fs/gfs2/glock.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
   4  *
   5  * This copyrighted material is made available to anyone wishing to use,
   6  * modify, copy, or redistribute it subject to the terms and conditions
   7  * of the GNU General Public License version 2.
   8  */
   9
  10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  11
  12 #include <linux/sched.h>
  13 #include <linux/slab.h>
  14 #include <linux/spinlock.h>
  15 #include <linux/buffer_head.h>
  16 #include <linux/delay.h>
  17 #include <linux/sort.h>
  18 #include <linux/jhash.h>
  19 #include <linux/kallsyms.h>
  20 #include <linux/gfs2_ondisk.h>
  21 #include <linux/list.h>
  22 #include <linux/wait.h>
  23 #include <linux/module.h>
  24 #include <asm/uaccess.h>
  25 #include <linux/seq_file.h>
  26 #include <linux/debugfs.h>
  27 #include <linux/kthread.h>
  28 #include <linux/freezer.h>
  29 #include <linux/workqueue.h>
  30 #include <linux/jiffies.h>
  31 #include <linux/rcupdate.h>
  32 #include <linux/rculist_bl.h>
  33 #include <linux/bit_spinlock.h>
  34 #include <linux/percpu.h>
  35 #include <linux/list_sort.h>
  36 #include <linux/lockref.h>
  37
  38 #include "gfs2.h"
  39 #include "incore.h"
  40 #include "glock.h"
  41 #include "glops.h"
  42 #include "inode.h"
  43 #include "lops.h"
  44 #include "meta_io.h"
  45 #include "quota.h"
  46 #include "super.h"
  47 #include "util.h"
  48 #include "bmap.h"
  49 #define CREATE_TRACE_POINTS
  50 #include "trace_gfs2.h"
  51
  52 struct gfs2_glock_iter {
  53         int hash;                       /* hash bucket index           */
  54         unsigned nhash;                 /* Index within current bucket */
  55         struct gfs2_sbd *sdp;           /* incore superblock           */
  56         struct gfs2_glock *gl;          /* current glock struct        */
  57         loff_t last_pos;                /* last position               */
  58 };
  59
  60 typedef void (*glock_examiner) (struct gfs2_glock * gl);
  61
  62 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
  63
  64 static struct dentry *gfs2_root;
  65 static struct workqueue_struct *glock_workqueue;
  66 struct workqueue_struct *gfs2_delete_workqueue;
  67 static LIST_HEAD(lru_list);
  68 static atomic_t lru_count = ATOMIC_INIT(0);
  69 static DEFINE_SPINLOCK(lru_lock);
  70
  71 #define GFS2_GL_HASH_SHIFT      15
  72 #define GFS2_GL_HASH_SIZE       (1 << GFS2_GL_HASH_SHIFT)
  73 #define GFS2_GL_HASH_MASK       (GFS2_GL_HASH_SIZE - 1)
  74
  75 static struct hlist_bl_head gl_hash_table[GFS2_GL_HASH_SIZE];
  76 static struct dentry *gfs2_root;
  77
  78 /**
  79  * gl_hash() - Turn glock number into hash bucket number
  80  * @lock: The glock number
  81  *
  82  * Returns: The number of the corresponding hash bucket
  83  */
  84
  85 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
  86                             const struct lm_lockname *name)
  87 {
  88         unsigned int h;
  89
  90         h = jhash(&name->ln_number, sizeof(u64), 0);
  91         h = jhash(&name->ln_type, sizeof(unsigned int), h);
  92         h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
  93         h &= GFS2_GL_HASH_MASK;
  94
  95         return h;
  96 }
  97
  98 static inline void spin_lock_bucket(unsigned int hash)
  99 {
 100         hlist_bl_lock(&gl_hash_table[hash]);
 101 }
 102
 103 static inline void spin_unlock_bucket(unsigned int hash)
 104 {
 105         hlist_bl_unlock(&gl_hash_table[hash]);
 106 }
 107
 108 static void gfs2_glock_dealloc(struct rcu_head *rcu)
 109 {
 110         struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
 111
 112         if (gl->gl_ops->go_flags & GLOF_ASPACE) {
 113                 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
 114         } else {
 115                 kfree(gl->gl_lksb.sb_lvbptr);
 116                 kmem_cache_free(gfs2_glock_cachep, gl);
 117         }
 118 }
 119
 120 void gfs2_glock_free(struct gfs2_glock *gl)
 121 {
 122         struct gfs2_sbd *sdp = gl->gl_sbd;
 123
 124         call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
 125         if (atomic_dec_and_test(&sdp->sd_glock_disposal))
 126                 wake_up(&sdp->sd_glock_wait);
 127 }
 128
 129 /**
 130  * gfs2_glock_hold() - increment reference count on glock
 131  * @gl: The glock to hold
 132  *
 133  */
 134
 135 static void gfs2_glock_hold(struct gfs2_glock *gl)
 136 {
 137         GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
 138         lockref_get(&gl->gl_lockref);
 139 }
 140
 141 /**
 142  * demote_ok - Check to see if it's ok to unlock a glock
 143  * @gl: the glock
 144  *
 145  * Returns: 1 if it's ok
 146  */
 147
 148 static int demote_ok(const struct gfs2_glock *gl)
 149 {
 150         const struct gfs2_glock_operations *glops = gl->gl_ops;
 151
 152         if (gl->gl_state == LM_ST_UNLOCKED)
 153                 return 0;
 154         if (!list_empty(&gl->gl_holders))
 155                 return 0;
 156         if (glops->go_demote_ok)
 157                 return glops->go_demote_ok(gl);
 158         return 1;
 159 }
 160
 161
 162 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
 163 {
 164         spin_lock(&lru_lock);
 165
 166         if (!list_empty(&gl->gl_lru))
 167                 list_del_init(&gl->gl_lru);
 168         else
 169                 atomic_inc(&lru_count);
 170
 171         list_add_tail(&gl->gl_lru, &lru_list);
 172         set_bit(GLF_LRU, &gl->gl_flags);
 173         spin_unlock(&lru_lock);
 174 }
 175
 176 static void __gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
 177 {
 178         if (!list_empty(&gl->gl_lru)) {
 179                 list_del_init(&gl->gl_lru);
 180                 atomic_dec(&lru_count);
 181                 clear_bit(GLF_LRU, &gl->gl_flags);
 182         }
 183 }
 184
 185 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
 186 {
 187         spin_lock(&lru_lock);
 188         __gfs2_glock_remove_from_lru(gl);
 189         spin_unlock(&lru_lock);
 190 }
 191
 192 /**
 193  * gfs2_glock_put() - Decrement reference count on glock
 194  * @gl: The glock to put
 195  *
 196  */
 197
 198 void gfs2_glock_put(struct gfs2_glock *gl)
 199 {
 200         struct gfs2_sbd *sdp = gl->gl_sbd;
 201         struct address_space *mapping = gfs2_glock2aspace(gl);
 202
 203         if (lockref_put_or_lock(&gl->gl_lockref))
 204                 return;
 205
 206         lockref_mark_dead(&gl->gl_lockref);
 207
 208         spin_lock(&lru_lock);
 209         __gfs2_glock_remove_from_lru(gl);
 210         spin_unlock(&lru_lock);
 211         spin_unlock(&gl->gl_lockref.lock);
 212         spin_lock_bucket(gl->gl_hash);
 213         hlist_bl_del_rcu(&gl->gl_list);
 214         spin_unlock_bucket(gl->gl_hash);
 215         GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
 216         GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
 217         trace_gfs2_glock_put(gl);
 218         sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
 219 }
 220
 221 /**
 222  * search_bucket() - Find struct gfs2_glock by lock number
 223  * @bucket: the bucket to search
 224  * @name: The lock name
 225  *
 226  * Returns: NULL, or the struct gfs2_glock with the requested number
 227  */
 228
 229 static struct gfs2_glock *search_bucket(unsigned int hash,
 230                                         const struct gfs2_sbd *sdp,
 231                                         const struct lm_lockname *name)
 232 {
 233         struct gfs2_glock *gl;
 234         struct hlist_bl_node *h;
 235
 236         hlist_bl_for_each_entry_rcu(gl, h, &gl_hash_table[hash], gl_list) {
 237                 if (!lm_name_equal(&gl->gl_name, name))
 238                         continue;
 239                 if (gl->gl_sbd != sdp)
 240                         continue;
 241                 if (lockref_get_not_dead(&gl->gl_lockref))
 242                         return gl;
 243         }
 244
 245         return NULL;
 246 }
 247
 248 /**
 249  * may_grant - check if its ok to grant a new lock
 250  * @gl: The glock
 251  * @gh: The lock request which we wish to grant
 252  *
 253  * Returns: true if its ok to grant the lock
 254  */
 255
 256 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
 257 {
 258         const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
 259         if ((gh->gh_state == LM_ST_EXCLUSIVE ||
 260              gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
 261                 return 0;
 262         if (gl->gl_state == gh->gh_state)
 263                 return 1;
 264         if (gh->gh_flags & GL_EXACT)
 265                 return 0;
 266         if (gl->gl_state == LM_ST_EXCLUSIVE) {
 267                 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
 268                         return 1;
 269                 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
 270                         return 1;
 271         }
 272         if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
 273                 return 1;
 274         return 0;
 275 }
 276
 277 static void gfs2_holder_wake(struct gfs2_holder *gh)
 278 {
 279         clear_bit(HIF_WAIT, &gh->gh_iflags);
 280         smp_mb__after_atomic();
 281         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
 282 }
 283
 284 /**
 285  * do_error - Something unexpected has happened during a lock request
 286  *
 287  */
 288
 289 static inline void do_error(struct gfs2_glock *gl, const int ret)
 290 {
 291         struct gfs2_holder *gh, *tmp;
 292
 293         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
 294                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 295                         continue;
 296                 if (ret & LM_OUT_ERROR)
 297                         gh->gh_error = -EIO;
 298                 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
 299                         gh->gh_error = GLR_TRYFAILED;
 300                 else
 301                         continue;
 302                 list_del_init(&gh->gh_list);
 303                 trace_gfs2_glock_queue(gh, 0);
 304                 gfs2_holder_wake(gh);
 305         }
 306 }
 307
 308 /**
 309  * do_promote - promote as many requests as possible on the current queue
 310  * @gl: The glock
 311  *
 312  * Returns: 1 if there is a blocked holder at the head of the list, or 2
 313  *          if a type specific operation is underway.
