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 | * dlmmod.c |
| 5 | * |
| 6 | * standalone DLM module |
| 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 | |
| 28 | #include <linux/module.h> |
| 29 | #include <linux/fs.h> |
| 30 | #include <linux/types.h> |
| 31 | #include <linux/slab.h> |
| 32 | #include <linux/highmem.h> |
| 33 | #include <linux/utsname.h> |
| 34 | #include <linux/init.h> |
| 35 | #include <linux/sysctl.h> |
| 36 | #include <linux/random.h> |
| 37 | #include <linux/blkdev.h> |
| 38 | #include <linux/socket.h> |
| 39 | #include <linux/inet.h> |
| 40 | #include <linux/spinlock.h> |
| 41 | #include <linux/delay.h> |
| 42 | |
| 43 | |
| 44 | #include "cluster/heartbeat.h" |
| 45 | #include "cluster/nodemanager.h" |
| 46 | #include "cluster/tcp.h" |
| 47 | |
| 48 | #include "dlmapi.h" |
| 49 | #include "dlmcommon.h" |
Adrian Bunk | 82353b5 | 2005-12-19 11:16:07 -0800 | [diff] [blame] | 50 | #include "dlmdomain.h" |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 51 | |
| 52 | #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_MASTER) |
| 53 | #include "cluster/masklog.h" |
| 54 | |
| 55 | enum dlm_mle_type { |
| 56 | DLM_MLE_BLOCK, |
| 57 | DLM_MLE_MASTER, |
| 58 | DLM_MLE_MIGRATION |
| 59 | }; |
| 60 | |
| 61 | struct dlm_lock_name |
| 62 | { |
| 63 | u8 len; |
| 64 | u8 name[DLM_LOCKID_NAME_MAX]; |
| 65 | }; |
| 66 | |
| 67 | struct dlm_master_list_entry |
| 68 | { |
| 69 | struct list_head list; |
| 70 | struct list_head hb_events; |
| 71 | struct dlm_ctxt *dlm; |
| 72 | spinlock_t spinlock; |
| 73 | wait_queue_head_t wq; |
| 74 | atomic_t woken; |
| 75 | struct kref mle_refs; |
Kurt Hackel | a2bf047 | 2006-04-27 18:51:26 -0700 | [diff] [blame] | 76 | int inuse; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 77 | unsigned long maybe_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; |
| 78 | unsigned long vote_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; |
| 79 | unsigned long response_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; |
| 80 | unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; |
| 81 | u8 master; |
| 82 | u8 new_master; |
| 83 | enum dlm_mle_type type; |
| 84 | struct o2hb_callback_func mle_hb_up; |
| 85 | struct o2hb_callback_func mle_hb_down; |
| 86 | union { |
| 87 | struct dlm_lock_resource *res; |
| 88 | struct dlm_lock_name name; |
| 89 | } u; |
| 90 | }; |
| 91 | |
| 92 | static void dlm_mle_node_down(struct dlm_ctxt *dlm, |
| 93 | struct dlm_master_list_entry *mle, |
| 94 | struct o2nm_node *node, |
| 95 | int idx); |
| 96 | static void dlm_mle_node_up(struct dlm_ctxt *dlm, |
| 97 | struct dlm_master_list_entry *mle, |
| 98 | struct o2nm_node *node, |
| 99 | int idx); |
| 100 | |
| 101 | static void dlm_assert_master_worker(struct dlm_work_item *item, void *data); |
| 102 | static int dlm_do_assert_master(struct dlm_ctxt *dlm, const char *lockname, |
| 103 | unsigned int namelen, void *nodemap, |
| 104 | u32 flags); |
| 105 | |
| 106 | static inline int dlm_mle_equal(struct dlm_ctxt *dlm, |
| 107 | struct dlm_master_list_entry *mle, |
| 108 | const char *name, |
| 109 | unsigned int namelen) |
| 110 | { |
| 111 | struct dlm_lock_resource *res; |
| 112 | |
| 113 | if (dlm != mle->dlm) |
| 114 | return 0; |
| 115 | |
| 116 | if (mle->type == DLM_MLE_BLOCK || |
| 117 | mle->type == DLM_MLE_MIGRATION) { |
| 118 | if (namelen != mle->u.name.len || |
| 119 | memcmp(name, mle->u.name.name, namelen)!=0) |
| 120 | return 0; |
| 121 | } else { |
| 122 | res = mle->u.res; |
| 123 | if (namelen != res->lockname.len || |
| 124 | memcmp(res->lockname.name, name, namelen) != 0) |
| 125 | return 0; |
| 126 | } |
| 127 | return 1; |
| 128 | } |
| 129 | |
Kurt Hackel | 9588371 | 2006-04-27 18:47:41 -0700 | [diff] [blame] | 130 | #define dlm_print_nodemap(m) _dlm_print_nodemap(m,#m) |
Mark Fasheh | 8a9343f | 2006-05-01 14:55:10 -0700 | [diff] [blame] | 131 | static void _dlm_print_nodemap(unsigned long *map, const char *mapname) |
Kurt Hackel | 9588371 | 2006-04-27 18:47:41 -0700 | [diff] [blame] | 132 | { |
| 133 | int i; |
| 134 | printk("%s=[ ", mapname); |
| 135 | for (i=0; i<O2NM_MAX_NODES; i++) |
| 136 | if (test_bit(i, map)) |
| 137 | printk("%d ", i); |
| 138 | printk("]"); |
| 139 | } |
| 140 | |
Mark Fasheh | 8a9343f | 2006-05-01 14:55:10 -0700 | [diff] [blame] | 141 | static void dlm_print_one_mle(struct dlm_master_list_entry *mle) |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 142 | { |
Kurt Hackel | 9588371 | 2006-04-27 18:47:41 -0700 | [diff] [blame] | 143 | int refs; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 144 | char *type; |
| 145 | char attached; |
| 146 | u8 master; |
| 147 | unsigned int namelen; |
| 148 | const char *name; |
| 149 | struct kref *k; |
Kurt Hackel | 9588371 | 2006-04-27 18:47:41 -0700 | [diff] [blame] | 150 | unsigned long *maybe = mle->maybe_map, |
| 151 | *vote = mle->vote_map, |
| 152 | *resp = mle->response_map, |
| 153 | *node = mle->node_map; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 154 | |
| 155 | k = &mle->mle_refs; |
| 156 | if (mle->type == DLM_MLE_BLOCK) |
| 157 | type = "BLK"; |
| 158 | else if (mle->type == DLM_MLE_MASTER) |
| 159 | type = "MAS"; |
| 160 | else |
| 161 | type = "MIG"; |
| 162 | refs = atomic_read(&k->refcount); |
| 163 | master = mle->master; |
| 164 | attached = (list_empty(&mle->hb_events) ? 'N' : 'Y'); |
| 165 | |
| 166 | if (mle->type != DLM_MLE_MASTER) { |
| 167 | namelen = mle->u.name.len; |
| 168 | name = mle->u.name.name; |
| 169 | } else { |
| 170 | namelen = mle->u.res->lockname.len; |
| 171 | name = mle->u.res->lockname.name; |
| 172 | } |
| 173 | |
Kurt Hackel | 9588371 | 2006-04-27 18:47:41 -0700 | [diff] [blame] | 174 | mlog(ML_NOTICE, "%.*s: %3s refs=%3d mas=%3u new=%3u evt=%c inuse=%d ", |
| 175 | namelen, name, type, refs, master, mle->new_master, attached, |
| 176 | mle->inuse); |
| 177 | dlm_print_nodemap(maybe); |
| 178 | printk(", "); |
| 179 | dlm_print_nodemap(vote); |
| 180 | printk(", "); |
| 181 | dlm_print_nodemap(resp); |
| 182 | printk(", "); |
| 183 | dlm_print_nodemap(node); |
| 184 | printk(", "); |
| 185 | printk("\n"); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 186 | } |
| 187 | |
Mark Fasheh | 8a9343f | 2006-05-01 14:55:10 -0700 | [diff] [blame] | 188 | #if 0 |
| 189 | /* Code here is included but defined out as it aids debugging */ |
| 190 | |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 191 | static void dlm_dump_mles(struct dlm_ctxt *dlm) |
| 192 | { |
| 193 | struct dlm_master_list_entry *mle; |
| 194 | struct list_head *iter; |
| 195 | |
| 196 | mlog(ML_NOTICE, "dumping all mles for domain %s:\n", dlm->name); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 197 | spin_lock(&dlm->master_lock); |
| 198 | list_for_each(iter, &dlm->master_list) { |
| 199 | mle = list_entry(iter, struct dlm_master_list_entry, list); |
| 200 | dlm_print_one_mle(mle); |
| 201 | } |
| 202 | spin_unlock(&dlm->master_lock); |
| 203 | } |
| 204 | |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 205 | int dlm_dump_all_mles(const char __user *data, unsigned int len) |
| 206 | { |
| 207 | struct list_head *iter; |
| 208 | struct dlm_ctxt *dlm; |
| 209 | |
| 210 | spin_lock(&dlm_domain_lock); |
| 211 | list_for_each(iter, &dlm_domains) { |
| 212 | dlm = list_entry (iter, struct dlm_ctxt, list); |
| 213 | mlog(ML_NOTICE, "found dlm: %p, name=%s\n", dlm, dlm->name); |
| 214 | dlm_dump_mles(dlm); |
| 215 | } |
| 216 | spin_unlock(&dlm_domain_lock); |
| 217 | return len; |
| 218 | } |
| 219 | EXPORT_SYMBOL_GPL(dlm_dump_all_mles); |
| 220 | |
| 221 | #endif /* 0 */ |
| 222 | |
| 223 | |
| 224 | static kmem_cache_t *dlm_mle_cache = NULL; |
| 225 | |
| 226 | |
| 227 | static void dlm_mle_release(struct kref *kref); |
| 228 | static void dlm_init_mle(struct dlm_master_list_entry *mle, |
| 229 | enum dlm_mle_type type, |
| 230 | struct dlm_ctxt *dlm, |
| 231 | struct dlm_lock_resource *res, |
| 232 | const char *name, |
| 233 | unsigned int namelen); |
| 234 | static void dlm_put_mle(struct dlm_master_list_entry *mle); |
| 235 | static void __dlm_put_mle(struct dlm_master_list_entry *mle); |
| 236 | static int dlm_find_mle(struct dlm_ctxt *dlm, |
| 237 | struct dlm_master_list_entry **mle, |
| 238 | char *name, unsigned int namelen); |
| 239 | |
| 240 | static int dlm_do_master_request(struct dlm_master_list_entry *mle, int to); |
| 241 | |
| 242 | |
| 243 | static int dlm_wait_for_lock_mastery(struct dlm_ctxt *dlm, |
| 244 | struct dlm_lock_resource *res, |
| 245 | struct dlm_master_list_entry *mle, |
| 246 | int *blocked); |
| 247 | static int dlm_restart_lock_mastery(struct dlm_ctxt *dlm, |
| 248 | struct dlm_lock_resource *res, |
| 249 | struct dlm_master_list_entry *mle, |
| 250 | int blocked); |
| 251 | static int dlm_add_migration_mle(struct dlm_ctxt *dlm, |
| 252 | struct dlm_lock_resource *res, |
| 253 | struct dlm_master_list_entry *mle, |
| 254 | struct dlm_master_list_entry **oldmle, |
| 255 | const char *name, unsigned int namelen, |
| 256 | u8 new_master, u8 master); |
| 257 | |
| 258 | static u8 dlm_pick_migration_target(struct dlm_ctxt *dlm, |
| 259 | struct dlm_lock_resource *res); |
| 260 | static void dlm_remove_nonlocal_locks(struct dlm_ctxt *dlm, |
| 261 | struct dlm_lock_resource *res); |
| 262 | static int dlm_mark_lockres_migrating(struct dlm_ctxt *dlm, |
| 263 | struct dlm_lock_resource *res, |
| 264 | u8 target); |
Kurt Hackel | c03872f | 2006-03-06 14:08:49 -0800 | [diff] [blame] | 265 | static int dlm_pre_master_reco_lockres(struct dlm_ctxt *dlm, |
| 266 | struct dlm_lock_resource *res); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 267 | |
| 268 | |
| 269 | int dlm_is_host_down(int errno) |
| 270 | { |
| 271 | switch (errno) { |
| 272 | case -EBADF: |
| 273 | case -ECONNREFUSED: |
| 274 | case -ENOTCONN: |
| 275 | case -ECONNRESET: |
| 276 | case -EPIPE: |
| 277 | case -EHOSTDOWN: |
| 278 | case -EHOSTUNREACH: |
| 279 | case -ETIMEDOUT: |
| 280 | case -ECONNABORTED: |
| 281 | case -ENETDOWN: |
| 282 | case -ENETUNREACH: |
| 283 | case -ENETRESET: |
| 284 | case -ESHUTDOWN: |
| 285 | case -ENOPROTOOPT: |
| 286 | case -EINVAL: /* if returned from our tcp code, |
| 287 | this means there is no socket */ |
| 288 | return 1; |
| 289 | } |
| 290 | return 0; |
| 291 | } |
| 292 | |
| 293 | |
| 294 | /* |
| 295 | * MASTER LIST FUNCTIONS |
| 296 | */ |
| 297 | |
| 298 | |
| 299 | /* |
| 300 | * regarding master list entries and heartbeat callbacks: |
| 301 | * |
| 302 | * in order to avoid sleeping and allocation that occurs in |
| 303 | * heartbeat, master list entries are simply attached to the |
| 304 | * dlm's established heartbeat callbacks. the mle is attached |
| 305 | * when it is created, and since the dlm->spinlock is held at |
| 306 | * that time, any heartbeat event will be properly discovered |
| 307 | * by the mle. the mle needs to be detached from the |
| 308 | * dlm->mle_hb_events list as soon as heartbeat events are no |
| 309 | * longer useful to the mle, and before the mle is freed. |
| 310 | * |
| 311 | * as a general rule, heartbeat events are no longer needed by |
| 312 | * the mle once an "answer" regarding the lock master has been |
| 313 | * received. |
| 314 | */ |
| 315 | static inline void __dlm_mle_attach_hb_events(struct dlm_ctxt *dlm, |
| 316 | struct dlm_master_list_entry *mle) |
| 317 | { |
| 318 | assert_spin_locked(&dlm->spinlock); |
| 319 | |
| 320 | list_add_tail(&mle->hb_events, &dlm->mle_hb_events); |
| 321 | } |
| 322 | |
| 323 | |
| 324 | static inline void __dlm_mle_detach_hb_events(struct dlm_ctxt *dlm, |
| 325 | struct dlm_master_list_entry *mle) |
| 326 | { |
| 327 | if (!list_empty(&mle->hb_events)) |
| 328 | list_del_init(&mle->hb_events); |
| 329 | } |
| 330 | |
| 331 | |
| 332 | static inline void dlm_mle_detach_hb_events(struct dlm_ctxt *dlm, |
| 333 | struct dlm_master_list_entry *mle) |
| 334 | { |
| 335 | spin_lock(&dlm->spinlock); |
| 336 | __dlm_mle_detach_hb_events(dlm, mle); |
| 337 | spin_unlock(&dlm->spinlock); |
| 338 | } |
| 339 | |
Kurt Hackel | a2bf047 | 2006-04-27 18:51:26 -0700 | [diff] [blame] | 340 | static void dlm_get_mle_inuse(struct dlm_master_list_entry *mle) |
| 341 | { |
| 342 | struct dlm_ctxt *dlm; |
| 343 | dlm = mle->dlm; |
| 344 | |
| 345 | assert_spin_locked(&dlm->spinlock); |
| 346 | assert_spin_locked(&dlm->master_lock); |
| 347 | mle->inuse++; |
| 348 | kref_get(&mle->mle_refs); |
| 349 | } |
| 350 | |
| 351 | static void dlm_put_mle_inuse(struct dlm_master_list_entry *mle) |
| 352 | { |
| 353 | struct dlm_ctxt *dlm; |
| 354 | dlm = mle->dlm; |
| 355 | |
| 356 | spin_lock(&dlm->spinlock); |
| 357 | spin_lock(&dlm->master_lock); |
| 358 | mle->inuse--; |
| 359 | __dlm_put_mle(mle); |
| 360 | spin_unlock(&dlm->master_lock); |
| 361 | spin_unlock(&dlm->spinlock); |
| 362 | |
| 363 | } |
| 364 | |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 365 | /* remove from list and free */ |
| 366 | static void __dlm_put_mle(struct dlm_master_list_entry *mle) |
| 367 | { |
| 368 | struct dlm_ctxt *dlm; |
| 369 | dlm = mle->dlm; |
| 370 | |
| 371 | assert_spin_locked(&dlm->spinlock); |
| 372 | assert_spin_locked(&dlm->master_lock); |
Kurt Hackel | aa85235 | 2006-04-27 19:04:49 -0700 | [diff] [blame] | 373 | if (!atomic_read(&mle->mle_refs.refcount)) { |
| 374 | /* this may or may not crash, but who cares. |
| 375 | * it's a BUG. */ |
| 376 | mlog(ML_ERROR, "bad mle: %p\n", mle); |
| 377 | dlm_print_one_mle(mle); |
| 378 | BUG(); |
| 379 | } else |
| 380 | kref_put(&mle->mle_refs, dlm_mle_release); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 381 | } |
| 382 | |
| 383 | |
| 384 | /* must not have any spinlocks coming in */ |
| 385 | static void dlm_put_mle(struct dlm_master_list_entry *mle) |
| 386 | { |
| 387 | struct dlm_ctxt *dlm; |
| 388 | dlm = mle->dlm; |
| 389 | |
| 390 | spin_lock(&dlm->spinlock); |
| 391 | spin_lock(&dlm->master_lock); |
| 392 | __dlm_put_mle(mle); |
| 393 | spin_unlock(&dlm->master_lock); |
| 394 | spin_unlock(&dlm->spinlock); |
| 395 | } |
| 396 | |
| 397 | static inline void dlm_get_mle(struct dlm_master_list_entry *mle) |
| 398 | { |
| 399 | kref_get(&mle->mle_refs); |
| 400 | } |
| 401 | |
| 402 | static void dlm_init_mle(struct dlm_master_list_entry *mle, |
| 403 | enum dlm_mle_type type, |
| 404 | struct dlm_ctxt *dlm, |
| 405 | struct dlm_lock_resource *res, |
| 406 | const char *name, |
| 407 | unsigned int namelen) |
| 408 | { |
| 409 | assert_spin_locked(&dlm->spinlock); |
| 410 | |
| 411 | mle->dlm = dlm; |
| 412 | mle->type = type; |
| 413 | INIT_LIST_HEAD(&mle->list); |
| 414 | INIT_LIST_HEAD(&mle->hb_events); |
| 415 | memset(mle->maybe_map, 0, sizeof(mle->maybe_map)); |
| 416 | spin_lock_init(&mle->spinlock); |
| 417 | init_waitqueue_head(&mle->wq); |
| 418 | atomic_set(&mle->woken, 0); |
| 419 | kref_init(&mle->mle_refs); |
| 420 | memset(mle->response_map, 0, sizeof(mle->response_map)); |
| 421 | mle->master = O2NM_MAX_NODES; |
| 422 | mle->new_master = O2NM_MAX_NODES; |
Kurt Hackel | a2bf047 | 2006-04-27 18:51:26 -0700 | [diff] [blame] | 423 | mle->inuse = 0; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 424 | |
