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
  30 #include "gfs2.h"
  31 #include "incore.h"
  32 #include "glock.h"
  33 #include "glops.h"
  34 #include "inode.h"
  35 #include "lops.h"
  36 #include "meta_io.h"
  37 #include "quota.h"
  38 #include "super.h"
  39 #include "util.h"
  40 #include "bmap.h"
  41 #define CREATE_TRACE_POINTS
  42 #include "trace_gfs2.h"
  43
  44 struct gfs2_gl_hash_bucket {
  45         struct hlist_head hb_list;
  46 };
  47
  48 struct gfs2_glock_iter {
  49         int hash;                       /* hash bucket index         */
  50         struct gfs2_sbd *sdp;           /* incore superblock         */
  51         struct gfs2_glock *gl;          /* current glock struct      */
  52         char string[512];               /* scratch space             */
  53 };
  54
  55 typedef void (*glock_examiner) (struct gfs2_glock * gl);
  56
  57 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp);
  58 static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl);
  59 #define GLOCK_BUG_ON(gl,x) do { if (unlikely(x)) { __dump_glock(NULL, gl); BUG(); } } while(0)
  60 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
  61
  62 static struct dentry *gfs2_root;
  63 static struct workqueue_struct *glock_workqueue;
  64 struct workqueue_struct *gfs2_delete_workqueue;
  65 static LIST_HEAD(lru_list);
  66 static atomic_t lru_count = ATOMIC_INIT(0);
  67 static DEFINE_SPINLOCK(lru_lock);
  68
  69 #define GFS2_GL_HASH_SHIFT      15
  70 #define GFS2_GL_HASH_SIZE       (1 << GFS2_GL_HASH_SHIFT)
  71 #define GFS2_GL_HASH_MASK       (GFS2_GL_HASH_SIZE - 1)
  72
  73 static struct gfs2_gl_hash_bucket gl_hash_table[GFS2_GL_HASH_SIZE];
  74 static struct dentry *gfs2_root;
  75
  76 /*
  77  * Despite what you might think, the numbers below are not arbitrary :-)
  78  * They are taken from the ipv4 routing hash code, which is well tested
  79  * and thus should be nearly optimal. Later on we might tweek the numbers
  80  * but for now this should be fine.
  81  *
  82  * The reason for putting the locks in a separate array from the list heads
  83  * is that we can have fewer locks than list heads and save memory. We use
  84  * the same hash function for both, but with a different hash mask.
  85  */
  86 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
  87         defined(CONFIG_PROVE_LOCKING)
  88
  89 #ifdef CONFIG_LOCKDEP
  90 # define GL_HASH_LOCK_SZ        256
  91 #else
  92 # if NR_CPUS >= 32
  93 #  define GL_HASH_LOCK_SZ       4096
  94 # elif NR_CPUS >= 16
  95 #  define GL_HASH_LOCK_SZ       2048
  96 # elif NR_CPUS >= 8
  97 #  define GL_HASH_LOCK_SZ       1024
  98 # elif NR_CPUS >= 4
  99 #  define GL_HASH_LOCK_SZ       512
 100 # else
 101 #  define GL_HASH_LOCK_SZ       256
 102 # endif
 103 #endif
 104
 105 /* We never want more locks than chains */
 106 #if GFS2_GL_HASH_SIZE < GL_HASH_LOCK_SZ
 107 # undef GL_HASH_LOCK_SZ
 108 # define GL_HASH_LOCK_SZ GFS2_GL_HASH_SIZE
 109 #endif
 110
 111 static rwlock_t gl_hash_locks[GL_HASH_LOCK_SZ];
 112
 113 static inline rwlock_t *gl_lock_addr(unsigned int x)
 114 {
 115         return &gl_hash_locks[x & (GL_HASH_LOCK_SZ-1)];
 116 }
 117 #else /* not SMP, so no spinlocks required */
 118 static inline rwlock_t *gl_lock_addr(unsigned int x)
 119 {
 120         return NULL;
 121 }
 122 #endif
 123
 124 /**
 125  * gl_hash() - Turn glock number into hash bucket number
 126  * @lock: The glock number
 127  *
 128  * Returns: The number of the corresponding hash bucket
 129  */
 130
 131 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
 132                             const struct lm_lockname *name)
 133 {
 134         unsigned int h;
 135
 136         h = jhash(&name->ln_number, sizeof(u64), 0);
 137         h = jhash(&name->ln_type, sizeof(unsigned int), h);
 138         h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
 139         h &= GFS2_GL_HASH_MASK;
 140
 141         return h;
 142 }
 143
 144 /**
 145  * glock_free() - Perform a few checks and then release struct gfs2_glock
 146  * @gl: The glock to release
 147  *
 148  * Also calls lock module to release its internal structure for this glock.