 314  */
 315
 316 static int do_promote(struct gfs2_glock *gl)
 317 __releases(&gl->gl_spin)
 318 __acquires(&gl->gl_spin)
 319 {
 320         const struct gfs2_glock_operations *glops = gl->gl_ops;
 321         struct gfs2_holder *gh, *tmp;
 322         int ret;
 323
 324 restart:
 325         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
 326                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 327                         continue;
 328                 if (may_grant(gl, gh)) {
 329                         if (gh->gh_list.prev == &gl->gl_holders &&
 330                             glops->go_lock) {
 331                                 spin_unlock(&gl->gl_spin);
 332                                 /* FIXME: eliminate this eventually */
 333                                 ret = glops->go_lock(gh);
 334                                 spin_lock(&gl->gl_spin);
 335                                 if (ret) {
 336                                         if (ret == 1)
 337                                                 return 2;
 338                                         gh->gh_error = ret;
 339                                         list_del_init(&gh->gh_list);
 340                                         trace_gfs2_glock_queue(gh, 0);
 341                                         gfs2_holder_wake(gh);
 342                                         goto restart;
 343                                 }
 344                                 set_bit(HIF_HOLDER, &gh->gh_iflags);
 345                                 trace_gfs2_promote(gh, 1);
 346                                 gfs2_holder_wake(gh);
 347                                 goto restart;
 348                         }
 349                         set_bit(HIF_HOLDER, &gh->gh_iflags);
 350                         trace_gfs2_promote(gh, 0);
 351                         gfs2_holder_wake(gh);
 352                         continue;
 353                 }
 354                 if (gh->gh_list.prev == &gl->gl_holders)
 355                         return 1;
 356                 do_error(gl, 0);
 357                 break;
 358         }
 359         return 0;
 360 }
 361
 362 /**
 363  * find_first_waiter - find the first gh that's waiting for the glock
 364  * @gl: the glock
 365  */
 366
 367 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
 368 {
 369         struct gfs2_holder *gh;
 370
 371         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
 372                 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
 373                         return gh;
 374         }
 375         return NULL;
 376 }
 377
 378 /**
 379  * state_change - record that the glock is now in a different state
 380  * @gl: the glock
 381  * @new_state the new state
 382  *
 383  */
 384
 385 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
 386 {
 387         int held1, held2;
 388
 389         held1 = (gl->gl_state != LM_ST_UNLOCKED);
 390         held2 = (new_state != LM_ST_UNLOCKED);
 391
 392         if (held1 != held2) {
 393                 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
 394                 if (held2)
 395                         gl->gl_lockref.count++;
 396                 else
 397                         gl->gl_lockref.count--;
 398         }
 399         if (held1 && held2 && list_empty(&gl->gl_holders))
 400                 clear_bit(GLF_QUEUED, &gl->gl_flags);
 401
 402         if (new_state != gl->gl_target)
 403                 /* shorten our minimum hold time */
 404                 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
 405                                        GL_GLOCK_MIN_HOLD);
 406         gl->gl_state = new_state;
 407         gl->gl_tchange = jiffies;
 408 }
 409
 410 static void gfs2_demote_wake(struct gfs2_glock *gl)
 411 {
 412         gl->gl_demote_state = LM_ST_EXCLUSIVE;
 413         clear_bit(GLF_DEMOTE, &gl->gl_flags);
 414         smp_mb__after_atomic();
 415         wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
 416 }
 417
 418 /**
 419  * finish_xmote - The DLM has replied to one of our lock requests
 420  * @gl: The glock
 421  * @ret: The status from the DLM
 422  *
 423  */
 424
 425 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
 426 {
 427         const struct gfs2_glock_operations *glops = gl->gl_ops;
 428         struct gfs2_holder *gh;
 429         unsigned state = ret & LM_OUT_ST_MASK;
 430         int rv;
 431
 432         spin_lock(&gl->gl_spin);
 433         trace_gfs2_glock_state_change(gl, state);
 434         state_change(gl, state);
 435         gh = find_first_waiter(gl);
 436
 437         /* Demote to UN request arrived during demote to SH or DF */
 438         if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
 439             state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
 440                 gl->gl_target = LM_ST_UNLOCKED;
 441
 442         /* Check for state != intended state */
 443         if (unlikely(state != gl->gl_target)) {
 444                 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
 445                         /* move to back of queue and try next entry */
 446                         if (ret & LM_OUT_CANCELED) {
 447                                 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
 448                                         list_move_tail(&gh->gh_list, &gl->gl_holders);
 449                                 gh = find_first_waiter(gl);
 450                                 gl->gl_target = gh->gh_state;
 451                                 goto retry;
 452                         }
 453                         /* Some error or failed "try lock" - report it */
 454                         if ((ret & LM_OUT_ERROR) ||
 455                             (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
 456                                 gl->gl_target = gl->gl_state;
 457                                 do_error(gl, ret);
 458                                 goto out;
 459                         }
 460                 }
 461                 switch(state) {
 462                 /* Unlocked due to conversion deadlock, try again */
 463                 case LM_ST_UNLOCKED:
 464 retry:
 465                         do_xmote(gl, gh, gl->gl_target);
 466                         break;
 467                 /* Conversion fails, unlock and try again */
 468                 case LM_ST_SHARED:
 469                 case LM_ST_DEFERRED:
 470                         do_xmote(gl, gh, LM_ST_UNLOCKED);
 471                         break;
 472                 default: /* Everything else */
 473                         pr_err("wanted %u got %u\n", gl->gl_target, state);
 474                         GLOCK_BUG_ON(gl, 1);
 475                 }
 476                 spin_unlock(&gl->gl_spin);
 477                 return;
 478         }
 479
 480         /* Fast path - we got what we asked for */
 481         if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
 482                 gfs2_demote_wake(gl);
 483         if (state != LM_ST_UNLOCKED) {
 484                 if (glops->go_xmote_bh) {
 485                         spin_unlock(&gl->gl_spin);
 486                         rv = glops->go_xmote_bh(gl, gh);
 487                         spin_lock(&gl->gl_spin);
 488                         if (rv) {
 489                                 do_error(gl, rv);
 490                                 goto out;
 491                         }
 492                 }
 493                 rv = do_promote(gl);
 494                 if (rv == 2)
 495                         goto out_locked;
 496         }
 497 out:
 498         clear_bit(GLF_LOCK, &gl->gl_flags);
 499 out_locked:
 500         spin_unlock(&gl->gl_spin);
 501 }
 502
 503 /**
 504  * do_xmote - Calls the DLM to change the state of a lock
 505  * @gl: The lock state
 506  * @gh: The holder (only for promotes)
 507  * @target: The target lock state
 508  *
 509  */
 510
 511 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
 512 __releases(&gl->gl_spin)
 513 __acquires(&gl->gl_spin)
 514 {
 515         const struct gfs2_glock_operations *glops = gl->gl_ops;
 516         struct gfs2_sbd *sdp = gl->gl_sbd;
 517         unsigned int lck_flags = gh ? gh->gh_flags : 0;
 518         int ret;
 519
 520         lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
 521                       LM_FLAG_PRIORITY);
 522         GLOCK_BUG_ON(gl, gl->gl_state == target);
 523         GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
 524         if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
 525             glops->go_inval) {
 526                 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 527                 do_error(gl, 0); /* Fail queued try locks */
 528         }
 529         gl->gl_req = target;
 530         set_bit(GLF_BLOCKING, &gl->gl_flags);
 531         if ((gl->gl_req == LM_ST_UNLOCKED) ||
 532             (gl->gl_state == LM_ST_EXCLUSIVE) ||
 533             (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
 534                 clear_bit(GLF_BLOCKING, &gl->gl_flags);
 535         spin_unlock(&gl->gl_spin);
 536         if (glops->go_sync)
 537                 glops->go_sync(gl);
 538         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 539                 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
 540         clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 541
 542         gfs2_glock_hold(gl);
 543         if (sdp->sd_lockstruct.ls_ops->lm_lock) {
 544                 /* lock_dlm */
 545                 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
 546                 if (ret) {
 547                         pr_err("lm_lock ret %d\n", ret);
 548                         GLOCK_BUG_ON(gl, 1);
 549                 }
 550         } else { /* lock_nolock */
 551                 finish_xmote(gl, target);
 552                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 553                         gfs2_glock_put(gl);
