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 = 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 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 = NULL;
 659         struct address_space *mapping;
 660         struct kmem_cache *cachep;
 661         int ret, tries = 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         ret = rhashtable_lookup_insert_fast(&gl_hash_table, &gl->gl_node,
 727                                             ht_parms);
 728         if (ret == 0) {
 729                 *glp = gl;
 730                 return 0;
 731         }
 732
 733         if (ret == -EEXIST) {
 734                 ret = 0;
 735                 rcu_read_lock();
 736                 tmp = rhashtable_lookup_fast(&gl_hash_table, &name, ht_parms);
 737                 if (tmp == NULL || !lockref_get_not_dead(&tmp->gl_lockref)) {
 738                         if (++tries < 100) {
 739                                 rcu_read_unlock();
 740                                 cond_resched();
 741                                 goto again;
 742                         }
 743                         tmp = NULL;
 744                         ret = -ENOMEM;
 745                 }
 746                 rcu_read_unlock();
 747         } else {
 748                 WARN_ON_ONCE(ret);
 749         }
 750         kfree(gl->gl_lksb.sb_lvbptr);
 751         kmem_cache_free(cachep, gl);
 752         atomic_dec(&sdp->sd_glock_disposal);
 753         *glp = tmp;
 754
 755         return ret;
 756 }
 757
 758 /**
 759  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
 760  * @gl: the glock
 761  * @state: the state we're requesting
 762  * @flags: the modifier flags
 763  * @gh: the holder structure
 764  *
 765  */
 766
 767 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
 768                       struct gfs2_holder *gh)
 769 {
 770         INIT_LIST_HEAD(&gh->gh_list);
 771         gh->gh_gl = gl;
 772         gh->gh_ip = _RET_IP_;
 773         gh->gh_owner_pid = get_pid(task_pid(current));
 774         gh->gh_state = state;
 775         gh->gh_flags = flags;
 776         gh->gh_error = 0;
 777         gh->gh_iflags = 0;
 778         gfs2_glock_hold(gl);
 779 }
 780
 781 /**
 782  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
 783  * @state: the state we're requesting
 784  * @flags: the modifier flags
 785  * @gh: the holder structure
 786  *
 787  * Don't mess with the glock.
 788  *
 789  */
 790
 791 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
 792 {
 793         gh->gh_state = state;
 794         gh->gh_flags = flags;
 795         gh->gh_iflags = 0;
 796         gh->gh_ip = _RET_IP_;
 797         put_pid(gh->gh_owner_pid);
 798         gh->gh_owner_pid = get_pid(task_pid(current));
 799 }
 800
 801 /**
 802  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
 803  * @gh: the holder structure
 804  *
 805  */
 806
 807 void gfs2_holder_uninit(struct gfs2_holder *gh)
 808 {
 809         put_pid(gh->gh_owner_pid);
 810         gfs2_glock_put(gh->gh_gl);
 811         gfs2_holder_mark_uninitialized(gh);
 812         gh->gh_ip = 0;
 813 }
 814
 815 /**
 816  * gfs2_glock_wait - wait on a glock acquisition
 817  * @gh: the glock holder
 818  *
 819  * Returns: 0 on success
 820  */
 821
 822 int gfs2_glock_wait(struct gfs2_holder *gh)
 823 {
 824         unsigned long time1 = jiffies;
 825
 826         might_sleep();
 827         wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
 828         if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
 829                 /* Lengthen the minimum hold time. */
 830                 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
 831                                               GL_GLOCK_HOLD_INCR,
 832                                               GL_GLOCK_MAX_HOLD);
 833         return gh->gh_error;
 834 }
 835
 836 /**
 837  * handle_callback - process a demote request
 838  * @gl: the glock
 839  * @state: the state the caller wants us to change to
 840  *
 841  * There are only two requests that we are going to see in actual
 842  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
 843  */
 844
 845 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
 846                             unsigned long delay, bool remote)
 847 {
 848         int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
 849
 850         set_bit(bit, &gl->gl_flags);
 851         if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
 852                 gl->gl_demote_state = state;
 853                 gl->gl_demote_time = jiffies;
 854         } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
 855                         gl->gl_demote_state != state) {
 856                 gl->gl_demote_state = LM_ST_UNLOCKED;
 857         }
 858         if (gl->gl_ops->go_callback)
 859                 gl->gl_ops->go_callback(gl, remote);
 860         trace_gfs2_demote_rq(gl, remote);
 861 }
 862
 863 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
 864 {
 865         struct va_format vaf;
 866         va_list args;
 867
 868         va_start(args, fmt);
 869
 870         if (seq) {
 871                 seq_vprintf(seq, fmt, args);
 872         } else {
 873                 vaf.fmt = fmt;
 874                 vaf.va = &args;
 875
 876                 pr_err("%pV", &vaf);
 877         }
 878
 879         va_end(args);
 880 }
 881
 882 /**
 883  * add_to_queue - Add a holder to the wait queue (but look for recursion)
 884  * @gh: the holder structure to add
 885  *
 886  * Eventually we should move the recursive locking trap to a
 887  * debugging option or something like that. This is the fast
 888  * path and needs to have the minimum number of distractions.
