linux/fs/ocfs2/dlmglue.c
<<
>>
Prefs
   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/kthread.h>
  31#include <linux/pagemap.h>
  32#include <linux/debugfs.h>
  33#include <linux/seq_file.h>
  34#include <linux/time.h>
  35#include <linux/quotaops.h>
  36
  37#define MLOG_MASK_PREFIX ML_DLM_GLUE
  38#include <cluster/masklog.h>
  39
  40#include "ocfs2.h"
  41#include "ocfs2_lockingver.h"
  42
  43#include "alloc.h"
  44#include "dcache.h"
  45#include "dlmglue.h"
  46#include "extent_map.h"
  47#include "file.h"
  48#include "heartbeat.h"
  49#include "inode.h"
  50#include "journal.h"
  51#include "stackglue.h"
  52#include "slot_map.h"
  53#include "super.h"
  54#include "uptodate.h"
  55#include "quota.h"
  56#include "refcounttree.h"
  57
  58#include "buffer_head_io.h"
  59
  60struct ocfs2_mask_waiter {
  61        struct list_head        mw_item;
  62        int                     mw_status;
  63        struct completion       mw_complete;
  64        unsigned long           mw_mask;
  65        unsigned long           mw_goal;
  66#ifdef CONFIG_OCFS2_FS_STATS
  67        ktime_t                 mw_lock_start;
  68#endif
  69};
  70
  71static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres);
  72static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres);
  73static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres);
  74static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres);
  75
  76/*
  77 * Return value from ->downconvert_worker functions.
  78 *
  79 * These control the precise actions of ocfs2_unblock_lock()
  80 * and ocfs2_process_blocked_lock()
  81 *
  82 */
  83enum ocfs2_unblock_action {
  84        UNBLOCK_CONTINUE        = 0, /* Continue downconvert */
  85        UNBLOCK_CONTINUE_POST   = 1, /* Continue downconvert, fire
  86                                      * ->post_unlock callback */
  87        UNBLOCK_STOP_POST       = 2, /* Do not downconvert, fire
  88                                      * ->post_unlock() callback. */
  89};
  90
  91struct ocfs2_unblock_ctl {
  92        int requeue;
  93        enum ocfs2_unblock_action unblock_action;
  94};
  95
  96/* Lockdep class keys */
  97struct lock_class_key lockdep_keys[OCFS2_NUM_LOCK_TYPES];
  98
  99static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
 100                                        int new_level);
 101static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres);
 102
 103static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
 104                                     int blocking);
 105
 106static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
 107                                       int blocking);
 108
 109static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
 110                                     struct ocfs2_lock_res *lockres);
 111
 112static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres);
 113
 114static int ocfs2_check_refcount_downconvert(struct ocfs2_lock_res *lockres,
 115                                            int new_level);
 116static int ocfs2_refcount_convert_worker(struct ocfs2_lock_res *lockres,
 117                                         int blocking);
 118
 119#define mlog_meta_lvb(__level, __lockres) ocfs2_dump_meta_lvb_info(__level, __PRETTY_FUNCTION__, __LINE__, __lockres)
 120
 121/* This aids in debugging situations where a bad LVB might be involved. */
 122static void ocfs2_dump_meta_lvb_info(u64 level,
 123                                     const char *function,
 124                                     unsigned int line,
 125                                     struct ocfs2_lock_res *lockres)
 126{
 127        struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
 128
 129        mlog(level, "LVB information for %s (called from %s:%u):\n",
 130             lockres->l_name, function, line);
 131        mlog(level, "version: %u, clusters: %u, generation: 0x%x\n",
 132             lvb->lvb_version, be32_to_cpu(lvb->lvb_iclusters),
 133             be32_to_cpu(lvb->lvb_igeneration));
 134        mlog(level, "size: %llu, uid %u, gid %u, mode 0x%x\n",
 135             (unsigned long long)be64_to_cpu(lvb->lvb_isize),
 136             be32_to_cpu(lvb->lvb_iuid), be32_to_cpu(lvb->lvb_igid),
 137             be16_to_cpu(lvb->lvb_imode));
 138        mlog(level, "nlink %u, atime_packed 0x%llx, ctime_packed 0x%llx, "
 139             "mtime_packed 0x%llx iattr 0x%x\n", be16_to_cpu(lvb->lvb_inlink),
 140             (long long)be64_to_cpu(lvb->lvb_iatime_packed),
 141             (long long)be64_to_cpu(lvb->lvb_ictime_packed),
 142             (long long)be64_to_cpu(lvb->lvb_imtime_packed),
 143             be32_to_cpu(lvb->lvb_iattr));
 144}
 145
 146
 147/*
 148 * OCFS2 Lock Resource Operations
 149 *
 150 * These fine tune the behavior of the generic dlmglue locking infrastructure.
 151 *
 152 * The most basic of lock types can point ->l_priv to their respective
 153 * struct ocfs2_super and allow the default actions to manage things.
 154 *
 155 * Right now, each lock type also needs to implement an init function,
 156 * and trivial lock/unlock wrappers. ocfs2_simple_drop_lockres()
 157 * should be called when the lock is no longer needed (i.e., object
 158 * destruction time).
 159 */
 160struct ocfs2_lock_res_ops {
 161        /*
 162         * Translate an ocfs2_lock_res * into an ocfs2_super *. Define
 163         * this callback if ->l_priv is not an ocfs2_super pointer
 164         */
 165        struct ocfs2_super * (*get_osb)(struct ocfs2_lock_res *);
 166
 167        /*
 168         * Optionally called in the downconvert thread after a
 169         * successful downconvert. The lockres will not be referenced
 170         * after this callback is called, so it is safe to free
 171         * memory, etc.
 172         *
 173         * The exact semantics of when this is called are controlled
 174         * by ->downconvert_worker()
 175         */
 176        void (*post_unlock)(struct ocfs2_super *, struct ocfs2_lock_res *);
 177
 178        /*
 179         * Allow a lock type to add checks to determine whether it is
 180         * safe to downconvert a lock. Return 0 to re-queue the
 181         * downconvert at a later time, nonzero to continue.
 182         *
 183         * For most locks, the default checks that there are no
 184         * incompatible holders are sufficient.
 185         *
 186         * Called with the lockres spinlock held.
 187         */
 188        int (*check_downconvert)(struct ocfs2_lock_res *, int);
 189
 190        /*
 191         * Allows a lock type to populate the lock value block. This
 192         * is called on downconvert, and when we drop a lock.
 193         *
 194         * Locks that want to use this should set LOCK_TYPE_USES_LVB
 195         * in the flags field.
 196         *
 197         * Called with the lockres spinlock held.
 198         */
 199        void (*set_lvb)(struct ocfs2_lock_res *);
 200
 201        /*
 202         * Called from the downconvert thread when it is determined
 203         * that a lock will be downconverted. This is called without
 204         * any locks held so the function can do work that might
 205         * schedule (syncing out data, etc).
 206         *
 207         * This should return any one of the ocfs2_unblock_action
 208         * values, depending on what it wants the thread to do.
 209         */
 210        int (*downconvert_worker)(struct ocfs2_lock_res *, int);
 211
 212        /*
 213         * LOCK_TYPE_* flags which describe the specific requirements
 214         * of a lock type. Descriptions of each individual flag follow.
 215         */
 216        int flags;
 217};
 218
 219/*
 220 * Some locks want to "refresh" potentially stale data when a
 221 * meaningful (PRMODE or EXMODE) lock level is first obtained. If this
 222 * flag is set, the OCFS2_LOCK_NEEDS_REFRESH flag will be set on the
 223 * individual lockres l_flags member from the ast function. It is
 224 * expected that the locking wrapper will clear the
 225 * OCFS2_LOCK_NEEDS_REFRESH flag when done.
 226 */
 227#define LOCK_TYPE_REQUIRES_REFRESH 0x1
 228
 229/*
 230 * Indicate that a lock type makes use of the lock value block. The
 231 * ->set_lvb lock type callback must be defined.
 232 */
 233#define LOCK_TYPE_USES_LVB              0x2
 234
 235static struct ocfs2_lock_res_ops ocfs2_inode_rw_lops = {
 236        .get_osb        = ocfs2_get_inode_osb,
 237        .flags          = 0,
 238};
 239
 240static struct ocfs2_lock_res_ops ocfs2_inode_inode_lops = {
 241        .get_osb        = ocfs2_get_inode_osb,
 242        .check_downconvert = ocfs2_check_meta_downconvert,
 243        .set_lvb        = ocfs2_set_meta_lvb,
 244        .downconvert_worker = ocfs2_data_convert_worker,
 245        .flags          = LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
 246};
 247
 248static struct ocfs2_lock_res_ops ocfs2_super_lops = {
 249        .flags          = LOCK_TYPE_REQUIRES_REFRESH,
 250};
 251
 252static struct ocfs2_lock_res_ops ocfs2_rename_lops = {
 253        .flags          = 0,
 254};
 255
 256static struct ocfs2_lock_res_ops ocfs2_nfs_sync_lops = {
 257        .flags          = 0,
 258};
 259
 260static struct ocfs2_lock_res_ops ocfs2_orphan_scan_lops = {
 261        .flags          = LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
 262};
 263
 264static struct ocfs2_lock_res_ops ocfs2_dentry_lops = {
 265        .get_osb        = ocfs2_get_dentry_osb,
 266        .post_unlock    = ocfs2_dentry_post_unlock,
 267        .downconvert_worker = ocfs2_dentry_convert_worker,
 268        .flags          = 0,
 269};
 270
 271static struct ocfs2_lock_res_ops ocfs2_inode_open_lops = {
 272        .get_osb        = ocfs2_get_inode_osb,
 273        .flags          = 0,
 274};
 275
 276static struct ocfs2_lock_res_ops ocfs2_flock_lops = {
 277        .get_osb        = ocfs2_get_file_osb,
 278        .flags          = 0,
 279};
 280
 281static struct ocfs2_lock_res_ops ocfs2_qinfo_lops = {
 282        .set_lvb        = ocfs2_set_qinfo_lvb,
 283        .get_osb        = ocfs2_get_qinfo_osb,
 284        .flags          = LOCK_TYPE_REQUIRES_REFRESH | LOCK_TYPE_USES_LVB,
 285};
 286
 287static struct ocfs2_lock_res_ops ocfs2_refcount_block_lops = {
 288        .check_downconvert = ocfs2_check_refcount_downconvert,
 289        .downconvert_worker = ocfs2_refcount_convert_worker,
 290        .flags          = 0,
 291};
 292
 293static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres)
 294{
 295        return lockres->l_type == OCFS2_LOCK_TYPE_META ||
 296                lockres->l_type == OCFS2_LOCK_TYPE_RW ||
 297                lockres->l_type == OCFS2_LOCK_TYPE_OPEN;
 298}
 299
 300static inline struct ocfs2_lock_res *ocfs2_lksb_to_lock_res(struct ocfs2_dlm_lksb *lksb)
 301{
 302        return container_of(lksb, struct ocfs2_lock_res, l_lksb);
 303}
 304
 305static inline struct inode *ocfs2_lock_res_inode(struct ocfs2_lock_res *lockres)
 306{
 307        BUG_ON(!ocfs2_is_inode_lock(lockres));
 308
 309        return (struct inode *) lockres->l_priv;
 310}
 311
 312static inline struct ocfs2_dentry_lock *ocfs2_lock_res_dl(struct ocfs2_lock_res *lockres)
 313{
 314        BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_DENTRY);
 315
 316        return (struct ocfs2_dentry_lock *)lockres->l_priv;
 317}
 318
 319static inline struct ocfs2_mem_dqinfo *ocfs2_lock_res_qinfo(struct ocfs2_lock_res *lockres)
 320{
 321        BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_QINFO);
 322
 323        return (struct ocfs2_mem_dqinfo *)lockres->l_priv;
 324}
 325
 326static inline struct ocfs2_refcount_tree *
 327ocfs2_lock_res_refcount_tree(struct ocfs2_lock_res *res)
 328{
 329        return container_of(res, struct ocfs2_refcount_tree, rf_lockres);
 330}
 331
 332static inline struct ocfs2_super *ocfs2_get_lockres_osb(struct ocfs2_lock_res *lockres)
 333{
 334        if (lockres->l_ops->get_osb)
 335                return lockres->l_ops->get_osb(lockres);
 336
 337        return (struct ocfs2_super *)lockres->l_priv;
 338}
 339
 340static int ocfs2_lock_create(struct ocfs2_super *osb,
 341                             struct ocfs2_lock_res *lockres,
 342                             int level,
 343                             u32 dlm_flags);
 344static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
 345                                                     int wanted);
 346static void __ocfs2_cluster_unlock(struct ocfs2_super *osb,
 347                                   struct ocfs2_lock_res *lockres,
 348                                   int level, unsigned long caller_ip);
 349static inline void ocfs2_cluster_unlock(struct ocfs2_super *osb,
 350                                        struct ocfs2_lock_res *lockres,
 351                                        int level)
 352{
 353        __ocfs2_cluster_unlock(osb, lockres, level, _RET_IP_);
 354}
 355
 356static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres);
 357static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres);
 358static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres);
 359static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, int level);
 360static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
 361                                        struct ocfs2_lock_res *lockres);
 362static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
 363                                                int convert);
 364#define ocfs2_log_dlm_error(_func, _err, _lockres) do {                                 \
 365        if ((_lockres)->l_type != OCFS2_LOCK_TYPE_DENTRY)                               \
 366                mlog(ML_ERROR, "DLM error %d while calling %s on resource %s\n",        \
 367                     _err, _func, _lockres->l_name);                                    \
 368        else                                                                            \
 369                mlog(ML_ERROR, "DLM error %d while calling %s on resource %.*s%08x\n",  \
 370                     _err, _func, OCFS2_DENTRY_LOCK_INO_START - 1, (_lockres)->l_name,  \
 371                     (unsigned int)ocfs2_get_dentry_lock_ino(_lockres));                \
 372} while (0)
 373static int ocfs2_downconvert_thread(void *arg);
 374static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
 375                                        struct ocfs2_lock_res *lockres);
 376static int ocfs2_inode_lock_update(struct inode *inode,
 377                                  struct buffer_head **bh);
 378static void ocfs2_drop_osb_locks(struct ocfs2_super *osb);
 379static inline int ocfs2_highest_compat_lock_level(int level);
 380static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
 381                                              int new_level);
 382static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
 383                                  struct ocfs2_lock_res *lockres,
 384                                  int new_level,
 385                                  int lvb,
 386                                  unsigned int generation);
 387static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
 388                                        struct ocfs2_lock_res *lockres);
 389static int ocfs2_cancel_convert(struct ocfs2_super *osb,
 390                                struct ocfs2_lock_res *lockres);
 391
 392
 393static void ocfs2_build_lock_name(enum ocfs2_lock_type type,
 394                                  u64 blkno,
 395                                  u32 generation,
 396                                  char *name)
 397{
 398        int len;
 399
 400        BUG_ON(type >= OCFS2_NUM_LOCK_TYPES);
 401
 402        len = snprintf(name, OCFS2_LOCK_ID_MAX_LEN, "%c%s%016llx%08x",
 403                       ocfs2_lock_type_char(type), OCFS2_LOCK_ID_PAD,
 404                       (long long)blkno, generation);
 405
 406        BUG_ON(len != (OCFS2_LOCK_ID_MAX_LEN - 1));
 407
 408        mlog(0, "built lock resource with name: %s\n", name);
 409}
 410
 411static DEFINE_SPINLOCK(ocfs2_dlm_tracking_lock);
 412
 413static void ocfs2_add_lockres_tracking(struct ocfs2_lock_res *res,
 414                                       struct ocfs2_dlm_debug *dlm_debug)
 415{
 416        mlog(0, "Add tracking for lockres %s\n", res->l_name);
 417
 418        spin_lock(&ocfs2_dlm_tracking_lock);
 419        list_add(&res->l_debug_list, &dlm_debug->d_lockres_tracking);
 420        spin_unlock(&ocfs2_dlm_tracking_lock);
 421}
 422
 423static void ocfs2_remove_lockres_tracking(struct ocfs2_lock_res *res)
 424{
 425        spin_lock(&ocfs2_dlm_tracking_lock);
 426        if (!list_empty(&res->l_debug_list))
 427                list_del_init(&res->l_debug_list);
 428        spin_unlock(&ocfs2_dlm_tracking_lock);
 429}
 430
 431#ifdef CONFIG_OCFS2_FS_STATS
 432static void ocfs2_init_lock_stats(struct ocfs2_lock_res *res)
 433{
 434        res->l_lock_refresh = 0;
 435        memset(&res->l_lock_prmode, 0, sizeof(struct ocfs2_lock_stats));
 436        memset(&res->l_lock_exmode, 0, sizeof(struct ocfs2_lock_stats));
 437}
 438
 439static void ocfs2_update_lock_stats(struct ocfs2_lock_res *res, int level,
 440                                    struct ocfs2_mask_waiter *mw, int ret)
 441{
 442        u32 usec;
 443        ktime_t kt;
 444        struct ocfs2_lock_stats *stats;
 445
 446        if (level == LKM_PRMODE)
 447                stats = &res->l_lock_prmode;
 448        else if (level == LKM_EXMODE)
 449                stats = &res->l_lock_exmode;
 450        else
 451                return;
 452
 453        kt = ktime_sub(ktime_get(), mw->mw_lock_start);
 454        usec = ktime_to_us(kt);
 455
 456        stats->ls_gets++;
 457        stats->ls_total += ktime_to_ns(kt);
 458        /* overflow */
 459        if (unlikely(stats->ls_gets == 0)) {
 460                stats->ls_gets++;
 461                stats->ls_total = ktime_to_ns(kt);
 462        }
 463
 464        if (stats->ls_max < usec)
 465                stats->ls_max = usec;
 466
 467        if (ret)
 468                stats->ls_fail++;
 469}
 470
 471static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres)
 472{
 473        lockres->l_lock_refresh++;
 474}
 475
 476static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw)
 477{
 478        mw->mw_lock_start = ktime_get();
 479}
 480#else
 481static inline void ocfs2_init_lock_stats(struct ocfs2_lock_res *res)
 482{
 483}
 484static inline void ocfs2_update_lock_stats(struct ocfs2_lock_res *res,
 485                           int level, struct ocfs2_mask_waiter *mw, int ret)
 486{
 487}
 488static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres)
 489{
 490}
 491static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw)
 492{
 493}
 494#endif
 495
 496static void ocfs2_lock_res_init_common(struct ocfs2_super *osb,
 497                                       struct ocfs2_lock_res *res,
 498                                       enum ocfs2_lock_type type,
 499                                       struct ocfs2_lock_res_ops *ops,
 500                                       void *priv)
 501{
 502        res->l_type          = type;
 503        res->l_ops           = ops;
 504        res->l_priv          = priv;
 505
 506        res->l_level         = DLM_LOCK_IV;
 507        res->l_requested     = DLM_LOCK_IV;
 508        res->l_blocking      = DLM_LOCK_IV;
 509        res->l_action        = OCFS2_AST_INVALID;
 510        res->l_unlock_action = OCFS2_UNLOCK_INVALID;
 511
 512        res->l_flags         = OCFS2_LOCK_INITIALIZED;
 513
 514        ocfs2_add_lockres_tracking(res, osb->osb_dlm_debug);
 515
 516        ocfs2_init_lock_stats(res);
 517#ifdef CONFIG_DEBUG_LOCK_ALLOC
 518        if (type != OCFS2_LOCK_TYPE_OPEN)
 519                lockdep_init_map(&res->l_lockdep_map, ocfs2_lock_type_strings[type],
 520                                 &lockdep_keys[type], 0);
 521        else
 522                res->l_lockdep_map.key = NULL;
 523#endif
 524}
 525
 526void ocfs2_lock_res_init_once(struct ocfs2_lock_res *res)
 527{
 528        /* This also clears out the lock status block */
 529        memset(res, 0, sizeof(struct ocfs2_lock_res));
 530        spin_lock_init(&res->l_lock);
 531        init_waitqueue_head(&res->l_event);
 532        INIT_LIST_HEAD(&res->l_blocked_list);
 533        INIT_LIST_HEAD(&res->l_mask_waiters);
 534}
 535
 536void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res,
 537                               enum ocfs2_lock_type type,
 538                               unsigned int generation,
 539                               struct inode *inode)
 540{
 541        struct ocfs2_lock_res_ops *ops;
 542
 543        switch(type) {
 544                case OCFS2_LOCK_TYPE_RW:
 545                        ops = &ocfs2_inode_rw_lops;
 546                        break;
 547                case OCFS2_LOCK_TYPE_META:
 548                        ops = &ocfs2_inode_inode_lops;
 549                        break;
 550                case OCFS2_LOCK_TYPE_OPEN:
 551                        ops = &ocfs2_inode_open_lops;
 552                        break;
 553                default:
 554                        mlog_bug_on_msg(1, "type: %d\n", type);
 555                        ops = NULL; /* thanks, gcc */
 556                        break;
 557        };
 558
 559        ocfs2_build_lock_name(type, OCFS2_I(inode)->ip_blkno,
 560                              generation, res->l_name);
 561        ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), res, type, ops, inode);
 562}
 563
 564static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres)
 565{
 566        struct inode *inode = ocfs2_lock_res_inode(lockres);
 567
 568        return OCFS2_SB(inode->i_sb);
 569}
 570
 571static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres)
 572{
 573        struct ocfs2_mem_dqinfo *info = lockres->l_priv;
 574
 575        return OCFS2_SB(info->dqi_gi.dqi_sb);
 576}
 577
 578static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres)
 579{
 580        struct ocfs2_file_private *fp = lockres->l_priv;
 581
 582        return OCFS2_SB(fp->fp_file->f_mapping->host->i_sb);
 583}
 584
 585static __u64 ocfs2_get_dentry_lock_ino(struct ocfs2_lock_res *lockres)
 586{
 587        __be64 inode_blkno_be;
 588
 589        memcpy(&inode_blkno_be, &lockres->l_name[OCFS2_DENTRY_LOCK_INO_START],
 590               sizeof(__be64));
 591
 592        return be64_to_cpu(inode_blkno_be);
 593}
 594
 595static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres)
 596{
 597        struct ocfs2_dentry_lock *dl = lockres->l_priv;
 598
 599        return OCFS2_SB(dl->dl_inode->i_sb);
 600}
 601
 602void ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock *dl,
 603                                u64 parent, struct inode *inode)
 604{
 605        int len;
 606        u64 inode_blkno = OCFS2_I(inode)->ip_blkno;
 607        __be64 inode_blkno_be = cpu_to_be64(inode_blkno);
 608        struct ocfs2_lock_res *lockres = &dl->dl_lockres;
 609
 610        ocfs2_lock_res_init_once(lockres);
 611
 612        /*
 613         * Unfortunately, the standard lock naming scheme won't work
 614         * here because we have two 16 byte values to use. Instead,
 615         * we'll stuff the inode number as a binary value. We still
 616         * want error prints to show something without garbling the
 617         * display, so drop a null byte in there before the inode
 618         * number. A future version of OCFS2 will likely use all
 619         * binary lock names. The stringified names have been a
 620         * tremendous aid in debugging, but now that the debugfs
 621         * interface exists, we can mangle things there if need be.