 554         }
 555
 556         spin_lock(&gl->gl_spin);
 557 }
 558
 559 /**
 560  * find_first_holder - find the first "holder" gh
 561  * @gl: the glock
 562  */
 563
 564 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
 565 {
 566         struct gfs2_holder *gh;
 567
 568         if (!list_empty(&gl->gl_holders)) {
 569                 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
 570                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 571                         return gh;
 572         }
 573         return NULL;
 574 }
 575
 576 /**
 577  * run_queue - do all outstanding tasks related to a glock
 578  * @gl: The glock in question
 579  * @nonblock: True if we must not block in run_queue
 580  *
 581  */
 582
 583 static void run_queue(struct gfs2_glock *gl, const int nonblock)
 584 __releases(&gl->gl_spin)
 585 __acquires(&gl->gl_spin)
 586 {
 587         struct gfs2_holder *gh = NULL;
 588         int ret;
 589
 590         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
 591                 return;
 592
 593         GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
 594
 595         if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
 596             gl->gl_demote_state != gl->gl_state) {
 597                 if (find_first_holder(gl))
 598                         goto out_unlock;
 599                 if (nonblock)
 600                         goto out_sched;
 601                 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
 602                 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
 603                 gl->gl_target = gl->gl_demote_state;
 604         } else {
 605                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
 606                         gfs2_demote_wake(gl);
 607                 ret = do_promote(gl);
 608                 if (ret == 0)
 609                         goto out_unlock;
 610                 if (ret == 2)
 611                         goto out;
 612                 gh = find_first_waiter(gl);
 613                 gl->gl_target = gh->gh_state;
 614                 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
 615                         do_error(gl, 0); /* Fail queued try locks */
 616         }
 617         do_xmote(gl, gh, gl->gl_target);
 618 out:
 619         return;
 620
 621 out_sched:
 622         clear_bit(GLF_LOCK, &gl->gl_flags);
 623         smp_mb__after_atomic();
 624         gl->gl_lockref.count++;
 625         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 626                 gl->gl_lockref.count--;
 627         return;
 628
 629 out_unlock:
 630         clear_bit(GLF_LOCK, &gl->gl_flags);
 631         smp_mb__after_atomic();
 632         return;
 633 }
 634
 635 static void delete_work_func(struct work_struct *work)
 636 {
 637         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
 638         struct gfs2_sbd *sdp = gl->gl_sbd;
 639         struct gfs2_inode *ip;
 640         struct inode *inode;
 641         u64 no_addr = gl->gl_name.ln_number;
 642
 643         ip = gl->gl_object;
 644         /* Note: Unsafe to dereference ip as we don't hold right refs/locks */
 645
 646         if (ip)
 647                 inode = gfs2_ilookup(sdp->sd_vfs, no_addr, 1);
 648         else
 649                 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
 650         if (inode && !IS_ERR(inode)) {
 651                 d_prune_aliases(inode);
 652                 iput(inode);
 653         }
 654         gfs2_glock_put(gl);
 655 }
 656
 657 static void glock_work_func(struct work_struct *work)
 658 {
 659         unsigned long delay = 0;
 660         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
 661         int drop_ref = 0;
 662
 663         if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
 664                 finish_xmote(gl, gl->gl_reply);
 665                 drop_ref = 1;
 666         }
 667         spin_lock(&gl->gl_spin);
 668         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
 669             gl->gl_state != LM_ST_UNLOCKED &&
 670             gl->gl_demote_state != LM_ST_EXCLUSIVE) {
 671                 unsigned long holdtime, now = jiffies;
 672
 673                 holdtime = gl->gl_tchange + gl->gl_hold_time;
 674                 if (time_before(now, holdtime))
 675                         delay = holdtime - now;
 676
 677                 if (!delay) {
 678                         clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
 679                         set_bit(GLF_DEMOTE, &gl->gl_flags);
 680                 }
 681         }
 682         run_queue(gl, 0);
 683         spin_unlock(&gl->gl_spin);
 684         if (!delay)
 685                 gfs2_glock_put(gl);
 686         else {
 687                 if (gl->gl_name.ln_type != LM_TYPE_INODE)
 688                         delay = 0;
 689                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
 690                         gfs2_glock_put(gl);
 691         }
 692         if (drop_ref)
 693                 gfs2_glock_put(gl);
 694 }
 695
 696 /**
 697  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
 698  * @sdp: The GFS2 superblock
 699  * @number: the lock number
 700  * @glops: The glock_operations to use
 701  * @create: If 0, don't create the glock if it doesn't exist
 702  * @glp: the glock is returned here
 703  *
 704  * This does not lock a glock, just finds/creates structures for one.
 705  *
 706  * Returns: errno
 707  */
 708
 709 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
 710                    const struct gfs2_glock_operations *glops, int create,
 711                    struct gfs2_glock **glp)
 712 {
 713         struct super_block *s = sdp->sd_vfs;
 714         struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
 715         struct gfs2_glock *gl, *tmp;
 716         unsigned int hash = gl_hash(sdp, &name);
 717         struct address_space *mapping;
 718         struct kmem_cache *cachep;
 719
 720         rcu_read_lock();
 721         gl = search_bucket(hash, sdp, &name);
 722         rcu_read_unlock();
 723
 724         *glp = gl;
 725         if (gl)
 726                 return 0;
 727         if (!create)
 728                 return -ENOENT;
 729
 730         if (glops->go_flags & GLOF_ASPACE)
 731                 cachep = gfs2_glock_aspace_cachep;
 732         else
 733                 cachep = gfs2_glock_cachep;
 734         gl = kmem_cache_alloc(cachep, GFP_NOFS);
 735         if (!gl)
 736                 return -ENOMEM;
 737
 738         memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
 739
 740         if (glops->go_flags & GLOF_LVB) {
 741                 gl->gl_lksb.sb_lvbptr = kzalloc(GFS2_MIN_LVB_SIZE, GFP_NOFS);
 742                 if (!gl->gl_lksb.sb_lvbptr) {
 743                         kmem_cache_free(cachep, gl);
 744                         return -ENOMEM;
 745                 }
 746         }
 747
 748         atomic_inc(&sdp->sd_glock_disposal);
 749         gl->gl_sbd = sdp;
 750         gl->gl_flags = 0;
 751         gl->gl_name = name;
 752         gl->gl_lockref.count = 1;
 753         gl->gl_state = LM_ST_UNLOCKED;
 754         gl->gl_target = LM_ST_UNLOCKED;
 755         gl->gl_demote_state = LM_ST_EXCLUSIVE;
 756         gl->gl_hash = hash;
 757         gl->gl_ops = glops;
 758         gl->gl_dstamp = ktime_set(0, 0);
 759         preempt_disable();
 760         /* We use the global stats to estimate the initial per-glock stats */
 761         gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
 762         preempt_enable();
 763         gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
 764         gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
 765         gl->gl_tchange = jiffies;
 766         gl->gl_object = NULL;
 767         gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
 768         INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
 769         INIT_WORK(&gl->gl_delete, delete_work_func);
 770
 771         mapping = gfs2_glock2aspace(gl);
 772         if (mapping) {
 773                 mapping->a_ops = &gfs2_meta_aops;
 774                 mapping->host = s->s_bdev->bd_inode;
 775                 mapping->flags = 0;
 776                 mapping_set_gfp_mask(mapping, GFP_NOFS);
 777                 mapping->private_data = NULL;
 778                 mapping->backing_dev_info = s->s_bdi;
 779                 mapping->writeback_index = 0;
 780         }
 781
 782         spin_lock_bucket(hash);
 783         tmp = search_bucket(hash, sdp, &name);
 784         if (tmp) {
 785                 spin_unlock_bucket(hash);
 786                 kfree(gl->gl_lksb.sb_lvbptr);
 787                 kmem_cache_free(cachep, gl);
 788                 atomic_dec(&sdp->sd_glock_disposal);
 789                 gl = tmp;
 790         } else {
 791                 hlist_bl_add_head_rcu(&gl->gl_list, &gl_hash_table[hash]);
 792                 spin_unlock_bucket(hash);
 793         }
 794
 795         *glp = gl;
 796
 797         return 0;
 798 }
 799
 800 /**
 801  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
 802  * @gl: the glock
 803  * @state: the state we're requesting
 804  * @flags: the modifier flags
 805  * @gh: the holder structure
 806  *
 807  */
 808
 809 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
 810                       struct gfs2_holder *gh)
 811 {
 812         INIT_LIST_HEAD(&gh->gh_list);
 813         gh->gh_gl = gl;
 814         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 815         gh->gh_owner_pid = get_pid(task_pid(current));
 816         gh->gh_state = state;
 817         gh->gh_flags = flags;
 818         gh->gh_error = 0;
 819         gh->gh_iflags = 0;
 820         gfs2_glock_hold(gl);
 821 }
 822
 823 /**
 824  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
 825  * @state: the state we're requesting
 826  * @flags: the modifier flags
 827  * @gh: the holder structure
 828  *
 829  * Don't mess with the glock.