 889  *
 890  */
 891
 892 static inline void add_to_queue(struct gfs2_holder *gh)
 893 __releases(&gl->gl_lockref.lock)
 894 __acquires(&gl->gl_lockref.lock)
 895 {
 896         struct gfs2_glock *gl = gh->gh_gl;
 897         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 898         struct list_head *insert_pt = NULL;
 899         struct gfs2_holder *gh2;
 900         int try_futile = 0;
 901
 902         BUG_ON(gh->gh_owner_pid == NULL);
 903         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
 904                 BUG();
 905
 906         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
 907                 if (test_bit(GLF_LOCK, &gl->gl_flags))
 908                         try_futile = !may_grant(gl, gh);
 909                 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 910                         goto fail;
 911         }
 912
 913         list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
 914                 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
 915                     (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
 916                         goto trap_recursive;
 917                 if (try_futile &&
 918                     !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
 919 fail:
 920                         gh->gh_error = GLR_TRYFAILED;
 921                         gfs2_holder_wake(gh);
 922                         return;
 923                 }
 924                 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
 925                         continue;
 926                 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
 927                         insert_pt = &gh2->gh_list;
 928         }
 929         set_bit(GLF_QUEUED, &gl->gl_flags);
 930         trace_gfs2_glock_queue(gh, 1);
 931         gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
 932         gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
 933         if (likely(insert_pt == NULL)) {
 934                 list_add_tail(&gh->gh_list, &gl->gl_holders);
 935                 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
 936                         goto do_cancel;
 937                 return;
 938         }
 939         list_add_tail(&gh->gh_list, insert_pt);
 940 do_cancel:
 941         gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
 942         if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
 943                 spin_unlock(&gl->gl_lockref.lock);
 944                 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
 945                         sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
 946                 spin_lock(&gl->gl_lockref.lock);
 947         }
 948         return;
 949
 950 trap_recursive:
 951         pr_err("original: %pSR\n", (void *)gh2->gh_ip);
 952         pr_err("pid: %d\n", pid_nr(gh2->gh_owner_pid));
 953         pr_err("lock type: %d req lock state : %d\n",
 954                gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
 955         pr_err("new: %pSR\n", (void *)gh->gh_ip);
 956         pr_err("pid: %d\n", pid_nr(gh->gh_owner_pid));
 957         pr_err("lock type: %d req lock state : %d\n",
 958                gh->gh_gl->gl_name.ln_type, gh->gh_state);
 959         gfs2_dump_glock(NULL, gl);
 960         BUG();
 961 }
 962
 963 /**
 964  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
 965  * @gh: the holder structure
 966  *
 967  * if (gh->gh_flags & GL_ASYNC), this never returns an error
 968  *
 969  * Returns: 0, GLR_TRYFAILED, or errno on failure
 970  */
 971
 972 int gfs2_glock_nq(struct gfs2_holder *gh)
 973 {
 974         struct gfs2_glock *gl = gh->gh_gl;
 975         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 976         int error = 0;
 977
 978         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 979                 return -EIO;
 980
 981         if (test_bit(GLF_LRU, &gl->gl_flags))
 982                 gfs2_glock_remove_from_lru(gl);
 983
 984         spin_lock(&gl->gl_lockref.lock);
 985         add_to_queue(gh);
 986         if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
 987                      test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
 988                 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
 989                 gl->gl_lockref.count++;
 990                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 991                         gl->gl_lockref.count--;
 992         }
 993         run_queue(gl, 1);
 994         spin_unlock(&gl->gl_lockref.lock);
 995
 996         if (!(gh->gh_flags & GL_ASYNC))
 997                 error = gfs2_glock_wait(gh);
 998
 999         return error;
1000 }
1001
1002 /**
1003  * gfs2_glock_poll - poll to see if an async request has been completed
1004  * @gh: the holder
1005  *
1006  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1007  */
1008
1009 int gfs2_glock_poll(struct gfs2_holder *gh)
1010 {
1011         return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1012 }
1013
1014 /**
1015  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1016  * @gh: the glock holder
1017  *
1018  */
1019
1020 void gfs2_glock_dq(struct gfs2_holder *gh)
1021 {
1022         struct gfs2_glock *gl = gh->gh_gl;
1023         const struct gfs2_glock_operations *glops = gl->gl_ops;
1024         unsigned delay = 0;
1025         int fast_path = 0;
1026
1027         spin_lock(&gl->gl_lockref.lock);
1028         if (gh->gh_flags & GL_NOCACHE)
1029                 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1030
1031         list_del_init(&gh->gh_list);
1032         clear_bit(HIF_HOLDER, &gh->gh_iflags);
1033         if (find_first_holder(gl) == NULL) {
1034                 if (glops->go_unlock) {
1035                         GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1036                         spin_unlock(&gl->gl_lockref.lock);
1037                         glops->go_unlock(gh);
1038                         spin_lock(&gl->gl_lockref.lock);
1039                         clear_bit(GLF_LOCK, &gl->gl_flags);
1040                 }
1041                 if (list_empty(&gl->gl_holders) &&
1042                     !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1043                     !test_bit(GLF_DEMOTE, &gl->gl_flags))
1044                         fast_path = 1;
1045         }
1046         if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl) &&
1047             (glops->go_flags & GLOF_LRU))
1048                 gfs2_glock_add_to_lru(gl);
1049
1050         trace_gfs2_glock_queue(gh, 0);
1051         spin_unlock(&gl->gl_lockref.lock);
1052         if (likely(fast_path))
1053                 return;
1054
1055         gfs2_glock_hold(gl);
1056         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1057             !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1058             gl->gl_name.ln_type == LM_TYPE_INODE)
1059                 delay = gl->gl_hold_time;
1060         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1061                 gfs2_glock_put(gl);