| 425 | if (mle->type == DLM_MLE_MASTER) { |
| 426 | BUG_ON(!res); |
| 427 | mle->u.res = res; |
| 428 | } else if (mle->type == DLM_MLE_BLOCK) { |
| 429 | BUG_ON(!name); |
| 430 | memcpy(mle->u.name.name, name, namelen); |
| 431 | mle->u.name.len = namelen; |
| 432 | } else /* DLM_MLE_MIGRATION */ { |
| 433 | BUG_ON(!name); |
| 434 | memcpy(mle->u.name.name, name, namelen); |
| 435 | mle->u.name.len = namelen; |
| 436 | } |
| 437 | |
| 438 | /* copy off the node_map and register hb callbacks on our copy */ |
| 439 | memcpy(mle->node_map, dlm->domain_map, sizeof(mle->node_map)); |
| 440 | memcpy(mle->vote_map, dlm->domain_map, sizeof(mle->vote_map)); |
| 441 | clear_bit(dlm->node_num, mle->vote_map); |
| 442 | clear_bit(dlm->node_num, mle->node_map); |
| 443 | |
| 444 | /* attach the mle to the domain node up/down events */ |
| 445 | __dlm_mle_attach_hb_events(dlm, mle); |
| 446 | } |
| 447 | |
| 448 | |
| 449 | /* returns 1 if found, 0 if not */ |
| 450 | static int dlm_find_mle(struct dlm_ctxt *dlm, |
| 451 | struct dlm_master_list_entry **mle, |
| 452 | char *name, unsigned int namelen) |
| 453 | { |
| 454 | struct dlm_master_list_entry *tmpmle; |
| 455 | struct list_head *iter; |
| 456 | |
| 457 | assert_spin_locked(&dlm->master_lock); |
| 458 | |
| 459 | list_for_each(iter, &dlm->master_list) { |
| 460 | tmpmle = list_entry(iter, struct dlm_master_list_entry, list); |
| 461 | if (!dlm_mle_equal(dlm, tmpmle, name, namelen)) |
| 462 | continue; |
| 463 | dlm_get_mle(tmpmle); |
| 464 | *mle = tmpmle; |
| 465 | return 1; |
| 466 | } |
| 467 | return 0; |
| 468 | } |
| 469 | |
| 470 | void dlm_hb_event_notify_attached(struct dlm_ctxt *dlm, int idx, int node_up) |
| 471 | { |
| 472 | struct dlm_master_list_entry *mle; |
| 473 | struct list_head *iter; |
| 474 | |
| 475 | assert_spin_locked(&dlm->spinlock); |
| 476 | |
| 477 | list_for_each(iter, &dlm->mle_hb_events) { |
| 478 | mle = list_entry(iter, struct dlm_master_list_entry, |
| 479 | hb_events); |
| 480 | if (node_up) |
| 481 | dlm_mle_node_up(dlm, mle, NULL, idx); |
| 482 | else |
| 483 | dlm_mle_node_down(dlm, mle, NULL, idx); |
| 484 | } |
| 485 | } |
| 486 | |
| 487 | static void dlm_mle_node_down(struct dlm_ctxt *dlm, |
| 488 | struct dlm_master_list_entry *mle, |
| 489 | struct o2nm_node *node, int idx) |
| 490 | { |
| 491 | spin_lock(&mle->spinlock); |
| 492 | |
| 493 | if (!test_bit(idx, mle->node_map)) |
| 494 | mlog(0, "node %u already removed from nodemap!\n", idx); |
| 495 | else |
| 496 | clear_bit(idx, mle->node_map); |
| 497 | |
| 498 | spin_unlock(&mle->spinlock); |
| 499 | } |
| 500 | |
| 501 | static void dlm_mle_node_up(struct dlm_ctxt *dlm, |
| 502 | struct dlm_master_list_entry *mle, |
| 503 | struct o2nm_node *node, int idx) |
| 504 | { |
| 505 | spin_lock(&mle->spinlock); |
| 506 | |
| 507 | if (test_bit(idx, mle->node_map)) |
| 508 | mlog(0, "node %u already in node map!\n", idx); |
| 509 | else |
| 510 | set_bit(idx, mle->node_map); |
| 511 | |
| 512 | spin_unlock(&mle->spinlock); |
| 513 | } |
| 514 | |
| 515 | |
| 516 | int dlm_init_mle_cache(void) |
| 517 | { |
| 518 | dlm_mle_cache = kmem_cache_create("dlm_mle_cache", |
| 519 | sizeof(struct dlm_master_list_entry), |
| 520 | 0, SLAB_HWCACHE_ALIGN, |
| 521 | NULL, NULL); |
| 522 | if (dlm_mle_cache == NULL) |
| 523 | return -ENOMEM; |
| 524 | return 0; |
| 525 | } |
| 526 | |
| 527 | void dlm_destroy_mle_cache(void) |
| 528 | { |
| 529 | if (dlm_mle_cache) |
| 530 | kmem_cache_destroy(dlm_mle_cache); |
| 531 | } |
| 532 | |
| 533 | static void dlm_mle_release(struct kref *kref) |
| 534 | { |
| 535 | struct dlm_master_list_entry *mle; |
| 536 | struct dlm_ctxt *dlm; |
| 537 | |
| 538 | mlog_entry_void(); |
| 539 | |
| 540 | mle = container_of(kref, struct dlm_master_list_entry, mle_refs); |
| 541 | dlm = mle->dlm; |
| 542 | |
| 543 | if (mle->type != DLM_MLE_MASTER) { |
| 544 | mlog(0, "calling mle_release for %.*s, type %d\n", |
| 545 | mle->u.name.len, mle->u.name.name, mle->type); |
| 546 | } else { |
| 547 | mlog(0, "calling mle_release for %.*s, type %d\n", |
| 548 | mle->u.res->lockname.len, |
| 549 | mle->u.res->lockname.name, mle->type); |
| 550 | } |
| 551 | assert_spin_locked(&dlm->spinlock); |
| 552 | assert_spin_locked(&dlm->master_lock); |
| 553 | |
| 554 | /* remove from list if not already */ |
| 555 | if (!list_empty(&mle->list)) |
| 556 | list_del_init(&mle->list); |
| 557 | |
| 558 | /* detach the mle from the domain node up/down events */ |
| 559 | __dlm_mle_detach_hb_events(dlm, mle); |
| 560 | |
| 561 | /* NOTE: kfree under spinlock here. |
| 562 | * if this is bad, we can move this to a freelist. */ |
| 563 | kmem_cache_free(dlm_mle_cache, mle); |
| 564 | } |
| 565 | |
| 566 | |
| 567 | /* |
| 568 | * LOCK RESOURCE FUNCTIONS |
| 569 | */ |
| 570 | |
| 571 | static void dlm_set_lockres_owner(struct dlm_ctxt *dlm, |
| 572 | struct dlm_lock_resource *res, |
| 573 | u8 owner) |
| 574 | { |
| 575 | assert_spin_locked(&res->spinlock); |
| 576 | |
| 577 | mlog_entry("%.*s, %u\n", res->lockname.len, res->lockname.name, owner); |
| 578 | |
| 579 | if (owner == dlm->node_num) |
| 580 | atomic_inc(&dlm->local_resources); |
| 581 | else if (owner == DLM_LOCK_RES_OWNER_UNKNOWN) |
| 582 | atomic_inc(&dlm->unknown_resources); |
| 583 | else |
| 584 | atomic_inc(&dlm->remote_resources); |
| 585 | |
| 586 | res->owner = owner; |
| 587 | } |
| 588 | |
| 589 | void dlm_change_lockres_owner(struct dlm_ctxt *dlm, |
| 590 | struct dlm_lock_resource *res, u8 owner) |
| 591 | { |
| 592 | assert_spin_locked(&res->spinlock); |
| 593 | |
| 594 | if (owner == res->owner) |
| 595 | return; |
| 596 | |
| 597 | if (res->owner == dlm->node_num) |
| 598 | atomic_dec(&dlm->local_resources); |
| 599 | else if (res->owner == DLM_LOCK_RES_OWNER_UNKNOWN) |
| 600 | atomic_dec(&dlm->unknown_resources); |
| 601 | else |
| 602 | atomic_dec(&dlm->remote_resources); |
| 603 | |
| 604 | dlm_set_lockres_owner(dlm, res, owner); |
| 605 | } |
| 606 | |
| 607 | |
| 608 | static void dlm_lockres_release(struct kref *kref) |
| 609 | { |
| 610 | struct dlm_lock_resource *res; |
| 611 | |
| 612 | res = container_of(kref, struct dlm_lock_resource, refs); |
| 613 | |
| 614 | /* This should not happen -- all lockres' have a name |
| 615 | * associated with them at init time. */ |
| 616 | BUG_ON(!res->lockname.name); |
| 617 | |
| 618 | mlog(0, "destroying lockres %.*s\n", res->lockname.len, |
| 619 | res->lockname.name); |
| 620 | |
Kurt Hackel | a7f90d8 | 2006-04-27 19:24:21 -0700 | [diff] [blame] | 621 | if (!hlist_unhashed(&res->hash_node) || |
| 622 | !list_empty(&res->granted) || |
| 623 | !list_empty(&res->converting) || |
| 624 | !list_empty(&res->blocked) || |
| 625 | !list_empty(&res->dirty) || |
| 626 | !list_empty(&res->recovering) || |
| 627 | !list_empty(&res->purge)) { |
| 628 | mlog(ML_ERROR, |
| 629 | "Going to BUG for resource %.*s." |
| 630 | " We're on a list! [%c%c%c%c%c%c%c]\n", |
| 631 | res->lockname.len, res->lockname.name, |
| 632 | !hlist_unhashed(&res->hash_node) ? 'H' : ' ', |
| 633 | !list_empty(&res->granted) ? 'G' : ' ', |
| 634 | !list_empty(&res->converting) ? 'C' : ' ', |
| 635 | !list_empty(&res->blocked) ? 'B' : ' ', |
| 636 | !list_empty(&res->dirty) ? 'D' : ' ', |
| 637 | !list_empty(&res->recovering) ? 'R' : ' ', |
| 638 | !list_empty(&res->purge) ? 'P' : ' '); |
| 639 | |
| 640 | dlm_print_one_lock_resource(res); |
| 641 | } |
| 642 | |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 643 | /* By the time we're ready to blow this guy away, we shouldn't |
| 644 | * be on any lists. */ |
Mark Fasheh | 81f2094 | 2006-02-28 17:31:22 -0800 | [diff] [blame] | 645 | BUG_ON(!hlist_unhashed(&res->hash_node)); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 646 | BUG_ON(!list_empty(&res->granted)); |
| 647 | BUG_ON(!list_empty(&res->converting)); |
| 648 | BUG_ON(!list_empty(&res->blocked)); |
| 649 | BUG_ON(!list_empty(&res->dirty)); |
| 650 | BUG_ON(!list_empty(&res->recovering)); |
| 651 | BUG_ON(!list_empty(&res->purge)); |
| 652 | |
| 653 | kfree(res->lockname.name); |
| 654 | |
| 655 | kfree(res); |
| 656 | } |
| 657 | |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 658 | void dlm_lockres_put(struct dlm_lock_resource *res) |
| 659 | { |
| 660 | kref_put(&res->refs, dlm_lockres_release); |
| 661 | } |
| 662 | |
| 663 | static void dlm_init_lockres(struct dlm_ctxt *dlm, |
| 664 | struct dlm_lock_resource *res, |
| 665 | const char *name, unsigned int namelen) |
| 666 | { |
| 667 | char *qname; |
| 668 | |
| 669 | /* If we memset here, we lose our reference to the kmalloc'd |
| 670 | * res->lockname.name, so be sure to init every field |
| 671 | * correctly! */ |
| 672 | |
| 673 | qname = (char *) res->lockname.name; |
| 674 | memcpy(qname, name, namelen); |
| 675 | |
| 676 | res->lockname.len = namelen; |
Mark Fasheh | a3d3329 | 2006-03-09 17:55:56 -0800 | [diff] [blame] | 677 | res->lockname.hash = dlm_lockid_hash(name, namelen); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 678 | |
| 679 | init_waitqueue_head(&res->wq); |
| 680 | spin_lock_init(&res->spinlock); |
Mark Fasheh | 81f2094 | 2006-02-28 17:31:22 -0800 | [diff] [blame] | 681 | INIT_HLIST_NODE(&res->hash_node); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 682 | INIT_LIST_HEAD(&res->granted); |
| 683 | INIT_LIST_HEAD(&res->converting); |
| 684 | INIT_LIST_HEAD(&res->blocked); |
| 685 | INIT_LIST_HEAD(&res->dirty); |
| 686 | INIT_LIST_HEAD(&res->recovering); |
| 687 | INIT_LIST_HEAD(&res->purge); |
| 688 | atomic_set(&res->asts_reserved, 0); |
| 689 | res->migration_pending = 0; |
| 690 | |
| 691 | kref_init(&res->refs); |
| 692 | |
| 693 | /* just for consistency */ |
| 694 | spin_lock(&res->spinlock); |
| 695 | dlm_set_lockres_owner(dlm, res, DLM_LOCK_RES_OWNER_UNKNOWN); |
| 696 | spin_unlock(&res->spinlock); |
| 697 | |
| 698 | res->state = DLM_LOCK_RES_IN_PROGRESS; |
| 699 | |
| 700 | res->last_used = 0; |
| 701 | |
| 702 | memset(res->lvb, 0, DLM_LVB_LEN); |
| 703 | } |
| 704 | |
| 705 | struct dlm_lock_resource *dlm_new_lockres(struct dlm_ctxt *dlm, |
| 706 | const char *name, |
| 707 | unsigned int namelen) |
| 708 | { |
| 709 | struct dlm_lock_resource *res; |
| 710 | |
Kurt Hackel | ad8100e | 2006-05-01 14:25:21 -0700 | [diff] [blame] | 711 | res = kmalloc(sizeof(struct dlm_lock_resource), GFP_NOFS); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 712 | if (!res) |
| 713 | return NULL; |
| 714 | |
Kurt Hackel | ad8100e | 2006-05-01 14:25:21 -0700 | [diff] [blame] | 715 | res->lockname.name = kmalloc(namelen, GFP_NOFS); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 716 | if (!res->lockname.name) { |
| 717 | kfree(res); |
| 718 | return NULL; |
| 719 | } |
| 720 | |
| 721 | dlm_init_lockres(dlm, res, name, namelen); |
| 722 | return res; |
| 723 | } |
| 724 | |
| 725 | /* |
| 726 | * lookup a lock resource by name. |
| 727 | * may already exist in the hashtable. |
| 728 | * lockid is null terminated |
| 729 | * |
| 730 | * if not, allocate enough for the lockres and for |
| 731 | * the temporary structure used in doing the mastering. |
| 732 | * |
| 733 | * also, do a lookup in the dlm->master_list to see |
| 734 | * if another node has begun mastering the same lock. |
| 735 | * if so, there should be a block entry in there |
| 736 | * for this name, and we should *not* attempt to master |
| 737 | * the lock here. need to wait around for that node |
| 738 | * to assert_master (or die). |
| 739 | * |
| 740 | */ |
| 741 | struct dlm_lock_resource * dlm_get_lock_resource(struct dlm_ctxt *dlm, |
| 742 | const char *lockid, |
Mark Fasheh | 3384f3d | 2006-09-08 11:38:29 -0700 | [diff] [blame] | 743 | int namelen, |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 744 | int flags) |
| 745 | { |
| 746 | struct dlm_lock_resource *tmpres=NULL, *res=NULL; |
| 747 | struct dlm_master_list_entry *mle = NULL; |
| 748 | struct dlm_master_list_entry *alloc_mle = NULL; |
| 749 | int blocked = 0; |
| 750 | int ret, nodenum; |
| 751 | struct dlm_node_iter iter; |
Mark Fasheh | 3384f3d | 2006-09-08 11:38:29 -0700 | [diff] [blame] | 752 | unsigned int hash; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 753 | int tries = 0; |
Kurt Hackel | c03872f | 2006-03-06 14:08:49 -0800 | [diff] [blame] | 754 | int bit, wait_on_recovery = 0; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 755 | |
| 756 | BUG_ON(!lockid); |
| 757 | |
Mark Fasheh | a3d3329 | 2006-03-09 17:55:56 -0800 | [diff] [blame] | 758 | hash = dlm_lockid_hash(lockid, namelen); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 759 | |
| 760 | mlog(0, "get lockres %s (len %d)\n", lockid, namelen); |
| 761 | |
| 762 | lookup: |
| 763 | spin_lock(&dlm->spinlock); |
Mark Fasheh | a3d3329 | 2006-03-09 17:55:56 -0800 | [diff] [blame] | 764 | tmpres = __dlm_lookup_lockres(dlm, lockid, namelen, hash); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 765 | if (tmpres) { |
| 766 | spin_unlock(&dlm->spinlock); |
| 767 | mlog(0, "found in hash!\n"); |
| 768 | if (res) |
| 769 | dlm_lockres_put(res); |
| 770 | res = tmpres; |
| 771 | goto leave; |
| 772 | } |
| 773 | |
| 774 | if (!res) { |
| 775 | spin_unlock(&dlm->spinlock); |
| 776 | mlog(0, "allocating a new resource\n"); |
| 777 | /* nothing found and we need to allocate one. */ |
| 778 | alloc_mle = (struct dlm_master_list_entry *) |
Kurt Hackel | ad8100e | 2006-05-01 14:25:21 -0700 | [diff] [blame] | 779 | kmem_cache_alloc(dlm_mle_cache, GFP_NOFS); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 780 | if (!alloc_mle) |
| 781 | goto leave; |
| 782 | res = dlm_new_lockres(dlm, lockid, namelen); |
| 783 | if (!res) |
| 784 | goto leave; |
| 785 | goto lookup; |
| 786 | } |
| 787 | |
| 788 | mlog(0, "no lockres found, allocated our own: %p\n", res); |
| 789 | |
| 790 | if (flags & LKM_LOCAL) { |
| 791 | /* caller knows it's safe to assume it's not mastered elsewhere |
| 792 | * DONE! return right away */ |
| 793 | spin_lock(&res->spinlock); |
| 794 | dlm_change_lockres_owner(dlm, res, dlm->node_num); |
| 795 | __dlm_insert_lockres(dlm, res); |
| 796 | spin_unlock(&res->spinlock); |
| 797 | spin_unlock(&dlm->spinlock); |
| 798 | /* lockres still marked IN_PROGRESS */ |
| 799 | goto wake_waiters; |
| 800 | } |
| 801 | |
| 802 | /* check master list to see if another node has started mastering it */ |
| 803 | spin_lock(&dlm->master_lock); |
| 804 | |
| 805 | /* if we found a block, wait for lock to be mastered by another node */ |
| 806 | blocked = dlm_find_mle(dlm, &mle, (char *)lockid, namelen); |
| 807 | if (blocked) { |
| 808 | if (mle->type == DLM_MLE_MASTER) { |
| 809 | mlog(ML_ERROR, "master entry for nonexistent lock!\n"); |
| 810 | BUG(); |
| 811 | } else if (mle->type == DLM_MLE_MIGRATION) { |
| 812 | /* migration is in progress! */ |
| 813 | /* the good news is that we now know the |
| 814 | * "current" master (mle->master). */ |
| 815 | |
| 816 | spin_unlock(&dlm->master_lock); |
| 817 | assert_spin_locked(&dlm->spinlock); |
| 818 | |
| 819 | /* set the lockres owner and hash it */ |
| 820 | spin_lock(&res->spinlock); |
| 821 | dlm_set_lockres_owner(dlm, res, mle->master); |
