fs/ocfs2/dlmglue.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * dlmglue.c
   5  *
   6  * Code which implements an OCFS2 specific interface to our DLM.
   7  *
   8  * Copyright (C) 2003, 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 #include <linux/types.h>
  27 #include <linux/slab.h>
  28 #include <linux/highmem.h>
  29 #include <linux/mm.h>
  30 #include <linux/smp_lock.h>
  31 #include <linux/crc32.h>
  32 #include <linux/kthread.h>
  33 #include <linux/pagemap.h>
  34 #include <linux/debugfs.h>
  35 #include <linux/seq_file.h>
  36
  37 #include <cluster/heartbeat.h>
  38 #include <cluster/nodemanager.h>
  39 #include <cluster/tcp.h>
  40
  41 #include <dlm/dlmapi.h>
  42
  43 #define MLOG_MASK_PREFIX ML_DLM_GLUE
  44 #include <cluster/masklog.h>
  45
  46 #include "ocfs2.h"
  47
  48 #include "alloc.h"
  49 #include "dlmglue.h"
  50 #include "extent_map.h"
  51 #include "heartbeat.h"
  52 #include "inode.h"
  53 #include "journal.h"
  54 #include "slot_map.h"
  55 #include "super.h"
  56 #include "uptodate.h"
  57 #include "vote.h"
  58
  59 #include "buffer_head_io.h"
  60
  61 struct ocfs2_mask_waiter {
  62         struct list_head        mw_item;
  63         int                     mw_status;
  64         struct completion       mw_complete;
  65         unsigned long           mw_mask;
  66         unsigned long           mw_goal;
  67 };
  68
  69 static void ocfs2_inode_ast_func(void *opaque);
  70 static void ocfs2_inode_bast_func(void *opaque,
  71                                   int level);
  72 static void ocfs2_super_ast_func(void *opaque);
  73 static void ocfs2_super_bast_func(void *opaque,
  74                                   int level);
  75 static void ocfs2_rename_ast_func(void *opaque);
  76 static void ocfs2_rename_bast_func(void *opaque,
  77                                    int level);
  78
  79 /* so far, all locks have gotten along with the same unlock ast */
  80 static void ocfs2_unlock_ast_func(void *opaque,
  81                                   enum dlm_status status);
  82 static int ocfs2_do_unblock_meta(struct inode *inode,
  83                                  int *requeue);
  84 static int ocfs2_unblock_meta(struct ocfs2_lock_res *lockres,
  85                               int *requeue);
  86 static int ocfs2_unblock_data(struct ocfs2_lock_res *lockres,
  87                               int *requeue);
  88 static int ocfs2_unblock_inode_lock(struct ocfs2_lock_res *lockres,
  89                               int *requeue);
  90 static int ocfs2_unblock_osb_lock(struct ocfs2_lock_res *lockres,
  91                                   int *requeue);
  92 typedef void (ocfs2_convert_worker_t)(struct ocfs2_lock_res *, int);
  93 static int ocfs2_generic_unblock_lock(struct ocfs2_super *osb,
  94                                       struct ocfs2_lock_res *lockres,
  95                                       int *requeue,
  96                                       ocfs2_convert_worker_t *worker);
  97
  98 struct ocfs2_lock_res_ops {
  99         void (*ast)(void *);
 100         void (*bast)(void *, int);
 101         void (*unlock_ast)(void *, enum dlm_status);
 102         int  (*unblock)(struct ocfs2_lock_res *, int *);
 103 };
 104
 105 static struct ocfs2_lock_res_ops ocfs2_inode_rw_lops = {
 106         .ast            = ocfs2_inode_ast_func,
 107         .bast           = ocfs2_inode_bast_func,
 108         .unlock_ast     = ocfs2_unlock_ast_func,
 109         .unblock        = ocfs2_unblock_inode_lock,
 110 };
 111
 112 static struct ocfs2_lock_res_ops ocfs2_inode_meta_lops = {
 113         .ast            = ocfs2_inode_ast_func,
 114         .bast           = ocfs2_inode_bast_func,
 115         .unlock_ast     = ocfs2_unlock_ast_func,
 116         .unblock        = ocfs2_unblock_meta,
 117 };
 118
 119 static void ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
 120                                       int blocking);
 121
 122 static struct ocfs2_lock_res_ops ocfs2_inode_data_lops = {
 123         .ast            = ocfs2_inode_ast_func,
 124         .bast           = ocfs2_inode_bast_func,
 125         .unlock_ast     = ocfs2_unlock_ast_func,
 126         .unblock        = ocfs2_unblock_data,
 127 };
 128
 129 static struct ocfs2_lock_res_ops ocfs2_super_lops = {
 130         .ast            = ocfs2_super_ast_func,
 131         .bast           = ocfs2_super_bast_func,
 132         .unlock_ast     = ocfs2_unlock_ast_func,
 133         .unblock        = ocfs2_unblock_osb_lock,
 134 };
 135
 136 static struct ocfs2_lock_res_ops ocfs2_rename_lops = {
 137         .ast            = ocfs2_rename_ast_func,
 138         .bast           = ocfs2_rename_bast_func,
 139         .unlock_ast     = ocfs2_unlock_ast_func,
 140         .unblock        = ocfs2_unblock_osb_lock,
 141 };
 142
 143 static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres)
 144 {
 145         return lockres->l_type == OCFS2_LOCK_TYPE_META ||
 146                 lockres->l_type == OCFS2_LOCK_TYPE_DATA ||
 147                 lockres->l_type == OCFS2_LOCK_TYPE_RW;
 148 }
 149
 150 static inline int ocfs2_is_super_lock(struct ocfs2_lock_res *lockres)
 151 {
 152         return lockres->l_type == OCFS2_LOCK_TYPE_SUPER;
 153 }
 154
 155 static inline int ocfs2_is_rename_lock(struct ocfs2_lock_res *lockres)
 156 {
 157         return lockres->l_type == OCFS2_LOCK_TYPE_RENAME;
 158 }
 159
 160 static inline struct ocfs2_super *ocfs2_lock_res_super(struct ocfs2_lock_res *lockres)
 161 {
 162         BUG_ON(!ocfs2_is_super_lock(lockres)
 163                && !ocfs2_is_rename_lock(lockres));
 164
 165         return (struct ocfs2_super *) lockres->l_priv;
 166 }
 167
 168 static inline struct inode *ocfs2_lock_res_inode(struct ocfs2_lock_res *lockres)
 169 {
 170         BUG_ON(!ocfs2_is_inode_lock(lockres));
 171
 172         return (struct inode *) lockres->l_priv;
 173 }
 174
 175 static int ocfs2_lock_create(struct ocfs2_super *osb,
 176                              struct ocfs2_lock_res *lockres,
 177                              int level,
 178                              int dlm_flags);
 179 static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
 180                                                      int wanted);
 181 static void ocfs2_cluster_unlock(struct ocfs2_super *osb,
 182                                  struct ocfs2_lock_res *lockres,
 183                                  int level);
 184 static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres);
 185 static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres);
 186 static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres);
 187 static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, int level);
 188 static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
 189                                         struct ocfs2_lock_res *lockres);
 190 static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
 191                                                 int convert);
 192 #define ocfs2_log_dlm_error(_func, _stat, _lockres) do {        \
 193         mlog(ML_ERROR, "Dlm error \"%s\" while calling %s on "  \
 194                 "resource %s: %s\n", dlm_errname(_stat), _func, \
 195                 _lockres->l_name, dlm_errmsg(_stat));           \
 196 } while (0)
 197 static void ocfs2_vote_on_unlock(struct ocfs2_super *osb,
 198                                  struct ocfs2_lock_res *lockres);
 199 static int ocfs2_meta_lock_update(struct inode *inode,
 200                                   struct buffer_head **bh);
 201 static void ocfs2_drop_osb_locks(struct ocfs2_super *osb);
 202 static inline int ocfs2_highest_compat_lock_level(int level);
 203 static inline int ocfs2_can_downconvert_meta_lock(struct inode *inode,
 204                                                   struct ocfs2_lock_res *lockres,
 205                                                   int new_level);
 206
 207 static char *ocfs2_lock_type_strings[] = {
 208         [OCFS2_LOCK_TYPE_META] = "Meta",
 209         [OCFS2_LOCK_TYPE_DATA] = "Data",
 210         [OCFS2_LOCK_TYPE_SUPER] = "Super",
 211         [OCFS2_LOCK_TYPE_RENAME] = "Rename",
 212         /* Need to differntiate from [R]ename.. serializing writes is the
 213          * important job it does, anyway. */
 214         [OCFS2_LOCK_TYPE_RW] = "Write/Read",
 215 };
 216
 217 static char *ocfs2_lock_type_string(enum ocfs2_lock_type type)
 218 {
 219         mlog_bug_on_msg(type >= OCFS2_NUM_LOCK_TYPES, "%d\n", type);
 220         return ocfs2_lock_type_strings[type];
 221 }
 222
 223 static void ocfs2_build_lock_name(enum ocfs2_lock_type type,
 224                                   u64 blkno,
 225                                   u32 generation,
 226                                   char *name)
 227 {
 228         int len;
 229
 230         mlog_entry_void();
 231
 232         BUG_ON(type >= OCFS2_NUM_LOCK_TYPES);
 233
 234         len = snprintf(name, OCFS2_LOCK_ID_MAX_LEN, "%c%s%016llx%08x",
 235                        ocfs2_lock_type_char(type), OCFS2_LOCK_ID_PAD,
 236                        (long long)blkno, generation);
 237
 238         BUG_ON(len != (OCFS2_LOCK_ID_MAX_LEN - 1));
 239
 240         mlog(0, "built lock resource with name: %s\n", name);
 241
 242         mlog_exit_void();
 243 }
 244
 245 static DEFINE_SPINLOCK(ocfs2_dlm_tracking_lock);
 246
 247 static void ocfs2_add_lockres_tracking(struct ocfs2_lock_res *res,
 248                                        struct ocfs2_dlm_debug *dlm_debug)
 249 {
 250         mlog(0, "Add tracking for lockres %s\n", res->l_name);
 251
 252         spin_lock(&ocfs2_dlm_tracking_lock);
 253         list_add(&res->l_debug_list, &dlm_debug->d_lockres_tracking);
 254         spin_unlock(&ocfs2_dlm_tracking_lock);
 255 }
 256
 257 static void ocfs2_remove_lockres_tracking(struct ocfs2_lock_res *res)
 258 {
 259         spin_lock(&ocfs2_dlm_tracking_lock);
 260         if (!list_empty(&res->l_debug_list))
 261                 list_del_init(&res->l_debug_list);
 262         spin_unlock(&ocfs2_dlm_tracking_lock);
 263 }
 264
 265 static void ocfs2_lock_res_init_common(struct ocfs2_super *osb,
 266                                        struct ocfs2_lock_res *res,
 267                                        enum ocfs2_lock_type type,
 268                                        u64 blkno,
 269                                        u32 generation,
 270                                        struct ocfs2_lock_res_ops *ops,
 271                                        void *priv)
 272 {
 273         ocfs2_build_lock_name(type, blkno, generation, res->l_name);
 274
 275         res->l_type          = type;
 276         res->l_ops           = ops;
 277         res->l_priv          = priv;
 278
 279         res->l_level         = LKM_IVMODE;
 280         res->l_requested     = LKM_IVMODE;
 281         res->l_blocking      = LKM_IVMODE;
 282         res->l_action        = OCFS2_AST_INVALID;
 283         res->l_unlock_action = OCFS2_UNLOCK_INVALID;
 284
 285         res->l_flags         = OCFS2_LOCK_INITIALIZED;
 286
 287         ocfs2_add_lockres_tracking(res, osb->osb_dlm_debug);
 288 }
 289
 290 void ocfs2_lock_res_init_once(struct ocfs2_lock_res *res)
 291 {
 292         /* This also clears out the lock status block */
 293         memset(res, 0, sizeof(struct ocfs2_lock_res));
 294         spin_lock_init(&res->l_lock);
 295         init_waitqueue_head(&res->l_event);
 296         INIT_LIST_HEAD(&res->l_blocked_list);
 297         INIT_LIST_HEAD(&res->l_mask_waiters);
 298 }
 299
 300 void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res,
 301                                enum ocfs2_lock_type type,
 302                                struct inode *inode)
 303 {
 304         struct ocfs2_lock_res_ops *ops;
 305
 306         switch(type) {
 307                 case OCFS2_LOCK_TYPE_RW:
 308                         ops = &ocfs2_inode_rw_lops;
 309                         break;
 310                 case OCFS2_LOCK_TYPE_META:
 311                         ops = &ocfs2_inode_meta_lops;
 312                         break;
 313                 case OCFS2_LOCK_TYPE_DATA:
 314                         ops = &ocfs2_inode_data_lops;
 315                         break;
 316                 default:
 317                         mlog_bug_on_msg(1, "type: %d\n", type);
 318                         ops = NULL; /* thanks, gcc */
 319                         break;
 320         };
 321
 322         ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), res, type,
 323                                    OCFS2_I(inode)->ip_blkno,
 324                                    inode->i_generation, ops, inode);
 325 }
 326
 327 static void ocfs2_super_lock_res_init(struct ocfs2_lock_res *res,
 328                                       struct ocfs2_super *osb)
 329 {
 330         /* Superblock lockres doesn't come from a slab so we call init
 331          * once on it manually.  */
 332         ocfs2_lock_res_init_once(res);
 333         ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_SUPER,
 334                                    OCFS2_SUPER_BLOCK_BLKNO, 0,
 335                                    &ocfs2_super_lops, osb);
 336 }
 337
 338 static void ocfs2_rename_lock_res_init(struct ocfs2_lock_res *res,
 339                                        struct ocfs2_super *osb)
 340 {
 341         /* Rename lockres doesn't come from a slab so we call init
 342          * once on it manually.  */
 343         ocfs2_lock_res_init_once(res);
 344         ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_RENAME, 0, 0,
 345                                    &ocfs2_rename_lops, osb);
 346 }
 347
 348 void ocfs2_lock_res_free(struct ocfs2_lock_res *res)
 349 {
 350         mlog_entry_void();
 351
 352         if (!(res->l_flags & OCFS2_LOCK_INITIALIZED))
 353                 return;
 354
 355         ocfs2_remove_lockres_tracking(res);
 356
 357         mlog_bug_on_msg(!list_empty(&res->l_blocked_list),
 358                         "Lockres %s is on the blocked list\n",
 359                         res->l_name);
 360         mlog_bug_on_msg(!list_empty(&res->l_mask_waiters),
 361                         "Lockres %s has mask waiters pending\n",
 362                         res->l_name);
 363         mlog_bug_on_msg(spin_is_locked(&res->l_lock),
 364                         "Lockres %s is locked\n",
 365                         res->l_name);
 366         mlog_bug_on_msg(res->l_ro_holders,
 367                         "Lockres %s has %u ro holders\n",
 368                         res->l_name, res->l_ro_holders);