 622         *
 623         * NOTE: We also drop the standard "pad" value (the total lock
 624         * name size stays the same though - the last part is all
 625         * zeros due to the memset in ocfs2_lock_res_init_once()
 626         */
 627        len = snprintf(lockres->l_name, OCFS2_DENTRY_LOCK_INO_START,
 628                       "%c%016llx",
 629                       ocfs2_lock_type_char(OCFS2_LOCK_TYPE_DENTRY),
 630                       (long long)parent);
 631
 632        BUG_ON(len != (OCFS2_DENTRY_LOCK_INO_START - 1));
 633
 634        memcpy(&lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], &inode_blkno_be,
 635               sizeof(__be64));
 636
 637        ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
 638                                   OCFS2_LOCK_TYPE_DENTRY, &ocfs2_dentry_lops,
 639                                   dl);
 640}
 641
 642static void ocfs2_super_lock_res_init(struct ocfs2_lock_res *res,
 643                                      struct ocfs2_super *osb)
 644{
 645        /* Superblock lockres doesn't come from a slab so we call init
 646         * once on it manually.  */
 647        ocfs2_lock_res_init_once(res);
 648        ocfs2_build_lock_name(OCFS2_LOCK_TYPE_SUPER, OCFS2_SUPER_BLOCK_BLKNO,
 649                              0, res->l_name);
 650        ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_SUPER,
 651                                   &ocfs2_super_lops, osb);
 652}
 653
 654static void ocfs2_rename_lock_res_init(struct ocfs2_lock_res *res,
 655                                       struct ocfs2_super *osb)
 656{
 657        /* Rename lockres doesn't come from a slab so we call init
 658         * once on it manually.  */
 659        ocfs2_lock_res_init_once(res);
 660        ocfs2_build_lock_name(OCFS2_LOCK_TYPE_RENAME, 0, 0, res->l_name);
 661        ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_RENAME,
 662                                   &ocfs2_rename_lops, osb);
 663}
 664
 665static void ocfs2_nfs_sync_lock_res_init(struct ocfs2_lock_res *res,
 666                                         struct ocfs2_super *osb)
 667{
 668        /* nfs_sync lockres doesn't come from a slab so we call init
 669         * once on it manually.  */
 670        ocfs2_lock_res_init_once(res);
 671        ocfs2_build_lock_name(OCFS2_LOCK_TYPE_NFS_SYNC, 0, 0, res->l_name);
 672        ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_NFS_SYNC,
 673                                   &ocfs2_nfs_sync_lops, osb);
 674}
 675
 676static void ocfs2_orphan_scan_lock_res_init(struct ocfs2_lock_res *res,
 677                                            struct ocfs2_super *osb)
 678{
 679        ocfs2_lock_res_init_once(res);
 680        ocfs2_build_lock_name(OCFS2_LOCK_TYPE_ORPHAN_SCAN, 0, 0, res->l_name);
 681        ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_ORPHAN_SCAN,
 682                                   &ocfs2_orphan_scan_lops, osb);
 683}
 684
 685void ocfs2_file_lock_res_init(struct ocfs2_lock_res *lockres,
 686                              struct ocfs2_file_private *fp)
 687{
 688        struct inode *inode = fp->fp_file->f_mapping->host;
 689        struct ocfs2_inode_info *oi = OCFS2_I(inode);
 690
 691        ocfs2_lock_res_init_once(lockres);
 692        ocfs2_build_lock_name(OCFS2_LOCK_TYPE_FLOCK, oi->ip_blkno,
 693                              inode->i_generation, lockres->l_name);
 694        ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
 695                                   OCFS2_LOCK_TYPE_FLOCK, &ocfs2_flock_lops,
 696                                   fp);
 697        lockres->l_flags |= OCFS2_LOCK_NOCACHE;
 698}
 699
 700void ocfs2_qinfo_lock_res_init(struct ocfs2_lock_res *lockres,
 701                               struct ocfs2_mem_dqinfo *info)
 702{
 703        ocfs2_lock_res_init_once(lockres);
 704        ocfs2_build_lock_name(OCFS2_LOCK_TYPE_QINFO, info->dqi_gi.dqi_type,
 705                              0, lockres->l_name);
 706        ocfs2_lock_res_init_common(OCFS2_SB(info->dqi_gi.dqi_sb), lockres,
 707                                   OCFS2_LOCK_TYPE_QINFO, &ocfs2_qinfo_lops,
 708                                   info);
 709}
 710
 711void ocfs2_refcount_lock_res_init(struct ocfs2_lock_res *lockres,
 712                                  struct ocfs2_super *osb, u64 ref_blkno,
 713                                  unsigned int generation)
 714{
 715        ocfs2_lock_res_init_once(lockres);
 716        ocfs2_build_lock_name(OCFS2_LOCK_TYPE_REFCOUNT, ref_blkno,
 717                              generation, lockres->l_name);
 718        ocfs2_lock_res_init_common(osb, lockres, OCFS2_LOCK_TYPE_REFCOUNT,
 719                                   &ocfs2_refcount_block_lops, osb);
 720}
 721
 722void ocfs2_lock_res_free(struct ocfs2_lock_res *res)
 723{
 724        if (!(res->l_flags & OCFS2_LOCK_INITIALIZED))
 725                return;
 726
 727        ocfs2_remove_lockres_tracking(res);
 728
 729        mlog_bug_on_msg(!list_empty(&res->l_blocked_list),
 730                        "Lockres %s is on the blocked list\n",
 731                        res->l_name);
 732        mlog_bug_on_msg(!list_empty(&res->l_mask_waiters),
 733                        "Lockres %s has mask waiters pending\n",
 734                        res->l_name);
 735        mlog_bug_on_msg(spin_is_locked(&res->l_lock),
 736                        "Lockres %s is locked\n",
 737                        res->l_name);
 738        mlog_bug_on_msg(res->l_ro_holders,
 739                        "Lockres %s has %u ro holders\n",
 740                        res->l_name, res->l_ro_holders);
 741        mlog_bug_on_msg(res->l_ex_holders,
 742                        "Lockres %s has %u ex holders\n",
 743                        res->l_name, res->l_ex_holders);
 744
 745        /* Need to clear out the lock status block for the dlm */
 746        memset(&res->l_lksb, 0, sizeof(res->l_lksb));
 747
 748        res->l_flags = 0UL;
 749}
 750
 751static inline void ocfs2_inc_holders(struct ocfs2_lock_res *lockres,
 752                                     int level)
 753{
 754        BUG_ON(!lockres);
 755
 756        switch(level) {
 757        case DLM_LOCK_EX:
 758                lockres->l_ex_holders++;
 759                break;
 760        case DLM_LOCK_PR:
 761                lockres->l_ro_holders++;
 762                break;
 763        default:
 764                BUG();
 765        }
 766}
 767
 768static inline void ocfs2_dec_holders(struct ocfs2_lock_res *lockres,
 769                                     int level)
 770{
 771        BUG_ON(!lockres);
 772
 773        switch(level) {
 774        case DLM_LOCK_EX:
 775                BUG_ON(!lockres->l_ex_holders);
 776                lockres->l_ex_holders--;
 777                break;
 778        case DLM_LOCK_PR:
 779                BUG_ON(!lockres->l_ro_holders);
 780                lockres->l_ro_holders--;
 781                break;
 782        default:
 783                BUG();
 784        }
 785}
 786
 787/* WARNING: This function lives in a world where the only three lock
 788 * levels are EX, PR, and NL. It *will* have to be adjusted when more
 789 * lock types are added. */
 790static inline int ocfs2_highest_compat_lock_level(int level)
 791{
 792        int new_level = DLM_LOCK_EX;
 793
 794        if (level == DLM_LOCK_EX)
 795                new_level = DLM_LOCK_NL;
 796        else if (level == DLM_LOCK_PR)
 797                new_level = DLM_LOCK_PR;
 798        return new_level;
 799}
 800
 801static void lockres_set_flags(struct ocfs2_lock_res *lockres,
 802                              unsigned long newflags)
 803{
 804        struct ocfs2_mask_waiter *mw, *tmp;
 805
 806        assert_spin_locked(&lockres->l_lock);
 807
 808        lockres->l_flags = newflags;
 809
 810        list_for_each_entry_safe(mw, tmp, &lockres->l_mask_waiters, mw_item) {
 811                if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
 812                        continue;
 813
 814                list_del_init(&mw->mw_item);
 815                mw->mw_status = 0;
 816                complete(&mw->mw_complete);
 817        }
 818}
 819static void lockres_or_flags(struct ocfs2_lock_res *lockres, unsigned long or)
 820{
 821        lockres_set_flags(lockres, lockres->l_flags | or);
 822}
 823static void lockres_clear_flags(struct ocfs2_lock_res *lockres,
 824                                unsigned long clear)
 825{
 826        lockres_set_flags(lockres, lockres->l_flags & ~clear);
 827}
 828
 829static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres)
 830{
 831        BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
 832        BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
 833        BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
 834        BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
 835
 836        lockres->l_level = lockres->l_requested;
 837        if (lockres->l_level <=
 838            ocfs2_highest_compat_lock_level(lockres->l_blocking)) {
 839                lockres->l_blocking = DLM_LOCK_NL;
 840                lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
 841        }
 842        lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
 843}
 844
 845static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres)
 846{
 847        BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
 848        BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
 849
 850        /* Convert from RO to EX doesn't really need anything as our
 851         * information is already up to data. Convert from NL to
 852         * *anything* however should mark ourselves as needing an
 853         * update */
 854        if (lockres->l_level == DLM_LOCK_NL &&
 855            lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
 856                lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
 857
 858        lockres->l_level = lockres->l_requested;
 859
 860        /*
 861         * We set the OCFS2_LOCK_UPCONVERT_FINISHING flag before clearing
 862         * the OCFS2_LOCK_BUSY flag to prevent the dc thread from
 863         * downconverting the lock before the upconvert has fully completed.
 864         */
 865        lockres_or_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
 866
 867        lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
 868}
 869
 870static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres)
 871{
 872        BUG_ON((!(lockres->l_flags & OCFS2_LOCK_BUSY)));
 873        BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
 874
 875        if (lockres->l_requested > DLM_LOCK_NL &&
 876            !(lockres->l_flags & OCFS2_LOCK_LOCAL) &&
 877            lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
 878                lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
 879
 880        lockres->l_level = lockres->l_requested;
 881        lockres_or_flags(lockres, OCFS2_LOCK_ATTACHED);
 882        lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
 883}
 884
 885static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres,
 886                                     int level)
 887{
 888        int needs_downconvert = 0;
 889
 890        assert_spin_locked(&lockres->l_lock);
 891
 892        if (level > lockres->l_blocking) {
 893                /* only schedule a downconvert if we haven't already scheduled
 894                 * one that goes low enough to satisfy the level we're
 895                 * blocking.  this also catches the case where we get
 896                 * duplicate BASTs */
 897                if (ocfs2_highest_compat_lock_level(level) <
 898                    ocfs2_highest_compat_lock_level(lockres->l_blocking))
 899                        needs_downconvert = 1;
 900
 901                lockres->l_blocking = level;
 902        }
 903
 904        mlog(ML_BASTS, "lockres %s, block %d, level %d, l_block %d, dwn %d\n",
 905             lockres->l_name, level, lockres->l_level, lockres->l_blocking,
 906             needs_downconvert);
 907
 908        if (needs_downconvert)
 909                lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
 910        mlog(0, "needs_downconvert = %d\n", needs_downconvert);
 911        return needs_downconvert;
 912}
 913
 914/*
 915 * OCFS2_LOCK_PENDING and l_pending_gen.
 916 *
 917 * Why does OCFS2_LOCK_PENDING exist?  To close a race between setting
 918 * OCFS2_LOCK_BUSY and calling ocfs2_dlm_lock().  See ocfs2_unblock_lock()
 919 * for more details on the race.
 920 *
 921 * OCFS2_LOCK_PENDING closes the race quite nicely.  However, it introduces
 922 * a race on itself.  In o2dlm, we can get the ast before ocfs2_dlm_lock()
 923 * returns.  The ast clears OCFS2_LOCK_BUSY, and must therefore clear
 924 * OCFS2_LOCK_PENDING at the same time.  When ocfs2_dlm_lock() returns,
 925 * the caller is going to try to clear PENDING again.  If nothing else is
 926 * happening, __lockres_clear_pending() sees PENDING is unset and does
 927 * nothing.
 928 *
 929 * But what if another path (eg downconvert thread) has just started a
 930 * new locking action?  The other path has re-set PENDING.  Our path
 931 * cannot clear PENDING, because that will re-open the original race
 932 * window.
 933 *
 934 * [Example]
 935 *
 936 * ocfs2_meta_lock()
 937 *  ocfs2_cluster_lock()
 938 *   set BUSY
 939 *   set PENDING
 940 *   drop l_lock
 941 *   ocfs2_dlm_lock()
 942 *    ocfs2_locking_ast()               ocfs2_downconvert_thread()
 943 *     clear PENDING                     ocfs2_unblock_lock()
 944 *                                        take_l_lock
 945 *                                        !BUSY
 946 *                                        ocfs2_prepare_downconvert()
 947 *                                         set BUSY
 948 *                                         set PENDING
 949 *                                        drop l_lock
 950 *   take l_lock
 951 *   clear PENDING
 952 *   drop l_lock
 953 *                      <window>
 954 *                                        ocfs2_dlm_lock()
 955 *
 956 * So as you can see, we now have a window where l_lock is not held,
 957 * PENDING is not set, and ocfs2_dlm_lock() has not been called.
 958 *
 959 * The core problem is that ocfs2_cluster_lock() has cleared the PENDING
 960 * set by ocfs2_prepare_downconvert().  That wasn't nice.
 961 *
 962 * To solve this we introduce l_pending_gen.  A call to
 963 * lockres_clear_pending() will only do so when it is passed a generation
 964 * number that matches the lockres.  lockres_set_pending() will return the
 965 * current generation number.  When ocfs2_cluster_lock() goes to clear
 966 * PENDING, it passes the generation it got from set_pending().  In our
 967 * example above, the generation numbers will *not* match.  Thus,
 968 * ocfs2_cluster_lock() will not clear the PENDING set by
 969 * ocfs2_prepare_downconvert().
 970 */
 971
 972/* Unlocked version for ocfs2_locking_ast() */
 973static void __lockres_clear_pending(struct ocfs2_lock_res *lockres,
 974                                    unsigned int generation,
 975                                    struct ocfs2_super *osb)
 976{
 977        assert_spin_locked(&lockres->l_lock);
 978
 979        /*
 980         * The ast and locking functions can race us here.  The winner
 981         * will clear pending, the loser will not.