| 822 | __dlm_insert_lockres(dlm, res); |
| 823 | spin_unlock(&res->spinlock); |
| 824 | spin_unlock(&dlm->spinlock); |
| 825 | |
| 826 | /* master is known, detach */ |
| 827 | dlm_mle_detach_hb_events(dlm, mle); |
| 828 | dlm_put_mle(mle); |
| 829 | mle = NULL; |
| 830 | goto wake_waiters; |
| 831 | } |
| 832 | } else { |
| 833 | /* go ahead and try to master lock on this node */ |
| 834 | mle = alloc_mle; |
| 835 | /* make sure this does not get freed below */ |
| 836 | alloc_mle = NULL; |
| 837 | dlm_init_mle(mle, DLM_MLE_MASTER, dlm, res, NULL, 0); |
| 838 | set_bit(dlm->node_num, mle->maybe_map); |
| 839 | list_add(&mle->list, &dlm->master_list); |
Kurt Hackel | c03872f | 2006-03-06 14:08:49 -0800 | [diff] [blame] | 840 | |
| 841 | /* still holding the dlm spinlock, check the recovery map |
| 842 | * to see if there are any nodes that still need to be |
| 843 | * considered. these will not appear in the mle nodemap |
| 844 | * but they might own this lockres. wait on them. */ |
| 845 | bit = find_next_bit(dlm->recovery_map, O2NM_MAX_NODES, 0); |
| 846 | if (bit < O2NM_MAX_NODES) { |
| 847 | mlog(ML_NOTICE, "%s:%.*s: at least one node (%d) to" |
| 848 | "recover before lock mastery can begin\n", |
| 849 | dlm->name, namelen, (char *)lockid, bit); |
| 850 | wait_on_recovery = 1; |
| 851 | } |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 852 | } |
| 853 | |
| 854 | /* at this point there is either a DLM_MLE_BLOCK or a |
| 855 | * DLM_MLE_MASTER on the master list, so it's safe to add the |
| 856 | * lockres to the hashtable. anyone who finds the lock will |
| 857 | * still have to wait on the IN_PROGRESS. */ |
| 858 | |
| 859 | /* finally add the lockres to its hash bucket */ |
| 860 | __dlm_insert_lockres(dlm, res); |
| 861 | /* get an extra ref on the mle in case this is a BLOCK |
| 862 | * if so, the creator of the BLOCK may try to put the last |
| 863 | * ref at this time in the assert master handler, so we |
| 864 | * need an extra one to keep from a bad ptr deref. */ |
Kurt Hackel | a2bf047 | 2006-04-27 18:51:26 -0700 | [diff] [blame] | 865 | dlm_get_mle_inuse(mle); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 866 | spin_unlock(&dlm->master_lock); |
| 867 | spin_unlock(&dlm->spinlock); |
| 868 | |
Kurt Hackel | e7e69eb | 2006-05-01 11:49:52 -0700 | [diff] [blame] | 869 | redo_request: |
Kurt Hackel | c03872f | 2006-03-06 14:08:49 -0800 | [diff] [blame] | 870 | while (wait_on_recovery) { |
| 871 | /* any cluster changes that occurred after dropping the |
| 872 | * dlm spinlock would be detectable be a change on the mle, |
| 873 | * so we only need to clear out the recovery map once. */ |
| 874 | if (dlm_is_recovery_lock(lockid, namelen)) { |
| 875 | mlog(ML_NOTICE, "%s: recovery map is not empty, but " |
| 876 | "must master $RECOVERY lock now\n", dlm->name); |
| 877 | if (!dlm_pre_master_reco_lockres(dlm, res)) |
| 878 | wait_on_recovery = 0; |
| 879 | else { |
| 880 | mlog(0, "%s: waiting 500ms for heartbeat state " |
| 881 | "change\n", dlm->name); |
| 882 | msleep(500); |
| 883 | } |
| 884 | continue; |
| 885 | } |
| 886 | |
| 887 | dlm_kick_recovery_thread(dlm); |
Kurt Hackel | aa087b8 | 2006-05-01 12:02:07 -0700 | [diff] [blame] | 888 | msleep(1000); |
Kurt Hackel | c03872f | 2006-03-06 14:08:49 -0800 | [diff] [blame] | 889 | dlm_wait_for_recovery(dlm); |
| 890 | |
| 891 | spin_lock(&dlm->spinlock); |
| 892 | bit = find_next_bit(dlm->recovery_map, O2NM_MAX_NODES, 0); |
| 893 | if (bit < O2NM_MAX_NODES) { |
| 894 | mlog(ML_NOTICE, "%s:%.*s: at least one node (%d) to" |
| 895 | "recover before lock mastery can begin\n", |
| 896 | dlm->name, namelen, (char *)lockid, bit); |
| 897 | wait_on_recovery = 1; |
| 898 | } else |
| 899 | wait_on_recovery = 0; |
| 900 | spin_unlock(&dlm->spinlock); |
Kurt Hackel | b7084ab | 2006-05-01 13:54:07 -0700 | [diff] [blame] | 901 | |
| 902 | if (wait_on_recovery) |
| 903 | dlm_wait_for_node_recovery(dlm, bit, 10000); |
Kurt Hackel | c03872f | 2006-03-06 14:08:49 -0800 | [diff] [blame] | 904 | } |
| 905 | |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 906 | /* must wait for lock to be mastered elsewhere */ |
| 907 | if (blocked) |
| 908 | goto wait; |
| 909 | |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 910 | ret = -EINVAL; |
| 911 | dlm_node_iter_init(mle->vote_map, &iter); |
| 912 | while ((nodenum = dlm_node_iter_next(&iter)) >= 0) { |
| 913 | ret = dlm_do_master_request(mle, nodenum); |
| 914 | if (ret < 0) |
| 915 | mlog_errno(ret); |
| 916 | if (mle->master != O2NM_MAX_NODES) { |
| 917 | /* found a master ! */ |
Kurt Hackel | 9c6510a | 2006-03-02 18:09:26 -0800 | [diff] [blame] | 918 | if (mle->master <= nodenum) |
| 919 | break; |
| 920 | /* if our master request has not reached the master |
| 921 | * yet, keep going until it does. this is how the |
| 922 | * master will know that asserts are needed back to |
| 923 | * the lower nodes. */ |
| 924 | mlog(0, "%s:%.*s: requests only up to %u but master " |
| 925 | "is %u, keep going\n", dlm->name, namelen, |
| 926 | lockid, nodenum, mle->master); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 927 | } |
| 928 | } |
| 929 | |
| 930 | wait: |
| 931 | /* keep going until the response map includes all nodes */ |
| 932 | ret = dlm_wait_for_lock_mastery(dlm, res, mle, &blocked); |
| 933 | if (ret < 0) { |
Kurt Hackel | e7e69eb | 2006-05-01 11:49:52 -0700 | [diff] [blame] | 934 | wait_on_recovery = 1; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 935 | mlog(0, "%s:%.*s: node map changed, redo the " |
| 936 | "master request now, blocked=%d\n", |
| 937 | dlm->name, res->lockname.len, |
| 938 | res->lockname.name, blocked); |
| 939 | if (++tries > 20) { |
| 940 | mlog(ML_ERROR, "%s:%.*s: spinning on " |
| 941 | "dlm_wait_for_lock_mastery, blocked=%d\n", |
| 942 | dlm->name, res->lockname.len, |
| 943 | res->lockname.name, blocked); |
| 944 | dlm_print_one_lock_resource(res); |
Mark Fasheh | 8a9343f | 2006-05-01 14:55:10 -0700 | [diff] [blame] | 945 | dlm_print_one_mle(mle); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 946 | tries = 0; |
| 947 | } |
| 948 | goto redo_request; |
| 949 | } |
| 950 | |
| 951 | mlog(0, "lockres mastered by %u\n", res->owner); |
| 952 | /* make sure we never continue without this */ |
| 953 | BUG_ON(res->owner == O2NM_MAX_NODES); |
| 954 | |
| 955 | /* master is known, detach if not already detached */ |
| 956 | dlm_mle_detach_hb_events(dlm, mle); |
| 957 | dlm_put_mle(mle); |
| 958 | /* put the extra ref */ |
Kurt Hackel | a2bf047 | 2006-04-27 18:51:26 -0700 | [diff] [blame] | 959 | dlm_put_mle_inuse(mle); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 960 | |
| 961 | wake_waiters: |
| 962 | spin_lock(&res->spinlock); |
| 963 | res->state &= ~DLM_LOCK_RES_IN_PROGRESS; |
| 964 | spin_unlock(&res->spinlock); |
| 965 | wake_up(&res->wq); |
| 966 | |
| 967 | leave: |
| 968 | /* need to free the unused mle */ |
| 969 | if (alloc_mle) |
| 970 | kmem_cache_free(dlm_mle_cache, alloc_mle); |
| 971 | |
| 972 | return res; |
| 973 | } |
| 974 | |
| 975 | |
| 976 | #define DLM_MASTERY_TIMEOUT_MS 5000 |
| 977 | |
| 978 | static int dlm_wait_for_lock_mastery(struct dlm_ctxt *dlm, |
| 979 | struct dlm_lock_resource *res, |
| 980 | struct dlm_master_list_entry *mle, |
| 981 | int *blocked) |
| 982 | { |
| 983 | u8 m; |
| 984 | int ret, bit; |
| 985 | int map_changed, voting_done; |
| 986 | int assert, sleep; |
| 987 | |
| 988 | recheck: |
| 989 | ret = 0; |
| 990 | assert = 0; |
| 991 | |
| 992 | /* check if another node has already become the owner */ |
| 993 | spin_lock(&res->spinlock); |
| 994 | if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN) { |
Kurt Hackel | 9c6510a | 2006-03-02 18:09:26 -0800 | [diff] [blame] | 995 | mlog(0, "%s:%.*s: owner is suddenly %u\n", dlm->name, |
| 996 | res->lockname.len, res->lockname.name, res->owner); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 997 | spin_unlock(&res->spinlock); |
Kurt Hackel | 9c6510a | 2006-03-02 18:09:26 -0800 | [diff] [blame] | 998 | /* this will cause the master to re-assert across |
| 999 | * the whole cluster, freeing up mles */ |
Kurt Hackel | 588e009 | 2006-05-01 11:22:06 -0700 | [diff] [blame] | 1000 | if (res->owner != dlm->node_num) { |
| 1001 | ret = dlm_do_master_request(mle, res->owner); |
| 1002 | if (ret < 0) { |
| 1003 | /* give recovery a chance to run */ |
| 1004 | mlog(ML_ERROR, "link to %u went down?: %d\n", res->owner, ret); |
| 1005 | msleep(500); |
| 1006 | goto recheck; |
| 1007 | } |
Kurt Hackel | 9c6510a | 2006-03-02 18:09:26 -0800 | [diff] [blame] | 1008 | } |
| 1009 | ret = 0; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1010 | goto leave; |
| 1011 | } |
| 1012 | spin_unlock(&res->spinlock); |
| 1013 | |
| 1014 | spin_lock(&mle->spinlock); |
| 1015 | m = mle->master; |
| 1016 | map_changed = (memcmp(mle->vote_map, mle->node_map, |
| 1017 | sizeof(mle->vote_map)) != 0); |
| 1018 | voting_done = (memcmp(mle->vote_map, mle->response_map, |
| 1019 | sizeof(mle->vote_map)) == 0); |
| 1020 | |
| 1021 | /* restart if we hit any errors */ |
| 1022 | if (map_changed) { |
| 1023 | int b; |
| 1024 | mlog(0, "%s: %.*s: node map changed, restarting\n", |
| 1025 | dlm->name, res->lockname.len, res->lockname.name); |
| 1026 | ret = dlm_restart_lock_mastery(dlm, res, mle, *blocked); |
| 1027 | b = (mle->type == DLM_MLE_BLOCK); |
| 1028 | if ((*blocked && !b) || (!*blocked && b)) { |
| 1029 | mlog(0, "%s:%.*s: status change: old=%d new=%d\n", |
| 1030 | dlm->name, res->lockname.len, res->lockname.name, |
| 1031 | *blocked, b); |
| 1032 | *blocked = b; |
| 1033 | } |
| 1034 | spin_unlock(&mle->spinlock); |
| 1035 | if (ret < 0) { |
| 1036 | mlog_errno(ret); |
| 1037 | goto leave; |
| 1038 | } |
| 1039 | mlog(0, "%s:%.*s: restart lock mastery succeeded, " |
| 1040 | "rechecking now\n", dlm->name, res->lockname.len, |
| 1041 | res->lockname.name); |
| 1042 | goto recheck; |
Kurt Hackel | aa85235 | 2006-04-27 19:04:49 -0700 | [diff] [blame] | 1043 | } else { |
| 1044 | if (!voting_done) { |
| 1045 | mlog(0, "map not changed and voting not done " |
| 1046 | "for %s:%.*s\n", dlm->name, res->lockname.len, |
| 1047 | res->lockname.name); |
| 1048 | } |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1049 | } |
| 1050 | |
| 1051 | if (m != O2NM_MAX_NODES) { |
| 1052 | /* another node has done an assert! |
| 1053 | * all done! */ |
| 1054 | sleep = 0; |
| 1055 | } else { |
| 1056 | sleep = 1; |
| 1057 | /* have all nodes responded? */ |
| 1058 | if (voting_done && !*blocked) { |
| 1059 | bit = find_next_bit(mle->maybe_map, O2NM_MAX_NODES, 0); |
| 1060 | if (dlm->node_num <= bit) { |
| 1061 | /* my node number is lowest. |
| 1062 | * now tell other nodes that I am |
| 1063 | * mastering this. */ |
| 1064 | mle->master = dlm->node_num; |
| 1065 | assert = 1; |
| 1066 | sleep = 0; |
| 1067 | } |
| 1068 | /* if voting is done, but we have not received |
| 1069 | * an assert master yet, we must sleep */ |
| 1070 | } |
| 1071 | } |
| 1072 | |
| 1073 | spin_unlock(&mle->spinlock); |
| 1074 | |
| 1075 | /* sleep if we haven't finished voting yet */ |
| 1076 | if (sleep) { |
| 1077 | unsigned long timeo = msecs_to_jiffies(DLM_MASTERY_TIMEOUT_MS); |
| 1078 | |
| 1079 | /* |
| 1080 | if (atomic_read(&mle->mle_refs.refcount) < 2) |
| 1081 | mlog(ML_ERROR, "mle (%p) refs=%d, name=%.*s\n", mle, |
| 1082 | atomic_read(&mle->mle_refs.refcount), |
| 1083 | res->lockname.len, res->lockname.name); |
| 1084 | */ |
| 1085 | atomic_set(&mle->woken, 0); |
| 1086 | (void)wait_event_timeout(mle->wq, |
| 1087 | (atomic_read(&mle->woken) == 1), |
| 1088 | timeo); |
| 1089 | if (res->owner == O2NM_MAX_NODES) { |
| 1090 | mlog(0, "waiting again\n"); |
| 1091 | goto recheck; |
| 1092 | } |
| 1093 | mlog(0, "done waiting, master is %u\n", res->owner); |
| 1094 | ret = 0; |
| 1095 | goto leave; |
| 1096 | } |
| 1097 | |
| 1098 | ret = 0; /* done */ |
| 1099 | if (assert) { |
| 1100 | m = dlm->node_num; |
| 1101 | mlog(0, "about to master %.*s here, this=%u\n", |
| 1102 | res->lockname.len, res->lockname.name, m); |
| 1103 | ret = dlm_do_assert_master(dlm, res->lockname.name, |
| 1104 | res->lockname.len, mle->vote_map, 0); |
| 1105 | if (ret) { |
| 1106 | /* This is a failure in the network path, |
| 1107 | * not in the response to the assert_master |
| 1108 | * (any nonzero response is a BUG on this node). |
| 1109 | * Most likely a socket just got disconnected |
| 1110 | * due to node death. */ |
| 1111 | mlog_errno(ret); |
| 1112 | } |
| 1113 | /* no longer need to restart lock mastery. |
| 1114 | * all living nodes have been contacted. */ |
| 1115 | ret = 0; |
| 1116 | } |
| 1117 | |
| 1118 | /* set the lockres owner */ |
| 1119 | spin_lock(&res->spinlock); |
| 1120 | dlm_change_lockres_owner(dlm, res, m); |
| 1121 | spin_unlock(&res->spinlock); |
| 1122 | |
| 1123 | leave: |
| 1124 | return ret; |
| 1125 | } |
| 1126 | |
| 1127 | struct dlm_bitmap_diff_iter |
| 1128 | { |
| 1129 | int curnode; |
| 1130 | unsigned long *orig_bm; |
| 1131 | unsigned long *cur_bm; |
| 1132 | unsigned long diff_bm[BITS_TO_LONGS(O2NM_MAX_NODES)]; |
| 1133 | }; |
| 1134 | |
| 1135 | enum dlm_node_state_change |
| 1136 | { |
| 1137 | NODE_DOWN = -1, |
| 1138 | NODE_NO_CHANGE = 0, |
| 1139 | NODE_UP |
| 1140 | }; |
| 1141 | |
| 1142 | static void dlm_bitmap_diff_iter_init(struct dlm_bitmap_diff_iter *iter, |
| 1143 | unsigned long *orig_bm, |
| 1144 | unsigned long *cur_bm) |
| 1145 | { |
| 1146 | unsigned long p1, p2; |
| 1147 | int i; |
| 1148 | |
| 1149 | iter->curnode = -1; |
| 1150 | iter->orig_bm = orig_bm; |
| 1151 | iter->cur_bm = cur_bm; |
| 1152 | |
| 1153 | for (i = 0; i < BITS_TO_LONGS(O2NM_MAX_NODES); i++) { |
| 1154 | p1 = *(iter->orig_bm + i); |
| 1155 | p2 = *(iter->cur_bm + i); |
| 1156 | iter->diff_bm[i] = (p1 & ~p2) | (p2 & ~p1); |
| 1157 | } |
| 1158 | } |
| 1159 | |
| 1160 | static int dlm_bitmap_diff_iter_next(struct dlm_bitmap_diff_iter *iter, |
| 1161 | enum dlm_node_state_change *state) |
| 1162 | { |
| 1163 | int bit; |
| 1164 | |
| 1165 | if (iter->curnode >= O2NM_MAX_NODES) |
| 1166 | return -ENOENT; |
| 1167 | |
| 1168 | bit = find_next_bit(iter->diff_bm, O2NM_MAX_NODES, |
| 1169 | iter->curnode+1); |
| 1170 | if (bit >= O2NM_MAX_NODES) { |
| 1171 | iter->curnode = O2NM_MAX_NODES; |
| 1172 | return -ENOENT; |
| 1173 | } |
| 1174 | |
| 1175 | /* if it was there in the original then this node died */ |
| 1176 | if (test_bit(bit, iter->orig_bm)) |
| 1177 | *state = NODE_DOWN; |
| 1178 | else |
| 1179 | *state = NODE_UP; |
| 1180 | |
| 1181 | iter->curnode = bit; |
| 1182 | return bit; |
| 1183 | } |
| 1184 | |
| 1185 | |
| 1186 | static int dlm_restart_lock_mastery(struct dlm_ctxt *dlm, |
| 1187 | struct dlm_lock_resource *res, |
| 1188 | struct dlm_master_list_entry *mle, |
| 1189 | int blocked) |
| 1190 | { |
| 1191 | struct dlm_bitmap_diff_iter bdi; |
| 1192 | enum dlm_node_state_change sc; |
| 1193 | int node; |
| 1194 | int ret = 0; |
| 1195 | |
| 1196 | mlog(0, "something happened such that the " |
| 1197 | "master process may need to be restarted!\n"); |
| 1198 | |
| 1199 | assert_spin_locked(&mle->spinlock); |
| 1200 | |
