]>
Commit | Line | Data |
---|---|---|
328970de | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
6714d8e8 KH |
2 | /* -*- mode: c; c-basic-offset: 8; -*- |
3 | * vim: noexpandtab sw=8 ts=8 sts=0: | |
4 | * | |
5 | * dlmdomain.c | |
6 | * | |
7 | * defines domain join / leave apis | |
8 | * | |
9 | * Copyright (C) 2004 Oracle. All rights reserved. | |
6714d8e8 KH |
10 | */ |
11 | ||
12 | #include <linux/module.h> | |
13 | #include <linux/types.h> | |
14 | #include <linux/slab.h> | |
15 | #include <linux/highmem.h> | |
6714d8e8 KH |
16 | #include <linux/init.h> |
17 | #include <linux/spinlock.h> | |
18 | #include <linux/delay.h> | |
19 | #include <linux/err.h> | |
6325b4a2 | 20 | #include <linux/debugfs.h> |
174cd4b1 | 21 | #include <linux/sched/signal.h> |
6714d8e8 KH |
22 | |
23 | #include "cluster/heartbeat.h" | |
24 | #include "cluster/nodemanager.h" | |
25 | #include "cluster/tcp.h" | |
26 | ||
27 | #include "dlmapi.h" | |
28 | #include "dlmcommon.h" | |
6714d8e8 | 29 | #include "dlmdomain.h" |
6325b4a2 | 30 | #include "dlmdebug.h" |
6714d8e8 | 31 | |
6714d8e8 KH |
32 | #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_DOMAIN) |
33 | #include "cluster/masklog.h" | |
34 | ||
1faf2894 SE |
35 | /* |
36 | * ocfs2 node maps are array of long int, which limits to send them freely | |
37 | * across the wire due to endianness issues. To workaround this, we convert | |
38 | * long ints to byte arrays. Following 3 routines are helper functions to | |
39 | * set/test/copy bits within those array of bytes | |
40 | */ | |
41 | static inline void byte_set_bit(u8 nr, u8 map[]) | |
42 | { | |
43 | map[nr >> 3] |= (1UL << (nr & 7)); | |
44 | } | |
45 | ||
46 | static inline int byte_test_bit(u8 nr, u8 map[]) | |
47 | { | |
48 | return ((1UL << (nr & 7)) & (map[nr >> 3])) != 0; | |
49 | } | |
50 | ||
51 | static inline void byte_copymap(u8 dmap[], unsigned long smap[], | |
52 | unsigned int sz) | |
53 | { | |
54 | unsigned int nn; | |
55 | ||
56 | if (!sz) | |
57 | return; | |
58 | ||
59 | memset(dmap, 0, ((sz + 7) >> 3)); | |
60 | for (nn = 0 ; nn < sz; nn++) | |
61 | if (test_bit(nn, smap)) | |
62 | byte_set_bit(nn, dmap); | |
63 | } | |
64 | ||
03d864c0 DP |
65 | static void dlm_free_pagevec(void **vec, int pages) |
66 | { | |
67 | while (pages--) | |
68 | free_page((unsigned long)vec[pages]); | |
69 | kfree(vec); | |
70 | } | |
71 | ||
72 | static void **dlm_alloc_pagevec(int pages) | |
73 | { | |
6da2ec56 | 74 | void **vec = kmalloc_array(pages, sizeof(void *), GFP_KERNEL); |
03d864c0 DP |
75 | int i; |
76 | ||
77 | if (!vec) | |
78 | return NULL; | |
79 | ||
80 | for (i = 0; i < pages; i++) | |
81 | if (!(vec[i] = (void *)__get_free_page(GFP_KERNEL))) | |
82 | goto out_free; | |
c8f33b6e | 83 | |
685f1adb | 84 | mlog(0, "Allocated DLM hash pagevec; %d pages (%lu expected), %lu buckets per page\n", |
f5a923d1 MF |
85 | pages, (unsigned long)DLM_HASH_PAGES, |
86 | (unsigned long)DLM_BUCKETS_PER_PAGE); | |
03d864c0 DP |
87 | return vec; |
88 | out_free: | |
89 | dlm_free_pagevec(vec, i); | |
90 | return NULL; | |
91 | } | |
92 | ||
6714d8e8 KH |
93 | /* |
94 | * | |
95 | * spinlock lock ordering: if multiple locks are needed, obey this ordering: | |
96 | * dlm_domain_lock | |
97 | * struct dlm_ctxt->spinlock | |
98 | * struct dlm_lock_resource->spinlock | |
99 | * struct dlm_ctxt->master_lock | |
100 | * struct dlm_ctxt->ast_lock | |
101 | * dlm_master_list_entry->spinlock | |
102 | * dlm_lock->spinlock | |
103 | * | |
104 | */ | |
105 | ||
34af946a | 106 | DEFINE_SPINLOCK(dlm_domain_lock); |
6714d8e8 KH |
107 | LIST_HEAD(dlm_domains); |
108 | static DECLARE_WAIT_QUEUE_HEAD(dlm_domain_events); | |
109 | ||
d24fbcda JB |
110 | /* |
111 | * The supported protocol version for DLM communication. Running domains | |
112 | * will have a negotiated version with the same major number and a minor | |
113 | * number equal or smaller. The dlm_ctxt->dlm_locking_proto field should | |
114 | * be used to determine what a running domain is actually using. | |
ea203441 SM |
115 | * |
116 | * New in version 1.1: | |
117 | * - Message DLM_QUERY_REGION added to support global heartbeat | |
18cfdf1b | 118 | * - Message DLM_QUERY_NODEINFO added to allow online node removes |
bddefdee SM |
119 | * New in version 1.2: |
120 | * - Message DLM_BEGIN_EXIT_DOMAIN_MSG added to mark start of exit domain | |
60d663cb | 121 | * New in version 1.3: |
122 | * - Message DLM_DEREF_LOCKRES_DONE added to inform non-master that the | |
123 | * refmap is cleared | |
d24fbcda JB |
124 | */ |
125 | static const struct dlm_protocol_version dlm_protocol = { | |
126 | .pv_major = 1, | |
60d663cb | 127 | .pv_minor = 3, |
d24fbcda JB |
128 | }; |
129 | ||
6714d8e8 KH |
130 | #define DLM_DOMAIN_BACKOFF_MS 200 |
131 | ||
d74c9803 KH |
132 | static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data, |
133 | void **ret_data); | |
134 | static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data, | |
135 | void **ret_data); | |
136 | static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data, | |
137 | void **ret_data); | |
ea203441 SM |
138 | static int dlm_query_region_handler(struct o2net_msg *msg, u32 len, |
139 | void *data, void **ret_data); | |
d74c9803 KH |
140 | static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data, |
141 | void **ret_data); | |
d24fbcda JB |
142 | static int dlm_protocol_compare(struct dlm_protocol_version *existing, |
143 | struct dlm_protocol_version *request); | |
6714d8e8 KH |
144 | |
145 | static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm); | |
146 | ||
e9f0b6a6 | 147 | void __dlm_unhash_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res) |
6714d8e8 | 148 | { |
e9f0b6a6 SM |
149 | if (hlist_unhashed(&res->hash_node)) |
150 | return; | |
151 | ||
152 | mlog(0, "%s: Unhash res %.*s\n", dlm->name, res->lockname.len, | |
153 | res->lockname.name); | |
154 | hlist_del_init(&res->hash_node); | |
155 | dlm_lockres_put(res); | |
6714d8e8 KH |
156 | } |
157 | ||
e9f0b6a6 | 158 | void __dlm_insert_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res) |
6714d8e8 | 159 | { |
81f2094a | 160 | struct hlist_head *bucket; |
6714d8e8 KH |
161 | |
162 | assert_spin_locked(&dlm->spinlock); | |
163 | ||
612645f7 | 164 | bucket = dlm_lockres_hash(dlm, res->lockname.hash); |
6714d8e8 KH |
165 | |
166 | /* get a reference for our hashtable */ | |
167 | dlm_lockres_get(res); | |
168 | ||
81f2094a | 169 | hlist_add_head(&res->hash_node, bucket); |
e9f0b6a6 SM |
170 | |
171 | mlog(0, "%s: Hash res %.*s\n", dlm->name, res->lockname.len, | |
172 | res->lockname.name); | |
6714d8e8 KH |
173 | } |
174 | ||
ba2bf218 KH |
175 | struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm, |
176 | const char *name, | |
177 | unsigned int len, | |
178 | unsigned int hash) | |
6714d8e8 | 179 | { |
81f2094a | 180 | struct hlist_head *bucket; |
df53cd3b | 181 | struct dlm_lock_resource *res; |
6714d8e8 | 182 | |
ef6b689b | 183 | mlog(0, "%.*s\n", len, name); |
6714d8e8 KH |
184 | |
185 | assert_spin_locked(&dlm->spinlock); | |
186 | ||
03d864c0 DP |
187 | bucket = dlm_lockres_hash(dlm, hash); |
188 | ||
df53cd3b | 189 | hlist_for_each_entry(res, bucket, hash_node) { |
4198985f DP |
190 | if (res->lockname.name[0] != name[0]) |
191 | continue; | |
192 | if (unlikely(res->lockname.len != len)) | |
193 | continue; | |
194 | if (memcmp(res->lockname.name + 1, name + 1, len - 1)) | |
195 | continue; | |
196 | dlm_lockres_get(res); | |
197 | return res; | |
6714d8e8 | 198 | } |
4198985f | 199 | return NULL; |
6714d8e8 KH |
200 | } |
201 | ||
ba2bf218 KH |
202 | /* intended to be called by functions which do not care about lock |
203 | * resources which are being purged (most net _handler functions). | |
204 | * this will return NULL for any lock resource which is found but | |
205 | * currently in the process of dropping its mastery reference. | |
206 | * use __dlm_lookup_lockres_full when you need the lock resource | |
207 | * regardless (e.g. dlm_get_lock_resource) */ | |
208 | struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm, | |
209 | const char *name, | |
210 | unsigned int len, | |
211 | unsigned int hash) | |
212 | { | |
213 | struct dlm_lock_resource *res = NULL; | |
214 | ||
ef6b689b | 215 | mlog(0, "%.*s\n", len, name); |
ba2bf218 KH |
216 | |
217 | assert_spin_locked(&dlm->spinlock); | |
218 | ||
219 | res = __dlm_lookup_lockres_full(dlm, name, len, hash); | |
220 | if (res) { | |
221 | spin_lock(&res->spinlock); | |
222 | if (res->state & DLM_LOCK_RES_DROPPING_REF) { | |
223 | spin_unlock(&res->spinlock); | |
224 | dlm_lockres_put(res); | |
225 | return NULL; | |
226 | } | |
227 | spin_unlock(&res->spinlock); | |
228 | } | |
229 | ||
230 | return res; | |
231 | } | |
232 | ||
6714d8e8 KH |
233 | struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm, |
234 | const char *name, | |
235 | unsigned int len) | |
236 | { | |
237 | struct dlm_lock_resource *res; | |
a3d33291 | 238 | unsigned int hash = dlm_lockid_hash(name, len); |
6714d8e8 KH |
239 | |
240 | spin_lock(&dlm->spinlock); | |
a3d33291 | 241 | res = __dlm_lookup_lockres(dlm, name, len, hash); |
6714d8e8 KH |
242 | spin_unlock(&dlm->spinlock); |
243 | return res; | |
244 | } | |
245 | ||
246 | static struct dlm_ctxt * __dlm_lookup_domain_full(const char *domain, int len) | |
247 | { | |
df53cd3b | 248 | struct dlm_ctxt *tmp; |
6714d8e8 KH |
249 | |
250 | assert_spin_locked(&dlm_domain_lock); | |
251 | ||
252 | /* tmp->name here is always NULL terminated, | |
253 | * but domain may not be! */ | |
df53cd3b | 254 | list_for_each_entry(tmp, &dlm_domains, list) { |
6714d8e8 KH |
255 | if (strlen(tmp->name) == len && |
256 | memcmp(tmp->name, domain, len)==0) | |
df53cd3b | 257 | return tmp; |
6714d8e8 KH |
258 | } |
259 | ||
df53cd3b | 260 | return NULL; |
6714d8e8 KH |
261 | } |
262 | ||
263 | /* For null terminated domain strings ONLY */ | |
264 | static struct dlm_ctxt * __dlm_lookup_domain(const char *domain) | |
265 | { | |
266 | assert_spin_locked(&dlm_domain_lock); | |
267 | ||
268 | return __dlm_lookup_domain_full(domain, strlen(domain)); | |
269 | } | |
270 | ||
271 | ||
272 | /* returns true on one of two conditions: | |
273 | * 1) the domain does not exist | |
274 | * 2) the domain exists and it's state is "joined" */ | |
275 | static int dlm_wait_on_domain_helper(const char *domain) | |
276 | { | |
277 | int ret = 0; | |
278 | struct dlm_ctxt *tmp = NULL; | |
279 | ||
280 | spin_lock(&dlm_domain_lock); | |
281 | ||
282 | tmp = __dlm_lookup_domain(domain); | |
283 | if (!tmp) | |
284 | ret = 1; | |
285 | else if (tmp->dlm_state == DLM_CTXT_JOINED) | |
286 | ret = 1; | |
287 | ||
288 | spin_unlock(&dlm_domain_lock); | |
289 | return ret; | |
290 | } | |
291 | ||
292 | static void dlm_free_ctxt_mem(struct dlm_ctxt *dlm) | |
293 | { | |
6325b4a2 SM |
294 | dlm_destroy_debugfs_subroot(dlm); |
