Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1 | /* -*- mode: c; c-basic-offset: 8; -*- |
| 2 | * vim: noexpandtab sw=8 ts=8 sts=0: |
| 3 | * |
| 4 | * dlmdomain.c |
| 5 | * |
| 6 | * defines domain join / leave apis |
| 7 | * |
| 8 | * Copyright (C) 2004 Oracle. All rights reserved. |
| 9 | * |
| 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. |
| 14 | * |
| 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. |
| 19 | * |
| 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. |
| 24 | * |
| 25 | */ |
| 26 | |
| 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 | |
| 37 | #include "cluster/heartbeat.h" |
| 38 | #include "cluster/nodemanager.h" |
| 39 | #include "cluster/tcp.h" |
| 40 | |
| 41 | #include "dlmapi.h" |
| 42 | #include "dlmcommon.h" |
| 43 | |
| 44 | #include "dlmdebug.h" |
| 45 | #include "dlmdomain.h" |
| 46 | |
| 47 | #include "dlmver.h" |
| 48 | |
| 49 | #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_DOMAIN) |
| 50 | #include "cluster/masklog.h" |
| 51 | |
| 52 | /* |
| 53 | * |
| 54 | * spinlock lock ordering: if multiple locks are needed, obey this ordering: |
| 55 | * dlm_domain_lock |
| 56 | * struct dlm_ctxt->spinlock |
| 57 | * struct dlm_lock_resource->spinlock |
| 58 | * struct dlm_ctxt->master_lock |
| 59 | * struct dlm_ctxt->ast_lock |
| 60 | * dlm_master_list_entry->spinlock |
| 61 | * dlm_lock->spinlock |
| 62 | * |
| 63 | */ |
| 64 | |
| 65 | spinlock_t dlm_domain_lock = SPIN_LOCK_UNLOCKED; |
| 66 | LIST_HEAD(dlm_domains); |
| 67 | static DECLARE_WAIT_QUEUE_HEAD(dlm_domain_events); |
| 68 | |
| 69 | #define DLM_DOMAIN_BACKOFF_MS 200 |
| 70 | |
| 71 | static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data); |
| 72 | static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data); |
| 73 | static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data); |
| 74 | static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data); |
| 75 | |
| 76 | static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm); |
| 77 | |
| 78 | void __dlm_unhash_lockres(struct dlm_lock_resource *lockres) |
| 79 | { |
Mark Fasheh | 81f2094 | 2006-02-28 17:31:22 -0800 | [diff] [blame] | 80 | hlist_del_init(&lockres->hash_node); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 81 | dlm_lockres_put(lockres); |
| 82 | } |
| 83 | |
| 84 | void __dlm_insert_lockres(struct dlm_ctxt *dlm, |
| 85 | struct dlm_lock_resource *res) |
| 86 | { |
Mark Fasheh | 81f2094 | 2006-02-28 17:31:22 -0800 | [diff] [blame] | 87 | struct hlist_head *bucket; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 88 | struct qstr *q; |
| 89 | |
| 90 | assert_spin_locked(&dlm->spinlock); |
| 91 | |
| 92 | q = &res->lockname; |
Mark Fasheh | 81f2094 | 2006-02-28 17:31:22 -0800 | [diff] [blame] | 93 | bucket = &(dlm->lockres_hash[q->hash % DLM_HASH_BUCKETS]); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 94 | |
| 95 | /* get a reference for our hashtable */ |
| 96 | dlm_lockres_get(res); |
| 97 | |
Mark Fasheh | 81f2094 | 2006-02-28 17:31:22 -0800 | [diff] [blame] | 98 | hlist_add_head(&res->hash_node, bucket); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 99 | } |
| 100 | |
| 101 | struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm, |
Mark Fasheh | a3d3329 | 2006-03-09 17:55:56 -0800 | [diff] [blame^] | 102 | const char *name, |
| 103 | unsigned int len, |
| 104 | unsigned int hash) |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 105 | { |
Mark Fasheh | 81f2094 | 2006-02-28 17:31:22 -0800 | [diff] [blame] | 106 | struct hlist_node *iter; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 107 | struct dlm_lock_resource *tmpres=NULL; |
Mark Fasheh | 81f2094 | 2006-02-28 17:31:22 -0800 | [diff] [blame] | 108 | struct hlist_head *bucket; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 109 | |
| 110 | mlog_entry("%.*s\n", len, name); |
| 111 | |
| 112 | assert_spin_locked(&dlm->spinlock); |
| 113 | |
Mark Fasheh | 81f2094 | 2006-02-28 17:31:22 -0800 | [diff] [blame] | 114 | bucket = &(dlm->lockres_hash[hash % DLM_HASH_BUCKETS]); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 115 | |
| 116 | /* check for pre-existing lock */ |
Mark Fasheh | 81f2094 | 2006-02-28 17:31:22 -0800 | [diff] [blame] | 117 | hlist_for_each(iter, bucket) { |
| 118 | tmpres = hlist_entry(iter, struct dlm_lock_resource, hash_node); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 119 | if (tmpres->lockname.len == len && |
| 120 | memcmp(tmpres->lockname.name, name, len) == 0) { |
| 121 | dlm_lockres_get(tmpres); |
| 122 | break; |
| 123 | } |
| 124 | |
| 125 | tmpres = NULL; |
| 126 | } |
| 127 | return tmpres; |
| 128 | } |
| 129 | |
| 130 | struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm, |
| 131 | const char *name, |
| 132 | unsigned int len) |
| 133 | { |
| 134 | struct dlm_lock_resource *res; |
Mark Fasheh | a3d3329 | 2006-03-09 17:55:56 -0800 | [diff] [blame^] | 135 | unsigned int hash = dlm_lockid_hash(name, len); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 136 | |
| 137 | spin_lock(&dlm->spinlock); |
Mark Fasheh | a3d3329 | 2006-03-09 17:55:56 -0800 | [diff] [blame^] | 138 | res = __dlm_lookup_lockres(dlm, name, len, hash); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 139 | spin_unlock(&dlm->spinlock); |
| 140 | return res; |
| 141 | } |
| 142 | |
| 143 | static struct dlm_ctxt * __dlm_lookup_domain_full(const char *domain, int len) |
| 144 | { |
| 145 | struct dlm_ctxt *tmp = NULL; |
| 146 | struct list_head *iter; |
| 147 | |
| 148 | assert_spin_locked(&dlm_domain_lock); |
| 149 | |
| 150 | /* tmp->name here is always NULL terminated, |
| 151 | * but domain may not be! */ |
| 152 | list_for_each(iter, &dlm_domains) { |
| 153 | tmp = list_entry (iter, struct dlm_ctxt, list); |
| 154 | if (strlen(tmp->name) == len && |
| 155 | memcmp(tmp->name, domain, len)==0) |
| 156 | break; |
| 157 | tmp = NULL; |
| 158 | } |
| 159 | |
| 160 | return tmp; |
| 161 | } |
| 162 | |
| 163 | /* For null terminated domain strings ONLY */ |
| 164 | static struct dlm_ctxt * __dlm_lookup_domain(const char *domain) |
| 165 | { |
| 166 | assert_spin_locked(&dlm_domain_lock); |
| 167 | |
| 168 | return __dlm_lookup_domain_full(domain, strlen(domain)); |
| 169 | } |
| 170 | |
| 171 | |
| 172 | /* returns true on one of two conditions: |
| 173 | * 1) the domain does not exist |
