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 #include <linux/sched.h>
  11 #include <linux/slab.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/buffer_head.h>
  14 #include <linux/delay.h>
  15 #include <linux/sort.h>
  16 #include <linux/jhash.h>
  17 #include <linux/kallsyms.h>
  18 #include <linux/gfs2_ondisk.h>
  19 #include <linux/list.h>
  20 #include <linux/wait.h>
  21 #include <linux/module.h>
  22 #include <asm/uaccess.h>
  23 #include <linux/seq_file.h>
  24 #include <linux/debugfs.h>
  25 #include <linux/kthread.h>
  26 #include <linux/freezer.h>
  27 #include <linux/workqueue.h>
  28 #include <linux/jiffies.h>
  29 #include <linux/rcupdate.h>
  30 #include <linux/rculist_bl.h>
  31 #include <linux/bit_spinlock.h>
  32 #include <linux/percpu.h>
  33 #include <linux/list_sort.h>
  34
  35 #include "gfs2.h"
  36 #include "incore.h"
  37 #include "glock.h"
  38 #include "glops.h"
  39 #include "inode.h"
  40 #include "lops.h"
  41 #include "meta_io.h"
  42 #include "quota.h"
  43 #include "super.h"
  44 #include "util.h"
  45 #include "bmap.h"
  46 #define CREATE_TRACE_POINTS
  47 #include "trace_gfs2.h"
  48
  49 struct gfs2_glock_iter {
  50         int hash;                       /* hash bucket index           */
  51         unsigned nhash;                 /* Index within current bucket */
  52         struct gfs2_sbd *sdp;           /* incore superblock           */
  53         struct gfs2_glock *gl;          /* current glock struct        */
  54         loff_t last_pos;                /* last position               */
  55 };
  56
  57 typedef void (*glock_examiner) (struct gfs2_glock * gl);
  58
  59 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
  60
  61 static struct dentry *gfs2_root;
  62 static struct workqueue_struct *glock_workqueue;
  63 struct workqueue_struct *gfs2_delete_workqueue;
  64 static LIST_HEAD(lru_list);
  65 static atomic_t lru_count = ATOMIC_INIT(0);
  66 static DEFINE_SPINLOCK(lru_lock);
  67
  68 #define GFS2_GL_HASH_SHIFT      15
  69 #define GFS2_GL_HASH_SIZE       (1 << GFS2_GL_HASH_SHIFT)
  70 #define GFS2_GL_HASH_MASK       (GFS2_GL_HASH_SIZE - 1)
  71
  72 static struct hlist_bl_head gl_hash_table[GFS2_GL_HASH_SIZE];
  73 static struct dentry *gfs2_root;
  74
  75 /**
  76  * gl_hash() - Turn glock number into hash bucket number
  77  * @lock: The glock number
  78  *
  79  * Returns: The number of the corresponding hash bucket
  80  */
  81
  82 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
  83                             const struct lm_lockname *name)
  84 {
  85         unsigned int h;
  86
  87         h = jhash(&name->ln_number, sizeof(u64), 0);
  88         h = jhash(&name->ln_type, sizeof(unsigned int), h);
  89         h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
  90         h &= GFS2_GL_HASH_MASK;
  91
  92         return h;
  93 }
  94
  95 static inline void spin_lock_bucket(unsigned int hash)
  96 {
  97         hlist_bl_lock(&gl_hash_table[hash]);
  98 }
  99
 100 static inline void spin_unlock_bucket(unsigned int hash)
 101 {
 102         hlist_bl_unlock(&gl_hash_table[hash]);
 103 }
 104
 105 static void gfs2_glock_dealloc(struct rcu_head *rcu)
 106 {
 107         struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
 108
 109         if (gl->gl_ops->go_flags & GLOF_ASPACE) {
 110                 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
 111         } else {
 112                 kfree(gl->gl_lksb.sb_lvbptr);
 113                 kmem_cache_free(gfs2_glock_cachep, gl);
 114         }
 115 }
 116
 117 void gfs2_glock_free(struct gfs2_glock *gl)
 118 {
 119         struct gfs2_sbd *sdp = gl->gl_sbd;
 120
 121         call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
 122         if (atomic_dec_and_test(&sdp->sd_glock_disposal))
 123                 wake_up(&sdp->sd_glock_wait);
 124 }
 125
 126 /**
 127  * gfs2_glock_hold() - increment reference count on glock
 128  * @gl: The glock to hold
 129  *
 130  */
 131
 132 void gfs2_glock_hold(struct gfs2_glock *gl)
 133 {
 134         GLOCK_BUG_ON(gl, atomic_read(&gl->gl_ref) == 0);
 135         atomic_inc(&gl->gl_ref);
 136 }
 137
 138 /**
 139  * demote_ok - Check to see if it's ok to unlock a glock
 140  * @gl: the glock
 141  *
 142  * Returns: 1 if it's ok
 143  */
 144
 145 static int demote_ok(const struct gfs2_glock *gl)
 146 {
 147         const struct gfs2_glock_operations *glops = gl->gl_ops;
 148
 149         if (gl->gl_state == LM_ST_UNLOCKED)
 150                 return 0;
 151         if (!list_empty(&gl->gl_holders))
 152                 return 0;
 153         if (glops->go_demote_ok)
 154                 return glops->go_demote_ok(gl);
 155         return 1;
 156 }
 157
 158
 159 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
 160 {
 161         spin_lock(&lru_lock);
 162
 163         if (!list_empty(&gl->gl_lru))
 164                 list_del_init(&gl->gl_lru);
 165         else
 166                 atomic_inc(&lru_count);
 167
 168         list_add_tail(&gl->gl_lru, &lru_list);
 169         set_bit(GLF_LRU, &gl->gl_flags);
 170         spin_unlock(&lru_lock);
 171 }
 172
 173 static void __gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
 174 {
 175         if (!list_empty(&gl->gl_lru)) {
 176                 list_del_init(&gl->gl_lru);
 177                 atomic_dec(&lru_count);
 178                 clear_bit(GLF_LRU, &gl->gl_flags);
 179         }
 180 }
 181
 182 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
 183 {
 184         spin_lock(&lru_lock);
 185         __gfs2_glock_remove_from_lru(gl);
 186         spin_unlock(&lru_lock);
 187 }
 188
 189 /**
 190  * gfs2_glock_put_nolock() - Decrement reference count on glock
 191  * @gl: The glock to put
 192  *
 193  * This function should only be used if the caller has its own reference
 194  * to the glock, in addition to the one it is dropping.
 195  */
 196
 197 void gfs2_glock_put_nolock(struct gfs2_glock *gl)
 198 {
 199         if (atomic_dec_and_test(&gl->gl_ref))
 200                 GLOCK_BUG_ON(gl, 1);
 201 }
 202
 203 /**
 204  * gfs2_glock_put() - Decrement reference count on glock
 205  * @gl: The glock to put
 206  *
 207  */
 208
 209 void gfs2_glock_put(struct gfs2_glock *gl)
 210 {
 211         struct gfs2_sbd *sdp = gl->gl_sbd;
 212         struct address_space *mapping = gfs2_glock2aspace(gl);
 213
 214         if (atomic_dec_and_lock(&gl->gl_ref, &lru_lock)) {
 215                 __gfs2_glock_remove_from_lru(gl);
 216                 spin_unlock(&lru_lock);
 217                 spin_lock_bucket(gl->gl_hash);
 218                 hlist_bl_del_rcu(&gl->gl_list);
 219                 spin_unlock_bucket(gl->gl_hash);
 220                 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
 221                 GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
 222                 trace_gfs2_glock_put(gl);
 223                 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
 224         }
 225 }
 226
 227 /**
 228  * search_bucket() - Find struct gfs2_glock by lock number
 229  * @bucket: the bucket to search
 230  * @name: The lock name
 231  *
 232  * Returns: NULL, or the struct gfs2_glock with the requested number
 233  */
 234
 235 static struct gfs2_glock *search_bucket(unsigned int hash,
 236                                         const struct gfs2_sbd *sdp,
 237                                         const struct lm_lockname *name)
 238 {
 239         struct gfs2_glock *gl;
 240         struct hlist_bl_node *h;
 241
 242         hlist_bl_for_each_entry_rcu(gl, h, &gl_hash_table[hash], gl_list) {
 243                 if (!lm_name_equal(&gl->gl_name, name))
 244                         continue;
 245                 if (gl->gl_sbd != sdp)
 246                         continue;
 247                 if (atomic_inc_not_zero(&gl->gl_ref))
 248                         return gl;
 249         }
 250
 251         return NULL;
 252 }
 253
 254 /**
 255  * may_grant - check if its ok to grant a new lock
 256  * @gl: The glock
 257  * @gh: The lock request which we wish to grant
 258  *
 259  * Returns: true if its ok to grant the lock
 260  */
 261
 262 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
 263 {
 264         const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
 265         if ((gh->gh_state == LM_ST_EXCLUSIVE ||
 266              gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
 267                 return 0;
 268         if (gl->gl_state == gh->gh_state)
 269                 return 1;
 270         if (gh->gh_flags & GL_EXACT)
 271                 return 0;
 272         if (gl->gl_state == LM_ST_EXCLUSIVE) {
 273                 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
 274                         return 1;
 275                 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
 276                         return 1;
 277         }
 278         if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
 279                 return 1;
 280         return 0;
 281 }
 282
 283 static void gfs2_holder_wake(struct gfs2_holder *gh)
 284 {
 285         clear_bit(HIF_WAIT, &gh->gh_iflags);
 286         smp_mb__after_clear_bit();
 287         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
 288 }
 289
 290 /**
 291  * do_error - Something unexpected has happened during a lock request
 292  *
 293  */
 294
 295 static inline void do_error(struct gfs2_glock *gl, const int ret)
 296 {
 297         struct gfs2_holder *gh, *tmp;
 298
 299         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
 300                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 301                         continue;
 302                 if (ret & LM_OUT_ERROR)
 303                         gh->gh_error = -EIO;
 304                 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
 305                         gh->gh_error = GLR_TRYFAILED;
 306                 else
 307                         continue;
 308                 list_del_init(&gh->gh_list);
 309                 trace_gfs2_glock_queue(gh, 0);
 310                 gfs2_holder_wake(gh);
 311         }
 312 }
 313
 314 /**
 315  * do_promote - promote as many requests as possible on the current queue
 316  * @gl: The glock
 317  *
 318  * Returns: 1 if there is a blocked holder at the head of the list, or 2
 319  *          if a type specific operation is underway.
