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 <asm/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       (1 << GFS2_GL_HASH_SHIFT)
  73
  74 static struct rhashtable_params ht_parms = {
  75         .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4,
  76         .key_len = sizeof(struct lm_lockname),
  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 static void gfs2_glock_dealloc(struct rcu_head *rcu)
  84 {
  85         struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
  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 }
  94
  95 void gfs2_glock_free(struct gfs2_glock *gl)
  96 {
  97         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
  98
  99         call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
 100         if (atomic_dec_and_test(&sdp->sd_glock_disposal))
 101                 wake_up(&sdp->sd_glock_wait);
 102 }
 103
 104 /**
 105  * gfs2_glock_hold() - increment reference count on glock
 106  * @gl: The glock to hold
 107  *
 108  */
 109
 110 static void gfs2_glock_hold(struct gfs2_glock *gl)
 111 {
 112         GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
 113         lockref_get(&gl->gl_lockref);
 114 }
 115
 116 /**
 117  * demote_ok - Check to see if it's ok to unlock a glock
 118  * @gl: the glock
 119  *
 120  * Returns: 1 if it's ok
 121  */
 122
 123 static int demote_ok(const struct gfs2_glock *gl)
 124 {
 125         const struct gfs2_glock_operations *glops = gl->gl_ops;
 126
 127         if (gl->gl_state == LM_ST_UNLOCKED)
 128                 return 0;
 129         if (!list_empty(&gl->gl_holders))
 130                 return 0;
 131         if (glops->go_demote_ok)
 132                 return glops->go_demote_ok(gl);
 133         return 1;
 134 }
 135
 136
 137 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
 138 {
 139         if (!(gl->gl_ops->go_flags & GLOF_LRU))
 140                 return;
 141
 142         spin_lock(&lru_lock);
 143
 144         list_del(&gl->gl_lru);
 145         list_add_tail(&gl->gl_lru, &lru_list);
 146
 147         if (!test_bit(GLF_LRU, &gl->gl_flags)) {
 148                 set_bit(GLF_LRU, &gl->gl_flags);
 149                 atomic_inc(&lru_count);
 150         }
 151
 152         spin_unlock(&lru_lock);
 153 }
 154
 155 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
 156 {
 157         spin_lock(&lru_lock);
 158         if (test_bit(GLF_LRU, &gl->gl_flags)) {
 159                 list_del_init(&gl->gl_lru);
 160                 atomic_dec(&lru_count);
 161                 clear_bit(GLF_LRU, &gl->gl_flags);
 162         }
 163         spin_unlock(&lru_lock);
 164 }
 165
 166 /**
 167  * gfs2_glock_put() - Decrement reference count on glock
 168  * @gl: The glock to put
 169  *
 170  */
 171
 172 void gfs2_glock_put(struct gfs2_glock *gl)
 173 {
 174         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 175         struct address_space *mapping = gfs2_glock2aspace(gl);
 176
 177         if (lockref_put_or_lock(&gl->gl_lockref))
 178                 return;
 179
 180         lockref_mark_dead(&gl->gl_lockref);
 181
 182         gfs2_glock_remove_from_lru(gl);
 183         spin_unlock(&gl->gl_lockref.lock);
 184         rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
 185         GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
 186         GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
 187         trace_gfs2_glock_put(gl);
 188         sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
 189 }
 190
 191 /**
 192  * may_grant - check if its ok to grant a new lock
 193  * @gl: The glock
 194  * @gh: The lock request which we wish to grant
 195  *
 196  * Returns: true if its ok to grant the lock
 197  */
 198
 199 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
 200 {
 201         const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
 202         if ((gh->gh_state == LM_ST_EXCLUSIVE ||
 203              gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
 204                 return 0;
 205         if (gl->gl_state == gh->gh_state)
 206                 return 1;
 207         if (gh->gh_flags & GL_EXACT)
 208                 return 0;
 209         if (gl->gl_state == LM_ST_EXCLUSIVE) {
 210                 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
 211                         return 1;
 212                 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
 213                         return 1;
 214         }
 215         if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
 216                 return 1;
 217         return 0;
 218 }
 219
 220 static void gfs2_holder_wake(struct gfs2_holder *gh)
 221 {
 222         clear_bit(HIF_WAIT, &gh->gh_iflags);
 223         smp_mb__after_atomic();
 224         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
 225 }
 226
 227 /**
 228  * do_error - Something unexpected has happened during a lock request
 229  *
 230  */
 231
 232 static inline void do_error(struct gfs2_glock *gl, const int ret)
 233 {
 234         struct gfs2_holder *gh, *tmp;
 235
 236         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
 237                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 238                         continue;
 239                 if (ret & LM_OUT_ERROR)
 240                         gh->gh_error = -EIO;
 241                 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
 242                         gh->gh_error = GLR_TRYFAILED;
 243                 else
 244                         continue;
 245                 list_del_init(&gh->gh_list);
 246                 trace_gfs2_glock_queue(gh, 0);
 247                 gfs2_holder_wake(gh);
 248         }
 249 }
 250
 251 /**
 252  * do_promote - promote as many requests as possible on the current queue
 253  * @gl: The glock
 254  *
 255  * Returns: 1 if there is a blocked holder at the head of the list, or 2
 256  *          if a type specific operation is underway.