| 174 | * 2) the domain exists and it's state is "joined" */ |
| 175 | static int dlm_wait_on_domain_helper(const char *domain) |
| 176 | { |
| 177 | int ret = 0; |
| 178 | struct dlm_ctxt *tmp = NULL; |
| 179 | |
| 180 | spin_lock(&dlm_domain_lock); |
| 181 | |
| 182 | tmp = __dlm_lookup_domain(domain); |
| 183 | if (!tmp) |
| 184 | ret = 1; |
| 185 | else if (tmp->dlm_state == DLM_CTXT_JOINED) |
| 186 | ret = 1; |
| 187 | |
| 188 | spin_unlock(&dlm_domain_lock); |
| 189 | return ret; |
| 190 | } |
| 191 | |
| 192 | static void dlm_free_ctxt_mem(struct dlm_ctxt *dlm) |
| 193 | { |
Mark Fasheh | 81f2094 | 2006-02-28 17:31:22 -0800 | [diff] [blame] | 194 | if (dlm->lockres_hash) |
| 195 | free_page((unsigned long) dlm->lockres_hash); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 196 | |
| 197 | if (dlm->name) |
| 198 | kfree(dlm->name); |
| 199 | |
| 200 | kfree(dlm); |
| 201 | } |
| 202 | |
| 203 | /* A little strange - this function will be called while holding |
| 204 | * dlm_domain_lock and is expected to be holding it on the way out. We |
| 205 | * will however drop and reacquire it multiple times */ |
| 206 | static void dlm_ctxt_release(struct kref *kref) |
| 207 | { |
| 208 | struct dlm_ctxt *dlm; |
| 209 | |
| 210 | dlm = container_of(kref, struct dlm_ctxt, dlm_refs); |
| 211 | |
| 212 | BUG_ON(dlm->num_joins); |
| 213 | BUG_ON(dlm->dlm_state == DLM_CTXT_JOINED); |
| 214 | |
| 215 | /* we may still be in the list if we hit an error during join. */ |
| 216 | list_del_init(&dlm->list); |
| 217 | |
| 218 | spin_unlock(&dlm_domain_lock); |
| 219 | |
| 220 | mlog(0, "freeing memory from domain %s\n", dlm->name); |
| 221 | |
| 222 | wake_up(&dlm_domain_events); |
| 223 | |
| 224 | dlm_free_ctxt_mem(dlm); |
| 225 | |
| 226 | spin_lock(&dlm_domain_lock); |
| 227 | } |
| 228 | |
| 229 | void dlm_put(struct dlm_ctxt *dlm) |
| 230 | { |
| 231 | spin_lock(&dlm_domain_lock); |
| 232 | kref_put(&dlm->dlm_refs, dlm_ctxt_release); |
| 233 | spin_unlock(&dlm_domain_lock); |
| 234 | } |
| 235 | |
| 236 | static void __dlm_get(struct dlm_ctxt *dlm) |
| 237 | { |
| 238 | kref_get(&dlm->dlm_refs); |
| 239 | } |
| 240 | |
| 241 | /* given a questionable reference to a dlm object, gets a reference if |
| 242 | * it can find it in the list, otherwise returns NULL in which case |
| 243 | * you shouldn't trust your pointer. */ |
| 244 | struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm) |
| 245 | { |
| 246 | struct list_head *iter; |
| 247 | struct dlm_ctxt *target = NULL; |
| 248 | |
| 249 | spin_lock(&dlm_domain_lock); |
| 250 | |
| 251 | list_for_each(iter, &dlm_domains) { |
| 252 | target = list_entry (iter, struct dlm_ctxt, list); |
| 253 | |
| 254 | if (target == dlm) { |
| 255 | __dlm_get(target); |
| 256 | break; |
| 257 | } |
| 258 | |
| 259 | target = NULL; |
| 260 | } |
| 261 | |
| 262 | spin_unlock(&dlm_domain_lock); |
| 263 | |
| 264 | return target; |
| 265 | } |
| 266 | |
| 267 | int dlm_domain_fully_joined(struct dlm_ctxt *dlm) |
| 268 | { |
| 269 | int ret; |
| 270 | |
| 271 | spin_lock(&dlm_domain_lock); |
| 272 | ret = (dlm->dlm_state == DLM_CTXT_JOINED) || |
| 273 | (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN); |
| 274 | spin_unlock(&dlm_domain_lock); |
| 275 | |
| 276 | return ret; |
| 277 | } |
| 278 | |
| 279 | static void dlm_complete_dlm_shutdown(struct dlm_ctxt *dlm) |
| 280 | { |
| 281 | dlm_unregister_domain_handlers(dlm); |
| 282 | dlm_complete_thread(dlm); |
| 283 | dlm_complete_recovery_thread(dlm); |
| 284 | |
| 285 | /* We've left the domain. Now we can take ourselves out of the |
| 286 | * list and allow the kref stuff to help us free the |
| 287 | * memory. */ |
| 288 | spin_lock(&dlm_domain_lock); |
| 289 | list_del_init(&dlm->list); |
| 290 | spin_unlock(&dlm_domain_lock); |
| 291 | |
| 292 | /* Wake up anyone waiting for us to remove this domain */ |
| 293 | wake_up(&dlm_domain_events); |
| 294 | } |
| 295 | |
| 296 | static void dlm_migrate_all_locks(struct dlm_ctxt *dlm) |
| 297 | { |
| 298 | int i; |
| 299 | struct dlm_lock_resource *res; |
| 300 | |
| 301 | mlog(0, "Migrating locks from domain %s\n", dlm->name); |
| 302 | restart: |
| 303 | spin_lock(&dlm->spinlock); |
Mark Fasheh | 81f2094 | 2006-02-28 17:31:22 -0800 | [diff] [blame] | 304 | for (i = 0; i < DLM_HASH_BUCKETS; i++) { |
| 305 | while (!hlist_empty(&dlm->lockres_hash[i])) { |
| 306 | res = hlist_entry(dlm->lockres_hash[i].first, |
| 307 | struct dlm_lock_resource, hash_node); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 308 | /* need reference when manually grabbing lockres */ |
| 309 | dlm_lockres_get(res); |
| 310 | /* this should unhash the lockres |
| 311 | * and exit with dlm->spinlock */ |
| 312 | mlog(0, "purging res=%p\n", res); |
| 313 | if (dlm_lockres_is_dirty(dlm, res)) { |
| 314 | /* HACK! this should absolutely go. |
| 315 | * need to figure out why some empty |
| 316 | * lockreses are still marked dirty */ |
| 317 | mlog(ML_ERROR, "lockres %.*s dirty!\n", |
| 318 | res->lockname.len, res->lockname.name); |
| 319 | |
| 320 | spin_unlock(&dlm->spinlock); |
| 321 | dlm_kick_thread(dlm, res); |
| 322 | wait_event(dlm->ast_wq, !dlm_lockres_is_dirty(dlm, res)); |
| 323 | dlm_lockres_put(res); |
| 324 | goto restart; |
| 325 | } |
| 326 | dlm_purge_lockres(dlm, res); |
| 327 | dlm_lockres_put(res); |
| 328 | } |
| 329 | } |
| 330 | spin_unlock(&dlm->spinlock); |
| 331 | |
| 332 | mlog(0, "DONE Migrating locks from domain %s\n", dlm->name); |
| 333 | } |
| 334 | |
| 335 | static int dlm_no_joining_node(struct dlm_ctxt *dlm) |
| 336 | { |
| 337 | int ret; |
| 338 | |
| 339 | spin_lock(&dlm->spinlock); |
| 340 | ret = dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN; |
| 341 | spin_unlock(&dlm->spinlock); |
| 342 | |
| 343 | return ret; |
| 344 | } |
| 345 | |
| 346 | static void dlm_mark_domain_leaving(struct dlm_ctxt *dlm) |
| 347 | { |
| 348 | /* Yikes, a double spinlock! I need domain_lock for the dlm |
| 349 | * state and the dlm spinlock for join state... Sorry! */ |
| 350 | again: |
| 351 | spin_lock(&dlm_domain_lock); |
| 352 | spin_lock(&dlm->spinlock); |
| 353 | |
| 354 | if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) { |
| 355 | mlog(0, "Node %d is joining, we wait on it.\n", |
| 356 | dlm->joining_node); |
| 357 | spin_unlock(&dlm->spinlock); |
| 358 | spin_unlock(&dlm_domain_lock); |
| 359 | |