 320  */
 321
 322 static int do_promote(struct gfs2_glock *gl)
 323 __releases(&gl->gl_spin)
 324 __acquires(&gl->gl_spin)
 325 {
 326         const struct gfs2_glock_operations *glops = gl->gl_ops;
 327         struct gfs2_holder *gh, *tmp;
 328         int ret;
 329
 330 restart:
 331         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
 332                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 333                         continue;
 334                 if (may_grant(gl, gh)) {
 335                         if (gh->gh_list.prev == &gl->gl_holders &&
 336                             glops->go_lock) {
 337                                 spin_unlock(&gl->gl_spin);
 338                                 /* FIXME: eliminate this eventually */
 339                                 ret = glops->go_lock(gh);
 340                                 spin_lock(&gl->gl_spin);
 341                                 if (ret) {
 342                                         if (ret == 1)
 343                                                 return 2;
 344                                         gh->gh_error = ret;
 345                                         list_del_init(&gh->gh_list);
 346                                         trace_gfs2_glock_queue(gh, 0);
 347                                         gfs2_holder_wake(gh);
 348                                         goto restart;
 349                                 }
 350                                 set_bit(HIF_HOLDER, &gh->gh_iflags);
 351                                 trace_gfs2_promote(gh, 1);
 352                                 gfs2_holder_wake(gh);
 353                                 goto restart;
 354                         }
 355                         set_bit(HIF_HOLDER, &gh->gh_iflags);
 356                         trace_gfs2_promote(gh, 0);
 357                         gfs2_holder_wake(gh);
 358                         continue;
 359                 }
 360                 if (gh->gh_list.prev == &gl->gl_holders)
 361                         return 1;
 362                 do_error(gl, 0);
 363                 break;
 364         }
 365         return 0;
 366 }
 367
 368 /**
 369  * find_first_waiter - find the first gh that's waiting for the glock
 370  * @gl: the glock
 371  */
 372
 373 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
 374 {
 375         struct gfs2_holder *gh;
 376
 377         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
 378                 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
 379                         return gh;
 380         }
 381         return NULL;
 382 }
 383
 384 /**
 385  * state_change - record that the glock is now in a different state
 386  * @gl: the glock
 387  * @new_state the new state
 388  *
 389  */
 390
 391 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
 392 {
 393         int held1, held2;
 394
 395         held1 = (gl->gl_state != LM_ST_UNLOCKED);
 396         held2 = (new_state != LM_ST_UNLOCKED);
 397
 398         if (held1 != held2) {
 399                 if (held2)
 400                         gfs2_glock_hold(gl);
 401                 else
 402                         gfs2_glock_put_nolock(gl);
 403         }
 404         if (held1 && held2 && list_empty(&gl->gl_holders))
 405                 clear_bit(GLF_QUEUED, &gl->gl_flags);
 406
 407         if (new_state != gl->gl_target)
 408                 /* shorten our minimum hold time */
 409                 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
 410                                        GL_GLOCK_MIN_HOLD);
 411         gl->gl_state = new_state;
 412         gl->gl_tchange = jiffies;
 413 }
 414
 415 static void gfs2_demote_wake(struct gfs2_glock *gl)
 416 {
 417         gl->gl_demote_state = LM_ST_EXCLUSIVE;
 418         clear_bit(GLF_DEMOTE, &gl->gl_flags);
 419         smp_mb__after_clear_bit();
 420         wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
 421 }
 422
 423 /**
 424  * finish_xmote - The DLM has replied to one of our lock requests
 425  * @gl: The glock
 426  * @ret: The status from the DLM
 427  *
 428  */
 429
 430 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
 431 {
 432         const struct gfs2_glock_operations *glops = gl->gl_ops;
 433         struct gfs2_holder *gh;
 434         unsigned state = ret & LM_OUT_ST_MASK;
 435         int rv;
 436
 437         spin_lock(&gl->gl_spin);
 438         trace_gfs2_glock_state_change(gl, state);
 439         state_change(gl, state);
 440         gh = find_first_waiter(gl);
 441
 442         /* Demote to UN request arrived during demote to SH or DF */
 443         if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
 444             state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
 445                 gl->gl_target = LM_ST_UNLOCKED;
 446
 447         /* Check for state != intended state */
 448         if (unlikely(state != gl->gl_target)) {
 449                 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
 450                         /* move to back of queue and try next entry */
 451                         if (ret & LM_OUT_CANCELED) {
 452                                 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
 453                                         list_move_tail(&gh->gh_list, &gl->gl_holders);
 454                                 gh = find_first_waiter(gl);
 455                                 gl->gl_target = gh->gh_state;
 456                                 goto retry;
 457                         }
 458                         /* Some error or failed "try lock" - report it */
 459                         if ((ret & LM_OUT_ERROR) ||
 460                             (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
 461                                 gl->gl_target = gl->gl_state;
 462                                 do_error(gl, ret);
 463                                 goto out;
 464                         }
 465                 }
 466                 switch(state) {
 467                 /* Unlocked due to conversion deadlock, try again */
 468                 case LM_ST_UNLOCKED:
 469 retry:
 470                         do_xmote(gl, gh, gl->gl_target);
 471                         break;
 472                 /* Conversion fails, unlock and try again */
 473                 case LM_ST_SHARED:
 474                 case LM_ST_DEFERRED:
 475                         do_xmote(gl, gh, LM_ST_UNLOCKED);
 476                         break;
 477                 default: /* Everything else */
 478                         printk(KERN_ERR "GFS2: wanted %u got %u\n", gl->gl_target, state);
 479                         GLOCK_BUG_ON(gl, 1);
 480                 }
 481                 spin_unlock(&gl->gl_spin);
 482                 return;
 483         }
 484
 485         /* Fast path - we got what we asked for */
 486         if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
 487                 gfs2_demote_wake(gl);
 488         if (state != LM_ST_UNLOCKED) {
 489                 if (glops->go_xmote_bh) {
 490                         spin_unlock(&gl->gl_spin);
 491                         rv = glops->go_xmote_bh(gl, gh);
 492                         spin_lock(&gl->gl_spin);
 493                         if (rv) {
 494                                 do_error(gl, rv);
 495                                 goto out;
 496                         }
 497                 }
 498                 rv = do_promote(gl);
 499                 if (rv == 2)
 500                         goto out_locked;
 501         }
 502 out:
 503         clear_bit(GLF_LOCK, &gl->gl_flags);
 504 out_locked:
 505         spin_unlock(&gl->gl_spin);
 506 }
 507
 508 /**
 509  * do_xmote - Calls the DLM to change the state of a lock
 510  * @gl: The lock state
 511  * @gh: The holder (only for promotes)
 512  * @target: The target lock state
 513  *
 514  */
 515
 516 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
 517 __releases(&gl->gl_spin)
 518 __acquires(&gl->gl_spin)
 519 {
 520         const struct gfs2_glock_operations *glops = gl->gl_ops;
 521         struct gfs2_sbd *sdp = gl->gl_sbd;
 522         unsigned int lck_flags = gh ? gh->gh_flags : 0;
 523         int ret;
 524
 525         lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
 526                       LM_FLAG_PRIORITY);
 527         GLOCK_BUG_ON(gl, gl->gl_state == target);
 528         GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
 529         if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
 530             glops->go_inval) {
 531                 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 532                 do_error(gl, 0); /* Fail queued try locks */
 533         }
 534         gl->gl_req = target;
 535         set_bit(GLF_BLOCKING, &gl->gl_flags);
 536         if ((gl->gl_req == LM_ST_UNLOCKED) ||
 537             (gl->gl_state == LM_ST_EXCLUSIVE) ||
 538             (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
 539                 clear_bit(GLF_BLOCKING, &gl->gl_flags);
 540         spin_unlock(&gl->gl_spin);
 541         if (glops->go_sync)
 542                 glops->go_sync(gl);
 543         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 544                 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
 545         clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 546
 547         gfs2_glock_hold(gl);
 548         if (sdp->sd_lockstruct.ls_ops->lm_lock) {
 549                 /* lock_dlm */
 550                 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
 551                 if (ret) {
 552                         printk(KERN_ERR "GFS2: lm_lock ret %d\n", ret);
 553                         GLOCK_BUG_ON(gl, 1);
 554                 }
 555         } else { /* lock_nolock */
 556                 finish_xmote(gl, target);
 557                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 558                         gfs2_glock_put(gl);
 559         }
