fs/gfs2/glock.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
   4  *
   5  * This copyrighted material is made available to anyone wishing to use,
   6  * modify, copy, or redistribute it subject to the terms and conditions
   7  * of the GNU General Public License version 2.
   8  */
   9
  10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  11
  12 #include <linux/sched.h>
  13 #include <linux/slab.h>
  14 #include <linux/spinlock.h>
  15 #include <linux/buffer_head.h>
  16 #include <linux/delay.h>
  17 #include <linux/sort.h>
  18 #include <linux/jhash.h>
  19 #include <linux/kallsyms.h>
  20 #include <linux/gfs2_ondisk.h>
  21 #include <linux/list.h>
  22 #include <linux/wait.h>
  23 #include <linux/module.h>
  24 #include <linux/uaccess.h>
  25 #include <linux/seq_file.h>
  26 #include <linux/debugfs.h>
  27 #include <linux/kthread.h>
  28 #include <linux/freezer.h>
  29 #include <linux/workqueue.h>
  30 #include <linux/jiffies.h>
  31 #include <linux/rcupdate.h>
  32 #include <linux/rculist_bl.h>
  33 #include <linux/bit_spinlock.h>
  34 #include <linux/percpu.h>
  35 #include <linux/list_sort.h>
  36 #include <linux/lockref.h>
  37 #include <linux/rhashtable.h>
  38
  39 #include "gfs2.h"
  40 #include "incore.h"
  41 #include "glock.h"
  42 #include "glops.h"
  43 #include "inode.h"
  44 #include "lops.h"
  45 #include "meta_io.h"
  46 #include "quota.h"
  47 #include "super.h"
  48 #include "util.h"
  49 #include "bmap.h"
  50 #define CREATE_TRACE_POINTS
  51 #include "trace_gfs2.h"
  52
  53 struct gfs2_glock_iter {
  54         struct gfs2_sbd *sdp;           /* incore superblock           */
  55         struct rhashtable_iter hti;     /* rhashtable iterator         */
  56         struct gfs2_glock *gl;          /* current glock struct        */
  57         loff_t last_pos;                /* last position               */
  58 };
  59
  60 typedef void (*glock_examiner) (struct gfs2_glock * gl);
  61
  62 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
  63
  64 static struct dentry *gfs2_root;
  65 static struct workqueue_struct *glock_workqueue;
  66 struct workqueue_struct *gfs2_delete_workqueue;
  67 static LIST_HEAD(lru_list);
  68 static atomic_t lru_count = ATOMIC_INIT(0);
  69 static DEFINE_SPINLOCK(lru_lock);
  70
  71 #define GFS2_GL_HASH_SHIFT      15
  72 #define GFS2_GL_HASH_SIZE       BIT(GFS2_GL_HASH_SHIFT)
  73
  74 static struct rhashtable_params ht_parms = {
  75         .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4,
  76         .key_len = offsetofend(struct lm_lockname, ln_type),
  77         .key_offset = offsetof(struct gfs2_glock, gl_name),
  78         .head_offset = offsetof(struct gfs2_glock, gl_node),
  79 };
  80
  81 static struct rhashtable gl_hash_table;
  82
  83 void gfs2_glock_free(struct gfs2_glock *gl)
  84 {
  85         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
  86
  87         if (gl->gl_ops->go_flags & GLOF_ASPACE) {
  88                 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
  89         } else {
  90                 kfree(gl->gl_lksb.sb_lvbptr);
  91                 kmem_cache_free(gfs2_glock_cachep, gl);
  92         }
  93         if (atomic_dec_and_test(&sdp->sd_glock_disposal))
  94                 wake_up(&sdp->sd_glock_wait);
  95 }
  96
  97 /**
  98  * gfs2_glock_hold() - increment reference count on glock
  99  * @gl: The glock to hold
 100  *
 101  */
 102
 103 static void gfs2_glock_hold(struct gfs2_glock *gl)
 104 {
 105         GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
 106         lockref_get(&gl->gl_lockref);
 107 }
 108
 109 /**
 110  * demote_ok - Check to see if it's ok to unlock a glock
 111  * @gl: the glock
 112  *
 113  * Returns: 1 if it's ok
 114  */
 115
 116 static int demote_ok(const struct gfs2_glock *gl)
 117 {
 118         const struct gfs2_glock_operations *glops = gl->gl_ops;
 119
 120         if (gl->gl_state == LM_ST_UNLOCKED)
 121                 return 0;
 122         if (!list_empty(&gl->gl_holders))
 123                 return 0;
 124         if (glops->go_demote_ok)
 125                 return glops->go_demote_ok(gl);
 126         return 1;
 127 }
 128
 129
 130 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
 131 {
 132         spin_lock(&lru_lock);
 133
 134         if (!list_empty(&gl->gl_lru))
 135                 list_del_init(&gl->gl_lru);
 136         else
 137                 atomic_inc(&lru_count);
 138
 139         list_add_tail(&gl->gl_lru, &lru_list);
 140         set_bit(GLF_LRU, &gl->gl_flags);
 141         spin_unlock(&lru_lock);
 142 }
 143
 144 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
 145 {
 146         spin_lock(&lru_lock);
 147         if (!list_empty(&gl->gl_lru)) {
 148                 list_del_init(&gl->gl_lru);
 149                 atomic_dec(&lru_count);
 150                 clear_bit(GLF_LRU, &gl->gl_flags);
 151         }
 152         spin_unlock(&lru_lock);
 153 }
 154
 155 /**
 156  * gfs2_glock_put() - Decrement reference count on glock
 157  * @gl: The glock to put
 158  *
 159  */
 160
 161 void gfs2_glock_put(struct gfs2_glock *gl)
 162 {
 163         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 164         struct address_space *mapping = gfs2_glock2aspace(gl);
 165
 166         if (lockref_put_or_lock(&gl->gl_lockref))
 167                 return;
 168
 169         lockref_mark_dead(&gl->gl_lockref);
 170
 171         gfs2_glock_remove_from_lru(gl);
 172         spin_unlock(&gl->gl_lockref.lock);
 173         rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
 174         GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
 175         GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
 176         trace_gfs2_glock_put(gl);
 177         sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
 178 }
 179
 180 /**
 181  * may_grant - check if its ok to grant a new lock
 182  * @gl: The glock
 183  * @gh: The lock request which we wish to grant
 184  *
 185  * Returns: true if its ok to grant the lock
 186  */
 187
 188 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
 189 {
 190         const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
 191         if ((gh->gh_state == LM_ST_EXCLUSIVE ||
 192              gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
 193                 return 0;
 194         if (gl->gl_state == gh->gh_state)
 195                 return 1;
 196         if (gh->gh_flags & GL_EXACT)
 197                 return 0;
 198         if (gl->gl_state == LM_ST_EXCLUSIVE) {
 199                 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
 200                         return 1;
 201                 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
 202                         return 1;
 203         }
 204         if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
 205                 return 1;
 206         return 0;
 207 }
 208
 209 static void gfs2_holder_wake(struct gfs2_holder *gh)
 210 {
 211         clear_bit(HIF_WAIT, &gh->gh_iflags);
 212         smp_mb__after_atomic();
 213         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
 214 }
 215
 216 /**
 217  * do_error - Something unexpected has happened during a lock request
 218  *
 219  */
 220
 221 static void do_error(struct gfs2_glock *gl, const int ret)
 222 {
 223         struct gfs2_holder *gh, *tmp;
 224
 225         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
 226                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 227                         continue;
 228                 if (ret & LM_OUT_ERROR)
 229                         gh->gh_error = -EIO;
 230                 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
 231                         gh->gh_error = GLR_TRYFAILED;
 232                 else
 233                         continue;
 234                 list_del_init(&gh->gh_list);
 235                 trace_gfs2_glock_queue(gh, 0);
 236                 gfs2_holder_wake(gh);
 237         }
 238 }
 239
 240 /**
 241  * do_promote - promote as many requests as possible on the current queue
 242  * @gl: The glock
 243  *
 244  * Returns: 1 if there is a blocked holder at the head of the list, or 2
 245  *          if a type specific operation is underway.