 257  */
 258
 259 static int do_promote(struct gfs2_glock *gl)
 260 __releases(&gl->gl_lockref.lock)
 261 __acquires(&gl->gl_lockref.lock)
 262 {
 263         const struct gfs2_glock_operations *glops = gl->gl_ops;
 264         struct gfs2_holder *gh, *tmp;
 265         int ret;
 266
 267 restart:
 268         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
 269                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 270                         continue;
 271                 if (may_grant(gl, gh)) {
 272                         if (gh->gh_list.prev == &gl->gl_holders &&
 273                             glops->go_lock) {
 274                                 spin_unlock(&gl->gl_lockref.lock);
 275                                 /* FIXME: eliminate this eventually */
 276                                 ret = glops->go_lock(gh);
 277                                 spin_lock(&gl->gl_lockref.lock);
 278                                 if (ret) {
 279                                         if (ret == 1)
 280                                                 return 2;
 281                                         gh->gh_error = ret;
 282                                         list_del_init(&gh->gh_list);
 283                                         trace_gfs2_glock_queue(gh, 0);
 284                                         gfs2_holder_wake(gh);
 285                                         goto restart;
 286                                 }
 287                                 set_bit(HIF_HOLDER, &gh->gh_iflags);
 288                                 trace_gfs2_promote(gh, 1);
 289                                 gfs2_holder_wake(gh);
 290                                 goto restart;
 291                         }
 292                         set_bit(HIF_HOLDER, &gh->gh_iflags);
 293                         trace_gfs2_promote(gh, 0);
 294                         gfs2_holder_wake(gh);
 295                         continue;
 296                 }
 297                 if (gh->gh_list.prev == &gl->gl_holders)
 298                         return 1;
 299                 do_error(gl, 0);
 300                 break;
 301         }
 302         return 0;
 303 }
 304
 305 /**
 306  * find_first_waiter - find the first gh that's waiting for the glock
 307  * @gl: the glock
 308  */
 309
 310 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
 311 {
 312         struct gfs2_holder *gh;
 313
 314         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
 315                 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
 316                         return gh;
 317         }
 318         return NULL;
 319 }
 320
 321 /**
 322  * state_change - record that the glock is now in a different state
 323  * @gl: the glock
 324  * @new_state the new state
 325  *
 326  */
 327
 328 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
 329 {
 330         int held1, held2;
 331
 332         held1 = (gl->gl_state != LM_ST_UNLOCKED);
 333         held2 = (new_state != LM_ST_UNLOCKED);
 334
 335         if (held1 != held2) {
 336                 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
 337                 if (held2)
 338                         gl->gl_lockref.count++;
 339                 else
 340                         gl->gl_lockref.count--;
 341         }
 342         if (held1 && held2 && list_empty(&gl->gl_holders))
 343                 clear_bit(GLF_QUEUED, &gl->gl_flags);
 344
 345         if (new_state != gl->gl_target)
 346                 /* shorten our minimum hold time */
 347                 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
 348                                        GL_GLOCK_MIN_HOLD);
 349         gl->gl_state = new_state;
 350         gl->gl_tchange = jiffies;
 351 }
 352
 353 static void gfs2_demote_wake(struct gfs2_glock *gl)
 354 {
 355         gl->gl_demote_state = LM_ST_EXCLUSIVE;
 356         clear_bit(GLF_DEMOTE, &gl->gl_flags);
 357         smp_mb__after_atomic();
 358         wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
 359 }
 360
 361 /**
 362  * finish_xmote - The DLM has replied to one of our lock requests
 363  * @gl: The glock
 364  * @ret: The status from the DLM
 365  *
 366  */
 367
 368 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
 369 {
 370         const struct gfs2_glock_operations *glops = gl->gl_ops;
 371         struct gfs2_holder *gh;
 372         unsigned state = ret & LM_OUT_ST_MASK;
 373         int rv;
 374
 375         spin_lock(&gl->gl_lockref.lock);
 376         trace_gfs2_glock_state_change(gl, state);
 377         state_change(gl, state);
 378         gh = find_first_waiter(gl);
 379
 380         /* Demote to UN request arrived during demote to SH or DF */
 381         if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
 382             state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
 383                 gl->gl_target = LM_ST_UNLOCKED;
 384
 385         /* Check for state != intended state */
 386         if (unlikely(state != gl->gl_target)) {
 387                 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
 388                         /* move to back of queue and try next entry */
 389                         if (ret & LM_OUT_CANCELED) {
 390                                 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
 391                                         list_move_tail(&gh->gh_list, &gl->gl_holders);
 392                                 gh = find_first_waiter(gl);
 393                                 gl->gl_target = gh->gh_state;
 394                                 goto retry;
 395                         }
 396                         /* Some error or failed "try lock" - report it */
 397                         if ((ret & LM_OUT_ERROR) ||
 398                             (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
 399                                 gl->gl_target = gl->gl_state;
 400                                 do_error(gl, ret);
 401                                 goto out;
 402                         }
 403                 }
 404                 switch(state) {
 405                 /* Unlocked due to conversion deadlock, try again */
 406                 case LM_ST_UNLOCKED:
 407 retry:
 408                         do_xmote(gl, gh, gl->gl_target);
 409                         break;
 410                 /* Conversion fails, unlock and try again */
 411                 case LM_ST_SHARED:
 412                 case LM_ST_DEFERRED:
 413                         do_xmote(gl, gh, LM_ST_UNLOCKED);
 414                         break;
 415                 default: /* Everything else */
 416                         pr_err("wanted %u got %u\n", gl->gl_target, state);
 417                         GLOCK_BUG_ON(gl, 1);
 418                 }
 419                 spin_unlock(&gl->gl_lockref.lock);
 420                 return;
 421         }
 422
 423         /* Fast path - we got what we asked for */
 424         if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
 425                 gfs2_demote_wake(gl);
 426         if (state != LM_ST_UNLOCKED) {
 427                 if (glops->go_xmote_bh) {
 428                         spin_unlock(&gl->gl_lockref.lock);
 429                         rv = glops->go_xmote_bh(gl, gh);
 430                         spin_lock(&gl->gl_lockref.lock);
 431                         if (rv) {
 432                                 do_error(gl, rv);
 433                                 goto out;
 434                         }
 435                 }
 436                 rv = do_promote(gl);
 437                 if (rv == 2)
 438                         goto out_locked;
 439         }
 440 out:
 441         clear_bit(GLF_LOCK, &gl->gl_flags);
 442 out_locked:
 443         spin_unlock(&gl->gl_lockref.lock);
 444 }
 445
 446 /**
 447  * do_xmote - Calls the DLM to change the state of a lock
 448  * @gl: The lock state
 449  * @gh: The holder (only for promotes)
 450  * @target: The target lock state
 451  *
 452  */
 453
 454 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