| 1201 | dlm_bitmap_diff_iter_init(&bdi, mle->vote_map, mle->node_map); |
| 1202 | node = dlm_bitmap_diff_iter_next(&bdi, &sc); |
| 1203 | while (node >= 0) { |
| 1204 | if (sc == NODE_UP) { |
Kurt Hackel | e2faea4 | 2006-01-12 14:24:55 -0800 | [diff] [blame] | 1205 | /* a node came up. clear any old vote from |
| 1206 | * the response map and set it in the vote map |
| 1207 | * then restart the mastery. */ |
| 1208 | mlog(ML_NOTICE, "node %d up while restarting\n", node); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1209 | |
| 1210 | /* redo the master request, but only for the new node */ |
| 1211 | mlog(0, "sending request to new node\n"); |
| 1212 | clear_bit(node, mle->response_map); |
| 1213 | set_bit(node, mle->vote_map); |
| 1214 | } else { |
| 1215 | mlog(ML_ERROR, "node down! %d\n", node); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1216 | if (blocked) { |
| 1217 | int lowest = find_next_bit(mle->maybe_map, |
| 1218 | O2NM_MAX_NODES, 0); |
| 1219 | |
| 1220 | /* act like it was never there */ |
| 1221 | clear_bit(node, mle->maybe_map); |
| 1222 | |
Kurt Hackel | e7e69eb | 2006-05-01 11:49:52 -0700 | [diff] [blame] | 1223 | if (node == lowest) { |
| 1224 | mlog(0, "expected master %u died" |
| 1225 | " while this node was blocked " |
| 1226 | "waiting on it!\n", node); |
| 1227 | lowest = find_next_bit(mle->maybe_map, |
| 1228 | O2NM_MAX_NODES, |
| 1229 | lowest+1); |
| 1230 | if (lowest < O2NM_MAX_NODES) { |
| 1231 | mlog(0, "%s:%.*s:still " |
| 1232 | "blocked. waiting on %u " |
| 1233 | "now\n", dlm->name, |
| 1234 | res->lockname.len, |
| 1235 | res->lockname.name, |
| 1236 | lowest); |
| 1237 | } else { |
| 1238 | /* mle is an MLE_BLOCK, but |
| 1239 | * there is now nothing left to |
| 1240 | * block on. we need to return |
| 1241 | * all the way back out and try |
| 1242 | * again with an MLE_MASTER. |
| 1243 | * dlm_do_local_recovery_cleanup |
| 1244 | * has already run, so the mle |
| 1245 | * refcount is ok */ |
| 1246 | mlog(0, "%s:%.*s: no " |
| 1247 | "longer blocking. try to " |
| 1248 | "master this here\n", |
| 1249 | dlm->name, |
| 1250 | res->lockname.len, |
| 1251 | res->lockname.name); |
| 1252 | mle->type = DLM_MLE_MASTER; |
| 1253 | mle->u.res = res; |
| 1254 | } |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1255 | } |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1256 | } |
| 1257 | |
Kurt Hackel | e7e69eb | 2006-05-01 11:49:52 -0700 | [diff] [blame] | 1258 | /* now blank out everything, as if we had never |
| 1259 | * contacted anyone */ |
| 1260 | memset(mle->maybe_map, 0, sizeof(mle->maybe_map)); |
| 1261 | memset(mle->response_map, 0, sizeof(mle->response_map)); |
| 1262 | /* reset the vote_map to the current node_map */ |
| 1263 | memcpy(mle->vote_map, mle->node_map, |
| 1264 | sizeof(mle->node_map)); |
| 1265 | /* put myself into the maybe map */ |
| 1266 | if (mle->type != DLM_MLE_BLOCK) |
| 1267 | set_bit(dlm->node_num, mle->maybe_map); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1268 | } |
| 1269 | ret = -EAGAIN; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1270 | node = dlm_bitmap_diff_iter_next(&bdi, &sc); |
| 1271 | } |
| 1272 | return ret; |
| 1273 | } |
| 1274 | |
| 1275 | |
| 1276 | /* |
| 1277 | * DLM_MASTER_REQUEST_MSG |
| 1278 | * |
| 1279 | * returns: 0 on success, |
| 1280 | * -errno on a network error |
| 1281 | * |
| 1282 | * on error, the caller should assume the target node is "dead" |
| 1283 | * |
| 1284 | */ |
| 1285 | |
| 1286 | static int dlm_do_master_request(struct dlm_master_list_entry *mle, int to) |
| 1287 | { |
| 1288 | struct dlm_ctxt *dlm = mle->dlm; |
| 1289 | struct dlm_master_request request; |
| 1290 | int ret, response=0, resend; |
| 1291 | |
| 1292 | memset(&request, 0, sizeof(request)); |
| 1293 | request.node_idx = dlm->node_num; |
| 1294 | |
| 1295 | BUG_ON(mle->type == DLM_MLE_MIGRATION); |
| 1296 | |
| 1297 | if (mle->type != DLM_MLE_MASTER) { |
| 1298 | request.namelen = mle->u.name.len; |
| 1299 | memcpy(request.name, mle->u.name.name, request.namelen); |
| 1300 | } else { |
| 1301 | request.namelen = mle->u.res->lockname.len; |
| 1302 | memcpy(request.name, mle->u.res->lockname.name, |
| 1303 | request.namelen); |
| 1304 | } |
| 1305 | |
| 1306 | again: |
| 1307 | ret = o2net_send_message(DLM_MASTER_REQUEST_MSG, dlm->key, &request, |
| 1308 | sizeof(request), to, &response); |
| 1309 | if (ret < 0) { |
| 1310 | if (ret == -ESRCH) { |
| 1311 | /* should never happen */ |
| 1312 | mlog(ML_ERROR, "TCP stack not ready!\n"); |
| 1313 | BUG(); |
| 1314 | } else if (ret == -EINVAL) { |
| 1315 | mlog(ML_ERROR, "bad args passed to o2net!\n"); |
| 1316 | BUG(); |
| 1317 | } else if (ret == -ENOMEM) { |
| 1318 | mlog(ML_ERROR, "out of memory while trying to send " |
| 1319 | "network message! retrying\n"); |
| 1320 | /* this is totally crude */ |
| 1321 | msleep(50); |
| 1322 | goto again; |
| 1323 | } else if (!dlm_is_host_down(ret)) { |
| 1324 | /* not a network error. bad. */ |
| 1325 | mlog_errno(ret); |
| 1326 | mlog(ML_ERROR, "unhandled error!"); |
| 1327 | BUG(); |
| 1328 | } |
| 1329 | /* all other errors should be network errors, |
| 1330 | * and likely indicate node death */ |
| 1331 | mlog(ML_ERROR, "link to %d went down!\n", to); |
| 1332 | goto out; |
| 1333 | } |
| 1334 | |
| 1335 | ret = 0; |
| 1336 | resend = 0; |
| 1337 | spin_lock(&mle->spinlock); |
| 1338 | switch (response) { |
| 1339 | case DLM_MASTER_RESP_YES: |
| 1340 | set_bit(to, mle->response_map); |
| 1341 | mlog(0, "node %u is the master, response=YES\n", to); |
| 1342 | mle->master = to; |
| 1343 | break; |
| 1344 | case DLM_MASTER_RESP_NO: |
| 1345 | mlog(0, "node %u not master, response=NO\n", to); |
| 1346 | set_bit(to, mle->response_map); |
| 1347 | break; |
| 1348 | case DLM_MASTER_RESP_MAYBE: |
| 1349 | mlog(0, "node %u not master, response=MAYBE\n", to); |
| 1350 | set_bit(to, mle->response_map); |
| 1351 | set_bit(to, mle->maybe_map); |
| 1352 | break; |
| 1353 | case DLM_MASTER_RESP_ERROR: |
| 1354 | mlog(0, "node %u hit an error, resending\n", to); |
| 1355 | resend = 1; |
| 1356 | response = 0; |
| 1357 | break; |
| 1358 | default: |
| 1359 | mlog(ML_ERROR, "bad response! %u\n", response); |
| 1360 | BUG(); |
| 1361 | } |
| 1362 | spin_unlock(&mle->spinlock); |
| 1363 | if (resend) { |
| 1364 | /* this is also totally crude */ |
| 1365 | msleep(50); |
| 1366 | goto again; |
| 1367 | } |
| 1368 | |
| 1369 | out: |
| 1370 | return ret; |
| 1371 | } |
| 1372 | |
| 1373 | /* |
| 1374 | * locks that can be taken here: |
| 1375 | * dlm->spinlock |
| 1376 | * res->spinlock |
| 1377 | * mle->spinlock |
| 1378 | * dlm->master_list |
| 1379 | * |
| 1380 | * if possible, TRIM THIS DOWN!!! |
| 1381 | */ |
| 1382 | int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data) |
| 1383 | { |
| 1384 | u8 response = DLM_MASTER_RESP_MAYBE; |
| 1385 | struct dlm_ctxt *dlm = data; |
Kurt Hackel | 9c6510a | 2006-03-02 18:09:26 -0800 | [diff] [blame] | 1386 | struct dlm_lock_resource *res = NULL; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1387 | struct dlm_master_request *request = (struct dlm_master_request *) msg->buf; |
| 1388 | struct dlm_master_list_entry *mle = NULL, *tmpmle = NULL; |
| 1389 | char *name; |
Mark Fasheh | a3d3329 | 2006-03-09 17:55:56 -0800 | [diff] [blame] | 1390 | unsigned int namelen, hash; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1391 | int found, ret; |
| 1392 | int set_maybe; |
Kurt Hackel | 9c6510a | 2006-03-02 18:09:26 -0800 | [diff] [blame] | 1393 | int dispatch_assert = 0; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1394 | |
| 1395 | if (!dlm_grab(dlm)) |
| 1396 | return DLM_MASTER_RESP_NO; |
| 1397 | |
| 1398 | if (!dlm_domain_fully_joined(dlm)) { |
| 1399 | response = DLM_MASTER_RESP_NO; |
| 1400 | goto send_response; |
| 1401 | } |
| 1402 | |
| 1403 | name = request->name; |
| 1404 | namelen = request->namelen; |
Mark Fasheh | a3d3329 | 2006-03-09 17:55:56 -0800 | [diff] [blame] | 1405 | hash = dlm_lockid_hash(name, namelen); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1406 | |
| 1407 | if (namelen > DLM_LOCKID_NAME_MAX) { |
| 1408 | response = DLM_IVBUFLEN; |
| 1409 | goto send_response; |
| 1410 | } |
| 1411 | |
| 1412 | way_up_top: |
| 1413 | spin_lock(&dlm->spinlock); |
Mark Fasheh | a3d3329 | 2006-03-09 17:55:56 -0800 | [diff] [blame] | 1414 | res = __dlm_lookup_lockres(dlm, name, namelen, hash); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1415 | if (res) { |
| 1416 | spin_unlock(&dlm->spinlock); |
| 1417 | |
| 1418 | /* take care of the easy cases up front */ |
| 1419 | spin_lock(&res->spinlock); |
| 1420 | if (res->state & DLM_LOCK_RES_RECOVERING) { |
| 1421 | spin_unlock(&res->spinlock); |
| 1422 | mlog(0, "returning DLM_MASTER_RESP_ERROR since res is " |
| 1423 | "being recovered\n"); |
| 1424 | response = DLM_MASTER_RESP_ERROR; |
| 1425 | if (mle) |
| 1426 | kmem_cache_free(dlm_mle_cache, mle); |
| 1427 | goto send_response; |
| 1428 | } |
| 1429 | |
| 1430 | if (res->owner == dlm->node_num) { |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1431 | spin_unlock(&res->spinlock); |
| 1432 | // mlog(0, "this node is the master\n"); |
| 1433 | response = DLM_MASTER_RESP_YES; |
| 1434 | if (mle) |
| 1435 | kmem_cache_free(dlm_mle_cache, mle); |
| 1436 | |
| 1437 | /* this node is the owner. |
| 1438 | * there is some extra work that needs to |
| 1439 | * happen now. the requesting node has |
| 1440 | * caused all nodes up to this one to |
| 1441 | * create mles. this node now needs to |
| 1442 | * go back and clean those up. */ |
Kurt Hackel | 9c6510a | 2006-03-02 18:09:26 -0800 | [diff] [blame] | 1443 | dispatch_assert = 1; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1444 | goto send_response; |
| 1445 | } else if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN) { |
| 1446 | spin_unlock(&res->spinlock); |
| 1447 | // mlog(0, "node %u is the master\n", res->owner); |
| 1448 | response = DLM_MASTER_RESP_NO; |
| 1449 | if (mle) |
| 1450 | kmem_cache_free(dlm_mle_cache, mle); |
| 1451 | goto send_response; |
| 1452 | } |
| 1453 | |
| 1454 | /* ok, there is no owner. either this node is |
| 1455 | * being blocked, or it is actively trying to |
| 1456 | * master this lock. */ |
| 1457 | if (!(res->state & DLM_LOCK_RES_IN_PROGRESS)) { |
| 1458 | mlog(ML_ERROR, "lock with no owner should be " |
| 1459 | "in-progress!\n"); |
| 1460 | BUG(); |
| 1461 | } |
| 1462 | |
| 1463 | // mlog(0, "lockres is in progress...\n"); |
| 1464 | spin_lock(&dlm->master_lock); |
| 1465 | found = dlm_find_mle(dlm, &tmpmle, name, namelen); |
| 1466 | if (!found) { |
| 1467 | mlog(ML_ERROR, "no mle found for this lock!\n"); |
| 1468 | BUG(); |
| 1469 | } |
| 1470 | set_maybe = 1; |
| 1471 | spin_lock(&tmpmle->spinlock); |
| 1472 | if (tmpmle->type == DLM_MLE_BLOCK) { |
| 1473 | // mlog(0, "this node is waiting for " |
| 1474 | // "lockres to be mastered\n"); |
| 1475 | response = DLM_MASTER_RESP_NO; |
| 1476 | } else if (tmpmle->type == DLM_MLE_MIGRATION) { |
| 1477 | mlog(0, "node %u is master, but trying to migrate to " |
| 1478 | "node %u.\n", tmpmle->master, tmpmle->new_master); |
| 1479 | if (tmpmle->master == dlm->node_num) { |
| 1480 | response = DLM_MASTER_RESP_YES; |
| 1481 | mlog(ML_ERROR, "no owner on lockres, but this " |
| 1482 | "node is trying to migrate it to %u?!\n", |
| 1483 | tmpmle->new_master); |
| 1484 | BUG(); |
| 1485 | } else { |
| 1486 | /* the real master can respond on its own */ |
| 1487 | response = DLM_MASTER_RESP_NO; |
| 1488 | } |
| 1489 | } else if (tmpmle->master != DLM_LOCK_RES_OWNER_UNKNOWN) { |
| 1490 | set_maybe = 0; |
Kurt Hackel | 9c6510a | 2006-03-02 18:09:26 -0800 | [diff] [blame] | 1491 | if (tmpmle->master == dlm->node_num) { |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1492 | response = DLM_MASTER_RESP_YES; |
Kurt Hackel | 9c6510a | 2006-03-02 18:09:26 -0800 | [diff] [blame] | 1493 | /* this node will be the owner. |
| 1494 | * go back and clean the mles on any |
| 1495 | * other nodes */ |
| 1496 | dispatch_assert = 1; |
| 1497 | } else |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1498 | response = DLM_MASTER_RESP_NO; |
| 1499 | } else { |
| 1500 | // mlog(0, "this node is attempting to " |
| 1501 | // "master lockres\n"); |
| 1502 | response = DLM_MASTER_RESP_MAYBE; |
| 1503 | } |
| 1504 | if (set_maybe) |
| 1505 | set_bit(request->node_idx, tmpmle->maybe_map); |
| 1506 | spin_unlock(&tmpmle->spinlock); |
| 1507 | |
| 1508 | spin_unlock(&dlm->master_lock); |
| 1509 | spin_unlock(&res->spinlock); |
| 1510 | |
| 1511 | /* keep the mle attached to heartbeat events */ |
| 1512 | dlm_put_mle(tmpmle); |
| 1513 | if (mle) |
| 1514 | kmem_cache_free(dlm_mle_cache, mle); |
| 1515 | goto send_response; |
| 1516 | } |
| 1517 | |
| 1518 | /* |
| 1519 | * lockres doesn't exist on this node |
| 1520 | * if there is an MLE_BLOCK, return NO |
| 1521 | * if there is an MLE_MASTER, return MAYBE |
| 1522 | * otherwise, add an MLE_BLOCK, return NO |
| 1523 | */ |
| 1524 | spin_lock(&dlm->master_lock); |
| 1525 | found = dlm_find_mle(dlm, &tmpmle, name, namelen); |
| 1526 | if (!found) { |
| 1527 | /* this lockid has never been seen on this node yet */ |
| 1528 | // mlog(0, "no mle found\n"); |
| 1529 | if (!mle) { |
| 1530 | spin_unlock(&dlm->master_lock); |
| 1531 | spin_unlock(&dlm->spinlock); |
| 1532 | |
| 1533 | mle = (struct dlm_master_list_entry *) |
Kurt Hackel | ad8100e | 2006-05-01 14:25:21 -0700 | [diff] [blame] | 1534 | kmem_cache_alloc(dlm_mle_cache, GFP_NOFS); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1535 | if (!mle) { |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1536 | response = DLM_MASTER_RESP_ERROR; |
Kurt Hackel | 9c6510a | 2006-03-02 18:09:26 -0800 | [diff] [blame] | 1537 | mlog_errno(-ENOMEM); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1538 | goto send_response; |
| 1539 | } |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1540 | goto way_up_top; |
| 1541 | } |
| 1542 | |
| 1543 | // mlog(0, "this is second time thru, already allocated, " |
| 1544 | // "add the block.\n"); |
Kurt Hackel | 41b8c8a | 2006-04-27 19:00:26 -0700 | [diff] [blame] | 1545 | dlm_init_mle(mle, DLM_MLE_BLOCK, dlm, NULL, name, namelen); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1546 | set_bit(request->node_idx, mle->maybe_map); |
| 1547 | list_add(&mle->list, &dlm->master_list); |
| 1548 | response = DLM_MASTER_RESP_NO; |
| 1549 | } else { |
| 1550 | // mlog(0, "mle was found\n"); |
| 1551 | set_maybe = 1; |
| 1552 | spin_lock(&tmpmle->spinlock); |
Kurt Hackel | 9c6510a | 2006-03-02 18:09:26 -0800 | [diff] [blame] | 1553 | if (tmpmle->master == dlm->node_num) { |
| 1554 | mlog(ML_ERROR, "no lockres, but an mle with this node as master!\n"); |
| 1555 | BUG(); |
| 1556 | } |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1557 | if (tmpmle->type == DLM_MLE_BLOCK) |
| 1558 | response = DLM_MASTER_RESP_NO; |
| 1559 | else if (tmpmle->type == DLM_MLE_MIGRATION) { |
| 1560 | mlog(0, "migration mle was found (%u->%u)\n", |
| 1561 | tmpmle->master, tmpmle->new_master); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1562 | /* real master can respond on its own */ |