| 360 | wait_event(dlm->dlm_join_events, dlm_no_joining_node(dlm)); |
| 361 | goto again; |
| 362 | } |
| 363 | |
| 364 | dlm->dlm_state = DLM_CTXT_LEAVING; |
| 365 | spin_unlock(&dlm->spinlock); |
| 366 | spin_unlock(&dlm_domain_lock); |
| 367 | } |
| 368 | |
| 369 | static void __dlm_print_nodes(struct dlm_ctxt *dlm) |
| 370 | { |
| 371 | int node = -1; |
| 372 | |
| 373 | assert_spin_locked(&dlm->spinlock); |
| 374 | |
| 375 | mlog(ML_NOTICE, "Nodes in my domain (\"%s\"):\n", dlm->name); |
| 376 | |
| 377 | while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES, |
| 378 | node + 1)) < O2NM_MAX_NODES) { |
| 379 | mlog(ML_NOTICE, " node %d\n", node); |
| 380 | } |
| 381 | } |
| 382 | |
| 383 | static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data) |
| 384 | { |
| 385 | struct dlm_ctxt *dlm = data; |
| 386 | unsigned int node; |
| 387 | struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf; |
| 388 | |
| 389 | mlog_entry("%p %u %p", msg, len, data); |
| 390 | |
| 391 | if (!dlm_grab(dlm)) |
| 392 | return 0; |
| 393 | |
| 394 | node = exit_msg->node_idx; |
| 395 | |
| 396 | mlog(0, "Node %u leaves domain %s\n", node, dlm->name); |
| 397 | |
| 398 | spin_lock(&dlm->spinlock); |
| 399 | clear_bit(node, dlm->domain_map); |
| 400 | __dlm_print_nodes(dlm); |
| 401 | |
| 402 | /* notify anything attached to the heartbeat events */ |
| 403 | dlm_hb_event_notify_attached(dlm, node, 0); |
| 404 | |
| 405 | spin_unlock(&dlm->spinlock); |
| 406 | |
| 407 | dlm_put(dlm); |
| 408 | |
| 409 | return 0; |
| 410 | } |
| 411 | |
| 412 | static int dlm_send_one_domain_exit(struct dlm_ctxt *dlm, |
| 413 | unsigned int node) |
| 414 | { |
| 415 | int status; |
| 416 | struct dlm_exit_domain leave_msg; |
| 417 | |
| 418 | mlog(0, "Asking node %u if we can leave the domain %s me = %u\n", |
| 419 | node, dlm->name, dlm->node_num); |
| 420 | |
| 421 | memset(&leave_msg, 0, sizeof(leave_msg)); |
| 422 | leave_msg.node_idx = dlm->node_num; |
| 423 | |
| 424 | status = o2net_send_message(DLM_EXIT_DOMAIN_MSG, dlm->key, |
| 425 | &leave_msg, sizeof(leave_msg), node, |
| 426 | NULL); |
| 427 | |
| 428 | mlog(0, "status return %d from o2net_send_message\n", status); |
| 429 | |
| 430 | return status; |
| 431 | } |
| 432 | |
| 433 | |
| 434 | static void dlm_leave_domain(struct dlm_ctxt *dlm) |
| 435 | { |
| 436 | int node, clear_node, status; |
| 437 | |
| 438 | /* At this point we've migrated away all our locks and won't |
| 439 | * accept mastership of new ones. The dlm is responsible for |
| 440 | * almost nothing now. We make sure not to confuse any joining |
| 441 | * nodes and then commence shutdown procedure. */ |
| 442 | |
| 443 | spin_lock(&dlm->spinlock); |
| 444 | /* Clear ourselves from the domain map */ |
| 445 | clear_bit(dlm->node_num, dlm->domain_map); |
| 446 | while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES, |
| 447 | 0)) < O2NM_MAX_NODES) { |
| 448 | /* Drop the dlm spinlock. This is safe wrt the domain_map. |
| 449 | * -nodes cannot be added now as the |
| 450 | * query_join_handlers knows to respond with OK_NO_MAP |
| 451 | * -we catch the right network errors if a node is |
| 452 | * removed from the map while we're sending him the |
| 453 | * exit message. */ |
| 454 | spin_unlock(&dlm->spinlock); |
| 455 | |
| 456 | clear_node = 1; |
| 457 | |
| 458 | status = dlm_send_one_domain_exit(dlm, node); |
| 459 | if (status < 0 && |
| 460 | status != -ENOPROTOOPT && |
| 461 | status != -ENOTCONN) { |
| 462 | mlog(ML_NOTICE, "Error %d sending domain exit message " |
| 463 | "to node %d\n", status, node); |
| 464 | |
| 465 | /* Not sure what to do here but lets sleep for |
| 466 | * a bit in case this was a transient |
| 467 | * error... */ |
| 468 | msleep(DLM_DOMAIN_BACKOFF_MS); |
| 469 | clear_node = 0; |
| 470 | } |
| 471 | |
| 472 | spin_lock(&dlm->spinlock); |
| 473 | /* If we're not clearing the node bit then we intend |
| 474 | * to loop back around to try again. */ |
| 475 | if (clear_node) |
| 476 | clear_bit(node, dlm->domain_map); |
| 477 | } |
| 478 | spin_unlock(&dlm->spinlock); |
| 479 | } |
| 480 | |
| 481 | int dlm_joined(struct dlm_ctxt *dlm) |
| 482 | { |
| 483 | int ret = 0; |
| 484 | |
| 485 | spin_lock(&dlm_domain_lock); |
| 486 | |
| 487 | if (dlm->dlm_state == DLM_CTXT_JOINED) |
| 488 | ret = 1; |
| 489 | |
| 490 | spin_unlock(&dlm_domain_lock); |
| 491 | |
| 492 | return ret; |
| 493 | } |
| 494 | |
| 495 | int dlm_shutting_down(struct dlm_ctxt *dlm) |
| 496 | { |
| 497 | int ret = 0; |
| 498 | |
| 499 | spin_lock(&dlm_domain_lock); |
| 500 | |
| 501 | if (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN) |
| 502 | ret = 1; |
| 503 | |
| 504 | spin_unlock(&dlm_domain_lock); |
| 505 | |
| 506 | return ret; |
| 507 | } |
| 508 | |
| 509 | void dlm_unregister_domain(struct dlm_ctxt *dlm) |
| 510 | { |
| 511 | int leave = 0; |
| 512 | |
| 513 | spin_lock(&dlm_domain_lock); |
| 514 | BUG_ON(dlm->dlm_state != DLM_CTXT_JOINED); |
| 515 | BUG_ON(!dlm->num_joins); |
| 516 | |
| 517 | dlm->num_joins--; |
| 518 | if (!dlm->num_joins) { |
| 519 | /* We mark it "in shutdown" now so new register |
| 520 | * requests wait until we've completely left the |
| 521 | * domain. Don't use DLM_CTXT_LEAVING yet as we still |
| 522 | * want new domain joins to communicate with us at |
| 523 | * least until we've completed migration of our |
| 524 | * resources. */ |
| 525 | dlm->dlm_state = DLM_CTXT_IN_SHUTDOWN; |
| 526 | leave = 1; |
| 527 | } |
| 528 | spin_unlock(&dlm_domain_lock); |
| 529 | |
| 530 | if (leave) { |
| 531 | mlog(0, "shutting down domain %s\n", dlm->name); |
| 532 | |
| 533 | /* We changed dlm state, notify the thread */ |
| 534 | dlm_kick_thread(dlm, NULL); |
| 535 | |
| 536 | dlm_migrate_all_locks(dlm); |
| 537 | dlm_mark_domain_leaving(dlm); |
| 538 | dlm_leave_domain(dlm); |
| 539 | dlm_complete_dlm_shutdown(dlm); |
| 540 | } |
| 541 | dlm_put(dlm); |
| 542 | } |
| 543 | EXPORT_SYMBOL_GPL(dlm_unregister_domain); |
| 544 | |
| 545 | static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data) |
| 546 | { |
| 547 | struct dlm_query_join_request *query; |
| 548 | enum dlm_query_join_response response; |
| 549 | struct dlm_ctxt *dlm = NULL; |
| 550 | |
| 551 | query = (struct dlm_query_join_request *) msg->buf; |
| 552 | |
| 553 | mlog(0, "node %u wants to join domain %s\n", query->node_idx, |
| 554 | query->domain); |