 455 __releases(&gl->gl_lockref.lock)
 456 __acquires(&gl->gl_lockref.lock)
 457 {
 458         const struct gfs2_glock_operations *glops = gl->gl_ops;
 459         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 460         unsigned int lck_flags = gh ? gh->gh_flags : 0;
 461         int ret;
 462
 463         lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
 464                       LM_FLAG_PRIORITY);
 465         GLOCK_BUG_ON(gl, gl->gl_state == target);
 466         GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
 467         if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
 468             glops->go_inval) {
 469                 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 470                 do_error(gl, 0); /* Fail queued try locks */
 471         }
 472         gl->gl_req = target;
 473         set_bit(GLF_BLOCKING, &gl->gl_flags);
 474         if ((gl->gl_req == LM_ST_UNLOCKED) ||
 475             (gl->gl_state == LM_ST_EXCLUSIVE) ||
 476             (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
 477                 clear_bit(GLF_BLOCKING, &gl->gl_flags);
 478         spin_unlock(&gl->gl_lockref.lock);
 479         if (glops->go_sync)
 480                 glops->go_sync(gl);
 481         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 482                 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
 483         clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 484
 485         gfs2_glock_hold(gl);
 486         if (sdp->sd_lockstruct.ls_ops->lm_lock) {
 487                 /* lock_dlm */
 488                 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
 489                 if (ret) {
 490                         pr_err("lm_lock ret %d\n", ret);
 491                         GLOCK_BUG_ON(gl, 1);
 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 gfs2_inode *ip;
 583         struct inode *inode;
 584         u64 no_addr = gl->gl_name.ln_number;
 585
 586         ip = gl->gl_object;
 587         /* Note: Unsafe to dereference ip as we don't hold right refs/locks */
 588
 589         if (ip)
 590                 inode = gfs2_ilookup(sdp->sd_vfs, no_addr, 1);
 591         else
 592                 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
 593         if (inode && !IS_ERR(inode)) {
 594                 d_prune_aliases(inode);
 595                 iput(inode);
 596         }
 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 = NULL;
 661         struct address_space *mapping;
 662         struct kmem_cache *cachep;
 663         int ret, tries = 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 = ktime_set(0, 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         ret = rhashtable_lookup_insert_fast(&gl_hash_table, &gl->gl_node,
 729                                             ht_parms);
 730         if (ret == 0) {
 731                 *glp = gl;
 732                 return 0;
 733         }
 734
 735         if (ret == -EEXIST) {
 736                 ret = 0;
 737                 rcu_read_lock();
 738                 tmp = rhashtable_lookup_fast(&gl_hash_table, &name, ht_parms);
 739                 if (tmp == NULL || !lockref_get_not_dead(&tmp->gl_lockref)) {
 740                         if (++tries < 100) {
 741                                 rcu_read_unlock();
 742                                 cond_resched();
 743                                 goto again;
 744                         }
 745                         tmp = NULL;
 746                         ret = -ENOMEM;
 747                 }
 748                 rcu_read_unlock();
 749         } else {
 750                 WARN_ON_ONCE(ret);
 751         }
 752         kfree(gl->gl_lksb.sb_lvbptr);
 753         kmem_cache_free(cachep, gl);
 754         if (atomic_dec_and_test(&sdp->sd_glock_disposal))
 755                 wake_up(&sdp->sd_glock_wait);
 756         *glp = tmp;
 757
 758         return ret;
 759 }
 760
 761 /**
 762  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
 763  * @gl: the glock
 764  * @state: the state we're requesting
 765  * @flags: the modifier flags
 766  * @gh: the holder structure
 767  *
 768  */
 769
 770 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
 771                       struct gfs2_holder *gh)
 772 {
 773         INIT_LIST_HEAD(&gh->gh_list);
 774         gh->gh_gl = gl;
 775         gh->gh_ip = _RET_IP_;
 776         gh->gh_owner_pid = get_pid(task_pid(current));
 777         gh->gh_state = state;
 778         gh->gh_flags = flags;
 779         gh->gh_error = 0;
 780         gh->gh_iflags = 0;
 781         gfs2_glock_hold(gl);
 782 }
 783
 784 /**
 785  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
 786  * @state: the state we're requesting
 787  * @flags: the modifier flags
 788  * @gh: the holder structure
 789  *
 790  * Don't mess with the glock.
 791  *
 792  */
 793
 794 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
 795 {
 796         gh->gh_state = state;
 797         gh->gh_flags = flags;
 798         gh->gh_iflags = 0;
 799         gh->gh_ip = _RET_IP_;
 800         put_pid(gh->gh_owner_pid);
 801         gh->gh_owner_pid = get_pid(task_pid(current));
 802 }
 803
 804 /**
 805  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
 806  * @gh: the holder structure
 807  *
 808  */
 809
 810 void gfs2_holder_uninit(struct gfs2_holder *gh)
 811 {
 812         put_pid(gh->gh_owner_pid);
 813         gfs2_glock_put(gh->gh_gl);
 814         gh->gh_gl = NULL;
 815         gh->gh_ip = 0;
 816 }
 817
 818 /**
 819  * gfs2_glock_wait - wait on a glock acquisition
 820  * @gh: the glock holder
 821  *
 822  * Returns: 0 on success
 823  */
 824
 825 int gfs2_glock_wait(struct gfs2_holder *gh)
 826 {
 827         unsigned long time1 = jiffies;
 828
 829         might_sleep();
 830         wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
 831         if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
 832                 /* Lengthen the minimum hold time. */
 833                 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
 834                                               GL_GLOCK_HOLD_INCR,
 835                                               GL_GLOCK_MAX_HOLD);
 836         return gh->gh_error;
 837 }
 838
 839 /**
 840  * handle_callback - process a demote request
 841  * @gl: the glock
 842  * @state: the state the caller wants us to change to
 843  *
 844  * There are only two requests that we are going to see in actual
 845  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
 846  */
 847
 848 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
 849                             unsigned long delay, bool remote)
 850 {
 851         int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
 852
 853         set_bit(bit, &gl->gl_flags);
 854         if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
 855                 gl->gl_demote_state = state;
 856                 gl->gl_demote_time = jiffies;
 857         } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
 858                         gl->gl_demote_state != state) {
 859                 gl->gl_demote_state = LM_ST_UNLOCKED;
 860         }
 861         if (gl->gl_ops->go_callback)
 862                 gl->gl_ops->go_callback(gl, remote);
 863         trace_gfs2_demote_rq(gl, remote);
 864 }
 865
 866 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
 867 {
 868         struct va_format vaf;
 869         va_list args;
 870
 871         va_start(args, fmt);
 872
 873         if (seq) {
 874                 seq_vprintf(seq, fmt, args);
 875         } else {
 876                 vaf.fmt = fmt;
 877                 vaf.va = &args;
 878
 879                 pr_err("%pV", &vaf);
 880         }
 881
 882         va_end(args);
 883 }
 884
 885 /**
 886  * add_to_queue - Add a holder to the wait queue (but look for recursion)
 887  * @gh: the holder structure to add
 888  *
 889  * Eventually we should move the recursive locking trap to a
 890  * debugging option or something like that. This is the fast
 891  * path and needs to have the minimum number of distractions.