1062 }
1063
1064 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1065 {
1066         struct gfs2_glock *gl = gh->gh_gl;
1067         gfs2_glock_dq(gh);
1068         might_sleep();
1069         wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1070 }
1071
1072 /**
1073  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1074  * @gh: the holder structure
1075  *
1076  */
1077
1078 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1079 {
1080         gfs2_glock_dq(gh);
1081         gfs2_holder_uninit(gh);
1082 }
1083
1084 /**
1085  * gfs2_glock_nq_num - acquire a glock based on lock number
1086  * @sdp: the filesystem
1087  * @number: the lock number
1088  * @glops: the glock operations for the type of glock
1089  * @state: the state to acquire the glock in
1090  * @flags: modifier flags for the acquisition
1091  * @gh: the struct gfs2_holder
1092  *
1093  * Returns: errno
1094  */
1095
1096 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1097                       const struct gfs2_glock_operations *glops,
1098                       unsigned int state, u16 flags, struct gfs2_holder *gh)
1099 {
1100         struct gfs2_glock *gl;
1101         int error;
1102
1103         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1104         if (!error) {
1105                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1106                 gfs2_glock_put(gl);
1107         }
1108
1109         return error;
1110 }
1111
1112 /**
1113  * glock_compare - Compare two struct gfs2_glock structures for sorting
1114  * @arg_a: the first structure
1115  * @arg_b: the second structure
1116  *
1117  */
1118
1119 static int glock_compare(const void *arg_a, const void *arg_b)
1120 {
1121         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1122         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1123         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1124         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1125
1126         if (a->ln_number > b->ln_number)
1127                 return 1;
1128         if (a->ln_number < b->ln_number)
1129                 return -1;
1130         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1131         return 0;
1132 }
1133
1134 /**
1135  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1136  * @num_gh: the number of structures
1137  * @ghs: an array of struct gfs2_holder structures
1138  *
1139  * Returns: 0 on success (all glocks acquired),
1140  *          errno on failure (no glocks acquired)
1141  */
1142
1143 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1144                      struct gfs2_holder **p)
1145 {
1146         unsigned int x;
1147         int error = 0;
1148
1149         for (x = 0; x < num_gh; x++)
1150                 p[x] = &ghs[x];
1151
1152         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1153
1154         for (x = 0; x < num_gh; x++) {
1155                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1156
1157                 error = gfs2_glock_nq(p[x]);
1158                 if (error) {
1159                         while (x--)
1160                                 gfs2_glock_dq(p[x]);
1161                         break;
1162                 }
1163         }
1164
1165         return error;
1166 }
1167
1168 /**
1169  * gfs2_glock_nq_m - acquire multiple glocks
1170  * @num_gh: the number of structures
1171  * @ghs: an array of struct gfs2_holder structures
1172  *
1173  *
1174  * Returns: 0 on success (all glocks acquired),
1175  *          errno on failure (no glocks acquired)
1176  */
1177
1178 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1179 {
1180         struct gfs2_holder *tmp[4];
1181         struct gfs2_holder **pph = tmp;
1182         int error = 0;
1183
1184         switch(num_gh) {
1185         case 0:
1186                 return 0;
1187         case 1:
1188                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1189                 return gfs2_glock_nq(ghs);
1190         default:
1191                 if (num_gh <= 4)
1192                         break;
1193                 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1194                 if (!pph)
1195                         return -ENOMEM;
1196         }
1197
1198         error = nq_m_sync(num_gh, ghs, pph);
1199
1200         if (pph != tmp)
1201                 kfree(pph);
1202
1203         return error;
1204 }
1205
1206 /**
1207  * gfs2_glock_dq_m - release multiple glocks
1208  * @num_gh: the number of structures
1209  * @ghs: an array of struct gfs2_holder structures
1210  *
1211  */
1212
1213 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1214 {
1215         while (num_gh--)
1216                 gfs2_glock_dq(&ghs[num_gh]);
1217 }
1218
1219 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1220 {
1221         unsigned long delay = 0;
1222         unsigned long holdtime;
1223         unsigned long now = jiffies;
1224
1225         gfs2_glock_hold(gl);
1226         holdtime = gl->gl_tchange + gl->gl_hold_time;
1227         if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1228             gl->gl_name.ln_type == LM_TYPE_INODE) {
1229                 if (time_before(now, holdtime))
1230                         delay = holdtime - now;
1231                 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1232                         delay = gl->gl_hold_time;
1233         }
1234
1235         spin_lock(&gl->gl_lockref.lock);
1236         handle_callback(gl, state, delay, true);
1237         spin_unlock(&gl->gl_lockref.lock);
1238         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1239                 gfs2_glock_put(gl);
1240 }
1241
1242 /**
1243  * gfs2_should_freeze - Figure out if glock should be frozen
1244  * @gl: The glock in question
1245  *
1246  * Glocks are not frozen if (a) the result of the dlm operation is
1247  * an error, (b) the locking operation was an unlock operation or
1248  * (c) if there is a "noexp" flagged request anywhere in the queue
1249  *
1250  * Returns: 1 if freezing should occur, 0 otherwise
1251  */
1252
1253 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1254 {
1255         const struct gfs2_holder *gh;
1256
1257         if (gl->gl_reply & ~LM_OUT_ST_MASK)
1258                 return 0;
1259         if (gl->gl_target == LM_ST_UNLOCKED)
1260                 return 0;
1261
1262         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1263                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1264                         continue;
1265                 if (LM_FLAG_NOEXP & gh->gh_flags)
1266                         return 0;
1267         }
1268
1269         return 1;
1270 }
1271
1272 /**
1273  * gfs2_glock_complete - Callback used by locking
1274  * @gl: Pointer to the glock
1275  * @ret: The return value from the dlm
1276  *
1277  * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1278  * to use a bitfield shared with other glock state fields.