295 | ||
81f2094a | 296 | if (dlm->lockres_hash) |
03d864c0 | 297 | dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES); |
6714d8e8 | 298 | |
e2b66ddc SM |
299 | if (dlm->master_hash) |
300 | dlm_free_pagevec((void **)dlm->master_hash, DLM_HASH_PAGES); | |
301 | ||
d787ab09 | 302 | kfree(dlm->name); |
6714d8e8 KH |
303 | kfree(dlm); |
304 | } | |
305 | ||
306 | /* A little strange - this function will be called while holding | |
307 | * dlm_domain_lock and is expected to be holding it on the way out. We | |
308 | * will however drop and reacquire it multiple times */ | |
309 | static void dlm_ctxt_release(struct kref *kref) | |
310 | { | |
311 | struct dlm_ctxt *dlm; | |
312 | ||
313 | dlm = container_of(kref, struct dlm_ctxt, dlm_refs); | |
314 | ||
315 | BUG_ON(dlm->num_joins); | |
316 | BUG_ON(dlm->dlm_state == DLM_CTXT_JOINED); | |
317 | ||
318 | /* we may still be in the list if we hit an error during join. */ | |
319 | list_del_init(&dlm->list); | |
320 | ||
321 | spin_unlock(&dlm_domain_lock); | |
322 | ||
323 | mlog(0, "freeing memory from domain %s\n", dlm->name); | |
324 | ||
325 | wake_up(&dlm_domain_events); | |
326 | ||
327 | dlm_free_ctxt_mem(dlm); | |
328 | ||
329 | spin_lock(&dlm_domain_lock); | |
330 | } | |
331 | ||
332 | void dlm_put(struct dlm_ctxt *dlm) | |
333 | { | |
334 | spin_lock(&dlm_domain_lock); | |
335 | kref_put(&dlm->dlm_refs, dlm_ctxt_release); | |
336 | spin_unlock(&dlm_domain_lock); | |
337 | } | |
338 | ||
339 | static void __dlm_get(struct dlm_ctxt *dlm) | |
340 | { | |
341 | kref_get(&dlm->dlm_refs); | |
342 | } | |
343 | ||
344 | /* given a questionable reference to a dlm object, gets a reference if | |
345 | * it can find it in the list, otherwise returns NULL in which case | |
346 | * you shouldn't trust your pointer. */ | |
347 | struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm) | |
348 | { | |
df53cd3b DF |
349 | struct dlm_ctxt *target; |
350 | struct dlm_ctxt *ret = NULL; | |
6714d8e8 KH |
351 | |
352 | spin_lock(&dlm_domain_lock); | |
353 | ||
df53cd3b | 354 | list_for_each_entry(target, &dlm_domains, list) { |
6714d8e8 KH |
355 | if (target == dlm) { |
356 | __dlm_get(target); | |
df53cd3b | 357 | ret = target; |
6714d8e8 KH |
358 | break; |
359 | } | |
6714d8e8 KH |
360 | } |
361 | ||
362 | spin_unlock(&dlm_domain_lock); | |
363 | ||
df53cd3b | 364 | return ret; |
6714d8e8 KH |
365 | } |
366 | ||
367 | int dlm_domain_fully_joined(struct dlm_ctxt *dlm) | |
368 | { | |
369 | int ret; | |
370 | ||
371 | spin_lock(&dlm_domain_lock); | |
372 | ret = (dlm->dlm_state == DLM_CTXT_JOINED) || | |
373 | (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN); | |
374 | spin_unlock(&dlm_domain_lock); | |
375 | ||
376 | return ret; | |
377 | } | |
378 | ||
3156d267 KH |
379 | static void dlm_destroy_dlm_worker(struct dlm_ctxt *dlm) |
380 | { | |
381 | if (dlm->dlm_worker) { | |
3156d267 KH |
382 | destroy_workqueue(dlm->dlm_worker); |
383 | dlm->dlm_worker = NULL; | |
384 | } | |
385 | } | |
386 | ||
6714d8e8 KH |
387 | static void dlm_complete_dlm_shutdown(struct dlm_ctxt *dlm) |
388 | { | |
389 | dlm_unregister_domain_handlers(dlm); | |
007dce53 | 390 | dlm_debug_shutdown(dlm); |
6714d8e8 KH |
391 | dlm_complete_thread(dlm); |
392 | dlm_complete_recovery_thread(dlm); | |
3156d267 | 393 | dlm_destroy_dlm_worker(dlm); |
6714d8e8 KH |
394 | |
395 | /* We've left the domain. Now we can take ourselves out of the | |
396 | * list and allow the kref stuff to help us free the | |
397 | * memory. */ | |
398 | spin_lock(&dlm_domain_lock); | |
399 | list_del_init(&dlm->list); | |
400 | spin_unlock(&dlm_domain_lock); | |
401 | ||
402 | /* Wake up anyone waiting for us to remove this domain */ | |
403 | wake_up(&dlm_domain_events); | |
404 | } | |
405 | ||
ba2bf218 | 406 | static int dlm_migrate_all_locks(struct dlm_ctxt *dlm) |
6714d8e8 | 407 | { |
ba2bf218 | 408 | int i, num, n, ret = 0; |
6714d8e8 | 409 | struct dlm_lock_resource *res; |
ba2bf218 KH |
410 | struct hlist_node *iter; |
411 | struct hlist_head *bucket; | |
412 | int dropped; | |
6714d8e8 KH |
413 | |
414 | mlog(0, "Migrating locks from domain %s\n", dlm->name); | |
ba2bf218 KH |
415 | |
416 | num = 0; | |
6714d8e8 | 417 | spin_lock(&dlm->spinlock); |
81f2094a | 418 | for (i = 0; i < DLM_HASH_BUCKETS; i++) { |
ba2bf218 KH |
419 | redo_bucket: |
420 | n = 0; | |
421 | bucket = dlm_lockres_hash(dlm, i); | |
422 | iter = bucket->first; | |
423 | while (iter) { | |
424 | n++; | |
425 | res = hlist_entry(iter, struct dlm_lock_resource, | |
426 | hash_node); | |
6714d8e8 | 427 | dlm_lockres_get(res); |
ba2bf218 KH |
428 | /* migrate, if necessary. this will drop the dlm |
429 | * spinlock and retake it if it does migration. */ | |
430 | dropped = dlm_empty_lockres(dlm, res); | |
431 | ||
432 | spin_lock(&res->spinlock); | |
66effd3c SM |
433 | if (dropped) |
434 | __dlm_lockres_calc_usage(dlm, res); | |
435 | else | |
436 | iter = res->hash_node.next; | |
ba2bf218 KH |
437 | spin_unlock(&res->spinlock); |
438 | ||
6714d8e8 | 439 | dlm_lockres_put(res); |
ba2bf218 | 440 | |
66effd3c SM |
441 | if (dropped) { |
442 | cond_resched_lock(&dlm->spinlock); | |
ba2bf218 | 443 | goto redo_bucket; |
66effd3c | 444 | } |
6714d8e8 | 445 | } |
0d01af6e | 446 | cond_resched_lock(&dlm->spinlock); |
ba2bf218 | 447 | num += n; |
6714d8e8 | 448 | } |
60c7ec9e | 449 | |
450 | if (!num) { | |
451 | if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) { | |
452 | mlog(0, "%s: perhaps there are more lock resources " | |
453 | "need to be migrated after dlm recovery\n", dlm->name); | |
454 | ret = -EAGAIN; | |
455 | } else { | |
456 | mlog(0, "%s: we won't do dlm recovery after migrating " | |
457 | "all lock resources\n", dlm->name); | |
458 | dlm->migrate_done = 1; | |
459 | } | |
460 | } | |
461 | ||
6714d8e8 | 462 | spin_unlock(&dlm->spinlock); |
ba2bf218 KH |
463 | wake_up(&dlm->dlm_thread_wq); |
464 | ||
465 | /* let the dlm thread take care of purging, keep scanning until | |
466 | * nothing remains in the hash */ | |
467 | if (num) { | |
468 | mlog(0, "%s: %d lock resources in hash last pass\n", | |
469 | dlm->name, num); | |
470 | ret = -EAGAIN; | |
471 | } | |
6714d8e8 | 472 | mlog(0, "DONE Migrating locks from domain %s\n", dlm->name); |
ba2bf218 | 473 | return ret; |
6714d8e8 KH |
474 | } |
475 | ||
476 | static int dlm_no_joining_node(struct dlm_ctxt *dlm) | |
477 | { | |
478 | int ret; | |
479 | ||
480 | spin_lock(&dlm->spinlock); | |
481 | ret = dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN; | |
482 | spin_unlock(&dlm->spinlock); | |
483 | ||
484 | return ret; | |
485 | } | |
486 | ||
bddefdee SM |
487 | static int dlm_begin_exit_domain_handler(struct o2net_msg *msg, u32 len, |
488 | void *data, void **ret_data) | |
489 | { | |
490 | struct dlm_ctxt *dlm = data; | |
491 | unsigned int node; | |
492 | struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf; | |
493 | ||
494 | if (!dlm_grab(dlm)) | |
495 | return 0; | |
496 | ||
497 | node = exit_msg->node_idx; | |
498 | mlog(0, "%s: Node %u sent a begin exit domain message\n", dlm->name, node); | |
499 | ||
500 | spin_lock(&dlm->spinlock); | |
501 | set_bit(node, dlm->exit_domain_map); | |
502 | spin_unlock(&dlm->spinlock); | |
503 | ||
504 | dlm_put(dlm); | |
505 | ||
506 | return 0; | |
507 | } | |
508 | ||
6714d8e8 KH |
509 | static void dlm_mark_domain_leaving(struct dlm_ctxt *dlm) |
510 | { | |
511 | /* Yikes, a double spinlock! I need domain_lock for the dlm | |
512 | * state and the dlm spinlock for join state... Sorry! */ | |
513 | again: | |
514 | spin_lock(&dlm_domain_lock); | |
515 | spin_lock(&dlm->spinlock); | |
516 | ||
517 | if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) { | |
518 | mlog(0, "Node %d is joining, we wait on it.\n", | |
519 | dlm->joining_node); | |
520 | spin_unlock(&dlm->spinlock); | |
521 | spin_unlock(&dlm_domain_lock); | |
522 | ||
523 | wait_event(dlm->dlm_join_events, dlm_no_joining_node(dlm)); | |
524 | goto again; | |
525 | } | |
526 | ||
527 | dlm->dlm_state = DLM_CTXT_LEAVING; | |
528 | spin_unlock(&dlm->spinlock); | |
529 | spin_unlock(&dlm_domain_lock); | |
530 | } | |
531 | ||
532 | static void __dlm_print_nodes(struct dlm_ctxt *dlm) | |
533 | { | |
8decab3c | 534 | int node = -1, num = 0; |
6714d8e8 KH |
535 | |
536 | assert_spin_locked(&dlm->spinlock); | |
537 | ||
8decab3c | 538 | printk("( "); |
6714d8e8 KH |
539 | while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES, |
540 | node + 1)) < O2NM_MAX_NODES) { | |
781ee3e2 | 541 | printk("%d ", node); |
8decab3c | 542 | ++num; |
6714d8e8 | 543 | } |
8decab3c | 544 | printk(") %u nodes\n", num); |
6714d8e8 KH |
545 | } |
546 | ||
d74c9803 KH |
547 | static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data, |
548 | void **ret_data) | |
6714d8e8 KH |
549 | { |
550 | struct dlm_ctxt *dlm = data; | |
551 | unsigned int node; | |
552 | struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf; | |
553 | ||
ef6b689b | 554 | mlog(0, "%p %u %p", msg, len, data); |
6714d8e8 KH |
555 | |
556 | if (!dlm_grab(dlm)) | |
557 | return 0; | |
558 | ||
559 | node = exit_msg->node_idx; | |
560 | ||
6714d8e8 KH |
561 | spin_lock(&dlm->spinlock); |
562 | clear_bit(node, dlm->domain_map); | |
bddefdee | 563 | clear_bit(node, dlm->exit_domain_map); |
8decab3c | 564 | printk(KERN_NOTICE "o2dlm: Node %u leaves domain %s ", node, dlm->name); |
6714d8e8 KH |
565 | __dlm_print_nodes(dlm); |
566 | ||
567 | /* notify anything attached to the heartbeat events */ | |
568 | dlm_hb_event_notify_attached(dlm, node, 0); | |
569 | ||
570 | spin_unlock(&dlm->spinlock); | |
571 | ||
572 | dlm_put(dlm); | |
573 | ||
574 | return 0; | |
575 | } | |
576 | ||
bddefdee | 577 | static int dlm_send_one_domain_exit(struct dlm_ctxt *dlm, u32 msg_type, |
6714d8e8 KH |
578 | unsigned int node) |
579 | { | |
580 | int status; | |
581 | struct dlm_exit_domain leave_msg; | |
582 | ||
bddefdee SM |
583 | mlog(0, "%s: Sending domain exit message %u to node %u\n", dlm->name, |
584 | msg_type, node); | |
6714d8e8 KH |
585 | |
586 | memset(&leave_msg, 0, sizeof(leave_msg)); | |
587 | leave_msg.node_idx = dlm->node_num; | |
588 | ||
bddefdee SM |
589 | status = o2net_send_message(msg_type, dlm->key, &leave_msg, |
590 | sizeof(leave_msg), node, NULL); | |
a5196ec5 | 591 | if (status < 0) |
bddefdee SM |
592 | mlog(ML_ERROR, "Error %d sending domain exit message %u " |
593 | "to node %u on domain %s\n", status, msg_type, node, | |
594 | dlm->name); | |
6714d8e8 KH |
595 | |
596 | return status; | |
597 | } | |
598 | ||
bddefdee SM |
599 | static void dlm_begin_exit_domain(struct dlm_ctxt *dlm) |
600 | { | |
601 | int node = -1; | |
602 | ||
603 | /* Support for begin exit domain was added in 1.2 */ | |
604 | if (dlm->dlm_locking_proto.pv_major == 1 && | |
605 | dlm->dlm_locking_proto.pv_minor < 2) | |
606 | return; | |
607 | ||