 892  *
 893  */
 894
 895 static inline void add_to_queue(struct gfs2_holder *gh)
 896 __releases(&gl->gl_lockref.lock)
 897 __acquires(&gl->gl_lockref.lock)
 898 {
 899         struct gfs2_glock *gl = gh->gh_gl;
 900         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 901         struct list_head *insert_pt = NULL;
 902         struct gfs2_holder *gh2;
 903         int try_futile = 0;
 904
 905         BUG_ON(gh->gh_owner_pid == NULL);
 906         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
 907                 BUG();
 908
 909         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
 910                 if (test_bit(GLF_LOCK, &gl->gl_flags))
 911                         try_futile = !may_grant(gl, gh);
 912                 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 913                         goto fail;
 914         }
 915
 916         list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
 917                 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
 918                     (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
 919                         goto trap_recursive;
 920                 if (try_futile &&
 921                     !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
 922 fail:
 923                         gh->gh_error = GLR_TRYFAILED;
 924                         gfs2_holder_wake(gh);
 925                         return;
 926                 }
 927                 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
 928                         continue;
 929                 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
 930                         insert_pt = &gh2->gh_list;
 931         }
 932         set_bit(GLF_QUEUED, &gl->gl_flags);
 933         trace_gfs2_glock_queue(gh, 1);
 934         gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
 935         gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
 936         if (likely(insert_pt == NULL)) {
 937                 list_add_tail(&gh->gh_list, &gl->gl_holders);
 938                 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
 939                         goto do_cancel;
 940                 return;
 941         }
 942         list_add_tail(&gh->gh_list, insert_pt);
 943 do_cancel:
 944         gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
 945         if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
 946                 spin_unlock(&gl->gl_lockref.lock);
 947                 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
 948                         sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
 949                 spin_lock(&gl->gl_lockref.lock);
 950         }
 951         return;
 952
 953 trap_recursive:
 954         pr_err("original: %pSR\n", (void *)gh2->gh_ip);
 955         pr_err("pid: %d\n", pid_nr(gh2->gh_owner_pid));
 956         pr_err("lock type: %d req lock state : %d\n",
 957                gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
 958         pr_err("new: %pSR\n", (void *)gh->gh_ip);
 959         pr_err("pid: %d\n", pid_nr(gh->gh_owner_pid));
 960         pr_err("lock type: %d req lock state : %d\n",
 961                gh->gh_gl->gl_name.ln_type, gh->gh_state);
 962         gfs2_dump_glock(NULL, gl);
 963         BUG();
 964 }
 965
 966 /**
 967  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
 968  * @gh: the holder structure
 969  *
 970  * if (gh->gh_flags & GL_ASYNC), this never returns an error
 971  *
 972  * Returns: 0, GLR_TRYFAILED, or errno on failure
 973  */
 974
 975 int gfs2_glock_nq(struct gfs2_holder *gh)
 976 {
 977         struct gfs2_glock *gl = gh->gh_gl;
 978         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 979         int error = 0;
 980
 981         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 982                 return -EIO;
 983
 984         if (test_bit(GLF_LRU, &gl->gl_flags))
 985                 gfs2_glock_remove_from_lru(gl);
 986
 987         spin_lock(&gl->gl_lockref.lock);
 988         add_to_queue(gh);
 989         if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
 990                      test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
 991                 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
 992                 gl->gl_lockref.count++;
 993                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 994                         gl->gl_lockref.count--;
 995         }
 996         run_queue(gl, 1);
 997         spin_unlock(&gl->gl_lockref.lock);
 998
 999         if (!(gh->gh_flags & GL_ASYNC))
1000                 error = gfs2_glock_wait(gh);
1001
1002         return error;
1003 }
1004
1005 /**
1006  * gfs2_glock_poll - poll to see if an async request has been completed
1007  * @gh: the holder
1008  *
1009  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1010  */
1011
1012 int gfs2_glock_poll(struct gfs2_holder *gh)
1013 {
1014         return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1015 }
1016
1017 /**
1018  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1019  * @gh: the glock holder
1020  *
1021  */
1022
1023 void gfs2_glock_dq(struct gfs2_holder *gh)
1024 {
1025         struct gfs2_glock *gl = gh->gh_gl;
1026         const struct gfs2_glock_operations *glops = gl->gl_ops;
1027         unsigned delay = 0;
1028         int fast_path = 0;
1029
1030         spin_lock(&gl->gl_lockref.lock);
1031         if (gh->gh_flags & GL_NOCACHE)
1032                 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1033
1034         list_del_init(&gh->gh_list);
1035         if (find_first_holder(gl) == NULL) {
1036                 if (glops->go_unlock) {
1037                         GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1038                         spin_unlock(&gl->gl_lockref.lock);
1039                         glops->go_unlock(gh);
1040                         spin_lock(&gl->gl_lockref.lock);
1041                         clear_bit(GLF_LOCK, &gl->gl_flags);
1042                 }
1043                 if (list_empty(&gl->gl_holders) &&
1044                     !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1045                     !test_bit(GLF_DEMOTE, &gl->gl_flags))
1046                         fast_path = 1;
1047         }
1048         if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1049                 gfs2_glock_add_to_lru(gl);
1050
1051         trace_gfs2_glock_queue(gh, 0);
1052         spin_unlock(&gl->gl_lockref.lock);
1053         if (likely(fast_path))
1054                 return;
1055
1056         gfs2_glock_hold(gl);
1057         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1058             !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1059             gl->gl_name.ln_type == LM_TYPE_INODE)
1060                 delay = gl->gl_hold_time;
1061         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1062                 gfs2_glock_put(gl);
1063 }
1064
1065 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1066 {
1067         struct gfs2_glock *gl = gh->gh_gl;
1068         gfs2_glock_dq(gh);
1069         might_sleep();
1070         wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1071 }
1072
1073 /**
1074  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1075  * @gh: the holder structure
1076  *
1077  */
1078
1079 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1080 {
1081         gfs2_glock_dq(gh);
1082         gfs2_holder_uninit(gh);
1083 }
1084
1085 /**
1086  * gfs2_glock_nq_num - acquire a glock based on lock number
1087  * @sdp: the filesystem
1088  * @number: the lock number
1089  * @glops: the glock operations for the type of glock
1090  * @state: the state to acquire the glock in
1091  * @flags: modifier flags for the acquisition
1092  * @gh: the struct gfs2_holder
1093  *
1094  * Returns: errno
1095  */
1096
1097 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1098                       const struct gfs2_glock_operations *glops,
1099                       unsigned int state, int flags, struct gfs2_holder *gh)
1100 {
1101         struct gfs2_glock *gl;
1102         int error;
1103
1104         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1105         if (!error) {
1106                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1107                 gfs2_glock_put(gl);
1108         }
1109
1110         return error;
1111 }
1112
1113 /**
1114  * glock_compare - Compare two struct gfs2_glock structures for sorting
1115  * @arg_a: the first structure
1116  * @arg_b: the second structure
1117  *
1118  */
1119
1120 static int glock_compare(const void *arg_a, const void *arg_b)
1121 {
1122         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1123         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1124         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1125         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1126