| 555 | |
| 556 | /* |
| 557 | * If heartbeat doesn't consider the node live, tell it |
| 558 | * to back off and try again. This gives heartbeat a chance |
| 559 | * to catch up. |
| 560 | */ |
| 561 | if (!o2hb_check_node_heartbeating(query->node_idx)) { |
| 562 | mlog(0, "node %u is not in our live map yet\n", |
| 563 | query->node_idx); |
| 564 | |
| 565 | response = JOIN_DISALLOW; |
| 566 | goto respond; |
| 567 | } |
| 568 | |
| 569 | response = JOIN_OK_NO_MAP; |
| 570 | |
| 571 | spin_lock(&dlm_domain_lock); |
| 572 | dlm = __dlm_lookup_domain_full(query->domain, query->name_len); |
| 573 | /* Once the dlm ctxt is marked as leaving then we don't want |
Kurt Hackel | e2faea4 | 2006-01-12 14:24:55 -0800 | [diff] [blame] | 574 | * to be put in someone's domain map. |
| 575 | * Also, explicitly disallow joining at certain troublesome |
| 576 | * times (ie. during recovery). */ |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 577 | if (dlm && dlm->dlm_state != DLM_CTXT_LEAVING) { |
Kurt Hackel | e2faea4 | 2006-01-12 14:24:55 -0800 | [diff] [blame] | 578 | int bit = query->node_idx; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 579 | spin_lock(&dlm->spinlock); |
| 580 | |
| 581 | if (dlm->dlm_state == DLM_CTXT_NEW && |
| 582 | dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN) { |
| 583 | /*If this is a brand new context and we |
| 584 | * haven't started our join process yet, then |
| 585 | * the other node won the race. */ |
| 586 | response = JOIN_OK_NO_MAP; |
| 587 | } else if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) { |
| 588 | /* Disallow parallel joins. */ |
| 589 | response = JOIN_DISALLOW; |
Kurt Hackel | e2faea4 | 2006-01-12 14:24:55 -0800 | [diff] [blame] | 590 | } else if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) { |
| 591 | mlog(ML_NOTICE, "node %u trying to join, but recovery " |
| 592 | "is ongoing.\n", bit); |
| 593 | response = JOIN_DISALLOW; |
| 594 | } else if (test_bit(bit, dlm->recovery_map)) { |
| 595 | mlog(ML_NOTICE, "node %u trying to join, but it " |
| 596 | "still needs recovery.\n", bit); |
| 597 | response = JOIN_DISALLOW; |
| 598 | } else if (test_bit(bit, dlm->domain_map)) { |
| 599 | mlog(ML_NOTICE, "node %u trying to join, but it " |
| 600 | "is still in the domain! needs recovery?\n", |
| 601 | bit); |
| 602 | response = JOIN_DISALLOW; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 603 | } else { |
| 604 | /* Alright we're fully a part of this domain |
| 605 | * so we keep some state as to who's joining |
| 606 | * and indicate to him that needs to be fixed |
| 607 | * up. */ |
| 608 | response = JOIN_OK; |
| 609 | __dlm_set_joining_node(dlm, query->node_idx); |
| 610 | } |
| 611 | |
| 612 | spin_unlock(&dlm->spinlock); |
| 613 | } |
| 614 | spin_unlock(&dlm_domain_lock); |
| 615 | |
| 616 | respond: |
| 617 | mlog(0, "We respond with %u\n", response); |
| 618 | |
| 619 | return response; |
| 620 | } |
| 621 | |
| 622 | static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data) |
| 623 | { |
| 624 | struct dlm_assert_joined *assert; |
| 625 | struct dlm_ctxt *dlm = NULL; |
| 626 | |
| 627 | assert = (struct dlm_assert_joined *) msg->buf; |
| 628 | |
| 629 | mlog(0, "node %u asserts join on domain %s\n", assert->node_idx, |
| 630 | assert->domain); |
| 631 | |
| 632 | spin_lock(&dlm_domain_lock); |
| 633 | dlm = __dlm_lookup_domain_full(assert->domain, assert->name_len); |
| 634 | /* XXX should we consider no dlm ctxt an error? */ |
| 635 | if (dlm) { |
| 636 | spin_lock(&dlm->spinlock); |
| 637 | |
| 638 | /* Alright, this node has officially joined our |
| 639 | * domain. Set him in the map and clean up our |
| 640 | * leftover join state. */ |
| 641 | BUG_ON(dlm->joining_node != assert->node_idx); |
| 642 | set_bit(assert->node_idx, dlm->domain_map); |
| 643 | __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN); |
| 644 | |
| 645 | __dlm_print_nodes(dlm); |
| 646 | |
| 647 | /* notify anything attached to the heartbeat events */ |
| 648 | dlm_hb_event_notify_attached(dlm, assert->node_idx, 1); |
| 649 | |
| 650 | spin_unlock(&dlm->spinlock); |
| 651 | } |
| 652 | spin_unlock(&dlm_domain_lock); |
| 653 | |
| 654 | return 0; |
| 655 | } |
| 656 | |
| 657 | static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data) |
| 658 | { |
| 659 | struct dlm_cancel_join *cancel; |
| 660 | struct dlm_ctxt *dlm = NULL; |
| 661 | |
| 662 | cancel = (struct dlm_cancel_join *) msg->buf; |
| 663 | |
| 664 | mlog(0, "node %u cancels join on domain %s\n", cancel->node_idx, |
| 665 | cancel->domain); |
| 666 | |
| 667 | spin_lock(&dlm_domain_lock); |
| 668 | dlm = __dlm_lookup_domain_full(cancel->domain, cancel->name_len); |
| 669 | |
| 670 | if (dlm) { |
| 671 | spin_lock(&dlm->spinlock); |
| 672 | |
| 673 | /* Yikes, this guy wants to cancel his join. No |
| 674 | * problem, we simply cleanup our join state. */ |
| 675 | BUG_ON(dlm->joining_node != cancel->node_idx); |
| 676 | __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN); |
| 677 | |
| 678 | spin_unlock(&dlm->spinlock); |
| 679 | } |
| 680 | spin_unlock(&dlm_domain_lock); |
| 681 | |
| 682 | return 0; |
| 683 | } |
| 684 | |
| 685 | static int dlm_send_one_join_cancel(struct dlm_ctxt *dlm, |
| 686 | unsigned int node) |
| 687 | { |
| 688 | int status; |
| 689 | struct dlm_cancel_join cancel_msg; |
| 690 | |
| 691 | memset(&cancel_msg, 0, sizeof(cancel_msg)); |
| 692 | cancel_msg.node_idx = dlm->node_num; |
| 693 | cancel_msg.name_len = strlen(dlm->name); |
| 694 | memcpy(cancel_msg.domain, dlm->name, cancel_msg.name_len); |
| 695 | |
| 696 | status = o2net_send_message(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY, |
| 697 | &cancel_msg, sizeof(cancel_msg), node, |
| 698 | NULL); |
| 699 | if (status < 0) { |
| 700 | mlog_errno(status); |
| 701 | goto bail; |
| 702 | } |
| 703 | |
| 704 | bail: |
| 705 | return status; |
| 706 | } |
| 707 | |
| 708 | /* map_size should be in bytes. */ |
| 709 | static int dlm_send_join_cancels(struct dlm_ctxt *dlm, |
| 710 | unsigned long *node_map, |
| 711 | unsigned int map_size) |
| 712 | { |
| 713 | int status, tmpstat; |
| 714 | unsigned int node; |
| 715 | |
| 716 | if (map_size != (BITS_TO_LONGS(O2NM_MAX_NODES) * |
| 717 | sizeof(unsigned long))) { |
| 718 | mlog(ML_ERROR, |
| 719 | "map_size %u != BITS_TO_LONGS(O2NM_MAX_NODES) %u\n", |