 369         mlog_bug_on_msg(res->l_ex_holders,
 370                         "Lockres %s has %u ex holders\n",
 371                         res->l_name, res->l_ex_holders);
 372
 373         /* Need to clear out the lock status block for the dlm */
 374         memset(&res->l_lksb, 0, sizeof(res->l_lksb));
 375
 376         res->l_flags = 0UL;
 377         mlog_exit_void();
 378 }
 379
 380 static inline void ocfs2_inc_holders(struct ocfs2_lock_res *lockres,
 381                                      int level)
 382 {
 383         mlog_entry_void();
 384
 385         BUG_ON(!lockres);
 386
 387         switch(level) {
 388         case LKM_EXMODE:
 389                 lockres->l_ex_holders++;
 390                 break;
 391         case LKM_PRMODE:
 392                 lockres->l_ro_holders++;
 393                 break;
 394         default:
 395                 BUG();
 396         }
 397
 398         mlog_exit_void();
 399 }
 400
 401 static inline void ocfs2_dec_holders(struct ocfs2_lock_res *lockres,
 402                                      int level)
 403 {
 404         mlog_entry_void();
 405
 406         BUG_ON(!lockres);
 407
 408         switch(level) {
 409         case LKM_EXMODE:
 410                 BUG_ON(!lockres->l_ex_holders);
 411                 lockres->l_ex_holders--;
 412                 break;
 413         case LKM_PRMODE:
 414                 BUG_ON(!lockres->l_ro_holders);
 415                 lockres->l_ro_holders--;
 416                 break;
 417         default:
 418                 BUG();
 419         }
 420         mlog_exit_void();
 421 }
 422
 423 /* WARNING: This function lives in a world where the only three lock
 424  * levels are EX, PR, and NL. It *will* have to be adjusted when more
 425  * lock types are added. */
 426 static inline int ocfs2_highest_compat_lock_level(int level)
 427 {
 428         int new_level = LKM_EXMODE;
 429
 430         if (level == LKM_EXMODE)
 431                 new_level = LKM_NLMODE;
 432         else if (level == LKM_PRMODE)
 433                 new_level = LKM_PRMODE;
 434         return new_level;
 435 }
 436
 437 static void lockres_set_flags(struct ocfs2_lock_res *lockres,
 438                               unsigned long newflags)
 439 {
 440         struct list_head *pos, *tmp;
 441         struct ocfs2_mask_waiter *mw;
 442
 443         assert_spin_locked(&lockres->l_lock);
 444
 445         lockres->l_flags = newflags;
 446
 447         list_for_each_safe(pos, tmp, &lockres->l_mask_waiters) {
 448                 mw = list_entry(pos, struct ocfs2_mask_waiter, mw_item);
 449                 if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
 450                         continue;
 451
 452                 list_del_init(&mw->mw_item);
 453                 mw->mw_status = 0;
 454                 complete(&mw->mw_complete);
 455         }
 456 }
 457 static void lockres_or_flags(struct ocfs2_lock_res *lockres, unsigned long or)
 458 {
 459         lockres_set_flags(lockres, lockres->l_flags | or);
 460 }
 461 static void lockres_clear_flags(struct ocfs2_lock_res *lockres,
 462                                 unsigned long clear)
 463 {
 464         lockres_set_flags(lockres, lockres->l_flags & ~clear);
 465 }
 466
 467 static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres)
 468 {
 469         mlog_entry_void();
 470
 471         BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
 472         BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
 473         BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
 474         BUG_ON(lockres->l_blocking <= LKM_NLMODE);
 475
 476         lockres->l_level = lockres->l_requested;
 477         if (lockres->l_level <=
 478             ocfs2_highest_compat_lock_level(lockres->l_blocking)) {
 479                 lockres->l_blocking = LKM_NLMODE;
 480                 lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
 481         }
 482         lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
 483
 484         mlog_exit_void();
 485 }
 486
 487 static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres)
 488 {
 489         mlog_entry_void();
 490
 491         BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
 492         BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
 493
 494         /* Convert from RO to EX doesn't really need anything as our
 495          * information is already up to data. Convert from NL to
 496          * *anything* however should mark ourselves as needing an
 497          * update */
 498         if (lockres->l_level == LKM_NLMODE)
 499                 lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
 500
 501         lockres->l_level = lockres->l_requested;
 502         lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
 503
 504         mlog_exit_void();
 505 }
 506
 507 static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres)
 508 {
 509         mlog_entry_void();
 510
 511         BUG_ON((!lockres->l_flags & OCFS2_LOCK_BUSY));
 512         BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
 513
 514         if (lockres->l_requested > LKM_NLMODE &&
 515             !(lockres->l_flags & OCFS2_LOCK_LOCAL))
 516                 lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
 517
 518         lockres->l_level = lockres->l_requested;
 519         lockres_or_flags(lockres, OCFS2_LOCK_ATTACHED);
 520         lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
 521
 522         mlog_exit_void();
 523 }
 524
 525 static void ocfs2_inode_ast_func(void *opaque)
 526 {
 527         struct ocfs2_lock_res *lockres = opaque;
 528         struct inode *inode;
 529         struct dlm_lockstatus *lksb;
 530         unsigned long flags;
 531
 532         mlog_entry_void();
 533
 534         inode = ocfs2_lock_res_inode(lockres);
 535
 536         mlog(0, "AST fired for inode %llu, l_action = %u, type = %s\n",
 537              (unsigned long long)OCFS2_I(inode)->ip_blkno, lockres->l_action,
 538              ocfs2_lock_type_string(lockres->l_type));
 539
 540         BUG_ON(!ocfs2_is_inode_lock(lockres));
 541
 542         spin_lock_irqsave(&lockres->l_lock, flags);
 543
 544         lksb = &(lockres->l_lksb);
 545         if (lksb->status != DLM_NORMAL) {
 546                 mlog(ML_ERROR, "ocfs2_inode_ast_func: lksb status value of %u "
 547                      "on inode %llu\n", lksb->status,
 548                      (unsigned long long)OCFS2_I(inode)->ip_blkno);
 549                 spin_unlock_irqrestore(&lockres->l_lock, flags);
 550                 mlog_exit_void();
 551                 return;
 552         }
 553
 554         switch(lockres->l_action) {
 555         case OCFS2_AST_ATTACH:
 556                 ocfs2_generic_handle_attach_action(lockres);
 557                 lockres_clear_flags(lockres, OCFS2_LOCK_LOCAL);
 558                 break;
 559         case OCFS2_AST_CONVERT:
 560                 ocfs2_generic_handle_convert_action(lockres);
 561                 break;
 562         case OCFS2_AST_DOWNCONVERT:
 563                 ocfs2_generic_handle_downconvert_action(lockres);
 564                 break;
 565         default:
 566                 mlog(ML_ERROR, "lockres %s: ast fired with invalid action: %u "
 567                      "lockres flags = 0x%lx, unlock action: %u\n",
 568                      lockres->l_name, lockres->l_action, lockres->l_flags,
 569                      lockres->l_unlock_action);
 570
 571                 BUG();
 572         }
 573
 574         /* data and rw locking ignores refresh flag for now. */
 575         if (lockres->l_type != OCFS2_LOCK_TYPE_META)
 576                 lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
 577
 578         /* set it to something invalid so if we get called again we
 579          * can catch it. */
 580         lockres->l_action = OCFS2_AST_INVALID;
 581         spin_unlock_irqrestore(&lockres->l_lock, flags);
 582         wake_up(&lockres->l_event);
 583
 584         mlog_exit_void();
 585 }
 586
 587 static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres,
 588                                      int level)
 589 {
 590         int needs_downconvert = 0;
 591         mlog_entry_void();
 592
 593         assert_spin_locked(&lockres->l_lock);
 594
 595         lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
 596
 597         if (level > lockres->l_blocking) {
 598                 /* only schedule a downconvert if we haven't already scheduled
 599                  * one that goes low enough to satisfy the level we're
 600                  * blocking.  this also catches the case where we get
 601                  * duplicate BASTs */
 602                 if (ocfs2_highest_compat_lock_level(level) <
 603                     ocfs2_highest_compat_lock_level(lockres->l_blocking))
 604                         needs_downconvert = 1;
 605
 606                 lockres->l_blocking = level;
 607         }
 608
 609         mlog_exit(needs_downconvert);
 610         return needs_downconvert;
 611 }
 612
 613 static void ocfs2_generic_bast_func(struct ocfs2_super *osb,
 614                                     struct ocfs2_lock_res *lockres,
 615                                     int level)
 616 {
 617         int needs_downconvert;
 618         unsigned long flags;
 619
 620         mlog_entry_void();
 621
 622         BUG_ON(level <= LKM_NLMODE);
 623
 624         spin_lock_irqsave(&lockres->l_lock, flags);
 625         needs_downconvert = ocfs2_generic_handle_bast(lockres, level);
 626         if (needs_downconvert)
 627                 ocfs2_schedule_blocked_lock(osb, lockres);
 628         spin_unlock_irqrestore(&lockres->l_lock, flags);
 629
 630         ocfs2_kick_vote_thread(osb);
 631
 632         wake_up(&lockres->l_event);
 633         mlog_exit_void();
 634 }
 635
 636 static void ocfs2_inode_bast_func(void *opaque, int level)
 637 {
 638         struct ocfs2_lock_res *lockres = opaque;
 639         struct inode *inode;
 640         struct ocfs2_super *osb;
 641
 642         mlog_entry_void();
 643
 644         BUG_ON(!ocfs2_is_inode_lock(lockres));
 645
 646         inode = ocfs2_lock_res_inode(lockres);
 647         osb = OCFS2_SB(inode->i_sb);
 648
 649         mlog(0, "BAST fired for inode %llu, blocking %d, level %d type %s\n",
 650              (unsigned long long)OCFS2_I(inode)->ip_blkno, level,
 651              lockres->l_level, ocfs2_lock_type_string(lockres->l_type));
 652
 653         ocfs2_generic_bast_func(osb, lockres, level);
 654
 655         mlog_exit_void();
 656 }
 657
 658 static void ocfs2_generic_ast_func(struct ocfs2_lock_res *lockres,
 659                                    int ignore_refresh)
 660 {
 661         struct dlm_lockstatus *lksb = &lockres->l_lksb;
 662         unsigned long flags;
 663
 664         spin_lock_irqsave(&lockres->l_lock, flags);
 665
 666         if (lksb->status != DLM_NORMAL) {
 667                 mlog(ML_ERROR, "lockres %s: lksb status value of %u!\n",
 668                      lockres->l_name, lksb->status);
 669                 spin_unlock_irqrestore(&lockres->l_lock, flags);
 670                 return;
 671         }
 672
 673         switch(lockres->l_action) {
 674         case OCFS2_AST_ATTACH:
 675                 ocfs2_generic_handle_attach_action(lockres);
 676                 break;
 677         case OCFS2_AST_CONVERT:
 678                 ocfs2_generic_handle_convert_action(lockres);
 679                 break;
 680         case OCFS2_AST_DOWNCONVERT:
 681                 ocfs2_generic_handle_downconvert_action(lockres);
 682                 break;
 683         default:
 684                 BUG();
 685         }
 686
 687         if (ignore_refresh)
 688                 lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
 689
 690         /* set it to something invalid so if we get called again we
 691          * can catch it. */
 692         lockres->l_action = OCFS2_AST_INVALID;
 693         spin_unlock_irqrestore(&lockres->l_lock, flags);
 694
 695         wake_up(&lockres->l_event);
 696 }
 697
 698 static void ocfs2_super_ast_func(void *opaque)
 699 {
 700         struct ocfs2_lock_res *lockres = opaque;
 701
 702         mlog_entry_void();
 703         mlog(0, "Superblock AST fired\n");
 704
 705         BUG_ON(!ocfs2_is_super_lock(lockres));
 706         ocfs2_generic_ast_func(lockres, 0);
 707
 708         mlog_exit_void();
 709 }
 710
 711 static void ocfs2_super_bast_func(void *opaque,
 712                                   int level)
 713 {
 714         struct ocfs2_lock_res *lockres = opaque;
 715         struct ocfs2_super *osb;
 716
 717         mlog_entry_void();
 718         mlog(0, "Superblock BAST fired\n");
 719
 720         BUG_ON(!ocfs2_is_super_lock(lockres));
 721         osb = ocfs2_lock_res_super(lockres);
 722         ocfs2_generic_bast_func(osb, lockres, level);
 723
 724         mlog_exit_void();
 725 }
 726
 727 static void ocfs2_rename_ast_func(void *opaque)
 728 {
 729         struct ocfs2_lock_res *lockres = opaque;
 730
 731         mlog_entry_void();
 732
 733         mlog(0, "Rename AST fired\n");
 734
 735         BUG_ON(!ocfs2_is_rename_lock(lockres));
 736
 737         ocfs2_generic_ast_func(lockres, 1);
 738
 739         mlog_exit_void();
 740 }
 741
 742 static void ocfs2_rename_bast_func(void *opaque,
 743                                    int level)
 744 {
 745         struct ocfs2_lock_res *lockres = opaque;
 746         struct ocfs2_super *osb;
 747
 748         mlog_entry_void();
 749
 750         mlog(0, "Rename BAST fired\n");
 751
 752         BUG_ON(!ocfs2_is_rename_lock(lockres));
 753
 754         osb = ocfs2_lock_res_super(lockres);
 755         ocfs2_generic_bast_func(osb, lockres, level);
 756
 757         mlog_exit_void();
 758 }
 759
 760 static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
 761                                                 int convert)
 762 {
 763         unsigned long flags;
 764
 765         mlog_entry_void();
 766         spin_lock_irqsave(&lockres->l_lock, flags);
 767         lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
 768         if (convert)
 769                 lockres->l_action = OCFS2_AST_INVALID;
 770         else
 771                 lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
 772         spin_unlock_irqrestore(&lockres->l_lock, flags);
 773
 774         wake_up(&lockres->l_event);
 775         mlog_exit_void();
 776 }
 777
 778 /* Note: If we detect another process working on the lock (i.e.,
 779  * OCFS2_LOCK_BUSY), we'll bail out returning 0. It's up to the caller
 780  * to do the right thing in that case.