1279  */
1280
1281 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1282 {
1283         struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1284
1285         spin_lock(&gl->gl_lockref.lock);
1286         gl->gl_reply = ret;
1287
1288         if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1289                 if (gfs2_should_freeze(gl)) {
1290                         set_bit(GLF_FROZEN, &gl->gl_flags);
1291                         spin_unlock(&gl->gl_lockref.lock);
1292                         return;
1293                 }
1294         }
1295
1296         gl->gl_lockref.count++;
1297         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1298         spin_unlock(&gl->gl_lockref.lock);
1299
1300         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1301                 gfs2_glock_put(gl);
1302 }
1303
1304 static int glock_cmp(void *priv, struct list_head *a, struct list_head *b)
1305 {
1306         struct gfs2_glock *gla, *glb;
1307
1308         gla = list_entry(a, struct gfs2_glock, gl_lru);
1309         glb = list_entry(b, struct gfs2_glock, gl_lru);
1310
1311         if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1312                 return 1;
1313         if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1314                 return -1;
1315
1316         return 0;
1317 }
1318
1319 /**
1320  * gfs2_dispose_glock_lru - Demote a list of glocks
1321  * @list: The list to dispose of
1322  *
1323  * Disposing of glocks may involve disk accesses, so that here we sort
1324  * the glocks by number (i.e. disk location of the inodes) so that if
1325  * there are any such accesses, they'll be sent in order (mostly).
1326  *
1327  * Must be called under the lru_lock, but may drop and retake this
1328  * lock. While the lru_lock is dropped, entries may vanish from the
1329  * list, but no new entries will appear on the list (since it is
1330  * private)
1331  */
1332
1333 static void gfs2_dispose_glock_lru(struct list_head *list)
1334 __releases(&lru_lock)
1335 __acquires(&lru_lock)
1336 {
1337         struct gfs2_glock *gl;
1338
1339         list_sort(NULL, list, glock_cmp);
1340
1341         while(!list_empty(list)) {
1342                 gl = list_entry(list->next, struct gfs2_glock, gl_lru);
1343                 list_del_init(&gl->gl_lru);
1344                 if (!spin_trylock(&gl->gl_lockref.lock)) {
1345 add_back_to_lru:
1346                         list_add(&gl->gl_lru, &lru_list);
1347                         atomic_inc(&lru_count);
1348                         continue;
1349                 }
1350                 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1351                         spin_unlock(&gl->gl_lockref.lock);
1352                         goto add_back_to_lru;
1353                 }
1354                 clear_bit(GLF_LRU, &gl->gl_flags);
1355                 gl->gl_lockref.count++;
1356                 if (demote_ok(gl))
1357                         handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1358                 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1359                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1360                         gl->gl_lockref.count--;
1361                 spin_unlock(&gl->gl_lockref.lock);
1362                 cond_resched_lock(&lru_lock);
1363         }
1364 }
1365
1366 /**
1367  * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1368  * @nr: The number of entries to scan
1369  *
1370  * This function selects the entries on the LRU which are able to
1371  * be demoted, and then kicks off the process by calling
1372  * gfs2_dispose_glock_lru() above.
1373  */
1374
1375 static long gfs2_scan_glock_lru(int nr)
1376 {
1377         struct gfs2_glock *gl;
1378         LIST_HEAD(skipped);
1379         LIST_HEAD(dispose);
1380         long freed = 0;
1381
1382         spin_lock(&lru_lock);
1383         while ((nr-- >= 0) && !list_empty(&lru_list)) {
1384                 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1385
1386                 /* Test for being demotable */
1387                 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1388                         list_move(&gl->gl_lru, &dispose);
1389                         atomic_dec(&lru_count);
1390                         freed++;
1391                         continue;
1392                 }
1393
1394                 list_move(&gl->gl_lru, &skipped);
1395         }
1396         list_splice(&skipped, &lru_list);
1397         if (!list_empty(&dispose))
1398                 gfs2_dispose_glock_lru(&dispose);
1399         spin_unlock(&lru_lock);
1400
1401         return freed;
1402 }
1403
1404 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1405                                             struct shrink_control *sc)
1406 {
1407         if (!(sc->gfp_mask & __GFP_FS))
1408                 return SHRINK_STOP;
1409         return gfs2_scan_glock_lru(sc->nr_to_scan);
1410 }
1411
1412 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1413                                              struct shrink_control *sc)
1414 {
1415         return vfs_pressure_ratio(atomic_read(&lru_count));
1416 }
1417
1418 static struct shrinker glock_shrinker = {
1419         .seeks = DEFAULT_SEEKS,
1420         .count_objects = gfs2_glock_shrink_count,
1421         .scan_objects = gfs2_glock_shrink_scan,
1422 };
1423
1424 /**
1425  * examine_bucket - Call a function for glock in a hash bucket
1426  * @examiner: the function
1427  * @sdp: the filesystem
1428  * @bucket: the bucket
1429  *
1430  * Note that the function can be called multiple times on the same
1431  * object.  So the user must ensure that the function can cope with
1432  * that.