| 720 | map_size, BITS_TO_LONGS(O2NM_MAX_NODES)); |
| 721 | return -EINVAL; |
| 722 | } |
| 723 | |
| 724 | status = 0; |
| 725 | node = -1; |
| 726 | while ((node = find_next_bit(node_map, O2NM_MAX_NODES, |
| 727 | node + 1)) < O2NM_MAX_NODES) { |
| 728 | if (node == dlm->node_num) |
| 729 | continue; |
| 730 | |
| 731 | tmpstat = dlm_send_one_join_cancel(dlm, node); |
| 732 | if (tmpstat) { |
| 733 | mlog(ML_ERROR, "Error return %d cancelling join on " |
| 734 | "node %d\n", tmpstat, node); |
| 735 | if (!status) |
| 736 | status = tmpstat; |
| 737 | } |
| 738 | } |
| 739 | |
| 740 | if (status) |
| 741 | mlog_errno(status); |
| 742 | return status; |
| 743 | } |
| 744 | |
| 745 | static int dlm_request_join(struct dlm_ctxt *dlm, |
| 746 | int node, |
| 747 | enum dlm_query_join_response *response) |
| 748 | { |
| 749 | int status, retval; |
| 750 | struct dlm_query_join_request join_msg; |
| 751 | |
| 752 | mlog(0, "querying node %d\n", node); |
| 753 | |
| 754 | memset(&join_msg, 0, sizeof(join_msg)); |
| 755 | join_msg.node_idx = dlm->node_num; |
| 756 | join_msg.name_len = strlen(dlm->name); |
| 757 | memcpy(join_msg.domain, dlm->name, join_msg.name_len); |
| 758 | |
| 759 | status = o2net_send_message(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, &join_msg, |
| 760 | sizeof(join_msg), node, &retval); |
| 761 | if (status < 0 && status != -ENOPROTOOPT) { |
| 762 | mlog_errno(status); |
| 763 | goto bail; |
| 764 | } |
| 765 | |
| 766 | /* -ENOPROTOOPT from the net code means the other side isn't |
| 767 | listening for our message type -- that's fine, it means |
| 768 | his dlm isn't up, so we can consider him a 'yes' but not |
| 769 | joined into the domain. */ |
| 770 | if (status == -ENOPROTOOPT) { |
| 771 | status = 0; |
| 772 | *response = JOIN_OK_NO_MAP; |
| 773 | } else if (retval == JOIN_DISALLOW || |
| 774 | retval == JOIN_OK || |
| 775 | retval == JOIN_OK_NO_MAP) { |
| 776 | *response = retval; |
| 777 | } else { |
| 778 | status = -EINVAL; |
| 779 | mlog(ML_ERROR, "invalid response %d from node %u\n", retval, |
| 780 | node); |
| 781 | } |
| 782 | |
| 783 | mlog(0, "status %d, node %d response is %d\n", status, node, |
| 784 | *response); |
| 785 | |
| 786 | bail: |
| 787 | return status; |
| 788 | } |
| 789 | |
| 790 | static int dlm_send_one_join_assert(struct dlm_ctxt *dlm, |
| 791 | unsigned int node) |
| 792 | { |
| 793 | int status; |
| 794 | struct dlm_assert_joined assert_msg; |
| 795 | |
| 796 | mlog(0, "Sending join assert to node %u\n", node); |
| 797 | |
| 798 | memset(&assert_msg, 0, sizeof(assert_msg)); |
| 799 | assert_msg.node_idx = dlm->node_num; |
| 800 | assert_msg.name_len = strlen(dlm->name); |
| 801 | memcpy(assert_msg.domain, dlm->name, assert_msg.name_len); |
| 802 | |
| 803 | status = o2net_send_message(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY, |
| 804 | &assert_msg, sizeof(assert_msg), node, |
| 805 | NULL); |
| 806 | if (status < 0) |
| 807 | mlog_errno(status); |
| 808 | |
| 809 | return status; |
| 810 | } |
| 811 | |
| 812 | static void dlm_send_join_asserts(struct dlm_ctxt *dlm, |
| 813 | unsigned long *node_map) |
| 814 | { |
| 815 | int status, node, live; |
| 816 | |
| 817 | status = 0; |
| 818 | node = -1; |
| 819 | while ((node = find_next_bit(node_map, O2NM_MAX_NODES, |
| 820 | node + 1)) < O2NM_MAX_NODES) { |
| 821 | if (node == dlm->node_num) |
| 822 | continue; |
| 823 | |
| 824 | do { |
| 825 | /* It is very important that this message be |
| 826 | * received so we spin until either the node |
| 827 | * has died or it gets the message. */ |
| 828 | status = dlm_send_one_join_assert(dlm, node); |
| 829 | |
| 830 | spin_lock(&dlm->spinlock); |
| 831 | live = test_bit(node, dlm->live_nodes_map); |
| 832 | spin_unlock(&dlm->spinlock); |
| 833 | |
| 834 | if (status) { |
| 835 | mlog(ML_ERROR, "Error return %d asserting " |
| 836 | "join on node %d\n", status, node); |
| 837 | |
| 838 | /* give us some time between errors... */ |
| 839 | if (live) |
| 840 | msleep(DLM_DOMAIN_BACKOFF_MS); |
| 841 | } |
| 842 | } while (status && live); |
| 843 | } |
| 844 | } |
| 845 | |
| 846 | struct domain_join_ctxt { |
| 847 | unsigned long live_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; |
| 848 | unsigned long yes_resp_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; |
| 849 | }; |
| 850 | |
| 851 | static int dlm_should_restart_join(struct dlm_ctxt *dlm, |
| 852 | struct domain_join_ctxt *ctxt, |
| 853 | enum dlm_query_join_response response) |
| 854 | { |
| 855 | int ret; |
| 856 | |
| 857 | if (response == JOIN_DISALLOW) { |
| 858 | mlog(0, "Latest response of disallow -- should restart\n"); |
| 859 | return 1; |
| 860 | } |
| 861 | |
| 862 | spin_lock(&dlm->spinlock); |
| 863 | /* For now, we restart the process if the node maps have |
| 864 | * changed at all */ |
| 865 | ret = memcmp(ctxt->live_map, dlm->live_nodes_map, |
| 866 | sizeof(dlm->live_nodes_map)); |
| 867 | spin_unlock(&dlm->spinlock); |
| 868 | |
| 869 | if (ret) |
| 870 | mlog(0, "Node maps changed -- should restart\n"); |
| 871 | |
| 872 | return ret; |
| 873 | } |
| 874 | |
| 875 | static int dlm_try_to_join_domain(struct dlm_ctxt *dlm) |
| 876 | { |
| 877 | int status = 0, tmpstat, node; |
| 878 | struct domain_join_ctxt *ctxt; |
| 879 | enum dlm_query_join_response response; |
| 880 | |
| 881 | mlog_entry("%p", dlm); |
| 882 | |
| 883 | ctxt = kcalloc(1, sizeof(*ctxt), GFP_KERNEL); |
| 884 | if (!ctxt) { |
| 885 | status = -ENOMEM; |
| 886 | mlog_errno(status); |
| 887 | goto bail; |
| 888 | } |
| 889 | |
| 890 | /* group sem locking should work for us here -- we're already |
| 891 | * registered for heartbeat events so filling this should be |
| 892 | * atomic wrt getting those handlers called. */ |
| 893 | o2hb_fill_node_map(dlm->live_nodes_map, sizeof(dlm->live_nodes_map)); |
| 894 | |
| 895 | spin_lock(&dlm->spinlock); |
| 896 | memcpy(ctxt->live_map, dlm->live_nodes_map, sizeof(ctxt->live_map)); |
| 897 | |
| 898 | __dlm_set_joining_node(dlm, dlm->node_num); |
| 899 | |
| 900 | spin_unlock(&dlm->spinlock); |
| 901 | |
| 902 | node = -1; |
| 903 | while ((node = find_next_bit(ctxt->live_map, O2NM_MAX_NODES, |
| 904 | node + 1)) < O2NM_MAX_NODES) { |
| 905 | if (node == dlm->node_num) |
| 906 | continue; |
| 907 | |
| 908 | status = dlm_request_join(dlm, node, &response); |
| 909 | if (status < 0) { |