1127         if (a->ln_number > b->ln_number)
1128                 return 1;
1129         if (a->ln_number < b->ln_number)
1130                 return -1;
1131         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1132         return 0;
1133 }
1134
1135 /**
1136  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1137  * @num_gh: the number of structures
1138  * @ghs: an array of struct gfs2_holder structures
1139  *
1140  * Returns: 0 on success (all glocks acquired),
1141  *          errno on failure (no glocks acquired)
1142  */
1143
1144 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1145                      struct gfs2_holder **p)
1146 {
1147         unsigned int x;
1148         int error = 0;
1149
1150         for (x = 0; x < num_gh; x++)
1151                 p[x] = &ghs[x];
1152
1153         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1154
1155         for (x = 0; x < num_gh; x++) {
1156                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1157
1158                 error = gfs2_glock_nq(p[x]);
1159                 if (error) {
1160                         while (x--)
1161                                 gfs2_glock_dq(p[x]);
1162                         break;
1163                 }
1164         }
1165
1166         return error;
1167 }
1168
1169 /**
1170  * gfs2_glock_nq_m - acquire multiple glocks
1171  * @num_gh: the number of structures
1172  * @ghs: an array of struct gfs2_holder structures
1173  *
1174  *
1175  * Returns: 0 on success (all glocks acquired),
1176  *          errno on failure (no glocks acquired)
1177  */
1178
1179 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1180 {
1181         struct gfs2_holder *tmp[4];
1182         struct gfs2_holder **pph = tmp;
1183         int error = 0;
1184
1185         switch(num_gh) {
1186         case 0:
1187                 return 0;
1188         case 1:
1189                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1190                 return gfs2_glock_nq(ghs);
1191         default:
1192                 if (num_gh <= 4)
1193                         break;
1194                 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1195                 if (!pph)
1196                         return -ENOMEM;
1197         }
1198
1199         error = nq_m_sync(num_gh, ghs, pph);
1200
1201         if (pph != tmp)
1202                 kfree(pph);
1203
1204         return error;
1205 }
1206
1207 /**
1208  * gfs2_glock_dq_m - release multiple glocks
1209  * @num_gh: the number of structures
1210  * @ghs: an array of struct gfs2_holder structures
1211  *
1212  */
1213
1214 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1215 {
1216         while (num_gh--)
1217                 gfs2_glock_dq(&ghs[num_gh]);
1218 }
1219
1220 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1221 {
1222         unsigned long delay = 0;
1223         unsigned long holdtime;
1224         unsigned long now = jiffies;
1225
1226         gfs2_glock_hold(gl);
1227         holdtime = gl->gl_tchange + gl->gl_hold_time;
1228         if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1229             gl->gl_name.ln_type == LM_TYPE_INODE) {
1230                 if (time_before(now, holdtime))
1231                         delay = holdtime - now;
1232                 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1233                         delay = gl->gl_hold_time;
1234         }
1235
1236         spin_lock(&gl->gl_lockref.lock);
1237         handle_callback(gl, state, delay, true);
1238         spin_unlock(&gl->gl_lockref.lock);
1239         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1240                 gfs2_glock_put(gl);
1241 }
1242
1243 /**
1244  * gfs2_should_freeze - Figure out if glock should be frozen
1245  * @gl: The glock in question
1246  *
1247  * Glocks are not frozen if (a) the result of the dlm operation is
1248  * an error, (b) the locking operation was an unlock operation or
1249  * (c) if there is a "noexp" flagged request anywhere in the queue
1250  *
1251  * Returns: 1 if freezing should occur, 0 otherwise
1252  */
1253
1254 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1255 {
1256         const struct gfs2_holder *gh;
1257
1258         if (gl->gl_reply & ~LM_OUT_ST_MASK)
1259                 return 0;
1260         if (gl->gl_target == LM_ST_UNLOCKED)
1261                 return 0;
1262
1263         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1264                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1265                         continue;
1266                 if (LM_FLAG_NOEXP & gh->gh_flags)
1267                         return 0;
1268         }
1269
1270         return 1;
1271 }
1272
1273 /**
1274  * gfs2_glock_complete - Callback used by locking
1275  * @gl: Pointer to the glock
1276  * @ret: The return value from the dlm
1277  *
1278  * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1279  * to use a bitfield shared with other glock state fields.
1280  */
1281
1282 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1283 {
1284         struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1285
1286         spin_lock(&gl->gl_lockref.lock);
1287         gl->gl_reply = ret;
1288
1289         if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1290                 if (gfs2_should_freeze(gl)) {
1291                         set_bit(GLF_FROZEN, &gl->gl_flags);
1292                         spin_unlock(&gl->gl_lockref.lock);
1293                         return;
1294                 }
1295         }
1296
1297         gl->gl_lockref.count++;
1298         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1299         spin_unlock(&gl->gl_lockref.lock);
1300
1301         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1302                 gfs2_glock_put(gl);
1303 }
1304
1305 static int glock_cmp(void *priv, struct list_head *a, struct list_head *b)
1306 {
1307         struct gfs2_glock *gla, *glb;
1308
1309         gla = list_entry(a, struct gfs2_glock, gl_lru);
1310         glb = list_entry(b, struct gfs2_glock, gl_lru);
1311
1312         if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1313                 return 1;
1314         if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1315                 return -1;
1316
1317         return 0;
1318 }
1319
1320 /**
1321  * gfs2_dispose_glock_lru - Demote a list of glocks
1322  * @list: The list to dispose of
1323  *
1324  * Disposing of glocks may involve disk accesses, so that here we sort
1325  * the glocks by number (i.e. disk location of the inodes) so that if
1326  * there are any such accesses, they'll be sent in order (mostly).
1327  *
1328  * Must be called under the lru_lock, but may drop and retake this
1329  * lock. While the lru_lock is dropped, entries may vanish from the
1330  * list, but no new entries will appear on the list (since it is
1331  * private)
1332  */
1333
1334 static void gfs2_dispose_glock_lru(struct list_head *list)
1335 __releases(&lru_lock)
1336 __acquires(&lru_lock)
1337 {
1338         struct gfs2_glock *gl;
1339
1340         list_sort(NULL, list, glock_cmp);
1341
1342         while(!list_empty(list)) {
1343                 gl = list_entry(list->next, struct gfs2_glock, gl_lru);
1344                 list_del_init(&gl->gl_lru);
1345                 clear_bit(GLF_LRU, &gl->gl_flags);
1346                 if (!spin_trylock(&gl->gl_lockref.lock)) {
1347 add_back_to_lru:
1348                         list_add(&gl->gl_lru, &lru_list);
1349                         set_bit(GLF_LRU, &gl->gl_flags);
1350                         atomic_inc(&lru_count);
1351                         continue;
1352                 }
1353                 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1354                         spin_unlock(&gl->gl_lockref.lock);
1355                         goto add_back_to_lru;
1356                 }
1357                 gl->gl_lockref.count++;
1358                 if (demote_ok(gl))
1359                         handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1360                 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1361                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1362                         gl->gl_lockref.count--;
1363                 spin_unlock(&gl->gl_lockref.lock);
1364                 cond_resched_lock(&lru_lock);
1365         }
1366 }
1367
1368 /**
1369  * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1370  * @nr: The number of entries to scan
1371  *
1372  * This function selects the entries on the LRU which are able to
1373  * be demoted, and then kicks off the process by calling
1374  * gfs2_dispose_glock_lru() above.