1433  */
1434
1435 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1436 {
1437         struct gfs2_glock *gl;
1438         struct rhashtable_iter iter;
1439
1440         rhashtable_walk_enter(&gl_hash_table, &iter);
1441
1442         do {
1443                 gl = ERR_PTR(rhashtable_walk_start(&iter));
1444                 if (gl)
1445                         continue;
1446
1447                 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl))
1448                         if ((gl->gl_name.ln_sbd == sdp) &&
1449                             lockref_get_not_dead(&gl->gl_lockref))
1450                                 examiner(gl);
1451
1452                 rhashtable_walk_stop(&iter);
1453         } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
1454
1455         rhashtable_walk_exit(&iter);
1456 }
1457
1458 /**
1459  * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1460  * @gl: The glock to thaw
1461  *
1462  */
1463
1464 static void thaw_glock(struct gfs2_glock *gl)
1465 {
1466         if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1467                 goto out;
1468         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1469         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0) {
1470 out:
1471                 gfs2_glock_put(gl);
1472         }
1473 }
1474
1475 /**
1476  * clear_glock - look at a glock and see if we can free it from glock cache
1477  * @gl: the glock to look at
1478  *
1479  */
1480
1481 static void clear_glock(struct gfs2_glock *gl)
1482 {
1483         gfs2_glock_remove_from_lru(gl);
1484
1485         spin_lock(&gl->gl_lockref.lock);
1486         if (gl->gl_state != LM_ST_UNLOCKED)
1487                 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1488         spin_unlock(&gl->gl_lockref.lock);
1489         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1490                 gfs2_glock_put(gl);
1491 }
1492
1493 /**
1494  * gfs2_glock_thaw - Thaw any frozen glocks
1495  * @sdp: The super block
1496  *
1497  */
1498
1499 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1500 {
1501         glock_hash_walk(thaw_glock, sdp);
1502 }
1503
1504 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1505 {
1506         spin_lock(&gl->gl_lockref.lock);
1507         gfs2_dump_glock(seq, gl);
1508         spin_unlock(&gl->gl_lockref.lock);
1509 }
1510
1511 static void dump_glock_func(struct gfs2_glock *gl)
1512 {
1513         dump_glock(NULL, gl);
1514 }
1515
1516 /**
1517  * gfs2_gl_hash_clear - Empty out the glock hash table
1518  * @sdp: the filesystem
1519  * @wait: wait until it's all gone
1520  *
1521  * Called when unmounting the filesystem.
1522  */
1523
1524 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1525 {
1526         set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1527         flush_workqueue(glock_workqueue);
1528         glock_hash_walk(clear_glock, sdp);
1529         flush_workqueue(glock_workqueue);
1530         wait_event_timeout(sdp->sd_glock_wait,
1531                            atomic_read(&sdp->sd_glock_disposal) == 0,
1532                            HZ * 600);
1533         glock_hash_walk(dump_glock_func, sdp);
1534 }
1535
1536 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1537 {
1538         struct gfs2_glock *gl = ip->i_gl;
1539         int ret;
1540
1541         ret = gfs2_truncatei_resume(ip);
1542         gfs2_assert_withdraw(gl->gl_name.ln_sbd, ret == 0);
1543
1544         spin_lock(&gl->gl_lockref.lock);
1545         clear_bit(GLF_LOCK, &gl->gl_flags);
1546         run_queue(gl, 1);
1547         spin_unlock(&gl->gl_lockref.lock);
1548 }
1549
1550 static const char *state2str(unsigned state)
1551 {
1552         switch(state) {
1553         case LM_ST_UNLOCKED:
1554                 return "UN";
1555         case LM_ST_SHARED:
1556                 return "SH";
1557         case LM_ST_DEFERRED:
1558                 return "DF";
1559         case LM_ST_EXCLUSIVE:
1560                 return "EX";
1561         }
1562         return "??";
1563 }
1564
1565 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
1566 {
1567         char *p = buf;
1568         if (flags & LM_FLAG_TRY)
1569                 *p++ = 't';
1570         if (flags & LM_FLAG_TRY_1CB)
1571                 *p++ = 'T';
1572         if (flags & LM_FLAG_NOEXP)
1573                 *p++ = 'e';
1574         if (flags & LM_FLAG_ANY)
1575                 *p++ = 'A';
1576         if (flags & LM_FLAG_PRIORITY)
1577                 *p++ = 'p';
1578         if (flags & GL_ASYNC)
1579                 *p++ = 'a';
1580         if (flags & GL_EXACT)
1581                 *p++ = 'E';
1582         if (flags & GL_NOCACHE)
1583                 *p++ = 'c';
1584         if (test_bit(HIF_HOLDER, &iflags))
1585                 *p++ = 'H';
1586         if (test_bit(HIF_WAIT, &iflags))
1587                 *p++ = 'W';
1588         if (test_bit(HIF_FIRST, &iflags))
1589                 *p++ = 'F';
1590         *p = 0;
1591         return buf;
1592 }
1593
1594 /**
1595  * dump_holder - print information about a glock holder
1596  * @seq: the seq_file struct
1597  * @gh: the glock holder
1598  *
1599  */
1600
1601 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1602 {
1603         struct task_struct *gh_owner = NULL;
1604         char flags_buf[32];
1605
1606         rcu_read_lock();
1607         if (gh->gh_owner_pid)
1608                 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1609         gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1610                        state2str(gh->gh_state),
1611                        hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1612                        gh->gh_error,