 982         */
 983        if (!(lockres->l_flags & OCFS2_LOCK_PENDING) ||
 984            (lockres->l_pending_gen != generation))
 985                return;
 986
 987        lockres_clear_flags(lockres, OCFS2_LOCK_PENDING);
 988        lockres->l_pending_gen++;
 989
 990        /*
 991         * The downconvert thread may have skipped us because we
 992         * were PENDING.  Wake it up.
 993         */
 994        if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
 995                ocfs2_wake_downconvert_thread(osb);
 996}
 997
 998/* Locked version for callers of ocfs2_dlm_lock() */
 999static void lockres_clear_pending(struct ocfs2_lock_res *lockres,
1000                                  unsigned int generation,
1001                                  struct ocfs2_super *osb)
1002{
1003        unsigned long flags;
1004
1005        spin_lock_irqsave(&lockres->l_lock, flags);
1006        __lockres_clear_pending(lockres, generation, osb);
1007        spin_unlock_irqrestore(&lockres->l_lock, flags);
1008}
1009
1010static unsigned int lockres_set_pending(struct ocfs2_lock_res *lockres)
1011{
1012        assert_spin_locked(&lockres->l_lock);
1013        BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
1014
1015        lockres_or_flags(lockres, OCFS2_LOCK_PENDING);
1016
1017        return lockres->l_pending_gen;
1018}
1019
1020static void ocfs2_blocking_ast(struct ocfs2_dlm_lksb *lksb, int level)
1021{
1022        struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb);
1023        struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
1024        int needs_downconvert;
1025        unsigned long flags;
1026
1027        BUG_ON(level <= DLM_LOCK_NL);
1028
1029        mlog(ML_BASTS, "BAST fired for lockres %s, blocking %d, level %d, "
1030             "type %s\n", lockres->l_name, level, lockres->l_level,
1031             ocfs2_lock_type_string(lockres->l_type));
1032
1033        /*
1034         * We can skip the bast for locks which don't enable caching -
1035         * they'll be dropped at the earliest possible time anyway.
1036         */
1037        if (lockres->l_flags & OCFS2_LOCK_NOCACHE)
1038                return;
1039
1040        spin_lock_irqsave(&lockres->l_lock, flags);
1041        needs_downconvert = ocfs2_generic_handle_bast(lockres, level);
1042        if (needs_downconvert)
1043                ocfs2_schedule_blocked_lock(osb, lockres);
1044        spin_unlock_irqrestore(&lockres->l_lock, flags);
1045
1046        wake_up(&lockres->l_event);
1047
1048        ocfs2_wake_downconvert_thread(osb);
1049}
1050
1051static void ocfs2_locking_ast(struct ocfs2_dlm_lksb *lksb)
1052{
1053        struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb);
1054        struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
1055        unsigned long flags;
1056        int status;
1057
1058        spin_lock_irqsave(&lockres->l_lock, flags);
1059
1060        status = ocfs2_dlm_lock_status(&lockres->l_lksb);
1061
1062        if (status == -EAGAIN) {
1063                lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1064                goto out;
1065        }
1066
1067        if (status) {
1068                mlog(ML_ERROR, "lockres %s: lksb status value of %d!\n",
1069                     lockres->l_name, status);
1070                spin_unlock_irqrestore(&lockres->l_lock, flags);
1071                return;
1072        }
1073
1074        mlog(ML_BASTS, "AST fired for lockres %s, action %d, unlock %d, "
1075             "level %d => %d\n", lockres->l_name, lockres->l_action,
1076             lockres->l_unlock_action, lockres->l_level, lockres->l_requested);
1077
1078        switch(lockres->l_action) {
1079        case OCFS2_AST_ATTACH:
1080                ocfs2_generic_handle_attach_action(lockres);
1081                lockres_clear_flags(lockres, OCFS2_LOCK_LOCAL);
1082                break;
1083        case OCFS2_AST_CONVERT:
1084                ocfs2_generic_handle_convert_action(lockres);
1085                break;
1086        case OCFS2_AST_DOWNCONVERT:
1087                ocfs2_generic_handle_downconvert_action(lockres);
1088                break;
1089        default:
1090                mlog(ML_ERROR, "lockres %s: AST fired with invalid action: %u, "
1091                     "flags 0x%lx, unlock: %u\n",
1092                     lockres->l_name, lockres->l_action, lockres->l_flags,
1093                     lockres->l_unlock_action);
1094                BUG();
1095        }
1096out:
1097        /* set it to something invalid so if we get called again we
1098         * can catch it. */
1099        lockres->l_action = OCFS2_AST_INVALID;
1100
1101        /* Did we try to cancel this lock?  Clear that state */
1102        if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT)
1103                lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1104
1105        /*
1106         * We may have beaten the locking functions here.  We certainly
1107         * know that dlm_lock() has been called :-)
1108         * Because we can't have two lock calls in flight at once, we
1109         * can use lockres->l_pending_gen.
1110         */
1111        __lockres_clear_pending(lockres, lockres->l_pending_gen,  osb);
1112
1113        wake_up(&lockres->l_event);
1114        spin_unlock_irqrestore(&lockres->l_lock, flags);
1115}
1116
1117static void ocfs2_unlock_ast(struct ocfs2_dlm_lksb *lksb, int error)
1118{
1119        struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb);
1120        unsigned long flags;
1121
1122        mlog(ML_BASTS, "UNLOCK AST fired for lockres %s, action = %d\n",
1123             lockres->l_name, lockres->l_unlock_action);
1124
1125        spin_lock_irqsave(&lockres->l_lock, flags);
1126        if (error) {
1127                mlog(ML_ERROR, "Dlm passes error %d for lock %s, "
1128                     "unlock_action %d\n", error, lockres->l_name,
1129                     lockres->l_unlock_action);
1130                spin_unlock_irqrestore(&lockres->l_lock, flags);
1131                return;
1132        }
1133
1134        switch(lockres->l_unlock_action) {
1135        case OCFS2_UNLOCK_CANCEL_CONVERT:
1136                mlog(0, "Cancel convert success for %s\n", lockres->l_name);
1137                lockres->l_action = OCFS2_AST_INVALID;
1138                /* Downconvert thread may have requeued this lock, we
1139                 * need to wake it. */
1140                if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
1141                        ocfs2_wake_downconvert_thread(ocfs2_get_lockres_osb(lockres));
1142                break;
1143        case OCFS2_UNLOCK_DROP_LOCK:
1144                lockres->l_level = DLM_LOCK_IV;
1145                break;
1146        default:
1147                BUG();
1148        }
1149
1150        lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1151        lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1152        wake_up(&lockres->l_event);
1153        spin_unlock_irqrestore(&lockres->l_lock, flags);
1154}
1155
1156/*
1157 * This is the filesystem locking protocol.  It provides the lock handling
1158 * hooks for the underlying DLM.  It has a maximum version number.
1159 * The version number allows interoperability with systems running at
1160 * the same major number and an equal or smaller minor number.
1161 *
1162 * Whenever the filesystem does new things with locks (adds or removes a
1163 * lock, orders them differently, does different things underneath a lock),
1164 * the version must be changed.  The protocol is negotiated when joining
1165 * the dlm domain.  A node may join the domain if its major version is
1166 * identical to all other nodes and its minor version is greater than
1167 * or equal to all other nodes.  When its minor version is greater than
1168 * the other nodes, it will run at the minor version specified by the
1169 * other nodes.
1170 *
1171 * If a locking change is made that will not be compatible with older
1172 * versions, the major number must be increased and the minor version set
1173 * to zero.  If a change merely adds a behavior that can be disabled when
1174 * speaking to older versions, the minor version must be increased.  If a
1175 * change adds a fully backwards compatible change (eg, LVB changes that
1176 * are just ignored by older versions), the version does not need to be
1177 * updated.
1178 */
1179static struct ocfs2_locking_protocol lproto = {
1180        .lp_max_version = {
1181                .pv_major = OCFS2_LOCKING_PROTOCOL_MAJOR,
1182                .pv_minor = OCFS2_LOCKING_PROTOCOL_MINOR,
1183        },
1184        .lp_lock_ast            = ocfs2_locking_ast,
1185        .lp_blocking_ast        = ocfs2_blocking_ast,
1186        .lp_unlock_ast          = ocfs2_unlock_ast,
1187};
1188
1189void ocfs2_set_locking_protocol(void)
1190{
1191        ocfs2_stack_glue_set_max_proto_version(&lproto.lp_max_version);
1192}
1193
1194static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
1195                                                int convert)
1196{
1197        unsigned long flags;
1198
1199        spin_lock_irqsave(&lockres->l_lock, flags);
1200        lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1201        lockres_clear_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
1202        if (convert)
1203                lockres->l_action = OCFS2_AST_INVALID;
1204        else
1205                lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1206        spin_unlock_irqrestore(&lockres->l_lock, flags);
1207
1208        wake_up(&lockres->l_event);
1209}
1210
1211/* Note: If we detect another process working on the lock (i.e.,
1212 * OCFS2_LOCK_BUSY), we'll bail out returning 0. It's up to the caller
1213 * to do the right thing in that case.
1214 */
1215static int ocfs2_lock_create(struct ocfs2_super *osb,
1216                             struct ocfs2_lock_res *lockres,
1217                             int level,
1218                             u32 dlm_flags)
1219{
1220        int ret = 0;
1221        unsigned long flags;
1222        unsigned int gen;
1223
1224        mlog(0, "lock %s, level = %d, flags = %u\n", lockres->l_name, level,
1225             dlm_flags);
1226
1227        spin_lock_irqsave(&lockres->l_lock, flags);
1228        if ((lockres->l_flags & OCFS2_LOCK_ATTACHED) ||
1229            (lockres->l_flags & OCFS2_LOCK_BUSY)) {
1230                spin_unlock_irqrestore(&lockres->l_lock, flags);
1231                goto bail;
1232        }
1233
1234        lockres->l_action = OCFS2_AST_ATTACH;
1235        lockres->l_requested = level;
1236        lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
1237        gen = lockres_set_pending(lockres);
1238        spin_unlock_irqrestore(&lockres->l_lock, flags);
1239
1240        ret = ocfs2_dlm_lock(osb->cconn,
1241                             level,
1242                             &lockres->l_lksb,
1243                             dlm_flags,
1244                             lockres->l_name,
1245                             OCFS2_LOCK_ID_MAX_LEN - 1);
1246        lockres_clear_pending(lockres, gen, osb);
1247        if (ret) {
1248                ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
1249                ocfs2_recover_from_dlm_error(lockres, 1);
1250        }
1251
1252        mlog(0, "lock %s, return from ocfs2_dlm_lock\n", lockres->l_name);
1253
1254bail:
1255        return ret;
1256}
1257
1258static inline int ocfs2_check_wait_flag(struct ocfs2_lock_res *lockres,
1259                                        int flag)
1260{
1261        unsigned long flags;
1262        int ret;
1263
1264        spin_lock_irqsave(&lockres->l_lock, flags);
1265        ret = lockres->l_flags & flag;
1266        spin_unlock_irqrestore(&lockres->l_lock, flags);
1267
1268        return ret;
1269}
1270
1271static inline void ocfs2_wait_on_busy_lock(struct ocfs2_lock_res *lockres)
1272
1273{
1274        wait_event(lockres->l_event,
1275                   !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_BUSY));
1276}
1277
1278static inline void ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res *lockres)
1279
1280{
1281        wait_event(lockres->l_event,
1282                   !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_REFRESHING));
1283}
1284
1285/* predict what lock level we'll be dropping down to on behalf
1286 * of another node, and return true if the currently wanted
1287 * level will be compatible with it. */
1288static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
1289                                                     int wanted)
1290{
1291        BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
1292
1293        return wanted <= ocfs2_highest_compat_lock_level(lockres->l_blocking);
1294}
1295
1296static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter *mw)
1297{
1298        INIT_LIST_HEAD(&mw->mw_item);
1299        init_completion(&mw->mw_complete);
1300        ocfs2_init_start_time(mw);
1301}
1302
1303static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter *mw)
1304{
1305        wait_for_completion(&mw->mw_complete);
1306        /* Re-arm the completion in case we want to wait on it again */
1307        INIT_COMPLETION(mw->mw_complete);
1308        return mw->mw_status;
1309}
1310
1311static void lockres_add_mask_waiter(struct ocfs2_lock_res *lockres,
1312                                    struct ocfs2_mask_waiter *mw,
1313                                    unsigned long mask,
1314                                    unsigned long goal)
1315{
1316        BUG_ON(!list_empty(&mw->mw_item));
1317
1318        assert_spin_locked(&lockres->l_lock);
1319
1320        list_add_tail(&mw->mw_item, &lockres->l_mask_waiters);
1321        mw->mw_mask = mask;
1322        mw->mw_goal = goal;
1323}
1324
1325/* returns 0 if the mw that was removed was already satisfied, -EBUSY
1326 * if the mask still hadn't reached its goal */
1327static int lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres,
1328                                      struct ocfs2_mask_waiter *mw)
1329{
1330        unsigned long flags;
1331        int ret = 0;
1332
1333        spin_lock_irqsave(&lockres->l_lock, flags);
1334        if (!list_empty(&mw->mw_item)) {
1335                if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
1336                        ret = -EBUSY;
1337
1338                list_del_init(&mw->mw_item);
1339                init_completion(&mw->mw_complete);
1340        }
1341        spin_unlock_irqrestore(&lockres->l_lock, flags);
1342
1343        return ret;
1344
1345}
1346
1347static int ocfs2_wait_for_mask_interruptible(struct ocfs2_mask_waiter *mw,
1348                                             struct ocfs2_lock_res *lockres)
1349{
1350        int ret;
1351
1352        ret = wait_for_completion_interruptible(&mw->mw_complete);
1353        if (ret)
1354                lockres_remove_mask_waiter(lockres, mw);
1355        else
1356                ret = mw->mw_status;
1357        /* Re-arm the completion in case we want to wait on it again */
1358        INIT_COMPLETION(mw->mw_complete);
1359        return ret;
1360}
1361
1362static int __ocfs2_cluster_lock(struct ocfs2_super *osb,
1363                                struct ocfs2_lock_res *lockres,
1364                                int level,
1365                                u32 lkm_flags,
1366                                int arg_flags,
1367                                int l_subclass,
1368                                unsigned long caller_ip)
1369{
1370        struct ocfs2_mask_waiter mw;
1371        int wait, catch_signals = !(osb->s_mount_opt & OCFS2_MOUNT_NOINTR);
1372        int ret = 0; /* gcc doesn't realize wait = 1 guarantees ret is set */
1373        unsigned long flags;
1374        unsigned int gen;
1375        int noqueue_attempted = 0;
1376
1377        ocfs2_init_mask_waiter(&mw);
1378
1379        if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
1380                lkm_flags |= DLM_LKF_VALBLK;
1381
1382again:
1383        wait = 0;
1384
1385        spin_lock_irqsave(&lockres->l_lock, flags);
1386
1387        if (catch_signals && signal_pending(current)) {
1388                ret = -ERESTARTSYS;
1389                goto unlock;
1390        }
1391
1392        mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING,
1393                        "Cluster lock called on freeing lockres %s! flags "
1394                        "0x%lx\n", lockres->l_name, lockres->l_flags);
1395
1396        /* We only compare against the currently granted level
1397         * here. If the lock is blocked waiting on a downconvert,
1398         * we'll get caught below. */
1399        if (lockres->l_flags & OCFS2_LOCK_BUSY &&
1400            level > lockres->l_level) {
1401                /* is someone sitting in dlm_lock? If so, wait on
1402                 * them. */
1403                lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1404                wait = 1;
1405                goto unlock;
1406        }
1407
1408        if (lockres->l_flags & OCFS2_LOCK_UPCONVERT_FINISHING) {
1409                /*
1410                 * We've upconverted. If the lock now has a level we can
1411                 * work with, we take it. If, however, the lock is not at the
1412                 * required level, we go thru the full cycle. One way this could
1413                 * happen is if a process requesting an upconvert to PR is
1414                 * closely followed by another requesting upconvert to an EX.
1415                 * If the process requesting EX lands here, we want it to
1416                 * continue attempting to upconvert and let the process
1417                 * requesting PR take the lock.
1418                 * If multiple processes request upconvert to PR, the first one
1419                 * here will take the lock. The others will have to go thru the
1420                 * OCFS2_LOCK_BLOCKED check to ensure that there is no pending
1421                 * downconvert request.
1422                 */
1423                if (level <= lockres->l_level)
1424                        goto update_holders;
1425        }
1426
1427        if (lockres->l_flags & OCFS2_LOCK_BLOCKED &&
1428            !ocfs2_may_continue_on_blocked_lock(lockres, level)) {
1429                /* is the lock is currently blocked on behalf of
1430                 * another node */
1431                lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BLOCKED, 0);
1432                wait = 1;
1433                goto unlock;
1434        }
1435
1436        if (level > lockres->l_level) {
1437                if (noqueue_attempted > 0) {
1438                        ret = -EAGAIN;
1439                        goto unlock;
1440                }
1441                if (lkm_flags & DLM_LKF_NOQUEUE)
1442                        noqueue_attempted = 1;
1443
1444                if (lockres->l_action != OCFS2_AST_INVALID)
1445                        mlog(ML_ERROR, "lockres %s has action %u pending\n",
1446                             lockres->l_name, lockres->l_action);
1447
1448                if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
1449                        lockres->l_action = OCFS2_AST_ATTACH;
1450                        lkm_flags &= ~DLM_LKF_CONVERT;
1451                } else {
1452                        lockres->l_action = OCFS2_AST_CONVERT;
1453                        lkm_flags |= DLM_LKF_CONVERT;
1454                }
1455
1456                lockres->l_requested = level;
1457                lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
1458                gen = lockres_set_pending(lockres);
1459                spin_unlock_irqrestore(&lockres->l_lock, flags);
1460
1461                BUG_ON(level == DLM_LOCK_IV);
1462                BUG_ON(level == DLM_LOCK_NL);
1463
1464                mlog(ML_BASTS, "lockres %s, convert from %d to %d\n",
1465                     lockres->l_name, lockres->l_level, level);
1466
1467                /* call dlm_lock to upgrade lock now */
1468                ret = ocfs2_dlm_lock(osb->cconn,
1469                                     level,
1470                                     &lockres->l_lksb,
1471                                     lkm_flags,
1472                                     lockres->l_name,
1473                                     OCFS2_LOCK_ID_MAX_LEN - 1);
1474                lockres_clear_pending(lockres, gen, osb);
1475                if (ret) {
1476                        if (!(lkm_flags & DLM_LKF_NOQUEUE) ||
1477                            (ret != -EAGAIN)) {
1478                                ocfs2_log_dlm_error("ocfs2_dlm_lock",
1479                                                    ret, lockres);
1480                        }
1481                        ocfs2_recover_from_dlm_error(lockres, 1);
1482                        goto out;
1483                }
1484
1485                mlog(0, "lock %s, successful return from ocfs2_dlm_lock\n",
1486                     lockres->l_name);
1487
1488                /* At this point we've gone inside the dlm and need to
1489                 * complete our work regardless. */
1490                catch_signals = 0;
1491
1492                /* wait for busy to clear and carry on */
1493                goto again;
1494        }
1495
1496update_holders:
1497        /* Ok, if we get here then we're good to go. */
1498        ocfs2_inc_holders(lockres, level);
1499
1500        ret = 0;
1501unlock:
1502        lockres_clear_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
1503
1504        spin_unlock_irqrestore(&lockres->l_lock, flags);
1505out:
1506        /*
1507         * This is helping work around a lock inversion between the page lock
1508         * and dlm locks.  One path holds the page lock while calling aops
1509         * which block acquiring dlm locks.  The voting thread holds dlm
1510         * locks while acquiring page locks while down converting data locks.