608 | /* | |
609 | * Unlike DLM_EXIT_DOMAIN_MSG, DLM_BEGIN_EXIT_DOMAIN_MSG is purely | |
610 | * informational. Meaning if a node does not receive the message, | |
611 | * so be it. | |
612 | */ | |
613 | spin_lock(&dlm->spinlock); | |
614 | while (1) { | |
615 | node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES, node + 1); | |
616 | if (node >= O2NM_MAX_NODES) | |
617 | break; | |
618 | if (node == dlm->node_num) | |
619 | continue; | |
620 | ||
621 | spin_unlock(&dlm->spinlock); | |
622 | dlm_send_one_domain_exit(dlm, DLM_BEGIN_EXIT_DOMAIN_MSG, node); | |
623 | spin_lock(&dlm->spinlock); | |
624 | } | |
625 | spin_unlock(&dlm->spinlock); | |
626 | } | |
6714d8e8 KH |
627 | |
628 | static void dlm_leave_domain(struct dlm_ctxt *dlm) | |
629 | { | |
630 | int node, clear_node, status; | |
631 | ||
632 | /* At this point we've migrated away all our locks and won't | |
633 | * accept mastership of new ones. The dlm is responsible for | |
634 | * almost nothing now. We make sure not to confuse any joining | |
635 | * nodes and then commence shutdown procedure. */ | |
636 | ||
637 | spin_lock(&dlm->spinlock); | |
638 | /* Clear ourselves from the domain map */ | |
639 | clear_bit(dlm->node_num, dlm->domain_map); | |
640 | while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES, | |
641 | 0)) < O2NM_MAX_NODES) { | |
642 | /* Drop the dlm spinlock. This is safe wrt the domain_map. | |
643 | * -nodes cannot be added now as the | |
644 | * query_join_handlers knows to respond with OK_NO_MAP | |
645 | * -we catch the right network errors if a node is | |
646 | * removed from the map while we're sending him the | |
647 | * exit message. */ | |
648 | spin_unlock(&dlm->spinlock); | |
649 | ||
650 | clear_node = 1; | |
651 | ||
bddefdee SM |
652 | status = dlm_send_one_domain_exit(dlm, DLM_EXIT_DOMAIN_MSG, |
653 | node); | |
6714d8e8 KH |
654 | if (status < 0 && |
655 | status != -ENOPROTOOPT && | |
656 | status != -ENOTCONN) { | |
657 | mlog(ML_NOTICE, "Error %d sending domain exit message " | |
658 | "to node %d\n", status, node); | |
659 | ||
660 | /* Not sure what to do here but lets sleep for | |
661 | * a bit in case this was a transient | |
662 | * error... */ | |
663 | msleep(DLM_DOMAIN_BACKOFF_MS); | |
664 | clear_node = 0; | |
665 | } | |
666 | ||
667 | spin_lock(&dlm->spinlock); | |
668 | /* If we're not clearing the node bit then we intend | |
669 | * to loop back around to try again. */ | |
670 | if (clear_node) | |
671 | clear_bit(node, dlm->domain_map); | |
672 | } | |
673 | spin_unlock(&dlm->spinlock); | |
674 | } | |
675 | ||
6714d8e8 KH |
676 | void dlm_unregister_domain(struct dlm_ctxt *dlm) |
677 | { | |
678 | int leave = 0; | |
29576f8b | 679 | struct dlm_lock_resource *res; |
6714d8e8 KH |
680 | |
681 | spin_lock(&dlm_domain_lock); | |
682 | BUG_ON(dlm->dlm_state != DLM_CTXT_JOINED); | |
683 | BUG_ON(!dlm->num_joins); | |
684 | ||
685 | dlm->num_joins--; | |
686 | if (!dlm->num_joins) { | |
687 | /* We mark it "in shutdown" now so new register | |
688 | * requests wait until we've completely left the | |
689 | * domain. Don't use DLM_CTXT_LEAVING yet as we still | |
690 | * want new domain joins to communicate with us at | |
691 | * least until we've completed migration of our | |
692 | * resources. */ | |
693 | dlm->dlm_state = DLM_CTXT_IN_SHUTDOWN; | |
694 | leave = 1; | |
695 | } | |
696 | spin_unlock(&dlm_domain_lock); | |
697 | ||
698 | if (leave) { | |
699 | mlog(0, "shutting down domain %s\n", dlm->name); | |
bddefdee | 700 | dlm_begin_exit_domain(dlm); |
6714d8e8 KH |
701 | |
702 | /* We changed dlm state, notify the thread */ | |
703 | dlm_kick_thread(dlm, NULL); | |
704 | ||
ba2bf218 | 705 | while (dlm_migrate_all_locks(dlm)) { |
2f5bf1f2 SM |
706 | /* Give dlm_thread time to purge the lockres' */ |
707 | msleep(500); | |
ba2bf218 KH |
708 | mlog(0, "%s: more migration to do\n", dlm->name); |
709 | } | |
29576f8b SM |
710 | |
711 | /* This list should be empty. If not, print remaining lockres */ | |
712 | if (!list_empty(&dlm->tracking_list)) { | |
713 | mlog(ML_ERROR, "Following lockres' are still on the " | |
714 | "tracking list:\n"); | |
715 | list_for_each_entry(res, &dlm->tracking_list, tracking) | |
716 | dlm_print_one_lock_resource(res); | |
717 | } | |
718 | ||
6714d8e8 KH |
719 | dlm_mark_domain_leaving(dlm); |
720 | dlm_leave_domain(dlm); | |
8decab3c | 721 | printk(KERN_NOTICE "o2dlm: Leaving domain %s\n", dlm->name); |
5dad6c39 | 722 | dlm_force_free_mles(dlm); |
6714d8e8 KH |
723 | dlm_complete_dlm_shutdown(dlm); |
724 | } | |
725 | dlm_put(dlm); | |
726 | } | |
727 | EXPORT_SYMBOL_GPL(dlm_unregister_domain); | |
728 | ||
d24fbcda JB |
729 | static int dlm_query_join_proto_check(char *proto_type, int node, |
730 | struct dlm_protocol_version *ours, | |
731 | struct dlm_protocol_version *request) | |
732 | { | |
733 | int rc; | |
734 | struct dlm_protocol_version proto = *request; | |
735 | ||
736 | if (!dlm_protocol_compare(ours, &proto)) { | |
737 | mlog(0, | |
738 | "node %u wanted to join with %s locking protocol " | |
739 | "%u.%u, we respond with %u.%u\n", | |
740 | node, proto_type, | |
741 | request->pv_major, | |
742 | request->pv_minor, | |
743 | proto.pv_major, proto.pv_minor); | |
744 | request->pv_minor = proto.pv_minor; | |
745 | rc = 0; | |
746 | } else { | |
747 | mlog(ML_NOTICE, | |
748 | "Node %u wanted to join with %s locking " | |
749 | "protocol %u.%u, but we have %u.%u, disallowing\n", | |
750 | node, proto_type, | |
751 | request->pv_major, | |
752 | request->pv_minor, | |
753 | ours->pv_major, | |
754 | ours->pv_minor); | |
755 | rc = 1; | |
756 | } | |
757 | ||
758 | return rc; | |
759 | } | |
760 | ||
0f71b7b4 JB |
761 | /* |
762 | * struct dlm_query_join_packet is made up of four one-byte fields. They | |
763 | * are effectively in big-endian order already. However, little-endian | |
764 | * machines swap them before putting the packet on the wire (because | |
765 | * query_join's response is a status, and that status is treated as a u32 | |
766 | * on the wire). Thus, a big-endian and little-endian machines will treat | |
767 | * this structure differently. | |
768 | * | |
769 | * The solution is to have little-endian machines swap the structure when | |
770 | * converting from the structure to the u32 representation. This will | |
771 | * result in the structure having the correct format on the wire no matter | |
772 | * the host endian format. | |
773 | */ | |
774 | static void dlm_query_join_packet_to_wire(struct dlm_query_join_packet *packet, | |
775 | u32 *wire) | |
776 | { | |
777 | union dlm_query_join_response response; | |
778 | ||
779 | response.packet = *packet; | |
85158410 | 780 | *wire = be32_to_cpu(response.intval); |
0f71b7b4 JB |
781 | } |
782 | ||
783 | static void dlm_query_join_wire_to_packet(u32 wire, | |
784 | struct dlm_query_join_packet *packet) | |
785 | { | |
786 | union dlm_query_join_response response; | |
787 | ||
788 | response.intval = cpu_to_be32(wire); | |
789 | *packet = response.packet; | |
790 | } | |
791 | ||
d74c9803 KH |
792 | static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data, |
793 | void **ret_data) | |
6714d8e8 KH |
794 | { |
795 | struct dlm_query_join_request *query; | |
0f71b7b4 JB |
796 | struct dlm_query_join_packet packet = { |
797 | .code = JOIN_DISALLOW, | |
d24fbcda | 798 | }; |
6714d8e8 | 799 | struct dlm_ctxt *dlm = NULL; |
0f71b7b4 | 800 | u32 response; |
1faf2894 | 801 | u8 nodenum; |
6714d8e8 KH |
802 | |
803 | query = (struct dlm_query_join_request *) msg->buf; | |
804 | ||
805 | mlog(0, "node %u wants to join domain %s\n", query->node_idx, | |
806 | query->domain); | |
807 | ||
808 | /* | |
809 | * If heartbeat doesn't consider the node live, tell it | |
810 | * to back off and try again. This gives heartbeat a chance | |
811 | * to catch up. | |
812 | */ | |
70e82a12 | 813 | if (!o2hb_check_node_heartbeating_no_sem(query->node_idx)) { |
6714d8e8 KH |
814 | mlog(0, "node %u is not in our live map yet\n", |
815 | query->node_idx); | |
816 | ||
0f71b7b4 | 817 | packet.code = JOIN_DISALLOW; |
6714d8e8 KH |
818 | goto respond; |
819 | } | |
820 | ||
0f71b7b4 | 821 | packet.code = JOIN_OK_NO_MAP; |
6714d8e8 KH |
822 | |
823 | spin_lock(&dlm_domain_lock); | |
824 | dlm = __dlm_lookup_domain_full(query->domain, query->name_len); | |
1faf2894 SE |
825 | if (!dlm) |
826 | goto unlock_respond; | |
827 | ||
828 | /* | |
829 | * There is a small window where the joining node may not see the | |
830 | * node(s) that just left but still part of the cluster. DISALLOW | |
831 | * join request if joining node has different node map. | |
832 | */ | |
833 | nodenum=0; | |
834 | while (nodenum < O2NM_MAX_NODES) { | |
835 | if (test_bit(nodenum, dlm->domain_map)) { | |
836 | if (!byte_test_bit(nodenum, query->node_map)) { | |
e4968476 SM |
837 | mlog(0, "disallow join as node %u does not " |
838 | "have node %u in its nodemap\n", | |
839 | query->node_idx, nodenum); | |
0f71b7b4 | 840 | packet.code = JOIN_DISALLOW; |
1faf2894 SE |
841 | goto unlock_respond; |
842 | } | |
843 | } | |
844 | nodenum++; | |
845 | } | |
846 | ||
6714d8e8 | 847 | /* Once the dlm ctxt is marked as leaving then we don't want |
2bd63216 | 848 | * to be put in someone's domain map. |
e2faea4c KH |
849 | * Also, explicitly disallow joining at certain troublesome |
850 | * times (ie. during recovery). */ | |
b3e3e5af | 851 | if (dlm->dlm_state != DLM_CTXT_LEAVING) { |
e2faea4c | 852 | int bit = query->node_idx; |
6714d8e8 KH |
853 | spin_lock(&dlm->spinlock); |
854 | ||
855 | if (dlm->dlm_state == DLM_CTXT_NEW && | |
856 | dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN) { | |
857 | /*If this is a brand new context and we | |
858 | * haven't started our join process yet, then | |
859 | * the other node won the race. */ | |
0f71b7b4 | 860 | packet.code = JOIN_OK_NO_MAP; |
6714d8e8 KH |
861 | } else if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) { |
862 | /* Disallow parallel joins. */ | |
0f71b7b4 | 863 | packet.code = JOIN_DISALLOW; |
e2faea4c | 864 | } else if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) { |
e4968476 | 865 | mlog(0, "node %u trying to join, but recovery " |
e2faea4c | 866 | "is ongoing.\n", bit); |
0f71b7b4 | 867 | packet.code = JOIN_DISALLOW; |
e2faea4c | 868 | } else if (test_bit(bit, dlm->recovery_map)) { |
e4968476 | 869 | mlog(0, "node %u trying to join, but it " |
e2faea4c | 870 | "still needs recovery.\n", bit); |
0f71b7b4 | 871 | packet.code = JOIN_DISALLOW; |
e2faea4c | 872 | } else if (test_bit(bit, dlm->domain_map)) { |
e4968476 | 873 | mlog(0, "node %u trying to join, but it " |
e2faea4c KH |
874 | "is still in the domain! needs recovery?\n", |
875 | bit); | |
0f71b7b4 | 876 | packet.code = JOIN_DISALLOW; |
6714d8e8 KH |
877 | } else { |
878 | /* Alright we're fully a part of this domain | |