 246  */
 247
 248 static int do_promote(struct gfs2_glock *gl)
 249 __releases(&gl->gl_lockref.lock)
 250 __acquires(&gl->gl_lockref.lock)
 251 {
 252         const struct gfs2_glock_operations *glops = gl->gl_ops;
 253         struct gfs2_holder *gh, *tmp;
 254         int ret;
 255
 256 restart:
 257         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
 258                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 259                         continue;
 260                 if (may_grant(gl, gh)) {
 261                         if (gh->gh_list.prev == &gl->gl_holders &&
 262                             glops->go_lock) {
 263                                 spin_unlock(&gl->gl_lockref.lock);
 264                                 /* FIXME: eliminate this eventually */
 265                                 ret = glops->go_lock(gh);
 266                                 spin_lock(&gl->gl_lockref.lock);
 267                                 if (ret) {
 268                                         if (ret == 1)
 269                                                 return 2;
 270                                         gh->gh_error = ret;
 271                                         list_del_init(&gh->gh_list);
 272                                         trace_gfs2_glock_queue(gh, 0);
 273                                         gfs2_holder_wake(gh);
 274                                         goto restart;
 275                                 }
 276                                 set_bit(HIF_HOLDER, &gh->gh_iflags);
 277                                 trace_gfs2_promote(gh, 1);
 278                                 gfs2_holder_wake(gh);
 279                                 goto restart;
 280                         }
 281                         set_bit(HIF_HOLDER, &gh->gh_iflags);
 282                         trace_gfs2_promote(gh, 0);
 283                         gfs2_holder_wake(gh);
 284                         continue;
 285                 }
 286                 if (gh->gh_list.prev == &gl->gl_holders)
 287                         return 1;
 288                 do_error(gl, 0);
 289                 break;
 290         }
 291         return 0;
 292 }
 293
 294 /**
 295  * find_first_waiter - find the first gh that's waiting for the glock
 296  * @gl: the glock
 297  */
 298
 299 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
 300 {
 301         struct gfs2_holder *gh;
 302
 303         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
 304                 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
 305                         return gh;
 306         }
 307         return NULL;
 308 }
 309
 310 /**
 311  * state_change - record that the glock is now in a different state
 312  * @gl: the glock
 313  * @new_state the new state
 314  *
 315  */
 316
 317 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
 318 {
 319         int held1, held2;
 320
 321         held1 = (gl->gl_state != LM_ST_UNLOCKED);
 322         held2 = (new_state != LM_ST_UNLOCKED);
 323
 324         if (held1 != held2) {
 325                 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
 326                 if (held2)
 327                         gl->gl_lockref.count++;
 328                 else
 329                         gl->gl_lockref.count--;
 330         }
 331         if (held1 && held2 && list_empty(&gl->gl_holders))
 332                 clear_bit(GLF_QUEUED, &gl->gl_flags);
 333
 334         if (new_state != gl->gl_target)
 335                 /* shorten our minimum hold time */
 336                 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
 337                                        GL_GLOCK_MIN_HOLD);
 338         gl->gl_state = new_state;
 339         gl->gl_tchange = jiffies;
 340 }
 341
 342 static void gfs2_demote_wake(struct gfs2_glock *gl)
 343 {
 344         gl->gl_demote_state = LM_ST_EXCLUSIVE;
 345         clear_bit(GLF_DEMOTE, &gl->gl_flags);
 346         smp_mb__after_atomic();
 347         wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
 348 }
 349
 350 /**
 351  * finish_xmote - The DLM has replied to one of our lock requests
 352  * @gl: The glock
 353  * @ret: The status from the DLM
 354  *
 355  */
 356
 357 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
 358 {
 359         const struct gfs2_glock_operations *glops = gl->gl_ops;
 360         struct gfs2_holder *gh;
 361         unsigned state = ret & LM_OUT_ST_MASK;
 362         int rv;
 363
 364         spin_lock(&gl->gl_lockref.lock);
 365         trace_gfs2_glock_state_change(gl, state);
 366         state_change(gl, state);
 367         gh = find_first_waiter(gl);
 368
 369         /* Demote to UN request arrived during demote to SH or DF */
 370         if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
 371             state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
 372                 gl->gl_target = LM_ST_UNLOCKED;
 373
 374         /* Check for state != intended state */
 375         if (unlikely(state != gl->gl_target)) {
 376                 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
 377                         /* move to back of queue and try next entry */
 378                         if (ret & LM_OUT_CANCELED) {
 379                                 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
 380                                         list_move_tail(&gh->gh_list, &gl->gl_holders);
 381                                 gh = find_first_waiter(gl);
 382                                 gl->gl_target = gh->gh_state;
 383                                 goto retry;
 384                         }
 385                         /* Some error or failed "try lock" - report it */
 386                         if ((ret & LM_OUT_ERROR) ||
 387                             (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
 388                                 gl->gl_target = gl->gl_state;
 389                                 do_error(gl, ret);
 390                                 goto out;
 391                         }
 392                 }
 393                 switch(state) {
 394                 /* Unlocked due to conversion deadlock, try again */
 395                 case LM_ST_UNLOCKED:
 396 retry:
 397                         do_xmote(gl, gh, gl->gl_target);
 398                         break;
 399                 /* Conversion fails, unlock and try again */
 400                 case LM_ST_SHARED:
 401                 case LM_ST_DEFERRED:
 402                         do_xmote(gl, gh, LM_ST_UNLOCKED);
 403                         break;
 404                 default: /* Everything else */
 405                         pr_err("wanted %u got %u\n", gl->gl_target, state);
 406                         GLOCK_BUG_ON(gl, 1);
 407                 }
 408                 spin_unlock(&gl->gl_lockref.lock);
 409                 return;
 410         }
 411
 412         /* Fast path - we got what we asked for */
 413         if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
 414                 gfs2_demote_wake(gl);
 415         if (state != LM_ST_UNLOCKED) {
 416                 if (glops->go_xmote_bh) {
 417                         spin_unlock(&gl->gl_lockref.lock);
 418                         rv = glops->go_xmote_bh(gl, gh);
 419                         spin_lock(&gl->gl_lockref.lock);
 420                         if (rv) {
 421                                 do_error(gl, rv);
 422                                 goto out;
 423                         }
 424                 }
 425                 rv = do_promote(gl);
 426                 if (rv == 2)
 427                         goto out_locked;
 428         }
 429 out:
 430         clear_bit(GLF_LOCK, &gl->gl_flags);
 431 out_locked:
 432         spin_unlock(&gl->gl_lockref.lock);
 433 }
 434
 435 /**
 436  * do_xmote - Calls the DLM to change the state of a lock
 437  * @gl: The lock state
 438  * @gh: The holder (only for promotes)
 439  * @target: The target lock state
 440  *
 441  */
 442
 443 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
 444 __releases(&gl->gl_lockref.lock)
 445 __acquires(&gl->gl_lockref.lock)
 446 {
 447         const struct gfs2_glock_operations *glops = gl->gl_ops;
 448         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 449         unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
 450         int ret;
 451
 452         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
 453             target != LM_ST_UNLOCKED)
 454                 return;
 455         lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
 456                       LM_FLAG_PRIORITY);
 457         GLOCK_BUG_ON(gl, gl->gl_state == target);
 458         GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
 459         if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
 460             glops->go_inval) {
 461                 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 462                 do_error(gl, 0); /* Fail queued try locks */
 463         }
 464         gl->gl_req = target;
 465         set_bit(GLF_BLOCKING, &gl->gl_flags);
 466         if ((gl->gl_req == LM_ST_UNLOCKED) ||
 467             (gl->gl_state == LM_ST_EXCLUSIVE) ||
 468             (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
 469                 clear_bit(GLF_BLOCKING, &gl->gl_flags);
 470         spin_unlock(&gl->gl_lockref.lock);
 471         if (glops->go_sync)
 472                 glops->go_sync(gl);
 473         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 474                 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
 475         clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 476
 477         gfs2_glock_hold(gl);
 478         if (sdp->sd_lockstruct.ls_ops->lm_lock) {
 479                 /* lock_dlm */
 480                 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
 481                 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
 482                     target == LM_ST_UNLOCKED &&
 483                     test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
 484                         finish_xmote(gl, target);
 485                         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 486                                 gfs2_glock_put(gl);
 487                 }
 488                 else if (ret) {
 489                         pr_err("lm_lock ret %d\n", ret);
 490                         GLOCK_BUG_ON(gl, !test_bit(SDF_SHUTDOWN,
 491                                                    &sdp->sd_flags));
 492                 }
 493         } else { /* lock_nolock */