1511         * This block is helping an aop path notice the inversion and back
1512         * off to unlock its page lock before trying the dlm lock again.
1513         */
1514        if (wait && arg_flags & OCFS2_LOCK_NONBLOCK &&
1515            mw.mw_mask & (OCFS2_LOCK_BUSY|OCFS2_LOCK_BLOCKED)) {
1516                wait = 0;
1517                if (lockres_remove_mask_waiter(lockres, &mw))
1518                        ret = -EAGAIN;
1519                else
1520                        goto again;
1521        }
1522        if (wait) {
1523                ret = ocfs2_wait_for_mask(&mw);
1524                if (ret == 0)
1525                        goto again;
1526                mlog_errno(ret);
1527        }
1528        ocfs2_update_lock_stats(lockres, level, &mw, ret);
1529
1530#ifdef CONFIG_DEBUG_LOCK_ALLOC
1531        if (!ret && lockres->l_lockdep_map.key != NULL) {
1532                if (level == DLM_LOCK_PR)
1533                        rwsem_acquire_read(&lockres->l_lockdep_map, l_subclass,
1534                                !!(arg_flags & OCFS2_META_LOCK_NOQUEUE),
1535                                caller_ip);
1536                else
1537                        rwsem_acquire(&lockres->l_lockdep_map, l_subclass,
1538                                !!(arg_flags & OCFS2_META_LOCK_NOQUEUE),
1539                                caller_ip);
1540        }
1541#endif
1542        return ret;
1543}
1544
1545static inline int ocfs2_cluster_lock(struct ocfs2_super *osb,
1546                                     struct ocfs2_lock_res *lockres,
1547                                     int level,
1548                                     u32 lkm_flags,
1549                                     int arg_flags)
1550{
1551        return __ocfs2_cluster_lock(osb, lockres, level, lkm_flags, arg_flags,
1552                                    0, _RET_IP_);
1553}
1554
1555
1556static void __ocfs2_cluster_unlock(struct ocfs2_super *osb,
1557                                   struct ocfs2_lock_res *lockres,
1558                                   int level,
1559                                   unsigned long caller_ip)
1560{
1561        unsigned long flags;
1562
1563        spin_lock_irqsave(&lockres->l_lock, flags);
1564        ocfs2_dec_holders(lockres, level);
1565        ocfs2_downconvert_on_unlock(osb, lockres);
1566        spin_unlock_irqrestore(&lockres->l_lock, flags);
1567#ifdef CONFIG_DEBUG_LOCK_ALLOC
1568        if (lockres->l_lockdep_map.key != NULL)
1569                rwsem_release(&lockres->l_lockdep_map, 1, caller_ip);
1570#endif
1571}
1572
1573static int ocfs2_create_new_lock(struct ocfs2_super *osb,
1574                                 struct ocfs2_lock_res *lockres,
1575                                 int ex,
1576                                 int local)
1577{
1578        int level =  ex ? DLM_LOCK_EX : DLM_LOCK_PR;
1579        unsigned long flags;
1580        u32 lkm_flags = local ? DLM_LKF_LOCAL : 0;
1581
1582        spin_lock_irqsave(&lockres->l_lock, flags);
1583        BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
1584        lockres_or_flags(lockres, OCFS2_LOCK_LOCAL);
1585        spin_unlock_irqrestore(&lockres->l_lock, flags);
1586
1587        return ocfs2_lock_create(osb, lockres, level, lkm_flags);
1588}
1589
1590/* Grants us an EX lock on the data and metadata resources, skipping
1591 * the normal cluster directory lookup. Use this ONLY on newly created
1592 * inodes which other nodes can't possibly see, and which haven't been
1593 * hashed in the inode hash yet. This can give us a good performance
1594 * increase as it'll skip the network broadcast normally associated
1595 * with creating a new lock resource. */
1596int ocfs2_create_new_inode_locks(struct inode *inode)
1597{
1598        int ret;
1599        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1600
1601        BUG_ON(!inode);
1602        BUG_ON(!ocfs2_inode_is_new(inode));
1603
1604        mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
1605
1606        /* NOTE: That we don't increment any of the holder counts, nor
1607         * do we add anything to a journal handle. Since this is
1608         * supposed to be a new inode which the cluster doesn't know
1609         * about yet, there is no need to.  As far as the LVB handling
1610         * is concerned, this is basically like acquiring an EX lock
1611         * on a resource which has an invalid one -- we'll set it
1612         * valid when we release the EX. */
1613
1614        ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_rw_lockres, 1, 1);
1615        if (ret) {
1616                mlog_errno(ret);
1617                goto bail;
1618        }
1619
1620        /*
1621         * We don't want to use DLM_LKF_LOCAL on a meta data lock as they
1622         * don't use a generation in their lock names.
1623         */
1624        ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_inode_lockres, 1, 0);
1625        if (ret) {
1626                mlog_errno(ret);
1627                goto bail;
1628        }
1629
1630        ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_open_lockres, 0, 0);
1631        if (ret) {
1632                mlog_errno(ret);
1633                goto bail;
1634        }
1635
1636bail:
1637        return ret;
1638}
1639
1640int ocfs2_rw_lock(struct inode *inode, int write)
1641{
1642        int status, level;
1643        struct ocfs2_lock_res *lockres;
1644        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1645
1646        BUG_ON(!inode);
1647
1648        mlog(0, "inode %llu take %s RW lock\n",
1649             (unsigned long long)OCFS2_I(inode)->ip_blkno,
1650             write ? "EXMODE" : "PRMODE");
1651
1652        if (ocfs2_mount_local(osb))
1653                return 0;
1654
1655        lockres = &OCFS2_I(inode)->ip_rw_lockres;
1656
1657        level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1658
1659        status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level, 0,
1660                                    0);
1661        if (status < 0)
1662                mlog_errno(status);
1663
1664        return status;
1665}
1666
1667void ocfs2_rw_unlock(struct inode *inode, int write)
1668{
1669        int level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1670        struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_rw_lockres;
1671        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1672
1673        mlog(0, "inode %llu drop %s RW lock\n",
1674             (unsigned long long)OCFS2_I(inode)->ip_blkno,
1675             write ? "EXMODE" : "PRMODE");
1676
1677        if (!ocfs2_mount_local(osb))
1678                ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
1679}
1680
1681/*
1682 * ocfs2_open_lock always get PR mode lock.
1683 */
1684int ocfs2_open_lock(struct inode *inode)
1685{
1686        int status = 0;
1687        struct ocfs2_lock_res *lockres;
1688        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1689
1690        BUG_ON(!inode);
1691
1692        mlog(0, "inode %llu take PRMODE open lock\n",
1693             (unsigned long long)OCFS2_I(inode)->ip_blkno);
1694
1695        if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb))
1696                goto out;
1697
1698        lockres = &OCFS2_I(inode)->ip_open_lockres;
1699
1700        status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres,
1701                                    DLM_LOCK_PR, 0, 0);
1702        if (status < 0)
1703                mlog_errno(status);
1704
1705out:
1706        return status;
1707}
1708
1709int ocfs2_try_open_lock(struct inode *inode, int write)
1710{
1711        int status = 0, level;
1712        struct ocfs2_lock_res *lockres;
1713        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1714
1715        BUG_ON(!inode);
1716
1717        mlog(0, "inode %llu try to take %s open lock\n",
1718             (unsigned long long)OCFS2_I(inode)->ip_blkno,
1719             write ? "EXMODE" : "PRMODE");
1720
1721        if (ocfs2_is_hard_readonly(osb)) {
1722                if (write)
1723                        status = -EROFS;
1724                goto out;
1725        }
1726
1727        if (ocfs2_mount_local(osb))
1728                goto out;
1729
1730        lockres = &OCFS2_I(inode)->ip_open_lockres;
1731
1732        level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
1733
1734        /*
1735         * The file system may already holding a PRMODE/EXMODE open lock.
1736         * Since we pass DLM_LKF_NOQUEUE, the request won't block waiting on
1737         * other nodes and the -EAGAIN will indicate to the caller that
1738         * this inode is still in use.
1739         */
1740        status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres,
1741                                    level, DLM_LKF_NOQUEUE, 0);
1742
1743out:
1744        return status;
1745}
1746
1747/*
1748 * ocfs2_open_unlock unlock PR and EX mode open locks.
1749 */
1750void ocfs2_open_unlock(struct inode *inode)
1751{
1752        struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_open_lockres;
1753        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1754
1755        mlog(0, "inode %llu drop open lock\n",
1756             (unsigned long long)OCFS2_I(inode)->ip_blkno);
1757
1758        if (ocfs2_mount_local(osb))
1759                goto out;
1760
1761        if(lockres->l_ro_holders)
1762                ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres,
1763                                     DLM_LOCK_PR);
1764        if(lockres->l_ex_holders)
1765                ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres,
1766                                     DLM_LOCK_EX);
1767
1768out:
1769        return;
1770}
1771
1772static int ocfs2_flock_handle_signal(struct ocfs2_lock_res *lockres,
1773                                     int level)
1774{
1775        int ret;
1776        struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
1777        unsigned long flags;
1778        struct ocfs2_mask_waiter mw;
1779
1780        ocfs2_init_mask_waiter(&mw);
1781
1782retry_cancel:
1783        spin_lock_irqsave(&lockres->l_lock, flags);
1784        if (lockres->l_flags & OCFS2_LOCK_BUSY) {
1785                ret = ocfs2_prepare_cancel_convert(osb, lockres);
1786                if (ret) {
1787                        spin_unlock_irqrestore(&lockres->l_lock, flags);
1788                        ret = ocfs2_cancel_convert(osb, lockres);
1789                        if (ret < 0) {
1790                                mlog_errno(ret);
1791                                goto out;
1792                        }
1793                        goto retry_cancel;
1794                }
1795                lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1796                spin_unlock_irqrestore(&lockres->l_lock, flags);
1797
1798                ocfs2_wait_for_mask(&mw);
1799                goto retry_cancel;
1800        }
1801
1802        ret = -ERESTARTSYS;
1803        /*
1804         * We may still have gotten the lock, in which case there's no
1805         * point to restarting the syscall.
1806         */
1807        if (lockres->l_level == level)
1808                ret = 0;
1809
1810        mlog(0, "Cancel returning %d. flags: 0x%lx, level: %d, act: %d\n", ret,
1811             lockres->l_flags, lockres->l_level, lockres->l_action);
1812
1813        spin_unlock_irqrestore(&lockres->l_lock, flags);
1814
1815out:
1816        return ret;
1817}
1818
1819/*
1820 * ocfs2_file_lock() and ocfs2_file_unlock() map to a single pair of
1821 * flock() calls. The locking approach this requires is sufficiently
1822 * different from all other cluster lock types that we implement a
1823 * separate path to the "low-level" dlm calls. In particular:
1824 *
1825 * - No optimization of lock levels is done - we take at exactly
1826 *   what's been requested.
1827 *
1828 * - No lock caching is employed. We immediately downconvert to
1829 *   no-lock at unlock time. This also means flock locks never go on
1830 *   the blocking list).
1831 *
1832 * - Since userspace can trivially deadlock itself with flock, we make
1833 *   sure to allow cancellation of a misbehaving applications flock()
1834 *   request.
1835 *
1836 * - Access to any flock lockres doesn't require concurrency, so we
1837 *   can simplify the code by requiring the caller to guarantee
1838 *   serialization of dlmglue flock calls.
1839 */
1840int ocfs2_file_lock(struct file *file, int ex, int trylock)
1841{
1842        int ret, level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
1843        unsigned int lkm_flags = trylock ? DLM_LKF_NOQUEUE : 0;
1844        unsigned long flags;
1845        struct ocfs2_file_private *fp = file->private_data;
1846        struct ocfs2_lock_res *lockres = &fp->fp_flock;
1847        struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
1848        struct ocfs2_mask_waiter mw;
1849
1850        ocfs2_init_mask_waiter(&mw);
1851
1852        if ((lockres->l_flags & OCFS2_LOCK_BUSY) ||
1853            (lockres->l_level > DLM_LOCK_NL)) {
1854                mlog(ML_ERROR,
1855                     "File lock \"%s\" has busy or locked state: flags: 0x%lx, "
1856                     "level: %u\n", lockres->l_name, lockres->l_flags,
1857                     lockres->l_level);
1858                return -EINVAL;
1859        }
1860
1861        spin_lock_irqsave(&lockres->l_lock, flags);
1862        if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
1863                lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1864                spin_unlock_irqrestore(&lockres->l_lock, flags);
1865
1866                /*
1867                 * Get the lock at NLMODE to start - that way we
1868                 * can cancel the upconvert request if need be.
1869                 */
1870                ret = ocfs2_lock_create(osb, lockres, DLM_LOCK_NL, 0);
1871                if (ret < 0) {
1872                        mlog_errno(ret);
1873                        goto out;
1874                }
1875
1876                ret = ocfs2_wait_for_mask(&mw);
1877                if (ret) {
1878                        mlog_errno(ret);
1879                        goto out;
1880                }
1881                spin_lock_irqsave(&lockres->l_lock, flags);
1882        }
1883
1884        lockres->l_action = OCFS2_AST_CONVERT;
1885        lkm_flags |= DLM_LKF_CONVERT;
1886        lockres->l_requested = level;
1887        lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
1888
1889        lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1890        spin_unlock_irqrestore(&lockres->l_lock, flags);
1891
1892        ret = ocfs2_dlm_lock(osb->cconn, level, &lockres->l_lksb, lkm_flags,
1893                             lockres->l_name, OCFS2_LOCK_ID_MAX_LEN - 1);
1894        if (ret) {
1895                if (!trylock || (ret != -EAGAIN)) {
1896                        ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
1897                        ret = -EINVAL;
1898                }
1899
1900                ocfs2_recover_from_dlm_error(lockres, 1);
1901                lockres_remove_mask_waiter(lockres, &mw);
1902                goto out;
1903        }
1904
1905        ret = ocfs2_wait_for_mask_interruptible(&mw, lockres);
1906        if (ret == -ERESTARTSYS) {
1907                /*
1908                 * Userspace can cause deadlock itself with
1909                 * flock(). Current behavior locally is to allow the
1910                 * deadlock, but abort the system call if a signal is
1911                 * received. We follow this example, otherwise a
1912                 * poorly written program could sit in kernel until
1913                 * reboot.
1914                 *
1915                 * Handling this is a bit more complicated for Ocfs2
1916                 * though. We can't exit this function with an
1917                 * outstanding lock request, so a cancel convert is
1918                 * required. We intentionally overwrite 'ret' - if the
1919                 * cancel fails and the lock was granted, it's easier
1920                 * to just bubble success back up to the user.
1921                 */
1922                ret = ocfs2_flock_handle_signal(lockres, level);
1923        } else if (!ret && (level > lockres->l_level)) {
1924                /* Trylock failed asynchronously */
1925                BUG_ON(!trylock);
1926                ret = -EAGAIN;
1927        }
1928
1929out:
1930
1931        mlog(0, "Lock: \"%s\" ex: %d, trylock: %d, returns: %d\n",
1932             lockres->l_name, ex, trylock, ret);
1933        return ret;
1934}
1935
1936void ocfs2_file_unlock(struct file *file)
1937{
1938        int ret;
1939        unsigned int gen;
1940        unsigned long flags;
1941        struct ocfs2_file_private *fp = file->private_data;
1942        struct ocfs2_lock_res *lockres = &fp->fp_flock;
1943        struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
1944        struct ocfs2_mask_waiter mw;
1945
1946        ocfs2_init_mask_waiter(&mw);
1947
1948        if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED))
1949                return;
1950
1951        if (lockres->l_level == DLM_LOCK_NL)
1952                return;
1953
1954        mlog(0, "Unlock: \"%s\" flags: 0x%lx, level: %d, act: %d\n",
1955             lockres->l_name, lockres->l_flags, lockres->l_level,
1956             lockres->l_action);
1957
1958        spin_lock_irqsave(&lockres->l_lock, flags);
1959        /*
1960         * Fake a blocking ast for the downconvert code.
1961         */
1962        lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
1963        lockres->l_blocking = DLM_LOCK_EX;
1964
1965        gen = ocfs2_prepare_downconvert(lockres, DLM_LOCK_NL);
1966        lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1967        spin_unlock_irqrestore(&lockres->l_lock, flags);
1968
1969        ret = ocfs2_downconvert_lock(osb, lockres, DLM_LOCK_NL, 0, gen);
1970        if (ret) {
1971                mlog_errno(ret);
1972                return;
1973        }
1974
1975        ret = ocfs2_wait_for_mask(&mw);
1976        if (ret)
1977                mlog_errno(ret);
1978}
1979
1980static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
1981                                        struct ocfs2_lock_res *lockres)
1982{
1983        int kick = 0;
1984
1985        /* If we know that another node is waiting on our lock, kick
1986         * the downconvert thread * pre-emptively when we reach a release
1987         * condition. */
1988        if (lockres->l_flags & OCFS2_LOCK_BLOCKED) {
1989                switch(lockres->l_blocking) {
1990                case DLM_LOCK_EX:
1991                        if (!lockres->l_ex_holders && !lockres->l_ro_holders)
1992                                kick = 1;
1993                        break;
1994                case DLM_LOCK_PR:
1995                        if (!lockres->l_ex_holders)
1996                                kick = 1;
1997                        break;
1998                default:
1999                        BUG();
2000                }
2001        }
2002
2003        if (kick)
2004                ocfs2_wake_downconvert_thread(osb);
2005}
2006
2007#define OCFS2_SEC_BITS   34
2008#define OCFS2_SEC_SHIFT  (64 - 34)
2009#define OCFS2_NSEC_MASK  ((1ULL << OCFS2_SEC_SHIFT) - 1)
2010
2011/* LVB only has room for 64 bits of time here so we pack it for
2012 * now. */
2013static u64 ocfs2_pack_timespec(struct timespec *spec)
2014{
2015        u64 res;
2016        u64 sec = spec->tv_sec;
2017        u32 nsec = spec->tv_nsec;
2018
2019        res = (sec << OCFS2_SEC_SHIFT) | (nsec & OCFS2_NSEC_MASK);
2020
2021        return res;
2022}
2023
2024/* Call this with the lockres locked. I am reasonably sure we don't
2025 * need ip_lock in this function as anyone who would be changing those
2026 * values is supposed to be blocked in ocfs2_inode_lock right now. */
2027static void __ocfs2_stuff_meta_lvb(struct inode *inode)
2028{
2029        struct ocfs2_inode_info *oi = OCFS2_I(inode);
2030        struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
2031        struct ocfs2_meta_lvb *lvb;
2032
2033        lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2034
2035        /*
2036         * Invalidate the LVB of a deleted inode - this way other
2037         * nodes are forced to go to disk and discover the new inode
2038         * status.