 149  *
 150  */
 151
 152 static void glock_free(struct gfs2_glock *gl)
 153 {
 154         struct gfs2_sbd *sdp = gl->gl_sbd;
 155         struct address_space *mapping = gfs2_glock2aspace(gl);
 156         struct kmem_cache *cachep = gfs2_glock_cachep;
 157
 158         GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
 159         trace_gfs2_glock_put(gl);
 160         if (mapping)
 161                 cachep = gfs2_glock_aspace_cachep;
 162         sdp->sd_lockstruct.ls_ops->lm_put_lock(cachep, gl);
 163 }
 164
 165 /**
 166  * gfs2_glock_hold() - increment reference count on glock
 167  * @gl: The glock to hold
 168  *
 169  */
 170
 171 void gfs2_glock_hold(struct gfs2_glock *gl)
 172 {
 173         GLOCK_BUG_ON(gl, atomic_read(&gl->gl_ref) == 0);
 174         atomic_inc(&gl->gl_ref);
 175 }
 176
 177 /**
 178  * demote_ok - Check to see if it's ok to unlock a glock
 179  * @gl: the glock
 180  *
 181  * Returns: 1 if it's ok
 182  */
 183
 184 static int demote_ok(const struct gfs2_glock *gl)
 185 {
 186         const struct gfs2_glock_operations *glops = gl->gl_ops;
 187
 188         if (gl->gl_state == LM_ST_UNLOCKED)
 189                 return 0;
 190         if (!list_empty(&gl->gl_holders))
 191                 return 0;
 192         if (glops->go_demote_ok)
 193                 return glops->go_demote_ok(gl);
 194         return 1;
 195 }
 196
 197 /**
 198  * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
 199  * @gl: the glock
 200  *
 201  */
 202
 203 static void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
 204 {
 205         int may_reclaim;
 206         may_reclaim = (demote_ok(gl) &&
 207                        (atomic_read(&gl->gl_ref) == 1 ||
 208                         (gl->gl_name.ln_type == LM_TYPE_INODE &&
 209                          atomic_read(&gl->gl_ref) <= 2)));
 210         spin_lock(&lru_lock);
 211         if (list_empty(&gl->gl_lru) && may_reclaim) {
 212                 list_add_tail(&gl->gl_lru, &lru_list);
 213                 atomic_inc(&lru_count);
 214         }
 215         spin_unlock(&lru_lock);
 216 }
 217
 218 /**
 219  * gfs2_glock_put_nolock() - Decrement reference count on glock
 220  * @gl: The glock to put
 221  *
 222  * This function should only be used if the caller has its own reference
 223  * to the glock, in addition to the one it is dropping.
 224  */
 225
 226 void gfs2_glock_put_nolock(struct gfs2_glock *gl)
 227 {
 228         if (atomic_dec_and_test(&gl->gl_ref))
 229                 GLOCK_BUG_ON(gl, 1);
 230         gfs2_glock_schedule_for_reclaim(gl);
 231 }
 232
 233 /**
 234  * gfs2_glock_put() - Decrement reference count on glock
 235  * @gl: The glock to put
 236  *
 237  */
 238
 239 int gfs2_glock_put(struct gfs2_glock *gl)
 240 {
 241         int rv = 0;
 242
 243         write_lock(gl_lock_addr(gl->gl_hash));
 244         if (atomic_dec_and_lock(&gl->gl_ref, &lru_lock)) {
 245                 hlist_del(&gl->gl_list);
 246                 if (!list_empty(&gl->gl_lru)) {
 247                         list_del_init(&gl->gl_lru);
 248                         atomic_dec(&lru_count);
 249                 }
 250                 spin_unlock(&lru_lock);
 251                 write_unlock(gl_lock_addr(gl->gl_hash));
 252                 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
 253                 glock_free(gl);
 254                 rv = 1;
 255                 goto out;
 256         }
 257         spin_lock(&gl->gl_spin);
 258         gfs2_glock_schedule_for_reclaim(gl);
 259         spin_unlock(&gl->gl_spin);
 260         write_unlock(gl_lock_addr(gl->gl_hash));
 261 out:
 262         return rv;
 263 }
 264
 265 /**
 266  * search_bucket() - Find struct gfs2_glock by lock number
 267  * @bucket: the bucket to search
 268  * @name: The lock name
 269  *
 270  * Returns: NULL, or the struct gfs2_glock with the requested number
 271  */
 272
 273 static struct gfs2_glock *search_bucket(unsigned int hash,
 274                                         const struct gfs2_sbd *sdp,
 275                                         const struct lm_lockname *name)
 276 {
 277         struct gfs2_glock *gl;
 278         struct hlist_node *h;
 279
 280         hlist_for_each_entry(gl, h, &gl_hash_table[hash].hb_list, gl_list) {
 281                 if (!lm_name_equal(&gl->gl_name, name))
 282                         continue;
 283                 if (gl->gl_sbd != sdp)
 284                         continue;
 285
 286                 atomic_inc(&gl->gl_ref);
 287
 288                 return gl;
 289         }
 290
 291         return NULL;
 292 }
 293
 294 /**
 295  * may_grant - check if its ok to grant a new lock
 296  * @gl: The glock
 297  * @gh: The lock request which we wish to grant
 298  *
 299  * Returns: true if its ok to grant the lock
 300  */
 301
 302 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
 303 {
 304         const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
 305         if ((gh->gh_state == LM_ST_EXCLUSIVE ||
 306              gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
 307                 return 0;
 308         if (gl->gl_state == gh->gh_state)
 309                 return 1;
 310         if (gh->gh_flags & GL_EXACT)
 311                 return 0;
 312         if (gl->gl_state == LM_ST_EXCLUSIVE) {
 313                 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
 314                         return 1;
 315                 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
 316                         return 1;
 317         }
 318         if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
 319                 return 1;
 320         return 0;
 321 }
 322
 323 static void gfs2_holder_wake(struct gfs2_holder *gh)
 324 {
 325         clear_bit(HIF_WAIT, &gh->gh_iflags);
 326         smp_mb__after_clear_bit();
 327         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
 328 }
 329
 330 /**
 331  * do_error - Something unexpected has happened during a lock request
 332  *
 333  */
 334
 335 static inline void do_error(struct gfs2_glock *gl, const int ret)
 336 {
 337         struct gfs2_holder *gh, *tmp;
 338
 339         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
 340                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 341                         continue;
 342                 if (ret & LM_OUT_ERROR)
 343                         gh->gh_error = -EIO;
 344                 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
 345                         gh->gh_error = GLR_TRYFAILED;
 346                 else
 347                         continue;
 348                 list_del_init(&gh->gh_list);
 349                 trace_gfs2_glock_queue(gh, 0);
 350                 gfs2_holder_wake(gh);
 351         }
 352 }
 353
 354 /**
 355  * do_promote - promote as many requests as possible on the current queue
 356  * @gl: The glock
 357  *
 358  * Returns: 1 if there is a blocked holder at the head of the list, or 2
 359  *          if a type specific operation is underway.