 830  *
 831  */
 832
 833 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
 834 {
 835         gh->gh_state = state;
 836         gh->gh_flags = flags;
 837         gh->gh_iflags = 0;
 838         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 839         if (gh->gh_owner_pid)
 840                 put_pid(gh->gh_owner_pid);
 841         gh->gh_owner_pid = get_pid(task_pid(current));
 842 }
 843
 844 /**
 845  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
 846  * @gh: the holder structure
 847  *
 848  */
 849
 850 void gfs2_holder_uninit(struct gfs2_holder *gh)
 851 {
 852         put_pid(gh->gh_owner_pid);
 853         gfs2_glock_put(gh->gh_gl);
 854         gh->gh_gl = NULL;
 855         gh->gh_ip = 0;
 856 }
 857
 858 /**
 859  * gfs2_glock_wait - wait on a glock acquisition
 860  * @gh: the glock holder
 861  *
 862  * Returns: 0 on success
 863  */
 864
 865 int gfs2_glock_wait(struct gfs2_holder *gh)
 866 {
 867         unsigned long time1 = jiffies;
 868
 869         might_sleep();
 870         wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
 871         if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
 872                 /* Lengthen the minimum hold time. */
 873                 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
 874                                               GL_GLOCK_HOLD_INCR,
 875                                               GL_GLOCK_MAX_HOLD);
 876         return gh->gh_error;
 877 }
 878
 879 /**
 880  * handle_callback - process a demote request
 881  * @gl: the glock
 882  * @state: the state the caller wants us to change to
 883  *
 884  * There are only two requests that we are going to see in actual
 885  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
 886  */
 887
 888 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
 889                             unsigned long delay, bool remote)
 890 {
 891         int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
 892
 893         set_bit(bit, &gl->gl_flags);
 894         if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
 895                 gl->gl_demote_state = state;
 896                 gl->gl_demote_time = jiffies;
 897         } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
 898                         gl->gl_demote_state != state) {
 899                 gl->gl_demote_state = LM_ST_UNLOCKED;
 900         }
 901         if (gl->gl_ops->go_callback)
 902                 gl->gl_ops->go_callback(gl, remote);
 903         trace_gfs2_demote_rq(gl, remote);
 904 }
 905
 906 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
 907 {
 908         struct va_format vaf;
 909         va_list args;
 910
 911         va_start(args, fmt);
 912
 913         if (seq) {
 914                 seq_vprintf(seq, fmt, args);
 915         } else {
 916                 vaf.fmt = fmt;
 917                 vaf.va = &args;
 918
 919                 pr_err("%pV", &vaf);
 920         }
 921
 922         va_end(args);
 923 }
 924
 925 /**
 926  * add_to_queue - Add a holder to the wait queue (but look for recursion)
 927  * @gh: the holder structure to add
 928  *
 929  * Eventually we should move the recursive locking trap to a
 930  * debugging option or something like that. This is the fast
 931  * path and needs to have the minimum number of distractions.
 932  *
 933  */
 934
 935 static inline void add_to_queue(struct gfs2_holder *gh)
 936 __releases(&gl->gl_spin)
 937 __acquires(&gl->gl_spin)
 938 {
 939         struct gfs2_glock *gl = gh->gh_gl;
 940         struct gfs2_sbd *sdp = gl->gl_sbd;
 941         struct list_head *insert_pt = NULL;
 942         struct gfs2_holder *gh2;
 943         int try_futile = 0;
 944
 945         BUG_ON(gh->gh_owner_pid == NULL);
 946         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
 947                 BUG();
 948
 949         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
 950                 if (test_bit(GLF_LOCK, &gl->gl_flags))
 951                         try_futile = !may_grant(gl, gh);
 952                 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 953                         goto fail;
 954         }
 955
 956         list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
 957                 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
 958                     (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
 959                         goto trap_recursive;
 960                 if (try_futile &&
 961                     !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
 962 fail:
 963                         gh->gh_error = GLR_TRYFAILED;
 964                         gfs2_holder_wake(gh);
 965                         return;
 966                 }
 967                 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
 968                         continue;
 969                 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
 970                         insert_pt = &gh2->gh_list;
 971         }
 972         set_bit(GLF_QUEUED, &gl->gl_flags);
 973         trace_gfs2_glock_queue(gh, 1);
 974         gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
 975         gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
 976         if (likely(insert_pt == NULL)) {
 977                 list_add_tail(&gh->gh_list, &gl->gl_holders);
 978                 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
 979                         goto do_cancel;
 980                 return;
 981         }
 982         list_add_tail(&gh->gh_list, insert_pt);
 983 do_cancel:
 984         gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
 985         if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
 986                 spin_unlock(&gl->gl_spin);
 987                 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
 988                         sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
 989                 spin_lock(&gl->gl_spin);
 990         }
 991         return;
 992
 993 trap_recursive:
 994         pr_err("original: %pSR\n", (void *)gh2->gh_ip);
 995         pr_err("pid: %d\n", pid_nr(gh2->gh_owner_pid));
 996         pr_err("lock type: %d req lock state : %d\n",
 997                gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
 998         pr_err("new: %pSR\n", (void *)gh->gh_ip);
 999         pr_err("pid: %d\n", pid_nr(gh->gh_owner_pid));
1000         pr_err("lock type: %d req lock state : %d\n",
1001                gh->gh_gl->gl_name.ln_type, gh->gh_state);
1002         gfs2_dump_glock(NULL, gl);
1003         BUG();
1004 }
1005
1006 /**
1007  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1008  * @gh: the holder structure
1009  *
1010  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1011  *
1012  * Returns: 0, GLR_TRYFAILED, or errno on failure
1013  */
1014
1015 int gfs2_glock_nq(struct gfs2_holder *gh)
1016 {
1017         struct gfs2_glock *gl = gh->gh_gl;
1018         struct gfs2_sbd *sdp = gl->gl_sbd;
1019         int error = 0;
1020
1021         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1022                 return -EIO;
1023
1024         if (test_bit(GLF_LRU, &gl->gl_flags))
1025                 gfs2_glock_remove_from_lru(gl);
1026
1027         spin_lock(&gl->gl_spin);
1028         add_to_queue(gh);
1029         if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1030                      test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1031                 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1032                 gl->gl_lockref.count++;
1033                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1034                         gl->gl_lockref.count--;
1035         }
1036         run_queue(gl, 1);
1037         spin_unlock(&gl->gl_spin);
1038
1039         if (!(gh->gh_flags & GL_ASYNC))
1040                 error = gfs2_glock_wait(gh);
1041
1042         return error;
1043 }
1044
1045 /**
1046  * gfs2_glock_poll - poll to see if an async request has been completed
1047  * @gh: the holder
1048  *
1049  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1050  */
1051
1052 int gfs2_glock_poll(struct gfs2_holder *gh)
1053 {
1054         return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1055 }
1056
1057 /**
1058  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1059  * @gh: the glock holder
1060  *
1061  */
1062
1063 void gfs2_glock_dq(struct gfs2_holder *gh)
1064 {
1065         struct gfs2_glock *gl = gh->gh_gl;
1066         const struct gfs2_glock_operations *glops = gl->gl_ops;