 560
 561         spin_lock(&gl->gl_spin);
 562 }
 563
 564 /**
 565  * find_first_holder - find the first "holder" gh
 566  * @gl: the glock
 567  */
 568
 569 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
 570 {
 571         struct gfs2_holder *gh;
 572
 573         if (!list_empty(&gl->gl_holders)) {
 574                 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
 575                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 576                         return gh;
 577         }
 578         return NULL;
 579 }
 580
 581 /**
 582  * run_queue - do all outstanding tasks related to a glock
 583  * @gl: The glock in question
 584  * @nonblock: True if we must not block in run_queue
 585  *
 586  */
 587
 588 static void run_queue(struct gfs2_glock *gl, const int nonblock)
 589 __releases(&gl->gl_spin)
 590 __acquires(&gl->gl_spin)
 591 {
 592         struct gfs2_holder *gh = NULL;
 593         int ret;
 594
 595         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
 596                 return;
 597
 598         GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
 599
 600         if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
 601             gl->gl_demote_state != gl->gl_state) {
 602                 if (find_first_holder(gl))
 603                         goto out_unlock;
 604                 if (nonblock)
 605                         goto out_sched;
 606                 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
 607                 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
 608                 gl->gl_target = gl->gl_demote_state;
 609         } else {
 610                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
 611                         gfs2_demote_wake(gl);
 612                 ret = do_promote(gl);
 613                 if (ret == 0)
 614                         goto out_unlock;
 615                 if (ret == 2)
 616                         goto out;
 617                 gh = find_first_waiter(gl);
 618                 gl->gl_target = gh->gh_state;
 619                 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
 620                         do_error(gl, 0); /* Fail queued try locks */
 621         }
 622         do_xmote(gl, gh, gl->gl_target);
 623 out:
 624         return;
 625
 626 out_sched:
 627         clear_bit(GLF_LOCK, &gl->gl_flags);
 628         smp_mb__after_clear_bit();
 629         gfs2_glock_hold(gl);
 630         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 631                 gfs2_glock_put_nolock(gl);
 632         return;
 633
 634 out_unlock:
 635         clear_bit(GLF_LOCK, &gl->gl_flags);
 636         smp_mb__after_clear_bit();
 637         return;
 638 }
 639
 640 static void delete_work_func(struct work_struct *work)
 641 {
 642         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
 643         struct gfs2_sbd *sdp = gl->gl_sbd;
 644         struct gfs2_inode *ip;
 645         struct inode *inode;
 646         u64 no_addr = gl->gl_name.ln_number;
 647
 648         ip = gl->gl_object;
 649         /* Note: Unsafe to dereference ip as we don't hold right refs/locks */
 650
 651         if (ip)
 652                 inode = gfs2_ilookup(sdp->sd_vfs, no_addr, 1);
 653         else
 654                 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
 655         if (inode && !IS_ERR(inode)) {
 656                 d_prune_aliases(inode);
 657                 iput(inode);
 658         }
 659         gfs2_glock_put(gl);
 660 }
 661
 662 static void glock_work_func(struct work_struct *work)
 663 {
 664         unsigned long delay = 0;
 665         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
 666         int drop_ref = 0;
 667
 668         if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
 669                 finish_xmote(gl, gl->gl_reply);
 670                 drop_ref = 1;
 671         }
 672         spin_lock(&gl->gl_spin);
 673         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
 674             gl->gl_state != LM_ST_UNLOCKED &&
 675             gl->gl_demote_state != LM_ST_EXCLUSIVE) {
 676                 unsigned long holdtime, now = jiffies;
 677
 678                 holdtime = gl->gl_tchange + gl->gl_hold_time;
 679                 if (time_before(now, holdtime))
 680                         delay = holdtime - now;
 681
 682                 if (!delay) {
 683                         clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
 684                         set_bit(GLF_DEMOTE, &gl->gl_flags);
 685                 }
 686         }
 687         run_queue(gl, 0);
 688         spin_unlock(&gl->gl_spin);
 689         if (!delay)
 690                 gfs2_glock_put(gl);
 691         else {
 692                 if (gl->gl_name.ln_type != LM_TYPE_INODE)
 693                         delay = 0;
 694                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
 695                         gfs2_glock_put(gl);
 696         }
 697         if (drop_ref)
 698                 gfs2_glock_put(gl);
 699 }
 700
 701 /**
 702  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
 703  * @sdp: The GFS2 superblock
 704  * @number: the lock number
 705  * @glops: The glock_operations to use
 706  * @create: If 0, don't create the glock if it doesn't exist
 707  * @glp: the glock is returned here
 708  *
 709  * This does not lock a glock, just finds/creates structures for one.
 710  *
 711  * Returns: errno
 712  */
 713
 714 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
 715                    const struct gfs2_glock_operations *glops, int create,
 716                    struct gfs2_glock **glp)
 717 {
 718         struct super_block *s = sdp->sd_vfs;
 719         struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
 720         struct gfs2_glock *gl, *tmp;
 721         unsigned int hash = gl_hash(sdp, &name);
 722         struct address_space *mapping;
 723         struct kmem_cache *cachep;
 724
 725         rcu_read_lock();
 726         gl = search_bucket(hash, sdp, &name);
 727         rcu_read_unlock();
 728
 729         *glp = gl;
 730         if (gl)
 731                 return 0;
 732         if (!create)
 733                 return -ENOENT;
 734
 735         if (glops->go_flags & GLOF_ASPACE)
 736                 cachep = gfs2_glock_aspace_cachep;
 737         else
 738                 cachep = gfs2_glock_cachep;
 739         gl = kmem_cache_alloc(cachep, GFP_KERNEL);
 740         if (!gl)
 741                 return -ENOMEM;
 742
 743         memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
 744
 745         if (glops->go_flags & GLOF_LVB) {
 746                 gl->gl_lksb.sb_lvbptr = kzalloc(GFS2_MIN_LVB_SIZE, GFP_KERNEL);
 747                 if (!gl->gl_lksb.sb_lvbptr) {
 748                         kmem_cache_free(cachep, gl);
 749                         return -ENOMEM;
 750                 }
 751         }
 752
 753         atomic_inc(&sdp->sd_glock_disposal);
 754         gl->gl_sbd = sdp;
 755         gl->gl_flags = 0;
 756         gl->gl_name = name;
 757         atomic_set(&gl->gl_ref, 1);
 758         gl->gl_state = LM_ST_UNLOCKED;
 759         gl->gl_target = LM_ST_UNLOCKED;
 760         gl->gl_demote_state = LM_ST_EXCLUSIVE;
 761         gl->gl_hash = hash;
 762         gl->gl_ops = glops;
 763         gl->gl_dstamp = ktime_set(0, 0);
 764         preempt_disable();
 765         /* We use the global stats to estimate the initial per-glock stats */
 766         gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
 767         preempt_enable();
 768         gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
 769         gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
 770         gl->gl_tchange = jiffies;
 771         gl->gl_object = NULL;
 772         gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
 773         INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
 774         INIT_WORK(&gl->gl_delete, delete_work_func);
 775
 776         mapping = gfs2_glock2aspace(gl);
 777         if (mapping) {
 778                 mapping->a_ops = &gfs2_meta_aops;
 779                 mapping->host = s->s_bdev->bd_inode;
 780                 mapping->flags = 0;
 781                 mapping_set_gfp_mask(mapping, GFP_NOFS);
 782                 mapping->private_data = NULL;
 783                 mapping->backing_dev_info = s->s_bdi;
 784                 mapping->writeback_index = 0;
 785         }
 786
 787         spin_lock_bucket(hash);
 788         tmp = search_bucket(hash, sdp, &name);
 789         if (tmp) {
 790                 spin_unlock_bucket(hash);
 791                 kfree(gl->gl_lksb.sb_lvbptr);
 792                 kmem_cache_free(cachep, gl);
 793                 atomic_dec(&sdp->sd_glock_disposal);
 794                 gl = tmp;
 795         } else {
 796                 hlist_bl_add_head_rcu(&gl->gl_list, &gl_hash_table[hash]);
 797                 spin_unlock_bucket(hash);
 798         }
 799
 800         *glp = gl;
 801
 802         return 0;
 803 }
 804
 805 /**
 806  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
 807  * @gl: the glock
 808  * @state: the state we're requesting
 809  * @flags: the modifier flags
 810  * @gh: the holder structure
 811  *
 812  */
 813
 814 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
 815                       struct gfs2_holder *gh)
 816 {
 817         INIT_LIST_HEAD(&gh->gh_list);
 818         gh->gh_gl = gl;
 819         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 820         gh->gh_owner_pid = get_pid(task_pid(current));
 821         gh->gh_state = state;
 822         gh->gh_flags = flags;
 823         gh->gh_error = 0;
 824         gh->gh_iflags = 0;
 825         gfs2_glock_hold(gl);
 826 }
 827
 828 /**
 829  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
 830  * @state: the state we're requesting
 831  * @flags: the modifier flags
 832  * @gh: the holder structure
 833  *
 834  * Don't mess with the glock.