| 910 | mlog_errno(status); |
| 911 | goto bail; |
| 912 | } |
| 913 | |
| 914 | /* Ok, either we got a response or the node doesn't have a |
| 915 | * dlm up. */ |
| 916 | if (response == JOIN_OK) |
| 917 | set_bit(node, ctxt->yes_resp_map); |
| 918 | |
| 919 | if (dlm_should_restart_join(dlm, ctxt, response)) { |
| 920 | status = -EAGAIN; |
| 921 | goto bail; |
| 922 | } |
| 923 | } |
| 924 | |
| 925 | mlog(0, "Yay, done querying nodes!\n"); |
| 926 | |
| 927 | /* Yay, everyone agree's we can join the domain. My domain is |
| 928 | * comprised of all nodes who were put in the |
| 929 | * yes_resp_map. Copy that into our domain map and send a join |
| 930 | * assert message to clean up everyone elses state. */ |
| 931 | spin_lock(&dlm->spinlock); |
| 932 | memcpy(dlm->domain_map, ctxt->yes_resp_map, |
| 933 | sizeof(ctxt->yes_resp_map)); |
| 934 | set_bit(dlm->node_num, dlm->domain_map); |
| 935 | spin_unlock(&dlm->spinlock); |
| 936 | |
| 937 | dlm_send_join_asserts(dlm, ctxt->yes_resp_map); |
| 938 | |
| 939 | /* Joined state *must* be set before the joining node |
| 940 | * information, otherwise the query_join handler may read no |
| 941 | * current joiner but a state of NEW and tell joining nodes |
| 942 | * we're not in the domain. */ |
| 943 | spin_lock(&dlm_domain_lock); |
| 944 | dlm->dlm_state = DLM_CTXT_JOINED; |
| 945 | dlm->num_joins++; |
| 946 | spin_unlock(&dlm_domain_lock); |
| 947 | |
| 948 | bail: |
| 949 | spin_lock(&dlm->spinlock); |
| 950 | __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN); |
| 951 | if (!status) |
| 952 | __dlm_print_nodes(dlm); |
| 953 | spin_unlock(&dlm->spinlock); |
| 954 | |
| 955 | if (ctxt) { |
| 956 | /* Do we need to send a cancel message to any nodes? */ |
| 957 | if (status < 0) { |
| 958 | tmpstat = dlm_send_join_cancels(dlm, |
| 959 | ctxt->yes_resp_map, |
| 960 | sizeof(ctxt->yes_resp_map)); |
| 961 | if (tmpstat < 0) |
| 962 | mlog_errno(tmpstat); |
| 963 | } |
| 964 | kfree(ctxt); |
| 965 | } |
| 966 | |
| 967 | mlog(0, "returning %d\n", status); |
| 968 | return status; |
| 969 | } |
| 970 | |
| 971 | static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm) |
| 972 | { |
| 973 | o2hb_unregister_callback(&dlm->dlm_hb_up); |
| 974 | o2hb_unregister_callback(&dlm->dlm_hb_down); |
| 975 | o2net_unregister_handler_list(&dlm->dlm_domain_handlers); |
| 976 | } |
| 977 | |
| 978 | static int dlm_register_domain_handlers(struct dlm_ctxt *dlm) |
| 979 | { |
| 980 | int status; |
| 981 | |
| 982 | mlog(0, "registering handlers.\n"); |
| 983 | |
| 984 | o2hb_setup_callback(&dlm->dlm_hb_down, O2HB_NODE_DOWN_CB, |
| 985 | dlm_hb_node_down_cb, dlm, DLM_HB_NODE_DOWN_PRI); |
| 986 | status = o2hb_register_callback(&dlm->dlm_hb_down); |
| 987 | if (status) |
| 988 | goto bail; |
| 989 | |
| 990 | o2hb_setup_callback(&dlm->dlm_hb_up, O2HB_NODE_UP_CB, |
| 991 | dlm_hb_node_up_cb, dlm, DLM_HB_NODE_UP_PRI); |
| 992 | status = o2hb_register_callback(&dlm->dlm_hb_up); |
| 993 | if (status) |
| 994 | goto bail; |
| 995 | |
| 996 | status = o2net_register_handler(DLM_MASTER_REQUEST_MSG, dlm->key, |
| 997 | sizeof(struct dlm_master_request), |
| 998 | dlm_master_request_handler, |
| 999 | dlm, &dlm->dlm_domain_handlers); |
| 1000 | if (status) |
| 1001 | goto bail; |
| 1002 | |
| 1003 | status = o2net_register_handler(DLM_ASSERT_MASTER_MSG, dlm->key, |
| 1004 | sizeof(struct dlm_assert_master), |
| 1005 | dlm_assert_master_handler, |
| 1006 | dlm, &dlm->dlm_domain_handlers); |
| 1007 | if (status) |
| 1008 | goto bail; |
| 1009 | |
| 1010 | status = o2net_register_handler(DLM_CREATE_LOCK_MSG, dlm->key, |
| 1011 | sizeof(struct dlm_create_lock), |
| 1012 | dlm_create_lock_handler, |
| 1013 | dlm, &dlm->dlm_domain_handlers); |
| 1014 | if (status) |
| 1015 | goto bail; |
| 1016 | |
| 1017 | status = o2net_register_handler(DLM_CONVERT_LOCK_MSG, dlm->key, |
| 1018 | DLM_CONVERT_LOCK_MAX_LEN, |
| 1019 | dlm_convert_lock_handler, |
| 1020 | dlm, &dlm->dlm_domain_handlers); |
| 1021 | if (status) |
| 1022 | goto bail; |
| 1023 | |
| 1024 | status = o2net_register_handler(DLM_UNLOCK_LOCK_MSG, dlm->key, |
| 1025 | DLM_UNLOCK_LOCK_MAX_LEN, |
| 1026 | dlm_unlock_lock_handler, |
| 1027 | dlm, &dlm->dlm_domain_handlers); |
| 1028 | if (status) |
| 1029 | goto bail; |
| 1030 | |
| 1031 | status = o2net_register_handler(DLM_PROXY_AST_MSG, dlm->key, |
| 1032 | DLM_PROXY_AST_MAX_LEN, |
| 1033 | dlm_proxy_ast_handler, |
| 1034 | dlm, &dlm->dlm_domain_handlers); |
| 1035 | if (status) |
| 1036 | goto bail; |
| 1037 | |
| 1038 | status = o2net_register_handler(DLM_EXIT_DOMAIN_MSG, dlm->key, |
| 1039 | sizeof(struct dlm_exit_domain), |
| 1040 | dlm_exit_domain_handler, |
| 1041 | dlm, &dlm->dlm_domain_handlers); |
| 1042 | if (status) |
| 1043 | goto bail; |
| 1044 | |
| 1045 | status = o2net_register_handler(DLM_MIGRATE_REQUEST_MSG, dlm->key, |
| 1046 | sizeof(struct dlm_migrate_request), |
| 1047 | dlm_migrate_request_handler, |
| 1048 | dlm, &dlm->dlm_domain_handlers); |
| 1049 | if (status) |
| 1050 | goto bail; |
| 1051 | |
| 1052 | status = o2net_register_handler(DLM_MIG_LOCKRES_MSG, dlm->key, |
| 1053 | DLM_MIG_LOCKRES_MAX_LEN, |
| 1054 | dlm_mig_lockres_handler, |
| 1055 | dlm, &dlm->dlm_domain_handlers); |
| 1056 | if (status) |
| 1057 | goto bail; |
| 1058 | |
| 1059 | status = o2net_register_handler(DLM_MASTER_REQUERY_MSG, dlm->key, |
| 1060 | sizeof(struct dlm_master_requery), |
| 1061 | dlm_master_requery_handler, |
| 1062 | dlm, &dlm->dlm_domain_handlers); |
| 1063 | if (status) |
| 1064 | goto bail; |
| 1065 | |
| 1066 | status = o2net_register_handler(DLM_LOCK_REQUEST_MSG, dlm->key, |
| 1067 | sizeof(struct dlm_lock_request), |
| 1068 | dlm_request_all_locks_handler, |
| 1069 | dlm, &dlm->dlm_domain_handlers); |
| 1070 | if (status) |
| 1071 | goto bail; |
| 1072 | |
| 1073 | status = o2net_register_handler(DLM_RECO_DATA_DONE_MSG, dlm->key, |
| 1074 | sizeof(struct dlm_reco_data_done), |
| 1075 | dlm_reco_data_done_handler, |
| 1076 | dlm, &dlm->dlm_domain_handlers); |
| 1077 | if (status) |
| 1078 | goto bail; |
| 1079 | |
| 1080 | status = o2net_register_handler(DLM_BEGIN_RECO_MSG, dlm->key, |
| 1081 | sizeof(struct dlm_begin_reco), |
| 1082 | dlm_begin_reco_handler, |
| 1083 | dlm, &dlm->dlm_domain_handlers); |
| 1084 | if (status) |
| 1085 | goto bail; |
| 1086 | |