 360  */
 361
 362 static int do_promote(struct gfs2_glock *gl)
 363 __releases(&gl->gl_spin)
 364 __acquires(&gl->gl_spin)
 365 {
 366         const struct gfs2_glock_operations *glops = gl->gl_ops;
 367         struct gfs2_holder *gh, *tmp;
 368         int ret;
 369
 370 restart:
 371         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
 372                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 373                         continue;
 374                 if (may_grant(gl, gh)) {
 375                         if (gh->gh_list.prev == &gl->gl_holders &&
 376                             glops->go_lock) {
 377                                 spin_unlock(&gl->gl_spin);
 378                                 /* FIXME: eliminate this eventually */
 379                                 ret = glops->go_lock(gh);
 380                                 spin_lock(&gl->gl_spin);
 381                                 if (ret) {
 382                                         if (ret == 1)
 383                                                 return 2;
 384                                         gh->gh_error = ret;
 385                                         list_del_init(&gh->gh_list);
 386                                         trace_gfs2_glock_queue(gh, 0);
 387                                         gfs2_holder_wake(gh);
 388                                         goto restart;
 389                                 }
 390                                 set_bit(HIF_HOLDER, &gh->gh_iflags);
 391                                 trace_gfs2_promote(gh, 1);
 392                                 gfs2_holder_wake(gh);
 393                                 goto restart;
 394                         }
 395                         set_bit(HIF_HOLDER, &gh->gh_iflags);
 396                         trace_gfs2_promote(gh, 0);
 397                         gfs2_holder_wake(gh);
 398                         continue;
 399                 }
 400                 if (gh->gh_list.prev == &gl->gl_holders)
 401                         return 1;
 402                 do_error(gl, 0);
 403                 break;
 404         }
 405         return 0;
 406 }
 407
 408 /**
 409  * find_first_waiter - find the first gh that's waiting for the glock
 410  * @gl: the glock
 411  */
 412
 413 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
 414 {
 415         struct gfs2_holder *gh;
 416
 417         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
 418                 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
 419                         return gh;
 420         }
 421         return NULL;
 422 }
 423
 424 /**
 425  * state_change - record that the glock is now in a different state
 426  * @gl: the glock
 427  * @new_state the new state
 428  *
 429  */
 430
 431 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
 432 {
 433         int held1, held2;
 434
 435         held1 = (gl->gl_state != LM_ST_UNLOCKED);
 436         held2 = (new_state != LM_ST_UNLOCKED);
 437
 438         if (held1 != held2) {
 439                 if (held2)
 440                         gfs2_glock_hold(gl);
 441                 else
 442                         gfs2_glock_put_nolock(gl);
 443         }
 444         if (held1 && held2 && list_empty(&gl->gl_holders))
 445                 clear_bit(GLF_QUEUED, &gl->gl_flags);
 446
 447         gl->gl_state = new_state;
 448         gl->gl_tchange = jiffies;
 449 }
 450
 451 static void gfs2_demote_wake(struct gfs2_glock *gl)
 452 {
 453         gl->gl_demote_state = LM_ST_EXCLUSIVE;
 454         clear_bit(GLF_DEMOTE, &gl->gl_flags);
 455         smp_mb__after_clear_bit();
 456         wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
 457 }
 458
 459 /**
 460  * finish_xmote - The DLM has replied to one of our lock requests
 461  * @gl: The glock
 462  * @ret: The status from the DLM
 463  *
 464  */
 465
 466 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
 467 {
 468         const struct gfs2_glock_operations *glops = gl->gl_ops;
 469         struct gfs2_holder *gh;
 470         unsigned state = ret & LM_OUT_ST_MASK;
 471         int rv;
 472
 473         spin_lock(&gl->gl_spin);
 474         trace_gfs2_glock_state_change(gl, state);
 475         state_change(gl, state);
 476         gh = find_first_waiter(gl);
 477
 478         /* Demote to UN request arrived during demote to SH or DF */
 479         if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
 480             state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
 481                 gl->gl_target = LM_ST_UNLOCKED;
 482
 483         /* Check for state != intended state */
 484         if (unlikely(state != gl->gl_target)) {
 485                 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
 486                         /* move to back of queue and try next entry */
 487                         if (ret & LM_OUT_CANCELED) {
 488                                 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
 489                                         list_move_tail(&gh->gh_list, &gl->gl_holders);
 490                                 gh = find_first_waiter(gl);
 491                                 gl->gl_target = gh->gh_state;
 492                                 goto retry;
 493                         }
 494                         /* Some error or failed "try lock" - report it */
 495                         if ((ret & LM_OUT_ERROR) ||
 496                             (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
 497                                 gl->gl_target = gl->gl_state;
 498                                 do_error(gl, ret);
 499                                 goto out;
 500                         }