2039         */
2040        if (oi->ip_flags & OCFS2_INODE_DELETED) {
2041                lvb->lvb_version = 0;
2042                goto out;
2043        }
2044
2045        lvb->lvb_version   = OCFS2_LVB_VERSION;
2046        lvb->lvb_isize     = cpu_to_be64(i_size_read(inode));
2047        lvb->lvb_iclusters = cpu_to_be32(oi->ip_clusters);
2048        lvb->lvb_iuid      = cpu_to_be32(i_uid_read(inode));
2049        lvb->lvb_igid      = cpu_to_be32(i_gid_read(inode));
2050        lvb->lvb_imode     = cpu_to_be16(inode->i_mode);
2051        lvb->lvb_inlink    = cpu_to_be16(inode->i_nlink);
2052        lvb->lvb_iatime_packed  =
2053                cpu_to_be64(ocfs2_pack_timespec(&inode->i_atime));
2054        lvb->lvb_ictime_packed =
2055                cpu_to_be64(ocfs2_pack_timespec(&inode->i_ctime));
2056        lvb->lvb_imtime_packed =
2057                cpu_to_be64(ocfs2_pack_timespec(&inode->i_mtime));
2058        lvb->lvb_iattr    = cpu_to_be32(oi->ip_attr);
2059        lvb->lvb_idynfeatures = cpu_to_be16(oi->ip_dyn_features);
2060        lvb->lvb_igeneration = cpu_to_be32(inode->i_generation);
2061
2062out:
2063        mlog_meta_lvb(0, lockres);
2064}
2065
2066static void ocfs2_unpack_timespec(struct timespec *spec,
2067                                  u64 packed_time)
2068{
2069        spec->tv_sec = packed_time >> OCFS2_SEC_SHIFT;
2070        spec->tv_nsec = packed_time & OCFS2_NSEC_MASK;
2071}
2072
2073static void ocfs2_refresh_inode_from_lvb(struct inode *inode)
2074{
2075        struct ocfs2_inode_info *oi = OCFS2_I(inode);
2076        struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
2077        struct ocfs2_meta_lvb *lvb;
2078
2079        mlog_meta_lvb(0, lockres);
2080
2081        lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2082
2083        /* We're safe here without the lockres lock... */
2084        spin_lock(&oi->ip_lock);
2085        oi->ip_clusters = be32_to_cpu(lvb->lvb_iclusters);
2086        i_size_write(inode, be64_to_cpu(lvb->lvb_isize));
2087
2088        oi->ip_attr = be32_to_cpu(lvb->lvb_iattr);
2089        oi->ip_dyn_features = be16_to_cpu(lvb->lvb_idynfeatures);
2090        ocfs2_set_inode_flags(inode);
2091
2092        /* fast-symlinks are a special case */
2093        if (S_ISLNK(inode->i_mode) && !oi->ip_clusters)
2094                inode->i_blocks = 0;
2095        else
2096                inode->i_blocks = ocfs2_inode_sector_count(inode);
2097
2098        i_uid_write(inode, be32_to_cpu(lvb->lvb_iuid));
2099        i_gid_write(inode, be32_to_cpu(lvb->lvb_igid));
2100        inode->i_mode    = be16_to_cpu(lvb->lvb_imode);
2101        set_nlink(inode, be16_to_cpu(lvb->lvb_inlink));
2102        ocfs2_unpack_timespec(&inode->i_atime,
2103                              be64_to_cpu(lvb->lvb_iatime_packed));
2104        ocfs2_unpack_timespec(&inode->i_mtime,
2105                              be64_to_cpu(lvb->lvb_imtime_packed));
2106        ocfs2_unpack_timespec(&inode->i_ctime,
2107                              be64_to_cpu(lvb->lvb_ictime_packed));
2108        spin_unlock(&oi->ip_lock);
2109}
2110
2111static inline int ocfs2_meta_lvb_is_trustable(struct inode *inode,
2112                                              struct ocfs2_lock_res *lockres)
2113{
2114        struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2115
2116        if (ocfs2_dlm_lvb_valid(&lockres->l_lksb)
2117            && lvb->lvb_version == OCFS2_LVB_VERSION
2118            && be32_to_cpu(lvb->lvb_igeneration) == inode->i_generation)
2119                return 1;
2120        return 0;
2121}
2122
2123/* Determine whether a lock resource needs to be refreshed, and
2124 * arbitrate who gets to refresh it.
2125 *
2126 *   0 means no refresh needed.
2127 *
2128 *   > 0 means you need to refresh this and you MUST call
2129 *   ocfs2_complete_lock_res_refresh afterwards. */
2130static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res *lockres)
2131{
2132        unsigned long flags;
2133        int status = 0;
2134
2135refresh_check:
2136        spin_lock_irqsave(&lockres->l_lock, flags);
2137        if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) {
2138                spin_unlock_irqrestore(&lockres->l_lock, flags);
2139                goto bail;
2140        }
2141
2142        if (lockres->l_flags & OCFS2_LOCK_REFRESHING) {
2143                spin_unlock_irqrestore(&lockres->l_lock, flags);
2144
2145                ocfs2_wait_on_refreshing_lock(lockres);
2146                goto refresh_check;
2147        }
2148
2149        /* Ok, I'll be the one to refresh this lock. */
2150        lockres_or_flags(lockres, OCFS2_LOCK_REFRESHING);
2151        spin_unlock_irqrestore(&lockres->l_lock, flags);
2152
2153        status = 1;
2154bail:
2155        mlog(0, "status %d\n", status);
2156        return status;
2157}
2158
2159/* If status is non zero, I'll mark it as not being in refresh
2160 * anymroe, but i won't clear the needs refresh flag. */
2161static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockres,
2162                                                   int status)
2163{
2164        unsigned long flags;
2165
2166        spin_lock_irqsave(&lockres->l_lock, flags);
2167        lockres_clear_flags(lockres, OCFS2_LOCK_REFRESHING);
2168        if (!status)
2169                lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
2170        spin_unlock_irqrestore(&lockres->l_lock, flags);
2171
2172        wake_up(&lockres->l_event);
2173}
2174
2175/* may or may not return a bh if it went to disk. */
2176static int ocfs2_inode_lock_update(struct inode *inode,
2177                                  struct buffer_head **bh)
2178{
2179        int status = 0;
2180        struct ocfs2_inode_info *oi = OCFS2_I(inode);
2181        struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
2182        struct ocfs2_dinode *fe;
2183        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2184
2185        if (ocfs2_mount_local(osb))
2186                goto bail;
2187
2188        spin_lock(&oi->ip_lock);
2189        if (oi->ip_flags & OCFS2_INODE_DELETED) {
2190                mlog(0, "Orphaned inode %llu was deleted while we "
2191                     "were waiting on a lock. ip_flags = 0x%x\n",
2192                     (unsigned long long)oi->ip_blkno, oi->ip_flags);
2193                spin_unlock(&oi->ip_lock);
2194                status = -ENOENT;
2195                goto bail;
2196        }
2197        spin_unlock(&oi->ip_lock);
2198
2199        if (!ocfs2_should_refresh_lock_res(lockres))
2200                goto bail;
2201
2202        /* This will discard any caching information we might have had
2203         * for the inode metadata. */
2204        ocfs2_metadata_cache_purge(INODE_CACHE(inode));
2205
2206        ocfs2_extent_map_trunc(inode, 0);
2207
2208        if (ocfs2_meta_lvb_is_trustable(inode, lockres)) {
2209                mlog(0, "Trusting LVB on inode %llu\n",
2210                     (unsigned long long)oi->ip_blkno);
2211                ocfs2_refresh_inode_from_lvb(inode);
2212        } else {
2213                /* Boo, we have to go to disk. */
2214                /* read bh, cast, ocfs2_refresh_inode */
2215                status = ocfs2_read_inode_block(inode, bh);
2216                if (status < 0) {
2217                        mlog_errno(status);
2218                        goto bail_refresh;
2219                }
2220                fe = (struct ocfs2_dinode *) (*bh)->b_data;
2221
2222                /* This is a good chance to make sure we're not
2223                 * locking an invalid object.  ocfs2_read_inode_block()
2224                 * already checked that the inode block is sane.
2225                 *
2226                 * We bug on a stale inode here because we checked
2227                 * above whether it was wiped from disk. The wiping
2228                 * node provides a guarantee that we receive that
2229                 * message and can mark the inode before dropping any
2230                 * locks associated with it. */
2231                mlog_bug_on_msg(inode->i_generation !=
2232                                le32_to_cpu(fe->i_generation),
2233                                "Invalid dinode %llu disk generation: %u "
2234                                "inode->i_generation: %u\n",
2235                                (unsigned long long)oi->ip_blkno,
2236                                le32_to_cpu(fe->i_generation),
2237                                inode->i_generation);
2238                mlog_bug_on_msg(le64_to_cpu(fe->i_dtime) ||
2239                                !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)),
2240                                "Stale dinode %llu dtime: %llu flags: 0x%x\n",
2241                                (unsigned long long)oi->ip_blkno,
2242                                (unsigned long long)le64_to_cpu(fe->i_dtime),
2243                                le32_to_cpu(fe->i_flags));
2244
2245                ocfs2_refresh_inode(inode, fe);
2246                ocfs2_track_lock_refresh(lockres);
2247        }
2248
2249        status = 0;
2250bail_refresh:
2251        ocfs2_complete_lock_res_refresh(lockres, status);
2252bail:
2253        return status;
2254}
2255
2256static int ocfs2_assign_bh(struct inode *inode,
2257                           struct buffer_head **ret_bh,
2258                           struct buffer_head *passed_bh)
2259{
2260        int status;
2261
2262        if (passed_bh) {
2263                /* Ok, the update went to disk for us, use the
2264                 * returned bh. */
2265                *ret_bh = passed_bh;
2266                get_bh(*ret_bh);
2267
2268                return 0;
2269        }
2270
2271        status = ocfs2_read_inode_block(inode, ret_bh);
2272        if (status < 0)
2273                mlog_errno(status);
2274
2275        return status;
2276}
2277
2278/*
2279 * returns < 0 error if the callback will never be called, otherwise
2280 * the result of the lock will be communicated via the callback.
2281 */
2282int ocfs2_inode_lock_full_nested(struct inode *inode,
2283                                 struct buffer_head **ret_bh,
2284                                 int ex,
2285                                 int arg_flags,
2286                                 int subclass)
2287{
2288        int status, level, acquired;
2289        u32 dlm_flags;
2290        struct ocfs2_lock_res *lockres = NULL;
2291        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2292        struct buffer_head *local_bh = NULL;
2293
2294        BUG_ON(!inode);
2295
2296        mlog(0, "inode %llu, take %s META lock\n",
2297             (unsigned long long)OCFS2_I(inode)->ip_blkno,
2298             ex ? "EXMODE" : "PRMODE");
2299
2300        status = 0;
2301        acquired = 0;
2302        /* We'll allow faking a readonly metadata lock for
2303         * rodevices. */
2304        if (ocfs2_is_hard_readonly(osb)) {
2305                if (ex)
2306                        status = -EROFS;
2307                goto getbh;
2308        }
2309
2310        if (ocfs2_mount_local(osb))
2311                goto local;
2312
2313        if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
2314                ocfs2_wait_for_recovery(osb);
2315
2316        lockres = &OCFS2_I(inode)->ip_inode_lockres;
2317        level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2318        dlm_flags = 0;
2319        if (arg_flags & OCFS2_META_LOCK_NOQUEUE)
2320                dlm_flags |= DLM_LKF_NOQUEUE;
2321
2322        status = __ocfs2_cluster_lock(osb, lockres, level, dlm_flags,
2323                                      arg_flags, subclass, _RET_IP_);
2324        if (status < 0) {
2325                if (status != -EAGAIN && status != -EIOCBRETRY)
2326                        mlog_errno(status);
2327                goto bail;
2328        }
2329
2330        /* Notify the error cleanup path to drop the cluster lock. */
2331        acquired = 1;
2332
2333        /* We wait twice because a node may have died while we were in
2334         * the lower dlm layers. The second time though, we've
2335         * committed to owning this lock so we don't allow signals to
2336         * abort the operation. */
2337        if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
2338                ocfs2_wait_for_recovery(osb);
2339
2340local:
2341        /*
2342         * We only see this flag if we're being called from
2343         * ocfs2_read_locked_inode(). It means we're locking an inode
2344         * which hasn't been populated yet, so clear the refresh flag
2345         * and let the caller handle it.
2346         */
2347        if (inode->i_state & I_NEW) {
2348                status = 0;
2349                if (lockres)
2350                        ocfs2_complete_lock_res_refresh(lockres, 0);
2351                goto bail;
2352        }
2353
2354        /* This is fun. The caller may want a bh back, or it may
2355         * not. ocfs2_inode_lock_update definitely wants one in, but
2356         * may or may not read one, depending on what's in the
2357         * LVB. The result of all of this is that we've *only* gone to
2358         * disk if we have to, so the complexity is worthwhile. */
2359        status = ocfs2_inode_lock_update(inode, &local_bh);
2360        if (status < 0) {
2361                if (status != -ENOENT)
2362                        mlog_errno(status);
2363                goto bail;
2364        }
2365getbh:
2366        if (ret_bh) {
2367                status = ocfs2_assign_bh(inode, ret_bh, local_bh);
2368                if (status < 0) {
2369                        mlog_errno(status);
2370                        goto bail;
2371                }
2372        }
2373
2374bail:
2375        if (status < 0) {
2376                if (ret_bh && (*ret_bh)) {
2377                        brelse(*ret_bh);
2378                        *ret_bh = NULL;
2379                }
2380                if (acquired)
2381                        ocfs2_inode_unlock(inode, ex);
2382        }
2383
2384        if (local_bh)
2385                brelse(local_bh);
2386
2387        return status;
2388}
2389
2390/*
2391 * This is working around a lock inversion between tasks acquiring DLM
2392 * locks while holding a page lock and the downconvert thread which
2393 * blocks dlm lock acquiry while acquiring page locks.
2394 *
2395 * ** These _with_page variantes are only intended to be called from aop
2396 * methods that hold page locks and return a very specific *positive* error
2397 * code that aop methods pass up to the VFS -- test for errors with != 0. **
2398 *
2399 * The DLM is called such that it returns -EAGAIN if it would have
2400 * blocked waiting for the downconvert thread.  In that case we unlock
2401 * our page so the downconvert thread can make progress.  Once we've
2402 * done this we have to return AOP_TRUNCATED_PAGE so the aop method
2403 * that called us can bubble that back up into the VFS who will then
2404 * immediately retry the aop call.
2405 *
2406 * We do a blocking lock and immediate unlock before returning, though, so that
2407 * the lock has a great chance of being cached on this node by the time the VFS
2408 * calls back to retry the aop.    This has a potential to livelock as nodes
2409 * ping locks back and forth, but that's a risk we're willing to take to avoid
2410 * the lock inversion simply.
2411 */
2412int ocfs2_inode_lock_with_page(struct inode *inode,
2413                              struct buffer_head **ret_bh,
2414                              int ex,
2415                              struct page *page)
2416{
2417        int ret;
2418
2419        ret = ocfs2_inode_lock_full(inode, ret_bh, ex, OCFS2_LOCK_NONBLOCK);
2420        if (ret == -EAGAIN) {
2421                unlock_page(page);
2422                if (ocfs2_inode_lock(inode, ret_bh, ex) == 0)
2423                        ocfs2_inode_unlock(inode, ex);
2424                ret = AOP_TRUNCATED_PAGE;
2425        }
2426
2427        return ret;
2428}
2429
2430int ocfs2_inode_lock_atime(struct inode *inode,
2431                          struct vfsmount *vfsmnt,
2432                          int *level)
2433{
2434        int ret;
2435
2436        ret = ocfs2_inode_lock(inode, NULL, 0);
2437        if (ret < 0) {
2438                mlog_errno(ret);
2439                return ret;
2440        }
2441
2442        /*
2443         * If we should update atime, we will get EX lock,
2444         * otherwise we just get PR lock.