1613                        gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1614                        gh_owner ? gh_owner->comm : "(ended)",
1615                        (void *)gh->gh_ip);
1616         rcu_read_unlock();
1617 }
1618
1619 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1620 {
1621         const unsigned long *gflags = &gl->gl_flags;
1622         char *p = buf;
1623
1624         if (test_bit(GLF_LOCK, gflags))
1625                 *p++ = 'l';
1626         if (test_bit(GLF_DEMOTE, gflags))
1627                 *p++ = 'D';
1628         if (test_bit(GLF_PENDING_DEMOTE, gflags))
1629                 *p++ = 'd';
1630         if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1631                 *p++ = 'p';
1632         if (test_bit(GLF_DIRTY, gflags))
1633                 *p++ = 'y';
1634         if (test_bit(GLF_LFLUSH, gflags))
1635                 *p++ = 'f';
1636         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1637                 *p++ = 'i';
1638         if (test_bit(GLF_REPLY_PENDING, gflags))
1639                 *p++ = 'r';
1640         if (test_bit(GLF_INITIAL, gflags))
1641                 *p++ = 'I';
1642         if (test_bit(GLF_FROZEN, gflags))
1643                 *p++ = 'F';
1644         if (test_bit(GLF_QUEUED, gflags))
1645                 *p++ = 'q';
1646         if (test_bit(GLF_LRU, gflags))
1647                 *p++ = 'L';
1648         if (gl->gl_object)
1649                 *p++ = 'o';
1650         if (test_bit(GLF_BLOCKING, gflags))
1651                 *p++ = 'b';
1652         *p = 0;
1653         return buf;
1654 }
1655
1656 /**
1657  * gfs2_dump_glock - print information about a glock
1658  * @seq: The seq_file struct
1659  * @gl: the glock
1660  *
1661  * The file format is as follows:
1662  * One line per object, capital letters are used to indicate objects
1663  * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1664  * other objects are indented by a single space and follow the glock to
1665  * which they are related. Fields are indicated by lower case letters
1666  * followed by a colon and the field value, except for strings which are in
1667  * [] so that its possible to see if they are composed of spaces for
1668  * example. The field's are n = number (id of the object), f = flags,
1669  * t = type, s = state, r = refcount, e = error, p = pid.
1670  *
1671  */
1672
1673 void gfs2_dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1674 {
1675         const struct gfs2_glock_operations *glops = gl->gl_ops;
1676         unsigned long long dtime;
1677         const struct gfs2_holder *gh;
1678         char gflags_buf[32];
1679
1680         dtime = jiffies - gl->gl_demote_time;
1681         dtime *= 1000000/HZ; /* demote time in uSec */
1682         if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1683                 dtime = 0;
1684         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",
1685                   state2str(gl->gl_state),
1686                   gl->gl_name.ln_type,
1687                   (unsigned long long)gl->gl_name.ln_number,
1688                   gflags2str(gflags_buf, gl),
1689                   state2str(gl->gl_target),
1690                   state2str(gl->gl_demote_state), dtime,
1691                   atomic_read(&gl->gl_ail_count),
1692                   atomic_read(&gl->gl_revokes),
1693                   (int)gl->gl_lockref.count, gl->gl_hold_time);
1694
1695         list_for_each_entry(gh, &gl->gl_holders, gh_list)
1696                 dump_holder(seq, gh);
1697
1698         if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1699                 glops->go_dump(seq, gl);
1700 }
1701
1702 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
1703 {
1704         struct gfs2_glock *gl = iter_ptr;
1705
1706         seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
1707                    gl->gl_name.ln_type,
1708                    (unsigned long long)gl->gl_name.ln_number,
1709                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
1710                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
1711                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
1712                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
1713                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
1714                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
1715                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
1716                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
1717         return 0;
1718 }
1719
1720 static const char *gfs2_gltype[] = {
1721         "type",
1722         "reserved",
1723         "nondisk",
1724         "inode",
1725         "rgrp",
1726         "meta",
1727         "iopen",
1728         "flock",
1729         "plock",
1730         "quota",
1731         "journal",
1732 };
1733
1734 static const char *gfs2_stype[] = {
1735         [GFS2_LKS_SRTT]         = "srtt",
1736         [GFS2_LKS_SRTTVAR]      = "srttvar",
1737         [GFS2_LKS_SRTTB]        = "srttb",
1738         [GFS2_LKS_SRTTVARB]     = "srttvarb",
1739         [GFS2_LKS_SIRT]         = "sirt",
1740         [GFS2_LKS_SIRTVAR]      = "sirtvar",
1741         [GFS2_LKS_DCOUNT]       = "dlm",
1742         [GFS2_LKS_QCOUNT]       = "queue",
1743 };
1744
1745 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1746
1747 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
1748 {
1749         struct gfs2_sbd *sdp = seq->private;
1750         loff_t pos = *(loff_t *)iter_ptr;
1751         unsigned index = pos >> 3;
1752         unsigned subindex = pos & 0x07;
1753         int i;
1754