 835  *
 836  */
 837
 838 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
 839 {
 840         gh->gh_state = state;
 841         gh->gh_flags = flags;
 842         gh->gh_iflags = 0;
 843         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 844         if (gh->gh_owner_pid)
 845                 put_pid(gh->gh_owner_pid);
 846         gh->gh_owner_pid = get_pid(task_pid(current));
 847 }
 848
 849 /**
 850  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
 851  * @gh: the holder structure
 852  *
 853  */
 854
 855 void gfs2_holder_uninit(struct gfs2_holder *gh)
 856 {
 857         put_pid(gh->gh_owner_pid);
 858         gfs2_glock_put(gh->gh_gl);
 859         gh->gh_gl = NULL;
 860         gh->gh_ip = 0;
 861 }
 862
 863 /**
 864  * gfs2_glock_holder_wait
 865  * @word: unused
 866  *
 867  * This function and gfs2_glock_demote_wait both show up in the WCHAN
 868  * field. Thus I've separated these otherwise identical functions in
 869  * order to be more informative to the user.
 870  */
 871
 872 static int gfs2_glock_holder_wait(void *word)
 873 {
 874         schedule();
 875         return 0;
 876 }
 877
 878 static int gfs2_glock_demote_wait(void *word)
 879 {
 880         schedule();
 881         return 0;
 882 }
 883
 884 /**
 885  * gfs2_glock_wait - wait on a glock acquisition
 886  * @gh: the glock holder
 887  *
 888  * Returns: 0 on success
 889  */
 890
 891 int gfs2_glock_wait(struct gfs2_holder *gh)
 892 {
 893         unsigned long time1 = jiffies;
 894
 895         might_sleep();
 896         wait_on_bit(&gh->gh_iflags, HIF_WAIT, gfs2_glock_holder_wait, TASK_UNINTERRUPTIBLE);
 897         if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
 898                 /* Lengthen the minimum hold time. */
 899                 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
 900                                               GL_GLOCK_HOLD_INCR,
 901                                               GL_GLOCK_MAX_HOLD);
 902         return gh->gh_error;
 903 }
 904
 905 /**
 906  * handle_callback - process a demote request
 907  * @gl: the glock
 908  * @state: the state the caller wants us to change to
 909  *
 910  * There are only two requests that we are going to see in actual
 911  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
 912  */
 913
 914 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
 915                             unsigned long delay)
 916 {
 917         int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
 918
 919         set_bit(bit, &gl->gl_flags);
 920         if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
 921                 gl->gl_demote_state = state;
 922                 gl->gl_demote_time = jiffies;
 923         } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
 924                         gl->gl_demote_state != state) {
 925                 gl->gl_demote_state = LM_ST_UNLOCKED;
 926         }
 927         if (gl->gl_ops->go_callback)
 928                 gl->gl_ops->go_callback(gl);
 929         trace_gfs2_demote_rq(gl);
 930 }
 931
 932 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
 933 {
 934         struct va_format vaf;
 935         va_list args;
 936
 937         va_start(args, fmt);
 938
 939         if (seq) {
 940                 seq_vprintf(seq, fmt, args);
 941         } else {
 942                 vaf.fmt = fmt;
 943                 vaf.va = &args;
 944
 945                 printk(KERN_ERR " %pV", &vaf);
 946         }
 947
 948         va_end(args);
 949 }
 950
 951 /**
 952  * add_to_queue - Add a holder to the wait queue (but look for recursion)
 953  * @gh: the holder structure to add
 954  *
 955  * Eventually we should move the recursive locking trap to a
 956  * debugging option or something like that. This is the fast
 957  * path and needs to have the minimum number of distractions.
 958  *
 959  */
 960
 961 static inline void add_to_queue(struct gfs2_holder *gh)
 962 __releases(&gl->gl_spin)
 963 __acquires(&gl->gl_spin)
 964 {
 965         struct gfs2_glock *gl = gh->gh_gl;
 966         struct gfs2_sbd *sdp = gl->gl_sbd;
 967         struct list_head *insert_pt = NULL;
 968         struct gfs2_holder *gh2;
 969         int try_futile = 0;
 970
 971         BUG_ON(gh->gh_owner_pid == NULL);
 972         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
 973                 BUG();
 974
 975         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
 976                 if (test_bit(GLF_LOCK, &gl->gl_flags))
 977                         try_futile = !may_grant(gl, gh);
 978                 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 979                         goto fail;
 980         }
 981
 982         list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
 983                 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
 984                     (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
 985                         goto trap_recursive;
 986                 if (try_futile &&
 987                     !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
 988 fail:
 989                         gh->gh_error = GLR_TRYFAILED;
 990                         gfs2_holder_wake(gh);
 991                         return;
 992                 }
 993                 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
 994                         continue;
 995                 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
 996                         insert_pt = &gh2->gh_list;
 997         }
 998         set_bit(GLF_QUEUED, &gl->gl_flags);
 999         trace_gfs2_glock_queue(gh, 1);
1000         gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1001         gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1002         if (likely(insert_pt == NULL)) {
1003                 list_add_tail(&gh->gh_list, &gl->gl_holders);
1004                 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1005                         goto do_cancel;
1006                 return;
1007         }
1008         list_add_tail(&gh->gh_list, insert_pt);
1009 do_cancel:
1010         gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
1011         if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1012                 spin_unlock(&gl->gl_spin);
1013                 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1014                         sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1015                 spin_lock(&gl->gl_spin);
1016         }
1017         return;
1018
1019 trap_recursive:
1020         print_symbol(KERN_ERR "original: %s\n", gh2->gh_ip);
1021         printk(KERN_ERR "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1022         printk(KERN_ERR "lock type: %d req lock state : %d\n",
1023                gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1024         print_symbol(KERN_ERR "new: %s\n", gh->gh_ip);
1025         printk(KERN_ERR "pid: %d\n", pid_nr(gh->gh_owner_pid));
1026         printk(KERN_ERR "lock type: %d req lock state : %d\n",
1027                gh->gh_gl->gl_name.ln_type, gh->gh_state);
1028         gfs2_dump_glock(NULL, gl);
1029         BUG();
1030 }
1031
1032 /**
1033  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1034  * @gh: the holder structure
1035  *
1036  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1037  *
1038  * Returns: 0, GLR_TRYFAILED, or errno on failure
1039  */
1040
1041 int gfs2_glock_nq(struct gfs2_holder *gh)
1042 {
1043         struct gfs2_glock *gl = gh->gh_gl;
1044         struct gfs2_sbd *sdp = gl->gl_sbd;
1045         int error = 0;
1046
1047         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1048                 return -EIO;
1049
1050         if (test_bit(GLF_LRU, &gl->gl_flags))
1051                 gfs2_glock_remove_from_lru(gl);
1052
1053         spin_lock(&gl->gl_spin);
1054         add_to_queue(gh);
1055         if ((LM_FLAG_NOEXP & gh->gh_flags) &&
1056             test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1057                 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1058         run_queue(gl, 1);
1059         spin_unlock(&gl->gl_spin);
1060
1061         if (!(gh->gh_flags & GL_ASYNC))
1062                 error = gfs2_glock_wait(gh);
1063
1064         return error;
1065 }
1066
1067 /**
1068  * gfs2_glock_poll - poll to see if an async request has been completed
1069  * @gh: the holder
1070  *
1071  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1072  */
1073
1074 int gfs2_glock_poll(struct gfs2_holder *gh)
1075 {
1076         return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1077 }
1078
1079 /**
1080  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1081  * @gh: the glock holder
1082  *
1083  */
1084
1085 void gfs2_glock_dq(struct gfs2_holder *gh)
1086 {
1087         struct gfs2_glock *gl = gh->gh_gl;
1088         const struct gfs2_glock_operations *glops = gl->gl_ops;