 494                 finish_xmote(gl, target);
 495                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 496                         gfs2_glock_put(gl);
 497         }
 498
 499         spin_lock(&gl->gl_lockref.lock);
 500 }
 501
 502 /**
 503  * find_first_holder - find the first "holder" gh
 504  * @gl: the glock
 505  */
 506
 507 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
 508 {
 509         struct gfs2_holder *gh;
 510
 511         if (!list_empty(&gl->gl_holders)) {
 512                 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
 513                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 514                         return gh;
 515         }
 516         return NULL;
 517 }
 518
 519 /**
 520  * run_queue - do all outstanding tasks related to a glock
 521  * @gl: The glock in question
 522  * @nonblock: True if we must not block in run_queue
 523  *
 524  */
 525
 526 static void run_queue(struct gfs2_glock *gl, const int nonblock)
 527 __releases(&gl->gl_lockref.lock)
 528 __acquires(&gl->gl_lockref.lock)
 529 {
 530         struct gfs2_holder *gh = NULL;
 531         int ret;
 532
 533         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
 534                 return;
 535
 536         GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
 537
 538         if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
 539             gl->gl_demote_state != gl->gl_state) {
 540                 if (find_first_holder(gl))
 541                         goto out_unlock;
 542                 if (nonblock)
 543                         goto out_sched;
 544                 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
 545                 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
 546                 gl->gl_target = gl->gl_demote_state;
 547         } else {
 548                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
 549                         gfs2_demote_wake(gl);
 550                 ret = do_promote(gl);
 551                 if (ret == 0)
 552                         goto out_unlock;
 553                 if (ret == 2)
 554                         goto out;
 555                 gh = find_first_waiter(gl);
 556                 gl->gl_target = gh->gh_state;
 557                 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
 558                         do_error(gl, 0); /* Fail queued try locks */
 559         }
 560         do_xmote(gl, gh, gl->gl_target);
 561 out:
 562         return;
 563
 564 out_sched:
 565         clear_bit(GLF_LOCK, &gl->gl_flags);
 566         smp_mb__after_atomic();
 567         gl->gl_lockref.count++;
 568         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 569                 gl->gl_lockref.count--;
 570         return;
 571
 572 out_unlock:
 573         clear_bit(GLF_LOCK, &gl->gl_flags);
 574         smp_mb__after_atomic();
 575         return;
 576 }
 577
 578 static void delete_work_func(struct work_struct *work)
 579 {
 580         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
 581         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 582         struct inode *inode;
 583         u64 no_addr = gl->gl_name.ln_number;
 584
 585         /* If someone's using this glock to create a new dinode, the block must
 586            have been freed by another node, then re-used, in which case our
 587            iopen callback is too late after the fact. Ignore it. */
 588         if (test_bit(GLF_INODE_CREATING, &gl->gl_flags))
 589                 goto out;
 590
 591         inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
 592         if (inode && !IS_ERR(inode)) {
 593                 d_prune_aliases(inode);
 594                 iput(inode);
 595         }
 596 out:
 597         gfs2_glock_put(gl);
 598 }
 599
 600 static void glock_work_func(struct work_struct *work)
 601 {
 602         unsigned long delay = 0;
 603         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
 604         int drop_ref = 0;
 605
 606         if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
 607                 finish_xmote(gl, gl->gl_reply);
 608                 drop_ref = 1;
 609         }
 610         spin_lock(&gl->gl_lockref.lock);
 611         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
 612             gl->gl_state != LM_ST_UNLOCKED &&
 613             gl->gl_demote_state != LM_ST_EXCLUSIVE) {
 614                 unsigned long holdtime, now = jiffies;
 615
 616                 holdtime = gl->gl_tchange + gl->gl_hold_time;
 617                 if (time_before(now, holdtime))
 618                         delay = holdtime - now;
 619
 620                 if (!delay) {
 621                         clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
 622                         set_bit(GLF_DEMOTE, &gl->gl_flags);
 623                 }
 624         }
 625         run_queue(gl, 0);
 626         spin_unlock(&gl->gl_lockref.lock);
 627         if (!delay)
 628                 gfs2_glock_put(gl);
 629         else {
 630                 if (gl->gl_name.ln_type != LM_TYPE_INODE)
 631                         delay = 0;
 632                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
 633                         gfs2_glock_put(gl);
 634         }
 635         if (drop_ref)
 636                 gfs2_glock_put(gl);
 637 }
 638
 639 /**
 640  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
 641  * @sdp: The GFS2 superblock
 642  * @number: the lock number
 643  * @glops: The glock_operations to use
 644  * @create: If 0, don't create the glock if it doesn't exist
 645  * @glp: the glock is returned here
 646  *
 647  * This does not lock a glock, just finds/creates structures for one.
 648  *
 649  * Returns: errno
 650  */
 651
 652 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
 653                    const struct gfs2_glock_operations *glops, int create,
 654                    struct gfs2_glock **glp)
 655 {
 656         struct super_block *s = sdp->sd_vfs;
 657         struct lm_lockname name = { .ln_number = number,
 658                                     .ln_type = glops->go_type,
 659                                     .ln_sbd = sdp };
 660         struct gfs2_glock *gl, *tmp;
 661         struct address_space *mapping;
 662         struct kmem_cache *cachep;
 663         int ret = 0;
 664
 665         rcu_read_lock();
 666         gl = rhashtable_lookup_fast(&gl_hash_table, &name, ht_parms);
 667         if (gl && !lockref_get_not_dead(&gl->gl_lockref))
 668                 gl = NULL;
 669         rcu_read_unlock();
 670
 671         *glp = gl;
 672         if (gl)
 673                 return 0;
 674         if (!create)
 675                 return -ENOENT;
 676
 677         if (glops->go_flags & GLOF_ASPACE)
 678                 cachep = gfs2_glock_aspace_cachep;
 679         else
 680                 cachep = gfs2_glock_cachep;
 681         gl = kmem_cache_alloc(cachep, GFP_NOFS);
 682         if (!gl)
 683                 return -ENOMEM;
 684
 685         memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
 686
 687         if (glops->go_flags & GLOF_LVB) {
 688                 gl->gl_lksb.sb_lvbptr = kzalloc(GFS2_MIN_LVB_SIZE, GFP_NOFS);
 689                 if (!gl->gl_lksb.sb_lvbptr) {
 690                         kmem_cache_free(cachep, gl);
 691                         return -ENOMEM;
 692                 }
 693         }
 694
 695         atomic_inc(&sdp->sd_glock_disposal);
 696         gl->gl_node.next = NULL;
 697         gl->gl_flags = 0;
 698         gl->gl_name = name;
 699         gl->gl_lockref.count = 1;
 700         gl->gl_state = LM_ST_UNLOCKED;
 701         gl->gl_target = LM_ST_UNLOCKED;
 702         gl->gl_demote_state = LM_ST_EXCLUSIVE;
 703         gl->gl_ops = glops;
 704         gl->gl_dstamp = 0;
 705         preempt_disable();
 706         /* We use the global stats to estimate the initial per-glock stats */
 707         gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
 708         preempt_enable();
 709         gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
 710         gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
 711         gl->gl_tchange = jiffies;
 712         gl->gl_object = NULL;
 713         gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
 714         INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
 715         INIT_WORK(&gl->gl_delete, delete_work_func);
 716
 717         mapping = gfs2_glock2aspace(gl);
 718         if (mapping) {
 719                 mapping->a_ops = &gfs2_meta_aops;
 720                 mapping->host = s->s_bdev->bd_inode;
 721                 mapping->flags = 0;
 722                 mapping_set_gfp_mask(mapping, GFP_NOFS);
 723                 mapping->private_data = NULL;
 724                 mapping->writeback_index = 0;
 725         }
 726
 727 again:
 728         rcu_read_lock();
 729         tmp = rhashtable_lookup_get_insert_fast(&gl_hash_table, &gl->gl_node,
 730                                                 ht_parms);
 731         if (!tmp) {
 732                 *glp = gl;
 733                 goto out;
 734         }
 735         if (IS_ERR(tmp)) {
 736                 ret = PTR_ERR(tmp);
 737                 goto out_free;
 738         }
 739         if (lockref_get_not_dead(&tmp->gl_lockref)) {
 740                 *glp = tmp;
 741                 goto out_free;
 742         }
 743         rcu_read_unlock();
 744         cond_resched();
 745         goto again;
 746
 747 out_free:
 748         kfree(gl->gl_lksb.sb_lvbptr);
 749         kmem_cache_free(cachep, gl);
 750         atomic_dec(&sdp->sd_glock_disposal);
 751
 752 out:
 753         rcu_read_unlock();
 754         return ret;
 755 }
 756
 757 /**
 758  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
 759  * @gl: the glock
 760  * @state: the state we're requesting
 761  * @flags: the modifier flags
 762  * @gh: the holder structure
 763  *
 764  */
 765
 766 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
 767                       struct gfs2_holder *gh)
 768 {
 769         INIT_LIST_HEAD(&gh->gh_list);
 770         gh->gh_gl = gl;
 771         gh->gh_ip = _RET_IP_;
 772         gh->gh_owner_pid = get_pid(task_pid(current));
 773         gh->gh_state = state;
 774         gh->gh_flags = flags;
 775         gh->gh_error = 0;
 776         gh->gh_iflags = 0;
 777         gfs2_glock_hold(gl);
 778 }
 779
 780 /**
 781  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
 782  * @state: the state we're requesting
 783  * @flags: the modifier flags
 784  * @gh: the holder structure
 785  *
 786  * Don't mess with the glock.