1067         unsigned delay = 0;
1068         int fast_path = 0;
1069
1070         spin_lock(&gl->gl_spin);
1071         if (gh->gh_flags & GL_NOCACHE)
1072                 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1073
1074         list_del_init(&gh->gh_list);
1075         if (find_first_holder(gl) == NULL) {
1076                 if (glops->go_unlock) {
1077                         GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1078                         spin_unlock(&gl->gl_spin);
1079                         glops->go_unlock(gh);
1080                         spin_lock(&gl->gl_spin);
1081                         clear_bit(GLF_LOCK, &gl->gl_flags);
1082                 }
1083                 if (list_empty(&gl->gl_holders) &&
1084                     !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1085                     !test_bit(GLF_DEMOTE, &gl->gl_flags))
1086                         fast_path = 1;
1087         }
1088         if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1089                 gfs2_glock_add_to_lru(gl);
1090
1091         trace_gfs2_glock_queue(gh, 0);
1092         spin_unlock(&gl->gl_spin);
1093         if (likely(fast_path))
1094                 return;
1095
1096         gfs2_glock_hold(gl);
1097         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1098             !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1099             gl->gl_name.ln_type == LM_TYPE_INODE)
1100                 delay = gl->gl_hold_time;
1101         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1102                 gfs2_glock_put(gl);
1103 }
1104
1105 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1106 {
1107         struct gfs2_glock *gl = gh->gh_gl;
1108         gfs2_glock_dq(gh);
1109         might_sleep();
1110         wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1111 }
1112
1113 /**
1114  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1115  * @gh: the holder structure
1116  *
1117  */
1118
1119 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1120 {
1121         gfs2_glock_dq(gh);
1122         gfs2_holder_uninit(gh);
1123 }
1124
1125 /**
1126  * gfs2_glock_nq_num - acquire a glock based on lock number
1127  * @sdp: the filesystem
1128  * @number: the lock number
1129  * @glops: the glock operations for the type of glock
1130  * @state: the state to acquire the glock in
1131  * @flags: modifier flags for the acquisition
1132  * @gh: the struct gfs2_holder
1133  *
1134  * Returns: errno
1135  */
1136
1137 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1138                       const struct gfs2_glock_operations *glops,
1139                       unsigned int state, int flags, struct gfs2_holder *gh)
1140 {
1141         struct gfs2_glock *gl;
1142         int error;
1143
1144         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1145         if (!error) {
1146                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1147                 gfs2_glock_put(gl);
1148         }
1149
1150         return error;
1151 }
1152
1153 /**
1154  * glock_compare - Compare two struct gfs2_glock structures for sorting
1155  * @arg_a: the first structure
1156  * @arg_b: the second structure
1157  *
1158  */
1159
1160 static int glock_compare(const void *arg_a, const void *arg_b)
1161 {
1162         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1163         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1164         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1165         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1166
1167         if (a->ln_number > b->ln_number)
1168                 return 1;
1169         if (a->ln_number < b->ln_number)
1170                 return -1;
1171         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1172         return 0;
1173 }
1174
1175 /**
1176  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1177  * @num_gh: the number of structures
1178  * @ghs: an array of struct gfs2_holder structures
1179  *
1180  * Returns: 0 on success (all glocks acquired),
1181  *          errno on failure (no glocks acquired)
1182  */
1183
1184 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1185                      struct gfs2_holder **p)
1186 {
1187         unsigned int x;
1188         int error = 0;
1189
1190         for (x = 0; x < num_gh; x++)
1191                 p[x] = &ghs[x];
1192
1193         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1194
1195         for (x = 0; x < num_gh; x++) {
1196                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1197
1198                 error = gfs2_glock_nq(p[x]);
1199                 if (error) {
1200                         while (x--)
1201                                 gfs2_glock_dq(p[x]);
1202                         break;
1203                 }
1204         }
1205
1206         return error;
1207 }
1208
1209 /**
1210  * gfs2_glock_nq_m - acquire multiple glocks
1211  * @num_gh: the number of structures
1212  * @ghs: an array of struct gfs2_holder structures
1213  *
1214  *
1215  * Returns: 0 on success (all glocks acquired),
1216  *          errno on failure (no glocks acquired)
1217  */
1218
1219 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1220 {
1221         struct gfs2_holder *tmp[4];
1222         struct gfs2_holder **pph = tmp;
1223         int error = 0;
1224
1225         switch(num_gh) {
1226         case 0:
1227                 return 0;
1228         case 1:
1229                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1230                 return gfs2_glock_nq(ghs);
1231         default:
1232                 if (num_gh <= 4)
1233                         break;
1234                 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1235                 if (!pph)
1236                         return -ENOMEM;
1237         }
1238
1239         error = nq_m_sync(num_gh, ghs, pph);
1240
1241         if (pph != tmp)
1242                 kfree(pph);
1243
1244         return error;
1245 }
1246
1247 /**
1248  * gfs2_glock_dq_m - release multiple glocks
1249  * @num_gh: the number of structures
1250  * @ghs: an array of struct gfs2_holder structures
1251  *
1252  */
1253
1254 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1255 {
1256         while (num_gh--)
1257                 gfs2_glock_dq(&ghs[num_gh]);
1258 }
1259
1260 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1261 {
1262         unsigned long delay = 0;
1263         unsigned long holdtime;
1264         unsigned long now = jiffies;
1265
1266         gfs2_glock_hold(gl);
1267         holdtime = gl->gl_tchange + gl->gl_hold_time;
1268         if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1269             gl->gl_name.ln_type == LM_TYPE_INODE) {
1270                 if (time_before(now, holdtime))
1271                         delay = holdtime - now;
1272                 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1273                         delay = gl->gl_hold_time;
1274         }
1275
1276         spin_lock(&gl->gl_spin);
1277         handle_callback(gl, state, delay, true);
1278         spin_unlock(&gl->gl_spin);
1279         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1280                 gfs2_glock_put(gl);
1281 }
1282
1283 /**
1284  * gfs2_should_freeze - Figure out if glock should be frozen
1285  * @gl: The glock in question
1286  *
1287  * Glocks are not frozen if (a) the result of the dlm operation is
1288  * an error, (b) the locking operation was an unlock operation or
1289  * (c) if there is a "noexp" flagged request anywhere in the queue
1290  *
1291  * Returns: 1 if freezing should occur, 0 otherwise
1292  */
1293
1294 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1295 {
1296         const struct gfs2_holder *gh;
1297
1298         if (gl->gl_reply & ~LM_OUT_ST_MASK)
1299                 return 0;
1300         if (gl->gl_target == LM_ST_UNLOCKED)
1301                 return 0;
1302
1303         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1304                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1305                         continue;
1306                 if (LM_FLAG_NOEXP & gh->gh_flags)
1307                         return 0;
1308         }
1309
1310         return 1;
1311 }
1312
1313 /**
1314  * gfs2_glock_complete - Callback used by locking
1315  * @gl: Pointer to the glock
1316  * @ret: The return value from the dlm
1317  *
1318  * The gl_reply field is under the gl_spin lock so that it is ok
1319  * to use a bitfield shared with other glock state fields.