1375  */
1376
1377 static long gfs2_scan_glock_lru(int nr)
1378 {
1379         struct gfs2_glock *gl;
1380         LIST_HEAD(skipped);
1381         LIST_HEAD(dispose);
1382         long freed = 0;
1383
1384         spin_lock(&lru_lock);
1385         while ((nr-- >= 0) && !list_empty(&lru_list)) {
1386                 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1387
1388                 /* Test for being demotable */
1389                 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1390                         list_move(&gl->gl_lru, &dispose);
1391                         atomic_dec(&lru_count);
1392                         freed++;
1393                         continue;
1394                 }
1395
1396                 list_move(&gl->gl_lru, &skipped);
1397         }
1398         list_splice(&skipped, &lru_list);
1399         if (!list_empty(&dispose))
1400                 gfs2_dispose_glock_lru(&dispose);
1401         spin_unlock(&lru_lock);
1402
1403         return freed;
1404 }
1405
1406 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1407                                             struct shrink_control *sc)
1408 {
1409         if (!(sc->gfp_mask & __GFP_FS))
1410                 return SHRINK_STOP;
1411         return gfs2_scan_glock_lru(sc->nr_to_scan);
1412 }
1413
1414 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1415                                              struct shrink_control *sc)
1416 {
1417         return vfs_pressure_ratio(atomic_read(&lru_count));
1418 }
1419
1420 static struct shrinker glock_shrinker = {
1421         .seeks = DEFAULT_SEEKS,
1422         .count_objects = gfs2_glock_shrink_count,
1423         .scan_objects = gfs2_glock_shrink_scan,
1424 };
1425
1426 /**
1427  * examine_bucket - Call a function for glock in a hash bucket
1428  * @examiner: the function
1429  * @sdp: the filesystem
1430  * @bucket: the bucket
1431  *
1432  */
1433
1434 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1435 {
1436         struct gfs2_glock *gl;
1437         struct rhash_head *pos, *next;
1438         const struct bucket_table *tbl;
1439         int i;
1440
1441         rcu_read_lock();
1442         tbl = rht_dereference_rcu(gl_hash_table.tbl, &gl_hash_table);
1443         for (i = 0; i < tbl->size; i++) {
1444                 rht_for_each_entry_safe(gl, pos, next, tbl, i, gl_node) {
1445                         if ((gl->gl_name.ln_sbd == sdp) &&
1446                             lockref_get_not_dead(&gl->gl_lockref))
1447                                 examiner(gl);
1448                 }
1449         }
1450         rcu_read_unlock();
1451         cond_resched();
1452 }
1453
1454 /**
1455  * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1456  * @gl: The glock to thaw
1457  *
1458  */
1459
1460 static void thaw_glock(struct gfs2_glock *gl)
1461 {
1462         if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1463                 goto out;
1464         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1465         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0) {
1466 out:
1467                 gfs2_glock_put(gl);
1468         }
1469 }
1470
1471 /**
1472  * clear_glock - look at a glock and see if we can free it from glock cache
1473  * @gl: the glock to look at
1474  *
1475  */
1476
1477 static void clear_glock(struct gfs2_glock *gl)
1478 {
1479         gfs2_glock_remove_from_lru(gl);
1480
1481         spin_lock(&gl->gl_lockref.lock);
1482         if (gl->gl_state != LM_ST_UNLOCKED)
1483                 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1484         spin_unlock(&gl->gl_lockref.lock);
1485         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1486                 gfs2_glock_put(gl);
1487 }
1488
1489 /**
1490  * gfs2_glock_thaw - Thaw any frozen glocks
1491  * @sdp: The super block
1492  *
1493  */
1494
1495 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1496 {
1497         glock_hash_walk(thaw_glock, sdp);
1498 }
1499
1500 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1501 {
1502         spin_lock(&gl->gl_lockref.lock);
1503         gfs2_dump_glock(seq, gl);
1504         spin_unlock(&gl->gl_lockref.lock);
1505 }
1506
1507 static void dump_glock_func(struct gfs2_glock *gl)
1508 {
1509         dump_glock(NULL, gl);
1510 }
1511
1512 /**
1513  * gfs2_gl_hash_clear - Empty out the glock hash table
1514  * @sdp: the filesystem
1515  * @wait: wait until it's all gone
1516  *
1517  * Called when unmounting the filesystem.
1518  */
1519
1520 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1521 {
1522         set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1523         flush_workqueue(glock_workqueue);
1524         glock_hash_walk(clear_glock, sdp);
1525         flush_workqueue(glock_workqueue);
1526         wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
1527         glock_hash_walk(dump_glock_func, sdp);
1528 }
1529
1530 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1531 {
1532         struct gfs2_glock *gl = ip->i_gl;
1533         int ret;
1534
1535         ret = gfs2_truncatei_resume(ip);
1536         gfs2_assert_withdraw(gl->gl_name.ln_sbd, ret == 0);
1537
1538         spin_lock(&gl->gl_lockref.lock);
1539         clear_bit(GLF_LOCK, &gl->gl_flags);
1540         run_queue(gl, 1);
1541         spin_unlock(&gl->gl_lockref.lock);
1542 }
1543
1544 static const char *state2str(unsigned state)
1545 {
1546         switch(state) {
1547         case LM_ST_UNLOCKED:
1548                 return "UN";
1549         case LM_ST_SHARED:
1550                 return "SH";
1551         case LM_ST_DEFERRED:
1552                 return "DF";
1553         case LM_ST_EXCLUSIVE:
1554                 return "EX";
1555         }
1556         return "??";
1557 }
1558
1559 static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1560 {
1561         char *p = buf;
1562         if (flags & LM_FLAG_TRY)
1563                 *p++ = 't';
1564         if (flags & LM_FLAG_TRY_1CB)
1565                 *p++ = 'T';
1566         if (flags & LM_FLAG_NOEXP)
1567                 *p++ = 'e';
1568         if (flags & LM_FLAG_ANY)
1569                 *p++ = 'A';
1570         if (flags & LM_FLAG_PRIORITY)
1571                 *p++ = 'p';
1572         if (flags & GL_ASYNC)
1573                 *p++ = 'a';
1574         if (flags & GL_EXACT)
1575                 *p++ = 'E';
1576         if (flags & GL_NOCACHE)
1577                 *p++ = 'c';
1578         if (test_bit(HIF_HOLDER, &iflags))
1579                 *p++ = 'H';
1580         if (test_bit(HIF_WAIT, &iflags))
1581                 *p++ = 'W';
1582         if (test_bit(HIF_FIRST, &iflags))
1583                 *p++ = 'F';
1584         *p = 0;
1585         return buf;
1586 }
1587
1588 /**
1589  * dump_holder - print information about a glock holder
1590  * @seq: the seq_file struct
1591  * @gh: the glock holder
1592  *
1593  */
1594
1595 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1596 {
1597         struct task_struct *gh_owner = NULL;
1598         char flags_buf[32];
1599
1600         rcu_read_lock();
1601         if (gh->gh_owner_pid)
1602                 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1603         gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1604                        state2str(gh->gh_state),
1605                        hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1606                        gh->gh_error,
1607                        gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1608                        gh_owner ? gh_owner->comm : "(ended)",
1609                        (void *)gh->gh_ip);
1610         rcu_read_unlock();
1611 }
1612
1613 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1614 {
1615         const unsigned long *gflags = &gl->gl_flags;
1616         char *p = buf;
1617
1618         if (test_bit(GLF_LOCK, gflags))
1619                 *p++ = 'l';
1620         if (test_bit(GLF_DEMOTE, gflags))
1621                 *p++ = 'D';
1622         if (test_bit(GLF_PENDING_DEMOTE, gflags))
1623                 *p++ = 'd';
1624         if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1625                 *p++ = 'p';
1626         if (test_bit(GLF_DIRTY, gflags))
1627                 *p++ = 'y';
1628         if (test_bit(GLF_LFLUSH, gflags))
1629                 *p++ = 'f';
1630         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1631                 *p++ = 'i';
1632         if (test_bit(GLF_REPLY_PENDING, gflags))
1633                 *p++ = 'r';
1634         if (test_bit(GLF_INITIAL, gflags))
1635                 *p++ = 'I';
1636         if (test_bit(GLF_FROZEN, gflags))
1637                 *p++ = 'F';
1638         if (test_bit(GLF_QUEUED, gflags))
1639                 *p++ = 'q';
1640         if (test_bit(GLF_LRU, gflags))
1641                 *p++ = 'L';
1642         if (gl->gl_object)
1643                 *p++ = 'o';
1644         if (test_bit(GLF_BLOCKING, gflags))
1645                 *p++ = 'b';
1646         *p = 0;
1647         return buf;
1648 }
1649
1650 /**
1651  * gfs2_dump_glock - print information about a glock
1652  * @seq: The seq_file struct
1653  * @gl: the glock
1654  *
1655  * The file format is as follows:
1656  * One line per object, capital letters are used to indicate objects
1657  * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1658  * other objects are indented by a single space and follow the glock to
1659  * which they are related. Fields are indicated by lower case letters
1660  * followed by a colon and the field value, except for strings which are in
1661  * [] so that its possible to see if they are composed of spaces for
1662  * example. The field's are n = number (id of the object), f = flags,
1663  * t = type, s = state, r = refcount, e = error, p = pid.