2445         */
2446        if (ocfs2_should_update_atime(inode, vfsmnt)) {
2447                struct buffer_head *bh = NULL;
2448
2449                ocfs2_inode_unlock(inode, 0);
2450                ret = ocfs2_inode_lock(inode, &bh, 1);
2451                if (ret < 0) {
2452                        mlog_errno(ret);
2453                        return ret;
2454                }
2455                *level = 1;
2456                if (ocfs2_should_update_atime(inode, vfsmnt))
2457                        ocfs2_update_inode_atime(inode, bh);
2458                if (bh)
2459                        brelse(bh);
2460        } else
2461                *level = 0;
2462
2463        return ret;
2464}
2465
2466void ocfs2_inode_unlock(struct inode *inode,
2467                       int ex)
2468{
2469        int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2470        struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_inode_lockres;
2471        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2472
2473        mlog(0, "inode %llu drop %s META lock\n",
2474             (unsigned long long)OCFS2_I(inode)->ip_blkno,
2475             ex ? "EXMODE" : "PRMODE");
2476
2477        if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb)) &&
2478            !ocfs2_mount_local(osb))
2479                ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
2480}
2481
2482int ocfs2_orphan_scan_lock(struct ocfs2_super *osb, u32 *seqno)
2483{
2484        struct ocfs2_lock_res *lockres;
2485        struct ocfs2_orphan_scan_lvb *lvb;
2486        int status = 0;
2487
2488        if (ocfs2_is_hard_readonly(osb))
2489                return -EROFS;
2490
2491        if (ocfs2_mount_local(osb))
2492                return 0;
2493
2494        lockres = &osb->osb_orphan_scan.os_lockres;
2495        status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0);
2496        if (status < 0)
2497                return status;
2498
2499        lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2500        if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) &&
2501            lvb->lvb_version == OCFS2_ORPHAN_LVB_VERSION)
2502                *seqno = be32_to_cpu(lvb->lvb_os_seqno);
2503        else
2504                *seqno = osb->osb_orphan_scan.os_seqno + 1;
2505
2506        return status;
2507}
2508
2509void ocfs2_orphan_scan_unlock(struct ocfs2_super *osb, u32 seqno)
2510{
2511        struct ocfs2_lock_res *lockres;
2512        struct ocfs2_orphan_scan_lvb *lvb;
2513
2514        if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb)) {
2515                lockres = &osb->osb_orphan_scan.os_lockres;
2516                lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2517                lvb->lvb_version = OCFS2_ORPHAN_LVB_VERSION;
2518                lvb->lvb_os_seqno = cpu_to_be32(seqno);
2519                ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
2520        }
2521}
2522
2523int ocfs2_super_lock(struct ocfs2_super *osb,
2524                     int ex)
2525{
2526        int status = 0;
2527        int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2528        struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
2529
2530        if (ocfs2_is_hard_readonly(osb))
2531                return -EROFS;
2532
2533        if (ocfs2_mount_local(osb))
2534                goto bail;
2535
2536        status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
2537        if (status < 0) {
2538                mlog_errno(status);
2539                goto bail;
2540        }
2541
2542        /* The super block lock path is really in the best position to
2543         * know when resources covered by the lock need to be
2544         * refreshed, so we do it here. Of course, making sense of
2545         * everything is up to the caller :) */
2546        status = ocfs2_should_refresh_lock_res(lockres);
2547        if (status < 0) {
2548                ocfs2_cluster_unlock(osb, lockres, level);
2549                mlog_errno(status);
2550                goto bail;
2551        }
2552        if (status) {
2553                status = ocfs2_refresh_slot_info(osb);
2554
2555                ocfs2_complete_lock_res_refresh(lockres, status);
2556
2557                if (status < 0) {
2558                        ocfs2_cluster_unlock(osb, lockres, level);
2559                        mlog_errno(status);
2560                }
2561                ocfs2_track_lock_refresh(lockres);
2562        }
2563bail:
2564        return status;
2565}
2566
2567void ocfs2_super_unlock(struct ocfs2_super *osb,
2568                        int ex)
2569{
2570        int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2571        struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
2572
2573        if (!ocfs2_mount_local(osb))
2574                ocfs2_cluster_unlock(osb, lockres, level);
2575}
2576
2577int ocfs2_rename_lock(struct ocfs2_super *osb)
2578{
2579        int status;
2580        struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
2581
2582        if (ocfs2_is_hard_readonly(osb))
2583                return -EROFS;
2584
2585        if (ocfs2_mount_local(osb))
2586                return 0;
2587
2588        status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0);
2589        if (status < 0)
2590                mlog_errno(status);
2591
2592        return status;
2593}
2594
2595void ocfs2_rename_unlock(struct ocfs2_super *osb)
2596{
2597        struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
2598
2599        if (!ocfs2_mount_local(osb))
2600                ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
2601}
2602
2603int ocfs2_nfs_sync_lock(struct ocfs2_super *osb, int ex)
2604{
2605        int status;
2606        struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres;
2607
2608        if (ocfs2_is_hard_readonly(osb))
2609                return -EROFS;
2610
2611        if (ocfs2_mount_local(osb))
2612                return 0;
2613
2614        status = ocfs2_cluster_lock(osb, lockres, ex ? LKM_EXMODE : LKM_PRMODE,
2615                                    0, 0);
2616        if (status < 0)
2617                mlog(ML_ERROR, "lock on nfs sync lock failed %d\n", status);
2618
2619        return status;
2620}
2621
2622void ocfs2_nfs_sync_unlock(struct ocfs2_super *osb, int ex)
2623{
2624        struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres;
2625
2626        if (!ocfs2_mount_local(osb))
2627                ocfs2_cluster_unlock(osb, lockres,
2628                                     ex ? LKM_EXMODE : LKM_PRMODE);
2629}
2630
2631int ocfs2_dentry_lock(struct dentry *dentry, int ex)
2632{
2633        int ret;
2634        int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2635        struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2636        struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2637
2638        BUG_ON(!dl);
2639
2640        if (ocfs2_is_hard_readonly(osb)) {
2641                if (ex)
2642                        return -EROFS;
2643                return 0;
2644        }
2645
2646        if (ocfs2_mount_local(osb))
2647                return 0;
2648
2649        ret = ocfs2_cluster_lock(osb, &dl->dl_lockres, level, 0, 0);
2650        if (ret < 0)
2651                mlog_errno(ret);
2652
2653        return ret;
2654}
2655
2656void ocfs2_dentry_unlock(struct dentry *dentry, int ex)
2657{
2658        int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
2659        struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2660        struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2661
2662        if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb))
2663                ocfs2_cluster_unlock(osb, &dl->dl_lockres, level);
2664}
2665
2666/* Reference counting of the dlm debug structure. We want this because
2667 * open references on the debug inodes can live on after a mount, so
2668 * we can't rely on the ocfs2_super to always exist. */
2669static void ocfs2_dlm_debug_free(struct kref *kref)
2670{
2671        struct ocfs2_dlm_debug *dlm_debug;
2672
2673        dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt);
2674
2675        kfree(dlm_debug);
2676}
2677
2678void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug)
2679{
2680        if (dlm_debug)
2681                kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free);
2682}
2683
2684static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug)
2685{
2686        kref_get(&debug->d_refcnt);
2687}
2688
2689struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void)
2690{
2691        struct ocfs2_dlm_debug *dlm_debug;
2692
2693        dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL);
2694        if (!dlm_debug) {
2695                mlog_errno(-ENOMEM);
2696                goto out;
2697        }
2698
2699        kref_init(&dlm_debug->d_refcnt);
2700        INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking);
2701        dlm_debug->d_locking_state = NULL;
2702out:
2703        return dlm_debug;
2704}
2705
2706/* Access to this is arbitrated for us via seq_file->sem. */
2707struct ocfs2_dlm_seq_priv {
2708        struct ocfs2_dlm_debug *p_dlm_debug;
2709        struct ocfs2_lock_res p_iter_res;
2710        struct ocfs2_lock_res p_tmp_res;
2711};
2712
2713static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start,
2714                                                 struct ocfs2_dlm_seq_priv *priv)
2715{
2716        struct ocfs2_lock_res *iter, *ret = NULL;
2717        struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug;
2718
2719        assert_spin_locked(&ocfs2_dlm_tracking_lock);
2720
2721        list_for_each_entry(iter, &start->l_debug_list, l_debug_list) {
2722                /* discover the head of the list */
2723                if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) {
2724                        mlog(0, "End of list found, %p\n", ret);
2725                        break;
2726                }
2727
2728                /* We track our "dummy" iteration lockres' by a NULL
2729                 * l_ops field. */
2730                if (iter->l_ops != NULL) {
2731                        ret = iter;
2732                        break;
2733                }
2734        }
2735
2736        return ret;
2737}
2738
2739static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos)
2740{
2741        struct ocfs2_dlm_seq_priv *priv = m->private;
2742        struct ocfs2_lock_res *iter;
2743
2744        spin_lock(&ocfs2_dlm_tracking_lock);
2745        iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv);
2746        if (iter) {
2747                /* Since lockres' have the lifetime of their container
2748                 * (which can be inodes, ocfs2_supers, etc) we want to
2749                 * copy this out to a temporary lockres while still
2750                 * under the spinlock. Obviously after this we can't
2751                 * trust any pointers on the copy returned, but that's
2752                 * ok as the information we want isn't typically held
2753                 * in them. */
2754                priv->p_tmp_res = *iter;
2755                iter = &priv->p_tmp_res;
2756        }
2757        spin_unlock(&ocfs2_dlm_tracking_lock);
2758
2759        return iter;
2760}
2761
2762static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v)
2763{
2764}
2765
2766static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos)
2767{
2768        struct ocfs2_dlm_seq_priv *priv = m->private;
2769        struct ocfs2_lock_res *iter = v;
2770        struct ocfs2_lock_res *dummy = &priv->p_iter_res;
2771
2772        spin_lock(&ocfs2_dlm_tracking_lock);
2773        iter = ocfs2_dlm_next_res(iter, priv);
2774        list_del_init(&dummy->l_debug_list);
2775        if (iter) {
2776                list_add(&dummy->l_debug_list, &iter->l_debug_list);
2777                priv->p_tmp_res = *iter;
2778                iter = &priv->p_tmp_res;
2779        }
2780        spin_unlock(&ocfs2_dlm_tracking_lock);
2781
2782        return iter;
2783}
2784
2785/*
2786 * Version is used by debugfs.ocfs2 to determine the format being used
2787 *
2788 * New in version 2
2789 *      - Lock stats printed
2790 * New in version 3
2791 *      - Max time in lock stats is in usecs (instead of nsecs)
2792 */
2793#define OCFS2_DLM_DEBUG_STR_VERSION 3
2794static int ocfs2_dlm_seq_show(struct seq_file *m, void *v)
2795{
2796        int i;
2797        char *lvb;
2798        struct ocfs2_lock_res *lockres = v;
2799
2800        if (!lockres)
2801                return -EINVAL;
2802
2803        seq_printf(m, "0x%x\t", OCFS2_DLM_DEBUG_STR_VERSION);
2804
2805        if (lockres->l_type == OCFS2_LOCK_TYPE_DENTRY)
2806                seq_printf(m, "%.*s%08x\t", OCFS2_DENTRY_LOCK_INO_START - 1,
2807                           lockres->l_name,
2808                           (unsigned int)ocfs2_get_dentry_lock_ino(lockres));
2809        else
2810                seq_printf(m, "%.*s\t", OCFS2_LOCK_ID_MAX_LEN, lockres->l_name);
2811
2812        seq_printf(m, "%d\t"
2813                   "0x%lx\t"
2814                   "0x%x\t"
2815                   "0x%x\t"
2816                   "%u\t"
2817                   "%u\t"
2818                   "%d\t"
2819                   "%d\t",
2820                   lockres->l_level,
2821                   lockres->l_flags,
2822                   lockres->l_action,
2823                   lockres->l_unlock_action,
2824                   lockres->l_ro_holders,
2825                   lockres->l_ex_holders,
2826                   lockres->l_requested,
2827                   lockres->l_blocking);
2828
2829        /* Dump the raw LVB */
2830        lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2831        for(i = 0; i < DLM_LVB_LEN; i++)
2832                seq_printf(m, "0x%x\t", lvb[i]);
2833
2834#ifdef CONFIG_OCFS2_FS_STATS
2835# define lock_num_prmode(_l)            ((_l)->l_lock_prmode.ls_gets)
2836# define lock_num_exmode(_l)            ((_l)->l_lock_exmode.ls_gets)
2837# define lock_num_prmode_failed(_l)     ((_l)->l_lock_prmode.ls_fail)
2838# define lock_num_exmode_failed(_l)     ((_l)->l_lock_exmode.ls_fail)
2839# define lock_total_prmode(_l)          ((_l)->l_lock_prmode.ls_total)
2840# define lock_total_exmode(_l)          ((_l)->l_lock_exmode.ls_total)
2841# define lock_max_prmode(_l)            ((_l)->l_lock_prmode.ls_max)
2842# define lock_max_exmode(_l)            ((_l)->l_lock_exmode.ls_max)
2843# define lock_refresh(_l)               ((_l)->l_lock_refresh)
2844#else
2845# define lock_num_prmode(_l)            (0)
2846# define lock_num_exmode(_l)            (0)
2847# define lock_num_prmode_failed(_l)     (0)
2848# define lock_num_exmode_failed(_l)     (0)
2849# define lock_total_prmode(_l)          (0ULL)
2850# define lock_total_exmode(_l)          (0ULL)
2851# define lock_max_prmode(_l)            (0)
2852# define lock_max_exmode(_l)            (0)
2853# define lock_refresh(_l)               (0)
2854#endif
2855        /* The following seq_print was added in version 2 of this output */
2856        seq_printf(m, "%u\t"
2857                   "%u\t"
2858                   "%u\t"
2859                   "%u\t"
2860                   "%llu\t"
2861                   "%llu\t"
2862                   "%u\t"
2863                   "%u\t"
2864                   "%u\t",
2865                   lock_num_prmode(lockres),
2866                   lock_num_exmode(lockres),
2867                   lock_num_prmode_failed(lockres),
2868                   lock_num_exmode_failed(lockres),
2869                   lock_total_prmode(lockres),
2870                   lock_total_exmode(lockres),
2871                   lock_max_prmode(lockres),
2872                   lock_max_exmode(lockres),
2873                   lock_refresh(lockres));
2874
2875        /* End the line */
2876        seq_printf(m, "\n");
2877        return 0;
2878}
2879
2880static const struct seq_operations ocfs2_dlm_seq_ops = {
2881        .start =        ocfs2_dlm_seq_start,
2882        .stop =         ocfs2_dlm_seq_stop,
2883        .next =         ocfs2_dlm_seq_next,
2884        .show =         ocfs2_dlm_seq_show,
2885};
2886
2887static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file)
2888{
2889        struct seq_file *seq = file->private_data;
2890        struct ocfs2_dlm_seq_priv *priv = seq->private;
2891        struct ocfs2_lock_res *res = &priv->p_iter_res;
2892
2893        ocfs2_remove_lockres_tracking(res);
2894        ocfs2_put_dlm_debug(priv->p_dlm_debug);
2895        return seq_release_private(inode, file);
2896}
2897
2898static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file)
2899{
2900        int ret;
2901        struct ocfs2_dlm_seq_priv *priv;
2902        struct seq_file *seq;
2903        struct ocfs2_super *osb;
2904
2905        priv = kzalloc(sizeof(struct ocfs2_dlm_seq_priv), GFP_KERNEL);
2906        if (!priv) {
2907                ret = -ENOMEM;
2908                mlog_errno(ret);
2909                goto out;
2910        }
2911        osb = inode->i_private;
2912        ocfs2_get_dlm_debug(osb->osb_dlm_debug);
2913        priv->p_dlm_debug = osb->osb_dlm_debug;
2914        INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list);
2915
2916        ret = seq_open(file, &ocfs2_dlm_seq_ops);
2917        if (ret) {
2918                kfree(priv);
2919                mlog_errno(ret);
2920                goto out;
2921        }
2922
2923        seq = file->private_data;
2924        seq->private = priv;
2925
2926        ocfs2_add_lockres_tracking(&priv->p_iter_res,
2927                                   priv->p_dlm_debug);
2928
2929out:
2930        return ret;
2931}
2932
2933static const struct file_operations ocfs2_dlm_debug_fops = {
2934        .open =         ocfs2_dlm_debug_open,
2935        .release =      ocfs2_dlm_debug_release,
2936        .read =         seq_read,
2937        .llseek =       seq_lseek,
2938};
2939
2940static int ocfs2_dlm_init_debug(struct ocfs2_super *osb)
2941{
2942        int ret = 0;
2943        struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
2944
2945        dlm_debug->d_locking_state = debugfs_create_file("locking_state",
2946                                                         S_IFREG|S_IRUSR,
2947                                                         osb->osb_debug_root,
2948                                                         osb,
2949                                                         &ocfs2_dlm_debug_fops);
2950        if (!dlm_debug->d_locking_state) {
2951                ret = -EINVAL;
2952                mlog(ML_ERROR,
2953                     "Unable to create locking state debugfs file.\n");
2954                goto out;
2955        }
2956
2957        ocfs2_get_dlm_debug(dlm_debug);
2958out:
2959        return ret;
2960}
2961
2962static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb)
2963{
2964        struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
2965
2966        if (dlm_debug) {
2967                debugfs_remove(dlm_debug->d_locking_state);
2968                ocfs2_put_dlm_debug(dlm_debug);
2969        }
2970}
2971
2972int ocfs2_dlm_init(struct ocfs2_super *osb)
2973{
2974        int status = 0;
2975        struct ocfs2_cluster_connection *conn = NULL;
2976
2977        if (ocfs2_mount_local(osb)) {
2978                osb->node_num = 0;
2979                goto local;
2980        }
2981
2982        status = ocfs2_dlm_init_debug(osb);
2983        if (status < 0) {
2984                mlog_errno(status);
2985                goto bail;
2986        }
2987
2988        /* launch downconvert thread */
2989        osb->dc_task = kthread_run(ocfs2_downconvert_thread, osb, "ocfs2dc");
2990        if (IS_ERR(osb->dc_task)) {
2991                status = PTR_ERR(osb->dc_task);
2992                osb->dc_task = NULL;
2993                mlog_errno(status);
2994                goto bail;
2995        }
2996
2997        /* for now, uuid == domain */
2998        status = ocfs2_cluster_connect(osb->osb_cluster_stack,
2999                                       osb->uuid_str,
3000                                       strlen(osb->uuid_str),
3001                                       &lproto, ocfs2_do_node_down, osb,
3002                                       &conn);
3003        if (status) {
3004                mlog_errno(status);
3005                goto bail;
3006        }
3007
3008        status = ocfs2_cluster_this_node(&osb->node_num);
3009        if (status < 0) {
3010                mlog_errno(status);
3011                mlog(ML_ERROR,
3012                     "could not find this host's node number\n");
3013                ocfs2_cluster_disconnect(conn, 0);
3014                goto bail;
3015        }
3016
3017local:
3018        ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb);
3019        ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb);
3020        ocfs2_nfs_sync_lock_res_init(&osb->osb_nfs_sync_lockres, osb);
3021        ocfs2_orphan_scan_lock_res_init(&osb->osb_orphan_scan.os_lockres, osb);
3022
3023        osb->cconn = conn;
3024
3025        status = 0;
3026bail:
3027        if (status < 0) {
3028                ocfs2_dlm_shutdown_debug(osb);
3029                if (osb->dc_task)
3030                        kthread_stop(osb->dc_task);
3031        }
3032
3033        return status;
3034}
3035
3036void ocfs2_dlm_shutdown(struct ocfs2_super *osb,
3037                        int hangup_pending)
3038{
3039        ocfs2_drop_osb_locks(osb);
3040
3041        /*
3042         * Now that we have dropped all locks and ocfs2_dismount_volume()
3043         * has disabled recovery, the DLM won't be talking to us.  It's
3044         * safe to tear things down before disconnecting the cluster.