1089         unsigned delay = 0;
1090         int fast_path = 0;
1091
1092         spin_lock(&gl->gl_spin);
1093         if (gh->gh_flags & GL_NOCACHE)
1094                 handle_callback(gl, LM_ST_UNLOCKED, 0);
1095
1096         list_del_init(&gh->gh_list);
1097         if (find_first_holder(gl) == NULL) {
1098                 if (glops->go_unlock) {
1099                         GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1100                         spin_unlock(&gl->gl_spin);
1101                         glops->go_unlock(gh);
1102                         spin_lock(&gl->gl_spin);
1103                         clear_bit(GLF_LOCK, &gl->gl_flags);
1104                 }
1105                 if (list_empty(&gl->gl_holders) &&
1106                     !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1107                     !test_bit(GLF_DEMOTE, &gl->gl_flags))
1108                         fast_path = 1;
1109         }
1110         if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1111                 gfs2_glock_add_to_lru(gl);
1112
1113         trace_gfs2_glock_queue(gh, 0);
1114         spin_unlock(&gl->gl_spin);
1115         if (likely(fast_path))
1116                 return;
1117
1118         gfs2_glock_hold(gl);
1119         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1120             !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1121             gl->gl_name.ln_type == LM_TYPE_INODE)
1122                 delay = gl->gl_hold_time;
1123         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1124                 gfs2_glock_put(gl);
1125 }
1126
1127 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1128 {
1129         struct gfs2_glock *gl = gh->gh_gl;
1130         gfs2_glock_dq(gh);
1131         might_sleep();
1132         wait_on_bit(&gl->gl_flags, GLF_DEMOTE, gfs2_glock_demote_wait, TASK_UNINTERRUPTIBLE);
1133 }
1134
1135 /**
1136  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1137  * @gh: the holder structure
1138  *
1139  */
1140
1141 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1142 {
1143         gfs2_glock_dq(gh);
1144         gfs2_holder_uninit(gh);
1145 }
1146
1147 /**
1148  * gfs2_glock_nq_num - acquire a glock based on lock number
1149  * @sdp: the filesystem
1150  * @number: the lock number
1151  * @glops: the glock operations for the type of glock
1152  * @state: the state to acquire the glock in
1153  * @flags: modifier flags for the acquisition
1154  * @gh: the struct gfs2_holder
1155  *
1156  * Returns: errno
1157  */
1158
1159 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1160                       const struct gfs2_glock_operations *glops,
1161                       unsigned int state, int flags, struct gfs2_holder *gh)
1162 {
1163         struct gfs2_glock *gl;
1164         int error;
1165
1166         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1167         if (!error) {
1168                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1169                 gfs2_glock_put(gl);
1170         }
1171
1172         return error;
1173 }
1174
1175 /**
1176  * glock_compare - Compare two struct gfs2_glock structures for sorting
1177  * @arg_a: the first structure
1178  * @arg_b: the second structure
1179  *
1180  */
1181
1182 static int glock_compare(const void *arg_a, const void *arg_b)
1183 {
1184         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1185         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1186         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1187         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1188
1189         if (a->ln_number > b->ln_number)
1190                 return 1;
1191         if (a->ln_number < b->ln_number)
1192                 return -1;
1193         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1194         return 0;
1195 }
1196
1197 /**
1198  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1199  * @num_gh: the number of structures
1200  * @ghs: an array of struct gfs2_holder structures
1201  *
1202  * Returns: 0 on success (all glocks acquired),
1203  *          errno on failure (no glocks acquired)
1204  */
1205
1206 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1207                      struct gfs2_holder **p)
1208 {
1209         unsigned int x;
1210         int error = 0;
1211
1212         for (x = 0; x < num_gh; x++)
1213                 p[x] = &ghs[x];
1214
1215         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1216
1217         for (x = 0; x < num_gh; x++) {
1218                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1219
1220                 error = gfs2_glock_nq(p[x]);
1221                 if (error) {
1222                         while (x--)
1223                                 gfs2_glock_dq(p[x]);
1224                         break;
1225                 }
1226         }
1227
1228         return error;
1229 }
1230
1231 /**
1232  * gfs2_glock_nq_m - acquire multiple glocks
1233  * @num_gh: the number of structures
1234  * @ghs: an array of struct gfs2_holder structures
1235  *
1236  *
1237  * Returns: 0 on success (all glocks acquired),
1238  *          errno on failure (no glocks acquired)
1239  */
1240
1241 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1242 {
1243         struct gfs2_holder *tmp[4];
1244         struct gfs2_holder **pph = tmp;
1245         int error = 0;
1246
1247         switch(num_gh) {
1248         case 0:
1249                 return 0;
1250         case 1:
1251                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1252                 return gfs2_glock_nq(ghs);
1253         default:
1254                 if (num_gh <= 4)
1255                         break;
1256                 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1257                 if (!pph)
1258                         return -ENOMEM;
1259         }
1260
1261         error = nq_m_sync(num_gh, ghs, pph);
1262
1263         if (pph != tmp)
1264                 kfree(pph);
1265
1266         return error;
1267 }
1268
1269 /**
1270  * gfs2_glock_dq_m - release multiple glocks
1271  * @num_gh: the number of structures
1272  * @ghs: an array of struct gfs2_holder structures
1273  *
1274  */
1275
1276 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1277 {
1278         while (num_gh--)
1279                 gfs2_glock_dq(&ghs[num_gh]);
1280 }
1281
1282 /**
1283  * gfs2_glock_dq_uninit_m - release multiple glocks
1284  * @num_gh: the number of structures
1285  * @ghs: an array of struct gfs2_holder structures
1286  *
1287  */
1288
1289 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1290 {
1291         while (num_gh--)
1292                 gfs2_glock_dq_uninit(&ghs[num_gh]);
1293 }
1294
1295 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1296 {
1297         unsigned long delay = 0;
1298         unsigned long holdtime;
1299         unsigned long now = jiffies;
1300
1301         gfs2_glock_hold(gl);
1302         holdtime = gl->gl_tchange + gl->gl_hold_time;
1303         if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1304             gl->gl_name.ln_type == LM_TYPE_INODE) {
1305                 if (time_before(now, holdtime))
1306                         delay = holdtime - now;
1307                 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1308                         delay = gl->gl_hold_time;
1309         }
1310
1311         spin_lock(&gl->gl_spin);
1312         handle_callback(gl, state, delay);
1313         spin_unlock(&gl->gl_spin);
1314         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1315                 gfs2_glock_put(gl);
1316 }
1317
1318 /**
1319  * gfs2_should_freeze - Figure out if glock should be frozen
1320  * @gl: The glock in question
1321  *
1322  * Glocks are not frozen if (a) the result of the dlm operation is
1323  * an error, (b) the locking operation was an unlock operation or
1324  * (c) if there is a "noexp" flagged request anywhere in the queue
1325  *
1326  * Returns: 1 if freezing should occur, 0 otherwise
1327  */
1328
1329 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1330 {
1331         const struct gfs2_holder *gh;
1332
1333         if (gl->gl_reply & ~LM_OUT_ST_MASK)
1334                 return 0;
1335         if (gl->gl_target == LM_ST_UNLOCKED)
1336                 return 0;
1337
1338         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1339                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1340                         continue;
1341                 if (LM_FLAG_NOEXP & gh->gh_flags)
1342                         return 0;
1343         }
1344
1345         return 1;
1346 }
1347
1348 /**
1349  * gfs2_glock_complete - Callback used by locking
1350  * @gl: Pointer to the glock
1351  * @ret: The return value from the dlm
1352  *
1353  * The gl_reply field is under the gl_spin lock so that it is ok
1354  * to use a bitfield shared with other glock state fields.