 501                 }
 502                 switch(state) {
 503                 /* Unlocked due to conversion deadlock, try again */
 504                 case LM_ST_UNLOCKED:
 505 retry:
 506                         do_xmote(gl, gh, gl->gl_target);
 507                         break;
 508                 /* Conversion fails, unlock and try again */
 509                 case LM_ST_SHARED:
 510                 case LM_ST_DEFERRED:
 511                         do_xmote(gl, gh, LM_ST_UNLOCKED);
 512                         break;
 513                 default: /* Everything else */
 514                         printk(KERN_ERR "GFS2: wanted %u got %u\n", gl->gl_target, state);
 515                         GLOCK_BUG_ON(gl, 1);
 516                 }
 517                 spin_unlock(&gl->gl_spin);
 518                 return;
 519         }
 520
 521         /* Fast path - we got what we asked for */
 522         if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
 523                 gfs2_demote_wake(gl);
 524         if (state != LM_ST_UNLOCKED) {
 525                 if (glops->go_xmote_bh) {
 526                         spin_unlock(&gl->gl_spin);
 527                         rv = glops->go_xmote_bh(gl, gh);
 528                         spin_lock(&gl->gl_spin);
 529                         if (rv) {
 530                                 do_error(gl, rv);
 531                                 goto out;
 532                         }
 533                 }
 534                 rv = do_promote(gl);
 535                 if (rv == 2)
 536                         goto out_locked;
 537         }
 538 out:
 539         clear_bit(GLF_LOCK, &gl->gl_flags);
 540 out_locked:
 541         spin_unlock(&gl->gl_spin);
 542 }
 543
 544 static unsigned int gfs2_lm_lock(struct gfs2_sbd *sdp, void *lock,
 545                                  unsigned int req_state,
 546                                  unsigned int flags)
 547 {
 548         int ret = LM_OUT_ERROR;
 549
 550         if (!sdp->sd_lockstruct.ls_ops->lm_lock)
 551                 return req_state == LM_ST_UNLOCKED ? 0 : req_state;
 552
 553         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 554                 ret = sdp->sd_lockstruct.ls_ops->lm_lock(lock,
 555                                                          req_state, flags);
 556         return ret;
 557 }
 558
 559 /**
 560  * do_xmote - Calls the DLM to change the state of a lock
 561  * @gl: The lock state
 562  * @gh: The holder (only for promotes)
 563  * @target: The target lock state
 564  *
 565  */
 566
 567 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
 568 __releases(&gl->gl_spin)
 569 __acquires(&gl->gl_spin)
 570 {
 571         const struct gfs2_glock_operations *glops = gl->gl_ops;
 572         struct gfs2_sbd *sdp = gl->gl_sbd;
 573         unsigned int lck_flags = gh ? gh->gh_flags : 0;
 574         int ret;
 575
 576         lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
 577                       LM_FLAG_PRIORITY);
 578         BUG_ON(gl->gl_state == target);
 579         BUG_ON(gl->gl_state == gl->gl_target);
 580         if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
 581             glops->go_inval) {
 582                 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 583                 do_error(gl, 0); /* Fail queued try locks */
 584         }
 585         spin_unlock(&gl->gl_spin);
 586         if (glops->go_xmote_th)
 587                 glops->go_xmote_th(gl);
 588         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 589                 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
 590         clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 591
 592         gfs2_glock_hold(gl);
 593         if (target != LM_ST_UNLOCKED && (gl->gl_state == LM_ST_SHARED ||
 594             gl->gl_state == LM_ST_DEFERRED) &&
 595             !(lck_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
 596                 lck_flags |= LM_FLAG_TRY_1CB;
 597         ret = gfs2_lm_lock(sdp, gl, target, lck_flags);
 598
 599         if (!(ret & LM_OUT_ASYNC)) {
 600                 finish_xmote(gl, ret);
 601                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 602                         gfs2_glock_put(gl);
 603         } else {
 604                 GLOCK_BUG_ON(gl, ret != LM_OUT_ASYNC);
 605         }
 606         spin_lock(&gl->gl_spin);
 607 }
 608
 609 /**
 610  * find_first_holder - find the first "holder" gh
 611  * @gl: the glock
 612  */
 613
 614 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
 615 {
 616         struct gfs2_holder *gh;
 617
 618         if (!list_empty(&gl->gl_holders)) {
 619                 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
 620                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 621                         return gh;
 622         }
 623         return NULL;
 624 }
 625
 626 /**
 627  * run_queue - do all outstanding tasks related to a glock
 628  * @gl: The glock in question
 629  * @nonblock: True if we must not block in run_queue
 630  *
 631  */
 632
 633 static void run_queue(struct gfs2_glock *gl, const int nonblock)
 634 __releases(&gl->gl_spin)
 635 __acquires(&gl->gl_spin)
 636 {
 637         struct gfs2_holder *gh = NULL;
 638         int ret;
 639
 640         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
 641                 return;
 642
 643         GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
 644
 645         if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
 646             gl->gl_demote_state != gl->gl_state) {
 647                 if (find_first_holder(gl))
 648                         goto out_unlock;