 787  *
 788  */
 789
 790 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
 791 {
 792         gh->gh_state = state;
 793         gh->gh_flags = flags;
 794         gh->gh_iflags = 0;
 795         gh->gh_ip = _RET_IP_;
 796         put_pid(gh->gh_owner_pid);
 797         gh->gh_owner_pid = get_pid(task_pid(current));
 798 }
 799
 800 /**
 801  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
 802  * @gh: the holder structure
 803  *
 804  */
 805
 806 void gfs2_holder_uninit(struct gfs2_holder *gh)
 807 {
 808         put_pid(gh->gh_owner_pid);
 809         gfs2_glock_put(gh->gh_gl);
 810         gfs2_holder_mark_uninitialized(gh);
 811         gh->gh_ip = 0;
 812 }
 813
 814 /**
 815  * gfs2_glock_wait - wait on a glock acquisition
 816  * @gh: the glock holder
 817  *
 818  * Returns: 0 on success
 819  */
 820
 821 int gfs2_glock_wait(struct gfs2_holder *gh)
 822 {
 823         unsigned long time1 = jiffies;
 824
 825         might_sleep();
 826         wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
 827         if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
 828                 /* Lengthen the minimum hold time. */
 829                 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
 830                                               GL_GLOCK_HOLD_INCR,
 831                                               GL_GLOCK_MAX_HOLD);
 832         return gh->gh_error;
 833 }
 834
 835 /**
 836  * handle_callback - process a demote request
 837  * @gl: the glock
 838  * @state: the state the caller wants us to change to
 839  *
 840  * There are only two requests that we are going to see in actual
 841  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
 842  */
 843
 844 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
 845                             unsigned long delay, bool remote)
 846 {
 847         int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
 848
 849         set_bit(bit, &gl->gl_flags);
 850         if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
 851                 gl->gl_demote_state = state;
 852                 gl->gl_demote_time = jiffies;
 853         } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
 854                         gl->gl_demote_state != state) {
 855                 gl->gl_demote_state = LM_ST_UNLOCKED;
 856         }
 857         if (gl->gl_ops->go_callback)
 858                 gl->gl_ops->go_callback(gl, remote);
 859         trace_gfs2_demote_rq(gl, remote);
 860 }
 861
 862 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
 863 {
 864         struct va_format vaf;
 865         va_list args;
 866
 867         va_start(args, fmt);
 868
 869         if (seq) {
 870                 seq_vprintf(seq, fmt, args);
 871         } else {
 872                 vaf.fmt = fmt;
 873                 vaf.va = &args;
 874
 875                 pr_err("%pV", &vaf);
 876         }
 877
 878         va_end(args);
 879 }
 880
 881 /**
 882  * add_to_queue - Add a holder to the wait queue (but look for recursion)
 883  * @gh: the holder structure to add
 884  *
 885  * Eventually we should move the recursive locking trap to a
 886  * debugging option or something like that. This is the fast
 887  * path and needs to have the minimum number of distractions.
 888  *
 889  */
 890
 891 static inline void add_to_queue(struct gfs2_holder *gh)
 892 __releases(&gl->gl_lockref.lock)
 893 __acquires(&gl->gl_lockref.lock)
 894 {
 895         struct gfs2_glock *gl = gh->gh_gl;
 896         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 897         struct list_head *insert_pt = NULL;
 898         struct gfs2_holder *gh2;
 899         int try_futile = 0;
 900
 901         BUG_ON(gh->gh_owner_pid == NULL);
 902         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
 903                 BUG();
 904
 905         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
 906                 if (test_bit(GLF_LOCK, &gl->gl_flags))
 907                         try_futile = !may_grant(gl, gh);
 908                 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 909                         goto fail;
 910         }
 911
 912         list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
 913                 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
 914                     (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
 915                         goto trap_recursive;
 916                 if (try_futile &&
 917                     !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
 918 fail:
 919                         gh->gh_error = GLR_TRYFAILED;
 920                         gfs2_holder_wake(gh);
 921                         return;
 922                 }
 923                 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
 924                         continue;
 925                 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
 926                         insert_pt = &gh2->gh_list;
 927         }
 928         set_bit(GLF_QUEUED, &gl->gl_flags);
 929         trace_gfs2_glock_queue(gh, 1);
 930         gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
 931         gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
 932         if (likely(insert_pt == NULL)) {
 933                 list_add_tail(&gh->gh_list, &gl->gl_holders);
 934                 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
 935                         goto do_cancel;
 936                 return;
 937         }
 938         list_add_tail(&gh->gh_list, insert_pt);
 939 do_cancel:
 940         gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
 941         if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
 942                 spin_unlock(&gl->gl_lockref.lock);
 943                 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
 944                         sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
 945                 spin_lock(&gl->gl_lockref.lock);
 946         }
 947         return;
 948
 949 trap_recursive:
 950         pr_err("original: %pSR\n", (void *)gh2->gh_ip);
 951         pr_err("pid: %d\n", pid_nr(gh2->gh_owner_pid));
 952         pr_err("lock type: %d req lock state : %d\n",
 953                gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
 954         pr_err("new: %pSR\n", (void *)gh->gh_ip);
 955         pr_err("pid: %d\n", pid_nr(gh->gh_owner_pid));
 956         pr_err("lock type: %d req lock state : %d\n",
 957                gh->gh_gl->gl_name.ln_type, gh->gh_state);
 958         gfs2_dump_glock(NULL, gl);
 959         BUG();
 960 }
 961
 962 /**
 963  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
 964  * @gh: the holder structure
 965  *
 966  * if (gh->gh_flags & GL_ASYNC), this never returns an error
 967  *
 968  * Returns: 0, GLR_TRYFAILED, or errno on failure
 969  */
 970
 971 int gfs2_glock_nq(struct gfs2_holder *gh)
 972 {
 973         struct gfs2_glock *gl = gh->gh_gl;
 974         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 975         int error = 0;
 976
 977         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 978                 return -EIO;
 979
 980         if (test_bit(GLF_LRU, &gl->gl_flags))
 981                 gfs2_glock_remove_from_lru(gl);
 982
 983         spin_lock(&gl->gl_lockref.lock);
 984         add_to_queue(gh);
 985         if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
 986                      test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
 987                 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
 988                 gl->gl_lockref.count++;
 989                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 990                         gl->gl_lockref.count--;
 991         }
 992         run_queue(gl, 1);
 993         spin_unlock(&gl->gl_lockref.lock);
 994
 995         if (!(gh->gh_flags & GL_ASYNC))
 996                 error = gfs2_glock_wait(gh);
 997
 998         return error;
 999 }
1000
1001 /**
1002  * gfs2_glock_poll - poll to see if an async request has been completed
1003  * @gh: the holder
1004  *
1005  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1006  */
1007
1008 int gfs2_glock_poll(struct gfs2_holder *gh)
1009 {
1010         return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1011 }
1012
1013 /**
1014  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1015  * @gh: the glock holder
1016  *
1017  */
1018
1019 void gfs2_glock_dq(struct gfs2_holder *gh)
1020 {
1021         struct gfs2_glock *gl = gh->gh_gl;
1022         const struct gfs2_glock_operations *glops = gl->gl_ops;
1023         unsigned delay = 0;
1024         int fast_path = 0;
1025
1026         spin_lock(&gl->gl_lockref.lock);
1027         if (gh->gh_flags & GL_NOCACHE)
1028                 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1029
1030         list_del_init(&gh->gh_list);
1031         clear_bit(HIF_HOLDER, &gh->gh_iflags);
1032         if (find_first_holder(gl) == NULL) {
1033                 if (glops->go_unlock) {
1034                         GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1035                         spin_unlock(&gl->gl_lockref.lock);
1036                         glops->go_unlock(gh);
1037                         spin_lock(&gl->gl_lockref.lock);
1038                         clear_bit(GLF_LOCK, &gl->gl_flags);
1039                 }
1040                 if (list_empty(&gl->gl_holders) &&