879 | * so we keep some state as to who's joining | |
880 | * and indicate to him that needs to be fixed | |
881 | * up. */ | |
d24fbcda JB |
882 | |
883 | /* Make sure we speak compatible locking protocols. */ | |
884 | if (dlm_query_join_proto_check("DLM", bit, | |
885 | &dlm->dlm_locking_proto, | |
886 | &query->dlm_proto)) { | |
0f71b7b4 | 887 | packet.code = JOIN_PROTOCOL_MISMATCH; |
d24fbcda JB |
888 | } else if (dlm_query_join_proto_check("fs", bit, |
889 | &dlm->fs_locking_proto, | |
890 | &query->fs_proto)) { | |
0f71b7b4 | 891 | packet.code = JOIN_PROTOCOL_MISMATCH; |
d24fbcda | 892 | } else { |
0f71b7b4 JB |
893 | packet.dlm_minor = query->dlm_proto.pv_minor; |
894 | packet.fs_minor = query->fs_proto.pv_minor; | |
895 | packet.code = JOIN_OK; | |
d24fbcda JB |
896 | __dlm_set_joining_node(dlm, query->node_idx); |
897 | } | |
6714d8e8 KH |
898 | } |
899 | ||
900 | spin_unlock(&dlm->spinlock); | |
901 | } | |
1faf2894 | 902 | unlock_respond: |
6714d8e8 KH |
903 | spin_unlock(&dlm_domain_lock); |
904 | ||
905 | respond: | |
0f71b7b4 | 906 | mlog(0, "We respond with %u\n", packet.code); |
6714d8e8 | 907 | |
0f71b7b4 JB |
908 | dlm_query_join_packet_to_wire(&packet, &response); |
909 | return response; | |
6714d8e8 KH |
910 | } |
911 | ||
d74c9803 KH |
912 | static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data, |
913 | void **ret_data) | |
6714d8e8 KH |
914 | { |
915 | struct dlm_assert_joined *assert; | |
916 | struct dlm_ctxt *dlm = NULL; | |
917 | ||
918 | assert = (struct dlm_assert_joined *) msg->buf; | |
919 | ||
920 | mlog(0, "node %u asserts join on domain %s\n", assert->node_idx, | |
921 | assert->domain); | |
922 | ||
923 | spin_lock(&dlm_domain_lock); | |
924 | dlm = __dlm_lookup_domain_full(assert->domain, assert->name_len); | |
925 | /* XXX should we consider no dlm ctxt an error? */ | |
926 | if (dlm) { | |
927 | spin_lock(&dlm->spinlock); | |
928 | ||
929 | /* Alright, this node has officially joined our | |
930 | * domain. Set him in the map and clean up our | |
931 | * leftover join state. */ | |
932 | BUG_ON(dlm->joining_node != assert->node_idx); | |
01c6222f X |
933 | |
934 | if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) { | |
935 | mlog(0, "dlm recovery is ongoing, disallow join\n"); | |
936 | spin_unlock(&dlm->spinlock); | |
937 | spin_unlock(&dlm_domain_lock); | |
938 | return -EAGAIN; | |
939 | } | |
940 | ||
6714d8e8 | 941 | set_bit(assert->node_idx, dlm->domain_map); |
bddefdee | 942 | clear_bit(assert->node_idx, dlm->exit_domain_map); |
6714d8e8 KH |
943 | __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN); |
944 | ||
8decab3c | 945 | printk(KERN_NOTICE "o2dlm: Node %u joins domain %s ", |
781ee3e2 | 946 | assert->node_idx, dlm->name); |
6714d8e8 KH |
947 | __dlm_print_nodes(dlm); |
948 | ||
949 | /* notify anything attached to the heartbeat events */ | |
950 | dlm_hb_event_notify_attached(dlm, assert->node_idx, 1); | |
951 | ||
952 | spin_unlock(&dlm->spinlock); | |
953 | } | |
954 | spin_unlock(&dlm_domain_lock); | |
955 | ||
956 | return 0; | |
957 | } | |
958 | ||
ea203441 | 959 | static int dlm_match_regions(struct dlm_ctxt *dlm, |
770c4d81 SM |
960 | struct dlm_query_region *qr, |
961 | char *local, int locallen) | |
ea203441 | 962 | { |
770c4d81 | 963 | char *remote = qr->qr_regions; |
ea203441 SM |
964 | char *l, *r; |
965 | int localnr, i, j, foundit; | |
966 | int status = 0; | |
967 | ||
968 | if (!o2hb_global_heartbeat_active()) { | |
969 | if (qr->qr_numregions) { | |
970 | mlog(ML_ERROR, "Domain %s: Joining node %d has global " | |
971 | "heartbeat enabled but local node %d does not\n", | |
972 | qr->qr_domain, qr->qr_node, dlm->node_num); | |
973 | status = -EINVAL; | |
974 | } | |
975 | goto bail; | |
976 | } | |
977 | ||
978 | if (o2hb_global_heartbeat_active() && !qr->qr_numregions) { | |
979 | mlog(ML_ERROR, "Domain %s: Local node %d has global " | |
980 | "heartbeat enabled but joining node %d does not\n", | |
981 | qr->qr_domain, dlm->node_num, qr->qr_node); | |
982 | status = -EINVAL; | |
983 | goto bail; | |
984 | } | |
985 | ||
986 | r = remote; | |
987 | for (i = 0; i < qr->qr_numregions; ++i) { | |
988 | mlog(0, "Region %.*s\n", O2HB_MAX_REGION_NAME_LEN, r); | |
989 | r += O2HB_MAX_REGION_NAME_LEN; | |
990 | } | |
991 | ||
770c4d81 SM |
992 | localnr = min(O2NM_MAX_REGIONS, locallen/O2HB_MAX_REGION_NAME_LEN); |
993 | localnr = o2hb_get_all_regions(local, (u8)localnr); | |
ea203441 SM |
994 | |
995 | /* compare local regions with remote */ | |
996 | l = local; | |
997 | for (i = 0; i < localnr; ++i) { | |
998 | foundit = 0; | |
999 | r = remote; | |
1000 | for (j = 0; j <= qr->qr_numregions; ++j) { | |
1001 | if (!memcmp(l, r, O2HB_MAX_REGION_NAME_LEN)) { | |
1002 | foundit = 1; | |
1003 | break; | |
1004 | } | |
1005 | r += O2HB_MAX_REGION_NAME_LEN; | |
1006 | } | |
1007 | if (!foundit) { | |
1008 | status = -EINVAL; | |
1009 | mlog(ML_ERROR, "Domain %s: Region '%.*s' registered " | |
1010 | "in local node %d but not in joining node %d\n", | |
1011 | qr->qr_domain, O2HB_MAX_REGION_NAME_LEN, l, | |
1012 | dlm->node_num, qr->qr_node); | |
1013 | goto bail; | |
1014 | } | |
1015 | l += O2HB_MAX_REGION_NAME_LEN; | |
1016 | } | |
1017 | ||
1018 | /* compare remote with local regions */ | |
1019 | r = remote; | |
1020 | for (i = 0; i < qr->qr_numregions; ++i) { | |
1021 | foundit = 0; | |
1022 | l = local; | |
1023 | for (j = 0; j < localnr; ++j) { | |
1024 | if (!memcmp(r, l, O2HB_MAX_REGION_NAME_LEN)) { | |
1025 | foundit = 1; | |
1026 | break; | |
1027 | } | |
1028 | l += O2HB_MAX_REGION_NAME_LEN; | |
1029 | } | |
1030 | if (!foundit) { | |
1031 | status = -EINVAL; | |
1032 | mlog(ML_ERROR, "Domain %s: Region '%.*s' registered " | |
1033 | "in joining node %d but not in local node %d\n", | |
1034 | qr->qr_domain, O2HB_MAX_REGION_NAME_LEN, r, | |
1035 | qr->qr_node, dlm->node_num); | |
1036 | goto bail; | |
1037 | } | |
1038 | r += O2HB_MAX_REGION_NAME_LEN; | |
1039 | } | |
1040 | ||
1041 | bail: | |
ea203441 SM |
1042 | return status; |
1043 | } | |
1044 | ||
1045 | static int dlm_send_regions(struct dlm_ctxt *dlm, unsigned long *node_map) | |
1046 | { | |
1047 | struct dlm_query_region *qr = NULL; | |
1048 | int status, ret = 0, i; | |
1049 | char *p; | |
1050 | ||
1051 | if (find_next_bit(node_map, O2NM_MAX_NODES, 0) >= O2NM_MAX_NODES) | |
1052 | goto bail; | |
1053 | ||
1054 | qr = kzalloc(sizeof(struct dlm_query_region), GFP_KERNEL); | |
1055 | if (!qr) { | |
1056 | ret = -ENOMEM; | |
1057 | mlog_errno(ret); | |
1058 | goto bail; | |
1059 | } | |
1060 | ||
1061 | qr->qr_node = dlm->node_num; | |
1062 | qr->qr_namelen = strlen(dlm->name); | |
1063 | memcpy(qr->qr_domain, dlm->name, qr->qr_namelen); | |
1064 | /* if local hb, the numregions will be zero */ | |
1065 | if (o2hb_global_heartbeat_active()) | |
1066 | qr->qr_numregions = o2hb_get_all_regions(qr->qr_regions, | |
1067 | O2NM_MAX_REGIONS); | |
1068 | ||
1069 | p = qr->qr_regions; | |
1070 | for (i = 0; i < qr->qr_numregions; ++i, p += O2HB_MAX_REGION_NAME_LEN) | |
1071 | mlog(0, "Region %.*s\n", O2HB_MAX_REGION_NAME_LEN, p); | |
1072 | ||
1073 | i = -1; | |
1074 | while ((i = find_next_bit(node_map, O2NM_MAX_NODES, | |
1075 | i + 1)) < O2NM_MAX_NODES) { | |
1076 | if (i == dlm->node_num) | |
1077 | continue; | |
1078 | ||
1079 | mlog(0, "Sending regions to node %d\n", i); | |
1080 | ||
1081 | ret = o2net_send_message(DLM_QUERY_REGION, DLM_MOD_KEY, qr, | |
1082 | sizeof(struct dlm_query_region), | |
1083 | i, &status); | |
1084 | if (ret >= 0) | |
1085 | ret = status; | |
1086 | if (ret) { | |
1087 | mlog(ML_ERROR, "Region mismatch %d, node %d\n", | |
1088 | ret, i); | |
1089 | break; | |
1090 | } | |
1091 | } | |
1092 | ||
1093 | bail: | |
1094 | kfree(qr); | |
1095 | return ret; | |
1096 | } | |
1097 | ||
1098 | static int dlm_query_region_handler(struct o2net_msg *msg, u32 len, | |
1099 | void *data, void **ret_data) | |
1100 | { | |
1101 | struct dlm_query_region *qr; | |
1102 | struct dlm_ctxt *dlm = NULL; | |
770c4d81 | 1103 | char *local = NULL; |
ea203441 | 1104 | int status = 0; |
ea203441 SM |
1105 | |
1106 | qr = (struct dlm_query_region *) msg->buf; | |
1107 | ||
1108 | mlog(0, "Node %u queries hb regions on domain %s\n", qr->qr_node, | |
1109 | qr->qr_domain); | |
1110 | ||
770c4d81 SM |
1111 | /* buffer used in dlm_mast_regions() */ |
1112 | local = kmalloc(sizeof(qr->qr_regions), GFP_KERNEL); | |
a35ad97c ZG |
1113 | if (!local) |
1114 | return -ENOMEM; | |
770c4d81 | 1115 | |
ea203441 SM |
1116 | status = -EINVAL; |
1117 | ||
1118 | spin_lock(&dlm_domain_lock); | |
1119 | dlm = __dlm_lookup_domain_full(qr->qr_domain, qr->qr_namelen); | |
1120 | if (!dlm) { | |
1121 | mlog(ML_ERROR, "Node %d queried hb regions on domain %s " | |
1122 | "before join domain\n", qr->qr_node, qr->qr_domain); | |
a35ad97c | 1123 | goto out_domain_lock; |
ea203441 SM |
1124 | } |
1125 | ||
1126 | spin_lock(&dlm->spinlock); | |
ea203441 SM |
1127 | if (dlm->joining_node != qr->qr_node) { |
1128 | mlog(ML_ERROR, "Node %d queried hb regions on domain %s " | |
1129 | "but joining node is %d\n", qr->qr_node, qr->qr_domain, | |
1130 | dlm->joining_node); | |
a35ad97c | 1131 | goto out_dlm_lock; |
ea203441 SM |
1132 | } |
1133 | ||
1134 | /* Support for global heartbeat was added in 1.1 */ | |
1135 | if (dlm->dlm_locking_proto.pv_major == 1 && | |
1136 | dlm->dlm_locking_proto.pv_minor == 0) { | |
1137 | mlog(ML_ERROR, "Node %d queried hb regions on domain %s " | |
1138 | "but active dlm protocol is %d.%d\n", qr->qr_node, | |
1139 | qr->qr_domain, dlm->dlm_locking_proto.pv_major, | |
1140 | dlm->dlm_locking_proto.pv_minor); | |
a35ad97c | 1141 | goto out_dlm_lock; |
ea203441 SM |
1142 | } |
1143 | ||
770c4d81 | 1144 | status = dlm_match_regions(dlm, qr, local, sizeof(qr->qr_regions)); |
ea203441 | 1145 | |
a35ad97c ZG |
1146 | out_dlm_lock: |
1147 | spin_unlock(&dlm->spinlock); | |
1148 | ||
1149 | out_domain_lock: | |
ea203441 SM |
1150 | spin_unlock(&dlm_domain_lock); |
1151 | ||
770c4d81 SM |
1152 | kfree(local); |
1153 | ||
ea203441 SM |
1154 | return status; |
1155 | } | |
1156 | ||
18cfdf1b SM |
1157 | static int dlm_match_nodes(struct dlm_ctxt *dlm, struct dlm_query_nodeinfo *qn) |
1158 | { | |
1159 | struct o2nm_node *local; | |
1160 | struct dlm_node_info *remote; | |
1161 | int i, j; | |
1162 | int status = 0; | |
1163 | ||
1164 | for (j = 0; j < qn->qn_numnodes; ++j) | |
1165 | mlog(0, "Node %3d, %pI4:%u\n", qn->qn_nodes[j].ni_nodenum, | |
1166 | &(qn->qn_nodes[j].ni_ipv4_address), | |
1167 | ntohs(qn->qn_nodes[j].ni_ipv4_port)); | |
1168 | ||
1169 | for (i = 0; i < O2NM_MAX_NODES && !status; ++i) { | |
1170 | local = o2nm_get_node_by_num(i); | |
1171 | remote = NULL; | |
1172 | for (j = 0; j < qn->qn_numnodes; ++j) { | |
1173 | if (qn->qn_nodes[j].ni_nodenum == i) { | |
1174 | remote = &(qn->qn_nodes[j]); | |