1355  */
1356
1357 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1358 {
1359         struct lm_lockstruct *ls = &gl->gl_sbd->sd_lockstruct;
1360
1361         spin_lock(&gl->gl_spin);
1362         gl->gl_reply = ret;
1363
1364         if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1365                 if (gfs2_should_freeze(gl)) {
1366                         set_bit(GLF_FROZEN, &gl->gl_flags);
1367                         spin_unlock(&gl->gl_spin);
1368                         return;
1369                 }
1370         }
1371
1372         spin_unlock(&gl->gl_spin);
1373         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1374         smp_wmb();
1375         gfs2_glock_hold(gl);
1376         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1377                 gfs2_glock_put(gl);
1378 }
1379
1380 static int glock_cmp(void *priv, struct list_head *a, struct list_head *b)
1381 {
1382         struct gfs2_glock *gla, *glb;
1383
1384         gla = list_entry(a, struct gfs2_glock, gl_lru);
1385         glb = list_entry(b, struct gfs2_glock, gl_lru);
1386
1387         if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1388                 return 1;
1389         if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1390                 return -1;
1391
1392         return 0;
1393 }
1394
1395 /**
1396  * gfs2_dispose_glock_lru - Demote a list of glocks
1397  * @list: The list to dispose of
1398  *
1399  * Disposing of glocks may involve disk accesses, so that here we sort
1400  * the glocks by number (i.e. disk location of the inodes) so that if
1401  * there are any such accesses, they'll be sent in order (mostly).
1402  *
1403  * Must be called under the lru_lock, but may drop and retake this
1404  * lock. While the lru_lock is dropped, entries may vanish from the
1405  * list, but no new entries will appear on the list (since it is
1406  * private)
1407  */
1408
1409 static void gfs2_dispose_glock_lru(struct list_head *list)
1410 __releases(&lru_lock)
1411 __acquires(&lru_lock)
1412 {
1413         struct gfs2_glock *gl;
1414
1415         list_sort(NULL, list, glock_cmp);
1416
1417         while(!list_empty(list)) {
1418                 gl = list_entry(list->next, struct gfs2_glock, gl_lru);
1419                 list_del_init(&gl->gl_lru);
1420                 clear_bit(GLF_LRU, &gl->gl_flags);
1421                 gfs2_glock_hold(gl);
1422                 spin_unlock(&lru_lock);
1423                 spin_lock(&gl->gl_spin);
1424                 if (demote_ok(gl))
1425                         handle_callback(gl, LM_ST_UNLOCKED, 0);
1426                 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1427                 smp_mb__after_clear_bit();
1428                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1429                         gfs2_glock_put_nolock(gl);
1430                 spin_unlock(&gl->gl_spin);
1431                 spin_lock(&lru_lock);
1432         }
1433 }
1434
1435 /**
1436  * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1437  * @nr: The number of entries to scan
1438  *
1439  * This function selects the entries on the LRU which are able to
1440  * be demoted, and then kicks off the process by calling
1441  * gfs2_dispose_glock_lru() above.
1442  */
1443
1444 static void gfs2_scan_glock_lru(int nr)
1445 {
1446         struct gfs2_glock *gl;
1447         LIST_HEAD(skipped);
1448         LIST_HEAD(dispose);
1449
1450         spin_lock(&lru_lock);
1451         while(nr && !list_empty(&lru_list)) {
1452                 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1453
1454                 /* Test for being demotable */
1455                 if (!test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1456                         list_move(&gl->gl_lru, &dispose);
1457                         atomic_dec(&lru_count);
1458                         nr--;
1459                         continue;
1460                 }
1461
1462                 list_move(&gl->gl_lru, &skipped);
1463         }
1464         list_splice(&skipped, &lru_list);
1465         if (!list_empty(&dispose))
1466                 gfs2_dispose_glock_lru(&dispose);
1467         spin_unlock(&lru_lock);
1468 }
1469
1470 static int gfs2_shrink_glock_memory(struct shrinker *shrink,
1471                                     struct shrink_control *sc)
1472 {
1473         if (sc->nr_to_scan) {
1474                 if (!(sc->gfp_mask & __GFP_FS))
1475                         return -1;
1476                 gfs2_scan_glock_lru(sc->nr_to_scan);
1477         }
1478
1479         return (atomic_read(&lru_count) / 100) * sysctl_vfs_cache_pressure;
1480 }
1481
1482 static struct shrinker glock_shrinker = {
1483         .shrink = gfs2_shrink_glock_memory,
1484         .seeks = DEFAULT_SEEKS,
1485 };
1486
1487 /**
1488  * examine_bucket - Call a function for glock in a hash bucket
1489  * @examiner: the function
1490  * @sdp: the filesystem
1491  * @bucket: the bucket
1492  *
1493  */
1494
1495 static void examine_bucket(glock_examiner examiner, const struct gfs2_sbd *sdp,
1496                           unsigned int hash)
1497 {
1498         struct gfs2_glock *gl;
1499         struct hlist_bl_head *head = &gl_hash_table[hash];
1500         struct hlist_bl_node *pos;
1501
1502         rcu_read_lock();
1503         hlist_bl_for_each_entry_rcu(gl, pos, head, gl_list) {
1504                 if ((gl->gl_sbd == sdp) && atomic_read(&gl->gl_ref))
1505                         examiner(gl);
1506         }
1507         rcu_read_unlock();
1508         cond_resched();
1509 }
1510
1511 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1512 {
1513         unsigned x;
1514
1515         for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1516                 examine_bucket(examiner, sdp, x);
1517 }
1518
1519
1520 /**
1521  * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1522  * @gl: The glock to thaw
1523  *
1524  * N.B. When we freeze a glock, we leave a ref to the glock outstanding,
1525  * so this has to result in the ref count being dropped by one.
1526  */
1527
1528 static void thaw_glock(struct gfs2_glock *gl)
1529 {
1530         if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1531                 return;
1532         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1533         gfs2_glock_hold(gl);
1534         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1535                 gfs2_glock_put(gl);
1536 }
1537
1538 /**
1539  * clear_glock - look at a glock and see if we can free it from glock cache
1540  * @gl: the glock to look at
1541  *
1542  */
1543
1544 static void clear_glock(struct gfs2_glock *gl)
1545 {
1546         gfs2_glock_remove_from_lru(gl);
1547
1548         spin_lock(&gl->gl_spin);
1549         if (gl->gl_state != LM_ST_UNLOCKED)
1550                 handle_callback(gl, LM_ST_UNLOCKED, 0);
1551         spin_unlock(&gl->gl_spin);
1552         gfs2_glock_hold(gl);
1553         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1554                 gfs2_glock_put(gl);
1555 }
1556
1557 /**
1558  * gfs2_glock_thaw - Thaw any frozen glocks
1559  * @sdp: The super block
1560  *
1561  */
1562
1563 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1564 {
1565         glock_hash_walk(thaw_glock, sdp);
1566 }
1567
1568 static int dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1569 {
1570         int ret;
1571         spin_lock(&gl->gl_spin);
1572         ret = gfs2_dump_glock(seq, gl);
1573         spin_unlock(&gl->gl_spin);
1574         return ret;
1575 }
1576
1577 static void dump_glock_func(struct gfs2_glock *gl)
1578 {
1579         dump_glock(NULL, gl);
1580 }
1581
1582 /**
1583  * gfs2_gl_hash_clear - Empty out the glock hash table
1584  * @sdp: the filesystem
1585  * @wait: wait until it's all gone
1586  *
1587  * Called when unmounting the filesystem.
