin_unlock(&dlm_domain_lock);
1698 new_ctxt = dlm_alloc_ctxt(domain, key);
1707 /* a little variable switch-a-roo here... */
1711 /* add the new domain */
1712 list_add_tail(&dlm->list, &dlm_domains);
1713 spin_unlock(&dlm_domain_lock);
1716 * Pass the locking protocol version into the join. If the join
1717 * succeeds, it will have the negotiated protocol set.
1719 dlm->dlm_locking_proto = dlm_protocol;
1720 dlm->fs_locking_proto = *fs_proto;
1722 ret = dlm_join_domain(dlm);
1729 /* Tell the caller what locking protocol we negotiated */
1730 *fs_proto = dlm->fs_locking_proto;
1735 dlm_free_ctxt_mem(new_ctxt);
1742 EXPORT_SYMBOL_GPL(dlm_register_domain);
1744 static LIST_HEAD(dlm_join_handlers);
1746 static void dlm_unregister_net_handlers(void)
1748 o2net_unregister_handler_list(&dlm_join_handlers);
1751 static int dlm_register_net_handlers(void)
1755 status = o2net_register_handler(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
1756 sizeof(struct dlm_query_join_request),
1757 dlm_query_join_handler,
1758 NULL, NULL, &dlm_join_handlers);
1762 status = o2net_register_handler(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1763 sizeof(struct dlm_assert_joined),
1764 dlm_assert_joined_handler,
1765 NULL, NULL, &dlm_join_handlers);
1769 status = o2net_register_handler(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
1770 sizeof(struct dlm_cancel_join),
1771 dlm_cancel_join_handler,
1772 NULL, NULL, &dlm_join_handlers);
1776 dlm_unregister_net_handlers();
1781 /* Domain eviction callback handling.
1783 * The file system requires notification of node death *before* the
1784 * dlm completes it's recovery work, otherwise it may be able to
1785 * acquire locks on resources requiring recovery. Since the dlm can
1786 * evict a node from it's domain *before* heartbeat fires, a similar
1787 * mechanism is required. */
1789 /* Eviction is not expected to happen often, so a per-domain lock is
1790 * not necessary. Eviction callbacks are allowed to sleep for short
1791 * periods of time. */
1792 static DECLARE_RWSEM(dlm_callback_sem);
1794 void dlm_fire_domain_eviction_callbacks(struct dlm_ctxt *dlm,
1797 struct list_head *iter;
1798 struct dlm_eviction_cb *cb;
1800 down_read(&dlm_callback_sem);
1801 list_for_each(iter, &dlm->dlm_eviction_callbacks) {
1802 cb = list_entry(iter, struct dlm_eviction_cb, ec_item);
1804 cb->ec_func(node_num, cb->ec_data);
1806 up_read(&dlm_callback_sem);
1809 void dlm_setup_eviction_cb(struct dlm_eviction_cb *cb,
1810 dlm_eviction_func *f,
1813 INIT_LIST_HEAD(&cb->ec_item);
1817 EXPORT_SYMBOL_GPL(dlm_setup_eviction_cb);
1819 void dlm_register_eviction_cb(struct dlm_ctxt *dlm,
1820 struct dlm_eviction_cb *cb)
1822 down_write(&dlm_callback_sem);
1823 list_add_tail(&cb->ec_item, &dlm->dlm_eviction_callbacks);
1824 up_write(&dlm_callback_sem);
1826 EXPORT_SYMBOL_GPL(dlm_register_eviction_cb);
1828 void dlm_unregister_eviction_cb(struct dlm_eviction_cb *cb)
1830 down_write(&dlm_callback_sem);
1831 list_del_init(&cb->ec_item);
1832 up_write(&dlm_callback_sem);
1834 EXPORT_SYMBOL_GPL(dlm_unregister_eviction_cb);
1836 static int __init dlm_init(void)
1840 dlm_print_version();
1842 status = dlm_init_mle_cache();
1844 mlog(ML_ERROR, "Could not create o2dlm_mle slabcache\n");
1848 status = dlm_init_master_caches();
1850 mlog(ML_ERROR, "Could not create o2dlm_lockres and "
1851 "o2dlm_lockname slabcaches\n");
1855 status = dlm_init_lock_cache();
1857 mlog(ML_ERROR, "Count not create o2dlm_lock slabcache\n");
1861 status = dlm_register_net_handlers();
1863 mlog(ML_ERROR, "Unable to register network handlers\n");
1867 status = dlm_create_debugfs_root();
1873 dlm_unregister_net_handlers();
1874 dlm_destroy_lock_cache();
1875 dlm_destroy_master_caches();
1876 dlm_destroy_mle_cache();
1880 static void __exit dlm_exit (void)
1882 dlm_destroy_debugfs_root();
1883 dlm_unregister_net_handlers();
1884 dlm_destroy_lock_cache();
1885 dlm_destroy_master_caches();
1886 dlm_destroy_mle_cache();
1889 MODULE_AUTHOR("Oracle");
1890 MODULE_LICENSE("GPL");
1892 module_init(dlm_init);
1893 module_exit(dlm_exit);