 649                 if (nonblock)
 650                         goto out_sched;
 651                 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
 652                 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
 653                 gl->gl_target = gl->gl_demote_state;
 654         } else {
 655                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
 656                         gfs2_demote_wake(gl);
 657                 ret = do_promote(gl);
 658                 if (ret == 0)
 659                         goto out_unlock;
 660                 if (ret == 2)
 661                         goto out;
 662                 gh = find_first_waiter(gl);
 663                 gl->gl_target = gh->gh_state;
 664                 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
 665                         do_error(gl, 0); /* Fail queued try locks */
 666         }
 667         do_xmote(gl, gh, gl->gl_target);
 668 out:
 669         return;
 670
 671 out_sched:
 672         clear_bit(GLF_LOCK, &gl->gl_flags);
 673         smp_mb__after_clear_bit();
 674         gfs2_glock_hold(gl);
 675         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 676                 gfs2_glock_put_nolock(gl);
 677         return;
 678
 679 out_unlock:
 680         clear_bit(GLF_LOCK, &gl->gl_flags);
 681         smp_mb__after_clear_bit();
 682         return;
 683 }
 684
 685 static void delete_work_func(struct work_struct *work)
 686 {
 687         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
 688         struct gfs2_sbd *sdp = gl->gl_sbd;
 689         struct gfs2_inode *ip = NULL;
 690         struct inode *inode;
 691         u64 no_addr = 0;
 692
 693         spin_lock(&gl->gl_spin);
 694         ip = (struct gfs2_inode *)gl->gl_object;
 695         if (ip)
 696                 no_addr = ip->i_no_addr;
 697         spin_unlock(&gl->gl_spin);
 698         if (ip) {
 699                 inode = gfs2_ilookup(sdp->sd_vfs, no_addr);
 700                 if (inode) {
 701                         d_prune_aliases(inode);
 702                         iput(inode);
 703                 }
 704         }
 705         gfs2_glock_put(gl);
 706 }
 707
 708 static void glock_work_func(struct work_struct *work)
 709 {
 710         unsigned long delay = 0;
 711         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
 712         int drop_ref = 0;
 713
 714         if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
 715                 finish_xmote(gl, gl->gl_reply);
 716                 drop_ref = 1;
 717         }
 718         spin_lock(&gl->gl_spin);
 719         if (test_and_clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
 720             gl->gl_state != LM_ST_UNLOCKED &&
 721             gl->gl_demote_state != LM_ST_EXCLUSIVE) {
 722                 unsigned long holdtime, now = jiffies;
 723                 holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
 724                 if (time_before(now, holdtime))
 725                         delay = holdtime - now;
 726                 set_bit(delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE, &gl->gl_flags);
 727         }
 728         run_queue(gl, 0);
 729         spin_unlock(&gl->gl_spin);
 730         if (!delay ||
 731             queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
 732                 gfs2_glock_put(gl);
 733         if (drop_ref)
 734                 gfs2_glock_put(gl);
 735 }
 736
 737 /**
 738  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
 739  * @sdp: The GFS2 superblock
 740  * @number: the lock number
 741  * @glops: The glock_operations to use
 742  * @create: If 0, don't create the glock if it doesn't exist
 743  * @glp: the glock is returned here
 744  *
 745  * This does not lock a glock, just finds/creates structures for one.
 746  *
 747  * Returns: errno
 748  */
 749
 750 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
 751                    const struct gfs2_glock_operations *glops, int create,
 752                    struct gfs2_glock **glp)
 753 {
 754         struct super_block *s = sdp->sd_vfs;
 755         struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
 756         struct gfs2_glock *gl, *tmp;
 757         unsigned int hash = gl_hash(sdp, &name);
 758         struct address_space *mapping;
 759
 760         read_lock(gl_lock_addr(hash));
 761         gl = search_bucket(hash, sdp, &name);
 762         read_unlock(gl_lock_addr(hash));
 763
 764         *glp = gl;
 765         if (gl)
 766                 return 0;
 767         if (!create)
 768                 return -ENOENT;
 769
 770         if (glops->go_flags & GLOF_ASPACE)
 771                 gl = kmem_cache_alloc(gfs2_glock_aspace_cachep, GFP_KERNEL);
 772         else
 773                 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
 774         if (!gl)
 775                 return -ENOMEM;
 776
 777         atomic_inc(&sdp->sd_glock_disposal);
 778         gl->gl_flags = 0;
 779         gl->gl_name = name;
 780         atomic_set(&gl->gl_ref, 1);
 781         gl->gl_state = LM_ST_UNLOCKED;
 782         gl->gl_target = LM_ST_UNLOCKED;
 783         gl->gl_demote_state = LM_ST_EXCLUSIVE;
 784         gl->gl_hash = hash;
 785         gl->gl_ops = glops;
 786         snprintf(gl->gl_strname, GDLM_STRNAME_BYTES, "%8x%16llx", name.ln_type, (unsigned long long)number);
 787         memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
 788         gl->gl_lksb.sb_lvbptr = gl->gl_lvb;
 789         gl->gl_tchange = jiffies;
 790         gl->gl_object = NULL;
 791         gl->gl_sbd = sdp;
 792         INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
 793         INIT_WORK(&gl->gl_delete, delete_work_func);