1755         if (index == 0 && subindex != 0)
1756                 return 0;
1757
1758         seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
1759                    (index == 0) ? "cpu": gfs2_stype[subindex]);
1760
1761         for_each_possible_cpu(i) {
1762                 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
1763
1764                 if (index == 0)
1765                         seq_printf(seq, " %15u", i);
1766                 else
1767                         seq_printf(seq, " %15llu", (unsigned long long)lkstats->
1768                                    lkstats[index - 1].stats[subindex]);
1769         }
1770         seq_putc(seq, '\n');
1771         return 0;
1772 }
1773
1774 int __init gfs2_glock_init(void)
1775 {
1776         int ret;
1777
1778         ret = rhashtable_init(&gl_hash_table, &ht_parms);
1779         if (ret < 0)
1780                 return ret;
1781
1782         glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1783                                           WQ_HIGHPRI | WQ_FREEZABLE, 0);
1784         if (!glock_workqueue) {
1785                 rhashtable_destroy(&gl_hash_table);
1786                 return -ENOMEM;
1787         }
1788         gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1789                                                 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1790                                                 0);
1791         if (!gfs2_delete_workqueue) {
1792                 destroy_workqueue(glock_workqueue);
1793                 rhashtable_destroy(&gl_hash_table);
1794                 return -ENOMEM;
1795         }
1796
1797         ret = register_shrinker(&glock_shrinker);
1798         if (ret) {
1799                 destroy_workqueue(gfs2_delete_workqueue);
1800                 destroy_workqueue(glock_workqueue);
1801                 rhashtable_destroy(&gl_hash_table);
1802                 return ret;
1803         }
1804
1805         return 0;
1806 }
1807
1808 void gfs2_glock_exit(void)
1809 {
1810         unregister_shrinker(&glock_shrinker);
1811         rhashtable_destroy(&gl_hash_table);
1812         destroy_workqueue(glock_workqueue);
1813         destroy_workqueue(gfs2_delete_workqueue);
1814 }
1815
1816 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1817 {
1818         while ((gi->gl = rhashtable_walk_next(&gi->hti))) {
1819                 if (IS_ERR(gi->gl)) {
1820                         if (PTR_ERR(gi->gl) == -EAGAIN)
1821                                 continue;
1822                         gi->gl = NULL;
1823                         return;
1824                 }
1825                 /* Skip entries for other sb and dead entries */
1826                 if (gi->sdp == gi->gl->gl_name.ln_sbd &&
1827                     !__lockref_is_dead(&gi->gl->gl_lockref))
1828                         return;
1829         }
1830 }
1831
1832 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1833 {
1834         struct gfs2_glock_iter *gi = seq->private;
1835         loff_t n = *pos;
1836         int ret;
1837
1838         if (gi->last_pos <= *pos)
1839                 n = (*pos - gi->last_pos);
1840
1841         ret = rhashtable_walk_start(&gi->hti);
1842         if (ret)
1843                 return NULL;
1844
1845         do {
1846                 gfs2_glock_iter_next(gi);
1847         } while (gi->gl && n--);
1848
1849         gi->last_pos = *pos;
1850         return gi->gl;
1851 }
1852
1853 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1854                                  loff_t *pos)
1855 {
1856         struct gfs2_glock_iter *gi = seq->private;
1857
1858         (*pos)++;
1859         gi->last_pos = *pos;
1860         gfs2_glock_iter_next(gi);
1861         return gi->gl;
1862 }
1863
1864 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1865 {
1866         struct gfs2_glock_iter *gi = seq->private;
1867
1868         gi->gl = NULL;
1869         rhashtable_walk_stop(&gi->hti);
1870 }
1871
1872 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1873 {
1874         dump_glock(seq, iter_ptr);
1875         return 0;
1876 }
1877
1878 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
1879 {
1880         preempt_disable();
1881         if (*pos >= GFS2_NR_SBSTATS)
1882                 return NULL;
1883         return pos;
1884 }
1885
1886 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
1887                                    loff_t *pos)
1888 {
1889         (*pos)++;
1890         if (*pos >= GFS2_NR_SBSTATS)
1891                 return NULL;
1892         return pos;
1893 }
1894
1895 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
1896 {
1897         preempt_enable();
1898 }
1899
1900 static const struct seq_operations gfs2_glock_seq_ops = {
1901         .start = gfs2_glock_seq_start,
1902         .next  = gfs2_glock_seq_next,
1903         .stop  = gfs2_glock_seq_stop,
1904         .show  = gfs2_glock_seq_show,
1905 };
1906
1907 static const struct seq_operations gfs2_glstats_seq_ops = {
1908         .start = gfs2_glock_seq_start,
1909         .next  = gfs2_glock_seq_next,
1910         .stop  = gfs2_glock_seq_stop,
1911         .show  = gfs2_glstats_seq_show,
1912 };
1913
1914 static const struct seq_operations gfs2_sbstats_seq_ops = {
1915         .start = gfs2_sbstats_seq_start,
1916         .next  = gfs2_sbstats_seq_next,
1917         .stop  = gfs2_sbstats_seq_stop,
1918         .show  = gfs2_sbstats_seq_show,
1919 };
1920
1921 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
1922
1923 static int gfs2_glocks_open(struct inode *inode, struct file *file)
1924 {
1925         int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1926                                    sizeof(struct gfs2_glock_iter));