| 1563 | response = DLM_MASTER_RESP_NO; |
Kurt Hackel | 9c6510a | 2006-03-02 18:09:26 -0800 | [diff] [blame] | 1564 | } else |
| 1565 | response = DLM_MASTER_RESP_MAYBE; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1566 | if (set_maybe) |
| 1567 | set_bit(request->node_idx, tmpmle->maybe_map); |
| 1568 | spin_unlock(&tmpmle->spinlock); |
| 1569 | } |
| 1570 | spin_unlock(&dlm->master_lock); |
| 1571 | spin_unlock(&dlm->spinlock); |
| 1572 | |
| 1573 | if (found) { |
| 1574 | /* keep the mle attached to heartbeat events */ |
| 1575 | dlm_put_mle(tmpmle); |
| 1576 | } |
| 1577 | send_response: |
Kurt Hackel | 9c6510a | 2006-03-02 18:09:26 -0800 | [diff] [blame] | 1578 | |
| 1579 | if (dispatch_assert) { |
| 1580 | if (response != DLM_MASTER_RESP_YES) |
| 1581 | mlog(ML_ERROR, "invalid response %d\n", response); |
| 1582 | if (!res) { |
| 1583 | mlog(ML_ERROR, "bad lockres while trying to assert!\n"); |
| 1584 | BUG(); |
| 1585 | } |
| 1586 | mlog(0, "%u is the owner of %.*s, cleaning everyone else\n", |
| 1587 | dlm->node_num, res->lockname.len, res->lockname.name); |
| 1588 | ret = dlm_dispatch_assert_master(dlm, res, 0, request->node_idx, |
| 1589 | DLM_ASSERT_MASTER_MLE_CLEANUP); |
| 1590 | if (ret < 0) { |
| 1591 | mlog(ML_ERROR, "failed to dispatch assert master work\n"); |
| 1592 | response = DLM_MASTER_RESP_ERROR; |
| 1593 | } |
| 1594 | } |
| 1595 | |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1596 | dlm_put(dlm); |
| 1597 | return response; |
| 1598 | } |
| 1599 | |
| 1600 | /* |
| 1601 | * DLM_ASSERT_MASTER_MSG |
| 1602 | */ |
| 1603 | |
| 1604 | |
| 1605 | /* |
| 1606 | * NOTE: this can be used for debugging |
| 1607 | * can periodically run all locks owned by this node |
| 1608 | * and re-assert across the cluster... |
| 1609 | */ |
| 1610 | static int dlm_do_assert_master(struct dlm_ctxt *dlm, const char *lockname, |
| 1611 | unsigned int namelen, void *nodemap, |
| 1612 | u32 flags) |
| 1613 | { |
| 1614 | struct dlm_assert_master assert; |
| 1615 | int to, tmpret; |
| 1616 | struct dlm_node_iter iter; |
| 1617 | int ret = 0; |
Kurt Hackel | 9c6510a | 2006-03-02 18:09:26 -0800 | [diff] [blame] | 1618 | int reassert; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1619 | |
| 1620 | BUG_ON(namelen > O2NM_MAX_NAME_LEN); |
Kurt Hackel | 9c6510a | 2006-03-02 18:09:26 -0800 | [diff] [blame] | 1621 | again: |
| 1622 | reassert = 0; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1623 | |
| 1624 | /* note that if this nodemap is empty, it returns 0 */ |
| 1625 | dlm_node_iter_init(nodemap, &iter); |
| 1626 | while ((to = dlm_node_iter_next(&iter)) >= 0) { |
| 1627 | int r = 0; |
Kurt Hackel | a9ee4c8 | 2006-04-27 19:26:15 -0700 | [diff] [blame] | 1628 | struct dlm_master_list_entry *mle = NULL; |
| 1629 | |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1630 | mlog(0, "sending assert master to %d (%.*s)\n", to, |
| 1631 | namelen, lockname); |
| 1632 | memset(&assert, 0, sizeof(assert)); |
| 1633 | assert.node_idx = dlm->node_num; |
| 1634 | assert.namelen = namelen; |
| 1635 | memcpy(assert.name, lockname, namelen); |
| 1636 | assert.flags = cpu_to_be32(flags); |
| 1637 | |
| 1638 | tmpret = o2net_send_message(DLM_ASSERT_MASTER_MSG, dlm->key, |
| 1639 | &assert, sizeof(assert), to, &r); |
| 1640 | if (tmpret < 0) { |
Kurt Hackel | 3b3b84a | 2006-05-01 14:31:37 -0700 | [diff] [blame] | 1641 | mlog(0, "assert_master returned %d!\n", tmpret); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1642 | if (!dlm_is_host_down(tmpret)) { |
Kurt Hackel | 3b3b84a | 2006-05-01 14:31:37 -0700 | [diff] [blame] | 1643 | mlog(ML_ERROR, "unhandled error=%d!\n", tmpret); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1644 | BUG(); |
| 1645 | } |
| 1646 | /* a node died. finish out the rest of the nodes. */ |
Kurt Hackel | 3b3b84a | 2006-05-01 14:31:37 -0700 | [diff] [blame] | 1647 | mlog(0, "link to %d went down!\n", to); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1648 | /* any nonzero status return will do */ |
| 1649 | ret = tmpret; |
| 1650 | } else if (r < 0) { |
| 1651 | /* ok, something horribly messed. kill thyself. */ |
| 1652 | mlog(ML_ERROR,"during assert master of %.*s to %u, " |
| 1653 | "got %d.\n", namelen, lockname, to, r); |
Kurt Hackel | a9ee4c8 | 2006-04-27 19:26:15 -0700 | [diff] [blame] | 1654 | spin_lock(&dlm->spinlock); |
| 1655 | spin_lock(&dlm->master_lock); |
| 1656 | if (dlm_find_mle(dlm, &mle, (char *)lockname, |
| 1657 | namelen)) { |
| 1658 | dlm_print_one_mle(mle); |
| 1659 | __dlm_put_mle(mle); |
| 1660 | } |
| 1661 | spin_unlock(&dlm->master_lock); |
| 1662 | spin_unlock(&dlm->spinlock); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1663 | BUG(); |
Kurt Hackel | 9c6510a | 2006-03-02 18:09:26 -0800 | [diff] [blame] | 1664 | } else if (r == EAGAIN) { |
| 1665 | mlog(0, "%.*s: node %u create mles on other " |
| 1666 | "nodes and requests a re-assert\n", |
| 1667 | namelen, lockname, to); |
| 1668 | reassert = 1; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1669 | } |
| 1670 | } |
| 1671 | |
Kurt Hackel | 9c6510a | 2006-03-02 18:09:26 -0800 | [diff] [blame] | 1672 | if (reassert) |
| 1673 | goto again; |
| 1674 | |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1675 | return ret; |
| 1676 | } |
| 1677 | |
| 1678 | /* |
| 1679 | * locks that can be taken here: |
| 1680 | * dlm->spinlock |
| 1681 | * res->spinlock |
| 1682 | * mle->spinlock |
| 1683 | * dlm->master_list |
| 1684 | * |
| 1685 | * if possible, TRIM THIS DOWN!!! |
| 1686 | */ |
| 1687 | int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data) |
| 1688 | { |
| 1689 | struct dlm_ctxt *dlm = data; |
| 1690 | struct dlm_master_list_entry *mle = NULL; |
| 1691 | struct dlm_assert_master *assert = (struct dlm_assert_master *)msg->buf; |
| 1692 | struct dlm_lock_resource *res = NULL; |
| 1693 | char *name; |
Mark Fasheh | a3d3329 | 2006-03-09 17:55:56 -0800 | [diff] [blame] | 1694 | unsigned int namelen, hash; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1695 | u32 flags; |
Kurt Hackel | 9c6510a | 2006-03-02 18:09:26 -0800 | [diff] [blame] | 1696 | int master_request = 0; |
| 1697 | int ret = 0; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1698 | |
| 1699 | if (!dlm_grab(dlm)) |
| 1700 | return 0; |
| 1701 | |
| 1702 | name = assert->name; |
| 1703 | namelen = assert->namelen; |
Mark Fasheh | a3d3329 | 2006-03-09 17:55:56 -0800 | [diff] [blame] | 1704 | hash = dlm_lockid_hash(name, namelen); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1705 | flags = be32_to_cpu(assert->flags); |
| 1706 | |
| 1707 | if (namelen > DLM_LOCKID_NAME_MAX) { |
| 1708 | mlog(ML_ERROR, "Invalid name length!"); |
| 1709 | goto done; |
| 1710 | } |
| 1711 | |
| 1712 | spin_lock(&dlm->spinlock); |
| 1713 | |
| 1714 | if (flags) |
| 1715 | mlog(0, "assert_master with flags: %u\n", flags); |
| 1716 | |
| 1717 | /* find the MLE */ |
| 1718 | spin_lock(&dlm->master_lock); |
| 1719 | if (!dlm_find_mle(dlm, &mle, name, namelen)) { |
| 1720 | /* not an error, could be master just re-asserting */ |
| 1721 | mlog(0, "just got an assert_master from %u, but no " |
| 1722 | "MLE for it! (%.*s)\n", assert->node_idx, |
| 1723 | namelen, name); |
| 1724 | } else { |
| 1725 | int bit = find_next_bit (mle->maybe_map, O2NM_MAX_NODES, 0); |
| 1726 | if (bit >= O2NM_MAX_NODES) { |
| 1727 | /* not necessarily an error, though less likely. |
| 1728 | * could be master just re-asserting. */ |
Kurt Hackel | aa85235 | 2006-04-27 19:04:49 -0700 | [diff] [blame] | 1729 | mlog(0, "no bits set in the maybe_map, but %u " |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1730 | "is asserting! (%.*s)\n", assert->node_idx, |
| 1731 | namelen, name); |
| 1732 | } else if (bit != assert->node_idx) { |
| 1733 | if (flags & DLM_ASSERT_MASTER_MLE_CLEANUP) { |
| 1734 | mlog(0, "master %u was found, %u should " |
| 1735 | "back off\n", assert->node_idx, bit); |
| 1736 | } else { |
| 1737 | /* with the fix for bug 569, a higher node |
| 1738 | * number winning the mastery will respond |
| 1739 | * YES to mastery requests, but this node |
| 1740 | * had no way of knowing. let it pass. */ |
Kurt Hackel | aa85235 | 2006-04-27 19:04:49 -0700 | [diff] [blame] | 1741 | mlog(0, "%u is the lowest node, " |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1742 | "%u is asserting. (%.*s) %u must " |
| 1743 | "have begun after %u won.\n", bit, |
| 1744 | assert->node_idx, namelen, name, bit, |
| 1745 | assert->node_idx); |
| 1746 | } |
| 1747 | } |
Kurt Hackel | 2d1a868 | 2006-04-27 19:01:35 -0700 | [diff] [blame] | 1748 | if (mle->type == DLM_MLE_MIGRATION) { |
| 1749 | if (flags & DLM_ASSERT_MASTER_MLE_CLEANUP) { |
| 1750 | mlog(0, "%s:%.*s: got cleanup assert" |
| 1751 | " from %u for migration\n", |
| 1752 | dlm->name, namelen, name, |
| 1753 | assert->node_idx); |
| 1754 | } else if (!(flags & DLM_ASSERT_MASTER_FINISH_MIGRATION)) { |
| 1755 | mlog(0, "%s:%.*s: got unrelated assert" |
| 1756 | " from %u for migration, ignoring\n", |
| 1757 | dlm->name, namelen, name, |
| 1758 | assert->node_idx); |
| 1759 | __dlm_put_mle(mle); |
| 1760 | spin_unlock(&dlm->master_lock); |
| 1761 | spin_unlock(&dlm->spinlock); |
| 1762 | goto done; |
| 1763 | } |
| 1764 | } |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1765 | } |
| 1766 | spin_unlock(&dlm->master_lock); |
| 1767 | |
| 1768 | /* ok everything checks out with the MLE |
| 1769 | * now check to see if there is a lockres */ |
Mark Fasheh | a3d3329 | 2006-03-09 17:55:56 -0800 | [diff] [blame] | 1770 | res = __dlm_lookup_lockres(dlm, name, namelen, hash); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1771 | if (res) { |
| 1772 | spin_lock(&res->spinlock); |
| 1773 | if (res->state & DLM_LOCK_RES_RECOVERING) { |
| 1774 | mlog(ML_ERROR, "%u asserting but %.*s is " |
| 1775 | "RECOVERING!\n", assert->node_idx, namelen, name); |
| 1776 | goto kill; |
| 1777 | } |
| 1778 | if (!mle) { |
Kurt Hackel | dc2ed19 | 2006-04-27 19:03:18 -0700 | [diff] [blame] | 1779 | if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN && |
| 1780 | res->owner != assert->node_idx) { |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1781 | mlog(ML_ERROR, "assert_master from " |
| 1782 | "%u, but current owner is " |
| 1783 | "%u! (%.*s)\n", |
| 1784 | assert->node_idx, res->owner, |
| 1785 | namelen, name); |
| 1786 | goto kill; |
| 1787 | } |
| 1788 | } else if (mle->type != DLM_MLE_MIGRATION) { |
| 1789 | if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN) { |
| 1790 | /* owner is just re-asserting */ |
| 1791 | if (res->owner == assert->node_idx) { |
| 1792 | mlog(0, "owner %u re-asserting on " |
| 1793 | "lock %.*s\n", assert->node_idx, |
| 1794 | namelen, name); |
| 1795 | goto ok; |
| 1796 | } |
| 1797 | mlog(ML_ERROR, "got assert_master from " |
| 1798 | "node %u, but %u is the owner! " |
| 1799 | "(%.*s)\n", assert->node_idx, |
| 1800 | res->owner, namelen, name); |
| 1801 | goto kill; |
| 1802 | } |
| 1803 | if (!(res->state & DLM_LOCK_RES_IN_PROGRESS)) { |
| 1804 | mlog(ML_ERROR, "got assert from %u, but lock " |
| 1805 | "with no owner should be " |
| 1806 | "in-progress! (%.*s)\n", |
| 1807 | assert->node_idx, |
| 1808 | namelen, name); |
| 1809 | goto kill; |
| 1810 | } |
| 1811 | } else /* mle->type == DLM_MLE_MIGRATION */ { |
| 1812 | /* should only be getting an assert from new master */ |
| 1813 | if (assert->node_idx != mle->new_master) { |
| 1814 | mlog(ML_ERROR, "got assert from %u, but " |
| 1815 | "new master is %u, and old master " |
| 1816 | "was %u (%.*s)\n", |
| 1817 | assert->node_idx, mle->new_master, |
| 1818 | mle->master, namelen, name); |
| 1819 | goto kill; |
| 1820 | } |
| 1821 | |
| 1822 | } |
| 1823 | ok: |
| 1824 | spin_unlock(&res->spinlock); |
| 1825 | } |
| 1826 | spin_unlock(&dlm->spinlock); |
| 1827 | |
| 1828 | // mlog(0, "woo! got an assert_master from node %u!\n", |
| 1829 | // assert->node_idx); |
| 1830 | if (mle) { |
Kurt Hackel | 9c6510a | 2006-03-02 18:09:26 -0800 | [diff] [blame] | 1831 | int extra_ref = 0; |
| 1832 | int nn = -1; |
Kurt Hackel | a2bf047 | 2006-04-27 18:51:26 -0700 | [diff] [blame] | 1833 | int rr, err = 0; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1834 | |
| 1835 | spin_lock(&mle->spinlock); |
Kurt Hackel | 9c6510a | 2006-03-02 18:09:26 -0800 | [diff] [blame] | 1836 | if (mle->type == DLM_MLE_BLOCK || mle->type == DLM_MLE_MIGRATION) |
| 1837 | extra_ref = 1; |
| 1838 | else { |
| 1839 | /* MASTER mle: if any bits set in the response map |
| 1840 | * then the calling node needs to re-assert to clear |
| 1841 | * up nodes that this node contacted */ |
| 1842 | while ((nn = find_next_bit (mle->response_map, O2NM_MAX_NODES, |
| 1843 | nn+1)) < O2NM_MAX_NODES) { |
| 1844 | if (nn != dlm->node_num && nn != assert->node_idx) |
| 1845 | master_request = 1; |
| 1846 | } |
| 1847 | } |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1848 | mle->master = assert->node_idx; |
| 1849 | atomic_set(&mle->woken, 1); |
| 1850 | wake_up(&mle->wq); |
| 1851 | spin_unlock(&mle->spinlock); |
| 1852 | |
Kurt Hackel | a2bf047 | 2006-04-27 18:51:26 -0700 | [diff] [blame] | 1853 | if (res) { |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1854 | spin_lock(&res->spinlock); |
Kurt Hackel | a2bf047 | 2006-04-27 18:51:26 -0700 | [diff] [blame] | 1855 | if (mle->type == DLM_MLE_MIGRATION) { |
| 1856 | mlog(0, "finishing off migration of lockres %.*s, " |
| 1857 | "from %u to %u\n", |
| 1858 | res->lockname.len, res->lockname.name, |
| 1859 | dlm->node_num, mle->new_master); |
| 1860 | res->state &= ~DLM_LOCK_RES_MIGRATING; |
| 1861 | dlm_change_lockres_owner(dlm, res, mle->new_master); |
| 1862 | BUG_ON(res->state & DLM_LOCK_RES_DIRTY); |
| 1863 | } else { |
| 1864 | dlm_change_lockres_owner(dlm, res, mle->master); |
| 1865 | } |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1866 | spin_unlock(&res->spinlock); |
| 1867 | } |
Kurt Hackel | a2bf047 | 2006-04-27 18:51:26 -0700 | [diff] [blame] | 1868 | |
| 1869 | /* master is known, detach if not already detached. |
| 1870 | * ensures that only one assert_master call will happen |
| 1871 | * on this mle. */ |
| 1872 | spin_lock(&dlm->spinlock); |
| 1873 | spin_lock(&dlm->master_lock); |
| 1874 | |
| 1875 | rr = atomic_read(&mle->mle_refs.refcount); |
| 1876 | if (mle->inuse > 0) { |
| 1877 | if (extra_ref && rr < 3) |
| 1878 | err = 1; |
| 1879 | else if (!extra_ref && rr < 2) |
| 1880 | err = 1; |
| 1881 | } else { |
| 1882 | if (extra_ref && rr < 2) |
| 1883 | err = 1; |
| 1884 | else if (!extra_ref && rr < 1) |
| 1885 | err = 1; |
| 1886 | } |
| 1887 | if (err) { |
| 1888 | mlog(ML_ERROR, "%s:%.*s: got assert master from %u " |
| 1889 | "that will mess up this node, refs=%d, extra=%d, " |
| 1890 | "inuse=%d\n", dlm->name, namelen, name, |
| 1891 | assert->node_idx, rr, extra_ref, mle->inuse); |
| 1892 | dlm_print_one_mle(mle); |
| 1893 | } |
| 1894 | list_del_init(&mle->list); |
| 1895 | __dlm_mle_detach_hb_events(dlm, mle); |
| 1896 | __dlm_put_mle(mle); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1897 | if (extra_ref) { |