1588  */
1589
1590 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1591 {
1592         set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1593         glock_hash_walk(clear_glock, sdp);
1594         flush_workqueue(glock_workqueue);
1595         wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
1596         glock_hash_walk(dump_glock_func, sdp);
1597 }
1598
1599 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1600 {
1601         struct gfs2_glock *gl = ip->i_gl;
1602         int ret;
1603
1604         ret = gfs2_truncatei_resume(ip);
1605         gfs2_assert_withdraw(gl->gl_sbd, ret == 0);
1606
1607         spin_lock(&gl->gl_spin);
1608         clear_bit(GLF_LOCK, &gl->gl_flags);
1609         run_queue(gl, 1);
1610         spin_unlock(&gl->gl_spin);
1611 }
1612
1613 static const char *state2str(unsigned state)
1614 {
1615         switch(state) {
1616         case LM_ST_UNLOCKED:
1617                 return "UN";
1618         case LM_ST_SHARED:
1619                 return "SH";
1620         case LM_ST_DEFERRED:
1621                 return "DF";
1622         case LM_ST_EXCLUSIVE:
1623                 return "EX";
1624         }
1625         return "??";
1626 }
1627
1628 static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1629 {
1630         char *p = buf;
1631         if (flags & LM_FLAG_TRY)
1632                 *p++ = 't';
1633         if (flags & LM_FLAG_TRY_1CB)
1634                 *p++ = 'T';
1635         if (flags & LM_FLAG_NOEXP)
1636                 *p++ = 'e';
1637         if (flags & LM_FLAG_ANY)
1638                 *p++ = 'A';
1639         if (flags & LM_FLAG_PRIORITY)
1640                 *p++ = 'p';
1641         if (flags & GL_ASYNC)
1642                 *p++ = 'a';
1643         if (flags & GL_EXACT)
1644                 *p++ = 'E';
1645         if (flags & GL_NOCACHE)
1646                 *p++ = 'c';
1647         if (test_bit(HIF_HOLDER, &iflags))
1648                 *p++ = 'H';
1649         if (test_bit(HIF_WAIT, &iflags))
1650                 *p++ = 'W';
1651         if (test_bit(HIF_FIRST, &iflags))
1652                 *p++ = 'F';
1653         *p = 0;
1654         return buf;
1655 }
1656
1657 /**
1658  * dump_holder - print information about a glock holder
1659  * @seq: the seq_file struct
1660  * @gh: the glock holder
1661  *
1662  * Returns: 0 on success, -ENOBUFS when we run out of space
1663  */
1664
1665 static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1666 {
1667         struct task_struct *gh_owner = NULL;
1668         char flags_buf[32];
1669
1670         if (gh->gh_owner_pid)
1671                 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1672         gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1673                        state2str(gh->gh_state),
1674                        hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1675                        gh->gh_error,
1676                        gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1677                        gh_owner ? gh_owner->comm : "(ended)",
1678                        (void *)gh->gh_ip);
1679         return 0;
1680 }
1681
1682 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1683 {
1684         const unsigned long *gflags = &gl->gl_flags;
1685         char *p = buf;
1686
1687         if (test_bit(GLF_LOCK, gflags))
1688                 *p++ = 'l';
1689         if (test_bit(GLF_DEMOTE, gflags))
1690                 *p++ = 'D';
1691         if (test_bit(GLF_PENDING_DEMOTE, gflags))
1692                 *p++ = 'd';
1693         if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1694                 *p++ = 'p';
1695         if (test_bit(GLF_DIRTY, gflags))
1696                 *p++ = 'y';
1697         if (test_bit(GLF_LFLUSH, gflags))
1698                 *p++ = 'f';
1699         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1700                 *p++ = 'i';
1701         if (test_bit(GLF_REPLY_PENDING, gflags))
1702                 *p++ = 'r';
1703         if (test_bit(GLF_INITIAL, gflags))
1704                 *p++ = 'I';
1705         if (test_bit(GLF_FROZEN, gflags))
1706                 *p++ = 'F';
1707         if (test_bit(GLF_QUEUED, gflags))
1708                 *p++ = 'q';
1709         if (test_bit(GLF_LRU, gflags))
1710                 *p++ = 'L';
1711         if (gl->gl_object)
1712                 *p++ = 'o';
1713         if (test_bit(GLF_BLOCKING, gflags))
1714                 *p++ = 'b';
1715         *p = 0;
1716         return buf;
1717 }
1718
1719 /**
1720  * gfs2_dump_glock - print information about a glock
1721  * @seq: The seq_file struct
1722  * @gl: the glock
1723  *
1724  * The file format is as follows:
1725  * One line per object, capital letters are used to indicate objects
1726  * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1727  * other objects are indented by a single space and follow the glock to
1728  * which they are related. Fields are indicated by lower case letters
1729  * followed by a colon and the field value, except for strings which are in
1730  * [] so that its possible to see if they are composed of spaces for
1731  * example. The field's are n = number (id of the object), f = flags,
1732  * t = type, s = state, r = refcount, e = error, p = pid.
1733  *
1734  * Returns: 0 on success, -ENOBUFS when we run out of space
1735  */
1736
1737 int gfs2_dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1738 {
1739         const struct gfs2_glock_operations *glops = gl->gl_ops;
1740         unsigned long long dtime;
1741         const struct gfs2_holder *gh;
1742         char gflags_buf[32];
1743         int error = 0;
1744
1745         dtime = jiffies - gl->gl_demote_time;
1746         dtime *= 1000000/HZ; /* demote time in uSec */
1747         if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1748                 dtime = 0;
1749         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",
1750                   state2str(gl->gl_state),
1751                   gl->gl_name.ln_type,
1752                   (unsigned long long)gl->gl_name.ln_number,
1753                   gflags2str(gflags_buf, gl),
1754                   state2str(gl->gl_target),
1755                   state2str(gl->gl_demote_state), dtime,
1756                   atomic_read(&gl->gl_ail_count),
1757                   atomic_read(&gl->gl_revokes),
1758                   atomic_read(&gl->gl_ref), gl->gl_hold_time);
1759
1760         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1761                 error = dump_holder(seq, gh);
1762                 if (error)
1763                         goto out;
1764         }
1765         if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1766                 error = glops->go_dump(seq, gl);
1767 out:
1768         return error;
1769 }
1770
1771 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
1772 {
1773         struct gfs2_glock *gl = iter_ptr;
1774
1775         seq_printf(seq, "G: n:%u/%llx rtt:%lld/%lld rttb:%lld/%lld irt:%lld/%lld dcnt: %lld qcnt: %lld\n",
1776                    gl->gl_name.ln_type,
1777                    (unsigned long long)gl->gl_name.ln_number,
1778                    (long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
1779                    (long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
1780                    (long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
1781                    (long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
1782                    (long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
1783                    (long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
1784                    (long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
1785                    (long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
1786         return 0;
1787 }
1788
1789 static const char *gfs2_gltype[] = {
1790         "type",
1791         "reserved",
1792         "nondisk",
1793         "inode",
1794         "rgrp",
1795         "meta",
1796         "iopen",
1797         "flock",
1798         "plock",
1799         "quota",
1800         "journal",
1801 };
1802
1803 static const char *gfs2_stype[] = {
1804         [GFS2_LKS_SRTT]         = "srtt",
1805         [GFS2_LKS_SRTTVAR]      = "srttvar",
1806         [GFS2_LKS_SRTTB]        = "srttb",
1807         [GFS2_LKS_SRTTVARB]     = "srttvarb",
1808         [GFS2_LKS_SIRT]         = "sirt",
1809         [GFS2_LKS_SIRTVAR]      = "sirtvar",
1810         [GFS2_LKS_DCOUNT]       = "dlm",
1811         [GFS2_LKS_QCOUNT]       = "queue",
1812 };
1813
1814 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1815
1816 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
1817 {
1818         struct gfs2_glock_iter *gi = seq->private;
1819         struct gfs2_sbd *sdp = gi->sdp;
1820         unsigned index = gi->hash >> 3;
1821         unsigned subindex = gi->hash & 0x07;
1822         s64 value;
1823         int i;
1824
1825         if (index == 0 && subindex != 0)
1826                 return 0;
1827
1828         seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
1829                    (index == 0) ? "cpu": gfs2_stype[subindex]);
1830
1831         for_each_possible_cpu(i) {
1832                 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
1833                 if (index == 0) {
1834                         value = i;
1835                 } else {
1836                         value = lkstats->lkstats[index - 1].stats[subindex];
1837                 }
1838                 seq_printf(seq, " %15lld", (long long)value);
1839         }
1840         seq_putc(seq, '\n');
1841         return 0;
1842 }
1843
1844 int __init gfs2_glock_init(void)
1845 {
1846         unsigned i;
1847         for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
1848                 INIT_HLIST_BL_HEAD(&gl_hash_table[i]);
1849         }
1850
1851         glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1852                                           WQ_HIGHPRI | WQ_FREEZABLE, 0);
1853         if (IS_ERR(glock_workqueue))
1854                 return PTR_ERR(glock_workqueue);
1855         gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1856                                                 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1857                                                 0);
1858         if (IS_ERR(gfs2_delete_workqueue)) {