1175 | break; | |
1176 | } | |
1177 | } | |
1178 | ||
1179 | if (!local && !remote) | |
1180 | continue; | |
1181 | ||
1182 | if ((local && !remote) || (!local && remote)) | |
1183 | status = -EINVAL; | |
1184 | ||
1185 | if (!status && | |
1186 | ((remote->ni_nodenum != local->nd_num) || | |
1187 | (remote->ni_ipv4_port != local->nd_ipv4_port) || | |
1188 | (remote->ni_ipv4_address != local->nd_ipv4_address))) | |
1189 | status = -EINVAL; | |
1190 | ||
1191 | if (status) { | |
1192 | if (remote && !local) | |
1193 | mlog(ML_ERROR, "Domain %s: Node %d (%pI4:%u) " | |
1194 | "registered in joining node %d but not in " | |
1195 | "local node %d\n", qn->qn_domain, | |
1196 | remote->ni_nodenum, | |
1197 | &(remote->ni_ipv4_address), | |
1198 | ntohs(remote->ni_ipv4_port), | |
1199 | qn->qn_nodenum, dlm->node_num); | |
1200 | if (local && !remote) | |
1201 | mlog(ML_ERROR, "Domain %s: Node %d (%pI4:%u) " | |
1202 | "registered in local node %d but not in " | |
1203 | "joining node %d\n", qn->qn_domain, | |
1204 | local->nd_num, &(local->nd_ipv4_address), | |
1205 | ntohs(local->nd_ipv4_port), | |
1206 | dlm->node_num, qn->qn_nodenum); | |
1207 | BUG_ON((!local && !remote)); | |
1208 | } | |
1209 | ||
1210 | if (local) | |
1211 | o2nm_node_put(local); | |
1212 | } | |
1213 | ||
1214 | return status; | |
1215 | } | |
1216 | ||
1217 | static int dlm_send_nodeinfo(struct dlm_ctxt *dlm, unsigned long *node_map) | |
1218 | { | |
1219 | struct dlm_query_nodeinfo *qn = NULL; | |
1220 | struct o2nm_node *node; | |
1221 | int ret = 0, status, count, i; | |
1222 | ||
1223 | if (find_next_bit(node_map, O2NM_MAX_NODES, 0) >= O2NM_MAX_NODES) | |
1224 | goto bail; | |
1225 | ||
1226 | qn = kzalloc(sizeof(struct dlm_query_nodeinfo), GFP_KERNEL); | |
1227 | if (!qn) { | |
1228 | ret = -ENOMEM; | |
1229 | mlog_errno(ret); | |
1230 | goto bail; | |
1231 | } | |
1232 | ||
1233 | for (i = 0, count = 0; i < O2NM_MAX_NODES; ++i) { | |
1234 | node = o2nm_get_node_by_num(i); | |
1235 | if (!node) | |
1236 | continue; | |
1237 | qn->qn_nodes[count].ni_nodenum = node->nd_num; | |
1238 | qn->qn_nodes[count].ni_ipv4_port = node->nd_ipv4_port; | |
1239 | qn->qn_nodes[count].ni_ipv4_address = node->nd_ipv4_address; | |
1240 | mlog(0, "Node %3d, %pI4:%u\n", node->nd_num, | |
1241 | &(node->nd_ipv4_address), ntohs(node->nd_ipv4_port)); | |
1242 | ++count; | |
1243 | o2nm_node_put(node); | |
1244 | } | |
1245 | ||
1246 | qn->qn_nodenum = dlm->node_num; | |
1247 | qn->qn_numnodes = count; | |
1248 | qn->qn_namelen = strlen(dlm->name); | |
1249 | memcpy(qn->qn_domain, dlm->name, qn->qn_namelen); | |
1250 | ||
1251 | i = -1; | |
1252 | while ((i = find_next_bit(node_map, O2NM_MAX_NODES, | |
1253 | i + 1)) < O2NM_MAX_NODES) { | |
1254 | if (i == dlm->node_num) | |
1255 | continue; | |
1256 | ||
1257 | mlog(0, "Sending nodeinfo to node %d\n", i); | |
1258 | ||
1259 | ret = o2net_send_message(DLM_QUERY_NODEINFO, DLM_MOD_KEY, | |
1260 | qn, sizeof(struct dlm_query_nodeinfo), | |
1261 | i, &status); | |
1262 | if (ret >= 0) | |
1263 | ret = status; | |
1264 | if (ret) { | |
1265 | mlog(ML_ERROR, "node mismatch %d, node %d\n", ret, i); | |
1266 | break; | |
1267 | } | |
1268 | } | |
1269 | ||
1270 | bail: | |
1271 | kfree(qn); | |
1272 | return ret; | |
1273 | } | |
1274 | ||
1275 | static int dlm_query_nodeinfo_handler(struct o2net_msg *msg, u32 len, | |
1276 | void *data, void **ret_data) | |
1277 | { | |
1278 | struct dlm_query_nodeinfo *qn; | |
1279 | struct dlm_ctxt *dlm = NULL; | |
1280 | int locked = 0, status = -EINVAL; | |
1281 | ||
1282 | qn = (struct dlm_query_nodeinfo *) msg->buf; | |
1283 | ||
1284 | mlog(0, "Node %u queries nodes on domain %s\n", qn->qn_nodenum, | |
1285 | qn->qn_domain); | |
1286 | ||
1287 | spin_lock(&dlm_domain_lock); | |
1288 | dlm = __dlm_lookup_domain_full(qn->qn_domain, qn->qn_namelen); | |
1289 | if (!dlm) { | |
1290 | mlog(ML_ERROR, "Node %d queried nodes on domain %s before " | |
1291 | "join domain\n", qn->qn_nodenum, qn->qn_domain); | |
1292 | goto bail; | |
1293 | } | |
1294 | ||
1295 | spin_lock(&dlm->spinlock); | |
1296 | locked = 1; | |
1297 | if (dlm->joining_node != qn->qn_nodenum) { | |
1298 | mlog(ML_ERROR, "Node %d queried nodes on domain %s but " | |
1299 | "joining node is %d\n", qn->qn_nodenum, qn->qn_domain, | |
1300 | dlm->joining_node); | |
1301 | goto bail; | |
1302 | } | |
1303 | ||
1304 | /* Support for node query was added in 1.1 */ | |
1305 | if (dlm->dlm_locking_proto.pv_major == 1 && | |
1306 | dlm->dlm_locking_proto.pv_minor == 0) { | |
1307 | mlog(ML_ERROR, "Node %d queried nodes on domain %s " | |
1308 | "but active dlm protocol is %d.%d\n", qn->qn_nodenum, | |
1309 | qn->qn_domain, dlm->dlm_locking_proto.pv_major, | |
1310 | dlm->dlm_locking_proto.pv_minor); | |
1311 | goto bail; | |
1312 | } | |
1313 | ||
1314 | status = dlm_match_nodes(dlm, qn); | |
1315 | ||
1316 | bail: | |
1317 | if (locked) | |
1318 | spin_unlock(&dlm->spinlock); | |
1319 | spin_unlock(&dlm_domain_lock); | |
1320 | ||
1321 | return status; | |
1322 | } | |
1323 | ||
d74c9803 KH |
1324 | static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data, |
1325 | void **ret_data) | |
6714d8e8 KH |
1326 | { |
1327 | struct dlm_cancel_join *cancel; | |
1328 | struct dlm_ctxt *dlm = NULL; | |
1329 | ||
1330 | cancel = (struct dlm_cancel_join *) msg->buf; | |
1331 | ||
1332 | mlog(0, "node %u cancels join on domain %s\n", cancel->node_idx, | |
1333 | cancel->domain); | |
1334 | ||
1335 | spin_lock(&dlm_domain_lock); | |
1336 | dlm = __dlm_lookup_domain_full(cancel->domain, cancel->name_len); | |
1337 | ||
1338 | if (dlm) { | |
1339 | spin_lock(&dlm->spinlock); | |
1340 | ||
1341 | /* Yikes, this guy wants to cancel his join. No | |
1342 | * problem, we simply cleanup our join state. */ | |
1343 | BUG_ON(dlm->joining_node != cancel->node_idx); | |
1344 | __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN); | |
1345 | ||
1346 | spin_unlock(&dlm->spinlock); | |
1347 | } | |
1348 | spin_unlock(&dlm_domain_lock); | |
1349 | ||
1350 | return 0; | |
1351 | } | |
1352 | ||
1353 | static int dlm_send_one_join_cancel(struct dlm_ctxt *dlm, | |
1354 | unsigned int node) | |
1355 | { | |
1356 | int status; | |
1357 | struct dlm_cancel_join cancel_msg; | |
1358 | ||
1359 | memset(&cancel_msg, 0, sizeof(cancel_msg)); | |
1360 | cancel_msg.node_idx = dlm->node_num; | |
1361 | cancel_msg.name_len = strlen(dlm->name); | |
1362 | memcpy(cancel_msg.domain, dlm->name, cancel_msg.name_len); | |
1363 | ||
1364 | status = o2net_send_message(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY, | |
1365 | &cancel_msg, sizeof(cancel_msg), node, | |
1366 | NULL); | |
1367 | if (status < 0) { | |
a5196ec5 WW |
1368 | mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to " |
1369 | "node %u\n", status, DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY, | |
1370 | node); | |
6714d8e8 KH |
1371 | goto bail; |
1372 | } | |
1373 | ||
1374 | bail: | |
1375 | return status; | |
1376 | } | |
1377 | ||
1378 | /* map_size should be in bytes. */ | |
1379 | static int dlm_send_join_cancels(struct dlm_ctxt *dlm, | |
1380 | unsigned long *node_map, | |
1381 | unsigned int map_size) | |
1382 | { | |
1383 | int status, tmpstat; | |
8d67d3c2 | 1384 | int node; |
6714d8e8 KH |
1385 | |
1386 | if (map_size != (BITS_TO_LONGS(O2NM_MAX_NODES) * | |
1387 | sizeof(unsigned long))) { | |
1388 | mlog(ML_ERROR, | |
1389 | "map_size %u != BITS_TO_LONGS(O2NM_MAX_NODES) %u\n", | |
3a4780a8 | 1390 | map_size, (unsigned)BITS_TO_LONGS(O2NM_MAX_NODES)); |
6714d8e8 KH |
1391 | return -EINVAL; |
1392 | } | |
1393 | ||
1394 | status = 0; | |
1395 | node = -1; | |
1396 | while ((node = find_next_bit(node_map, O2NM_MAX_NODES, | |
1397 | node + 1)) < O2NM_MAX_NODES) { | |
1398 | if (node == dlm->node_num) | |
1399 | continue; | |
1400 | ||
1401 | tmpstat = dlm_send_one_join_cancel(dlm, node); | |
1402 | if (tmpstat) { | |
1403 | mlog(ML_ERROR, "Error return %d cancelling join on " | |
1404 | "node %d\n", tmpstat, node); | |
1405 | if (!status) | |
1406 | status = tmpstat; | |
1407 | } | |
1408 | } | |
1409 | ||
1410 | if (status) | |
1411 | mlog_errno(status); | |
1412 | return status; | |
1413 | } | |
1414 | ||
1415 | static int dlm_request_join(struct dlm_ctxt *dlm, | |
1416 | int node, | |
d24fbcda | 1417 | enum dlm_query_join_response_code *response) |
6714d8e8 | 1418 | { |
d24fbcda | 1419 | int status; |
6714d8e8 | 1420 | struct dlm_query_join_request join_msg; |
0f71b7b4 JB |
1421 | struct dlm_query_join_packet packet; |
1422 | u32 join_resp; | |
6714d8e8 KH |
1423 | |
1424 | mlog(0, "querying node %d\n", node); | |
1425 | ||
1426 | memset(&join_msg, 0, sizeof(join_msg)); | |
1427 | join_msg.node_idx = dlm->node_num; | |
1428 | join_msg.name_len = strlen(dlm->name); | |
1429 | memcpy(join_msg.domain, dlm->name, join_msg.name_len); | |
d24fbcda JB |
1430 | join_msg.dlm_proto = dlm->dlm_locking_proto; |
1431 | join_msg.fs_proto = dlm->fs_locking_proto; | |
6714d8e8 | 1432 | |
1faf2894 SE |
1433 | /* copy live node map to join message */ |
1434 | byte_copymap(join_msg.node_map, dlm->live_nodes_map, O2NM_MAX_NODES); | |
1435 | ||
6714d8e8 | 1436 | status = o2net_send_message(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, &join_msg, |
a5196ec5 | 1437 | sizeof(join_msg), node, &join_resp); |
6714d8e8 | 1438 | if (status < 0 && status != -ENOPROTOOPT) { |
a5196ec5 WW |
1439 | mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to " |
1440 | "node %u\n", status, DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, | |
1441 | node); | |
6714d8e8 KH |
1442 | goto bail; |
1443 | } | |
0f71b7b4 | 1444 | dlm_query_join_wire_to_packet(join_resp, &packet); |
6714d8e8 KH |
1445 | |
1446 | /* -ENOPROTOOPT from the net code means the other side isn't | |
1447 | listening for our message type -- that's fine, it means | |
1448 | his dlm isn't up, so we can consider him a 'yes' but not | |
1449 | joined into the domain. */ | |
1450 | if (status == -ENOPROTOOPT) { | |
1451 | status = 0; | |
1452 | *response = JOIN_OK_NO_MAP; | |
6714d8e8 | 1453 | } else { |
72f6fe1f NZ |
1454 | *response = packet.code; |
1455 | switch (packet.code) { | |
1456 | case JOIN_DISALLOW: | |
1457 | case JOIN_OK_NO_MAP: | |
1458 | break; | |
1459 | case JOIN_PROTOCOL_MISMATCH: | |
1460 | mlog(ML_NOTICE, | |
1461 | "This node requested DLM locking protocol %u.%u and " | |
1462 | "filesystem locking protocol %u.%u. At least one of " | |
1463 | "the protocol versions on node %d is not compatible, " | |
1464 | "disconnecting\n", | |
1465 | dlm->dlm_locking_proto.pv_major, | |
1466 | dlm->dlm_locking_proto.pv_minor, | |
1467 | dlm->fs_locking_proto.pv_major, | |
1468 | dlm->fs_locking_proto.pv_minor, | |
1469 | node); | |
1470 | status = -EPROTO; | |
1471 | break; | |
1472 | case JOIN_OK: | |
1473 | /* Use the same locking protocol as the remote node */ | |