1320  */
1321
1322 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1323 {
1324         struct lm_lockstruct *ls = &gl->gl_sbd->sd_lockstruct;
1325
1326         spin_lock(&gl->gl_spin);
1327         gl->gl_reply = ret;
1328
1329         if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1330                 if (gfs2_should_freeze(gl)) {
1331                         set_bit(GLF_FROZEN, &gl->gl_flags);
1332                         spin_unlock(&gl->gl_spin);
1333                         return;
1334                 }
1335         }
1336
1337         gl->gl_lockref.count++;
1338         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1339         spin_unlock(&gl->gl_spin);
1340
1341         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1342                 gfs2_glock_put(gl);
1343 }
1344
1345 static int glock_cmp(void *priv, struct list_head *a, struct list_head *b)
1346 {
1347         struct gfs2_glock *gla, *glb;
1348
1349         gla = list_entry(a, struct gfs2_glock, gl_lru);
1350         glb = list_entry(b, struct gfs2_glock, gl_lru);
1351
1352         if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1353                 return 1;
1354         if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1355                 return -1;
1356
1357         return 0;
1358 }
1359
1360 /**
1361  * gfs2_dispose_glock_lru - Demote a list of glocks
1362  * @list: The list to dispose of
1363  *
1364  * Disposing of glocks may involve disk accesses, so that here we sort
1365  * the glocks by number (i.e. disk location of the inodes) so that if
1366  * there are any such accesses, they'll be sent in order (mostly).
1367  *
1368  * Must be called under the lru_lock, but may drop and retake this
1369  * lock. While the lru_lock is dropped, entries may vanish from the
1370  * list, but no new entries will appear on the list (since it is
1371  * private)
1372  */
1373
1374 static void gfs2_dispose_glock_lru(struct list_head *list)
1375 __releases(&lru_lock)
1376 __acquires(&lru_lock)
1377 {
1378         struct gfs2_glock *gl;
1379
1380         list_sort(NULL, list, glock_cmp);
1381
1382         while(!list_empty(list)) {
1383                 gl = list_entry(list->next, struct gfs2_glock, gl_lru);
1384                 list_del_init(&gl->gl_lru);
1385                 if (!spin_trylock(&gl->gl_spin)) {
1386 add_back_to_lru:
1387                         list_add(&gl->gl_lru, &lru_list);
1388                         atomic_inc(&lru_count);
1389                         continue;
1390                 }
1391                 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1392                         spin_unlock(&gl->gl_spin);
1393                         goto add_back_to_lru;
1394                 }
1395                 clear_bit(GLF_LRU, &gl->gl_flags);
1396                 gl->gl_lockref.count++;
1397                 if (demote_ok(gl))
1398                         handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1399                 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1400                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1401                         gl->gl_lockref.count--;
1402                 spin_unlock(&gl->gl_spin);
1403                 cond_resched_lock(&lru_lock);
1404         }
1405 }
1406
1407 /**
1408  * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1409  * @nr: The number of entries to scan
1410  *
1411  * This function selects the entries on the LRU which are able to
1412  * be demoted, and then kicks off the process by calling
1413  * gfs2_dispose_glock_lru() above.
1414  */
1415
1416 static long gfs2_scan_glock_lru(int nr)
1417 {
1418         struct gfs2_glock *gl;
1419         LIST_HEAD(skipped);
1420         LIST_HEAD(dispose);
1421         long freed = 0;
1422
1423         spin_lock(&lru_lock);
1424         while ((nr-- >= 0) && !list_empty(&lru_list)) {
1425                 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1426
1427                 /* Test for being demotable */
1428                 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1429                         list_move(&gl->gl_lru, &dispose);
1430                         atomic_dec(&lru_count);
1431                         freed++;
1432                         continue;
1433                 }
1434
1435                 list_move(&gl->gl_lru, &skipped);
1436         }
1437         list_splice(&skipped, &lru_list);
1438         if (!list_empty(&dispose))
1439                 gfs2_dispose_glock_lru(&dispose);
1440         spin_unlock(&lru_lock);
1441
1442         return freed;
1443 }
1444
1445 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1446                                             struct shrink_control *sc)
1447 {
1448         if (!(sc->gfp_mask & __GFP_FS))
1449                 return SHRINK_STOP;
1450         return gfs2_scan_glock_lru(sc->nr_to_scan);
1451 }
1452
1453 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1454                                              struct shrink_control *sc)
1455 {
1456         return vfs_pressure_ratio(atomic_read(&lru_count));
1457 }
1458
1459 static struct shrinker glock_shrinker = {
1460         .seeks = DEFAULT_SEEKS,
1461         .count_objects = gfs2_glock_shrink_count,
1462         .scan_objects = gfs2_glock_shrink_scan,
1463 };
1464
1465 /**
1466  * examine_bucket - Call a function for glock in a hash bucket
1467  * @examiner: the function
1468  * @sdp: the filesystem
1469  * @bucket: the bucket
1470  *
1471  */
1472
1473 static void examine_bucket(glock_examiner examiner, const struct gfs2_sbd *sdp,
1474                           unsigned int hash)
1475 {
1476         struct gfs2_glock *gl;
1477         struct hlist_bl_head *head = &gl_hash_table[hash];
1478         struct hlist_bl_node *pos;
1479
1480         rcu_read_lock();
1481         hlist_bl_for_each_entry_rcu(gl, pos, head, gl_list) {
1482                 if ((gl->gl_sbd == sdp) && lockref_get_not_dead(&gl->gl_lockref))
1483                         examiner(gl);
1484         }
1485         rcu_read_unlock();
1486         cond_resched();
1487 }
1488
1489 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1490 {
1491         unsigned x;
1492
1493         for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1494                 examine_bucket(examiner, sdp, x);
1495 }
1496
1497
1498 /**
1499  * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1500  * @gl: The glock to thaw
1501  *
1502  */
1503
1504 static void thaw_glock(struct gfs2_glock *gl)
1505 {
1506         if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1507                 goto out;
1508         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1509         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0) {
1510 out:
1511                 gfs2_glock_put(gl);
1512         }
1513 }
1514
1515 /**
1516  * clear_glock - look at a glock and see if we can free it from glock cache
1517  * @gl: the glock to look at
1518  *
1519  */
1520
1521 static void clear_glock(struct gfs2_glock *gl)
1522 {
1523         gfs2_glock_remove_from_lru(gl);
1524
1525         spin_lock(&gl->gl_spin);
1526         if (gl->gl_state != LM_ST_UNLOCKED)
1527                 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1528         spin_unlock(&gl->gl_spin);
1529         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1530                 gfs2_glock_put(gl);
1531 }
1532
1533 /**
1534  * gfs2_glock_thaw - Thaw any frozen glocks
1535  * @sdp: The super block
1536  *
1537  */
1538
1539 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1540 {
1541         glock_hash_walk(thaw_glock, sdp);
1542 }
1543
1544 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1545 {
1546         spin_lock(&gl->gl_spin);
1547         gfs2_dump_glock(seq, gl);
1548         spin_unlock(&gl->gl_spin);
1549 }
1550
1551 static void dump_glock_func(struct gfs2_glock *gl)
1552 {
1553         dump_glock(NULL, gl);
1554 }
1555
1556 /**
1557  * gfs2_gl_hash_clear - Empty out the glock hash table
1558  * @sdp: the filesystem
1559  * @wait: wait until it's all gone
1560  *
1561  * Called when unmounting the filesystem.