| 1087 | status = o2net_register_handler(DLM_FINALIZE_RECO_MSG, dlm->key, |
| 1088 | sizeof(struct dlm_finalize_reco), |
| 1089 | dlm_finalize_reco_handler, |
| 1090 | dlm, &dlm->dlm_domain_handlers); |
| 1091 | if (status) |
| 1092 | goto bail; |
| 1093 | |
| 1094 | bail: |
| 1095 | if (status) |
| 1096 | dlm_unregister_domain_handlers(dlm); |
| 1097 | |
| 1098 | return status; |
| 1099 | } |
| 1100 | |
| 1101 | static int dlm_join_domain(struct dlm_ctxt *dlm) |
| 1102 | { |
| 1103 | int status; |
| 1104 | |
| 1105 | BUG_ON(!dlm); |
| 1106 | |
| 1107 | mlog(0, "Join domain %s\n", dlm->name); |
| 1108 | |
| 1109 | status = dlm_register_domain_handlers(dlm); |
| 1110 | if (status) { |
| 1111 | mlog_errno(status); |
| 1112 | goto bail; |
| 1113 | } |
| 1114 | |
| 1115 | status = dlm_launch_thread(dlm); |
| 1116 | if (status < 0) { |
| 1117 | mlog_errno(status); |
| 1118 | goto bail; |
| 1119 | } |
| 1120 | |
| 1121 | status = dlm_launch_recovery_thread(dlm); |
| 1122 | if (status < 0) { |
| 1123 | mlog_errno(status); |
| 1124 | goto bail; |
| 1125 | } |
| 1126 | |
| 1127 | do { |
| 1128 | unsigned int backoff; |
| 1129 | status = dlm_try_to_join_domain(dlm); |
| 1130 | |
| 1131 | /* If we're racing another node to the join, then we |
| 1132 | * need to back off temporarily and let them |
| 1133 | * complete. */ |
| 1134 | if (status == -EAGAIN) { |
| 1135 | if (signal_pending(current)) { |
| 1136 | status = -ERESTARTSYS; |
| 1137 | goto bail; |
| 1138 | } |
| 1139 | |
| 1140 | /* |
| 1141 | * <chip> After you! |
| 1142 | * <dale> No, after you! |
| 1143 | * <chip> I insist! |
| 1144 | * <dale> But you first! |
| 1145 | * ... |
| 1146 | */ |
| 1147 | backoff = (unsigned int)(jiffies & 0x3); |
| 1148 | backoff *= DLM_DOMAIN_BACKOFF_MS; |
| 1149 | mlog(0, "backoff %d\n", backoff); |
| 1150 | msleep(backoff); |
| 1151 | } |
| 1152 | } while (status == -EAGAIN); |
| 1153 | |
| 1154 | if (status < 0) { |
| 1155 | mlog_errno(status); |
| 1156 | goto bail; |
| 1157 | } |
| 1158 | |
| 1159 | status = 0; |
| 1160 | bail: |
| 1161 | wake_up(&dlm_domain_events); |
| 1162 | |
| 1163 | if (status) { |
| 1164 | dlm_unregister_domain_handlers(dlm); |
| 1165 | dlm_complete_thread(dlm); |
| 1166 | dlm_complete_recovery_thread(dlm); |
| 1167 | } |
| 1168 | |
| 1169 | return status; |
| 1170 | } |
| 1171 | |
| 1172 | static struct dlm_ctxt *dlm_alloc_ctxt(const char *domain, |
| 1173 | u32 key) |
| 1174 | { |
| 1175 | int i; |
| 1176 | struct dlm_ctxt *dlm = NULL; |
| 1177 | |
| 1178 | dlm = kcalloc(1, sizeof(*dlm), GFP_KERNEL); |
| 1179 | if (!dlm) { |
| 1180 | mlog_errno(-ENOMEM); |
| 1181 | goto leave; |
| 1182 | } |
| 1183 | |
| 1184 | dlm->name = kmalloc(strlen(domain) + 1, GFP_KERNEL); |
| 1185 | if (dlm->name == NULL) { |
| 1186 | mlog_errno(-ENOMEM); |
| 1187 | kfree(dlm); |
| 1188 | dlm = NULL; |
| 1189 | goto leave; |
| 1190 | } |
| 1191 | |
Mark Fasheh | 81f2094 | 2006-02-28 17:31:22 -0800 | [diff] [blame] | 1192 | dlm->lockres_hash = (struct hlist_head *) __get_free_page(GFP_KERNEL); |
| 1193 | if (!dlm->lockres_hash) { |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1194 | mlog_errno(-ENOMEM); |
| 1195 | kfree(dlm->name); |
| 1196 | kfree(dlm); |
| 1197 | dlm = NULL; |
| 1198 | goto leave; |
| 1199 | } |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1200 | |
Mark Fasheh | 81f2094 | 2006-02-28 17:31:22 -0800 | [diff] [blame] | 1201 | for (i=0; i<DLM_HASH_BUCKETS; i++) |
| 1202 | INIT_HLIST_HEAD(&dlm->lockres_hash[i]); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1203 | |
| 1204 | strcpy(dlm->name, domain); |
| 1205 | dlm->key = key; |
| 1206 | dlm->node_num = o2nm_this_node(); |
| 1207 | |
| 1208 | spin_lock_init(&dlm->spinlock); |
| 1209 | spin_lock_init(&dlm->master_lock); |
| 1210 | spin_lock_init(&dlm->ast_lock); |
| 1211 | INIT_LIST_HEAD(&dlm->list); |
| 1212 | INIT_LIST_HEAD(&dlm->dirty_list); |
| 1213 | INIT_LIST_HEAD(&dlm->reco.resources); |
| 1214 | INIT_LIST_HEAD(&dlm->reco.received); |
| 1215 | INIT_LIST_HEAD(&dlm->reco.node_data); |
| 1216 | INIT_LIST_HEAD(&dlm->purge_list); |
| 1217 | INIT_LIST_HEAD(&dlm->dlm_domain_handlers); |
| 1218 | dlm->reco.state = 0; |
| 1219 | |
| 1220 | INIT_LIST_HEAD(&dlm->pending_asts); |
| 1221 | INIT_LIST_HEAD(&dlm->pending_basts); |
| 1222 | |
| 1223 | mlog(0, "dlm->recovery_map=%p, &(dlm->recovery_map[0])=%p\n", |
| 1224 | dlm->recovery_map, &(dlm->recovery_map[0])); |
| 1225 | |
| 1226 | memset(dlm->recovery_map, 0, sizeof(dlm->recovery_map)); |
| 1227 | memset(dlm->live_nodes_map, 0, sizeof(dlm->live_nodes_map)); |
| 1228 | memset(dlm->domain_map, 0, sizeof(dlm->domain_map)); |
| 1229 | |
| 1230 | dlm->dlm_thread_task = NULL; |
| 1231 | dlm->dlm_reco_thread_task = NULL; |
| 1232 | init_waitqueue_head(&dlm->dlm_thread_wq); |
| 1233 | init_waitqueue_head(&dlm->dlm_reco_thread_wq); |
| 1234 | init_waitqueue_head(&dlm->reco.event); |
| 1235 | init_waitqueue_head(&dlm->ast_wq); |
| 1236 | init_waitqueue_head(&dlm->migration_wq); |
| 1237 | INIT_LIST_HEAD(&dlm->master_list); |
| 1238 | INIT_LIST_HEAD(&dlm->mle_hb_events); |
| 1239 | |
| 1240 | dlm->joining_node = DLM_LOCK_RES_OWNER_UNKNOWN; |
| 1241 | init_waitqueue_head(&dlm->dlm_join_events); |
| 1242 | |
| 1243 | dlm->reco.new_master = O2NM_INVALID_NODE_NUM; |
| 1244 | dlm->reco.dead_node = O2NM_INVALID_NODE_NUM; |
| 1245 | atomic_set(&dlm->local_resources, 0); |
| 1246 | atomic_set(&dlm->remote_resources, 0); |
| 1247 | atomic_set(&dlm->unknown_resources, 0); |
| 1248 | |
| 1249 | spin_lock_init(&dlm->work_lock); |
| 1250 | INIT_LIST_HEAD(&dlm->work_list); |
| 1251 | INIT_WORK(&dlm->dispatched_work, dlm_dispatch_work, dlm); |
| 1252 | |
| 1253 | kref_init(&dlm->dlm_refs); |
| 1254 | dlm->dlm_state = DLM_CTXT_NEW; |
| 1255 | |
| 1256 | INIT_LIST_HEAD(&dlm->dlm_eviction_callbacks); |
| 1257 | |
| 1258 | mlog(0, "context init: refcount %u\n", |
| 1259 | atomic_read(&dlm->dlm_refs.refcount)); |
| 1260 | |
| 1261 | leave: |
| 1262 | return dlm; |
| 1263 | } |
| 1264 | |
| 1265 | /* |
| 1266 | * dlm_register_domain: one-time setup per "domain" |
| 1267 | */ |
| 1268 | struct dlm_ctxt * dlm_register_domain(const char *domain, |
| 1269 | u32 key) |
| 1270 | { |
| 1271 | int ret; |
| 1272 | struct dlm_ctxt *dlm = NULL; |
| 1273 | struct dlm_ctxt *new_ctxt = NULL; |
| 1274 | |
| 1275 | if (strlen(domain) > O2NM_MAX_NAME_LEN) { |
| 1276 | ret = -ENAMETOOLONG; |