| 1898 | /* the assert master message now balances the extra |
| 1899 | * ref given by the master / migration request message. |
| 1900 | * if this is the last put, it will be removed |
| 1901 | * from the list. */ |
Kurt Hackel | a2bf047 | 2006-04-27 18:51:26 -0700 | [diff] [blame] | 1902 | __dlm_put_mle(mle); |
| 1903 | } |
| 1904 | spin_unlock(&dlm->master_lock); |
| 1905 | spin_unlock(&dlm->spinlock); |
| 1906 | } else if (res) { |
| 1907 | if (res->owner != assert->node_idx) { |
| 1908 | mlog(0, "assert_master from %u, but current " |
| 1909 | "owner is %u (%.*s), no mle\n", assert->node_idx, |
| 1910 | res->owner, namelen, name); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1911 | } |
| 1912 | } |
| 1913 | |
| 1914 | done: |
Kurt Hackel | 9c6510a | 2006-03-02 18:09:26 -0800 | [diff] [blame] | 1915 | ret = 0; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1916 | if (res) |
| 1917 | dlm_lockres_put(res); |
| 1918 | dlm_put(dlm); |
Kurt Hackel | 9c6510a | 2006-03-02 18:09:26 -0800 | [diff] [blame] | 1919 | if (master_request) { |
| 1920 | mlog(0, "need to tell master to reassert\n"); |
| 1921 | ret = EAGAIN; // positive. negative would shoot down the node. |
| 1922 | } |
| 1923 | return ret; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1924 | |
| 1925 | kill: |
| 1926 | /* kill the caller! */ |
Kurt Hackel | a9ee4c8 | 2006-04-27 19:26:15 -0700 | [diff] [blame] | 1927 | mlog(ML_ERROR, "Bad message received from another node. Dumping state " |
| 1928 | "and killing the other node now! This node is OK and can continue.\n"); |
| 1929 | __dlm_print_one_lock_resource(res); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1930 | spin_unlock(&res->spinlock); |
| 1931 | spin_unlock(&dlm->spinlock); |
| 1932 | dlm_lockres_put(res); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1933 | dlm_put(dlm); |
| 1934 | return -EINVAL; |
| 1935 | } |
| 1936 | |
| 1937 | int dlm_dispatch_assert_master(struct dlm_ctxt *dlm, |
| 1938 | struct dlm_lock_resource *res, |
| 1939 | int ignore_higher, u8 request_from, u32 flags) |
| 1940 | { |
| 1941 | struct dlm_work_item *item; |
Kurt Hackel | ad8100e | 2006-05-01 14:25:21 -0700 | [diff] [blame] | 1942 | item = kcalloc(1, sizeof(*item), GFP_NOFS); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1943 | if (!item) |
| 1944 | return -ENOMEM; |
| 1945 | |
| 1946 | |
| 1947 | /* queue up work for dlm_assert_master_worker */ |
| 1948 | dlm_grab(dlm); /* get an extra ref for the work item */ |
| 1949 | dlm_init_work_item(dlm, item, dlm_assert_master_worker, NULL); |
| 1950 | item->u.am.lockres = res; /* already have a ref */ |
| 1951 | /* can optionally ignore node numbers higher than this node */ |
| 1952 | item->u.am.ignore_higher = ignore_higher; |
| 1953 | item->u.am.request_from = request_from; |
| 1954 | item->u.am.flags = flags; |
| 1955 | |
Kurt Hackel | 9c6510a | 2006-03-02 18:09:26 -0800 | [diff] [blame] | 1956 | if (ignore_higher) |
| 1957 | mlog(0, "IGNORE HIGHER: %.*s\n", res->lockname.len, |
| 1958 | res->lockname.name); |
| 1959 | |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1960 | spin_lock(&dlm->work_lock); |
| 1961 | list_add_tail(&item->list, &dlm->work_list); |
| 1962 | spin_unlock(&dlm->work_lock); |
| 1963 | |
Kurt Hackel | 3156d26 | 2006-05-01 14:39:29 -0700 | [diff] [blame] | 1964 | queue_work(dlm->dlm_worker, &dlm->dispatched_work); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1965 | return 0; |
| 1966 | } |
| 1967 | |
| 1968 | static void dlm_assert_master_worker(struct dlm_work_item *item, void *data) |
| 1969 | { |
| 1970 | struct dlm_ctxt *dlm = data; |
| 1971 | int ret = 0; |
| 1972 | struct dlm_lock_resource *res; |
| 1973 | unsigned long nodemap[BITS_TO_LONGS(O2NM_MAX_NODES)]; |
| 1974 | int ignore_higher; |
| 1975 | int bit; |
| 1976 | u8 request_from; |
| 1977 | u32 flags; |
| 1978 | |
| 1979 | dlm = item->dlm; |
| 1980 | res = item->u.am.lockres; |
| 1981 | ignore_higher = item->u.am.ignore_higher; |
| 1982 | request_from = item->u.am.request_from; |
| 1983 | flags = item->u.am.flags; |
| 1984 | |
| 1985 | spin_lock(&dlm->spinlock); |
| 1986 | memcpy(nodemap, dlm->domain_map, sizeof(nodemap)); |
| 1987 | spin_unlock(&dlm->spinlock); |
| 1988 | |
| 1989 | clear_bit(dlm->node_num, nodemap); |
| 1990 | if (ignore_higher) { |
| 1991 | /* if is this just to clear up mles for nodes below |
| 1992 | * this node, do not send the message to the original |
| 1993 | * caller or any node number higher than this */ |
| 1994 | clear_bit(request_from, nodemap); |
| 1995 | bit = dlm->node_num; |
| 1996 | while (1) { |
| 1997 | bit = find_next_bit(nodemap, O2NM_MAX_NODES, |
| 1998 | bit+1); |
| 1999 | if (bit >= O2NM_MAX_NODES) |
| 2000 | break; |
| 2001 | clear_bit(bit, nodemap); |
| 2002 | } |
| 2003 | } |
| 2004 | |
Kurt Hackel | 3640748 | 2006-05-01 13:32:27 -0700 | [diff] [blame] | 2005 | /* |
| 2006 | * If we're migrating this lock to someone else, we are no |
| 2007 | * longer allowed to assert out own mastery. OTOH, we need to |
| 2008 | * prevent migration from starting while we're still asserting |
| 2009 | * our dominance. The reserved ast delays migration. |
| 2010 | */ |
| 2011 | spin_lock(&res->spinlock); |
| 2012 | if (res->state & DLM_LOCK_RES_MIGRATING) { |
| 2013 | mlog(0, "Someone asked us to assert mastery, but we're " |
| 2014 | "in the middle of migration. Skipping assert, " |
| 2015 | "the new master will handle that.\n"); |
| 2016 | spin_unlock(&res->spinlock); |
| 2017 | goto put; |
| 2018 | } else |
| 2019 | __dlm_lockres_reserve_ast(res); |
| 2020 | spin_unlock(&res->spinlock); |
| 2021 | |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2022 | /* this call now finishes out the nodemap |
| 2023 | * even if one or more nodes die */ |
| 2024 | mlog(0, "worker about to master %.*s here, this=%u\n", |
| 2025 | res->lockname.len, res->lockname.name, dlm->node_num); |
| 2026 | ret = dlm_do_assert_master(dlm, res->lockname.name, |
| 2027 | res->lockname.len, |
| 2028 | nodemap, flags); |
| 2029 | if (ret < 0) { |
| 2030 | /* no need to restart, we are done */ |
Kurt Hackel | 3b3b84a | 2006-05-01 14:31:37 -0700 | [diff] [blame] | 2031 | if (!dlm_is_host_down(ret)) |
| 2032 | mlog_errno(ret); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2033 | } |
| 2034 | |
Kurt Hackel | 3640748 | 2006-05-01 13:32:27 -0700 | [diff] [blame] | 2035 | /* Ok, we've asserted ourselves. Let's let migration start. */ |
| 2036 | dlm_lockres_release_ast(dlm, res); |
| 2037 | |
| 2038 | put: |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2039 | dlm_lockres_put(res); |
| 2040 | |
| 2041 | mlog(0, "finished with dlm_assert_master_worker\n"); |
| 2042 | } |
| 2043 | |
Kurt Hackel | c03872f | 2006-03-06 14:08:49 -0800 | [diff] [blame] | 2044 | /* SPECIAL CASE for the $RECOVERY lock used by the recovery thread. |
| 2045 | * We cannot wait for node recovery to complete to begin mastering this |
| 2046 | * lockres because this lockres is used to kick off recovery! ;-) |
| 2047 | * So, do a pre-check on all living nodes to see if any of those nodes |
| 2048 | * think that $RECOVERY is currently mastered by a dead node. If so, |
| 2049 | * we wait a short time to allow that node to get notified by its own |
| 2050 | * heartbeat stack, then check again. All $RECOVERY lock resources |
| 2051 | * mastered by dead nodes are purged when the hearbeat callback is |
| 2052 | * fired, so we can know for sure that it is safe to continue once |
| 2053 | * the node returns a live node or no node. */ |
| 2054 | static int dlm_pre_master_reco_lockres(struct dlm_ctxt *dlm, |
| 2055 | struct dlm_lock_resource *res) |
| 2056 | { |
| 2057 | struct dlm_node_iter iter; |
| 2058 | int nodenum; |
| 2059 | int ret = 0; |
| 2060 | u8 master = DLM_LOCK_RES_OWNER_UNKNOWN; |
| 2061 | |
| 2062 | spin_lock(&dlm->spinlock); |
| 2063 | dlm_node_iter_init(dlm->domain_map, &iter); |
| 2064 | spin_unlock(&dlm->spinlock); |
| 2065 | |
| 2066 | while ((nodenum = dlm_node_iter_next(&iter)) >= 0) { |
| 2067 | /* do not send to self */ |
| 2068 | if (nodenum == dlm->node_num) |
| 2069 | continue; |
| 2070 | ret = dlm_do_master_requery(dlm, res, nodenum, &master); |
| 2071 | if (ret < 0) { |
| 2072 | mlog_errno(ret); |
| 2073 | if (!dlm_is_host_down(ret)) |
| 2074 | BUG(); |
| 2075 | /* host is down, so answer for that node would be |
| 2076 | * DLM_LOCK_RES_OWNER_UNKNOWN. continue. */ |
Kurt Hackel | f42a100 | 2006-05-01 11:53:33 -0700 | [diff] [blame] | 2077 | ret = 0; |
Kurt Hackel | c03872f | 2006-03-06 14:08:49 -0800 | [diff] [blame] | 2078 | } |
| 2079 | |
| 2080 | if (master != DLM_LOCK_RES_OWNER_UNKNOWN) { |
| 2081 | /* check to see if this master is in the recovery map */ |
| 2082 | spin_lock(&dlm->spinlock); |
| 2083 | if (test_bit(master, dlm->recovery_map)) { |
| 2084 | mlog(ML_NOTICE, "%s: node %u has not seen " |
| 2085 | "node %u go down yet, and thinks the " |
| 2086 | "dead node is mastering the recovery " |
| 2087 | "lock. must wait.\n", dlm->name, |
| 2088 | nodenum, master); |
| 2089 | ret = -EAGAIN; |
| 2090 | } |
| 2091 | spin_unlock(&dlm->spinlock); |
| 2092 | mlog(0, "%s: reco lock master is %u\n", dlm->name, |
| 2093 | master); |
| 2094 | break; |
| 2095 | } |
| 2096 | } |
| 2097 | return ret; |
| 2098 | } |
| 2099 | |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2100 | |
| 2101 | /* |
| 2102 | * DLM_MIGRATE_LOCKRES |
| 2103 | */ |
| 2104 | |
| 2105 | |
| 2106 | int dlm_migrate_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, |
| 2107 | u8 target) |
| 2108 | { |
| 2109 | struct dlm_master_list_entry *mle = NULL; |
| 2110 | struct dlm_master_list_entry *oldmle = NULL; |
| 2111 | struct dlm_migratable_lockres *mres = NULL; |
| 2112 | int ret = -EINVAL; |
| 2113 | const char *name; |
| 2114 | unsigned int namelen; |
| 2115 | int mle_added = 0; |
| 2116 | struct list_head *queue, *iter; |
| 2117 | int i; |
| 2118 | struct dlm_lock *lock; |
| 2119 | int empty = 1; |
| 2120 | |
| 2121 | if (!dlm_grab(dlm)) |
| 2122 | return -EINVAL; |
| 2123 | |
| 2124 | name = res->lockname.name; |
| 2125 | namelen = res->lockname.len; |
| 2126 | |
| 2127 | mlog(0, "migrating %.*s to %u\n", namelen, name, target); |
| 2128 | |
| 2129 | /* |
| 2130 | * ensure this lockres is a proper candidate for migration |
| 2131 | */ |
| 2132 | spin_lock(&res->spinlock); |
| 2133 | if (res->owner == DLM_LOCK_RES_OWNER_UNKNOWN) { |
| 2134 | mlog(0, "cannot migrate lockres with unknown owner!\n"); |
| 2135 | spin_unlock(&res->spinlock); |
| 2136 | goto leave; |
| 2137 | } |
| 2138 | if (res->owner != dlm->node_num) { |
| 2139 | mlog(0, "cannot migrate lockres this node doesn't own!\n"); |
| 2140 | spin_unlock(&res->spinlock); |
| 2141 | goto leave; |
| 2142 | } |
| 2143 | mlog(0, "checking queues...\n"); |
| 2144 | queue = &res->granted; |
| 2145 | for (i=0; i<3; i++) { |
| 2146 | list_for_each(iter, queue) { |
| 2147 | lock = list_entry (iter, struct dlm_lock, list); |
| 2148 | empty = 0; |
| 2149 | if (lock->ml.node == dlm->node_num) { |
| 2150 | mlog(0, "found a lock owned by this node " |
| 2151 | "still on the %s queue! will not " |
| 2152 | "migrate this lockres\n", |
| 2153 | i==0 ? "granted" : |
| 2154 | (i==1 ? "converting" : "blocked")); |
| 2155 | spin_unlock(&res->spinlock); |
| 2156 | ret = -ENOTEMPTY; |
| 2157 | goto leave; |
| 2158 | } |
| 2159 | } |
| 2160 | queue++; |
| 2161 | } |
| 2162 | mlog(0, "all locks on this lockres are nonlocal. continuing\n"); |
| 2163 | spin_unlock(&res->spinlock); |
| 2164 | |
| 2165 | /* no work to do */ |
| 2166 | if (empty) { |
| 2167 | mlog(0, "no locks were found on this lockres! done!\n"); |
| 2168 | ret = 0; |
| 2169 | goto leave; |
| 2170 | } |
| 2171 | |
| 2172 | /* |
| 2173 | * preallocate up front |
| 2174 | * if this fails, abort |
| 2175 | */ |
| 2176 | |
| 2177 | ret = -ENOMEM; |
Kurt Hackel | ad8100e | 2006-05-01 14:25:21 -0700 | [diff] [blame] | 2178 | mres = (struct dlm_migratable_lockres *) __get_free_page(GFP_NOFS); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2179 | if (!mres) { |
| 2180 | mlog_errno(ret); |
| 2181 | goto leave; |
| 2182 | } |
| 2183 | |
| 2184 | mle = (struct dlm_master_list_entry *) kmem_cache_alloc(dlm_mle_cache, |
Kurt Hackel | ad8100e | 2006-05-01 14:25:21 -0700 | [diff] [blame] | 2185 | GFP_NOFS); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2186 | if (!mle) { |
| 2187 | mlog_errno(ret); |
| 2188 | goto leave; |
| 2189 | } |
| 2190 | ret = 0; |
| 2191 | |
| 2192 | /* |
| 2193 | * find a node to migrate the lockres to |
| 2194 | */ |
| 2195 | |
| 2196 | mlog(0, "picking a migration node\n"); |
| 2197 | spin_lock(&dlm->spinlock); |
| 2198 | /* pick a new node */ |
| 2199 | if (!test_bit(target, dlm->domain_map) || |
| 2200 | target >= O2NM_MAX_NODES) { |
| 2201 | target = dlm_pick_migration_target(dlm, res); |
| 2202 | } |
| 2203 | mlog(0, "node %u chosen for migration\n", target); |
| 2204 | |
| 2205 | if (target >= O2NM_MAX_NODES || |
| 2206 | !test_bit(target, dlm->domain_map)) { |
| 2207 | /* target chosen is not alive */ |
| 2208 | ret = -EINVAL; |
| 2209 | } |
| 2210 | |
| 2211 | if (ret) { |
| 2212 | spin_unlock(&dlm->spinlock); |
| 2213 | goto fail; |
| 2214 | } |
| 2215 | |
| 2216 | mlog(0, "continuing with target = %u\n", target); |
| 2217 | |
| 2218 | /* |
| 2219 | * clear any existing master requests and |
| 2220 | * add the migration mle to the list |
| 2221 | */ |
| 2222 | spin_lock(&dlm->master_lock); |
| 2223 | ret = dlm_add_migration_mle(dlm, res, mle, &oldmle, name, |
| 2224 | namelen, target, dlm->node_num); |
| 2225 | spin_unlock(&dlm->master_lock); |
| 2226 | spin_unlock(&dlm->spinlock); |
| 2227 | |
| 2228 | if (ret == -EEXIST) { |
| 2229 | mlog(0, "another process is already migrating it\n"); |
| 2230 | goto fail; |
| 2231 | } |
| 2232 | mle_added = 1; |
| 2233 | |
| 2234 | /* |
| 2235 | * set the MIGRATING flag and flush asts |
| 2236 | * if we fail after this we need to re-dirty the lockres |
| 2237 | */ |
| 2238 | if (dlm_mark_lockres_migrating(dlm, res, target) < 0) { |
| 2239 | mlog(ML_ERROR, "tried to migrate %.*s to %u, but " |
| 2240 | "the target went down.\n", res->lockname.len, |
| 2241 | res->lockname.name, target); |
| 2242 | spin_lock(&res->spinlock); |
| 2243 | res->state &= ~DLM_LOCK_RES_MIGRATING; |
| 2244 | spin_unlock(&res->spinlock); |
| 2245 | ret = -EINVAL; |
| 2246 | } |
| 2247 | |
| 2248 | fail: |
| 2249 | if (oldmle) { |
| 2250 | /* master is known, detach if not already detached */ |
| 2251 | dlm_mle_detach_hb_events(dlm, oldmle); |
| 2252 | dlm_put_mle(oldmle); |
| 2253 | } |
| 2254 | |
| 2255 | if (ret < 0) { |
| 2256 | if (mle_added) { |
| 2257 | dlm_mle_detach_hb_events(dlm, mle); |
| 2258 | dlm_put_mle(mle); |
| 2259 | } else if (mle) { |
| 2260 | kmem_cache_free(dlm_mle_cache, mle); |
| 2261 | } |
| 2262 | goto leave; |
| 2263 | } |
| 2264 | |
| 2265 | /* |
| 2266 | * at this point, we have a migration target, an mle |
| 2267 | * in the master list, and the MIGRATING flag set on |
| 2268 | * the lockres |
| 2269 | */ |
| 2270 | |
| 2271 | |
| 2272 | /* get an extra reference on the mle. |
| 2273 | * otherwise the assert_master from the new |
| 2274 | * master will destroy this. |