1664  *
1665  */
1666
1667 void gfs2_dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1668 {
1669         const struct gfs2_glock_operations *glops = gl->gl_ops;
1670         unsigned long long dtime;
1671         const struct gfs2_holder *gh;
1672         char gflags_buf[32];
1673
1674         dtime = jiffies - gl->gl_demote_time;
1675         dtime *= 1000000/HZ; /* demote time in uSec */
1676         if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1677                 dtime = 0;
1678         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",
1679                   state2str(gl->gl_state),
1680                   gl->gl_name.ln_type,
1681                   (unsigned long long)gl->gl_name.ln_number,
1682                   gflags2str(gflags_buf, gl),
1683                   state2str(gl->gl_target),
1684                   state2str(gl->gl_demote_state), dtime,
1685                   atomic_read(&gl->gl_ail_count),
1686                   atomic_read(&gl->gl_revokes),
1687                   (int)gl->gl_lockref.count, gl->gl_hold_time);
1688
1689         list_for_each_entry(gh, &gl->gl_holders, gh_list)
1690                 dump_holder(seq, gh);
1691
1692         if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1693                 glops->go_dump(seq, gl);
1694 }
1695
1696 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
1697 {
1698         struct gfs2_glock *gl = iter_ptr;
1699
1700         seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
1701                    gl->gl_name.ln_type,
1702                    (unsigned long long)gl->gl_name.ln_number,
1703                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
1704                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
1705                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
1706                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
1707                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
1708                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
1709                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
1710                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
1711         return 0;
1712 }
1713
1714 static const char *gfs2_gltype[] = {
1715         "type",
1716         "reserved",
1717         "nondisk",
1718         "inode",
1719         "rgrp",
1720         "meta",
1721         "iopen",
1722         "flock",
1723         "plock",
1724         "quota",
1725         "journal",
1726 };
1727
1728 static const char *gfs2_stype[] = {
1729         [GFS2_LKS_SRTT]         = "srtt",
1730         [GFS2_LKS_SRTTVAR]      = "srttvar",
1731         [GFS2_LKS_SRTTB]        = "srttb",
1732         [GFS2_LKS_SRTTVARB]     = "srttvarb",
1733         [GFS2_LKS_SIRT]         = "sirt",
1734         [GFS2_LKS_SIRTVAR]      = "sirtvar",
1735         [GFS2_LKS_DCOUNT]       = "dlm",
1736         [GFS2_LKS_QCOUNT]       = "queue",
1737 };
1738
1739 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1740
1741 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
1742 {
1743         struct gfs2_sbd *sdp = seq->private;
1744         loff_t pos = *(loff_t *)iter_ptr;
1745         unsigned index = pos >> 3;
1746         unsigned subindex = pos & 0x07;
1747         int i;
1748
1749         if (index == 0 && subindex != 0)
1750                 return 0;
1751
1752         seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
1753                    (index == 0) ? "cpu": gfs2_stype[subindex]);
1754
1755         for_each_possible_cpu(i) {
1756                 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
1757
1758                 if (index == 0)
1759                         seq_printf(seq, " %15u", i);
1760                 else
1761                         seq_printf(seq, " %15llu", (unsigned long long)lkstats->
1762                                    lkstats[index - 1].stats[subindex]);
1763         }
1764         seq_putc(seq, '\n');
1765         return 0;
1766 }
1767
1768 int __init gfs2_glock_init(void)
1769 {
1770         int ret;
1771
1772         ret = rhashtable_init(&gl_hash_table, &ht_parms);
1773         if (ret < 0)
1774                 return ret;
1775
1776         glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1777                                           WQ_HIGHPRI | WQ_FREEZABLE, 0);
1778         if (!glock_workqueue) {
1779                 rhashtable_destroy(&gl_hash_table);
1780                 return -ENOMEM;
1781         }
1782         gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1783                                                 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1784                                                 0);
1785         if (!gfs2_delete_workqueue) {
1786                 destroy_workqueue(glock_workqueue);
1787                 rhashtable_destroy(&gl_hash_table);
1788                 return -ENOMEM;
1789         }
1790
1791         register_shrinker(&glock_shrinker);
1792
1793         return 0;
1794 }
1795
1796 void gfs2_glock_exit(void)
1797 {
1798         unregister_shrinker(&glock_shrinker);
1799         rhashtable_destroy(&gl_hash_table);
1800         destroy_workqueue(glock_workqueue);
1801         destroy_workqueue(gfs2_delete_workqueue);
1802 }
1803
1804 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1805 {
1806         while ((gi->gl = rhashtable_walk_next(&gi->hti))) {
1807                 if (IS_ERR(gi->gl)) {
1808                         if (PTR_ERR(gi->gl) == -EAGAIN)
1809                                 continue;
1810                         gi->gl = NULL;
1811                         return;
1812                 }
1813                 /* Skip entries for other sb and dead entries */
1814                 if (gi->sdp == gi->gl->gl_name.ln_sbd &&
1815                     !__lockref_is_dead(&gi->gl->gl_lockref))
1816                         return;
1817         }
1818 }
1819
1820 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1821 {
1822         struct gfs2_glock_iter *gi = seq->private;
1823         loff_t n = *pos;
1824
1825         if (rhashtable_walk_init(&gl_hash_table, &gi->hti) != 0)
1826                 return NULL;
1827         if (rhashtable_walk_start(&gi->hti) != 0)
1828                 return NULL;
1829
1830         do {
1831                 gfs2_glock_iter_next(gi);
1832         } while (gi->gl && n--);
1833
1834         gi->last_pos = *pos;
1835
1836         return gi->gl;
1837 }
1838
1839 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1840                                  loff_t *pos)
1841 {
1842         struct gfs2_glock_iter *gi = seq->private;
1843
1844         (*pos)++;
1845         gi->last_pos = *pos;
1846         gfs2_glock_iter_next(gi);
1847
1848         return gi->gl;
1849 }
1850
1851 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1852 {
1853         struct gfs2_glock_iter *gi = seq->private;
1854
1855         gi->gl = NULL;
1856         if (gi->hti.walker) {
1857                 rhashtable_walk_stop(&gi->hti);
1858                 rhashtable_walk_exit(&gi->hti);
1859         }
1860 }