 781  */
 782 static int ocfs2_lock_create(struct ocfs2_super *osb,
 783                              struct ocfs2_lock_res *lockres,
 784                              int level,
 785                              int dlm_flags)
 786 {
 787         int ret = 0;
 788         enum dlm_status status;
 789         unsigned long flags;
 790
 791         mlog_entry_void();
 792
 793         mlog(0, "lock %s, level = %d, flags = %d\n", lockres->l_name, level,
 794              dlm_flags);
 795
 796         spin_lock_irqsave(&lockres->l_lock, flags);
 797         if ((lockres->l_flags & OCFS2_LOCK_ATTACHED) ||
 798             (lockres->l_flags & OCFS2_LOCK_BUSY)) {
 799                 spin_unlock_irqrestore(&lockres->l_lock, flags);
 800                 goto bail;
 801         }
 802
 803         lockres->l_action = OCFS2_AST_ATTACH;
 804         lockres->l_requested = level;
 805         lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
 806         spin_unlock_irqrestore(&lockres->l_lock, flags);
 807
 808         status = dlmlock(osb->dlm,
 809                          level,
 810                          &lockres->l_lksb,
 811                          dlm_flags,
 812                          lockres->l_name,
 813                          lockres->l_ops->ast,
 814                          lockres,
 815                          lockres->l_ops->bast);
 816         if (status != DLM_NORMAL) {
 817                 ocfs2_log_dlm_error("dlmlock", status, lockres);
 818                 ret = -EINVAL;
 819                 ocfs2_recover_from_dlm_error(lockres, 1);
 820         }
 821
 822         mlog(0, "lock %s, successfull return from dlmlock\n", lockres->l_name);
 823
 824 bail:
 825         mlog_exit(ret);
 826         return ret;
 827 }
 828
 829 static inline int ocfs2_check_wait_flag(struct ocfs2_lock_res *lockres,
 830                                         int flag)
 831 {
 832         unsigned long flags;
 833         int ret;
 834
 835         spin_lock_irqsave(&lockres->l_lock, flags);
 836         ret = lockres->l_flags & flag;
 837         spin_unlock_irqrestore(&lockres->l_lock, flags);
 838
 839         return ret;
 840 }
 841
 842 static inline void ocfs2_wait_on_busy_lock(struct ocfs2_lock_res *lockres)
 843
 844 {
 845         wait_event(lockres->l_event,
 846                    !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_BUSY));
 847 }
 848
 849 static inline void ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res *lockres)
 850
 851 {
 852         wait_event(lockres->l_event,
 853                    !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_REFRESHING));
 854 }
 855
 856 /* predict what lock level we'll be dropping down to on behalf
 857  * of another node, and return true if the currently wanted
 858  * level will be compatible with it. */
 859 static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
 860                                                      int wanted)
 861 {
 862         BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
 863
 864         return wanted <= ocfs2_highest_compat_lock_level(lockres->l_blocking);
 865 }
 866
 867 static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter *mw)
 868 {
 869         INIT_LIST_HEAD(&mw->mw_item);
 870         init_completion(&mw->mw_complete);
 871 }
 872
 873 static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter *mw)
 874 {
 875         wait_for_completion(&mw->mw_complete);
 876         /* Re-arm the completion in case we want to wait on it again */
 877         INIT_COMPLETION(mw->mw_complete);
 878         return mw->mw_status;
 879 }
 880
 881 static void lockres_add_mask_waiter(struct ocfs2_lock_res *lockres,
 882                                     struct ocfs2_mask_waiter *mw,
 883                                     unsigned long mask,
 884                                     unsigned long goal)
 885 {
 886         BUG_ON(!list_empty(&mw->mw_item));
 887
 888         assert_spin_locked(&lockres->l_lock);
 889
 890         list_add_tail(&mw->mw_item, &lockres->l_mask_waiters);
 891         mw->mw_mask = mask;
 892         mw->mw_goal = goal;
 893 }
 894
 895 /* returns 0 if the mw that was removed was already satisfied, -EBUSY
 896  * if the mask still hadn't reached its goal */
 897 static int lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres,
 898                                       struct ocfs2_mask_waiter *mw)
 899 {
 900         unsigned long flags;
 901         int ret = 0;
 902
 903         spin_lock_irqsave(&lockres->l_lock, flags);
 904         if (!list_empty(&mw->mw_item)) {
 905                 if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
 906                         ret = -EBUSY;
 907
 908                 list_del_init(&mw->mw_item);
 909                 init_completion(&mw->mw_complete);
 910         }
 911         spin_unlock_irqrestore(&lockres->l_lock, flags);
 912
 913         return ret;
 914
 915 }
 916
 917 static int ocfs2_cluster_lock(struct ocfs2_super *osb,
 918                               struct ocfs2_lock_res *lockres,
 919                               int level,
 920                               int lkm_flags,
 921                               int arg_flags)
 922 {
 923         struct ocfs2_mask_waiter mw;
 924         enum dlm_status status;
 925         int wait, catch_signals = !(osb->s_mount_opt & OCFS2_MOUNT_NOINTR);
 926         int ret = 0; /* gcc doesn't realize wait = 1 guarantees ret is set */
 927         unsigned long flags;
 928
 929         mlog_entry_void();
 930
 931         ocfs2_init_mask_waiter(&mw);
 932
 933 again:
 934         wait = 0;
 935
 936         if (catch_signals && signal_pending(current)) {
 937                 ret = -ERESTARTSYS;
 938                 goto out;
 939         }
 940
 941         spin_lock_irqsave(&lockres->l_lock, flags);
 942
 943         mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING,
 944                         "Cluster lock called on freeing lockres %s! flags "
 945                         "0x%lx\n", lockres->l_name, lockres->l_flags);
 946
 947         /* We only compare against the currently granted level
 948          * here. If the lock is blocked waiting on a downconvert,
 949          * we'll get caught below. */
 950         if (lockres->l_flags & OCFS2_LOCK_BUSY &&
 951             level > lockres->l_level) {
 952                 /* is someone sitting in dlm_lock? If so, wait on
 953                  * them. */
 954                 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
 955                 wait = 1;
 956                 goto unlock;
 957         }
 958
 959         if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
 960                 /* lock has not been created yet. */
 961                 spin_unlock_irqrestore(&lockres->l_lock, flags);
 962
 963                 ret = ocfs2_lock_create(osb, lockres, LKM_NLMODE, 0);
 964                 if (ret < 0) {
 965                         mlog_errno(ret);
 966                         goto out;
 967                 }
 968                 goto again;
 969         }
 970
 971         if (lockres->l_flags & OCFS2_LOCK_BLOCKED &&
 972             !ocfs2_may_continue_on_blocked_lock(lockres, level)) {
 973                 /* is the lock is currently blocked on behalf of
 974                  * another node */
 975                 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BLOCKED, 0);
 976                 wait = 1;
 977                 goto unlock;
 978         }
 979
 980         if (level > lockres->l_level) {
 981                 if (lockres->l_action != OCFS2_AST_INVALID)
 982                         mlog(ML_ERROR, "lockres %s has action %u pending\n",
 983                              lockres->l_name, lockres->l_action);
 984
 985                 lockres->l_action = OCFS2_AST_CONVERT;
 986                 lockres->l_requested = level;
 987                 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
 988                 spin_unlock_irqrestore(&lockres->l_lock, flags);
 989
 990                 BUG_ON(level == LKM_IVMODE);
 991                 BUG_ON(level == LKM_NLMODE);
 992
 993                 mlog(0, "lock %s, convert from %d to level = %d\n",
 994                      lockres->l_name, lockres->l_level, level);
 995
 996                 /* call dlm_lock to upgrade lock now */
 997                 status = dlmlock(osb->dlm,
 998                                  level,
 999                                  &lockres->l_lksb,
1000                                  lkm_flags|LKM_CONVERT|LKM_VALBLK,
1001                                  lockres->l_name,
1002                                  lockres->l_ops->ast,
1003                                  lockres,
1004                                  lockres->l_ops->bast);
1005                 if (status != DLM_NORMAL) {
1006                         if ((lkm_flags & LKM_NOQUEUE) &&
1007                             (status == DLM_NOTQUEUED))
1008                                 ret = -EAGAIN;
1009                         else {
1010                                 ocfs2_log_dlm_error("dlmlock", status,
1011                                                     lockres);
1012                                 ret = -EINVAL;
1013                         }
1014                         ocfs2_recover_from_dlm_error(lockres, 1);
1015                         goto out;
1016                 }
1017
1018                 mlog(0, "lock %s, successfull return from dlmlock\n",
1019                      lockres->l_name);
1020
1021                 /* At this point we've gone inside the dlm and need to
1022                  * complete our work regardless. */
1023                 catch_signals = 0;
1024
1025                 /* wait for busy to clear and carry on */
1026                 goto again;
1027         }
1028
1029         /* Ok, if we get here then we're good to go. */
1030         ocfs2_inc_holders(lockres, level);
1031
1032         ret = 0;
1033 unlock:
1034         spin_unlock_irqrestore(&lockres->l_lock, flags);
1035 out:
1036         /*
1037          * This is helping work around a lock inversion between the page lock
1038          * and dlm locks.  One path holds the page lock while calling aops
1039          * which block acquiring dlm locks.  The voting thread holds dlm
1040          * locks while acquiring page locks while down converting data locks.
1041          * This block is helping an aop path notice the inversion and back
1042          * off to unlock its page lock before trying the dlm lock again.