1041                     !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1042                     !test_bit(GLF_DEMOTE, &gl->gl_flags))
1043                         fast_path = 1;
1044         }
1045         if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl) &&
1046             (glops->go_flags & GLOF_LRU))
1047                 gfs2_glock_add_to_lru(gl);
1048
1049         trace_gfs2_glock_queue(gh, 0);
1050         spin_unlock(&gl->gl_lockref.lock);
1051         if (likely(fast_path))
1052                 return;
1053
1054         gfs2_glock_hold(gl);
1055         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1056             !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1057             gl->gl_name.ln_type == LM_TYPE_INODE)
1058                 delay = gl->gl_hold_time;
1059         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1060                 gfs2_glock_put(gl);
1061 }
1062
1063 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1064 {
1065         struct gfs2_glock *gl = gh->gh_gl;
1066         gfs2_glock_dq(gh);
1067         might_sleep();
1068         wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1069 }
1070
1071 /**
1072  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1073  * @gh: the holder structure
1074  *
1075  */
1076
1077 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1078 {
1079         gfs2_glock_dq(gh);
1080         gfs2_holder_uninit(gh);
1081 }
1082
1083 /**
1084  * gfs2_glock_nq_num - acquire a glock based on lock number
1085  * @sdp: the filesystem
1086  * @number: the lock number
1087  * @glops: the glock operations for the type of glock
1088  * @state: the state to acquire the glock in
1089  * @flags: modifier flags for the acquisition
1090  * @gh: the struct gfs2_holder
1091  *
1092  * Returns: errno
1093  */
1094
1095 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1096                       const struct gfs2_glock_operations *glops,
1097                       unsigned int state, u16 flags, struct gfs2_holder *gh)
1098 {
1099         struct gfs2_glock *gl;
1100         int error;
1101
1102         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1103         if (!error) {
1104                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1105                 gfs2_glock_put(gl);
1106         }
1107
1108         return error;
1109 }
1110
1111 /**
1112  * glock_compare - Compare two struct gfs2_glock structures for sorting
1113  * @arg_a: the first structure
1114  * @arg_b: the second structure
1115  *
1116  */
1117
1118 static int glock_compare(const void *arg_a, const void *arg_b)
1119 {
1120         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1121         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1122         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1123         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1124
1125         if (a->ln_number > b->ln_number)
1126                 return 1;
1127         if (a->ln_number < b->ln_number)
1128                 return -1;
1129         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1130         return 0;
1131 }
1132
1133 /**
1134  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1135  * @num_gh: the number of structures
1136  * @ghs: an array of struct gfs2_holder structures
1137  *
1138  * Returns: 0 on success (all glocks acquired),
1139  *          errno on failure (no glocks acquired)
1140  */
1141
1142 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1143                      struct gfs2_holder **p)
1144 {
1145         unsigned int x;
1146         int error = 0;
1147
1148         for (x = 0; x < num_gh; x++)
1149                 p[x] = &ghs[x];
1150
1151         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1152
1153         for (x = 0; x < num_gh; x++) {
1154                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1155
1156                 error = gfs2_glock_nq(p[x]);
1157                 if (error) {
1158                         while (x--)
1159                                 gfs2_glock_dq(p[x]);
1160                         break;
1161                 }
1162         }
1163
1164         return error;
1165 }
1166
1167 /**
1168  * gfs2_glock_nq_m - acquire multiple glocks
1169  * @num_gh: the number of structures
1170  * @ghs: an array of struct gfs2_holder structures
1171  *
1172  *
1173  * Returns: 0 on success (all glocks acquired),
1174  *          errno on failure (no glocks acquired)
1175  */
1176
1177 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1178 {
1179         struct gfs2_holder *tmp[4];
1180         struct gfs2_holder **pph = tmp;
1181         int error = 0;
1182
1183         switch(num_gh) {
1184         case 0:
1185                 return 0;
1186         case 1:
1187                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1188                 return gfs2_glock_nq(ghs);
1189         default:
1190                 if (num_gh <= 4)
1191                         break;
1192                 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1193                 if (!pph)
1194                         return -ENOMEM;
1195         }
1196
1197         error = nq_m_sync(num_gh, ghs, pph);
1198
1199         if (pph != tmp)
1200                 kfree(pph);
1201
1202         return error;
1203 }
1204
1205 /**
1206  * gfs2_glock_dq_m - release multiple glocks
1207  * @num_gh: the number of structures
1208  * @ghs: an array of struct gfs2_holder structures
1209  *
1210  */
1211
1212 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1213 {
1214         while (num_gh--)
1215                 gfs2_glock_dq(&ghs[num_gh]);
1216 }
1217
1218 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1219 {
1220         unsigned long delay = 0;
1221         unsigned long holdtime;
1222         unsigned long now = jiffies;
1223
1224         gfs2_glock_hold(gl);
1225         holdtime = gl->gl_tchange + gl->gl_hold_time;
1226         if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1227             gl->gl_name.ln_type == LM_TYPE_INODE) {
1228                 if (time_before(now, holdtime))
1229                         delay = holdtime - now;
1230                 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1231                         delay = gl->gl_hold_time;
1232         }
1233
1234         spin_lock(&gl->gl_lockref.lock);
1235         handle_callback(gl, state, delay, true);
1236         spin_unlock(&gl->gl_lockref.lock);
1237         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1238                 gfs2_glock_put(gl);
1239 }
1240
1241 /**
1242  * gfs2_should_freeze - Figure out if glock should be frozen
1243  * @gl: The glock in question
1244  *
1245  * Glocks are not frozen if (a) the result of the dlm operation is
1246  * an error, (b) the locking operation was an unlock operation or
1247  * (c) if there is a "noexp" flagged request anywhere in the queue
1248  *
1249  * Returns: 1 if freezing should occur, 0 otherwise
1250  */
1251
1252 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1253 {
1254         const struct gfs2_holder *gh;
1255
1256         if (gl->gl_reply & ~LM_OUT_ST_MASK)
1257                 return 0;
1258         if (gl->gl_target == LM_ST_UNLOCKED)
1259                 return 0;
1260
1261         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1262                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1263                         continue;
1264                 if (LM_FLAG_NOEXP & gh->gh_flags)
1265                         return 0;
1266         }
1267
1268         return 1;
1269 }
1270
1271 /**
1272  * gfs2_glock_complete - Callback used by locking
1273  * @gl: Pointer to the glock
1274  * @ret: The return value from the dlm
1275  *
1276  * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1277  * to use a bitfield shared with other glock state fields.
1278  */
1279
1280 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1281 {
1282         struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1283
1284         spin_lock(&gl->gl_lockref.lock);
1285         gl->gl_reply = ret;
1286
1287         if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1288                 if (gfs2_should_freeze(gl)) {
1289                         set_bit(GLF_FROZEN, &gl->gl_flags);
1290                         spin_unlock(&gl->gl_lockref.lock);
1291                         return;
1292                 }
1293         }
1294
1295         gl->gl_lockref.count++;
1296         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1297         spin_unlock(&gl->gl_lockref.lock);
1298
1299         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1300                 gfs2_glock_put(gl);
1301 }
1302
1303 static int glock_cmp(void *priv, struct list_head *a, struct list_head *b)
1304 {
1305         struct gfs2_glock *gla, *glb;
1306
1307         gla = list_entry(a, struct gfs2_glock, gl_lru);
1308         glb = list_entry(b, struct gfs2_glock, gl_lru);
1309
1310         if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1311                 return 1;
1312         if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1313                 return -1;
1314
1315         return 0;
1316 }
1317
1318 /**
1319  * gfs2_dispose_glock_lru - Demote a list of glocks
1320  * @list: The list to dispose of
1321  *
1322  * Disposing of glocks may involve disk accesses, so that here we sort
1323  * the glocks by number (i.e. disk location of the inodes) so that if
1324  * there are any such accesses, they'll be sent in order (mostly).