1474 | dlm->dlm_locking_proto.pv_minor = packet.dlm_minor; | |
1475 | dlm->fs_locking_proto.pv_minor = packet.fs_minor; | |
1476 | mlog(0, | |
1477 | "Node %d responds JOIN_OK with DLM locking protocol " | |
1478 | "%u.%u and fs locking protocol %u.%u\n", | |
1479 | node, | |
1480 | dlm->dlm_locking_proto.pv_major, | |
1481 | dlm->dlm_locking_proto.pv_minor, | |
1482 | dlm->fs_locking_proto.pv_major, | |
1483 | dlm->fs_locking_proto.pv_minor); | |
1484 | break; | |
1485 | default: | |
1486 | status = -EINVAL; | |
1487 | mlog(ML_ERROR, "invalid response %d from node %u\n", | |
1488 | packet.code, node); | |
1489 | /* Reset response to JOIN_DISALLOW */ | |
1490 | *response = JOIN_DISALLOW; | |
1491 | break; | |
1492 | } | |
6714d8e8 KH |
1493 | } |
1494 | ||
1495 | mlog(0, "status %d, node %d response is %d\n", status, node, | |
0f71b7b4 | 1496 | *response); |
6714d8e8 KH |
1497 | |
1498 | bail: | |
1499 | return status; | |
1500 | } | |
1501 | ||
1502 | static int dlm_send_one_join_assert(struct dlm_ctxt *dlm, | |
1503 | unsigned int node) | |
1504 | { | |
1505 | int status; | |
01c6222f | 1506 | int ret; |
6714d8e8 KH |
1507 | struct dlm_assert_joined assert_msg; |
1508 | ||
1509 | mlog(0, "Sending join assert to node %u\n", node); | |
1510 | ||
1511 | memset(&assert_msg, 0, sizeof(assert_msg)); | |
1512 | assert_msg.node_idx = dlm->node_num; | |
1513 | assert_msg.name_len = strlen(dlm->name); | |
1514 | memcpy(assert_msg.domain, dlm->name, assert_msg.name_len); | |
1515 | ||
1516 | status = o2net_send_message(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY, | |
1517 | &assert_msg, sizeof(assert_msg), node, | |
01c6222f | 1518 | &ret); |
6714d8e8 | 1519 | if (status < 0) |
a5196ec5 WW |
1520 | mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to " |
1521 | "node %u\n", status, DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY, | |
1522 | node); | |
01c6222f X |
1523 | else |
1524 | status = ret; | |
6714d8e8 KH |
1525 | |
1526 | return status; | |
1527 | } | |
1528 | ||
1529 | static void dlm_send_join_asserts(struct dlm_ctxt *dlm, | |
1530 | unsigned long *node_map) | |
1531 | { | |
1532 | int status, node, live; | |
1533 | ||
1534 | status = 0; | |
1535 | node = -1; | |
1536 | while ((node = find_next_bit(node_map, O2NM_MAX_NODES, | |
1537 | node + 1)) < O2NM_MAX_NODES) { | |
1538 | if (node == dlm->node_num) | |
1539 | continue; | |
1540 | ||
1541 | do { | |
1542 | /* It is very important that this message be | |
1543 | * received so we spin until either the node | |
1544 | * has died or it gets the message. */ | |
1545 | status = dlm_send_one_join_assert(dlm, node); | |
1546 | ||
1547 | spin_lock(&dlm->spinlock); | |
1548 | live = test_bit(node, dlm->live_nodes_map); | |
1549 | spin_unlock(&dlm->spinlock); | |
1550 | ||
1551 | if (status) { | |
1552 | mlog(ML_ERROR, "Error return %d asserting " | |
1553 | "join on node %d\n", status, node); | |
1554 | ||
1555 | /* give us some time between errors... */ | |
1556 | if (live) | |
1557 | msleep(DLM_DOMAIN_BACKOFF_MS); | |
1558 | } | |
1559 | } while (status && live); | |
1560 | } | |
1561 | } | |
1562 | ||
1563 | struct domain_join_ctxt { | |
1564 | unsigned long live_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; | |
1565 | unsigned long yes_resp_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; | |
1566 | }; | |
1567 | ||
1568 | static int dlm_should_restart_join(struct dlm_ctxt *dlm, | |
1569 | struct domain_join_ctxt *ctxt, | |
d24fbcda | 1570 | enum dlm_query_join_response_code response) |
6714d8e8 KH |
1571 | { |
1572 | int ret; | |
1573 | ||
1574 | if (response == JOIN_DISALLOW) { | |
1575 | mlog(0, "Latest response of disallow -- should restart\n"); | |
1576 | return 1; | |
1577 | } | |
1578 | ||
1579 | spin_lock(&dlm->spinlock); | |
1580 | /* For now, we restart the process if the node maps have | |
1581 | * changed at all */ | |
1582 | ret = memcmp(ctxt->live_map, dlm->live_nodes_map, | |
1583 | sizeof(dlm->live_nodes_map)); | |
1584 | spin_unlock(&dlm->spinlock); | |
1585 | ||
1586 | if (ret) | |
1587 | mlog(0, "Node maps changed -- should restart\n"); | |
1588 | ||
1589 | return ret; | |
1590 | } | |
1591 | ||
1592 | static int dlm_try_to_join_domain(struct dlm_ctxt *dlm) | |
1593 | { | |
1594 | int status = 0, tmpstat, node; | |
1595 | struct domain_join_ctxt *ctxt; | |
d24fbcda | 1596 | enum dlm_query_join_response_code response = JOIN_DISALLOW; |
6714d8e8 | 1597 | |
ef6b689b | 1598 | mlog(0, "%p", dlm); |
6714d8e8 | 1599 | |
cd861280 | 1600 | ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL); |
6714d8e8 KH |
1601 | if (!ctxt) { |
1602 | status = -ENOMEM; | |
1603 | mlog_errno(status); | |
1604 | goto bail; | |
1605 | } | |
1606 | ||
1607 | /* group sem locking should work for us here -- we're already | |
1608 | * registered for heartbeat events so filling this should be | |
1609 | * atomic wrt getting those handlers called. */ | |
1610 | o2hb_fill_node_map(dlm->live_nodes_map, sizeof(dlm->live_nodes_map)); | |
1611 | ||
1612 | spin_lock(&dlm->spinlock); | |
1613 | memcpy(ctxt->live_map, dlm->live_nodes_map, sizeof(ctxt->live_map)); | |
1614 | ||
1615 | __dlm_set_joining_node(dlm, dlm->node_num); | |
1616 | ||
1617 | spin_unlock(&dlm->spinlock); | |
1618 | ||
1619 | node = -1; | |
1620 | while ((node = find_next_bit(ctxt->live_map, O2NM_MAX_NODES, | |
1621 | node + 1)) < O2NM_MAX_NODES) { | |
1622 | if (node == dlm->node_num) | |
1623 | continue; | |
1624 | ||
1625 | status = dlm_request_join(dlm, node, &response); | |
1626 | if (status < 0) { | |
1627 | mlog_errno(status); | |
1628 | goto bail; | |
1629 | } | |
1630 | ||
1631 | /* Ok, either we got a response or the node doesn't have a | |
1632 | * dlm up. */ | |
1633 | if (response == JOIN_OK) | |
1634 | set_bit(node, ctxt->yes_resp_map); | |
1635 | ||
1636 | if (dlm_should_restart_join(dlm, ctxt, response)) { | |
1637 | status = -EAGAIN; | |
1638 | goto bail; | |
1639 | } | |
1640 | } | |
1641 | ||
1642 | mlog(0, "Yay, done querying nodes!\n"); | |
1643 | ||
1644 | /* Yay, everyone agree's we can join the domain. My domain is | |
1645 | * comprised of all nodes who were put in the | |
1646 | * yes_resp_map. Copy that into our domain map and send a join | |
1647 | * assert message to clean up everyone elses state. */ | |
1648 | spin_lock(&dlm->spinlock); | |
1649 | memcpy(dlm->domain_map, ctxt->yes_resp_map, | |
1650 | sizeof(ctxt->yes_resp_map)); | |
1651 | set_bit(dlm->node_num, dlm->domain_map); | |
1652 | spin_unlock(&dlm->spinlock); | |
1653 | ||
18cfdf1b | 1654 | /* Support for global heartbeat and node info was added in 1.1 */ |
4da6dc29 SM |
1655 | if (dlm->dlm_locking_proto.pv_major > 1 || |
1656 | dlm->dlm_locking_proto.pv_minor > 0) { | |
18cfdf1b SM |
1657 | status = dlm_send_nodeinfo(dlm, ctxt->yes_resp_map); |
1658 | if (status) { | |
1659 | mlog_errno(status); | |
1660 | goto bail; | |
1661 | } | |
ea203441 SM |
1662 | status = dlm_send_regions(dlm, ctxt->yes_resp_map); |
1663 | if (status) { | |
1664 | mlog_errno(status); | |
1665 | goto bail; | |
1666 | } | |
1667 | } | |
1668 | ||
6714d8e8 KH |
1669 | dlm_send_join_asserts(dlm, ctxt->yes_resp_map); |
1670 | ||
1671 | /* Joined state *must* be set before the joining node | |
1672 | * information, otherwise the query_join handler may read no | |
1673 | * current joiner but a state of NEW and tell joining nodes | |
1674 | * we're not in the domain. */ | |
1675 | spin_lock(&dlm_domain_lock); | |
1676 | dlm->dlm_state = DLM_CTXT_JOINED; | |
1677 | dlm->num_joins++; | |
1678 | spin_unlock(&dlm_domain_lock); | |
1679 | ||
1680 | bail: | |
1681 | spin_lock(&dlm->spinlock); | |
1682 | __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN); | |
8decab3c SM |
1683 | if (!status) { |
1684 | printk(KERN_NOTICE "o2dlm: Joining domain %s ", dlm->name); | |
6714d8e8 | 1685 | __dlm_print_nodes(dlm); |
8decab3c | 1686 | } |
6714d8e8 KH |
1687 | spin_unlock(&dlm->spinlock); |
1688 | ||
1689 | if (ctxt) { | |
1690 | /* Do we need to send a cancel message to any nodes? */ | |
1691 | if (status < 0) { | |
1692 | tmpstat = dlm_send_join_cancels(dlm, | |
1693 | ctxt->yes_resp_map, | |
1694 | sizeof(ctxt->yes_resp_map)); | |
1695 | if (tmpstat < 0) | |
1696 | mlog_errno(tmpstat); | |
1697 | } | |
1698 | kfree(ctxt); | |
1699 | } | |
1700 | ||
1701 | mlog(0, "returning %d\n", status); | |
1702 | return status; | |
1703 | } | |
1704 | ||
1705 | static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm) | |
1706 | { | |
58a3158a SM |
1707 | o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_up); |
1708 | o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_down); | |
6714d8e8 KH |
1709 | o2net_unregister_handler_list(&dlm->dlm_domain_handlers); |
1710 | } | |
1711 | ||
1712 | static int dlm_register_domain_handlers(struct dlm_ctxt *dlm) | |
1713 | { | |
1714 | int status; | |
1715 | ||
1716 | mlog(0, "registering handlers.\n"); | |
1717 | ||
1718 | o2hb_setup_callback(&dlm->dlm_hb_down, O2HB_NODE_DOWN_CB, | |
1719 | dlm_hb_node_down_cb, dlm, DLM_HB_NODE_DOWN_PRI); | |
cdd09f49 JQ |
1720 | o2hb_setup_callback(&dlm->dlm_hb_up, O2HB_NODE_UP_CB, |
1721 | dlm_hb_node_up_cb, dlm, DLM_HB_NODE_UP_PRI); | |
1722 | ||
58a3158a | 1723 | status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_down); |
6714d8e8 KH |
1724 | if (status) |
1725 | goto bail; | |
1726 | ||
58a3158a | 1727 | status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_up); |
6714d8e8 KH |
1728 | if (status) |
1729 | goto bail; | |
1730 | ||
1731 | status = o2net_register_handler(DLM_MASTER_REQUEST_MSG, dlm->key, | |
1732 | sizeof(struct dlm_master_request), | |
1733 | dlm_master_request_handler, | |
d74c9803 | 1734 | dlm, NULL, &dlm->dlm_domain_handlers); |
6714d8e8 KH |
1735 | if (status) |
1736 | goto bail; | |
1737 | ||
1738 | status = o2net_register_handler(DLM_ASSERT_MASTER_MSG, dlm->key, | |
1739 | sizeof(struct dlm_assert_master), | |
1740 | dlm_assert_master_handler, | |
3b8118cf KH |
1741 | dlm, dlm_assert_master_post_handler, |
1742 | &dlm->dlm_domain_handlers); | |
6714d8e8 KH |
1743 | if (status) |
1744 | goto bail; | |
1745 | ||
1746 | status = o2net_register_handler(DLM_CREATE_LOCK_MSG, dlm->key, | |
1747 | sizeof(struct dlm_create_lock), | |
1748 | dlm_create_lock_handler, | |
d74c9803 | 1749 | dlm, NULL, &dlm->dlm_domain_handlers); |
6714d8e8 KH |
1750 | if (status) |
1751 | goto bail; | |
1752 | ||
1753 | status = o2net_register_handler(DLM_CONVERT_LOCK_MSG, dlm->key, | |
1754 | DLM_CONVERT_LOCK_MAX_LEN, | |
1755 | dlm_convert_lock_handler, | |
d74c9803 | 1756 | dlm, NULL, &dlm->dlm_domain_handlers); |
6714d8e8 KH |
1757 | if (status) |
1758 | goto bail; | |
1759 | ||
1760 | status = o2net_register_handler(DLM_UNLOCK_LOCK_MSG, dlm->key, | |
1761 | DLM_UNLOCK_LOCK_MAX_LEN, | |
1762 | dlm_unlock_lock_handler, | |
d74c9803 | 1763 | dlm, NULL, &dlm->dlm_domain_handlers); |