1043          */
1044         if (wait && arg_flags & OCFS2_LOCK_NONBLOCK &&
1045             mw.mw_mask & (OCFS2_LOCK_BUSY|OCFS2_LOCK_BLOCKED)) {
1046                 wait = 0;
1047                 if (lockres_remove_mask_waiter(lockres, &mw))
1048                         ret = -EAGAIN;
1049                 else
1050                         goto again;
1051         }
1052         if (wait) {
1053                 ret = ocfs2_wait_for_mask(&mw);
1054                 if (ret == 0)
1055                         goto again;
1056                 mlog_errno(ret);
1057         }
1058
1059         mlog_exit(ret);
1060         return ret;
1061 }
1062
1063 static void ocfs2_cluster_unlock(struct ocfs2_super *osb,
1064                                  struct ocfs2_lock_res *lockres,
1065                                  int level)
1066 {
1067         unsigned long flags;
1068
1069         mlog_entry_void();
1070         spin_lock_irqsave(&lockres->l_lock, flags);
1071         ocfs2_dec_holders(lockres, level);
1072         ocfs2_vote_on_unlock(osb, lockres);
1073         spin_unlock_irqrestore(&lockres->l_lock, flags);
1074         mlog_exit_void();
1075 }
1076
1077 static int ocfs2_create_new_inode_lock(struct inode *inode,
1078                                        struct ocfs2_lock_res *lockres)
1079 {
1080         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1081         unsigned long flags;
1082
1083         spin_lock_irqsave(&lockres->l_lock, flags);
1084         BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
1085         lockres_or_flags(lockres, OCFS2_LOCK_LOCAL);
1086         spin_unlock_irqrestore(&lockres->l_lock, flags);
1087
1088         return ocfs2_lock_create(osb, lockres, LKM_EXMODE, LKM_LOCAL);
1089 }
1090
1091 /* Grants us an EX lock on the data and metadata resources, skipping
1092  * the normal cluster directory lookup. Use this ONLY on newly created
1093  * inodes which other nodes can't possibly see, and which haven't been
1094  * hashed in the inode hash yet. This can give us a good performance
1095  * increase as it'll skip the network broadcast normally associated
1096  * with creating a new lock resource. */
1097 int ocfs2_create_new_inode_locks(struct inode *inode)
1098 {
1099         int ret;
1100
1101         BUG_ON(!inode);
1102         BUG_ON(!ocfs2_inode_is_new(inode));
1103
1104         mlog_entry_void();
1105
1106         mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
1107
1108         /* NOTE: That we don't increment any of the holder counts, nor
1109          * do we add anything to a journal handle. Since this is
1110          * supposed to be a new inode which the cluster doesn't know
1111          * about yet, there is no need to.  As far as the LVB handling
1112          * is concerned, this is basically like acquiring an EX lock
1113          * on a resource which has an invalid one -- we'll set it
1114          * valid when we release the EX. */
1115
1116         ret = ocfs2_create_new_inode_lock(inode,
1117                                           &OCFS2_I(inode)->ip_rw_lockres);
1118         if (ret) {
1119                 mlog_errno(ret);
1120                 goto bail;
1121         }
1122
1123         ret = ocfs2_create_new_inode_lock(inode,
1124                                           &OCFS2_I(inode)->ip_meta_lockres);
1125         if (ret) {
1126                 mlog_errno(ret);
1127                 goto bail;
1128         }
1129
1130         ret = ocfs2_create_new_inode_lock(inode,
1131                                           &OCFS2_I(inode)->ip_data_lockres);
1132         if (ret) {
1133                 mlog_errno(ret);
1134                 goto bail;
1135         }
1136
1137 bail:
1138         mlog_exit(ret);
1139         return ret;
1140 }
1141
1142 int ocfs2_rw_lock(struct inode *inode, int write)
1143 {
1144         int status, level;
1145         struct ocfs2_lock_res *lockres;
1146
1147         BUG_ON(!inode);
1148
1149         mlog_entry_void();
1150
1151         mlog(0, "inode %llu take %s RW lock\n",
1152              (unsigned long long)OCFS2_I(inode)->ip_blkno,
1153              write ? "EXMODE" : "PRMODE");
1154
1155         lockres = &OCFS2_I(inode)->ip_rw_lockres;
1156
1157         level = write ? LKM_EXMODE : LKM_PRMODE;
1158
1159         status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level, 0,
1160                                     0);
1161         if (status < 0)
1162                 mlog_errno(status);
1163
1164         mlog_exit(status);
1165         return status;
1166 }
1167
1168 void ocfs2_rw_unlock(struct inode *inode, int write)
1169 {
1170         int level = write ? LKM_EXMODE : LKM_PRMODE;
1171         struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_rw_lockres;
1172
1173         mlog_entry_void();
1174
1175         mlog(0, "inode %llu drop %s RW lock\n",
1176              (unsigned long long)OCFS2_I(inode)->ip_blkno,
1177              write ? "EXMODE" : "PRMODE");
1178
1179         ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
1180
1181         mlog_exit_void();
1182 }
1183
1184 int ocfs2_data_lock_full(struct inode *inode,
1185                          int write,
1186                          int arg_flags)
1187 {
1188         int status = 0, level;
1189         struct ocfs2_lock_res *lockres;
1190
1191         BUG_ON(!inode);
1192
1193         mlog_entry_void();
1194
1195         mlog(0, "inode %llu take %s DATA lock\n",
1196              (unsigned long long)OCFS2_I(inode)->ip_blkno,
1197              write ? "EXMODE" : "PRMODE");
1198
1199         /* We'll allow faking a readonly data lock for
1200          * rodevices. */
1201         if (ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb))) {
1202                 if (write) {
1203                         status = -EROFS;
1204                         mlog_errno(status);
1205                 }
1206                 goto out;
1207         }
1208
1209         lockres = &OCFS2_I(inode)->ip_data_lockres;
1210
1211         level = write ? LKM_EXMODE : LKM_PRMODE;
1212
1213         status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level,
1214                                     0, arg_flags);
1215         if (status < 0 && status != -EAGAIN)
1216                 mlog_errno(status);
1217
1218 out:
1219         mlog_exit(status);
1220         return status;
1221 }
1222
1223 /* see ocfs2_meta_lock_with_page() */
1224 int ocfs2_data_lock_with_page(struct inode *inode,
1225                               int write,
1226                               struct page *page)
1227 {
1228         int ret;
1229
1230         ret = ocfs2_data_lock_full(inode, write, OCFS2_LOCK_NONBLOCK);
1231         if (ret == -EAGAIN) {
1232                 unlock_page(page);
1233                 if (ocfs2_data_lock(inode, write) == 0)
1234                         ocfs2_data_unlock(inode, write);
1235                 ret = AOP_TRUNCATED_PAGE;
1236         }
1237
1238         return ret;
1239 }
1240
1241 static void ocfs2_vote_on_unlock(struct ocfs2_super *osb,
1242                                  struct ocfs2_lock_res *lockres)
1243 {
1244         int kick = 0;
1245
1246         mlog_entry_void();
1247
1248         /* If we know that another node is waiting on our lock, kick
1249          * the vote thread * pre-emptively when we reach a release
1250          * condition. */
1251         if (lockres->l_flags & OCFS2_LOCK_BLOCKED) {
1252                 switch(lockres->l_blocking) {
1253                 case LKM_EXMODE:
1254                         if (!lockres->l_ex_holders && !lockres->l_ro_holders)
1255                                 kick = 1;
1256                         break;
1257                 case LKM_PRMODE:
1258                         if (!lockres->l_ex_holders)
1259                                 kick = 1;
1260                         break;
1261                 default:
1262                         BUG();
1263                 }
1264         }
1265
1266         if (kick)
1267                 ocfs2_kick_vote_thread(osb);
1268
1269         mlog_exit_void();
1270 }
1271
1272 void ocfs2_data_unlock(struct inode *inode,
1273                        int write)
1274 {
1275         int level = write ? LKM_EXMODE : LKM_PRMODE;
1276         struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_data_lockres;
1277
1278         mlog_entry_void();
1279
1280         mlog(0, "inode %llu drop %s DATA lock\n",
1281              (unsigned long long)OCFS2_I(inode)->ip_blkno,
1282              write ? "EXMODE" : "PRMODE");
1283
1284         if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb)))
1285                 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
1286
1287         mlog_exit_void();
1288 }
1289
1290 #define OCFS2_SEC_BITS   34
1291 #define OCFS2_SEC_SHIFT  (64 - 34)
1292 #define OCFS2_NSEC_MASK  ((1ULL << OCFS2_SEC_SHIFT) - 1)
1293
1294 /* LVB only has room for 64 bits of time here so we pack it for
1295  * now. */
1296 static u64 ocfs2_pack_timespec(struct timespec *spec)
1297 {
1298         u64 res;
1299         u64 sec = spec->tv_sec;
1300         u32 nsec = spec->tv_nsec;
1301
1302         res = (sec << OCFS2_SEC_SHIFT) | (nsec & OCFS2_NSEC_MASK);
1303
1304         return res;
1305 }
1306
1307 /* Call this with the lockres locked. I am reasonably sure we don't
1308  * need ip_lock in this function as anyone who would be changing those
1309  * values is supposed to be blocked in ocfs2_meta_lock right now. */
1310 static void __ocfs2_stuff_meta_lvb(struct inode *inode)
1311 {
1312         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1313         struct ocfs2_lock_res *lockres = &oi->ip_meta_lockres;
1314         struct ocfs2_meta_lvb *lvb;
1315
1316         mlog_entry_void();
1317
1318         lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;
1319
1320         lvb->lvb_version   = cpu_to_be32(OCFS2_LVB_VERSION);
1321         lvb->lvb_isize     = cpu_to_be64(i_size_read(inode));
1322         lvb->lvb_iclusters = cpu_to_be32(oi->ip_clusters);
1323         lvb->lvb_iuid      = cpu_to_be32(inode->i_uid);
1324         lvb->lvb_igid      = cpu_to_be32(inode->i_gid);
1325         lvb->lvb_imode     = cpu_to_be16(inode->i_mode);
1326         lvb->lvb_inlink    = cpu_to_be16(inode->i_nlink);
1327         lvb->lvb_iatime_packed  =
1328                 cpu_to_be64(ocfs2_pack_timespec(&inode->i_atime));
1329         lvb->lvb_ictime_packed =
1330                 cpu_to_be64(ocfs2_pack_timespec(&inode->i_ctime));
1331         lvb->lvb_imtime_packed =
1332                 cpu_to_be64(ocfs2_pack_timespec(&inode->i_mtime));
1333
1334         mlog_meta_lvb(0, lockres);
1335
1336         mlog_exit_void();
1337 }
1338
1339 static void ocfs2_unpack_timespec(struct timespec *spec,
1340                                   u64 packed_time)
1341 {
1342         spec->tv_sec = packed_time >> OCFS2_SEC_SHIFT;
1343         spec->tv_nsec = packed_time & OCFS2_NSEC_MASK;
1344 }
1345
1346 static void ocfs2_refresh_inode_from_lvb(struct inode *inode)
1347 {
1348         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1349         struct ocfs2_lock_res *lockres = &oi->ip_meta_lockres;
1350         struct ocfs2_meta_lvb *lvb;
1351
1352         mlog_entry_void();
1353
1354         mlog_meta_lvb(0, lockres);
1355
1356         lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;
1357
1358         /* We're safe here without the lockres lock... */
1359         spin_lock(&oi->ip_lock);
1360         oi->ip_clusters = be32_to_cpu(lvb->lvb_iclusters);
1361         i_size_write(inode, be64_to_cpu(lvb->lvb_isize));
1362
1363         /* fast-symlinks are a special case */
1364         if (S_ISLNK(inode->i_mode) && !oi->ip_clusters)
1365                 inode->i_blocks = 0;
1366         else
1367                 inode->i_blocks =
1368                         ocfs2_align_bytes_to_sectors(i_size_read(inode));
1369
1370         inode->i_uid     = be32_to_cpu(lvb->lvb_iuid);
1371         inode->i_gid     = be32_to_cpu(lvb->lvb_igid);
1372         inode->i_mode    = be16_to_cpu(lvb->lvb_imode);
1373         inode->i_nlink   = be16_to_cpu(lvb->lvb_inlink);
1374         ocfs2_unpack_timespec(&inode->i_atime,
1375                               be64_to_cpu(lvb->lvb_iatime_packed));
1376         ocfs2_unpack_timespec(&inode->i_mtime,
1377                               be64_to_cpu(lvb->lvb_imtime_packed));
1378         ocfs2_unpack_timespec(&inode->i_ctime,
1379                               be64_to_cpu(lvb->lvb_ictime_packed));
1380         spin_unlock(&oi->ip_lock);
1381
1382         mlog_exit_void();
1383 }
1384
1385 static inline int ocfs2_meta_lvb_is_trustable(struct ocfs2_lock_res *lockres)
1386 {
1387         struct ocfs2_meta_lvb *lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;
1388
1389         if (be32_to_cpu(lvb->lvb_version) == OCFS2_LVB_VERSION)
1390                 return 1;
1391         return 0;
1392 }
1393
1394 /* Determine whether a lock resource needs to be refreshed, and
1395  * arbitrate who gets to refresh it.
1396  *
1397  *   0 means no refresh needed.