 794
 795         mapping = gfs2_glock2aspace(gl);
 796         if (mapping) {
 797                 mapping->a_ops = &gfs2_meta_aops;
 798                 mapping->host = s->s_bdev->bd_inode;
 799                 mapping->flags = 0;
 800                 mapping_set_gfp_mask(mapping, GFP_NOFS);
 801                 mapping->assoc_mapping = NULL;
 802                 mapping->backing_dev_info = s->s_bdi;
 803                 mapping->writeback_index = 0;
 804         }
 805
 806         write_lock(gl_lock_addr(hash));
 807         tmp = search_bucket(hash, sdp, &name);
 808         if (tmp) {
 809                 write_unlock(gl_lock_addr(hash));
 810                 glock_free(gl);
 811                 gl = tmp;
 812         } else {
 813                 hlist_add_head(&gl->gl_list, &gl_hash_table[hash].hb_list);
 814                 write_unlock(gl_lock_addr(hash));
 815         }
 816
 817         *glp = gl;
 818
 819         return 0;
 820 }
 821
 822 /**
 823  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
 824  * @gl: the glock
 825  * @state: the state we're requesting
 826  * @flags: the modifier flags
 827  * @gh: the holder structure
 828  *
 829  */
 830
 831 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
 832                       struct gfs2_holder *gh)
 833 {
 834         INIT_LIST_HEAD(&gh->gh_list);
 835         gh->gh_gl = gl;
 836         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 837         gh->gh_owner_pid = get_pid(task_pid(current));
 838         gh->gh_state = state;
 839         gh->gh_flags = flags;
 840         gh->gh_error = 0;
 841         gh->gh_iflags = 0;
 842         gfs2_glock_hold(gl);
 843 }
 844
 845 /**
 846  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
 847  * @state: the state we're requesting
 848  * @flags: the modifier flags
 849  * @gh: the holder structure
 850  *
 851  * Don't mess with the glock.
 852  *
 853  */
 854
 855 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
 856 {
 857         gh->gh_state = state;
 858         gh->gh_flags = flags;
 859         gh->gh_iflags = 0;
 860         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 861         if (gh->gh_owner_pid)
 862                 put_pid(gh->gh_owner_pid);
 863         gh->gh_owner_pid = get_pid(task_pid(current));
 864 }
 865
 866 /**
 867  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
 868  * @gh: the holder structure
 869  *
 870  */
 871
 872 void gfs2_holder_uninit(struct gfs2_holder *gh)
 873 {
 874         put_pid(gh->gh_owner_pid);
 875         gfs2_glock_put(gh->gh_gl);
 876         gh->gh_gl = NULL;
 877         gh->gh_ip = 0;
 878 }
 879
 880 /**
 881  * gfs2_glock_holder_wait
 882  * @word: unused
 883  *
 884  * This function and gfs2_glock_demote_wait both show up in the WCHAN
 885  * field. Thus I've separated these otherwise identical functions in
 886  * order to be more informative to the user.
 887  */
 888
 889 static int gfs2_glock_holder_wait(void *word)
 890 {
 891         schedule();
 892         return 0;
 893 }
 894
 895 static int gfs2_glock_demote_wait(void *word)
 896 {
 897         schedule();
 898         return 0;
 899 }
 900
 901 static void wait_on_holder(struct gfs2_holder *gh)
 902 {
 903         might_sleep();
 904         wait_on_bit(&gh->gh_iflags, HIF_WAIT, gfs2_glock_holder_wait, TASK_UNINTERRUPTIBLE);
 905 }
 906
 907 static void wait_on_demote(struct gfs2_glock *gl)
 908 {
 909         might_sleep();
 910         wait_on_bit(&gl->gl_flags, GLF_DEMOTE, gfs2_glock_demote_wait, TASK_UNINTERRUPTIBLE);
 911 }
 912
 913 /**
 914  * handle_callback - process a demote request
 915  * @gl: the glock
 916  * @state: the state the caller wants us to change to
 917  *
 918  * There are only two requests that we are going to see in actual
 919  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
 920  */
 921
 922 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
 923                             unsigned long delay)
 924 {
 925         int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
 926
 927         set_bit(bit, &gl->gl_flags);
 928         if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
 929                 gl->gl_demote_state = state;
 930                 gl->gl_demote_time = jiffies;
 931         } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
 932                         gl->gl_demote_state != state) {
 933                 gl->gl_demote_state = LM_ST_UNLOCKED;
 934         }
 935         if (gl->gl_ops->go_callback)
 936                 gl->gl_ops->go_callback(gl);
 937         trace_gfs2_demote_rq(gl);
 938 }
 939
 940 /**
 941  * gfs2_glock_wait - wait on a glock acquisition
 942  * @gh: the glock holder
 943  *
 944  * Returns: 0 on success
 945  */
 946
 947 int gfs2_glock_wait(struct gfs2_holder *gh)
 948 {
 949         wait_on_holder(gh);
 950         return gh->gh_error;
 951 }
 952
 953 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
 954 {
 955         va_list args;
 956
 957         va_start(args, fmt);
 958         if (seq) {
 959                 struct gfs2_glock_iter *gi = seq->private;
 960                 vsprintf(gi->string, fmt, args);
 961                 seq_printf(seq, gi->string);
 962         } else {
 963                 printk(KERN_ERR " ");
 964                 vprintk(fmt, args);
 965         }
 966         va_end(args);
 967 }
 968
 969 /**
 970  * add_to_queue - Add a holder to the wait queue (but look for recursion)
 971  * @gh: the holder structure to add
 972  *
 973  * Eventually we should move the recursive locking trap to a
 974  * debugging option or something like that. This is the fast
 975  * path and needs to have the minimum number of distractions.