| 2275 | * also, make sure that all callers of dlm_get_mle |
| 2276 | * take both dlm->spinlock and dlm->master_lock */ |
| 2277 | spin_lock(&dlm->spinlock); |
| 2278 | spin_lock(&dlm->master_lock); |
Kurt Hackel | a2bf047 | 2006-04-27 18:51:26 -0700 | [diff] [blame] | 2279 | dlm_get_mle_inuse(mle); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2280 | spin_unlock(&dlm->master_lock); |
| 2281 | spin_unlock(&dlm->spinlock); |
| 2282 | |
| 2283 | /* notify new node and send all lock state */ |
| 2284 | /* call send_one_lockres with migration flag. |
| 2285 | * this serves as notice to the target node that a |
| 2286 | * migration is starting. */ |
| 2287 | ret = dlm_send_one_lockres(dlm, res, mres, target, |
| 2288 | DLM_MRES_MIGRATION); |
| 2289 | |
| 2290 | if (ret < 0) { |
| 2291 | mlog(0, "migration to node %u failed with %d\n", |
| 2292 | target, ret); |
| 2293 | /* migration failed, detach and clean up mle */ |
| 2294 | dlm_mle_detach_hb_events(dlm, mle); |
| 2295 | dlm_put_mle(mle); |
Kurt Hackel | a2bf047 | 2006-04-27 18:51:26 -0700 | [diff] [blame] | 2296 | dlm_put_mle_inuse(mle); |
| 2297 | spin_lock(&res->spinlock); |
| 2298 | res->state &= ~DLM_LOCK_RES_MIGRATING; |
| 2299 | spin_unlock(&res->spinlock); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2300 | goto leave; |
| 2301 | } |
| 2302 | |
| 2303 | /* at this point, the target sends a message to all nodes, |
| 2304 | * (using dlm_do_migrate_request). this node is skipped since |
| 2305 | * we had to put an mle in the list to begin the process. this |
| 2306 | * node now waits for target to do an assert master. this node |
| 2307 | * will be the last one notified, ensuring that the migration |
| 2308 | * is complete everywhere. if the target dies while this is |
| 2309 | * going on, some nodes could potentially see the target as the |
| 2310 | * master, so it is important that my recovery finds the migration |
| 2311 | * mle and sets the master to UNKNONWN. */ |
| 2312 | |
| 2313 | |
| 2314 | /* wait for new node to assert master */ |
| 2315 | while (1) { |
| 2316 | ret = wait_event_interruptible_timeout(mle->wq, |
| 2317 | (atomic_read(&mle->woken) == 1), |
| 2318 | msecs_to_jiffies(5000)); |
| 2319 | |
| 2320 | if (ret >= 0) { |
| 2321 | if (atomic_read(&mle->woken) == 1 || |
| 2322 | res->owner == target) |
| 2323 | break; |
| 2324 | |
| 2325 | mlog(0, "timed out during migration\n"); |
Kurt Hackel | e2faea4 | 2006-01-12 14:24:55 -0800 | [diff] [blame] | 2326 | /* avoid hang during shutdown when migrating lockres |
| 2327 | * to a node which also goes down */ |
| 2328 | if (dlm_is_node_dead(dlm, target)) { |
Kurt Hackel | aa85235 | 2006-04-27 19:04:49 -0700 | [diff] [blame] | 2329 | mlog(0, "%s:%.*s: expected migration " |
| 2330 | "target %u is no longer up, restarting\n", |
Kurt Hackel | e2faea4 | 2006-01-12 14:24:55 -0800 | [diff] [blame] | 2331 | dlm->name, res->lockname.len, |
| 2332 | res->lockname.name, target); |
| 2333 | ret = -ERESTARTSYS; |
| 2334 | } |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2335 | } |
| 2336 | if (ret == -ERESTARTSYS) { |
| 2337 | /* migration failed, detach and clean up mle */ |
| 2338 | dlm_mle_detach_hb_events(dlm, mle); |
| 2339 | dlm_put_mle(mle); |
Kurt Hackel | a2bf047 | 2006-04-27 18:51:26 -0700 | [diff] [blame] | 2340 | dlm_put_mle_inuse(mle); |
| 2341 | spin_lock(&res->spinlock); |
| 2342 | res->state &= ~DLM_LOCK_RES_MIGRATING; |
| 2343 | spin_unlock(&res->spinlock); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2344 | goto leave; |
| 2345 | } |
| 2346 | /* TODO: if node died: stop, clean up, return error */ |
| 2347 | } |
| 2348 | |
| 2349 | /* all done, set the owner, clear the flag */ |
| 2350 | spin_lock(&res->spinlock); |
| 2351 | dlm_set_lockres_owner(dlm, res, target); |
| 2352 | res->state &= ~DLM_LOCK_RES_MIGRATING; |
| 2353 | dlm_remove_nonlocal_locks(dlm, res); |
| 2354 | spin_unlock(&res->spinlock); |
| 2355 | wake_up(&res->wq); |
| 2356 | |
| 2357 | /* master is known, detach if not already detached */ |
| 2358 | dlm_mle_detach_hb_events(dlm, mle); |
Kurt Hackel | a2bf047 | 2006-04-27 18:51:26 -0700 | [diff] [blame] | 2359 | dlm_put_mle_inuse(mle); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2360 | ret = 0; |
| 2361 | |
| 2362 | dlm_lockres_calc_usage(dlm, res); |
| 2363 | |
| 2364 | leave: |
| 2365 | /* re-dirty the lockres if we failed */ |
| 2366 | if (ret < 0) |
| 2367 | dlm_kick_thread(dlm, res); |
| 2368 | |
| 2369 | /* TODO: cleanup */ |
| 2370 | if (mres) |
| 2371 | free_page((unsigned long)mres); |
| 2372 | |
| 2373 | dlm_put(dlm); |
| 2374 | |
| 2375 | mlog(0, "returning %d\n", ret); |
| 2376 | return ret; |
| 2377 | } |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2378 | |
| 2379 | int dlm_lock_basts_flushed(struct dlm_ctxt *dlm, struct dlm_lock *lock) |
| 2380 | { |
| 2381 | int ret; |
| 2382 | spin_lock(&dlm->ast_lock); |
| 2383 | spin_lock(&lock->spinlock); |
| 2384 | ret = (list_empty(&lock->bast_list) && !lock->bast_pending); |
| 2385 | spin_unlock(&lock->spinlock); |
| 2386 | spin_unlock(&dlm->ast_lock); |
| 2387 | return ret; |
| 2388 | } |
| 2389 | |
| 2390 | static int dlm_migration_can_proceed(struct dlm_ctxt *dlm, |
| 2391 | struct dlm_lock_resource *res, |
| 2392 | u8 mig_target) |
| 2393 | { |
| 2394 | int can_proceed; |
| 2395 | spin_lock(&res->spinlock); |
| 2396 | can_proceed = !!(res->state & DLM_LOCK_RES_MIGRATING); |
| 2397 | spin_unlock(&res->spinlock); |
| 2398 | |
| 2399 | /* target has died, so make the caller break out of the |
| 2400 | * wait_event, but caller must recheck the domain_map */ |
| 2401 | spin_lock(&dlm->spinlock); |
| 2402 | if (!test_bit(mig_target, dlm->domain_map)) |
| 2403 | can_proceed = 1; |
| 2404 | spin_unlock(&dlm->spinlock); |
| 2405 | return can_proceed; |
| 2406 | } |
| 2407 | |
| 2408 | int dlm_lockres_is_dirty(struct dlm_ctxt *dlm, struct dlm_lock_resource *res) |
| 2409 | { |
| 2410 | int ret; |
| 2411 | spin_lock(&res->spinlock); |
| 2412 | ret = !!(res->state & DLM_LOCK_RES_DIRTY); |
| 2413 | spin_unlock(&res->spinlock); |
| 2414 | return ret; |
| 2415 | } |
| 2416 | |
| 2417 | |
| 2418 | static int dlm_mark_lockres_migrating(struct dlm_ctxt *dlm, |
| 2419 | struct dlm_lock_resource *res, |
| 2420 | u8 target) |
| 2421 | { |
| 2422 | int ret = 0; |
| 2423 | |
| 2424 | mlog(0, "dlm_mark_lockres_migrating: %.*s, from %u to %u\n", |
| 2425 | res->lockname.len, res->lockname.name, dlm->node_num, |
| 2426 | target); |
| 2427 | /* need to set MIGRATING flag on lockres. this is done by |
| 2428 | * ensuring that all asts have been flushed for this lockres. */ |
| 2429 | spin_lock(&res->spinlock); |
| 2430 | BUG_ON(res->migration_pending); |
| 2431 | res->migration_pending = 1; |
| 2432 | /* strategy is to reserve an extra ast then release |
| 2433 | * it below, letting the release do all of the work */ |
| 2434 | __dlm_lockres_reserve_ast(res); |
| 2435 | spin_unlock(&res->spinlock); |
| 2436 | |
| 2437 | /* now flush all the pending asts.. hang out for a bit */ |
| 2438 | dlm_kick_thread(dlm, res); |
| 2439 | wait_event(dlm->ast_wq, !dlm_lockres_is_dirty(dlm, res)); |
| 2440 | dlm_lockres_release_ast(dlm, res); |
| 2441 | |
| 2442 | mlog(0, "about to wait on migration_wq, dirty=%s\n", |
| 2443 | res->state & DLM_LOCK_RES_DIRTY ? "yes" : "no"); |
| 2444 | /* if the extra ref we just put was the final one, this |
| 2445 | * will pass thru immediately. otherwise, we need to wait |
| 2446 | * for the last ast to finish. */ |
| 2447 | again: |
| 2448 | ret = wait_event_interruptible_timeout(dlm->migration_wq, |
| 2449 | dlm_migration_can_proceed(dlm, res, target), |
| 2450 | msecs_to_jiffies(1000)); |
| 2451 | if (ret < 0) { |
| 2452 | mlog(0, "woken again: migrating? %s, dead? %s\n", |
| 2453 | res->state & DLM_LOCK_RES_MIGRATING ? "yes":"no", |
| 2454 | test_bit(target, dlm->domain_map) ? "no":"yes"); |
| 2455 | } else { |
| 2456 | mlog(0, "all is well: migrating? %s, dead? %s\n", |
| 2457 | res->state & DLM_LOCK_RES_MIGRATING ? "yes":"no", |
| 2458 | test_bit(target, dlm->domain_map) ? "no":"yes"); |
| 2459 | } |
| 2460 | if (!dlm_migration_can_proceed(dlm, res, target)) { |
| 2461 | mlog(0, "trying again...\n"); |
| 2462 | goto again; |
| 2463 | } |
| 2464 | |
| 2465 | /* did the target go down or die? */ |
| 2466 | spin_lock(&dlm->spinlock); |
| 2467 | if (!test_bit(target, dlm->domain_map)) { |
| 2468 | mlog(ML_ERROR, "aha. migration target %u just went down\n", |
| 2469 | target); |
| 2470 | ret = -EHOSTDOWN; |
| 2471 | } |
| 2472 | spin_unlock(&dlm->spinlock); |
| 2473 | |
| 2474 | /* |
| 2475 | * at this point: |
| 2476 | * |
| 2477 | * o the DLM_LOCK_RES_MIGRATING flag is set |
| 2478 | * o there are no pending asts on this lockres |
| 2479 | * o all processes trying to reserve an ast on this |
| 2480 | * lockres must wait for the MIGRATING flag to clear |
| 2481 | */ |
| 2482 | return ret; |
| 2483 | } |
| 2484 | |
| 2485 | /* last step in the migration process. |
| 2486 | * original master calls this to free all of the dlm_lock |
| 2487 | * structures that used to be for other nodes. */ |
| 2488 | static void dlm_remove_nonlocal_locks(struct dlm_ctxt *dlm, |
| 2489 | struct dlm_lock_resource *res) |
| 2490 | { |
| 2491 | struct list_head *iter, *iter2; |
| 2492 | struct list_head *queue = &res->granted; |
| 2493 | int i; |
| 2494 | struct dlm_lock *lock; |
| 2495 | |
| 2496 | assert_spin_locked(&res->spinlock); |
| 2497 | |
| 2498 | BUG_ON(res->owner == dlm->node_num); |
| 2499 | |
| 2500 | for (i=0; i<3; i++) { |
| 2501 | list_for_each_safe(iter, iter2, queue) { |
| 2502 | lock = list_entry (iter, struct dlm_lock, list); |
| 2503 | if (lock->ml.node != dlm->node_num) { |
| 2504 | mlog(0, "putting lock for node %u\n", |
| 2505 | lock->ml.node); |
| 2506 | /* be extra careful */ |
| 2507 | BUG_ON(!list_empty(&lock->ast_list)); |
| 2508 | BUG_ON(!list_empty(&lock->bast_list)); |
| 2509 | BUG_ON(lock->ast_pending); |
| 2510 | BUG_ON(lock->bast_pending); |
| 2511 | list_del_init(&lock->list); |
| 2512 | dlm_lock_put(lock); |
| 2513 | } |
| 2514 | } |
| 2515 | queue++; |
| 2516 | } |
| 2517 | } |
| 2518 | |
| 2519 | /* for now this is not too intelligent. we will |
| 2520 | * need stats to make this do the right thing. |
| 2521 | * this just finds the first lock on one of the |
| 2522 | * queues and uses that node as the target. */ |
| 2523 | static u8 dlm_pick_migration_target(struct dlm_ctxt *dlm, |
| 2524 | struct dlm_lock_resource *res) |
| 2525 | { |
| 2526 | int i; |
| 2527 | struct list_head *queue = &res->granted; |
| 2528 | struct list_head *iter; |
| 2529 | struct dlm_lock *lock; |
| 2530 | int nodenum; |
| 2531 | |
| 2532 | assert_spin_locked(&dlm->spinlock); |
| 2533 | |
| 2534 | spin_lock(&res->spinlock); |
| 2535 | for (i=0; i<3; i++) { |
| 2536 | list_for_each(iter, queue) { |
| 2537 | /* up to the caller to make sure this node |
| 2538 | * is alive */ |
| 2539 | lock = list_entry (iter, struct dlm_lock, list); |
| 2540 | if (lock->ml.node != dlm->node_num) { |
| 2541 | spin_unlock(&res->spinlock); |
| 2542 | return lock->ml.node; |
| 2543 | } |
| 2544 | } |
| 2545 | queue++; |
| 2546 | } |
| 2547 | spin_unlock(&res->spinlock); |
| 2548 | mlog(0, "have not found a suitable target yet! checking domain map\n"); |
| 2549 | |
| 2550 | /* ok now we're getting desperate. pick anyone alive. */ |
| 2551 | nodenum = -1; |
| 2552 | while (1) { |
| 2553 | nodenum = find_next_bit(dlm->domain_map, |
| 2554 | O2NM_MAX_NODES, nodenum+1); |
| 2555 | mlog(0, "found %d in domain map\n", nodenum); |
| 2556 | if (nodenum >= O2NM_MAX_NODES) |
| 2557 | break; |
| 2558 | if (nodenum != dlm->node_num) { |
| 2559 | mlog(0, "picking %d\n", nodenum); |
| 2560 | return nodenum; |
| 2561 | } |
| 2562 | } |
| 2563 | |
| 2564 | mlog(0, "giving up. no master to migrate to\n"); |
| 2565 | return DLM_LOCK_RES_OWNER_UNKNOWN; |
| 2566 | } |
| 2567 | |
| 2568 | |
| 2569 | |
| 2570 | /* this is called by the new master once all lockres |
| 2571 | * data has been received */ |
| 2572 | static int dlm_do_migrate_request(struct dlm_ctxt *dlm, |
| 2573 | struct dlm_lock_resource *res, |
| 2574 | u8 master, u8 new_master, |
| 2575 | struct dlm_node_iter *iter) |
| 2576 | { |
| 2577 | struct dlm_migrate_request migrate; |
| 2578 | int ret, status = 0; |
| 2579 | int nodenum; |
| 2580 | |
| 2581 | memset(&migrate, 0, sizeof(migrate)); |
| 2582 | migrate.namelen = res->lockname.len; |
| 2583 | memcpy(migrate.name, res->lockname.name, migrate.namelen); |
| 2584 | migrate.new_master = new_master; |
| 2585 | migrate.master = master; |
| 2586 | |
| 2587 | ret = 0; |
| 2588 | |
| 2589 | /* send message to all nodes, except the master and myself */ |
| 2590 | while ((nodenum = dlm_node_iter_next(iter)) >= 0) { |
| 2591 | if (nodenum == master || |
| 2592 | nodenum == new_master) |
| 2593 | continue; |
| 2594 | |
| 2595 | ret = o2net_send_message(DLM_MIGRATE_REQUEST_MSG, dlm->key, |
| 2596 | &migrate, sizeof(migrate), nodenum, |
| 2597 | &status); |
| 2598 | if (ret < 0) |
| 2599 | mlog_errno(ret); |
| 2600 | else if (status < 0) { |
| 2601 | mlog(0, "migrate request (node %u) returned %d!\n", |
| 2602 | nodenum, status); |
| 2603 | ret = status; |
| 2604 | } |
| 2605 | } |
| 2606 | |
| 2607 | if (ret < 0) |
| 2608 | mlog_errno(ret); |
| 2609 | |
| 2610 | mlog(0, "returning ret=%d\n", ret); |
| 2611 | return ret; |
| 2612 | } |
| 2613 | |
| 2614 | |
| 2615 | /* if there is an existing mle for this lockres, we now know who the master is. |
| 2616 | * (the one who sent us *this* message) we can clear it up right away. |
| 2617 | * since the process that put the mle on the list still has a reference to it, |
| 2618 | * we can unhash it now, set the master and wake the process. as a result, |
| 2619 | * we will have no mle in the list to start with. now we can add an mle for |
| 2620 | * the migration and this should be the only one found for those scanning the |
| 2621 | * list. */ |
| 2622 | int dlm_migrate_request_handler(struct o2net_msg *msg, u32 len, void *data) |
| 2623 | { |
| 2624 | struct dlm_ctxt *dlm = data; |
| 2625 | struct dlm_lock_resource *res = NULL; |
| 2626 | struct dlm_migrate_request *migrate = (struct dlm_migrate_request *) msg->buf; |
| 2627 | struct dlm_master_list_entry *mle = NULL, *oldmle = NULL; |
| 2628 | const char *name; |
Mark Fasheh | a3d3329 | 2006-03-09 17:55:56 -0800 | [diff] [blame] | 2629 | unsigned int namelen, hash; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2630 | int ret = 0; |
| 2631 | |
| 2632 | if (!dlm_grab(dlm)) |
| 2633 | return -EINVAL; |
| 2634 | |
| 2635 | name = migrate->name; |
| 2636 | namelen = migrate->namelen; |
Mark Fasheh | a3d3329 | 2006-03-09 17:55:56 -0800 | [diff] [blame] | 2637 | hash = dlm_lockid_hash(name, namelen); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2638 | |
| 2639 | /* preallocate.. if this fails, abort */ |
| 2640 | mle = (struct dlm_master_list_entry *) kmem_cache_alloc(dlm_mle_cache, |
Kurt Hackel | ad8100e | 2006-05-01 14:25:21 -0700 | [diff] [blame] | 2641 | GFP_NOFS); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2642 | |
| 2643 | if (!mle) { |