1398  *
1399  *   > 0 means you need to refresh this and you MUST call
1400  *   ocfs2_complete_lock_res_refresh afterwards. */
1401 static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res *lockres)
1402 {
1403         unsigned long flags;
1404         int status = 0;
1405
1406         mlog_entry_void();
1407
1408 refresh_check:
1409         spin_lock_irqsave(&lockres->l_lock, flags);
1410         if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) {
1411                 spin_unlock_irqrestore(&lockres->l_lock, flags);
1412                 goto bail;
1413         }
1414
1415         if (lockres->l_flags & OCFS2_LOCK_REFRESHING) {
1416                 spin_unlock_irqrestore(&lockres->l_lock, flags);
1417
1418                 ocfs2_wait_on_refreshing_lock(lockres);
1419                 goto refresh_check;
1420         }
1421
1422         /* Ok, I'll be the one to refresh this lock. */
1423         lockres_or_flags(lockres, OCFS2_LOCK_REFRESHING);
1424         spin_unlock_irqrestore(&lockres->l_lock, flags);
1425
1426         status = 1;
1427 bail:
1428         mlog_exit(status);
1429         return status;
1430 }
1431
1432 /* If status is non zero, I'll mark it as not being in refresh
1433  * anymroe, but i won't clear the needs refresh flag. */
1434 static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockres,
1435                                                    int status)
1436 {
1437         unsigned long flags;
1438         mlog_entry_void();
1439
1440         spin_lock_irqsave(&lockres->l_lock, flags);
1441         lockres_clear_flags(lockres, OCFS2_LOCK_REFRESHING);
1442         if (!status)
1443                 lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
1444         spin_unlock_irqrestore(&lockres->l_lock, flags);
1445
1446         wake_up(&lockres->l_event);
1447
1448         mlog_exit_void();
1449 }
1450
1451 /* may or may not return a bh if it went to disk. */
1452 static int ocfs2_meta_lock_update(struct inode *inode,
1453                                   struct buffer_head **bh)
1454 {
1455         int status = 0;
1456         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1457         struct ocfs2_lock_res *lockres;
1458         struct ocfs2_dinode *fe;
1459
1460         mlog_entry_void();
1461
1462         spin_lock(&oi->ip_lock);
1463         if (oi->ip_flags & OCFS2_INODE_DELETED) {
1464                 mlog(0, "Orphaned inode %llu was deleted while we "
1465                      "were waiting on a lock. ip_flags = 0x%x\n",
1466                      (unsigned long long)oi->ip_blkno, oi->ip_flags);
1467                 spin_unlock(&oi->ip_lock);
1468                 status = -ENOENT;
1469                 goto bail;
1470         }
1471         spin_unlock(&oi->ip_lock);
1472
1473         lockres = &oi->ip_meta_lockres;
1474
1475         if (!ocfs2_should_refresh_lock_res(lockres))
1476                 goto bail;
1477
1478         /* This will discard any caching information we might have had
1479          * for the inode metadata. */
1480         ocfs2_metadata_cache_purge(inode);
1481
1482         /* will do nothing for inode types that don't use the extent
1483          * map (directories, bitmap files, etc) */
1484         ocfs2_extent_map_trunc(inode, 0);
1485
1486         if (ocfs2_meta_lvb_is_trustable(lockres)) {
1487                 mlog(0, "Trusting LVB on inode %llu\n",
1488                      (unsigned long long)oi->ip_blkno);
1489                 ocfs2_refresh_inode_from_lvb(inode);
1490         } else {
1491                 /* Boo, we have to go to disk. */
1492                 /* read bh, cast, ocfs2_refresh_inode */
1493                 status = ocfs2_read_block(OCFS2_SB(inode->i_sb), oi->ip_blkno,
1494                                           bh, OCFS2_BH_CACHED, inode);
1495                 if (status < 0) {
1496                         mlog_errno(status);
1497                         goto bail_refresh;
1498                 }
1499                 fe = (struct ocfs2_dinode *) (*bh)->b_data;
1500
1501                 /* This is a good chance to make sure we're not
1502                  * locking an invalid object.
1503                  *
1504                  * We bug on a stale inode here because we checked
1505                  * above whether it was wiped from disk. The wiping
1506                  * node provides a guarantee that we receive that
1507                  * message and can mark the inode before dropping any
1508                  * locks associated with it. */
1509                 if (!OCFS2_IS_VALID_DINODE(fe)) {
1510                         OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
1511                         status = -EIO;
1512                         goto bail_refresh;
1513                 }
1514                 mlog_bug_on_msg(inode->i_generation !=
1515                                 le32_to_cpu(fe->i_generation),
1516                                 "Invalid dinode %llu disk generation: %u "
1517                                 "inode->i_generation: %u\n",
1518                                 (unsigned long long)oi->ip_blkno,
1519                                 le32_to_cpu(fe->i_generation),
1520                                 inode->i_generation);
1521                 mlog_bug_on_msg(le64_to_cpu(fe->i_dtime) ||
1522                                 !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)),
1523                                 "Stale dinode %llu dtime: %llu flags: 0x%x\n",
1524                                 (unsigned long long)oi->ip_blkno,
1525                                 (unsigned long long)le64_to_cpu(fe->i_dtime),
1526                                 le32_to_cpu(fe->i_flags));
1527
1528                 ocfs2_refresh_inode(inode, fe);
1529         }
1530
1531         status = 0;
1532 bail_refresh:
1533         ocfs2_complete_lock_res_refresh(lockres, status);
1534 bail:
1535         mlog_exit(status);
1536         return status;
1537 }
1538
1539 static int ocfs2_assign_bh(struct inode *inode,
1540                            struct buffer_head **ret_bh,
1541                            struct buffer_head *passed_bh)
1542 {
1543         int status;
1544
1545         if (passed_bh) {
1546                 /* Ok, the update went to disk for us, use the
1547                  * returned bh. */
1548                 *ret_bh = passed_bh;
1549                 get_bh(*ret_bh);
1550
1551                 return 0;
1552         }
1553
1554         status = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1555                                   OCFS2_I(inode)->ip_blkno,
1556                                   ret_bh,
1557                                   OCFS2_BH_CACHED,
1558                                   inode);
1559         if (status < 0)
1560                 mlog_errno(status);
1561
1562         return status;
1563 }
1564
1565 /*
1566  * returns < 0 error if the callback will never be called, otherwise
1567  * the result of the lock will be communicated via the callback.
1568  */
1569 int ocfs2_meta_lock_full(struct inode *inode,
1570                          struct ocfs2_journal_handle *handle,
1571                          struct buffer_head **ret_bh,
1572                          int ex,
1573                          int arg_flags)
1574 {
1575         int status, level, dlm_flags, acquired;
1576         struct ocfs2_lock_res *lockres;
1577         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1578         struct buffer_head *local_bh = NULL;
1579
1580         BUG_ON(!inode);
1581
1582         mlog_entry_void();
1583
1584         mlog(0, "inode %llu, take %s META lock\n",
1585              (unsigned long long)OCFS2_I(inode)->ip_blkno,
1586              ex ? "EXMODE" : "PRMODE");
1587
1588         status = 0;
1589         acquired = 0;
1590         /* We'll allow faking a readonly metadata lock for
1591          * rodevices. */
1592         if (ocfs2_is_hard_readonly(osb)) {
1593                 if (ex)
1594                         status = -EROFS;
1595                 goto bail;
1596         }
1597
1598         if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
1599                 wait_event(osb->recovery_event,
1600                            ocfs2_node_map_is_empty(osb, &osb->recovery_map));
1601
1602         acquired = 0;
1603         lockres = &OCFS2_I(inode)->ip_meta_lockres;
1604         level = ex ? LKM_EXMODE : LKM_PRMODE;
1605         dlm_flags = 0;
1606         if (arg_flags & OCFS2_META_LOCK_NOQUEUE)
1607                 dlm_flags |= LKM_NOQUEUE;
1608
1609         status = ocfs2_cluster_lock(osb, lockres, level, dlm_flags, arg_flags);
1610         if (status < 0) {
1611                 if (status != -EAGAIN && status != -EIOCBRETRY)
1612                         mlog_errno(status);
1613                 goto bail;
1614         }
1615
1616         /* Notify the error cleanup path to drop the cluster lock. */
1617         acquired = 1;
1618
1619         /* We wait twice because a node may have died while we were in
1620          * the lower dlm layers. The second time though, we've
1621          * committed to owning this lock so we don't allow signals to
1622          * abort the operation. */
1623         if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
1624                 wait_event(osb->recovery_event,
1625                            ocfs2_node_map_is_empty(osb, &osb->recovery_map));
1626
1627         /* This is fun. The caller may want a bh back, or it may
1628          * not. ocfs2_meta_lock_update definitely wants one in, but
1629          * may or may not read one, depending on what's in the
1630          * LVB. The result of all of this is that we've *only* gone to
1631          * disk if we have to, so the complexity is worthwhile. */
1632         status = ocfs2_meta_lock_update(inode, &local_bh);
1633         if (status < 0) {
1634                 if (status != -ENOENT)
1635                         mlog_errno(status);
1636                 goto bail;
1637         }
1638
1639         if (ret_bh) {
1640                 status = ocfs2_assign_bh(inode, ret_bh, local_bh);
1641                 if (status < 0) {
1642                         mlog_errno(status);
1643                         goto bail;
1644                 }
1645         }
1646
1647         if (handle) {
1648                 status = ocfs2_handle_add_lock(handle, inode);
1649                 if (status < 0)
1650                         mlog_errno(status);
1651         }
1652
1653 bail:
1654         if (status < 0) {
1655                 if (ret_bh && (*ret_bh)) {
1656                         brelse(*ret_bh);
1657                         *ret_bh = NULL;
1658                 }
1659                 if (acquired)
1660                         ocfs2_meta_unlock(inode, ex);
1661         }
1662
1663         if (local_bh)
1664                 brelse(local_bh);
1665
1666         mlog_exit(status);
1667         return status;
1668 }
1669
1670 /*
1671  * This is working around a lock inversion between tasks acquiring DLM locks
1672  * while holding a page lock and the vote thread which blocks dlm lock acquiry
1673  * while acquiring page locks.
1674  *
1675  * ** These _with_page variantes are only intended to be called from aop
1676  * methods that hold page locks and return a very specific *positive* error
1677  * code that aop methods pass up to the VFS -- test for errors with != 0. **
1678  *
1679  * The DLM is called such that it returns -EAGAIN if it would have blocked
1680  * waiting for the vote thread.  In that case we unlock our page so the vote
1681  * thread can make progress.  Once we've done this we have to return
1682  * AOP_TRUNCATED_PAGE so the aop method that called us can bubble that back up
1683  * into the VFS who will then immediately retry the aop call.
1684  *
1685  * We do a blocking lock and immediate unlock before returning, though, so that
1686  * the lock has a great chance of being cached on this node by the time the VFS
1687  * calls back to retry the aop.    This has a potential to livelock as nodes
1688  * ping locks back and forth, but that's a risk we're willing to take to avoid
1689  * the lock inversion simply.