3045         */
3046
3047        if (osb->dc_task) {
3048                kthread_stop(osb->dc_task);
3049                osb->dc_task = NULL;
3050        }
3051
3052        ocfs2_lock_res_free(&osb->osb_super_lockres);
3053        ocfs2_lock_res_free(&osb->osb_rename_lockres);
3054        ocfs2_lock_res_free(&osb->osb_nfs_sync_lockres);
3055        ocfs2_lock_res_free(&osb->osb_orphan_scan.os_lockres);
3056
3057        ocfs2_cluster_disconnect(osb->cconn, hangup_pending);
3058        osb->cconn = NULL;
3059
3060        ocfs2_dlm_shutdown_debug(osb);
3061}
3062
3063static int ocfs2_drop_lock(struct ocfs2_super *osb,
3064                           struct ocfs2_lock_res *lockres)
3065{
3066        int ret;
3067        unsigned long flags;
3068        u32 lkm_flags = 0;
3069
3070        /* We didn't get anywhere near actually using this lockres. */
3071        if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED))
3072                goto out;
3073
3074        if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
3075                lkm_flags |= DLM_LKF_VALBLK;
3076
3077        spin_lock_irqsave(&lockres->l_lock, flags);
3078
3079        mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING),
3080                        "lockres %s, flags 0x%lx\n",
3081                        lockres->l_name, lockres->l_flags);
3082
3083        while (lockres->l_flags & OCFS2_LOCK_BUSY) {
3084                mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = "
3085                     "%u, unlock_action = %u\n",
3086                     lockres->l_name, lockres->l_flags, lockres->l_action,
3087                     lockres->l_unlock_action);
3088
3089                spin_unlock_irqrestore(&lockres->l_lock, flags);
3090
3091                /* XXX: Today we just wait on any busy
3092                 * locks... Perhaps we need to cancel converts in the
3093                 * future? */
3094                ocfs2_wait_on_busy_lock(lockres);
3095
3096                spin_lock_irqsave(&lockres->l_lock, flags);
3097        }
3098
3099        if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
3100                if (lockres->l_flags & OCFS2_LOCK_ATTACHED &&
3101                    lockres->l_level == DLM_LOCK_EX &&
3102                    !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
3103                        lockres->l_ops->set_lvb(lockres);
3104        }
3105
3106        if (lockres->l_flags & OCFS2_LOCK_BUSY)
3107                mlog(ML_ERROR, "destroying busy lock: \"%s\"\n",
3108                     lockres->l_name);
3109        if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
3110                mlog(0, "destroying blocked lock: \"%s\"\n", lockres->l_name);
3111
3112        if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
3113                spin_unlock_irqrestore(&lockres->l_lock, flags);
3114                goto out;
3115        }
3116
3117        lockres_clear_flags(lockres, OCFS2_LOCK_ATTACHED);
3118
3119        /* make sure we never get here while waiting for an ast to
3120         * fire. */
3121        BUG_ON(lockres->l_action != OCFS2_AST_INVALID);
3122
3123        /* is this necessary? */
3124        lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
3125        lockres->l_unlock_action = OCFS2_UNLOCK_DROP_LOCK;
3126        spin_unlock_irqrestore(&lockres->l_lock, flags);
3127
3128        mlog(0, "lock %s\n", lockres->l_name);
3129
3130        ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb, lkm_flags);
3131        if (ret) {
3132                ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
3133                mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags);
3134                ocfs2_dlm_dump_lksb(&lockres->l_lksb);
3135                BUG();
3136        }
3137        mlog(0, "lock %s, successful return from ocfs2_dlm_unlock\n",
3138             lockres->l_name);
3139
3140        ocfs2_wait_on_busy_lock(lockres);
3141out:
3142        return 0;
3143}
3144
3145/* Mark the lockres as being dropped. It will no longer be
3146 * queued if blocking, but we still may have to wait on it
3147 * being dequeued from the downconvert thread before we can consider
3148 * it safe to drop.
3149 *
3150 * You can *not* attempt to call cluster_lock on this lockres anymore. */
3151void ocfs2_mark_lockres_freeing(struct ocfs2_lock_res *lockres)
3152{
3153        int status;
3154        struct ocfs2_mask_waiter mw;
3155        unsigned long flags;
3156
3157        ocfs2_init_mask_waiter(&mw);
3158
3159        spin_lock_irqsave(&lockres->l_lock, flags);
3160        lockres->l_flags |= OCFS2_LOCK_FREEING;
3161        while (lockres->l_flags & OCFS2_LOCK_QUEUED) {
3162                lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0);
3163                spin_unlock_irqrestore(&lockres->l_lock, flags);
3164
3165                mlog(0, "Waiting on lockres %s\n", lockres->l_name);
3166
3167                status = ocfs2_wait_for_mask(&mw);
3168                if (status)
3169                        mlog_errno(status);
3170
3171                spin_lock_irqsave(&lockres->l_lock, flags);
3172        }
3173        spin_unlock_irqrestore(&lockres->l_lock, flags);
3174}
3175
3176void ocfs2_simple_drop_lockres(struct ocfs2_super *osb,
3177                               struct ocfs2_lock_res *lockres)
3178{
3179        int ret;
3180
3181        ocfs2_mark_lockres_freeing(lockres);
3182        ret = ocfs2_drop_lock(osb, lockres);
3183        if (ret)
3184                mlog_errno(ret);
3185}
3186
3187static void ocfs2_drop_osb_locks(struct ocfs2_super *osb)
3188{
3189        ocfs2_simple_drop_lockres(osb, &osb->osb_super_lockres);
3190        ocfs2_simple_drop_lockres(osb, &osb->osb_rename_lockres);
3191        ocfs2_simple_drop_lockres(osb, &osb->osb_nfs_sync_lockres);
3192        ocfs2_simple_drop_lockres(osb, &osb->osb_orphan_scan.os_lockres);
3193}
3194
3195int ocfs2_drop_inode_locks(struct inode *inode)
3196{
3197        int status, err;
3198
3199        /* No need to call ocfs2_mark_lockres_freeing here -
3200         * ocfs2_clear_inode has done it for us. */
3201
3202        err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
3203                              &OCFS2_I(inode)->ip_open_lockres);
3204        if (err < 0)
3205                mlog_errno(err);
3206
3207        status = err;
3208
3209        err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
3210                              &OCFS2_I(inode)->ip_inode_lockres);
3211        if (err < 0)
3212                mlog_errno(err);
3213        if (err < 0 && !status)
3214                status = err;
3215
3216        err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
3217                              &OCFS2_I(inode)->ip_rw_lockres);
3218        if (err < 0)
3219                mlog_errno(err);
3220        if (err < 0 && !status)
3221                status = err;
3222
3223        return status;
3224}
3225
3226static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
3227                                              int new_level)
3228{
3229        assert_spin_locked(&lockres->l_lock);
3230
3231        BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
3232
3233        if (lockres->l_level <= new_level) {
3234                mlog(ML_ERROR, "lockres %s, lvl %d <= %d, blcklst %d, mask %d, "
3235                     "type %d, flags 0x%lx, hold %d %d, act %d %d, req %d, "
3236                     "block %d, pgen %d\n", lockres->l_name, lockres->l_level,
3237                     new_level, list_empty(&lockres->l_blocked_list),
3238                     list_empty(&lockres->l_mask_waiters), lockres->l_type,
3239                     lockres->l_flags, lockres->l_ro_holders,
3240                     lockres->l_ex_holders, lockres->l_action,
3241                     lockres->l_unlock_action, lockres->l_requested,
3242                     lockres->l_blocking, lockres->l_pending_gen);
3243                BUG();
3244        }
3245
3246        mlog(ML_BASTS, "lockres %s, level %d => %d, blocking %d\n",
3247             lockres->l_name, lockres->l_level, new_level, lockres->l_blocking);
3248
3249        lockres->l_action = OCFS2_AST_DOWNCONVERT;
3250        lockres->l_requested = new_level;
3251        lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
3252        return lockres_set_pending(lockres);
3253}
3254
3255static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
3256                                  struct ocfs2_lock_res *lockres,
3257                                  int new_level,
3258                                  int lvb,
3259                                  unsigned int generation)
3260{
3261        int ret;
3262        u32 dlm_flags = DLM_LKF_CONVERT;
3263
3264        mlog(ML_BASTS, "lockres %s, level %d => %d\n", lockres->l_name,
3265             lockres->l_level, new_level);
3266
3267        if (lvb)
3268                dlm_flags |= DLM_LKF_VALBLK;
3269
3270        ret = ocfs2_dlm_lock(osb->cconn,
3271                             new_level,
3272                             &lockres->l_lksb,
3273                             dlm_flags,
3274                             lockres->l_name,
3275                             OCFS2_LOCK_ID_MAX_LEN - 1);
3276        lockres_clear_pending(lockres, generation, osb);
3277        if (ret) {
3278                ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
3279                ocfs2_recover_from_dlm_error(lockres, 1);
3280                goto bail;
3281        }
3282
3283        ret = 0;
3284bail:
3285        return ret;
3286}
3287
3288/* returns 1 when the caller should unlock and call ocfs2_dlm_unlock */
3289static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
3290                                        struct ocfs2_lock_res *lockres)
3291{
3292        assert_spin_locked(&lockres->l_lock);
3293
3294        if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
3295                /* If we're already trying to cancel a lock conversion
3296                 * then just drop the spinlock and allow the caller to
3297                 * requeue this lock. */
3298                mlog(ML_BASTS, "lockres %s, skip convert\n", lockres->l_name);
3299                return 0;
3300        }
3301
3302        /* were we in a convert when we got the bast fire? */
3303        BUG_ON(lockres->l_action != OCFS2_AST_CONVERT &&
3304               lockres->l_action != OCFS2_AST_DOWNCONVERT);
3305        /* set things up for the unlockast to know to just
3306         * clear out the ast_action and unset busy, etc. */
3307        lockres->l_unlock_action = OCFS2_UNLOCK_CANCEL_CONVERT;
3308
3309        mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_BUSY),
3310                        "lock %s, invalid flags: 0x%lx\n",
3311                        lockres->l_name, lockres->l_flags);
3312
3313        mlog(ML_BASTS, "lockres %s\n", lockres->l_name);
3314
3315        return 1;
3316}
3317
3318static int ocfs2_cancel_convert(struct ocfs2_super *osb,
3319                                struct ocfs2_lock_res *lockres)
3320{
3321        int ret;
3322
3323        ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb,
3324                               DLM_LKF_CANCEL);
3325        if (ret) {
3326                ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
3327                ocfs2_recover_from_dlm_error(lockres, 0);
3328        }
3329
3330        mlog(ML_BASTS, "lockres %s\n", lockres->l_name);
3331
3332        return ret;
3333}
3334
3335static int ocfs2_unblock_lock(struct ocfs2_super *osb,
3336                              struct ocfs2_lock_res *lockres,
3337                              struct ocfs2_unblock_ctl *ctl)
3338{
3339        unsigned long flags;
3340        int blocking;
3341        int new_level;
3342        int level;
3343        int ret = 0;
3344        int set_lvb = 0;
3345        unsigned int gen;
3346
3347        spin_lock_irqsave(&lockres->l_lock, flags);
3348
3349recheck:
3350        /*
3351         * Is it still blocking? If not, we have no more work to do.
3352         */
3353        if (!(lockres->l_flags & OCFS2_LOCK_BLOCKED)) {
3354                BUG_ON(lockres->l_blocking != DLM_LOCK_NL);
3355                spin_unlock_irqrestore(&lockres->l_lock, flags);
3356                ret = 0;
3357                goto leave;
3358        }
3359
3360        if (lockres->l_flags & OCFS2_LOCK_BUSY) {
3361                /* XXX
3362                 * This is a *big* race.  The OCFS2_LOCK_PENDING flag
3363                 * exists entirely for one reason - another thread has set
3364                 * OCFS2_LOCK_BUSY, but has *NOT* yet called dlm_lock().
3365                 *
3366                 * If we do ocfs2_cancel_convert() before the other thread
3367                 * calls dlm_lock(), our cancel will do nothing.  We will
3368                 * get no ast, and we will have no way of knowing the
3369                 * cancel failed.  Meanwhile, the other thread will call
3370                 * into dlm_lock() and wait...forever.
3371                 *
3372                 * Why forever?  Because another node has asked for the
3373                 * lock first; that's why we're here in unblock_lock().
3374                 *
3375                 * The solution is OCFS2_LOCK_PENDING.  When PENDING is
3376                 * set, we just requeue the unblock.  Only when the other
3377                 * thread has called dlm_lock() and cleared PENDING will
3378                 * we then cancel their request.
3379                 *
3380                 * All callers of dlm_lock() must set OCFS2_DLM_PENDING
3381                 * at the same time they set OCFS2_DLM_BUSY.  They must
3382                 * clear OCFS2_DLM_PENDING after dlm_lock() returns.
3383                 */
3384                if (lockres->l_flags & OCFS2_LOCK_PENDING) {
3385                        mlog(ML_BASTS, "lockres %s, ReQ: Pending\n",
3386                             lockres->l_name);
3387                        goto leave_requeue;
3388                }
3389
3390                ctl->requeue = 1;
3391                ret = ocfs2_prepare_cancel_convert(osb, lockres);
3392                spin_unlock_irqrestore(&lockres->l_lock, flags);
3393                if (ret) {
3394                        ret = ocfs2_cancel_convert(osb, lockres);
3395                        if (ret < 0)
3396                                mlog_errno(ret);
3397                }
3398                goto leave;
3399        }
3400
3401        /*
3402         * This prevents livelocks. OCFS2_LOCK_UPCONVERT_FINISHING flag is
3403         * set when the ast is received for an upconvert just before the
3404         * OCFS2_LOCK_BUSY flag is cleared. Now if the fs received a bast
3405         * on the heels of the ast, we want to delay the downconvert just
3406         * enough to allow the up requestor to do its task. Because this
3407         * lock is in the blocked queue, the lock will be downconverted
3408         * as soon as the requestor is done with the lock.
3409         */
3410        if (lockres->l_flags & OCFS2_LOCK_UPCONVERT_FINISHING)
3411                goto leave_requeue;
3412
3413        /*
3414         * How can we block and yet be at NL?  We were trying to upconvert
3415         * from NL and got canceled.  The code comes back here, and now
3416         * we notice and clear BLOCKING.
3417         */
3418        if (lockres->l_level == DLM_LOCK_NL) {
3419                BUG_ON(lockres->l_ex_holders || lockres->l_ro_holders);
3420                mlog(ML_BASTS, "lockres %s, Aborting dc\n", lockres->l_name);
3421                lockres->l_blocking = DLM_LOCK_NL;
3422                lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
3423                spin_unlock_irqrestore(&lockres->l_lock, flags);
3424                goto leave;
3425        }
3426
3427        /* if we're blocking an exclusive and we have *any* holders,
3428         * then requeue. */
3429        if ((lockres->l_blocking == DLM_LOCK_EX)
3430            && (lockres->l_ex_holders || lockres->l_ro_holders)) {
3431                mlog(ML_BASTS, "lockres %s, ReQ: EX/PR Holders %u,%u\n",
3432                     lockres->l_name, lockres->l_ex_holders,
3433                     lockres->l_ro_holders);
3434                goto leave_requeue;
3435        }
3436
3437        /* If it's a PR we're blocking, then only
3438         * requeue if we've got any EX holders */
3439        if (lockres->l_blocking == DLM_LOCK_PR &&
3440            lockres->l_ex_holders) {
3441                mlog(ML_BASTS, "lockres %s, ReQ: EX Holders %u\n",
3442                     lockres->l_name, lockres->l_ex_holders);
3443                goto leave_requeue;
3444        }
3445
3446        /*
3447         * Can we get a lock in this state if the holder counts are
3448         * zero? The meta data unblock code used to check this.
3449         */
3450        if ((lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
3451            && (lockres->l_flags & OCFS2_LOCK_REFRESHING)) {
3452                mlog(ML_BASTS, "lockres %s, ReQ: Lock Refreshing\n",
3453                     lockres->l_name);
3454                goto leave_requeue;
3455        }
3456
3457        new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking);
3458
3459        if (lockres->l_ops->check_downconvert
3460            && !lockres->l_ops->check_downconvert(lockres, new_level)) {
3461                mlog(ML_BASTS, "lockres %s, ReQ: Checkpointing\n",
3462                     lockres->l_name);
3463                goto leave_requeue;
3464        }
3465
3466        /* If we get here, then we know that there are no more
3467         * incompatible holders (and anyone asking for an incompatible
3468         * lock is blocked). We can now downconvert the lock */
3469        if (!lockres->l_ops->downconvert_worker)
3470                goto downconvert;
3471
3472        /* Some lockres types want to do a bit of work before
3473         * downconverting a lock. Allow that here. The worker function
3474         * may sleep, so we save off a copy of what we're blocking as
3475         * it may change while we're not holding the spin lock. */
3476        blocking = lockres->l_blocking;
3477        level = lockres->l_level;
3478        spin_unlock_irqrestore(&lockres->l_lock, flags);
3479
3480        ctl->unblock_action = lockres->l_ops->downconvert_worker(lockres, blocking);
3481
3482        if (ctl->unblock_action == UNBLOCK_STOP_POST) {
3483                mlog(ML_BASTS, "lockres %s, UNBLOCK_STOP_POST\n",
3484                     lockres->l_name);
3485                goto leave;
3486        }
3487
3488        spin_lock_irqsave(&lockres->l_lock, flags);
3489        if ((blocking != lockres->l_blocking) || (level != lockres->l_level)) {
3490                /* If this changed underneath us, then we can't drop
3491                 * it just yet. */
3492                mlog(ML_BASTS, "lockres %s, block=%d:%d, level=%d:%d, "
3493                     "Recheck\n", lockres->l_name, blocking,
3494                     lockres->l_blocking, level, lockres->l_level);
3495                goto recheck;
3496        }
3497
3498downconvert:
3499        ctl->requeue = 0;
3500
3501        if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
3502                if (lockres->l_level == DLM_LOCK_EX)
3503                        set_lvb = 1;
3504
3505                /*
3506                 * We only set the lvb if the lock has been fully
3507                 * refreshed - otherwise we risk setting stale
3508                 * data. Otherwise, there's no need to actually clear
3509                 * out the lvb here as it's value is still valid.
3510                 */
3511                if (set_lvb && !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
3512                        lockres->l_ops->set_lvb(lockres);
3513        }
3514
3515        gen = ocfs2_prepare_downconvert(lockres, new_level);
3516        spin_unlock_irqrestore(&lockres->l_lock, flags);
3517        ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb,
3518                                     gen);
3519
3520leave:
3521        if (ret)
3522                mlog_errno(ret);
3523        return ret;
3524
3525leave_requeue:
3526        spin_unlock_irqrestore(&lockres->l_lock, flags);
3527        ctl->requeue = 1;
3528
3529        return 0;
3530}
3531
3532static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
3533                                     int blocking)
3534{
3535        struct inode *inode;
3536        struct address_space *mapping;
3537        struct ocfs2_inode_info *oi;
3538
3539        inode = ocfs2_lock_res_inode(lockres);
3540        mapping = inode->i_mapping;
3541
3542        if (S_ISDIR(inode->i_mode)) {
3543                oi = OCFS2_I(inode);
3544                oi->ip_dir_lock_gen++;
3545                mlog(0, "generation: %u\n", oi->ip_dir_lock_gen);
3546                goto out;
3547        }
3548
3549        if (!S_ISREG(inode->i_mode))
3550                goto out;
3551
3552        /*
3553         * We need this before the filemap_fdatawrite() so that it can
3554         * transfer the dirty bit from the PTE to the
3555         * page. Unfortunately this means that even for EX->PR
3556         * downconverts, we'll lose our mappings and have to build
3557         * them up again.