1859                 destroy_workqueue(glock_workqueue);
1860                 return PTR_ERR(gfs2_delete_workqueue);
1861         }
1862
1863         register_shrinker(&glock_shrinker);
1864
1865         return 0;
1866 }
1867
1868 void gfs2_glock_exit(void)
1869 {
1870         unregister_shrinker(&glock_shrinker);
1871         destroy_workqueue(glock_workqueue);
1872         destroy_workqueue(gfs2_delete_workqueue);
1873 }
1874
1875 static inline struct gfs2_glock *glock_hash_chain(unsigned hash)
1876 {
1877         return hlist_bl_entry(hlist_bl_first_rcu(&gl_hash_table[hash]),
1878                               struct gfs2_glock, gl_list);
1879 }
1880
1881 static inline struct gfs2_glock *glock_hash_next(struct gfs2_glock *gl)
1882 {
1883         return hlist_bl_entry(rcu_dereference(gl->gl_list.next),
1884                               struct gfs2_glock, gl_list);
1885 }
1886
1887 static int gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1888 {
1889         struct gfs2_glock *gl;
1890
1891         do {
1892                 gl = gi->gl;
1893                 if (gl) {
1894                         gi->gl = glock_hash_next(gl);
1895                         gi->nhash++;
1896                 } else {
1897                         if (gi->hash >= GFS2_GL_HASH_SIZE) {
1898                                 rcu_read_unlock();
1899                                 return 1;
1900                         }
1901                         gi->gl = glock_hash_chain(gi->hash);
1902                         gi->nhash = 0;
1903                 }
1904                 while (gi->gl == NULL) {
1905                         gi->hash++;
1906                         if (gi->hash >= GFS2_GL_HASH_SIZE) {
1907                                 rcu_read_unlock();
1908                                 return 1;
1909                         }
1910                         gi->gl = glock_hash_chain(gi->hash);
1911                         gi->nhash = 0;
1912                 }
1913         /* Skip entries for other sb and dead entries */
1914         } while (gi->sdp != gi->gl->gl_sbd || atomic_read(&gi->gl->gl_ref) == 0);
1915
1916         return 0;
1917 }
1918
1919 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1920 {
1921         struct gfs2_glock_iter *gi = seq->private;
1922         loff_t n = *pos;
1923
1924         if (gi->last_pos <= *pos)
1925                 n = gi->nhash + (*pos - gi->last_pos);
1926         else
1927                 gi->hash = 0;
1928
1929         gi->nhash = 0;
1930         rcu_read_lock();
1931
1932         do {
1933                 if (gfs2_glock_iter_next(gi))
1934                         return NULL;
1935         } while (n--);
1936
1937         gi->last_pos = *pos;
1938         return gi->gl;
1939 }
1940
1941 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1942                                  loff_t *pos)
1943 {
1944         struct gfs2_glock_iter *gi = seq->private;
1945
1946         (*pos)++;
1947         gi->last_pos = *pos;
1948         if (gfs2_glock_iter_next(gi))
1949                 return NULL;
1950
1951         return gi->gl;
1952 }
1953
1954 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1955 {
1956         struct gfs2_glock_iter *gi = seq->private;
1957
1958         if (gi->gl)
1959                 rcu_read_unlock();
1960         gi->gl = NULL;
1961 }
1962
1963 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1964 {
1965         return dump_glock(seq, iter_ptr);
1966 }
1967
1968 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
1969 {
1970         struct gfs2_glock_iter *gi = seq->private;
1971
1972         gi->hash = *pos;
1973         if (*pos >= GFS2_NR_SBSTATS)
1974                 return NULL;
1975         preempt_disable();
1976         return SEQ_START_TOKEN;
1977 }
1978
1979 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
1980                                    loff_t *pos)
1981 {
1982         struct gfs2_glock_iter *gi = seq->private;
1983         (*pos)++;
1984         gi->hash++;
1985         if (gi->hash >= GFS2_NR_SBSTATS) {
1986                 preempt_enable();
1987                 return NULL;
1988         }
1989         return SEQ_START_TOKEN;
1990 }
1991
1992 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
1993 {
1994         preempt_enable();
1995 }
1996
1997 static const struct seq_operations gfs2_glock_seq_ops = {
1998         .start = gfs2_glock_seq_start,
1999         .next  = gfs2_glock_seq_next,
2000         .stop  = gfs2_glock_seq_stop,
2001         .show  = gfs2_glock_seq_show,
2002 };
2003
2004 static const struct seq_operations gfs2_glstats_seq_ops = {
2005         .start = gfs2_glock_seq_start,
2006         .next  = gfs2_glock_seq_next,
2007         .stop  = gfs2_glock_seq_stop,
2008         .show  = gfs2_glstats_seq_show,
2009 };
2010
2011 static const struct seq_operations gfs2_sbstats_seq_ops = {
2012         .start = gfs2_sbstats_seq_start,
2013         .next  = gfs2_sbstats_seq_next,
2014         .stop  = gfs2_sbstats_seq_stop,
2015         .show  = gfs2_sbstats_seq_show,
2016 };
2017
2018 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2019
2020 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2021 {
2022         int ret = seq_open_private(file, &gfs2_glock_seq_ops,
2023                                    sizeof(struct gfs2_glock_iter));
2024         if (ret == 0) {
2025                 struct seq_file *seq = file->private_data;
2026                 struct gfs2_glock_iter *gi = seq->private;
2027                 gi->sdp = inode->i_private;
2028                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2029                 if (seq->buf)
2030                         seq->size = GFS2_SEQ_GOODSIZE;
2031         }
2032         return ret;
2033 }
2034
2035 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2036 {
2037         int ret = seq_open_private(file, &gfs2_glstats_seq_ops,
2038                                    sizeof(struct gfs2_glock_iter));
2039         if (ret == 0) {
2040                 struct seq_file *seq = file->private_data;
2041                 struct gfs2_glock_iter *gi = seq->private;
2042                 gi->sdp = inode->i_private;
2043                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2044                 if (seq->buf)
2045                         seq->size = GFS2_SEQ_GOODSIZE;
2046         }
2047         return ret;
2048 }
2049
2050 static int gfs2_sbstats_open(struct inode *inode, struct file *file)
2051 {
2052         int ret = seq_open_private(file, &gfs2_sbstats_seq_ops,
2053                                    sizeof(struct gfs2_glock_iter));
2054         if (ret == 0) {
2055                 struct seq_file *seq = file->private_data;
2056                 struct gfs2_glock_iter *gi = seq->private;
2057                 gi->sdp = inode->i_private;
2058         }
2059         return ret;
2060 }
2061
2062 static const struct file_operations gfs2_glocks_fops = {
2063         .owner   = THIS_MODULE,
2064         .open    = gfs2_glocks_open,
2065         .read    = seq_read,
2066         .llseek  = seq_lseek,
2067         .release = seq_release_private,
2068 };
2069
2070 static const struct file_operations gfs2_glstats_fops = {
2071         .owner   = THIS_MODULE,
2072         .open    = gfs2_glstats_open,
2073         .read    = seq_read,
2074         .llseek  = seq_lseek,
2075         .release = seq_release_private,
2076 };
2077
2078 static const struct file_operations gfs2_sbstats_fops = {
2079         .owner   = THIS_MODULE,
2080         .open    = gfs2_sbstats_open,
2081         .read    = seq_read,
2082         .llseek  = seq_lseek,
2083         .release = seq_release_private,
2084 };
2085
2086 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2087 {
2088         sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2089         if (!sdp->debugfs_dir)
2090                 return -ENOMEM;
2091         sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
2092                                                          S_IFREG | S_IRUGO,
2093                                                          sdp->debugfs_dir, sdp,
2094                                                          &gfs2_glocks_fops);
2095         if (!sdp->debugfs_dentry_glocks)
2096                 goto fail;
2097
2098         sdp->debugfs_dentry_glstats = debugfs_create_file("glstats",
2099                                                         S_IFREG | S_IRUGO,
2100                                                         sdp->debugfs_dir, sdp,
2101                                                         &gfs2_glstats_fops);
2102         if (!sdp->debugfs_dentry_glstats)
2103                 goto fail;
2104
2105         sdp->debugfs_dentry_sbstats = debugfs_create_file("sbstats",
2106                                                         S_IFREG | S_IRUGO,
2107                                                         sdp->debugfs_dir, sdp,
2108                                                         &gfs2_sbstats_fops);
2109         if (!sdp->debugfs_dentry_sbstats)
2110                 goto fail;
2111
2112         return 0;
2113 fail:
2114         gfs2_delete_debugfs_file(sdp);
2115         return -ENOMEM;
2116 }
2117
2118 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2119 {
2120         if (sdp->debugfs_dir) {
2121                 if (sdp->debugfs_dentry_glocks) {
2122                         debugfs_remove(sdp->debugfs_dentry_glocks);
2123                         sdp->debugfs_dentry_glocks = NULL;
2124                 }
2125                 if (sdp->debugfs_dentry_glstats) {
2126                         debugfs_remove(sdp->debugfs_dentry_glstats);
2127                         sdp->debugfs_dentry_glstats = NULL;
2128                 }
2129                 if (sdp->debugfs_dentry_sbstats) {
2130                         debugfs_remove(sdp->debugfs_dentry_sbstats);
2131                         sdp->debugfs_dentry_sbstats = NULL;
2132                 }
2133                 debugfs_remove(sdp->debugfs_dir);
2134                 sdp->debugfs_dir = NULL;
2135         }
2136 }
2137
2138 int gfs2_register_debugfs(void)
2139 {
2140         gfs2_root = debugfs_create_dir("gfs2", NULL);
2141         return gfs2_root ? 0 : -ENOMEM;
2142 }
2143
2144 void gfs2_unregister_debugfs(void)
2145 {
2146         debugfs_remove(gfs2_root);
2147         gfs2_root = NULL;
2148 }