1690  */
1691 int ocfs2_meta_lock_with_page(struct inode *inode,
1692                               struct ocfs2_journal_handle *handle,
1693                               struct buffer_head **ret_bh,
1694                               int ex,
1695                               struct page *page)
1696 {
1697         int ret;
1698
1699         ret = ocfs2_meta_lock_full(inode, handle, ret_bh, ex,
1700                                    OCFS2_LOCK_NONBLOCK);
1701         if (ret == -EAGAIN) {
1702                 unlock_page(page);
1703                 if (ocfs2_meta_lock(inode, handle, ret_bh, ex) == 0)
1704                         ocfs2_meta_unlock(inode, ex);
1705                 ret = AOP_TRUNCATED_PAGE;
1706         }
1707
1708         return ret;
1709 }
1710
1711 void ocfs2_meta_unlock(struct inode *inode,
1712                        int ex)
1713 {
1714         int level = ex ? LKM_EXMODE : LKM_PRMODE;
1715         struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_meta_lockres;
1716
1717         mlog_entry_void();
1718
1719         mlog(0, "inode %llu drop %s META lock\n",
1720              (unsigned long long)OCFS2_I(inode)->ip_blkno,
1721              ex ? "EXMODE" : "PRMODE");
1722
1723         if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb)))
1724                 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
1725
1726         mlog_exit_void();
1727 }
1728
1729 int ocfs2_super_lock(struct ocfs2_super *osb,
1730                      int ex)
1731 {
1732         int status;
1733         int level = ex ? LKM_EXMODE : LKM_PRMODE;
1734         struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
1735         struct buffer_head *bh;
1736         struct ocfs2_slot_info *si = osb->slot_info;
1737
1738         mlog_entry_void();
1739
1740         if (ocfs2_is_hard_readonly(osb))
1741                 return -EROFS;
1742
1743         status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
1744         if (status < 0) {
1745                 mlog_errno(status);
1746                 goto bail;
1747         }
1748
1749         /* The super block lock path is really in the best position to
1750          * know when resources covered by the lock need to be
1751          * refreshed, so we do it here. Of course, making sense of
1752          * everything is up to the caller :) */
1753         status = ocfs2_should_refresh_lock_res(lockres);
1754         if (status < 0) {
1755                 mlog_errno(status);
1756                 goto bail;
1757         }
1758         if (status) {
1759                 bh = si->si_bh;
1760                 status = ocfs2_read_block(osb, bh->b_blocknr, &bh, 0,
1761                                           si->si_inode);
1762                 if (status == 0)
1763                         ocfs2_update_slot_info(si);
1764
1765                 ocfs2_complete_lock_res_refresh(lockres, status);
1766
1767                 if (status < 0)
1768                         mlog_errno(status);
1769         }
1770 bail:
1771         mlog_exit(status);
1772         return status;
1773 }
1774
1775 void ocfs2_super_unlock(struct ocfs2_super *osb,
1776                         int ex)
1777 {
1778         int level = ex ? LKM_EXMODE : LKM_PRMODE;
1779         struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
1780
1781         ocfs2_cluster_unlock(osb, lockres, level);
1782 }
1783
1784 int ocfs2_rename_lock(struct ocfs2_super *osb)
1785 {
1786         int status;
1787         struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
1788
1789         if (ocfs2_is_hard_readonly(osb))
1790                 return -EROFS;
1791
1792         status = ocfs2_cluster_lock(osb, lockres, LKM_EXMODE, 0, 0);
1793         if (status < 0)
1794                 mlog_errno(status);
1795
1796         return status;
1797 }
1798
1799 void ocfs2_rename_unlock(struct ocfs2_super *osb)
1800 {
1801         struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
1802
1803         ocfs2_cluster_unlock(osb, lockres, LKM_EXMODE);
1804 }
1805
1806 /* Reference counting of the dlm debug structure. We want this because
1807  * open references on the debug inodes can live on after a mount, so
1808  * we can't rely on the ocfs2_super to always exist. */
1809 static void ocfs2_dlm_debug_free(struct kref *kref)
1810 {
1811         struct ocfs2_dlm_debug *dlm_debug;
1812
1813         dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt);
1814
1815         kfree(dlm_debug);
1816 }
1817
1818 void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug)
1819 {
1820         if (dlm_debug)
1821                 kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free);
1822 }
1823
1824 static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug)
1825 {
1826         kref_get(&debug->d_refcnt);
1827 }
1828
1829 struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void)
1830 {
1831         struct ocfs2_dlm_debug *dlm_debug;
1832
1833         dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL);
1834         if (!dlm_debug) {
1835                 mlog_errno(-ENOMEM);
1836                 goto out;
1837         }
1838
1839         kref_init(&dlm_debug->d_refcnt);
1840         INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking);
1841         dlm_debug->d_locking_state = NULL;
1842 out:
1843         return dlm_debug;
1844 }
1845
1846 /* Access to this is arbitrated for us via seq_file->sem. */
1847 struct ocfs2_dlm_seq_priv {
1848         struct ocfs2_dlm_debug *p_dlm_debug;
1849         struct ocfs2_lock_res p_iter_res;
1850         struct ocfs2_lock_res p_tmp_res;
1851 };
1852
1853 static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start,
1854                                                  struct ocfs2_dlm_seq_priv *priv)
1855 {
1856         struct ocfs2_lock_res *iter, *ret = NULL;
1857         struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug;
1858
1859         assert_spin_locked(&ocfs2_dlm_tracking_lock);
1860
1861         list_for_each_entry(iter, &start->l_debug_list, l_debug_list) {
1862                 /* discover the head of the list */
1863                 if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) {
1864                         mlog(0, "End of list found, %p\n", ret);
1865                         break;
1866                 }
1867
1868                 /* We track our "dummy" iteration lockres' by a NULL
1869                  * l_ops field. */
1870                 if (iter->l_ops != NULL) {
1871                         ret = iter;
1872                         break;
1873                 }
1874         }
1875
1876         return ret;
1877 }
1878
1879 static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos)
1880 {
1881         struct ocfs2_dlm_seq_priv *priv = m->private;
1882         struct ocfs2_lock_res *iter;
1883
1884         spin_lock(&ocfs2_dlm_tracking_lock);
1885         iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv);
1886         if (iter) {
1887                 /* Since lockres' have the lifetime of their container
1888                  * (which can be inodes, ocfs2_supers, etc) we want to
1889                  * copy this out to a temporary lockres while still
1890                  * under the spinlock. Obviously after this we can't
1891                  * trust any pointers on the copy returned, but that's
1892                  * ok as the information we want isn't typically held
1893                  * in them. */
1894                 priv->p_tmp_res = *iter;
1895                 iter = &priv->p_tmp_res;
1896         }
1897         spin_unlock(&ocfs2_dlm_tracking_lock);
1898
1899         return iter;
1900 }
1901
1902 static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v)
1903 {
1904 }
1905
1906 static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos)
1907 {
1908         struct ocfs2_dlm_seq_priv *priv = m->private;
1909         struct ocfs2_lock_res *iter = v;
1910         struct ocfs2_lock_res *dummy = &priv->p_iter_res;
1911
1912         spin_lock(&ocfs2_dlm_tracking_lock);
1913         iter = ocfs2_dlm_next_res(iter, priv);
1914         list_del_init(&dummy->l_debug_list);
1915         if (iter) {
1916                 list_add(&dummy->l_debug_list, &iter->l_debug_list);
1917                 priv->p_tmp_res = *iter;
1918                 iter = &priv->p_tmp_res;
1919         }
1920         spin_unlock(&ocfs2_dlm_tracking_lock);
1921
1922         return iter;
1923 }
1924
1925 /* So that debugfs.ocfs2 can determine which format is being used */
1926 #define OCFS2_DLM_DEBUG_STR_VERSION 1
1927 static int ocfs2_dlm_seq_show(struct seq_file *m, void *v)
1928 {
1929         int i;
1930         char *lvb;
1931         struct ocfs2_lock_res *lockres = v;
1932
1933         if (!lockres)
1934                 return -EINVAL;
1935
1936         seq_printf(m, "0x%x\t"
1937                    "%.*s\t"
1938                    "%d\t"
1939                    "0x%lx\t"
1940                    "0x%x\t"
1941                    "0x%x\t"
1942                    "%u\t"
1943                    "%u\t"
1944                    "%d\t"
1945                    "%d\t",
1946                    OCFS2_DLM_DEBUG_STR_VERSION,
1947                    OCFS2_LOCK_ID_MAX_LEN, lockres->l_name,
1948                    lockres->l_level,
1949                    lockres->l_flags,
1950                    lockres->l_action,
1951                    lockres->l_unlock_action,
1952                    lockres->l_ro_holders,
1953                    lockres->l_ex_holders,
1954                    lockres->l_requested,
1955                    lockres->l_blocking);
1956
1957         /* Dump the raw LVB */
1958         lvb = lockres->l_lksb.lvb;
1959         for(i = 0; i < DLM_LVB_LEN; i++)
1960                 seq_printf(m, "0x%x\t", lvb[i]);
1961
1962         /* End the line */
1963         seq_printf(m, "\n");
1964         return 0;
1965 }
1966
1967 static struct seq_operations ocfs2_dlm_seq_ops = {
1968         .start =        ocfs2_dlm_seq_start,
1969         .stop =         ocfs2_dlm_seq_stop,
1970         .next =         ocfs2_dlm_seq_next,
1971         .show =         ocfs2_dlm_seq_show,
1972 };
1973
1974 static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file)
1975 {
1976         struct seq_file *seq = (struct seq_file *) file->private_data;
1977         struct ocfs2_dlm_seq_priv *priv = seq->private;
1978         struct ocfs2_lock_res *res = &priv->p_iter_res;
1979
1980         ocfs2_remove_lockres_tracking(res);
1981         ocfs2_put_dlm_debug(priv->p_dlm_debug);
1982         return seq_release_private(inode, file);
1983 }
1984
1985 static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file)
1986 {
1987         int ret;
1988         struct ocfs2_dlm_seq_priv *priv;
1989         struct seq_file *seq;
1990         struct ocfs2_super *osb;
1991
1992         priv = kzalloc(sizeof(struct ocfs2_dlm_seq_priv), GFP_KERNEL);
1993         if (!priv) {
1994                 ret = -ENOMEM;
1995                 mlog_errno(ret);
1996                 goto out;
1997         }
1998         osb = (struct ocfs2_super *) inode->u.generic_ip;
1999         ocfs2_get_dlm_debug(osb->osb_dlm_debug);
2000         priv->p_dlm_debug = osb->osb_dlm_debug;
2001         INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list);
2002
2003         ret = seq_open(file, &ocfs2_dlm_seq_ops);
2004         if (ret) {
2005                 kfree(priv);
2006                 mlog_errno(ret);
2007                 goto out;
2008         }
2009
2010         seq = (struct seq_file *) file->private_data;
2011         seq->private = priv;
2012
2013         ocfs2_add_lockres_tracking(&priv->p_iter_res,
2014                                    priv->p_dlm_debug);
2015
2016 out:
2017         return ret;
2018 }
2019
2020 static const struct file_operations ocfs2_dlm_debug_fops = {
2021         .open =         ocfs2_dlm_debug_open,
2022         .release =      ocfs2_dlm_debug_release,
2023         .read =         seq_read,
2024         .llseek =       seq_lseek,
2025 };
2026
2027 static int ocfs2_dlm_init_debug(struct ocfs2_super *osb)
2028 {
2029         int ret = 0;
2030         struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
2031
2032         dlm_debug->d_locking_state = debugfs_create_file("locking_state",
2033                                                          S_IFREG|S_IRUSR,
2034                                                          osb->osb_debug_root,
2035                                                          osb,
2036                                                          &ocfs2_dlm_debug_fops);
2037         if (!dlm_debug->d_locking_state) {
2038                 ret = -EINVAL;
2039                 mlog(ML_ERROR,
2040                      "Unable to create locking state debugfs file.\n");
2041                 goto out;
2042         }
2043
2044         ocfs2_get_dlm_debug(dlm_debug);
2045 out:
2046         return ret;
2047 }
2048
2049 static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb)
2050 {
2051         struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
2052
2053         if (dlm_debug) {
2054                 debugfs_remove(dlm_debug->d_locking_state);
2055                 ocfs2_put_dlm_debug(dlm_debug);
2056         }
2057 }
2058
2059 int ocfs2_dlm_init(struct ocfs2_super *osb)
2060 {
2061         int status;
2062         u32 dlm_key;
2063         struct dlm_ctxt *dlm;
2064
2065         mlog_entry_void();
2066
2067         status = ocfs2_dlm_init_debug(osb);
2068         if (status < 0) {
2069                 mlog_errno(status);
2070                 goto bail;
2071         }
2072
2073         /* launch vote thread */
2074         osb->vote_task = kthread_run(ocfs2_vote_thread, osb, "ocfs2vote-%d",
2075                                      osb->osb_id);
2076         if (IS_ERR(osb->vote_task)) {
2077                 status = PTR_ERR(osb->vote_task);
2078                 osb->vote_task = NULL;
2079                 mlog_errno(status);
2080                 goto bail;
2081         }
2082
2083         /* used by the dlm code to make message headers unique, each
2084          * node in this domain must agree on this. */
2085         dlm_key = crc32_le(0, osb->uuid_str, strlen(osb->uuid_str));
2086
2087         /* for now, uuid == domain */
2088         dlm = dlm_register_domain(osb->uuid_str, dlm_key);
2089         if (IS_ERR(dlm)) {
2090                 status = PTR_ERR(dlm);
2091                 mlog_errno(status);
2092                 goto bail;
2093         }
2094
2095         ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb);
2096         ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb);
2097
2098         dlm_register_eviction_cb(dlm, &osb->osb_eviction_cb);
2099
2100         osb->dlm = dlm;
2101
2102         status = 0;
2103 bail:
2104         if (status < 0) {
2105                 ocfs2_dlm_shutdown_debug(osb);
2106                 if (osb->vote_task)
2107                         kthread_stop(osb->vote_task);
2108         }
2109
2110         mlog_exit(status);
2111         return status;
2112 }
2113
2114 void ocfs2_dlm_shutdown(struct ocfs2_super *osb)
2115 {
2116         mlog_entry_void();
2117
2118         dlm_unregister_eviction_cb(&osb->osb_eviction_cb);
2119
2120         ocfs2_drop_osb_locks(osb);
2121
2122         if (osb->vote_task) {
2123                 kthread_stop(osb->vote_task);
2124                 osb->vote_task = NULL;
2125         }
2126
2127         ocfs2_lock_res_free(&osb->osb_super_lockres);
2128         ocfs2_lock_res_free(&osb->osb_rename_lockres);
2129
2130         dlm_unregister_domain(osb->dlm);
2131         osb->dlm = NULL;
2132
2133         ocfs2_dlm_shutdown_debug(osb);
2134
2135         mlog_exit_void();
2136 }
2137
2138 static void ocfs2_unlock_ast_func(void *opaque, enum dlm_status status)
2139 {
2140         struct ocfs2_lock_res *lockres = opaque;
2141         unsigned long flags;
2142
2143         mlog_entry_void();
2144
2145         mlog(0, "UNLOCK AST called on lock %s, action = %d\n", lockres->l_name,
2146              lockres->l_unlock_action);
2147
2148         spin_lock_irqsave(&lockres->l_lock, flags);
2149         /* We tried to cancel a convert request, but it was already
2150          * granted. All we want to do here is clear our unlock
2151          * state. The wake_up call done at the bottom is redundant
2152          * (ocfs2_prepare_cancel_convert doesn't sleep on this) but doesn't
2153          * hurt anything anyway */
2154         if (status == DLM_CANCELGRANT &&
2155             lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
2156                 mlog(0, "Got cancelgrant for %s\n", lockres->l_name);
2157
2158                 /* We don't clear the busy flag in this case as it
2159                  * should have been cleared by the ast which the dlm
2160                  * has called. */
2161                 goto complete_unlock;
2162         }
2163
2164         if (status != DLM_NORMAL) {
2165                 mlog(ML_ERROR, "Dlm passes status %d for lock %s, "
2166                      "unlock_action %d\n", status, lockres->l_name,
2167                      lockres->l_unlock_action);
2168                 spin_unlock_irqrestore(&lockres->l_lock, flags);
2169                 return;
2170         }
2171
2172         switch(lockres->l_unlock_action) {
2173         case OCFS2_UNLOCK_CANCEL_CONVERT:
2174                 mlog(0, "Cancel convert success for %s\n", lockres->l_name);
2175                 lockres->l_action = OCFS2_AST_INVALID;
2176                 break;
2177         case OCFS2_UNLOCK_DROP_LOCK:
2178                 lockres->l_level = LKM_IVMODE;
2179                 break;
2180         default:
2181                 BUG();
2182         }
2183
2184         lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
2185 complete_unlock:
2186         lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
2187         spin_unlock_irqrestore(&lockres->l_lock, flags);
2188
2189         wake_up(&lockres->l_event);
2190
2191         mlog_exit_void();
2192 }
2193
2194 typedef void (ocfs2_pre_drop_cb_t)(struct ocfs2_lock_res *, void *);
2195
2196 struct drop_lock_cb {
2197         ocfs2_pre_drop_cb_t     *drop_func;
2198         void                    *drop_data;
2199 };
2200
2201 static int ocfs2_drop_lock(struct ocfs2_super *osb,
2202                            struct ocfs2_lock_res *lockres,
2203                            struct drop_lock_cb *dcb)
2204 {
2205         enum dlm_status status;
2206         unsigned long flags;
2207
2208         /* We didn't get anywhere near actually using this lockres. */
2209         if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED))
2210                 goto out;
2211
2212         spin_lock_irqsave(&lockres->l_lock, flags);
2213
2214         mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING),
2215                         "lockres %s, flags 0x%lx\n",
2216                         lockres->l_name, lockres->l_flags);
2217
2218         while (lockres->l_flags & OCFS2_LOCK_BUSY) {
2219                 mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = "
2220                      "%u, unlock_action = %u\n",
2221                      lockres->l_name, lockres->l_flags, lockres->l_action,
2222                      lockres->l_unlock_action);
2223
2224                 spin_unlock_irqrestore(&lockres->l_lock, flags);
2225
2226                 /* XXX: Today we just wait on any busy
2227                  * locks... Perhaps we need to cancel converts in the
2228                  * future? */
2229                 ocfs2_wait_on_busy_lock(lockres);
2230
2231                 spin_lock_irqsave(&lockres->l_lock, flags);
2232         }
2233
2234         if (dcb)
2235                 dcb->drop_func(lockres, dcb->drop_data);
2236
2237         if (lockres->l_flags & OCFS2_LOCK_BUSY)
2238                 mlog(ML_ERROR, "destroying busy lock: \"%s\"\n",
2239                      lockres->l_name);
2240         if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
2241                 mlog(0, "destroying blocked lock: \"%s\"\n", lockres->l_name);
2242
2243         if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
2244                 spin_unlock_irqrestore(&lockres->l_lock, flags);
2245                 goto out;
2246         }
2247
2248         lockres_clear_flags(lockres, OCFS2_LOCK_ATTACHED);
2249
2250         /* make sure we never get here while waiting for an ast to
2251          * fire. */
2252         BUG_ON(lockres->l_action != OCFS2_AST_INVALID);
2253
2254         /* is this necessary? */
2255         lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
2256         lockres->l_unlock_action = OCFS2_UNLOCK_DROP_LOCK;
2257         spin_unlock_irqrestore(&lockres->l_lock, flags);
2258
2259         mlog(0, "lock %s\n", lockres->l_name);
2260
2261         status = dlmunlock(osb->dlm, &lockres->l_lksb, LKM_VALBLK,
2262                            lockres->l_ops->unlock_ast, lockres);
2263         if (status != DLM_NORMAL) {
2264                 ocfs2_log_dlm_error("dlmunlock", status, lockres);
2265                 mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags);
2266                 dlm_print_one_lock(lockres->l_lksb.lockid);
2267                 BUG();
2268         }
2269         mlog(0, "lock %s, successfull return from dlmunlock\n",
2270              lockres->l_name);
2271
2272         ocfs2_wait_on_busy_lock(lockres);
2273 out:
2274         mlog_exit(0);
2275         return 0;
2276 }
2277
2278 /* Mark the lockres as being dropped. It will no longer be
2279  * queued if blocking, but we still may have to wait on it
2280  * being dequeued from the vote thread before we can consider
2281  * it safe to drop.