 976  *
 977  */
 978
 979 static inline void add_to_queue(struct gfs2_holder *gh)
 980 __releases(&gl->gl_spin)
 981 __acquires(&gl->gl_spin)
 982 {
 983         struct gfs2_glock *gl = gh->gh_gl;
 984         struct gfs2_sbd *sdp = gl->gl_sbd;
 985         struct list_head *insert_pt = NULL;
 986         struct gfs2_holder *gh2;
 987         int try_lock = 0;
 988
 989         BUG_ON(gh->gh_owner_pid == NULL);
 990         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
 991                 BUG();
 992
 993         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
 994                 if (test_bit(GLF_LOCK, &gl->gl_flags))
 995                         try_lock = 1;
 996                 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 997                         goto fail;
 998         }
 999
1000         list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1001                 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
1002                     (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
1003                         goto trap_recursive;
1004                 if (try_lock &&
1005                     !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) &&
1006                     !may_grant(gl, gh)) {
1007 fail:
1008                         gh->gh_error = GLR_TRYFAILED;
1009                         gfs2_holder_wake(gh);
1010                         return;
1011                 }
1012                 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1013                         continue;
1014                 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1015                         insert_pt = &gh2->gh_list;
1016         }
1017         set_bit(GLF_QUEUED, &gl->gl_flags);
1018         if (likely(insert_pt == NULL)) {
1019                 list_add_tail(&gh->gh_list, &gl->gl_holders);
1020                 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1021                         goto do_cancel;
1022                 return;
1023         }
1024         trace_gfs2_glock_queue(gh, 1);
1025         list_add_tail(&gh->gh_list, insert_pt);
1026 do_cancel:
1027         gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
1028         if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1029                 spin_unlock(&gl->gl_spin);
1030                 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1031                         sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1032                 spin_lock(&gl->gl_spin);
1033         }
1034         return;
1035
1036 trap_recursive:
1037         print_symbol(KERN_ERR "original: %s\n", gh2->gh_ip);
1038         printk(KERN_ERR "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1039         printk(KERN_ERR "lock type: %d req lock state : %d\n",
1040                gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1041         print_symbol(KERN_ERR "new: %s\n", gh->gh_ip);
1042         printk(KERN_ERR "pid: %d\n", pid_nr(gh->gh_owner_pid));
1043         printk(KERN_ERR "lock type: %d req lock state : %d\n",
1044                gh->gh_gl->gl_name.ln_type, gh->gh_state);
1045         __dump_glock(NULL, gl);
1046         BUG();
1047 }
1048
1049 /**
1050  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1051  * @gh: the holder structure
1052  *
1053  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1054  *
1055  * Returns: 0, GLR_TRYFAILED, or errno on failure
1056  */
1057
1058 int gfs2_glock_nq(struct gfs2_holder *gh)
1059 {
1060         struct gfs2_glock *gl = gh->gh_gl;
1061         struct gfs2_sbd *sdp = gl->gl_sbd;
1062         int error = 0;
1063
1064         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1065                 return -EIO;
1066
1067         spin_lock(&gl->gl_spin);
1068         add_to_queue(gh);
1069         if ((LM_FLAG_NOEXP & gh->gh_flags) &&
1070             test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1071                 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1072         run_queue(gl, 1);
1073         spin_unlock(&gl->gl_spin);
1074
1075         if (!(gh->gh_flags & GL_ASYNC))
1076                 error = gfs2_glock_wait(gh);
1077
1078         return error;
1079 }
1080
1081 /**
1082  * gfs2_glock_poll - poll to see if an async request has been completed
1083  * @gh: the holder
1084  *
1085  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1086  */
1087
1088 int gfs2_glock_poll(struct gfs2_holder *gh)
1089 {
1090         return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1091 }
1092
1093 /**
1094  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1095  * @gh: the glock holder
1096  *
1097  */
1098
1099 void gfs2_glock_dq(struct gfs2_holder *gh)
1100 {
1101         struct gfs2_glock *gl = gh->gh_gl;
1102         const struct gfs2_glock_operations *glops = gl->gl_ops;
1103         unsigned delay = 0;
1104         int fast_path = 0;
1105
1106         spin_lock(&gl->gl_spin);
1107         if (gh->gh_flags & GL_NOCACHE)
1108                 handle_callback(gl, LM_ST_UNLOCKED, 0);
1109
1110         list_del_init(&gh->gh_list);
1111         if (find_first_holder(gl) == NULL) {
1112                 if (glops->go_unlock) {
1113                         GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1114                         spin_unlock(&gl->gl_spin);
1115                         glops->go_unlock(gh);
1116                         spin_lock(&gl->gl_spin);
1117                         clear_bit(GLF_LOCK, &gl->gl_flags);
1118                 }
1119                 if (list_empty(&gl->gl_holders) &&
1120                     !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1121                     !test_bit(GLF_DEMOTE, &gl->gl_flags))
1122                         fast_path = 1;
1123         }
1124         trace_gfs2_glock_queue(gh, 0);
1125         spin_unlock(&gl->gl_spin);
1126         if (likely(fast_path))
1127                 return;
1128
1129         gfs2_glock_hold(gl);
1130         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1131             !test_bit(GLF_DEMOTE, &gl->gl_flags))
1132                 delay = gl->gl_ops->go_min_hold_time;