6714d8e8 KH |
1764 | if (status) |
1765 | goto bail; | |
1766 | ||
1767 | status = o2net_register_handler(DLM_PROXY_AST_MSG, dlm->key, | |
1768 | DLM_PROXY_AST_MAX_LEN, | |
1769 | dlm_proxy_ast_handler, | |
d74c9803 | 1770 | dlm, NULL, &dlm->dlm_domain_handlers); |
6714d8e8 KH |
1771 | if (status) |
1772 | goto bail; | |
1773 | ||
1774 | status = o2net_register_handler(DLM_EXIT_DOMAIN_MSG, dlm->key, | |
1775 | sizeof(struct dlm_exit_domain), | |
1776 | dlm_exit_domain_handler, | |
d74c9803 | 1777 | dlm, NULL, &dlm->dlm_domain_handlers); |
6714d8e8 KH |
1778 | if (status) |
1779 | goto bail; | |
1780 | ||
ba2bf218 KH |
1781 | status = o2net_register_handler(DLM_DEREF_LOCKRES_MSG, dlm->key, |
1782 | sizeof(struct dlm_deref_lockres), | |
1783 | dlm_deref_lockres_handler, | |
d74c9803 | 1784 | dlm, NULL, &dlm->dlm_domain_handlers); |
ba2bf218 KH |
1785 | if (status) |
1786 | goto bail; | |
1787 | ||
6714d8e8 KH |
1788 | status = o2net_register_handler(DLM_MIGRATE_REQUEST_MSG, dlm->key, |
1789 | sizeof(struct dlm_migrate_request), | |
1790 | dlm_migrate_request_handler, | |
d74c9803 | 1791 | dlm, NULL, &dlm->dlm_domain_handlers); |
6714d8e8 KH |
1792 | if (status) |
1793 | goto bail; | |
1794 | ||
1795 | status = o2net_register_handler(DLM_MIG_LOCKRES_MSG, dlm->key, | |
1796 | DLM_MIG_LOCKRES_MAX_LEN, | |
1797 | dlm_mig_lockres_handler, | |
d74c9803 | 1798 | dlm, NULL, &dlm->dlm_domain_handlers); |
6714d8e8 KH |
1799 | if (status) |
1800 | goto bail; | |
1801 | ||
1802 | status = o2net_register_handler(DLM_MASTER_REQUERY_MSG, dlm->key, | |
1803 | sizeof(struct dlm_master_requery), | |
1804 | dlm_master_requery_handler, | |
d74c9803 | 1805 | dlm, NULL, &dlm->dlm_domain_handlers); |
6714d8e8 KH |
1806 | if (status) |
1807 | goto bail; | |
1808 | ||
1809 | status = o2net_register_handler(DLM_LOCK_REQUEST_MSG, dlm->key, | |
1810 | sizeof(struct dlm_lock_request), | |
1811 | dlm_request_all_locks_handler, | |
d74c9803 | 1812 | dlm, NULL, &dlm->dlm_domain_handlers); |
6714d8e8 KH |
1813 | if (status) |
1814 | goto bail; | |
1815 | ||
1816 | status = o2net_register_handler(DLM_RECO_DATA_DONE_MSG, dlm->key, | |
1817 | sizeof(struct dlm_reco_data_done), | |
1818 | dlm_reco_data_done_handler, | |
d74c9803 | 1819 | dlm, NULL, &dlm->dlm_domain_handlers); |
6714d8e8 KH |
1820 | if (status) |
1821 | goto bail; | |
1822 | ||
1823 | status = o2net_register_handler(DLM_BEGIN_RECO_MSG, dlm->key, | |
1824 | sizeof(struct dlm_begin_reco), | |
1825 | dlm_begin_reco_handler, | |
d74c9803 | 1826 | dlm, NULL, &dlm->dlm_domain_handlers); |
6714d8e8 KH |
1827 | if (status) |
1828 | goto bail; | |
1829 | ||
1830 | status = o2net_register_handler(DLM_FINALIZE_RECO_MSG, dlm->key, | |
1831 | sizeof(struct dlm_finalize_reco), | |
1832 | dlm_finalize_reco_handler, | |
d74c9803 | 1833 | dlm, NULL, &dlm->dlm_domain_handlers); |
6714d8e8 KH |
1834 | if (status) |
1835 | goto bail; | |
1836 | ||
bddefdee SM |
1837 | status = o2net_register_handler(DLM_BEGIN_EXIT_DOMAIN_MSG, dlm->key, |
1838 | sizeof(struct dlm_exit_domain), | |
1839 | dlm_begin_exit_domain_handler, | |
1840 | dlm, NULL, &dlm->dlm_domain_handlers); | |
60d663cb | 1841 | if (status) |
1842 | goto bail; | |
bddefdee | 1843 | |
60d663cb | 1844 | status = o2net_register_handler(DLM_DEREF_LOCKRES_DONE, dlm->key, |
1845 | sizeof(struct dlm_deref_lockres_done), | |
1846 | dlm_deref_lockres_done_handler, | |
1847 | dlm, NULL, &dlm->dlm_domain_handlers); | |
6714d8e8 KH |
1848 | bail: |
1849 | if (status) | |
1850 | dlm_unregister_domain_handlers(dlm); | |
1851 | ||
1852 | return status; | |
1853 | } | |
1854 | ||
1855 | static int dlm_join_domain(struct dlm_ctxt *dlm) | |
1856 | { | |
1857 | int status; | |
0dd82141 SM |
1858 | unsigned int backoff; |
1859 | unsigned int total_backoff = 0; | |
5afc44e2 | 1860 | char wq_name[O2NM_MAX_NAME_LEN]; |
6714d8e8 KH |
1861 | |
1862 | BUG_ON(!dlm); | |
1863 | ||
1864 | mlog(0, "Join domain %s\n", dlm->name); | |
1865 | ||
1866 | status = dlm_register_domain_handlers(dlm); | |
1867 | if (status) { | |
1868 | mlog_errno(status); | |
1869 | goto bail; | |
1870 | } | |
1871 | ||
181a9a04 | 1872 | status = dlm_launch_thread(dlm); |
007dce53 SM |
1873 | if (status < 0) { |
1874 | mlog_errno(status); | |
1875 | goto bail; | |
1876 | } | |
1877 | ||
181a9a04 | 1878 | status = dlm_launch_recovery_thread(dlm); |
6714d8e8 KH |
1879 | if (status < 0) { |
1880 | mlog_errno(status); | |
1881 | goto bail; | |
1882 | } | |
1883 | ||
e581595e | 1884 | dlm_debug_init(dlm); |
6714d8e8 | 1885 | |
5afc44e2 | 1886 | snprintf(wq_name, O2NM_MAX_NAME_LEN, "dlm_wq-%s", dlm->name); |
055fdcff | 1887 | dlm->dlm_worker = alloc_workqueue(wq_name, WQ_MEM_RECLAIM, 0); |
3156d267 KH |
1888 | if (!dlm->dlm_worker) { |
1889 | status = -ENOMEM; | |
1890 | mlog_errno(status); | |
1891 | goto bail; | |
1892 | } | |
1893 | ||
6714d8e8 | 1894 | do { |
6714d8e8 KH |
1895 | status = dlm_try_to_join_domain(dlm); |
1896 | ||
1897 | /* If we're racing another node to the join, then we | |
1898 | * need to back off temporarily and let them | |
1899 | * complete. */ | |
0dd82141 | 1900 | #define DLM_JOIN_TIMEOUT_MSECS 90000 |
6714d8e8 KH |
1901 | if (status == -EAGAIN) { |
1902 | if (signal_pending(current)) { | |
1903 | status = -ERESTARTSYS; | |
1904 | goto bail; | |
1905 | } | |
1906 | ||
7567c148 | 1907 | if (total_backoff > DLM_JOIN_TIMEOUT_MSECS) { |
0dd82141 SM |
1908 | status = -ERESTARTSYS; |
1909 | mlog(ML_NOTICE, "Timed out joining dlm domain " | |
1910 | "%s after %u msecs\n", dlm->name, | |
7567c148 | 1911 | total_backoff); |
0dd82141 SM |
1912 | goto bail; |
1913 | } | |
1914 | ||
6714d8e8 KH |
1915 | /* |
1916 | * <chip> After you! | |
1917 | * <dale> No, after you! | |
1918 | * <chip> I insist! | |
1919 | * <dale> But you first! | |
1920 | * ... | |
1921 | */ | |
1922 | backoff = (unsigned int)(jiffies & 0x3); | |
1923 | backoff *= DLM_DOMAIN_BACKOFF_MS; | |
0dd82141 | 1924 | total_backoff += backoff; |
6714d8e8 KH |
1925 | mlog(0, "backoff %d\n", backoff); |
1926 | msleep(backoff); | |
1927 | } | |
1928 | } while (status == -EAGAIN); | |
1929 | ||
1930 | if (status < 0) { | |
1931 | mlog_errno(status); | |
1932 | goto bail; | |
1933 | } | |
1934 | ||
1935 | status = 0; | |
1936 | bail: | |
1937 | wake_up(&dlm_domain_events); | |
1938 | ||
1939 | if (status) { | |
1940 | dlm_unregister_domain_handlers(dlm); | |
007dce53 | 1941 | dlm_debug_shutdown(dlm); |
6714d8e8 KH |
1942 | dlm_complete_thread(dlm); |
1943 | dlm_complete_recovery_thread(dlm); | |
3156d267 | 1944 | dlm_destroy_dlm_worker(dlm); |
6714d8e8 KH |
1945 | } |
1946 | ||
1947 | return status; | |
1948 | } | |
1949 | ||
1950 | static struct dlm_ctxt *dlm_alloc_ctxt(const char *domain, | |
1951 | u32 key) | |
1952 | { | |
1953 | int i; | |
6325b4a2 | 1954 | int ret; |
6714d8e8 KH |
1955 | struct dlm_ctxt *dlm = NULL; |
1956 | ||
cd861280 | 1957 | dlm = kzalloc(sizeof(*dlm), GFP_KERNEL); |
6714d8e8 | 1958 | if (!dlm) { |
190a7721 JQ |
1959 | ret = -ENOMEM; |
1960 | mlog_errno(ret); | |
6714d8e8 KH |
1961 | goto leave; |
1962 | } | |
1963 | ||
316ce2ba | 1964 | dlm->name = kstrdup(domain, GFP_KERNEL); |
6714d8e8 | 1965 | if (dlm->name == NULL) { |
190a7721 JQ |
1966 | ret = -ENOMEM; |
1967 | mlog_errno(ret); | |
6714d8e8 KH |
1968 | goto leave; |
1969 | } | |
1970 | ||
03d864c0 | 1971 | dlm->lockres_hash = (struct hlist_head **)dlm_alloc_pagevec(DLM_HASH_PAGES); |
81f2094a | 1972 | if (!dlm->lockres_hash) { |
190a7721 JQ |
1973 | ret = -ENOMEM; |
1974 | mlog_errno(ret); | |
6714d8e8 KH |
1975 | goto leave; |
1976 | } | |
6714d8e8 | 1977 | |
03d864c0 DP |
1978 | for (i = 0; i < DLM_HASH_BUCKETS; i++) |
1979 | INIT_HLIST_HEAD(dlm_lockres_hash(dlm, i)); | |
6714d8e8 | 1980 | |
e2b66ddc SM |
1981 | dlm->master_hash = (struct hlist_head **) |
1982 | dlm_alloc_pagevec(DLM_HASH_PAGES); | |
1983 | if (!dlm->master_hash) { | |
190a7721 JQ |
1984 | ret = -ENOMEM; |
1985 | mlog_errno(ret); | |
e2b66ddc SM |
1986 | goto leave; |
1987 | } | |
1988 | ||
1989 | for (i = 0; i < DLM_HASH_BUCKETS; i++) | |
1990 | INIT_HLIST_HEAD(dlm_master_hash(dlm, i)); | |
1991 | ||
6714d8e8 KH |
1992 | dlm->key = key; |
1993 | dlm->node_num = o2nm_this_node(); | |
1994 | ||
6325b4a2 | 1995 | ret = dlm_create_debugfs_subroot(dlm); |
190a7721 | 1996 | if (ret < 0) |
6325b4a2 | 1997 | goto leave; |
6325b4a2 | 1998 | |
6714d8e8 KH |
1999 | spin_lock_init(&dlm->spinlock); |
2000 | spin_lock_init(&dlm->master_lock); | |
2001 | spin_lock_init(&dlm->ast_lock); | |
b0d4f817 | 2002 | spin_lock_init(&dlm->track_lock); |
6714d8e8 KH |
2003 | INIT_LIST_HEAD(&dlm->list); |
2004 | INIT_LIST_HEAD(&dlm->dirty_list); | |
2005 | INIT_LIST_HEAD(&dlm->reco.resources); | |
6714d8e8 KH |
2006 | INIT_LIST_HEAD(&dlm->reco.node_data); |
2007 | INIT_LIST_HEAD(&dlm->purge_list); | |
2008 | INIT_LIST_HEAD(&dlm->dlm_domain_handlers); | |
29576f8b | 2009 | INIT_LIST_HEAD(&dlm->tracking_list); |
6714d8e8 KH |
2010 | dlm->reco.state = 0; |
2011 | ||
2012 | INIT_LIST_HEAD(&dlm->pending_asts); | |
2013 | INIT_LIST_HEAD(&dlm->pending_basts); | |
2014 | ||
2015 | mlog(0, "dlm->recovery_map=%p, &(dlm->recovery_map[0])=%p\n", | |
2016 | dlm->recovery_map, &(dlm->recovery_map[0])); | |
2017 | ||
2018 | memset(dlm->recovery_map, 0, sizeof(dlm->recovery_map)); | |
2019 | memset(dlm->live_nodes_map, 0, sizeof(dlm->live_nodes_map)); | |
2020 | memset(dlm->domain_map, 0, sizeof(dlm->domain_map)); | |
2021 | ||
2022 | dlm->dlm_thread_task = NULL; | |
2023 | dlm->dlm_reco_thread_task = NULL; | |
3156d267 | 2024 | dlm->dlm_worker = NULL; |
6714d8e8 KH |
2025 | init_waitqueue_head(&dlm->dlm_thread_wq); |
2026 | init_waitqueue_head(&dlm->dlm_reco_thread_wq); | |
2027 | init_waitqueue_head(&dlm->reco.event); | |
2028 | init_waitqueue_head(&dlm->ast_wq); | |
2029 | init_waitqueue_head(&dlm->migration_wq); | |
6714d8e8 KH |
2030 | INIT_LIST_HEAD(&dlm->mle_hb_events); |
2031 | ||
2032 | dlm->joining_node = DLM_LOCK_RES_OWNER_UNKNOWN; | |
2033 | init_waitqueue_head(&dlm->dlm_join_events); | |
2034 | ||
60c7ec9e | 2035 | dlm->migrate_done = 0; |
2036 | ||
6714d8e8 KH |
2037 | dlm->reco.new_master = O2NM_INVALID_NODE_NUM; |
2038 | dlm->reco.dead_node = O2NM_INVALID_NODE_NUM; | |
6714d8e8 | 2039 | |
6800791a SM |
2040 | atomic_set(&dlm->res_tot_count, 0); |
2041 | atomic_set(&dlm->res_cur_count, 0); | |
2041d8fd SM |
2042 | for (i = 0; i < DLM_MLE_NUM_TYPES; ++i) { |
2043 | atomic_set(&dlm->mle_tot_count[i], 0); | |
2044 | atomic_set(&dlm->mle_cur_count[i], 0); | |
2045 | } | |
2046 | ||
6714d8e8 KH |
2047 | spin_lock_init(&dlm->work_lock); |
2048 | INIT_LIST_HEAD(&dlm->work_list); | |
c4028958 | 2049 | INIT_WORK(&dlm->dispatched_work, dlm_dispatch_work); |
6714d8e8 KH |
2050 | |
2051 | kref_init(&dlm->dlm_refs); | |
2052 | dlm->dlm_state = DLM_CTXT_NEW; | |
2053 | ||
2054 | INIT_LIST_HEAD(&dlm->dlm_eviction_callbacks); | |
2055 | ||