3558         */
3559        unmap_mapping_range(mapping, 0, 0, 0);
3560
3561        if (filemap_fdatawrite(mapping)) {
3562                mlog(ML_ERROR, "Could not sync inode %llu for downconvert!",
3563                     (unsigned long long)OCFS2_I(inode)->ip_blkno);
3564        }
3565        sync_mapping_buffers(mapping);
3566        if (blocking == DLM_LOCK_EX) {
3567                truncate_inode_pages(mapping, 0);
3568        } else {
3569                /* We only need to wait on the I/O if we're not also
3570                 * truncating pages because truncate_inode_pages waits
3571                 * for us above. We don't truncate pages if we're
3572                 * blocking anything < EXMODE because we want to keep
3573                 * them around in that case. */
3574                filemap_fdatawait(mapping);
3575        }
3576
3577out:
3578        return UNBLOCK_CONTINUE;
3579}
3580
3581static int ocfs2_ci_checkpointed(struct ocfs2_caching_info *ci,
3582                                 struct ocfs2_lock_res *lockres,
3583                                 int new_level)
3584{
3585        int checkpointed = ocfs2_ci_fully_checkpointed(ci);
3586
3587        BUG_ON(new_level != DLM_LOCK_NL && new_level != DLM_LOCK_PR);
3588        BUG_ON(lockres->l_level != DLM_LOCK_EX && !checkpointed);
3589
3590        if (checkpointed)
3591                return 1;
3592
3593        ocfs2_start_checkpoint(OCFS2_SB(ocfs2_metadata_cache_get_super(ci)));
3594        return 0;
3595}
3596
3597static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
3598                                        int new_level)
3599{
3600        struct inode *inode = ocfs2_lock_res_inode(lockres);
3601
3602        return ocfs2_ci_checkpointed(INODE_CACHE(inode), lockres, new_level);
3603}
3604
3605static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres)
3606{
3607        struct inode *inode = ocfs2_lock_res_inode(lockres);
3608
3609        __ocfs2_stuff_meta_lvb(inode);
3610}
3611
3612/*
3613 * Does the final reference drop on our dentry lock. Right now this
3614 * happens in the downconvert thread, but we could choose to simplify the
3615 * dlmglue API and push these off to the ocfs2_wq in the future.
3616 */
3617static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
3618                                     struct ocfs2_lock_res *lockres)
3619{
3620        struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
3621        ocfs2_dentry_lock_put(osb, dl);
3622}
3623
3624/*
3625 * d_delete() matching dentries before the lock downconvert.
3626 *
3627 * At this point, any process waiting to destroy the
3628 * dentry_lock due to last ref count is stopped by the
3629 * OCFS2_LOCK_QUEUED flag.
3630 *
3631 * We have two potential problems
3632 *
3633 * 1) If we do the last reference drop on our dentry_lock (via dput)
3634 *    we'll wind up in ocfs2_release_dentry_lock(), waiting on
3635 *    the downconvert to finish. Instead we take an elevated
3636 *    reference and push the drop until after we've completed our
3637 *    unblock processing.
3638 *
3639 * 2) There might be another process with a final reference,
3640 *    waiting on us to finish processing. If this is the case, we
3641 *    detect it and exit out - there's no more dentries anyway.
3642 */
3643static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
3644                                       int blocking)
3645{
3646        struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
3647        struct ocfs2_inode_info *oi = OCFS2_I(dl->dl_inode);
3648        struct dentry *dentry;
3649        unsigned long flags;
3650        int extra_ref = 0;
3651
3652        /*
3653         * This node is blocking another node from getting a read
3654         * lock. This happens when we've renamed within a
3655         * directory. We've forced the other nodes to d_delete(), but
3656         * we never actually dropped our lock because it's still
3657         * valid. The downconvert code will retain a PR for this node,
3658         * so there's no further work to do.
3659         */
3660        if (blocking == DLM_LOCK_PR)
3661                return UNBLOCK_CONTINUE;
3662
3663        /*
3664         * Mark this inode as potentially orphaned. The code in
3665         * ocfs2_delete_inode() will figure out whether it actually
3666         * needs to be freed or not.
3667         */
3668        spin_lock(&oi->ip_lock);
3669        oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
3670        spin_unlock(&oi->ip_lock);
3671
3672        /*
3673         * Yuck. We need to make sure however that the check of
3674         * OCFS2_LOCK_FREEING and the extra reference are atomic with
3675         * respect to a reference decrement or the setting of that
3676         * flag.
3677         */
3678        spin_lock_irqsave(&lockres->l_lock, flags);
3679        spin_lock(&dentry_attach_lock);
3680        if (!(lockres->l_flags & OCFS2_LOCK_FREEING)
3681            && dl->dl_count) {
3682                dl->dl_count++;
3683                extra_ref = 1;
3684        }
3685        spin_unlock(&dentry_attach_lock);
3686        spin_unlock_irqrestore(&lockres->l_lock, flags);
3687
3688        mlog(0, "extra_ref = %d\n", extra_ref);
3689
3690        /*
3691         * We have a process waiting on us in ocfs2_dentry_iput(),
3692         * which means we can't have any more outstanding
3693         * aliases. There's no need to do any more work.
3694         */
3695        if (!extra_ref)
3696                return UNBLOCK_CONTINUE;
3697
3698        spin_lock(&dentry_attach_lock);
3699        while (1) {
3700                dentry = ocfs2_find_local_alias(dl->dl_inode,
3701                                                dl->dl_parent_blkno, 1);
3702                if (!dentry)
3703                        break;
3704                spin_unlock(&dentry_attach_lock);
3705
3706                mlog(0, "d_delete(%.*s);\n", dentry->d_name.len,
3707                     dentry->d_name.name);
3708
3709                /*
3710                 * The following dcache calls may do an
3711                 * iput(). Normally we don't want that from the
3712                 * downconverting thread, but in this case it's ok
3713                 * because the requesting node already has an
3714                 * exclusive lock on the inode, so it can't be queued
3715                 * for a downconvert.
3716                 */
3717                d_delete(dentry);
3718                dput(dentry);
3719
3720                spin_lock(&dentry_attach_lock);
3721        }
3722        spin_unlock(&dentry_attach_lock);
3723
3724        /*
3725         * If we are the last holder of this dentry lock, there is no
3726         * reason to downconvert so skip straight to the unlock.
3727         */
3728        if (dl->dl_count == 1)
3729                return UNBLOCK_STOP_POST;
3730
3731        return UNBLOCK_CONTINUE_POST;
3732}
3733
3734static int ocfs2_check_refcount_downconvert(struct ocfs2_lock_res *lockres,
3735                                            int new_level)
3736{
3737        struct ocfs2_refcount_tree *tree =
3738                                ocfs2_lock_res_refcount_tree(lockres);
3739
3740        return ocfs2_ci_checkpointed(&tree->rf_ci, lockres, new_level);
3741}
3742
3743static int ocfs2_refcount_convert_worker(struct ocfs2_lock_res *lockres,
3744                                         int blocking)
3745{
3746        struct ocfs2_refcount_tree *tree =
3747                                ocfs2_lock_res_refcount_tree(lockres);
3748
3749        ocfs2_metadata_cache_purge(&tree->rf_ci);
3750
3751        return UNBLOCK_CONTINUE;
3752}
3753
3754static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres)
3755{
3756        struct ocfs2_qinfo_lvb *lvb;
3757        struct ocfs2_mem_dqinfo *oinfo = ocfs2_lock_res_qinfo(lockres);
3758        struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb,
3759                                            oinfo->dqi_gi.dqi_type);
3760
3761        lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
3762        lvb->lvb_version = OCFS2_QINFO_LVB_VERSION;
3763        lvb->lvb_bgrace = cpu_to_be32(info->dqi_bgrace);
3764        lvb->lvb_igrace = cpu_to_be32(info->dqi_igrace);
3765        lvb->lvb_syncms = cpu_to_be32(oinfo->dqi_syncms);
3766        lvb->lvb_blocks = cpu_to_be32(oinfo->dqi_gi.dqi_blocks);
3767        lvb->lvb_free_blk = cpu_to_be32(oinfo->dqi_gi.dqi_free_blk);
3768        lvb->lvb_free_entry = cpu_to_be32(oinfo->dqi_gi.dqi_free_entry);
3769}
3770
3771void ocfs2_qinfo_unlock(struct ocfs2_mem_dqinfo *oinfo, int ex)
3772{
3773        struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
3774        struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb);
3775        int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
3776
3777        if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb))
3778                ocfs2_cluster_unlock(osb, lockres, level);
3779}
3780
3781static int ocfs2_refresh_qinfo(struct ocfs2_mem_dqinfo *oinfo)
3782{
3783        struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb,
3784                                            oinfo->dqi_gi.dqi_type);
3785        struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
3786        struct ocfs2_qinfo_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
3787        struct buffer_head *bh = NULL;
3788        struct ocfs2_global_disk_dqinfo *gdinfo;
3789        int status = 0;
3790
3791        if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) &&
3792            lvb->lvb_version == OCFS2_QINFO_LVB_VERSION) {
3793                info->dqi_bgrace = be32_to_cpu(lvb->lvb_bgrace);
3794                info->dqi_igrace = be32_to_cpu(lvb->lvb_igrace);
3795                oinfo->dqi_syncms = be32_to_cpu(lvb->lvb_syncms);
3796                oinfo->dqi_gi.dqi_blocks = be32_to_cpu(lvb->lvb_blocks);
3797                oinfo->dqi_gi.dqi_free_blk = be32_to_cpu(lvb->lvb_free_blk);
3798                oinfo->dqi_gi.dqi_free_entry =
3799                                        be32_to_cpu(lvb->lvb_free_entry);
3800        } else {
3801                status = ocfs2_read_quota_phys_block(oinfo->dqi_gqinode,
3802                                                     oinfo->dqi_giblk, &bh);
3803                if (status) {
3804                        mlog_errno(status);
3805                        goto bail;
3806                }
3807                gdinfo = (struct ocfs2_global_disk_dqinfo *)
3808                                        (bh->b_data + OCFS2_GLOBAL_INFO_OFF);
3809                info->dqi_bgrace = le32_to_cpu(gdinfo->dqi_bgrace);
3810                info->dqi_igrace = le32_to_cpu(gdinfo->dqi_igrace);
3811                oinfo->dqi_syncms = le32_to_cpu(gdinfo->dqi_syncms);
3812                oinfo->dqi_gi.dqi_blocks = le32_to_cpu(gdinfo->dqi_blocks);
3813                oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(gdinfo->dqi_free_blk);
3814                oinfo->dqi_gi.dqi_free_entry =
3815                                        le32_to_cpu(gdinfo->dqi_free_entry);
3816                brelse(bh);
3817                ocfs2_track_lock_refresh(lockres);
3818        }
3819
3820bail:
3821        return status;
3822}
3823
3824/* Lock quota info, this function expects at least shared lock on the quota file
3825 * so that we can safely refresh quota info from disk. */
3826int ocfs2_qinfo_lock(struct ocfs2_mem_dqinfo *oinfo, int ex)
3827{
3828        struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
3829        struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb);
3830        int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
3831        int status = 0;
3832
3833        /* On RO devices, locking really isn't needed... */
3834        if (ocfs2_is_hard_readonly(osb)) {
3835                if (ex)
3836                        status = -EROFS;
3837                goto bail;
3838        }
3839        if (ocfs2_mount_local(osb))
3840                goto bail;
3841
3842        status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
3843        if (status < 0) {
3844                mlog_errno(status);
3845                goto bail;
3846        }
3847        if (!ocfs2_should_refresh_lock_res(lockres))
3848                goto bail;
3849        /* OK, we have the lock but we need to refresh the quota info */
3850        status = ocfs2_refresh_qinfo(oinfo);
3851        if (status)
3852                ocfs2_qinfo_unlock(oinfo, ex);
3853        ocfs2_complete_lock_res_refresh(lockres, status);
3854bail:
3855        return status;
3856}
3857
3858int ocfs2_refcount_lock(struct ocfs2_refcount_tree *ref_tree, int ex)
3859{
3860        int status;
3861        int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
3862        struct ocfs2_lock_res *lockres = &ref_tree->rf_lockres;
3863        struct ocfs2_super *osb = lockres->l_priv;
3864
3865
3866        if (ocfs2_is_hard_readonly(osb))
3867                return -EROFS;
3868
3869        if (ocfs2_mount_local(osb))
3870                return 0;
3871
3872        status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
3873        if (status < 0)
3874                mlog_errno(status);
3875
3876        return status;
3877}
3878
3879void ocfs2_refcount_unlock(struct ocfs2_refcount_tree *ref_tree, int ex)
3880{
3881        int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
3882        struct ocfs2_lock_res *lockres = &ref_tree->rf_lockres;
3883        struct ocfs2_super *osb = lockres->l_priv;
3884
3885        if (!ocfs2_mount_local(osb))
3886                ocfs2_cluster_unlock(osb, lockres, level);
3887}
3888
3889static void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
3890                                       struct ocfs2_lock_res *lockres)
3891{
3892        int status;
3893        struct ocfs2_unblock_ctl ctl = {0, 0,};
3894        unsigned long flags;
3895
3896        /* Our reference to the lockres in this function can be
3897         * considered valid until we remove the OCFS2_LOCK_QUEUED
3898         * flag. */
3899
3900        BUG_ON(!lockres);
3901        BUG_ON(!lockres->l_ops);
3902
3903        mlog(ML_BASTS, "lockres %s blocked\n", lockres->l_name);
3904
3905        /* Detect whether a lock has been marked as going away while
3906         * the downconvert thread was processing other things. A lock can
3907         * still be marked with OCFS2_LOCK_FREEING after this check,
3908         * but short circuiting here will still save us some
3909         * performance. */
3910        spin_lock_irqsave(&lockres->l_lock, flags);
3911        if (lockres->l_flags & OCFS2_LOCK_FREEING)
3912                goto unqueue;
3913        spin_unlock_irqrestore(&lockres->l_lock, flags);
3914
3915        status = ocfs2_unblock_lock(osb, lockres, &ctl);
3916        if (status < 0)
3917                mlog_errno(status);
3918
3919        spin_lock_irqsave(&lockres->l_lock, flags);
3920unqueue:
3921        if (lockres->l_flags & OCFS2_LOCK_FREEING || !ctl.requeue) {
3922                lockres_clear_flags(lockres, OCFS2_LOCK_QUEUED);
3923        } else
3924                ocfs2_schedule_blocked_lock(osb, lockres);
3925
3926        mlog(ML_BASTS, "lockres %s, requeue = %s.\n", lockres->l_name,
3927             ctl.requeue ? "yes" : "no");
3928        spin_unlock_irqrestore(&lockres->l_lock, flags);
3929
3930        if (ctl.unblock_action != UNBLOCK_CONTINUE
3931            && lockres->l_ops->post_unlock)
3932                lockres->l_ops->post_unlock(osb, lockres);
3933}
3934
3935static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
3936                                        struct ocfs2_lock_res *lockres)
3937{
3938        unsigned long flags;
3939
3940        assert_spin_locked(&lockres->l_lock);
3941
3942        if (lockres->l_flags & OCFS2_LOCK_FREEING) {
3943                /* Do not schedule a lock for downconvert when it's on
3944                 * the way to destruction - any nodes wanting access
3945                 * to the resource will get it soon. */
3946                mlog(ML_BASTS, "lockres %s won't be scheduled: flags 0x%lx\n",
3947                     lockres->l_name, lockres->l_flags);
3948                return;
3949        }
3950
3951        lockres_or_flags(lockres, OCFS2_LOCK_QUEUED);
3952
3953        spin_lock_irqsave(&osb->dc_task_lock, flags);
3954        if (list_empty(&lockres->l_blocked_list)) {
3955                list_add_tail(&lockres->l_blocked_list,
3956                              &osb->blocked_lock_list);
3957                osb->blocked_lock_count++;
3958        }
3959        spin_unlock_irqrestore(&osb->dc_task_lock, flags);
3960}
3961
3962static void ocfs2_downconvert_thread_do_work(struct ocfs2_super *osb)
3963{
3964        unsigned long processed;
3965        unsigned long flags;
3966        struct ocfs2_lock_res *lockres;
3967
3968        spin_lock_irqsave(&osb->dc_task_lock, flags);
3969        /* grab this early so we know to try again if a state change and
3970         * wake happens part-way through our work  */
3971        osb->dc_work_sequence = osb->dc_wake_sequence;
3972
3973        processed = osb->blocked_lock_count;
3974        while (processed) {
3975                BUG_ON(list_empty(&osb->blocked_lock_list));
3976
3977                lockres = list_entry(osb->blocked_lock_list.next,
3978                                     struct ocfs2_lock_res, l_blocked_list);
3979                list_del_init(&lockres->l_blocked_list);
3980                osb->blocked_lock_count--;
3981                spin_unlock_irqrestore(&osb->dc_task_lock, flags);
3982
3983                BUG_ON(!processed);
3984                processed--;
3985
3986                ocfs2_process_blocked_lock(osb, lockres);
3987
3988                spin_lock_irqsave(&osb->dc_task_lock, flags);
3989        }
3990        spin_unlock_irqrestore(&osb->dc_task_lock, flags);
3991}
3992
3993static int ocfs2_downconvert_thread_lists_empty(struct ocfs2_super *osb)
3994{
3995        int empty = 0;
3996        unsigned long flags;
3997
3998        spin_lock_irqsave(&osb->dc_task_lock, flags);
3999        if (list_empty(&osb->blocked_lock_list))
4000                empty = 1;
4001
4002        spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4003        return empty;
4004}
4005
4006static int ocfs2_downconvert_thread_should_wake(struct ocfs2_super *osb)
4007{
4008        int should_wake = 0;
4009        unsigned long flags;
4010
4011        spin_lock_irqsave(&osb->dc_task_lock, flags);
4012        if (osb->dc_work_sequence != osb->dc_wake_sequence)
4013                should_wake = 1;
4014        spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4015
4016        return should_wake;
4017}
4018
4019static int ocfs2_downconvert_thread(void *arg)
4020{
4021        int status = 0;
4022        struct ocfs2_super *osb = arg;
4023
4024        /* only quit once we've been asked to stop and there is no more
4025         * work available */
4026        while (!(kthread_should_stop() &&
4027                ocfs2_downconvert_thread_lists_empty(osb))) {
4028
4029                wait_event_interruptible(osb->dc_event,
4030                                         ocfs2_downconvert_thread_should_wake(osb) ||
4031                                         kthread_should_stop());
4032
4033                mlog(0, "downconvert_thread: awoken\n");
4034
4035                ocfs2_downconvert_thread_do_work(osb);
4036        }
4037
4038        osb->dc_task = NULL;
4039        return status;
4040}
4041
4042void ocfs2_wake_downconvert_thread(struct ocfs2_super *osb)
4043{
4044        unsigned long flags;
4045
4046        spin_lock_irqsave(&osb->dc_task_lock, flags);
4047        /* make sure the voting thread gets a swipe at whatever changes
4048         * the caller may have made to the voting state */
4049        osb->dc_wake_sequence++;
4050        spin_unlock_irqrestore(&osb->dc_task_lock, flags);
4051        wake_up(&osb->dc_event);
4052}
4053