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