1562  */
1563
1564 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1565 {
1566         set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1567         flush_workqueue(glock_workqueue);
1568         glock_hash_walk(clear_glock, sdp);
1569         flush_workqueue(glock_workqueue);
1570         wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
1571         glock_hash_walk(dump_glock_func, sdp);
1572 }
1573
1574 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1575 {
1576         struct gfs2_glock *gl = ip->i_gl;
1577         int ret;
1578
1579         ret = gfs2_truncatei_resume(ip);
1580         gfs2_assert_withdraw(gl->gl_sbd, ret == 0);
1581
1582         spin_lock(&gl->gl_spin);
1583         clear_bit(GLF_LOCK, &gl->gl_flags);
1584         run_queue(gl, 1);
1585         spin_unlock(&gl->gl_spin);
1586 }
1587
1588 static const char *state2str(unsigned state)
1589 {
1590         switch(state) {
1591         case LM_ST_UNLOCKED:
1592                 return "UN";
1593         case LM_ST_SHARED:
1594                 return "SH";
1595         case LM_ST_DEFERRED:
1596                 return "DF";
1597         case LM_ST_EXCLUSIVE:
1598                 return "EX";
1599         }
1600         return "??";
1601 }
1602
1603 static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1604 {
1605         char *p = buf;
1606         if (flags & LM_FLAG_TRY)
1607                 *p++ = 't';
1608         if (flags & LM_FLAG_TRY_1CB)
1609                 *p++ = 'T';
1610         if (flags & LM_FLAG_NOEXP)
1611                 *p++ = 'e';
1612         if (flags & LM_FLAG_ANY)
1613                 *p++ = 'A';
1614         if (flags & LM_FLAG_PRIORITY)
1615                 *p++ = 'p';
1616         if (flags & GL_ASYNC)
1617                 *p++ = 'a';
1618         if (flags & GL_EXACT)
1619                 *p++ = 'E';
1620         if (flags & GL_NOCACHE)
1621                 *p++ = 'c';
1622         if (test_bit(HIF_HOLDER, &iflags))
1623                 *p++ = 'H';
1624         if (test_bit(HIF_WAIT, &iflags))
1625                 *p++ = 'W';
1626         if (test_bit(HIF_FIRST, &iflags))
1627                 *p++ = 'F';
1628         *p = 0;
1629         return buf;
1630 }
1631
1632 /**
1633  * dump_holder - print information about a glock holder
1634  * @seq: the seq_file struct
1635  * @gh: the glock holder
1636  *
1637  */
1638
1639 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1640 {
1641         struct task_struct *gh_owner = NULL;
1642         char flags_buf[32];
1643
1644         rcu_read_lock();
1645         if (gh->gh_owner_pid)
1646                 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1647         gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1648                        state2str(gh->gh_state),
1649                        hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1650                        gh->gh_error,
1651                        gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1652                        gh_owner ? gh_owner->comm : "(ended)",
1653                        (void *)gh->gh_ip);
1654         rcu_read_unlock();
1655 }
1656
1657 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1658 {
1659         const unsigned long *gflags = &gl->gl_flags;
1660         char *p = buf;
1661
1662         if (test_bit(GLF_LOCK, gflags))
1663                 *p++ = 'l';
1664         if (test_bit(GLF_DEMOTE, gflags))
1665                 *p++ = 'D';
1666         if (test_bit(GLF_PENDING_DEMOTE, gflags))
1667                 *p++ = 'd';
1668         if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1669                 *p++ = 'p';
1670         if (test_bit(GLF_DIRTY, gflags))
1671                 *p++ = 'y';
1672         if (test_bit(GLF_LFLUSH, gflags))
1673                 *p++ = 'f';
1674         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1675                 *p++ = 'i';
1676         if (test_bit(GLF_REPLY_PENDING, gflags))
1677                 *p++ = 'r';
1678         if (test_bit(GLF_INITIAL, gflags))
1679                 *p++ = 'I';
1680         if (test_bit(GLF_FROZEN, gflags))
1681                 *p++ = 'F';
1682         if (test_bit(GLF_QUEUED, gflags))
1683                 *p++ = 'q';
1684         if (test_bit(GLF_LRU, gflags))
1685                 *p++ = 'L';
1686         if (gl->gl_object)
1687                 *p++ = 'o';
1688         if (test_bit(GLF_BLOCKING, gflags))
1689                 *p++ = 'b';
1690         *p = 0;
1691         return buf;
1692 }
1693
1694 /**
1695  * gfs2_dump_glock - print information about a glock
1696  * @seq: The seq_file struct
1697  * @gl: the glock
1698  *
1699  * The file format is as follows:
1700  * One line per object, capital letters are used to indicate objects
1701  * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1702  * other objects are indented by a single space and follow the glock to
1703  * which they are related. Fields are indicated by lower case letters
1704  * followed by a colon and the field value, except for strings which are in
1705  * [] so that its possible to see if they are composed of spaces for
1706  * example. The field's are n = number (id of the object), f = flags,
1707  * t = type, s = state, r = refcount, e = error, p = pid.
1708  *
1709  */
1710
1711 void gfs2_dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1712 {
1713         const struct gfs2_glock_operations *glops = gl->gl_ops;
1714         unsigned long long dtime;
1715         const struct gfs2_holder *gh;
1716         char gflags_buf[32];
1717
1718         dtime = jiffies - gl->gl_demote_time;
1719         dtime *= 1000000/HZ; /* demote time in uSec */
1720         if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1721                 dtime = 0;
1722         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",
1723                   state2str(gl->gl_state),
1724                   gl->gl_name.ln_type,
1725                   (unsigned long long)gl->gl_name.ln_number,
1726                   gflags2str(gflags_buf, gl),
1727                   state2str(gl->gl_target),
1728                   state2str(gl->gl_demote_state), dtime,
1729                   atomic_read(&gl->gl_ail_count),
1730                   atomic_read(&gl->gl_revokes),
1731                   (int)gl->gl_lockref.count, gl->gl_hold_time);
1732
1733         list_for_each_entry(gh, &gl->gl_holders, gh_list)
1734                 dump_holder(seq, gh);
1735
1736         if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1737                 glops->go_dump(seq, gl);
1738 }
1739
1740 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
1741 {
1742         struct gfs2_glock *gl = iter_ptr;
1743
1744         seq_printf(seq, "G: n:%u/%llx rtt:%lld/%lld rttb:%lld/%lld irt:%lld/%lld dcnt: %lld qcnt: %lld\n",
1745                    gl->gl_name.ln_type,
1746                    (unsigned long long)gl->gl_name.ln_number,
1747                    (long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
1748                    (long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
1749                    (long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
1750                    (long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
1751                    (long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
1752                    (long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
1753                    (long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
1754                    (long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
1755         return 0;
1756 }
1757
1758 static const char *gfs2_gltype[] = {
1759         "type",
1760         "reserved",
1761         "nondisk",
1762         "inode",
1763         "rgrp",
1764         "meta",
1765         "iopen",
1766         "flock",
1767         "plock",
1768         "quota",
1769         "journal",
1770 };
1771
1772 static const char *gfs2_stype[] = {
1773         [GFS2_LKS_SRTT]         = "srtt",
1774         [GFS2_LKS_SRTTVAR]      = "srttvar",
1775         [GFS2_LKS_SRTTB]        = "srttb",
1776         [GFS2_LKS_SRTTVARB]     = "srttvarb",
1777         [GFS2_LKS_SIRT]         = "sirt",
1778         [GFS2_LKS_SIRTVAR]      = "sirtvar",
1779         [GFS2_LKS_DCOUNT]       = "dlm",
1780         [GFS2_LKS_QCOUNT]       = "queue",
1781 };
1782
1783 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1784
1785 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
1786 {
1787         struct gfs2_glock_iter *gi = seq->private;
1788         struct gfs2_sbd *sdp = gi->sdp;
1789         unsigned index = gi->hash >> 3;
1790         unsigned subindex = gi->hash & 0x07;
1791         s64 value;
1792         int i;
1793
1794         if (index == 0 && subindex != 0)
1795                 return 0;
1796
1797         seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
1798                    (index == 0) ? "cpu": gfs2_stype[subindex]);
1799
1800         for_each_possible_cpu(i) {
1801                 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
1802                 if (index == 0) {
1803                         value = i;
1804                 } else {
1805                         value = lkstats->lkstats[index - 1].stats[subindex];
1806                 }
1807                 seq_printf(seq, " %15lld", (long long)value);
1808         }
1809         seq_putc(seq, '\n');
1810         return 0;
1811 }
1812
1813 int __init gfs2_glock_init(void)
1814 {
1815         unsigned i;
1816         for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
1817                 INIT_HLIST_BL_HEAD(&gl_hash_table[i]);
1818         }
1819
1820         glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1821                                           WQ_HIGHPRI | WQ_FREEZABLE, 0);
1822         if (!glock_workqueue)
1823                 return -ENOMEM;
1824         gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1825                                                 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1826                                                 0);
1827         if (!gfs2_delete_workqueue) {
1828                 destroy_workqueue(glock_workqueue);
1829                 return -ENOMEM;
1830         }
1831