1927         if (ret == 0) {
1928                 struct seq_file *seq = file->private_data;
1929                 struct gfs2_glock_iter *gi = seq->private;
1930
1931                 gi->sdp = inode->i_private;
1932                 gi->last_pos = 0;
1933                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
1934                 if (seq->buf)
1935                         seq->size = GFS2_SEQ_GOODSIZE;
1936                 gi->gl = NULL;
1937                 ret = rhashtable_walk_init(&gl_hash_table, &gi->hti, GFP_KERNEL);
1938         }
1939         return ret;
1940 }
1941
1942 static int gfs2_glocks_release(struct inode *inode, struct file *file)
1943 {
1944         struct seq_file *seq = file->private_data;
1945         struct gfs2_glock_iter *gi = seq->private;
1946
1947         gi->gl = NULL;
1948         rhashtable_walk_exit(&gi->hti);
1949         return seq_release_private(inode, file);
1950 }
1951
1952 static int gfs2_glstats_open(struct inode *inode, struct file *file)
1953 {
1954         int ret = seq_open_private(file, &gfs2_glstats_seq_ops,
1955                                    sizeof(struct gfs2_glock_iter));
1956         if (ret == 0) {
1957                 struct seq_file *seq = file->private_data;
1958                 struct gfs2_glock_iter *gi = seq->private;
1959                 gi->sdp = inode->i_private;
1960                 gi->last_pos = 0;
1961                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
1962                 if (seq->buf)
1963                         seq->size = GFS2_SEQ_GOODSIZE;
1964                 gi->gl = NULL;
1965                 ret = rhashtable_walk_init(&gl_hash_table, &gi->hti, GFP_KERNEL);
1966         }
1967         return ret;
1968 }
1969
1970 static int gfs2_sbstats_open(struct inode *inode, struct file *file)
1971 {
1972         int ret = seq_open(file, &gfs2_sbstats_seq_ops);
1973         if (ret == 0) {
1974                 struct seq_file *seq = file->private_data;
1975                 seq->private = inode->i_private;  /* sdp */
1976         }
1977         return ret;
1978 }
1979
1980 static const struct file_operations gfs2_glocks_fops = {
1981         .owner   = THIS_MODULE,
1982         .open    = gfs2_glocks_open,
1983         .read    = seq_read,
1984         .llseek  = seq_lseek,
1985         .release = gfs2_glocks_release,
1986 };
1987
1988 static const struct file_operations gfs2_glstats_fops = {
1989         .owner   = THIS_MODULE,
1990         .open    = gfs2_glstats_open,
1991         .read    = seq_read,
1992         .llseek  = seq_lseek,
1993         .release = gfs2_glocks_release,
1994 };
1995
1996 static const struct file_operations gfs2_sbstats_fops = {
1997         .owner   = THIS_MODULE,
1998         .open    = gfs2_sbstats_open,
1999         .read    = seq_read,
2000         .llseek  = seq_lseek,
2001         .release = seq_release,
2002 };
2003
2004 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2005 {
2006         struct dentry *dent;
2007
2008         dent = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2009         if (IS_ERR_OR_NULL(dent))
2010                 goto fail;
2011         sdp->debugfs_dir = dent;
2012
2013         dent = debugfs_create_file("glocks",
2014                                    S_IFREG | S_IRUGO,
2015                                    sdp->debugfs_dir, sdp,
2016                                    &gfs2_glocks_fops);
2017         if (IS_ERR_OR_NULL(dent))
2018                 goto fail;
2019         sdp->debugfs_dentry_glocks = dent;
2020
2021         dent = debugfs_create_file("glstats",
2022                                    S_IFREG | S_IRUGO,
2023                                    sdp->debugfs_dir, sdp,
2024                                    &gfs2_glstats_fops);
2025         if (IS_ERR_OR_NULL(dent))
2026                 goto fail;
2027         sdp->debugfs_dentry_glstats = dent;
2028
2029         dent = debugfs_create_file("sbstats",
2030                                    S_IFREG | S_IRUGO,
2031                                    sdp->debugfs_dir, sdp,
2032                                    &gfs2_sbstats_fops);
2033         if (IS_ERR_OR_NULL(dent))
2034                 goto fail;
2035         sdp->debugfs_dentry_sbstats = dent;
2036
2037         return 0;
2038 fail:
2039         gfs2_delete_debugfs_file(sdp);
2040         return dent ? PTR_ERR(dent) : -ENOMEM;
2041 }
2042
2043 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2044 {
2045         if (sdp->debugfs_dir) {
2046                 if (sdp->debugfs_dentry_glocks) {
2047                         debugfs_remove(sdp->debugfs_dentry_glocks);
2048                         sdp->debugfs_dentry_glocks = NULL;
2049                 }
2050                 if (sdp->debugfs_dentry_glstats) {
2051                         debugfs_remove(sdp->debugfs_dentry_glstats);
2052                         sdp->debugfs_dentry_glstats = NULL;
2053                 }
2054                 if (sdp->debugfs_dentry_sbstats) {
2055                         debugfs_remove(sdp->debugfs_dentry_sbstats);
2056                         sdp->debugfs_dentry_sbstats = NULL;
2057                 }
2058                 debugfs_remove(sdp->debugfs_dir);
2059                 sdp->debugfs_dir = NULL;
2060         }
2061 }
2062
2063 int gfs2_register_debugfs(void)
2064 {
2065         gfs2_root = debugfs_create_dir("gfs2", NULL);
2066         if (IS_ERR(gfs2_root))
2067                 return PTR_ERR(gfs2_root);
2068         return gfs2_root ? 0 : -ENOMEM;
2069 }
2070
2071 void gfs2_unregister_debugfs(void)
2072 {
2073         debugfs_remove(gfs2_root);
2074         gfs2_root = NULL;
2075 }