1325  *
1326  * Must be called under the lru_lock, but may drop and retake this
1327  * lock. While the lru_lock is dropped, entries may vanish from the
1328  * list, but no new entries will appear on the list (since it is
1329  * private)
1330  */
1331
1332 static void gfs2_dispose_glock_lru(struct list_head *list)
1333 __releases(&lru_lock)
1334 __acquires(&lru_lock)
1335 {
1336         struct gfs2_glock *gl;
1337
1338         list_sort(NULL, list, glock_cmp);
1339
1340         while(!list_empty(list)) {
1341                 gl = list_entry(list->next, struct gfs2_glock, gl_lru);
1342                 list_del_init(&gl->gl_lru);
1343                 if (!spin_trylock(&gl->gl_lockref.lock)) {
1344 add_back_to_lru:
1345                         list_add(&gl->gl_lru, &lru_list);
1346                         atomic_inc(&lru_count);
1347                         continue;
1348                 }
1349                 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1350                         spin_unlock(&gl->gl_lockref.lock);
1351                         goto add_back_to_lru;
1352                 }
1353                 clear_bit(GLF_LRU, &gl->gl_flags);
1354                 gl->gl_lockref.count++;
1355                 if (demote_ok(gl))
1356                         handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1357                 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1358                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1359                         gl->gl_lockref.count--;
1360                 spin_unlock(&gl->gl_lockref.lock);
1361                 cond_resched_lock(&lru_lock);
1362         }
1363 }
1364
1365 /**
1366  * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1367  * @nr: The number of entries to scan
1368  *
1369  * This function selects the entries on the LRU which are able to
1370  * be demoted, and then kicks off the process by calling
1371  * gfs2_dispose_glock_lru() above.
1372  */
1373
1374 static long gfs2_scan_glock_lru(int nr)
1375 {
1376         struct gfs2_glock *gl;
1377         LIST_HEAD(skipped);
1378         LIST_HEAD(dispose);
1379         long freed = 0;
1380
1381         spin_lock(&lru_lock);
1382         while ((nr-- >= 0) && !list_empty(&lru_list)) {
1383                 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1384
1385                 /* Test for being demotable */
1386                 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1387                         list_move(&gl->gl_lru, &dispose);
1388                         atomic_dec(&lru_count);
1389                         freed++;
1390                         continue;
1391                 }
1392
1393                 list_move(&gl->gl_lru, &skipped);
1394         }
1395         list_splice(&skipped, &lru_list);
1396         if (!list_empty(&dispose))
1397                 gfs2_dispose_glock_lru(&dispose);
1398         spin_unlock(&lru_lock);
1399
1400         return freed;
1401 }
1402
1403 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1404                                             struct shrink_control *sc)
1405 {
1406         if (!(sc->gfp_mask & __GFP_FS))
1407                 return SHRINK_STOP;
1408         return gfs2_scan_glock_lru(sc->nr_to_scan);
1409 }
1410
1411 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1412                                              struct shrink_control *sc)
1413 {
1414         return vfs_pressure_ratio(atomic_read(&lru_count));
1415 }
1416
1417 static struct shrinker glock_shrinker = {
1418         .seeks = DEFAULT_SEEKS,
1419         .count_objects = gfs2_glock_shrink_count,
1420         .scan_objects = gfs2_glock_shrink_scan,
1421 };
1422
1423 /**
1424  * examine_bucket - Call a function for glock in a hash bucket
1425  * @examiner: the function
1426  * @sdp: the filesystem
1427  * @bucket: the bucket
1428  *
1429  * Note that the function can be called multiple times on the same
1430  * object.  So the user must ensure that the function can cope with
1431  * that.
1432  */
1433
1434 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1435 {
1436         struct gfs2_glock *gl;
1437         struct rhashtable_iter iter;
1438
1439         rhashtable_walk_enter(&gl_hash_table, &iter);
1440
1441         do {
1442                 gl = ERR_PTR(rhashtable_walk_start(&iter));
1443                 if (gl)
1444                         continue;
1445
1446                 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl))
1447                         if ((gl->gl_name.ln_sbd == sdp) &&
1448                             lockref_get_not_dead(&gl->gl_lockref))
1449                                 examiner(gl);
1450
1451                 rhashtable_walk_stop(&iter);
1452         } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
1453
1454         rhashtable_walk_exit(&iter);
1455 }
1456
1457 /**
1458  * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1459  * @gl: The glock to thaw
1460  *
1461  */
1462
1463 static void thaw_glock(struct gfs2_glock *gl)
1464 {
1465         if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1466                 goto out;
1467         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1468         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0) {
1469 out:
1470                 gfs2_glock_put(gl);
1471         }
1472 }
1473
1474 /**
1475  * clear_glock - look at a glock and see if we can free it from glock cache
1476  * @gl: the glock to look at
1477  *
1478  */
1479
1480 static void clear_glock(struct gfs2_glock *gl)
1481 {
1482         gfs2_glock_remove_from_lru(gl);
1483
1484         spin_lock(&gl->gl_lockref.lock);
1485         if (gl->gl_state != LM_ST_UNLOCKED)
1486                 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1487         spin_unlock(&gl->gl_lockref.lock);
1488         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1489                 gfs2_glock_put(gl);
1490 }
1491
1492 /**
1493  * gfs2_glock_thaw - Thaw any frozen glocks
1494  * @sdp: The super block
1495  *
1496  */
1497
1498 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1499 {
1500         glock_hash_walk(thaw_glock, sdp);
1501 }
1502
1503 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1504 {
1505         spin_lock(&gl->gl_lockref.lock);
1506         gfs2_dump_glock(seq, gl);
1507         spin_unlock(&gl->gl_lockref.lock);
1508 }
1509
1510 static void dump_glock_func(struct gfs2_glock *gl)
1511 {
1512         dump_glock(NULL, gl);
1513 }
1514
1515 /**
1516  * gfs2_gl_hash_clear - Empty out the glock hash table
1517  * @sdp: the filesystem
1518  * @wait: wait until it's all gone
1519  *
1520  * Called when unmounting the filesystem.