1832         register_shrinker(&glock_shrinker);
1833
1834         return 0;
1835 }
1836
1837 void gfs2_glock_exit(void)
1838 {
1839         unregister_shrinker(&glock_shrinker);
1840         destroy_workqueue(glock_workqueue);
1841         destroy_workqueue(gfs2_delete_workqueue);
1842 }
1843
1844 static inline struct gfs2_glock *glock_hash_chain(unsigned hash)
1845 {
1846         return hlist_bl_entry(hlist_bl_first_rcu(&gl_hash_table[hash]),
1847                               struct gfs2_glock, gl_list);
1848 }
1849
1850 static inline struct gfs2_glock *glock_hash_next(struct gfs2_glock *gl)
1851 {
1852         return hlist_bl_entry(rcu_dereference(gl->gl_list.next),
1853                               struct gfs2_glock, gl_list);
1854 }
1855
1856 static int gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1857 {
1858         struct gfs2_glock *gl;
1859
1860         do {
1861                 gl = gi->gl;
1862                 if (gl) {
1863                         gi->gl = glock_hash_next(gl);
1864                         gi->nhash++;
1865                 } else {
1866                         if (gi->hash >= GFS2_GL_HASH_SIZE) {
1867                                 rcu_read_unlock();
1868                                 return 1;
1869                         }
1870                         gi->gl = glock_hash_chain(gi->hash);
1871                         gi->nhash = 0;
1872                 }
1873                 while (gi->gl == NULL) {
1874                         gi->hash++;
1875                         if (gi->hash >= GFS2_GL_HASH_SIZE) {
1876                                 rcu_read_unlock();
1877                                 return 1;
1878                         }
1879                         gi->gl = glock_hash_chain(gi->hash);
1880                         gi->nhash = 0;
1881                 }
1882         /* Skip entries for other sb and dead entries */
1883         } while (gi->sdp != gi->gl->gl_sbd ||
1884                  __lockref_is_dead(&gi->gl->gl_lockref));
1885
1886         return 0;
1887 }
1888
1889 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1890 {
1891         struct gfs2_glock_iter *gi = seq->private;
1892         loff_t n = *pos;
1893
1894         if (gi->last_pos <= *pos)
1895                 n = gi->nhash + (*pos - gi->last_pos);
1896         else
1897                 gi->hash = 0;
1898
1899         gi->nhash = 0;
1900         rcu_read_lock();
1901
1902         do {
1903                 if (gfs2_glock_iter_next(gi))
1904                         return NULL;
1905         } while (n--);
1906
1907         gi->last_pos = *pos;
1908         return gi->gl;
1909 }
1910
1911 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1912                                  loff_t *pos)
1913 {
1914         struct gfs2_glock_iter *gi = seq->private;
1915
1916         (*pos)++;
1917         gi->last_pos = *pos;
1918         if (gfs2_glock_iter_next(gi))
1919                 return NULL;
1920
1921         return gi->gl;
1922 }
1923
1924 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1925 {
1926         struct gfs2_glock_iter *gi = seq->private;
1927
1928         if (gi->gl)
1929                 rcu_read_unlock();
1930         gi->gl = NULL;
1931 }
1932
1933 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1934 {
1935         dump_glock(seq, iter_ptr);
1936         return 0;
1937 }
1938
1939 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
1940 {
1941         struct gfs2_glock_iter *gi = seq->private;
1942
1943         gi->hash = *pos;
1944         if (*pos >= GFS2_NR_SBSTATS)
1945                 return NULL;
1946         preempt_disable();
1947         return SEQ_START_TOKEN;
1948 }
1949
1950 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
1951                                    loff_t *pos)
1952 {
1953         struct gfs2_glock_iter *gi = seq->private;
1954         (*pos)++;
1955         gi->hash++;
1956         if (gi->hash >= GFS2_NR_SBSTATS) {
1957                 preempt_enable();
1958                 return NULL;
1959         }
1960         return SEQ_START_TOKEN;
1961 }
1962
1963 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
1964 {
1965         preempt_enable();
1966 }
1967
1968 static const struct seq_operations gfs2_glock_seq_ops = {
1969         .start = gfs2_glock_seq_start,
1970         .next  = gfs2_glock_seq_next,
1971         .stop  = gfs2_glock_seq_stop,
1972         .show  = gfs2_glock_seq_show,
1973 };
1974
1975 static const struct seq_operations gfs2_glstats_seq_ops = {
1976         .start = gfs2_glock_seq_start,
1977         .next  = gfs2_glock_seq_next,
1978         .stop  = gfs2_glock_seq_stop,
1979         .show  = gfs2_glstats_seq_show,
1980 };
1981
1982 static const struct seq_operations gfs2_sbstats_seq_ops = {
1983         .start = gfs2_sbstats_seq_start,
1984         .next  = gfs2_sbstats_seq_next,
1985         .stop  = gfs2_sbstats_seq_stop,
1986         .show  = gfs2_sbstats_seq_show,
1987 };
1988
1989 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
1990
1991 static int gfs2_glocks_open(struct inode *inode, struct file *file)
1992 {
1993         int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1994                                    sizeof(struct gfs2_glock_iter));
1995         if (ret == 0) {
1996                 struct seq_file *seq = file->private_data;
1997                 struct gfs2_glock_iter *gi = seq->private;
1998                 gi->sdp = inode->i_private;
1999                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2000                 if (seq->buf)
2001                         seq->size = GFS2_SEQ_GOODSIZE;
2002         }
2003         return ret;
2004 }
2005
2006 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2007 {
2008         int ret = seq_open_private(file, &gfs2_glstats_seq_ops,
2009                                    sizeof(struct gfs2_glock_iter));
2010         if (ret == 0) {
2011                 struct seq_file *seq = file->private_data;
2012                 struct gfs2_glock_iter *gi = seq->private;
2013                 gi->sdp = inode->i_private;
2014                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2015                 if (seq->buf)
2016                         seq->size = GFS2_SEQ_GOODSIZE;
2017         }
2018         return ret;
2019 }
2020
2021 static int gfs2_sbstats_open(struct inode *inode, struct file *file)
2022 {
2023         int ret = seq_open_private(file, &gfs2_sbstats_seq_ops,
2024                                    sizeof(struct gfs2_glock_iter));
2025         if (ret == 0) {
2026                 struct seq_file *seq = file->private_data;
2027                 struct gfs2_glock_iter *gi = seq->private;
2028                 gi->sdp = inode->i_private;
2029         }
2030         return ret;
2031 }
2032
2033 static const struct file_operations gfs2_glocks_fops = {
2034         .owner   = THIS_MODULE,
2035         .open    = gfs2_glocks_open,
2036         .read    = seq_read,
2037         .llseek  = seq_lseek,
2038         .release = seq_release_private,
2039 };
2040
2041 static const struct file_operations gfs2_glstats_fops = {
2042         .owner   = THIS_MODULE,
2043         .open    = gfs2_glstats_open,
2044         .read    = seq_read,
2045         .llseek  = seq_lseek,
2046         .release = seq_release_private,
2047 };
2048
2049 static const struct file_operations gfs2_sbstats_fops = {
2050         .owner   = THIS_MODULE,
2051         .open    = gfs2_sbstats_open,
2052         .read    = seq_read,
2053         .llseek  = seq_lseek,
2054         .release = seq_release_private,
2055 };
2056
2057 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2058 {
2059         sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2060         if (!sdp->debugfs_dir)
2061                 return -ENOMEM;
2062         sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
2063                                                          S_IFREG | S_IRUGO,
2064                                                          sdp->debugfs_dir, sdp,
2065                                                          &gfs2_glocks_fops);
2066         if (!sdp->debugfs_dentry_glocks)
2067                 goto fail;
2068
2069         sdp->debugfs_dentry_glstats = debugfs_create_file("glstats",
2070                                                         S_IFREG | S_IRUGO,
2071                                                         sdp->debugfs_dir, sdp,
2072                                                         &gfs2_glstats_fops);
2073         if (!sdp->debugfs_dentry_glstats)
2074                 goto fail;
2075
2076         sdp->debugfs_dentry_sbstats = debugfs_create_file("sbstats",
2077                                                         S_IFREG | S_IRUGO,
2078                                                         sdp->debugfs_dir, sdp,
2079                                                         &gfs2_sbstats_fops);
2080         if (!sdp->debugfs_dentry_sbstats)
2081                 goto fail;
2082
2083         return 0;
2084 fail:
2085         gfs2_delete_debugfs_file(sdp);
2086         return -ENOMEM;
2087 }
2088
2089 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2090 {
2091         if (sdp->debugfs_dir) {
2092                 if (sdp->debugfs_dentry_glocks) {
2093                         debugfs_remove(sdp->debugfs_dentry_glocks);
2094                         sdp->debugfs_dentry_glocks = NULL;
2095                 }
2096                 if (sdp->debugfs_dentry_glstats) {
2097                         debugfs_remove(sdp->debugfs_dentry_glstats);
2098                         sdp->debugfs_dentry_glstats = NULL;
2099                 }
2100                 if (sdp->debugfs_dentry_sbstats) {
2101                         debugfs_remove(sdp->debugfs_dentry_sbstats);
2102                         sdp->debugfs_dentry_sbstats = NULL;
2103                 }
2104                 debugfs_remove(sdp->debugfs_dir);
2105                 sdp->debugfs_dir = NULL;
2106         }
2107 }
2108
2109 int gfs2_register_debugfs(void)
2110 {
2111         gfs2_root = debugfs_create_dir("gfs2", NULL);
2112         return gfs2_root ? 0 : -ENOMEM;
2113 }
2114
2115 void gfs2_unregister_debugfs(void)
2116 {
2117         debugfs_remove(gfs2_root);
2118         gfs2_root = NULL;
2119 }