2056 | mlog(0, "context init: refcount %u\n", | |
2c935bc5 | 2057 | kref_read(&dlm->dlm_refs)); |
6714d8e8 KH |
2058 | |
2059 | leave: | |
190a7721 JQ |
2060 | if (ret < 0 && dlm) { |
2061 | if (dlm->master_hash) | |
2062 | dlm_free_pagevec((void **)dlm->master_hash, | |
2063 | DLM_HASH_PAGES); | |
2064 | ||
2065 | if (dlm->lockres_hash) | |
2066 | dlm_free_pagevec((void **)dlm->lockres_hash, | |
2067 | DLM_HASH_PAGES); | |
2068 | ||
2069 | kfree(dlm->name); | |
2070 | kfree(dlm); | |
2071 | dlm = NULL; | |
2072 | } | |
6714d8e8 KH |
2073 | return dlm; |
2074 | } | |
2075 | ||
2076 | /* | |
d24fbcda JB |
2077 | * Compare a requested locking protocol version against the current one. |
2078 | * | |
2079 | * If the major numbers are different, they are incompatible. | |
2080 | * If the current minor is greater than the request, they are incompatible. | |
2081 | * If the current minor is less than or equal to the request, they are | |
2082 | * compatible, and the requester should run at the current minor version. | |
2083 | */ | |
2084 | static int dlm_protocol_compare(struct dlm_protocol_version *existing, | |
2085 | struct dlm_protocol_version *request) | |
2086 | { | |
2087 | if (existing->pv_major != request->pv_major) | |
2088 | return 1; | |
2089 | ||
2090 | if (existing->pv_minor > request->pv_minor) | |
2091 | return 1; | |
2092 | ||
2093 | if (existing->pv_minor < request->pv_minor) | |
2094 | request->pv_minor = existing->pv_minor; | |
2095 | ||
2096 | return 0; | |
2097 | } | |
2098 | ||
2099 | /* | |
2100 | * dlm_register_domain: one-time setup per "domain". | |
2101 | * | |
2102 | * The filesystem passes in the requested locking version via proto. | |
2103 | * If registration was successful, proto will contain the negotiated | |
2104 | * locking protocol. | |
6714d8e8 KH |
2105 | */ |
2106 | struct dlm_ctxt * dlm_register_domain(const char *domain, | |
d24fbcda JB |
2107 | u32 key, |
2108 | struct dlm_protocol_version *fs_proto) | |
6714d8e8 KH |
2109 | { |
2110 | int ret; | |
2111 | struct dlm_ctxt *dlm = NULL; | |
2112 | struct dlm_ctxt *new_ctxt = NULL; | |
2113 | ||
e372357b | 2114 | if (strlen(domain) >= O2NM_MAX_NAME_LEN) { |
6714d8e8 KH |
2115 | ret = -ENAMETOOLONG; |
2116 | mlog(ML_ERROR, "domain name length too long\n"); | |
2117 | goto leave; | |
2118 | } | |
2119 | ||
6714d8e8 KH |
2120 | mlog(0, "register called for domain \"%s\"\n", domain); |
2121 | ||
2122 | retry: | |
2123 | dlm = NULL; | |
2124 | if (signal_pending(current)) { | |
2125 | ret = -ERESTARTSYS; | |
2126 | mlog_errno(ret); | |
2127 | goto leave; | |
2128 | } | |
2129 | ||
2130 | spin_lock(&dlm_domain_lock); | |
2131 | ||
2132 | dlm = __dlm_lookup_domain(domain); | |
2133 | if (dlm) { | |
2134 | if (dlm->dlm_state != DLM_CTXT_JOINED) { | |
2135 | spin_unlock(&dlm_domain_lock); | |
2136 | ||
2137 | mlog(0, "This ctxt is not joined yet!\n"); | |
2138 | wait_event_interruptible(dlm_domain_events, | |
2139 | dlm_wait_on_domain_helper( | |
2140 | domain)); | |
2141 | goto retry; | |
2142 | } | |
2143 | ||
d24fbcda | 2144 | if (dlm_protocol_compare(&dlm->fs_locking_proto, fs_proto)) { |
6469272c | 2145 | spin_unlock(&dlm_domain_lock); |
d24fbcda JB |
2146 | mlog(ML_ERROR, |
2147 | "Requested locking protocol version is not " | |
2148 | "compatible with already registered domain " | |
2149 | "\"%s\"\n", domain); | |
2150 | ret = -EPROTO; | |
2151 | goto leave; | |
2152 | } | |
2153 | ||
6714d8e8 KH |
2154 | __dlm_get(dlm); |
2155 | dlm->num_joins++; | |
2156 | ||
2157 | spin_unlock(&dlm_domain_lock); | |
2158 | ||
2159 | ret = 0; | |
2160 | goto leave; | |
2161 | } | |
2162 | ||
2163 | /* doesn't exist */ | |
2164 | if (!new_ctxt) { | |
2165 | spin_unlock(&dlm_domain_lock); | |
2166 | ||
2167 | new_ctxt = dlm_alloc_ctxt(domain, key); | |
2168 | if (new_ctxt) | |
2169 | goto retry; | |
2170 | ||
2171 | ret = -ENOMEM; | |
2172 | mlog_errno(ret); | |
2173 | goto leave; | |
2174 | } | |
2175 | ||
2176 | /* a little variable switch-a-roo here... */ | |
2177 | dlm = new_ctxt; | |
2178 | new_ctxt = NULL; | |
2179 | ||
2180 | /* add the new domain */ | |
2181 | list_add_tail(&dlm->list, &dlm_domains); | |
2182 | spin_unlock(&dlm_domain_lock); | |
2183 | ||
d24fbcda JB |
2184 | /* |
2185 | * Pass the locking protocol version into the join. If the join | |
2186 | * succeeds, it will have the negotiated protocol set. | |
2187 | */ | |
2188 | dlm->dlm_locking_proto = dlm_protocol; | |
2189 | dlm->fs_locking_proto = *fs_proto; | |
2190 | ||
6714d8e8 KH |
2191 | ret = dlm_join_domain(dlm); |
2192 | if (ret) { | |
2193 | mlog_errno(ret); | |
2194 | dlm_put(dlm); | |
2195 | goto leave; | |
2196 | } | |
2197 | ||
d24fbcda JB |
2198 | /* Tell the caller what locking protocol we negotiated */ |
2199 | *fs_proto = dlm->fs_locking_proto; | |
2200 | ||
6714d8e8 KH |
2201 | ret = 0; |
2202 | leave: | |
2203 | if (new_ctxt) | |
2204 | dlm_free_ctxt_mem(new_ctxt); | |
2205 | ||
2206 | if (ret < 0) | |
2207 | dlm = ERR_PTR(ret); | |
2208 | ||
2209 | return dlm; | |
2210 | } | |
2211 | EXPORT_SYMBOL_GPL(dlm_register_domain); | |
2212 | ||
2213 | static LIST_HEAD(dlm_join_handlers); | |
2214 | ||
2215 | static void dlm_unregister_net_handlers(void) | |
2216 | { | |
2217 | o2net_unregister_handler_list(&dlm_join_handlers); | |
2218 | } | |
2219 | ||
2220 | static int dlm_register_net_handlers(void) | |
2221 | { | |
2222 | int status = 0; | |
2223 | ||
2224 | status = o2net_register_handler(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, | |
2225 | sizeof(struct dlm_query_join_request), | |
2226 | dlm_query_join_handler, | |
d74c9803 | 2227 | NULL, NULL, &dlm_join_handlers); |
6714d8e8 KH |
2228 | if (status) |
2229 | goto bail; | |
2230 | ||
2231 | status = o2net_register_handler(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY, | |
2232 | sizeof(struct dlm_assert_joined), | |
2233 | dlm_assert_joined_handler, | |
d74c9803 | 2234 | NULL, NULL, &dlm_join_handlers); |
6714d8e8 KH |
2235 | if (status) |
2236 | goto bail; | |
2237 | ||
2238 | status = o2net_register_handler(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY, | |
2239 | sizeof(struct dlm_cancel_join), | |
2240 | dlm_cancel_join_handler, | |
d74c9803 | 2241 | NULL, NULL, &dlm_join_handlers); |
ea203441 SM |
2242 | if (status) |
2243 | goto bail; | |
2244 | ||
2245 | status = o2net_register_handler(DLM_QUERY_REGION, DLM_MOD_KEY, | |
2246 | sizeof(struct dlm_query_region), | |
2247 | dlm_query_region_handler, | |
2248 | NULL, NULL, &dlm_join_handlers); | |
6714d8e8 | 2249 | |
18cfdf1b SM |
2250 | if (status) |
2251 | goto bail; | |
2252 | ||
2253 | status = o2net_register_handler(DLM_QUERY_NODEINFO, DLM_MOD_KEY, | |
2254 | sizeof(struct dlm_query_nodeinfo), | |
2255 | dlm_query_nodeinfo_handler, | |
2256 | NULL, NULL, &dlm_join_handlers); | |
6714d8e8 KH |
2257 | bail: |
2258 | if (status < 0) | |
2259 | dlm_unregister_net_handlers(); | |
2260 | ||
2261 | return status; | |
2262 | } | |
2263 | ||
2264 | /* Domain eviction callback handling. | |
2265 | * | |
2266 | * The file system requires notification of node death *before* the | |
2267 | * dlm completes it's recovery work, otherwise it may be able to | |
2268 | * acquire locks on resources requiring recovery. Since the dlm can | |
2269 | * evict a node from it's domain *before* heartbeat fires, a similar | |
2270 | * mechanism is required. */ | |
2271 | ||
2272 | /* Eviction is not expected to happen often, so a per-domain lock is | |
2273 | * not necessary. Eviction callbacks are allowed to sleep for short | |
2274 | * periods of time. */ | |
2275 | static DECLARE_RWSEM(dlm_callback_sem); | |
2276 | ||
2277 | void dlm_fire_domain_eviction_callbacks(struct dlm_ctxt *dlm, | |
2278 | int node_num) | |
2279 | { | |
6714d8e8 KH |
2280 | struct dlm_eviction_cb *cb; |
2281 | ||
2282 | down_read(&dlm_callback_sem); | |
df53cd3b | 2283 | list_for_each_entry(cb, &dlm->dlm_eviction_callbacks, ec_item) { |
6714d8e8 KH |
2284 | cb->ec_func(node_num, cb->ec_data); |
2285 | } | |
2286 | up_read(&dlm_callback_sem); | |
2287 | } | |
2288 | ||
2289 | void dlm_setup_eviction_cb(struct dlm_eviction_cb *cb, | |
2290 | dlm_eviction_func *f, | |
2291 | void *data) | |
2292 | { | |
2293 | INIT_LIST_HEAD(&cb->ec_item); | |
2294 | cb->ec_func = f; | |
2295 | cb->ec_data = data; | |
2296 | } | |
2297 | EXPORT_SYMBOL_GPL(dlm_setup_eviction_cb); | |
2298 | ||
2299 | void dlm_register_eviction_cb(struct dlm_ctxt *dlm, | |
2300 | struct dlm_eviction_cb *cb) | |
2301 | { | |
2302 | down_write(&dlm_callback_sem); | |
2303 | list_add_tail(&cb->ec_item, &dlm->dlm_eviction_callbacks); | |
2304 | up_write(&dlm_callback_sem); | |
2305 | } | |
2306 | EXPORT_SYMBOL_GPL(dlm_register_eviction_cb); | |
2307 | ||
2308 | void dlm_unregister_eviction_cb(struct dlm_eviction_cb *cb) | |
2309 | { | |
2310 | down_write(&dlm_callback_sem); | |
2311 | list_del_init(&cb->ec_item); | |
2312 | up_write(&dlm_callback_sem); | |
2313 | } | |
2314 | EXPORT_SYMBOL_GPL(dlm_unregister_eviction_cb); | |
2315 | ||
2316 | static int __init dlm_init(void) | |
2317 | { | |
2318 | int status; | |
2319 | ||
6714d8e8 | 2320 | status = dlm_init_mle_cache(); |
12eb0035 SM |
2321 | if (status) { |
2322 | mlog(ML_ERROR, "Could not create o2dlm_mle slabcache\n"); | |
724bdca9 SM |
2323 | goto error; |
2324 | } | |
2325 | ||
2326 | status = dlm_init_master_caches(); | |
2327 | if (status) { | |
2328 | mlog(ML_ERROR, "Could not create o2dlm_lockres and " | |
2329 | "o2dlm_lockname slabcaches\n"); | |
2330 | goto error; | |
2331 | } | |
2332 | ||
2333 | status = dlm_init_lock_cache(); | |
2334 | if (status) { | |
2335 | mlog(ML_ERROR, "Count not create o2dlm_lock slabcache\n"); | |
2336 | goto error; | |
12eb0035 | 2337 | } |
6714d8e8 KH |
2338 | |
2339 | status = dlm_register_net_handlers(); | |
2340 | if (status) { | |
724bdca9 SM |
2341 | mlog(ML_ERROR, "Unable to register network handlers\n"); |
2342 | goto error; | |
6714d8e8 KH |
2343 | } |
2344 | ||
e581595e | 2345 | dlm_create_debugfs_root(); |
6325b4a2 | 2346 | |
6714d8e8 | 2347 | return 0; |
724bdca9 | 2348 | error: |
6325b4a2 | 2349 | dlm_unregister_net_handlers(); |
724bdca9 SM |
2350 | dlm_destroy_lock_cache(); |
2351 | dlm_destroy_master_caches(); | |
2352 | dlm_destroy_mle_cache(); | |
2353 | return -1; | |
6714d8e8 KH |
2354 | } |
2355 | ||
2356 | static void __exit dlm_exit (void) | |
2357 | { | |
6325b4a2 | 2358 | dlm_destroy_debugfs_root(); |
6714d8e8 | 2359 | dlm_unregister_net_handlers(); |
724bdca9 SM |
2360 | dlm_destroy_lock_cache(); |
2361 | dlm_destroy_master_caches(); | |
6714d8e8 KH |
2362 | dlm_destroy_mle_cache(); |
2363 | } | |
2364 | ||
2365 | MODULE_AUTHOR("Oracle"); | |
2366 | MODULE_LICENSE("GPL"); | |
ff8fb335 | 2367 | MODULE_DESCRIPTION("OCFS2 Distributed Lock Management"); |
6714d8e8 KH |
2368 | |
2369 | module_init(dlm_init); | |
2370 | module_exit(dlm_exit); |