1861
1862 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1863 {
1864         dump_glock(seq, iter_ptr);
1865         return 0;
1866 }
1867
1868 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
1869 {
1870         preempt_disable();
1871         if (*pos >= GFS2_NR_SBSTATS)
1872                 return NULL;
1873         return pos;
1874 }
1875
1876 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
1877                                    loff_t *pos)
1878 {
1879         (*pos)++;
1880         if (*pos >= GFS2_NR_SBSTATS)
1881                 return NULL;
1882         return pos;
1883 }
1884
1885 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
1886 {
1887         preempt_enable();
1888 }
1889
1890 static const struct seq_operations gfs2_glock_seq_ops = {
1891         .start = gfs2_glock_seq_start,
1892         .next  = gfs2_glock_seq_next,
1893         .stop  = gfs2_glock_seq_stop,
1894         .show  = gfs2_glock_seq_show,
1895 };
1896
1897 static const struct seq_operations gfs2_glstats_seq_ops = {
1898         .start = gfs2_glock_seq_start,
1899         .next  = gfs2_glock_seq_next,
1900         .stop  = gfs2_glock_seq_stop,
1901         .show  = gfs2_glstats_seq_show,
1902 };
1903
1904 static const struct seq_operations gfs2_sbstats_seq_ops = {
1905         .start = gfs2_sbstats_seq_start,
1906         .next  = gfs2_sbstats_seq_next,
1907         .stop  = gfs2_sbstats_seq_stop,
1908         .show  = gfs2_sbstats_seq_show,
1909 };
1910
1911 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
1912
1913 static int gfs2_glocks_open(struct inode *inode, struct file *file)
1914 {
1915         int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1916                                    sizeof(struct gfs2_glock_iter));
1917         if (ret == 0) {
1918                 struct seq_file *seq = file->private_data;
1919                 struct gfs2_glock_iter *gi = seq->private;
1920
1921                 gi->sdp = inode->i_private;
1922                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
1923                 if (seq->buf)
1924                         seq->size = GFS2_SEQ_GOODSIZE;
1925                 gi->gl = NULL;
1926         }
1927         return ret;
1928 }
1929
1930 static int gfs2_glocks_release(struct inode *inode, struct file *file)
1931 {
1932         struct seq_file *seq = file->private_data;
1933         struct gfs2_glock_iter *gi = seq->private;
1934
1935         gi->gl = NULL;
1936         return seq_release_private(inode, file);
1937 }
1938
1939 static int gfs2_glstats_open(struct inode *inode, struct file *file)
1940 {
1941         int ret = seq_open_private(file, &gfs2_glstats_seq_ops,
1942                                    sizeof(struct gfs2_glock_iter));
1943         if (ret == 0) {
1944                 struct seq_file *seq = file->private_data;
1945                 struct gfs2_glock_iter *gi = seq->private;
1946                 gi->sdp = inode->i_private;
1947                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
1948                 if (seq->buf)
1949                         seq->size = GFS2_SEQ_GOODSIZE;
1950                 gi->gl = NULL;
1951         }
1952         return ret;
1953 }
1954
1955 static int gfs2_sbstats_open(struct inode *inode, struct file *file)
1956 {
1957         int ret = seq_open(file, &gfs2_sbstats_seq_ops);
1958         if (ret == 0) {
1959                 struct seq_file *seq = file->private_data;
1960                 seq->private = inode->i_private;  /* sdp */
1961         }
1962         return ret;
1963 }
1964
1965 static const struct file_operations gfs2_glocks_fops = {
1966         .owner   = THIS_MODULE,
1967         .open    = gfs2_glocks_open,
1968         .read    = seq_read,
1969         .llseek  = seq_lseek,
1970         .release = gfs2_glocks_release,
1971 };
1972
1973 static const struct file_operations gfs2_glstats_fops = {
1974         .owner   = THIS_MODULE,
1975         .open    = gfs2_glstats_open,
1976         .read    = seq_read,
1977         .llseek  = seq_lseek,
1978         .release = gfs2_glocks_release,
1979 };
1980
1981 static const struct file_operations gfs2_sbstats_fops = {
1982         .owner   = THIS_MODULE,
1983         .open    = gfs2_sbstats_open,
1984         .read    = seq_read,
1985         .llseek  = seq_lseek,
1986         .release = seq_release,
1987 };
1988
1989 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
1990 {
1991         struct dentry *dent;
1992
1993         dent = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
1994         if (IS_ERR_OR_NULL(dent))
1995                 goto fail;
1996         sdp->debugfs_dir = dent;
1997
1998         dent = debugfs_create_file("glocks",
1999                                    S_IFREG | S_IRUGO,
2000                                    sdp->debugfs_dir, sdp,
2001                                    &gfs2_glocks_fops);
2002         if (IS_ERR_OR_NULL(dent))
2003                 goto fail;
2004         sdp->debugfs_dentry_glocks = dent;
2005
2006         dent = debugfs_create_file("glstats",
2007                                    S_IFREG | S_IRUGO,
2008                                    sdp->debugfs_dir, sdp,
2009                                    &gfs2_glstats_fops);
2010         if (IS_ERR_OR_NULL(dent))
2011                 goto fail;
2012         sdp->debugfs_dentry_glstats = dent;
2013
2014         dent = debugfs_create_file("sbstats",
2015                                    S_IFREG | S_IRUGO,
2016                                    sdp->debugfs_dir, sdp,
2017                                    &gfs2_sbstats_fops);
2018         if (IS_ERR_OR_NULL(dent))
2019                 goto fail;
2020         sdp->debugfs_dentry_sbstats = dent;
2021
2022         return 0;
2023 fail:
2024         gfs2_delete_debugfs_file(sdp);
2025         return dent ? PTR_ERR(dent) : -ENOMEM;
2026 }
2027
2028 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2029 {
2030         if (sdp->debugfs_dir) {
2031                 if (sdp->debugfs_dentry_glocks) {
2032                         debugfs_remove(sdp->debugfs_dentry_glocks);
2033                         sdp->debugfs_dentry_glocks = NULL;
2034                 }
2035                 if (sdp->debugfs_dentry_glstats) {
2036                         debugfs_remove(sdp->debugfs_dentry_glstats);
2037                         sdp->debugfs_dentry_glstats = NULL;
2038                 }
2039                 if (sdp->debugfs_dentry_sbstats) {
2040                         debugfs_remove(sdp->debugfs_dentry_sbstats);
2041                         sdp->debugfs_dentry_sbstats = NULL;
2042                 }
2043                 debugfs_remove(sdp->debugfs_dir);
2044                 sdp->debugfs_dir = NULL;
2045         }
2046 }
2047
2048 int gfs2_register_debugfs(void)
2049 {
2050         gfs2_root = debugfs_create_dir("gfs2", NULL);
2051         if (IS_ERR(gfs2_root))
2052                 return PTR_ERR(gfs2_root);
2053         return gfs2_root ? 0 : -ENOMEM;
2054 }
2055
2056 void gfs2_unregister_debugfs(void)
2057 {
2058         debugfs_remove(gfs2_root);
2059         gfs2_root = NULL;
2060 }