2282  *
2283  * You can *not* attempt to call cluster_lock on this lockres anymore. */
2284 void ocfs2_mark_lockres_freeing(struct ocfs2_lock_res *lockres)
2285 {
2286         int status;
2287         struct ocfs2_mask_waiter mw;
2288         unsigned long flags;
2289
2290         ocfs2_init_mask_waiter(&mw);
2291
2292         spin_lock_irqsave(&lockres->l_lock, flags);
2293         lockres->l_flags |= OCFS2_LOCK_FREEING;
2294         while (lockres->l_flags & OCFS2_LOCK_QUEUED) {
2295                 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0);
2296                 spin_unlock_irqrestore(&lockres->l_lock, flags);
2297
2298                 mlog(0, "Waiting on lockres %s\n", lockres->l_name);
2299
2300                 status = ocfs2_wait_for_mask(&mw);
2301                 if (status)
2302                         mlog_errno(status);
2303
2304                 spin_lock_irqsave(&lockres->l_lock, flags);
2305         }
2306         spin_unlock_irqrestore(&lockres->l_lock, flags);
2307 }
2308
2309 static void ocfs2_drop_osb_locks(struct ocfs2_super *osb)
2310 {
2311         int status;
2312
2313         mlog_entry_void();
2314
2315         ocfs2_mark_lockres_freeing(&osb->osb_super_lockres);
2316
2317         status = ocfs2_drop_lock(osb, &osb->osb_super_lockres, NULL);
2318         if (status < 0)
2319                 mlog_errno(status);
2320
2321         ocfs2_mark_lockres_freeing(&osb->osb_rename_lockres);
2322
2323         status = ocfs2_drop_lock(osb, &osb->osb_rename_lockres, NULL);
2324         if (status < 0)
2325                 mlog_errno(status);
2326
2327         mlog_exit(status);
2328 }
2329
2330 static void ocfs2_meta_pre_drop(struct ocfs2_lock_res *lockres, void *data)
2331 {
2332         struct inode *inode = data;
2333
2334         /* the metadata lock requires a bit more work as we have an
2335          * LVB to worry about. */
2336         if (lockres->l_flags & OCFS2_LOCK_ATTACHED &&
2337             lockres->l_level == LKM_EXMODE &&
2338             !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
2339                 __ocfs2_stuff_meta_lvb(inode);
2340 }
2341
2342 int ocfs2_drop_inode_locks(struct inode *inode)
2343 {
2344         int status, err;
2345         struct drop_lock_cb meta_dcb = { ocfs2_meta_pre_drop, inode, };
2346
2347         mlog_entry_void();
2348
2349         /* No need to call ocfs2_mark_lockres_freeing here -
2350          * ocfs2_clear_inode has done it for us. */
2351
2352         err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
2353                               &OCFS2_I(inode)->ip_data_lockres,
2354                               NULL);
2355         if (err < 0)
2356                 mlog_errno(err);
2357
2358         status = err;
2359
2360         err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
2361                               &OCFS2_I(inode)->ip_meta_lockres,
2362                               &meta_dcb);
2363         if (err < 0)
2364                 mlog_errno(err);
2365         if (err < 0 && !status)
2366                 status = err;
2367
2368         err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
2369                               &OCFS2_I(inode)->ip_rw_lockres,
2370                               NULL);
2371         if (err < 0)
2372                 mlog_errno(err);
2373         if (err < 0 && !status)
2374                 status = err;
2375
2376         mlog_exit(status);
2377         return status;
2378 }
2379
2380 static void ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
2381                                       int new_level)
2382 {
2383         assert_spin_locked(&lockres->l_lock);
2384
2385         BUG_ON(lockres->l_blocking <= LKM_NLMODE);
2386
2387         if (lockres->l_level <= new_level) {
2388                 mlog(ML_ERROR, "lockres->l_level (%u) <= new_level (%u)\n",
2389                      lockres->l_level, new_level);
2390                 BUG();
2391         }
2392
2393         mlog(0, "lock %s, new_level = %d, l_blocking = %d\n",
2394              lockres->l_name, new_level, lockres->l_blocking);
2395
2396         lockres->l_action = OCFS2_AST_DOWNCONVERT;
2397         lockres->l_requested = new_level;
2398         lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
2399 }
2400
2401 static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
2402                                   struct ocfs2_lock_res *lockres,
2403                                   int new_level,
2404                                   int lvb)
2405 {
2406         int ret, dlm_flags = LKM_CONVERT;
2407         enum dlm_status status;
2408
2409         mlog_entry_void();
2410
2411         if (lvb)
2412                 dlm_flags |= LKM_VALBLK;
2413
2414         status = dlmlock(osb->dlm,
2415                          new_level,
2416                          &lockres->l_lksb,
2417                          dlm_flags,
2418                          lockres->l_name,
2419                          lockres->l_ops->ast,
2420                          lockres,
2421                          lockres->l_ops->bast);
2422         if (status != DLM_NORMAL) {
2423                 ocfs2_log_dlm_error("dlmlock", status, lockres);
2424                 ret = -EINVAL;
2425                 ocfs2_recover_from_dlm_error(lockres, 1);
2426                 goto bail;
2427         }
2428
2429         ret = 0;
2430 bail:
2431         mlog_exit(ret);
2432         return ret;
2433 }
2434
2435 /* returns 1 when the caller should unlock and call dlmunlock */
2436 static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
2437                                         struct ocfs2_lock_res *lockres)
2438 {
2439         assert_spin_locked(&lockres->l_lock);
2440
2441         mlog_entry_void();
2442         mlog(0, "lock %s\n", lockres->l_name);
2443
2444         if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
2445                 /* If we're already trying to cancel a lock conversion
2446                  * then just drop the spinlock and allow the caller to
2447                  * requeue this lock. */
2448
2449                 mlog(0, "Lockres %s, skip convert\n", lockres->l_name);
2450                 return 0;
2451         }
2452
2453         /* were we in a convert when we got the bast fire? */
2454         BUG_ON(lockres->l_action != OCFS2_AST_CONVERT &&
2455                lockres->l_action != OCFS2_AST_DOWNCONVERT);
2456         /* set things up for the unlockast to know to just
2457          * clear out the ast_action and unset busy, etc. */
2458         lockres->l_unlock_action = OCFS2_UNLOCK_CANCEL_CONVERT;
2459
2460         mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_BUSY),
2461                         "lock %s, invalid flags: 0x%lx\n",
2462                         lockres->l_name, lockres->l_flags);
2463
2464         return 1;
2465 }
2466
2467 static int ocfs2_cancel_convert(struct ocfs2_super *osb,
2468                                 struct ocfs2_lock_res *lockres)
2469 {
2470         int ret;
2471         enum dlm_status status;
2472
2473         mlog_entry_void();
2474         mlog(0, "lock %s\n", lockres->l_name);
2475
2476         ret = 0;
2477         status = dlmunlock(osb->dlm,
2478                            &lockres->l_lksb,
2479                            LKM_CANCEL,
2480                            lockres->l_ops->unlock_ast,
2481                            lockres);
2482         if (status != DLM_NORMAL) {
2483                 ocfs2_log_dlm_error("dlmunlock", status, lockres);
2484                 ret = -EINVAL;
2485                 ocfs2_recover_from_dlm_error(lockres, 0);
2486         }
2487
2488         mlog(0, "lock %s return from dlmunlock\n", lockres->l_name);
2489
2490         mlog_exit(ret);
2491         return ret;
2492 }
2493
2494 static inline int ocfs2_can_downconvert_meta_lock(struct inode *inode,
2495                                                   struct ocfs2_lock_res *lockres,
2496                                                   int new_level)
2497 {
2498         int ret;
2499
2500         mlog_entry_void();
2501
2502         BUG_ON(new_level != LKM_NLMODE && new_level != LKM_PRMODE);
2503
2504         if (lockres->l_flags & OCFS2_LOCK_REFRESHING) {
2505                 ret = 0;
2506                 mlog(0, "lockres %s currently being refreshed -- backing "
2507                      "off!\n", lockres->l_name);
2508         } else if (new_level == LKM_PRMODE)
2509                 ret = !lockres->l_ex_holders &&
2510                         ocfs2_inode_fully_checkpointed(inode);
2511         else /* Must be NLMODE we're converting to. */
2512                 ret = !lockres->l_ro_holders && !lockres->l_ex_holders &&
2513                         ocfs2_inode_fully_checkpointed(inode);
2514
2515         mlog_exit(ret);
2516         return ret;
2517 }
2518
2519 static int ocfs2_do_unblock_meta(struct inode *inode,
2520                                  int *requeue)
2521 {
2522         int new_level;
2523         int set_lvb = 0;
2524         int ret = 0;
2525         struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_meta_lockres;
2526         unsigned long flags;
2527
2528         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2529
2530         mlog_entry_void();
2531
2532         spin_lock_irqsave(&lockres->l_lock, flags);
2533
2534         BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
2535
2536         mlog(0, "l_level=%d, l_blocking=%d\n", lockres->l_level,
2537              lockres->l_blocking);
2538
2539         BUG_ON(lockres->l_level != LKM_EXMODE &&
2540                lockres->l_level != LKM_PRMODE);
2541
2542         if (lockres->l_flags & OCFS2_LOCK_BUSY) {
2543                 *requeue = 1;
2544                 ret = ocfs2_prepare_cancel_convert(osb, lockres);
2545                 spin_unlock_irqrestore(&lockres->l_lock, flags);
2546                 if (ret) {
2547                         ret = ocfs2_cancel_convert(osb, lockres);
2548                         if (ret < 0)
2549                                 mlog_errno(ret);
2550                 }
2551                 goto leave;
2552         }
2553
2554         new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking);
2555
2556         mlog(0, "l_level=%d, l_blocking=%d, new_level=%d\n",
2557              lockres->l_level, lockres->l_blocking, new_level);
2558
2559         if (ocfs2_can_downconvert_meta_lock(inode, lockres, new_level)) {
2560                 if (lockres->l_level == LKM_EXMODE)
2561                         set_lvb = 1;
2562
2563                 /* If the lock hasn't been refreshed yet (rare), then
2564                  * our memory inode values are old and we skip
2565                  * stuffing the lvb. There's no need to actually clear
2566                  * out the lvb here as it's value is still valid. */
2567                 if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) {
2568                         if (set_lvb)
2569                                 __ocfs2_stuff_meta_lvb(inode);
2570                 } else
2571                         mlog(0, "lockres %s: downconverting stale lock!\n",
2572                              lockres->l_name);
2573
2574                 mlog(0, "calling ocfs2_downconvert_lock with l_level=%d, "
2575                      "l_blocking=%d, new_level=%d\n",
2576                      lockres->l_level, lockres->l_blocking, new_level);
2577
2578                 ocfs2_prepare_downconvert(lockres, new_level);
2579                 spin_unlock_irqrestore(&lockres->l_lock, flags);
2580                 ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb);
2581                 goto leave;
2582         }
2583         if (!ocfs2_inode_fully_checkpointed(inode))
2584                 ocfs2_start_checkpoint(osb);
2585
2586         *requeue = 1;
2587         spin_unlock_irqrestore(&lockres->l_lock, flags);
2588         ret = 0;
2589 leave:
2590         mlog_exit(ret);
2591         return ret;
2592 }
2593
2594 static int ocfs2_generic_unblock_lock(struct ocfs2_super *osb,