1521  */
1522
1523 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1524 {
1525         set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1526         flush_workqueue(glock_workqueue);
1527         glock_hash_walk(clear_glock, sdp);
1528         flush_workqueue(glock_workqueue);
1529         wait_event_timeout(sdp->sd_glock_wait,
1530                            atomic_read(&sdp->sd_glock_disposal) == 0,
1531                            HZ * 600);
1532         glock_hash_walk(dump_glock_func, sdp);
1533 }
1534
1535 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1536 {
1537         struct gfs2_glock *gl = ip->i_gl;
1538         int ret;
1539
1540         ret = gfs2_truncatei_resume(ip);
1541         gfs2_assert_withdraw(gl->gl_name.ln_sbd, ret == 0);
1542
1543         spin_lock(&gl->gl_lockref.lock);
1544         clear_bit(GLF_LOCK, &gl->gl_flags);
1545         run_queue(gl, 1);
1546         spin_unlock(&gl->gl_lockref.lock);
1547 }
1548
1549 static const char *state2str(unsigned state)
1550 {
1551         switch(state) {
1552         case LM_ST_UNLOCKED:
1553                 return "UN";
1554         case LM_ST_SHARED:
1555                 return "SH";
1556         case LM_ST_DEFERRED:
1557                 return "DF";
1558         case LM_ST_EXCLUSIVE:
1559                 return "EX";
1560         }
1561         return "??";
1562 }
1563
1564 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
1565 {
1566         char *p = buf;
1567         if (flags & LM_FLAG_TRY)
1568                 *p++ = 't';
1569         if (flags & LM_FLAG_TRY_1CB)
1570                 *p++ = 'T';
1571         if (flags & LM_FLAG_NOEXP)
1572                 *p++ = 'e';
1573         if (flags & LM_FLAG_ANY)
1574                 *p++ = 'A';
1575         if (flags & LM_FLAG_PRIORITY)
1576                 *p++ = 'p';
1577         if (flags & GL_ASYNC)
1578                 *p++ = 'a';
1579         if (flags & GL_EXACT)
1580                 *p++ = 'E';
1581         if (flags & GL_NOCACHE)
1582                 *p++ = 'c';
1583         if (test_bit(HIF_HOLDER, &iflags))
1584                 *p++ = 'H';
1585         if (test_bit(HIF_WAIT, &iflags))
1586                 *p++ = 'W';
1587         if (test_bit(HIF_FIRST, &iflags))
1588                 *p++ = 'F';
1589         *p = 0;
1590         return buf;
1591 }
1592
1593 /**
1594  * dump_holder - print information about a glock holder
1595  * @seq: the seq_file struct
1596  * @gh: the glock holder
1597  *
1598  */
1599
1600 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1601 {
1602         struct task_struct *gh_owner = NULL;
1603         char flags_buf[32];
1604
1605         rcu_read_lock();
1606         if (gh->gh_owner_pid)
1607                 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1608         gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1609                        state2str(gh->gh_state),
1610                        hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1611                        gh->gh_error,
1612                        gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1613                        gh_owner ? gh_owner->comm : "(ended)",
1614                        (void *)gh->gh_ip);
1615         rcu_read_unlock();
1616 }
1617
1618 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1619 {
1620         const unsigned long *gflags = &gl->gl_flags;
1621         char *p = buf;
1622
1623         if (test_bit(GLF_LOCK, gflags))
1624                 *p++ = 'l';
1625         if (test_bit(GLF_DEMOTE, gflags))
1626                 *p++ = 'D';
1627         if (test_bit(GLF_PENDING_DEMOTE, gflags))
1628                 *p++ = 'd';
1629         if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1630                 *p++ = 'p';
1631         if (test_bit(GLF_DIRTY, gflags))
1632                 *p++ = 'y';
1633         if (test_bit(GLF_LFLUSH, gflags))
1634                 *p++ = 'f';
1635         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1636                 *p++ = 'i';
1637         if (test_bit(GLF_REPLY_PENDING, gflags))
1638                 *p++ = 'r';
1639         if (test_bit(GLF_INITIAL, gflags))
1640                 *p++ = 'I';
1641         if (test_bit(GLF_FROZEN, gflags))
1642                 *p++ = 'F';
1643         if (test_bit(GLF_QUEUED, gflags))
1644                 *p++ = 'q';
1645         if (test_bit(GLF_LRU, gflags))
1646                 *p++ = 'L';
1647         if (gl->gl_object)
1648                 *p++ = 'o';
1649         if (test_bit(GLF_BLOCKING, gflags))
1650                 *p++ = 'b';
1651         *p = 0;
1652         return buf;
1653 }
1654
1655 /**
1656  * gfs2_dump_glock - print information about a glock
1657  * @seq: The seq_file struct
1658  * @gl: the glock
1659  *
1660  * The file format is as follows:
1661  * One line per object, capital letters are used to indicate objects
1662  * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1663  * other objects are indented by a single space and follow the glock to
1664  * which they are related. Fields are indicated by lower case letters
1665  * followed by a colon and the field value, except for strings which are in
1666  * [] so that its possible to see if they are composed of spaces for
1667  * example. The field's are n = number (id of the object), f = flags,
1668  * t = type, s = state, r = refcount, e = error, p = pid.
1669  *
1670  */
1671
1672 void gfs2_dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1673 {
1674         const struct gfs2_glock_operations *glops = gl->gl_ops;
1675         unsigned long long dtime;
1676         const struct gfs2_holder *gh;
1677         char gflags_buf[32];
1678
1679         dtime = jiffies - gl->gl_demote_time;
1680         dtime *= 1000000/HZ; /* demote time in uSec */
1681         if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1682                 dtime = 0;
1683         gfs2_print_dbg(seq, "G:  s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d v:%d r:%d m:%ld\n",
1684                   state2str(gl->gl_state),
1685                   gl->gl_name.ln_type,
1686                   (unsigned long long)gl->gl_name.ln_number,
1687                   gflags2str(gflags_buf, gl),
1688                   state2str(gl->gl_target),
1689                   state2str(gl->gl_demote_state), dtime,
1690                   atomic_read(&gl->gl_ail_count),
1691                   atomic_read(&gl->gl_revokes),
1692                   (int)gl->gl_lockref.count, gl->gl_hold_time);
1693
1694         list_for_each_entry(gh, &gl->gl_holders, gh_list)
1695                 dump_holder(seq, gh);
1696
1697         if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1698                 glops->go_dump(seq, gl);
1699 }
1700
1701 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
1702 {
1703         struct gfs2_glock *gl = iter_ptr;
1704
1705         seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
1706                    gl->gl_name.ln_type,
1707                    (unsigned long long)gl->gl_name.ln_number,
1708                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
1709                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
1710                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
1711                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
1712                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
1713                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
1714                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
1715                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
1716         return 0;
1717 }
1718
1719 static const char *gfs2_gltype[] = {
1720         "type",
1721         "reserved",
1722         "nondisk",
1723         "inode",
1724         "rgrp",
1725         "meta",
1726         "iopen",
1727         "flock",
1728         "plock",
1729         "quota",
1730         "journal",
1731 };
1732
1733 static const char *gfs2_stype[] = {
1734         [GFS2_LKS_SRTT]         = "srtt",
1735         [GFS2_LKS_SRTTVAR]      = "srttvar",
1736         [GFS2_LKS_SRTTB]        = "srttb",
1737         [GFS2_LKS_SRTTVARB]     = "srttvarb",
1738         [GFS2_LKS_SIRT]         = "sirt",
1739         [GFS2_LKS_SIRTVAR]      = "sirtvar",
1740         [GFS2_LKS_DCOUNT]       = "dlm",
1741         [GFS2_LKS_QCOUNT]       = "queue",
1742 };
1743
1744 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1745
1746 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
1747 {
1748         struct gfs2_sbd *sdp = seq->private;
1749         loff_t pos = *(loff_t *)iter_ptr;
1750         unsigned index = pos >> 3;
1751         unsigned subindex = pos & 0x07;
1752         int i;
1753
1754         if (index == 0 && subindex != 0)
1755                 return 0;
1756
1757         seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
1758                    (index == 0) ? "cpu": gfs2_stype[subindex]);
1759
1760         for_each_possible_cpu(i) {
1761                 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
1762
1763                 if (index == 0)
1764                         seq_printf(seq, " %15u", i);
1765                 else
1766                         seq_printf(seq, " %15llu", (unsigned long long)lkstats->
1767                                    lkstats[index - 1].stats[subindex]);
1768         }
1769         seq_putc(seq, '\n');
1770         return 0;
1771 }
1772
1773 int __init gfs2_glock_init(void)
1774 {
1775         int ret;
1776
1777         ret = rhashtable_init(&gl_hash_table, &ht_parms);
1778         if (ret < 0)
1779                 return ret;
1780
1781         glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1782                                           WQ_HIGHPRI | WQ_FREEZABLE, 0);
1783         if (!glock_workqueue) {