| 1277 | mlog(ML_ERROR, "domain name length too long\n"); |
| 1278 | goto leave; |
| 1279 | } |
| 1280 | |
| 1281 | if (!o2hb_check_local_node_heartbeating()) { |
| 1282 | mlog(ML_ERROR, "the local node has not been configured, or is " |
| 1283 | "not heartbeating\n"); |
| 1284 | ret = -EPROTO; |
| 1285 | goto leave; |
| 1286 | } |
| 1287 | |
| 1288 | mlog(0, "register called for domain \"%s\"\n", domain); |
| 1289 | |
| 1290 | retry: |
| 1291 | dlm = NULL; |
| 1292 | if (signal_pending(current)) { |
| 1293 | ret = -ERESTARTSYS; |
| 1294 | mlog_errno(ret); |
| 1295 | goto leave; |
| 1296 | } |
| 1297 | |
| 1298 | spin_lock(&dlm_domain_lock); |
| 1299 | |
| 1300 | dlm = __dlm_lookup_domain(domain); |
| 1301 | if (dlm) { |
| 1302 | if (dlm->dlm_state != DLM_CTXT_JOINED) { |
| 1303 | spin_unlock(&dlm_domain_lock); |
| 1304 | |
| 1305 | mlog(0, "This ctxt is not joined yet!\n"); |
| 1306 | wait_event_interruptible(dlm_domain_events, |
| 1307 | dlm_wait_on_domain_helper( |
| 1308 | domain)); |
| 1309 | goto retry; |
| 1310 | } |
| 1311 | |
| 1312 | __dlm_get(dlm); |
| 1313 | dlm->num_joins++; |
| 1314 | |
| 1315 | spin_unlock(&dlm_domain_lock); |
| 1316 | |
| 1317 | ret = 0; |
| 1318 | goto leave; |
| 1319 | } |
| 1320 | |
| 1321 | /* doesn't exist */ |
| 1322 | if (!new_ctxt) { |
| 1323 | spin_unlock(&dlm_domain_lock); |
| 1324 | |
| 1325 | new_ctxt = dlm_alloc_ctxt(domain, key); |
| 1326 | if (new_ctxt) |
| 1327 | goto retry; |
| 1328 | |
| 1329 | ret = -ENOMEM; |
| 1330 | mlog_errno(ret); |
| 1331 | goto leave; |
| 1332 | } |
| 1333 | |
| 1334 | /* a little variable switch-a-roo here... */ |
| 1335 | dlm = new_ctxt; |
| 1336 | new_ctxt = NULL; |
| 1337 | |
| 1338 | /* add the new domain */ |
| 1339 | list_add_tail(&dlm->list, &dlm_domains); |
| 1340 | spin_unlock(&dlm_domain_lock); |
| 1341 | |
| 1342 | ret = dlm_join_domain(dlm); |
| 1343 | if (ret) { |
| 1344 | mlog_errno(ret); |
| 1345 | dlm_put(dlm); |
| 1346 | goto leave; |
| 1347 | } |
| 1348 | |
| 1349 | ret = 0; |
| 1350 | leave: |
| 1351 | if (new_ctxt) |
| 1352 | dlm_free_ctxt_mem(new_ctxt); |
| 1353 | |
| 1354 | if (ret < 0) |
| 1355 | dlm = ERR_PTR(ret); |
| 1356 | |
| 1357 | return dlm; |
| 1358 | } |
| 1359 | EXPORT_SYMBOL_GPL(dlm_register_domain); |
| 1360 | |
| 1361 | static LIST_HEAD(dlm_join_handlers); |
| 1362 | |
| 1363 | static void dlm_unregister_net_handlers(void) |
| 1364 | { |
| 1365 | o2net_unregister_handler_list(&dlm_join_handlers); |
| 1366 | } |
| 1367 | |
| 1368 | static int dlm_register_net_handlers(void) |
| 1369 | { |
| 1370 | int status = 0; |
| 1371 | |
| 1372 | status = o2net_register_handler(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, |
| 1373 | sizeof(struct dlm_query_join_request), |
| 1374 | dlm_query_join_handler, |
| 1375 | NULL, &dlm_join_handlers); |
| 1376 | if (status) |
| 1377 | goto bail; |
| 1378 | |
| 1379 | status = o2net_register_handler(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY, |
| 1380 | sizeof(struct dlm_assert_joined), |
| 1381 | dlm_assert_joined_handler, |
| 1382 | NULL, &dlm_join_handlers); |
| 1383 | if (status) |
| 1384 | goto bail; |
| 1385 | |
| 1386 | status = o2net_register_handler(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY, |
| 1387 | sizeof(struct dlm_cancel_join), |
| 1388 | dlm_cancel_join_handler, |
| 1389 | NULL, &dlm_join_handlers); |
| 1390 | |
| 1391 | bail: |
| 1392 | if (status < 0) |
| 1393 | dlm_unregister_net_handlers(); |
| 1394 | |
| 1395 | return status; |
| 1396 | } |
| 1397 | |
| 1398 | /* Domain eviction callback handling. |
| 1399 | * |
| 1400 | * The file system requires notification of node death *before* the |
| 1401 | * dlm completes it's recovery work, otherwise it may be able to |
| 1402 | * acquire locks on resources requiring recovery. Since the dlm can |
| 1403 | * evict a node from it's domain *before* heartbeat fires, a similar |
| 1404 | * mechanism is required. */ |
| 1405 | |
| 1406 | /* Eviction is not expected to happen often, so a per-domain lock is |
| 1407 | * not necessary. Eviction callbacks are allowed to sleep for short |
| 1408 | * periods of time. */ |
| 1409 | static DECLARE_RWSEM(dlm_callback_sem); |
| 1410 | |
| 1411 | void dlm_fire_domain_eviction_callbacks(struct dlm_ctxt *dlm, |
| 1412 | int node_num) |
| 1413 | { |
| 1414 | struct list_head *iter; |
| 1415 | struct dlm_eviction_cb *cb; |
| 1416 | |
| 1417 | down_read(&dlm_callback_sem); |
| 1418 | list_for_each(iter, &dlm->dlm_eviction_callbacks) { |
| 1419 | cb = list_entry(iter, struct dlm_eviction_cb, ec_item); |
| 1420 | |
| 1421 | cb->ec_func(node_num, cb->ec_data); |
| 1422 | } |
| 1423 | up_read(&dlm_callback_sem); |
| 1424 | } |
| 1425 | |
| 1426 | void dlm_setup_eviction_cb(struct dlm_eviction_cb *cb, |
| 1427 | dlm_eviction_func *f, |
| 1428 | void *data) |
| 1429 | { |
| 1430 | INIT_LIST_HEAD(&cb->ec_item); |
| 1431 | cb->ec_func = f; |
| 1432 | cb->ec_data = data; |
| 1433 | } |
| 1434 | EXPORT_SYMBOL_GPL(dlm_setup_eviction_cb); |
| 1435 | |
| 1436 | void dlm_register_eviction_cb(struct dlm_ctxt *dlm, |
| 1437 | struct dlm_eviction_cb *cb) |
| 1438 | { |
| 1439 | down_write(&dlm_callback_sem); |
| 1440 | list_add_tail(&cb->ec_item, &dlm->dlm_eviction_callbacks); |
| 1441 | up_write(&dlm_callback_sem); |
| 1442 | } |
| 1443 | EXPORT_SYMBOL_GPL(dlm_register_eviction_cb); |
| 1444 | |
| 1445 | void dlm_unregister_eviction_cb(struct dlm_eviction_cb *cb) |
| 1446 | { |
| 1447 | down_write(&dlm_callback_sem); |
| 1448 | list_del_init(&cb->ec_item); |
| 1449 | up_write(&dlm_callback_sem); |
| 1450 | } |
| 1451 | EXPORT_SYMBOL_GPL(dlm_unregister_eviction_cb); |
| 1452 | |
| 1453 | static int __init dlm_init(void) |
| 1454 | { |
| 1455 | int status; |
| 1456 | |
| 1457 | dlm_print_version(); |
| 1458 | |
| 1459 | status = dlm_init_mle_cache(); |
| 1460 | if (status) |
| 1461 | return -1; |
| 1462 | |
| 1463 | status = dlm_register_net_handlers(); |
| 1464 | if (status) { |
| 1465 | dlm_destroy_mle_cache(); |
| 1466 | return -1; |
| 1467 | } |
| 1468 | |
| 1469 | return 0; |
| 1470 | } |
| 1471 | |
| 1472 | static void __exit dlm_exit (void) |
| 1473 | { |
| 1474 | dlm_unregister_net_handlers(); |
| 1475 | dlm_destroy_mle_cache(); |
| 1476 | } |
| 1477 | |
| 1478 | MODULE_AUTHOR("Oracle"); |
| 1479 | MODULE_LICENSE("GPL"); |
| 1480 | |
| 1481 | module_init(dlm_init); |
| 1482 | module_exit(dlm_exit); |