2595                                       struct ocfs2_lock_res *lockres,
2596                                       int *requeue,
2597                                       ocfs2_convert_worker_t *worker)
2598 {
2599         unsigned long flags;
2600         int blocking;
2601         int new_level;
2602         int ret = 0;
2603
2604         mlog_entry_void();
2605
2606         spin_lock_irqsave(&lockres->l_lock, flags);
2607
2608         BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
2609
2610 recheck:
2611         if (lockres->l_flags & OCFS2_LOCK_BUSY) {
2612                 *requeue = 1;
2613                 ret = ocfs2_prepare_cancel_convert(osb, lockres);
2614                 spin_unlock_irqrestore(&lockres->l_lock, flags);
2615                 if (ret) {
2616                         ret = ocfs2_cancel_convert(osb, lockres);
2617                         if (ret < 0)
2618                                 mlog_errno(ret);
2619                 }
2620                 goto leave;
2621         }
2622
2623         /* if we're blocking an exclusive and we have *any* holders,
2624          * then requeue. */
2625         if ((lockres->l_blocking == LKM_EXMODE)
2626             && (lockres->l_ex_holders || lockres->l_ro_holders)) {
2627                 spin_unlock_irqrestore(&lockres->l_lock, flags);
2628                 *requeue = 1;
2629                 ret = 0;
2630                 goto leave;
2631         }
2632
2633         /* If it's a PR we're blocking, then only
2634          * requeue if we've got any EX holders */
2635         if (lockres->l_blocking == LKM_PRMODE &&
2636             lockres->l_ex_holders) {
2637                 spin_unlock_irqrestore(&lockres->l_lock, flags);
2638                 *requeue = 1;
2639                 ret = 0;
2640                 goto leave;
2641         }
2642
2643         /* If we get here, then we know that there are no more
2644          * incompatible holders (and anyone asking for an incompatible
2645          * lock is blocked). We can now downconvert the lock */
2646         if (!worker)
2647                 goto downconvert;
2648
2649         /* Some lockres types want to do a bit of work before
2650          * downconverting a lock. Allow that here. The worker function
2651          * may sleep, so we save off a copy of what we're blocking as
2652          * it may change while we're not holding the spin lock. */
2653         blocking = lockres->l_blocking;
2654         spin_unlock_irqrestore(&lockres->l_lock, flags);
2655
2656         worker(lockres, blocking);
2657
2658         spin_lock_irqsave(&lockres->l_lock, flags);
2659         if (blocking != lockres->l_blocking) {
2660                 /* If this changed underneath us, then we can't drop
2661                  * it just yet. */
2662                 goto recheck;
2663         }
2664
2665 downconvert:
2666         *requeue = 0;
2667         new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking);
2668
2669         ocfs2_prepare_downconvert(lockres, new_level);
2670         spin_unlock_irqrestore(&lockres->l_lock, flags);
2671         ret = ocfs2_downconvert_lock(osb, lockres, new_level, 0);
2672 leave:
2673         mlog_exit(ret);
2674         return ret;
2675 }
2676
2677 static void ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
2678                                       int blocking)
2679 {
2680         struct inode *inode;
2681         struct address_space *mapping;
2682
2683         mlog_entry_void();
2684
2685         inode = ocfs2_lock_res_inode(lockres);
2686         mapping = inode->i_mapping;
2687
2688         if (filemap_fdatawrite(mapping)) {
2689                 mlog(ML_ERROR, "Could not sync inode %llu for downconvert!",
2690                      (unsigned long long)OCFS2_I(inode)->ip_blkno);
2691         }
2692         sync_mapping_buffers(mapping);
2693         if (blocking == LKM_EXMODE) {
2694                 truncate_inode_pages(mapping, 0);
2695                 unmap_mapping_range(mapping, 0, 0, 0);
2696         } else {
2697                 /* We only need to wait on the I/O if we're not also
2698                  * truncating pages because truncate_inode_pages waits
2699                  * for us above. We don't truncate pages if we're
2700                  * blocking anything < EXMODE because we want to keep
2701                  * them around in that case. */
2702                 filemap_fdatawait(mapping);
2703         }
2704
2705         mlog_exit_void();
2706 }
2707
2708 int ocfs2_unblock_data(struct ocfs2_lock_res *lockres,
2709                        int *requeue)
2710 {
2711         int status;
2712         struct inode *inode;
2713         struct ocfs2_super *osb;
2714
2715         mlog_entry_void();
2716
2717         inode = ocfs2_lock_res_inode(lockres);
2718         osb = OCFS2_SB(inode->i_sb);
2719
2720         mlog(0, "unblock inode %llu\n",
2721              (unsigned long long)OCFS2_I(inode)->ip_blkno);
2722
2723         status = ocfs2_generic_unblock_lock(osb,
2724                                             lockres,
2725                                             requeue,
2726                                             ocfs2_data_convert_worker);
2727         if (status < 0)
2728                 mlog_errno(status);
2729
2730         mlog(0, "inode %llu, requeue = %d\n",
2731              (unsigned long long)OCFS2_I(inode)->ip_blkno, *requeue);
2732
2733         mlog_exit(status);
2734         return status;
2735 }
2736
2737 static int ocfs2_unblock_inode_lock(struct ocfs2_lock_res *lockres,
2738                                     int *requeue)
2739 {
2740         int status;
2741         struct inode *inode;
2742
2743         mlog_entry_void();
2744
2745         mlog(0, "Unblock lockres %s\n", lockres->l_name);
2746
2747         inode  = ocfs2_lock_res_inode(lockres);
2748
2749         status = ocfs2_generic_unblock_lock(OCFS2_SB(inode->i_sb),
2750                                             lockres,
2751                                             requeue,
2752                                             NULL);
2753         if (status < 0)
2754                 mlog_errno(status);
2755
2756         mlog_exit(status);
2757         return status;
2758 }
2759
2760
2761 int ocfs2_unblock_meta(struct ocfs2_lock_res *lockres,
2762                        int *requeue)
2763 {
2764         int status;
2765         struct inode *inode;
2766
2767         mlog_entry_void();
2768
2769         inode = ocfs2_lock_res_inode(lockres);
2770
2771         mlog(0, "unblock inode %llu\n",
2772              (unsigned long long)OCFS2_I(inode)->ip_blkno);
2773
2774         status = ocfs2_do_unblock_meta(inode, requeue);
2775         if (status < 0)
2776                 mlog_errno(status);
2777
2778         mlog(0, "inode %llu, requeue = %d\n",
2779              (unsigned long long)OCFS2_I(inode)->ip_blkno, *requeue);
2780
2781         mlog_exit(status);
2782         return status;
2783 }
2784
2785 /* Generic unblock function for any lockres whose private data is an
2786  * ocfs2_super pointer. */
2787 static int ocfs2_unblock_osb_lock(struct ocfs2_lock_res *lockres,
2788                                   int *requeue)
2789 {
2790         int status;
2791         struct ocfs2_super *osb;
2792
2793         mlog_entry_void();
2794
2795         mlog(0, "Unblock lockres %s\n", lockres->l_name);
2796
2797         osb = ocfs2_lock_res_super(lockres);
2798
2799         status = ocfs2_generic_unblock_lock(osb,
2800                                             lockres,
2801                                             requeue,
2802                                             NULL);
2803         if (status < 0)
2804                 mlog_errno(status);
2805
2806         mlog_exit(status);
2807         return status;
2808 }
2809
2810 void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
2811                                 struct ocfs2_lock_res *lockres)
2812 {
2813         int status;
2814         int requeue = 0;
2815         unsigned long flags;
2816
2817         /* Our reference to the lockres in this function can be
2818          * considered valid until we remove the OCFS2_LOCK_QUEUED
2819          * flag. */
2820
2821         mlog_entry_void();
2822
2823         BUG_ON(!lockres);
2824         BUG_ON(!lockres->l_ops);
2825         BUG_ON(!lockres->l_ops->unblock);
2826
2827         mlog(0, "lockres %s blocked.\n", lockres->l_name);
2828
2829         /* Detect whether a lock has been marked as going away while
2830          * the vote thread was processing other things. A lock can
2831          * still be marked with OCFS2_LOCK_FREEING after this check,
2832          * but short circuiting here will still save us some
2833          * performance. */
2834         spin_lock_irqsave(&lockres->l_lock, flags);
2835         if (lockres->l_flags & OCFS2_LOCK_FREEING)
2836                 goto unqueue;
2837         spin_unlock_irqrestore(&lockres->l_lock, flags);
2838
2839         status = lockres->l_ops->unblock(lockres, &requeue);
2840         if (status < 0)
2841                 mlog_errno(status);
2842
2843         spin_lock_irqsave(&lockres->l_lock, flags);
2844 unqueue:
2845         if (lockres->l_flags & OCFS2_LOCK_FREEING || !requeue) {
2846                 lockres_clear_flags(lockres, OCFS2_LOCK_QUEUED);
2847         } else
2848                 ocfs2_schedule_blocked_lock(osb, lockres);
2849
2850         mlog(0, "lockres %s, requeue = %s.\n", lockres->l_name,
2851              requeue ? "yes" : "no");
2852         spin_unlock_irqrestore(&lockres->l_lock, flags);
2853
2854         mlog_exit_void();
2855 }
2856
2857 static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
2858                                         struct ocfs2_lock_res *lockres)
2859 {
2860         mlog_entry_void();
2861
2862         assert_spin_locked(&lockres->l_lock);
2863
2864         if (lockres->l_flags & OCFS2_LOCK_FREEING) {
2865                 /* Do not schedule a lock for downconvert when it's on
2866                  * the way to destruction - any nodes wanting access
2867                  * to the resource will get it soon. */
2868                 mlog(0, "Lockres %s won't be scheduled: flags 0x%lx\n",
2869                      lockres->l_name, lockres->l_flags);
2870                 return;
2871         }
2872
2873         lockres_or_flags(lockres, OCFS2_LOCK_QUEUED);
2874
2875         spin_lock(&osb->vote_task_lock);
2876         if (list_empty(&lockres->l_blocked_list)) {
2877                 list_add_tail(&lockres->l_blocked_list,
2878                               &osb->blocked_lock_list);
2879                 osb->blocked_lock_count++;
2880         }
2881         spin_unlock(&osb->vote_task_lock);
2882
2883         mlog_exit_void();
2884 }
2885
2886 /* This aids in debugging situations where a bad LVB might be involved. */
2887 void ocfs2_dump_meta_lvb_info(u64 level,
2888                               const char *function,
2889                               unsigned int line,
2890                               struct ocfs2_lock_res *lockres)
2891 {
2892         struct ocfs2_meta_lvb *lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;
2893
2894         mlog(level, "LVB information for %s (called from %s:%u):\n",
2895              lockres->l_name, function, line);
2896         mlog(level, "version: %u, clusters: %u\n",
2897              be32_to_cpu(lvb->lvb_version), be32_to_cpu(lvb->lvb_iclusters));
2898         mlog(level, "size: %llu, uid %u, gid %u, mode 0x%x\n",
2899              (unsigned long long)be64_to_cpu(lvb->lvb_isize),
2900              be32_to_cpu(lvb->lvb_iuid), be32_to_cpu(lvb->lvb_igid),
2901              be16_to_cpu(lvb->lvb_imode));
2902         mlog(level, "nlink %u, atime_packed 0x%llx, ctime_packed 0x%llx, "
2903              "mtime_packed 0x%llx\n", be16_to_cpu(lvb->lvb_inlink),
2904              (long long)be64_to_cpu(lvb->lvb_iatime_packed),
2905              (long long)be64_to_cpu(lvb->lvb_ictime_packed),
2906              (long long)be64_to_cpu(lvb->lvb_imtime_packed));
2907 }