1784                 rhashtable_destroy(&gl_hash_table);
1785                 return -ENOMEM;
1786         }
1787         gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1788                                                 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1789                                                 0);
1790         if (!gfs2_delete_workqueue) {
1791                 destroy_workqueue(glock_workqueue);
1792                 rhashtable_destroy(&gl_hash_table);
1793                 return -ENOMEM;
1794         }
1795
1796         ret = register_shrinker(&glock_shrinker);
1797         if (ret) {
1798                 destroy_workqueue(gfs2_delete_workqueue);
1799                 destroy_workqueue(glock_workqueue);
1800                 rhashtable_destroy(&gl_hash_table);
1801                 return ret;
1802         }
1803
1804         return 0;
1805 }
1806
1807 void gfs2_glock_exit(void)
1808 {
1809         unregister_shrinker(&glock_shrinker);
1810         rhashtable_destroy(&gl_hash_table);
1811         destroy_workqueue(glock_workqueue);
1812         destroy_workqueue(gfs2_delete_workqueue);
1813 }
1814
1815 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1816 {
1817         while ((gi->gl = rhashtable_walk_next(&gi->hti))) {
1818                 if (IS_ERR(gi->gl)) {
1819                         if (PTR_ERR(gi->gl) == -EAGAIN)
1820                                 continue;
1821                         gi->gl = NULL;
1822                         return;
1823                 }
1824                 /* Skip entries for other sb and dead entries */
1825                 if (gi->sdp == gi->gl->gl_name.ln_sbd &&
1826                     !__lockref_is_dead(&gi->gl->gl_lockref))
1827                         return;
1828         }
1829 }
1830
1831 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1832 {
1833         struct gfs2_glock_iter *gi = seq->private;
1834         loff_t n = *pos;
1835         int ret;
1836
1837         if (gi->last_pos <= *pos)
1838                 n = (*pos - gi->last_pos);
1839
1840         ret = rhashtable_walk_start(&gi->hti);
1841         if (ret)
1842                 return NULL;
1843
1844         do {
1845                 gfs2_glock_iter_next(gi);
1846         } while (gi->gl && n--);
1847
1848         gi->last_pos = *pos;
1849         return gi->gl;
1850 }
1851
1852 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1853                                  loff_t *pos)
1854 {
1855         struct gfs2_glock_iter *gi = seq->private;
1856
1857         (*pos)++;
1858         gi->last_pos = *pos;
1859         gfs2_glock_iter_next(gi);
1860         return gi->gl;
1861 }
1862
1863 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1864 {
1865         struct gfs2_glock_iter *gi = seq->private;
1866
1867         gi->gl = NULL;
1868         rhashtable_walk_stop(&gi->hti);
1869 }
1870
1871 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1872 {
1873         dump_glock(seq, iter_ptr);
1874         return 0;
1875 }
1876
1877 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
1878 {
1879         preempt_disable();
1880         if (*pos >= GFS2_NR_SBSTATS)
1881                 return NULL;
1882         return pos;
1883 }
1884
1885 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
1886                                    loff_t *pos)
1887 {
1888         (*pos)++;
1889         if (*pos >= GFS2_NR_SBSTATS)
1890                 return NULL;
1891         return pos;
1892 }
1893
1894 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
1895 {
1896         preempt_enable();
1897 }
1898
1899 static const struct seq_operations gfs2_glock_seq_ops = {
1900         .start = gfs2_glock_seq_start,
1901         .next  = gfs2_glock_seq_next,
1902         .stop  = gfs2_glock_seq_stop,
1903         .show  = gfs2_glock_seq_show,
1904 };
1905
1906 static const struct seq_operations gfs2_glstats_seq_ops = {
1907         .start = gfs2_glock_seq_start,
1908         .next  = gfs2_glock_seq_next,
1909         .stop  = gfs2_glock_seq_stop,
1910         .show  = gfs2_glstats_seq_show,
1911 };
1912
1913 static const struct seq_operations gfs2_sbstats_seq_ops = {
1914         .start = gfs2_sbstats_seq_start,
1915         .next  = gfs2_sbstats_seq_next,
1916         .stop  = gfs2_sbstats_seq_stop,
1917         .show  = gfs2_sbstats_seq_show,
1918 };
1919
1920 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
1921
1922 static int __gfs2_glocks_open(struct inode *inode, struct file *file,
1923                               const struct seq_operations *ops)
1924 {
1925         int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
1926         if (ret == 0) {
1927                 struct seq_file *seq = file->private_data;
1928                 struct gfs2_glock_iter *gi = seq->private;
1929
1930                 gi->sdp = inode->i_private;
1931                 gi->last_pos = 0;
1932                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
1933                 if (seq->buf)
1934                         seq->size = GFS2_SEQ_GOODSIZE;
1935                 gi->gl = NULL;
1936                 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
1937         }
1938         return ret;
1939 }
1940
1941 static int gfs2_glocks_open(struct inode *inode, struct file *file)
1942 {
1943         return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
1944 }
1945
1946 static int gfs2_glocks_release(struct inode *inode, struct file *file)
1947 {
1948         struct seq_file *seq = file->private_data;
1949         struct gfs2_glock_iter *gi = seq->private;
1950
1951         gi->gl = NULL;
1952         rhashtable_walk_exit(&gi->hti);
1953         return seq_release_private(inode, file);
1954 }
1955
1956 static int gfs2_glstats_open(struct inode *inode, struct file *file)
1957 {
1958         return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
1959 }
1960
1961 static int gfs2_sbstats_open(struct inode *inode, struct file *file)
1962 {
1963         int ret = seq_open(file, &gfs2_sbstats_seq_ops);
1964         if (ret == 0) {
1965                 struct seq_file *seq = file->private_data;
1966                 seq->private = inode->i_private;  /* sdp */
1967         }
1968         return ret;
1969 }
1970
1971 static const struct file_operations gfs2_glocks_fops = {
1972         .owner   = THIS_MODULE,
1973         .open    = gfs2_glocks_open,
1974         .read    = seq_read,
1975         .llseek  = seq_lseek,
1976         .release = gfs2_glocks_release,
1977 };
1978
1979 static const struct file_operations gfs2_glstats_fops = {
1980         .owner   = THIS_MODULE,
1981         .open    = gfs2_glstats_open,
1982         .read    = seq_read,
1983         .llseek  = seq_lseek,
1984         .release = gfs2_glocks_release,
1985 };
1986
1987 static const struct file_operations gfs2_sbstats_fops = {
1988         .owner   = THIS_MODULE,
1989         .open    = gfs2_sbstats_open,
1990         .read    = seq_read,
1991         .llseek  = seq_lseek,
1992         .release = seq_release,
1993 };
1994
1995 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
1996 {
1997         struct dentry *dent;
1998
1999         dent = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2000         if (IS_ERR_OR_NULL(dent))
2001                 goto fail;
2002         sdp->debugfs_dir = dent;
2003
2004         dent = debugfs_create_file("glocks",
2005                                    S_IFREG | S_IRUGO,
2006                                    sdp->debugfs_dir, sdp,
2007                                    &gfs2_glocks_fops);
2008         if (IS_ERR_OR_NULL(dent))
2009                 goto fail;
2010         sdp->debugfs_dentry_glocks = dent;
2011
2012         dent = debugfs_create_file("glstats",
2013                                    S_IFREG | S_IRUGO,
2014                                    sdp->debugfs_dir, sdp,
2015                                    &gfs2_glstats_fops);
2016         if (IS_ERR_OR_NULL(dent))
2017                 goto fail;
2018         sdp->debugfs_dentry_glstats = dent;
2019
2020         dent = debugfs_create_file("sbstats",
2021                                    S_IFREG | S_IRUGO,
2022                                    sdp->debugfs_dir, sdp,
2023                                    &gfs2_sbstats_fops);
2024         if (IS_ERR_OR_NULL(dent))
2025                 goto fail;
2026         sdp->debugfs_dentry_sbstats = dent;
2027
2028         return 0;
2029 fail:
2030         gfs2_delete_debugfs_file(sdp);
2031         return dent ? PTR_ERR(dent) : -ENOMEM;
2032 }
2033
2034 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2035 {
2036         if (sdp->debugfs_dir) {
2037                 if (sdp->debugfs_dentry_glocks) {
2038                         debugfs_remove(sdp->debugfs_dentry_glocks);
2039                         sdp->debugfs_dentry_glocks = NULL;
2040                 }
2041                 if (sdp->debugfs_dentry_glstats) {
2042                         debugfs_remove(sdp->debugfs_dentry_glstats);
2043                         sdp->debugfs_dentry_glstats = NULL;
2044                 }
2045                 if (sdp->debugfs_dentry_sbstats) {
2046                         debugfs_remove(sdp->debugfs_dentry_sbstats);
2047                         sdp->debugfs_dentry_sbstats = NULL;
2048                 }
2049                 debugfs_remove(sdp->debugfs_dir);
2050                 sdp->debugfs_dir = NULL;
2051         }
2052 }
2053
2054 int gfs2_register_debugfs(void)
2055 {
2056         gfs2_root = debugfs_create_dir("gfs2", NULL);
2057         if (IS_ERR(gfs2_root))
2058                 return PTR_ERR(gfs2_root);
2059         return gfs2_root ? 0 : -ENOMEM;
2060 }
2061
2062 void gfs2_unregister_debugfs(void)
2063 {
2064         debugfs_remove(gfs2_root);
2065         gfs2_root = NULL;
2066 }