| 2644 | ret = -ENOMEM; |
| 2645 | goto leave; |
| 2646 | } |
| 2647 | |
| 2648 | /* check for pre-existing lock */ |
| 2649 | spin_lock(&dlm->spinlock); |
Mark Fasheh | a3d3329 | 2006-03-09 17:55:56 -0800 | [diff] [blame] | 2650 | res = __dlm_lookup_lockres(dlm, name, namelen, hash); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2651 | spin_lock(&dlm->master_lock); |
| 2652 | |
| 2653 | if (res) { |
| 2654 | spin_lock(&res->spinlock); |
| 2655 | if (res->state & DLM_LOCK_RES_RECOVERING) { |
| 2656 | /* if all is working ok, this can only mean that we got |
| 2657 | * a migrate request from a node that we now see as |
| 2658 | * dead. what can we do here? drop it to the floor? */ |
| 2659 | spin_unlock(&res->spinlock); |
| 2660 | mlog(ML_ERROR, "Got a migrate request, but the " |
| 2661 | "lockres is marked as recovering!"); |
| 2662 | kmem_cache_free(dlm_mle_cache, mle); |
| 2663 | ret = -EINVAL; /* need a better solution */ |
| 2664 | goto unlock; |
| 2665 | } |
| 2666 | res->state |= DLM_LOCK_RES_MIGRATING; |
| 2667 | spin_unlock(&res->spinlock); |
| 2668 | } |
| 2669 | |
| 2670 | /* ignore status. only nonzero status would BUG. */ |
| 2671 | ret = dlm_add_migration_mle(dlm, res, mle, &oldmle, |
| 2672 | name, namelen, |
| 2673 | migrate->new_master, |
| 2674 | migrate->master); |
| 2675 | |
| 2676 | unlock: |
| 2677 | spin_unlock(&dlm->master_lock); |
| 2678 | spin_unlock(&dlm->spinlock); |
| 2679 | |
| 2680 | if (oldmle) { |
| 2681 | /* master is known, detach if not already detached */ |
| 2682 | dlm_mle_detach_hb_events(dlm, oldmle); |
| 2683 | dlm_put_mle(oldmle); |
| 2684 | } |
| 2685 | |
| 2686 | if (res) |
| 2687 | dlm_lockres_put(res); |
| 2688 | leave: |
| 2689 | dlm_put(dlm); |
| 2690 | return ret; |
| 2691 | } |
| 2692 | |
| 2693 | /* must be holding dlm->spinlock and dlm->master_lock |
| 2694 | * when adding a migration mle, we can clear any other mles |
| 2695 | * in the master list because we know with certainty that |
| 2696 | * the master is "master". so we remove any old mle from |
| 2697 | * the list after setting it's master field, and then add |
| 2698 | * the new migration mle. this way we can hold with the rule |
| 2699 | * of having only one mle for a given lock name at all times. */ |
| 2700 | static int dlm_add_migration_mle(struct dlm_ctxt *dlm, |
| 2701 | struct dlm_lock_resource *res, |
| 2702 | struct dlm_master_list_entry *mle, |
| 2703 | struct dlm_master_list_entry **oldmle, |
| 2704 | const char *name, unsigned int namelen, |
| 2705 | u8 new_master, u8 master) |
| 2706 | { |
| 2707 | int found; |
| 2708 | int ret = 0; |
| 2709 | |
| 2710 | *oldmle = NULL; |
| 2711 | |
| 2712 | mlog_entry_void(); |
| 2713 | |
| 2714 | assert_spin_locked(&dlm->spinlock); |
| 2715 | assert_spin_locked(&dlm->master_lock); |
| 2716 | |
| 2717 | /* caller is responsible for any ref taken here on oldmle */ |
| 2718 | found = dlm_find_mle(dlm, oldmle, (char *)name, namelen); |
| 2719 | if (found) { |
| 2720 | struct dlm_master_list_entry *tmp = *oldmle; |
| 2721 | spin_lock(&tmp->spinlock); |
| 2722 | if (tmp->type == DLM_MLE_MIGRATION) { |
| 2723 | if (master == dlm->node_num) { |
| 2724 | /* ah another process raced me to it */ |
| 2725 | mlog(0, "tried to migrate %.*s, but some " |
| 2726 | "process beat me to it\n", |
| 2727 | namelen, name); |
| 2728 | ret = -EEXIST; |
| 2729 | } else { |
| 2730 | /* bad. 2 NODES are trying to migrate! */ |
| 2731 | mlog(ML_ERROR, "migration error mle: " |
| 2732 | "master=%u new_master=%u // request: " |
| 2733 | "master=%u new_master=%u // " |
| 2734 | "lockres=%.*s\n", |
| 2735 | tmp->master, tmp->new_master, |
| 2736 | master, new_master, |
| 2737 | namelen, name); |
| 2738 | BUG(); |
| 2739 | } |
| 2740 | } else { |
| 2741 | /* this is essentially what assert_master does */ |
| 2742 | tmp->master = master; |
| 2743 | atomic_set(&tmp->woken, 1); |
| 2744 | wake_up(&tmp->wq); |
| 2745 | /* remove it from the list so that only one |
| 2746 | * mle will be found */ |
| 2747 | list_del_init(&tmp->list); |
Kurt Hackel | da01ad0 | 2006-04-27 18:53:04 -0700 | [diff] [blame] | 2748 | __dlm_mle_detach_hb_events(dlm, mle); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2749 | } |
| 2750 | spin_unlock(&tmp->spinlock); |
| 2751 | } |
| 2752 | |
| 2753 | /* now add a migration mle to the tail of the list */ |
| 2754 | dlm_init_mle(mle, DLM_MLE_MIGRATION, dlm, res, name, namelen); |
| 2755 | mle->new_master = new_master; |
| 2756 | mle->master = master; |
| 2757 | /* do this for consistency with other mle types */ |
| 2758 | set_bit(new_master, mle->maybe_map); |
| 2759 | list_add(&mle->list, &dlm->master_list); |
| 2760 | |
| 2761 | return ret; |
| 2762 | } |
| 2763 | |
| 2764 | |
| 2765 | void dlm_clean_master_list(struct dlm_ctxt *dlm, u8 dead_node) |
| 2766 | { |
| 2767 | struct list_head *iter, *iter2; |
| 2768 | struct dlm_master_list_entry *mle; |
| 2769 | struct dlm_lock_resource *res; |
Mark Fasheh | a3d3329 | 2006-03-09 17:55:56 -0800 | [diff] [blame] | 2770 | unsigned int hash; |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2771 | |
| 2772 | mlog_entry("dlm=%s, dead node=%u\n", dlm->name, dead_node); |
| 2773 | top: |
| 2774 | assert_spin_locked(&dlm->spinlock); |
| 2775 | |
| 2776 | /* clean the master list */ |
| 2777 | spin_lock(&dlm->master_lock); |
| 2778 | list_for_each_safe(iter, iter2, &dlm->master_list) { |
| 2779 | mle = list_entry(iter, struct dlm_master_list_entry, list); |
| 2780 | |
| 2781 | BUG_ON(mle->type != DLM_MLE_BLOCK && |
| 2782 | mle->type != DLM_MLE_MASTER && |
| 2783 | mle->type != DLM_MLE_MIGRATION); |
| 2784 | |
| 2785 | /* MASTER mles are initiated locally. the waiting |
| 2786 | * process will notice the node map change |
| 2787 | * shortly. let that happen as normal. */ |
| 2788 | if (mle->type == DLM_MLE_MASTER) |
| 2789 | continue; |
| 2790 | |
| 2791 | |
| 2792 | /* BLOCK mles are initiated by other nodes. |
| 2793 | * need to clean up if the dead node would have |
| 2794 | * been the master. */ |
| 2795 | if (mle->type == DLM_MLE_BLOCK) { |
| 2796 | int bit; |
| 2797 | |
| 2798 | spin_lock(&mle->spinlock); |
| 2799 | bit = find_next_bit(mle->maybe_map, O2NM_MAX_NODES, 0); |
| 2800 | if (bit != dead_node) { |
| 2801 | mlog(0, "mle found, but dead node %u would " |
| 2802 | "not have been master\n", dead_node); |
| 2803 | spin_unlock(&mle->spinlock); |
| 2804 | } else { |
| 2805 | /* must drop the refcount by one since the |
| 2806 | * assert_master will never arrive. this |
| 2807 | * may result in the mle being unlinked and |
| 2808 | * freed, but there may still be a process |
| 2809 | * waiting in the dlmlock path which is fine. */ |
Kurt Hackel | 3b3b84a | 2006-05-01 14:31:37 -0700 | [diff] [blame] | 2810 | mlog(0, "node %u was expected master\n", |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2811 | dead_node); |
| 2812 | atomic_set(&mle->woken, 1); |
| 2813 | spin_unlock(&mle->spinlock); |
| 2814 | wake_up(&mle->wq); |
Kurt Hackel | f671c09 | 2006-02-14 11:45:21 -0800 | [diff] [blame] | 2815 | /* do not need events any longer, so detach |
| 2816 | * from heartbeat */ |
| 2817 | __dlm_mle_detach_hb_events(dlm, mle); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2818 | __dlm_put_mle(mle); |
| 2819 | } |
| 2820 | continue; |
| 2821 | } |
| 2822 | |
| 2823 | /* everything else is a MIGRATION mle */ |
| 2824 | |
| 2825 | /* the rule for MIGRATION mles is that the master |
| 2826 | * becomes UNKNOWN if *either* the original or |
| 2827 | * the new master dies. all UNKNOWN lockreses |
| 2828 | * are sent to whichever node becomes the recovery |
| 2829 | * master. the new master is responsible for |
| 2830 | * determining if there is still a master for |
| 2831 | * this lockres, or if he needs to take over |
| 2832 | * mastery. either way, this node should expect |
| 2833 | * another message to resolve this. */ |
| 2834 | if (mle->master != dead_node && |
| 2835 | mle->new_master != dead_node) |
| 2836 | continue; |
| 2837 | |
| 2838 | /* if we have reached this point, this mle needs to |
| 2839 | * be removed from the list and freed. */ |
| 2840 | |
| 2841 | /* remove from the list early. NOTE: unlinking |
| 2842 | * list_head while in list_for_each_safe */ |
Kurt Hackel | da01ad0 | 2006-04-27 18:53:04 -0700 | [diff] [blame] | 2843 | __dlm_mle_detach_hb_events(dlm, mle); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2844 | spin_lock(&mle->spinlock); |
| 2845 | list_del_init(&mle->list); |
| 2846 | atomic_set(&mle->woken, 1); |
| 2847 | spin_unlock(&mle->spinlock); |
| 2848 | wake_up(&mle->wq); |
| 2849 | |
Kurt Hackel | aa85235 | 2006-04-27 19:04:49 -0700 | [diff] [blame] | 2850 | mlog(0, "%s: node %u died during migration from " |
| 2851 | "%u to %u!\n", dlm->name, dead_node, |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2852 | mle->master, mle->new_master); |
| 2853 | /* if there is a lockres associated with this |
| 2854 | * mle, find it and set its owner to UNKNOWN */ |
Mark Fasheh | a3d3329 | 2006-03-09 17:55:56 -0800 | [diff] [blame] | 2855 | hash = dlm_lockid_hash(mle->u.name.name, mle->u.name.len); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2856 | res = __dlm_lookup_lockres(dlm, mle->u.name.name, |
Mark Fasheh | a3d3329 | 2006-03-09 17:55:56 -0800 | [diff] [blame] | 2857 | mle->u.name.len, hash); |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2858 | if (res) { |
| 2859 | /* unfortunately if we hit this rare case, our |
| 2860 | * lock ordering is messed. we need to drop |
| 2861 | * the master lock so that we can take the |
| 2862 | * lockres lock, meaning that we will have to |
| 2863 | * restart from the head of list. */ |
| 2864 | spin_unlock(&dlm->master_lock); |
| 2865 | |
| 2866 | /* move lockres onto recovery list */ |
| 2867 | spin_lock(&res->spinlock); |
| 2868 | dlm_set_lockres_owner(dlm, res, |
| 2869 | DLM_LOCK_RES_OWNER_UNKNOWN); |
| 2870 | dlm_move_lockres_to_recovery_list(dlm, res); |
| 2871 | spin_unlock(&res->spinlock); |
| 2872 | dlm_lockres_put(res); |
| 2873 | |
Kurt Hackel | f671c09 | 2006-02-14 11:45:21 -0800 | [diff] [blame] | 2874 | /* about to get rid of mle, detach from heartbeat */ |
| 2875 | __dlm_mle_detach_hb_events(dlm, mle); |
| 2876 | |
Kurt Hackel | 6714d8e | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 2877 | /* dump the mle */ |
| 2878 | spin_lock(&dlm->master_lock); |
| 2879 | __dlm_put_mle(mle); |
| 2880 | spin_unlock(&dlm->master_lock); |
| 2881 | |
| 2882 | /* restart */ |
| 2883 | goto top; |
| 2884 | } |
| 2885 | |
| 2886 | /* this may be the last reference */ |
| 2887 | __dlm_put_mle(mle); |
| 2888 | } |
| 2889 | spin_unlock(&dlm->master_lock); |
| 2890 | } |
| 2891 | |
| 2892 | |
| 2893 | int dlm_finish_migration(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, |
| 2894 | u8 old_master) |
| 2895 | { |
| 2896 | struct dlm_node_iter iter; |
| 2897 | int ret = 0; |
| 2898 | |
| 2899 | spin_lock(&dlm->spinlock); |
| 2900 | dlm_node_iter_init(dlm->domain_map, &iter); |
| 2901 | clear_bit(old_master, iter.node_map); |
| 2902 | clear_bit(dlm->node_num, iter.node_map); |
| 2903 | spin_unlock(&dlm->spinlock); |
| 2904 | |
| 2905 | mlog(0, "now time to do a migrate request to other nodes\n"); |
| 2906 | ret = dlm_do_migrate_request(dlm, res, old_master, |
| 2907 | dlm->node_num, &iter); |
| 2908 | if (ret < 0) { |
| 2909 | mlog_errno(ret); |
| 2910 | goto leave; |
| 2911 | } |
| 2912 | |
| 2913 | mlog(0, "doing assert master of %.*s to all except the original node\n", |
| 2914 | res->lockname.len, res->lockname.name); |
| 2915 | /* this call now finishes out the nodemap |
| 2916 | * even if one or more nodes die */ |
| 2917 | ret = dlm_do_assert_master(dlm, res->lockname.name, |
| 2918 | res->lockname.len, iter.node_map, |
| 2919 | DLM_ASSERT_MASTER_FINISH_MIGRATION); |
| 2920 | if (ret < 0) { |
| 2921 | /* no longer need to retry. all living nodes contacted. */ |
| 2922 | mlog_errno(ret); |
| 2923 | ret = 0; |
| 2924 | } |
| 2925 | |
| 2926 | memset(iter.node_map, 0, sizeof(iter.node_map)); |
| 2927 | set_bit(old_master, iter.node_map); |
| 2928 | mlog(0, "doing assert master of %.*s back to %u\n", |
| 2929 | res->lockname.len, res->lockname.name, old_master); |
| 2930 | ret = dlm_do_assert_master(dlm, res->lockname.name, |
| 2931 | res->lockname.len, iter.node_map, |
| 2932 | DLM_ASSERT_MASTER_FINISH_MIGRATION); |
| 2933 | if (ret < 0) { |
| 2934 | mlog(0, "assert master to original master failed " |
| 2935 | "with %d.\n", ret); |
| 2936 | /* the only nonzero status here would be because of |
| 2937 | * a dead original node. we're done. */ |
| 2938 | ret = 0; |
| 2939 | } |
| 2940 | |
| 2941 | /* all done, set the owner, clear the flag */ |
| 2942 | spin_lock(&res->spinlock); |
| 2943 | dlm_set_lockres_owner(dlm, res, dlm->node_num); |
| 2944 | res->state &= ~DLM_LOCK_RES_MIGRATING; |
| 2945 | spin_unlock(&res->spinlock); |
| 2946 | /* re-dirty it on the new master */ |
| 2947 | dlm_kick_thread(dlm, res); |
| 2948 | wake_up(&res->wq); |
| 2949 | leave: |
| 2950 | return ret; |
| 2951 | } |
| 2952 | |
| 2953 | /* |
| 2954 | * LOCKRES AST REFCOUNT |
| 2955 | * this is integral to migration |
| 2956 | */ |
| 2957 | |
| 2958 | /* for future intent to call an ast, reserve one ahead of time. |
| 2959 | * this should be called only after waiting on the lockres |
| 2960 | * with dlm_wait_on_lockres, and while still holding the |
| 2961 | * spinlock after the call. */ |
| 2962 | void __dlm_lockres_reserve_ast(struct dlm_lock_resource *res) |
| 2963 | { |
| 2964 | assert_spin_locked(&res->spinlock); |
| 2965 | if (res->state & DLM_LOCK_RES_MIGRATING) { |
| 2966 | __dlm_print_one_lock_resource(res); |
| 2967 | } |
| 2968 | BUG_ON(res->state & DLM_LOCK_RES_MIGRATING); |
| 2969 | |
| 2970 | atomic_inc(&res->asts_reserved); |
| 2971 | } |
| 2972 | |
| 2973 | /* |
| 2974 | * used to drop the reserved ast, either because it went unused, |
| 2975 | * or because the ast/bast was actually called. |
| 2976 | * |
| 2977 | * also, if there is a pending migration on this lockres, |
| 2978 | * and this was the last pending ast on the lockres, |
| 2979 | * atomically set the MIGRATING flag before we drop the lock. |
| 2980 | * this is how we ensure that migration can proceed with no |
| 2981 | * asts in progress. note that it is ok if the state of the |
| 2982 | * queues is such that a lock should be granted in the future |
| 2983 | * or that a bast should be fired, because the new master will |
| 2984 | * shuffle the lists on this lockres as soon as it is migrated. |
| 2985 | */ |
| 2986 | void dlm_lockres_release_ast(struct dlm_ctxt *dlm, |
| 2987 | struct dlm_lock_resource *res) |
| 2988 | { |
| 2989 | if (!atomic_dec_and_lock(&res->asts_reserved, &res->spinlock)) |
| 2990 | return; |
| 2991 | |
| 2992 | if (!res->migration_pending) { |
| 2993 | spin_unlock(&res->spinlock); |
| 2994 | return; |
| 2995 | } |
| 2996 | |
| 2997 | BUG_ON(res->state & DLM_LOCK_RES_MIGRATING); |
| 2998 | res->migration_pending = 0; |
| 2999 | res->state |= DLM_LOCK_RES_MIGRATING; |
| 3000 | spin_unlock(&res->spinlock); |
| 3001 | wake_up(&res->wq); |
| 3002 | wake_up(&dlm->migration_wq); |
| 3003 | } |