1133         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1134                 gfs2_glock_put(gl);
1135 }
1136
1137 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1138 {
1139         struct gfs2_glock *gl = gh->gh_gl;
1140         gfs2_glock_dq(gh);
1141         wait_on_demote(gl);
1142 }
1143
1144 /**
1145  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1146  * @gh: the holder structure
1147  *
1148  */
1149
1150 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1151 {
1152         gfs2_glock_dq(gh);
1153         gfs2_holder_uninit(gh);
1154 }
1155
1156 /**
1157  * gfs2_glock_nq_num - acquire a glock based on lock number
1158  * @sdp: the filesystem
1159  * @number: the lock number
1160  * @glops: the glock operations for the type of glock
1161  * @state: the state to acquire the glock in
1162  * @flags: modifier flags for the aquisition
1163  * @gh: the struct gfs2_holder
1164  *
1165  * Returns: errno
1166  */
1167
1168 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1169                       const struct gfs2_glock_operations *glops,
1170                       unsigned int state, int flags, struct gfs2_holder *gh)
1171 {
1172         struct gfs2_glock *gl;
1173         int error;
1174
1175         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1176         if (!error) {
1177                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1178                 gfs2_glock_put(gl);
1179         }
1180
1181         return error;
1182 }
1183
1184 /**
1185  * glock_compare - Compare two struct gfs2_glock structures for sorting
1186  * @arg_a: the first structure
1187  * @arg_b: the second structure
1188  *
1189  */
1190
1191 static int glock_compare(const void *arg_a, const void *arg_b)
1192 {
1193         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1194         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1195         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1196         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1197
1198         if (a->ln_number > b->ln_number)
1199                 return 1;
1200         if (a->ln_number < b->ln_number)
1201                 return -1;
1202         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1203         return 0;
1204 }
1205
1206 /**
1207  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1208  * @num_gh: the number of structures
1209  * @ghs: an array of struct gfs2_holder structures
1210  *
1211  * Returns: 0 on success (all glocks acquired),
1212  *          errno on failure (no glocks acquired)
1213  */
1214
1215 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1216                      struct gfs2_holder **p)
1217 {
1218         unsigned int x;
1219         int error = 0;
1220
1221         for (x = 0; x < num_gh; x++)
1222                 p[x] = &ghs[x];
1223
1224         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1225
1226         for (x = 0; x < num_gh; x++) {
1227                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1228
1229                 error = gfs2_glock_nq(p[x]);
1230                 if (error) {
1231                         while (x--)
1232                                 gfs2_glock_dq(p[x]);
1233                         break;
1234                 }
1235         }
1236
1237         return error;
1238 }
1239
1240 /**
1241  * gfs2_glock_nq_m - acquire multiple glocks
1242  * @num_gh: the number of structures
1243  * @ghs: an array of struct gfs2_holder structures
1244  *
1245  *
1246  * Returns: 0 on success (all glocks acquired),
1247  *          errno on failure (no glocks acquired)
1248  */
1249
1250 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1251 {
1252         struct gfs2_holder *tmp[4];
1253         struct gfs2_holder **pph = tmp;
1254         int error = 0;
1255
1256         switch(num_gh) {
1257         case 0:
1258                 return 0;
1259         case 1:
1260                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1261                 return gfs2_glock_nq(ghs);
1262         default:
1263                 if (num_gh <= 4)
1264                         break;
1265                 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1266                 if (!pph)
1267                         return -ENOMEM;
1268         }
1269
1270         error = nq_m_sync(num_gh, ghs, pph);
1271
1272         if (pph != tmp)
1273                 kfree(pph);
1274
1275         return error;
1276 }
1277
1278 /**
1279  * gfs2_glock_dq_m - release multiple glocks
1280  * @num_gh: the number of structures
1281  * @ghs: an array of struct gfs2_holder structures
1282  *
1283  */
1284
1285 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1286 {
1287         unsigned int x;
1288
1289         for (x = 0; x < num_gh; x++)
1290                 gfs2_glock_dq(&ghs[x]);
1291 }
1292
1293 /**
1294  * gfs2_glock_dq_uninit_m - release multiple glocks
1295  * @num_gh: the number of structures
1296  * @ghs: an array of struct gfs2_holder structures
1297  *
1298  */
1299
1300 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1301 {
1302         unsigned int x;
1303
1304         for (x = 0; x < num_gh; x++)
1305                 gfs2_glock_dq_uninit(&ghs[x]);
1306 }
1307
1308 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1309 {
1310         unsigned long delay = 0;
1311         unsigned long holdtime;
1312         unsigned long now = jiffies;
1313
1314         gfs2_glock_hold(gl);
1315         holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
1316         if (test_bit(GLF_QUEUED, &gl->gl_flags)) {
1317                 if (time_before(now, holdtime))
1318                         delay = holdtime - now;
1319                 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1320                         delay = gl->gl_ops->go_min_hold_time;
1321         }
1322
1323         spin_lock(&gl->gl_spin);
1324         handle_callback(gl, state, delay);
1325         spin_unlock(&gl->gl_spin);
1326         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1327                 gfs2_glock_put(gl);
1328 }
1329
1330 /**
1331  * gfs2_should_freeze - Figure out if glock should be frozen
1332  * @gl: The glock in question
1333  *
1334  * Glocks are not frozen if (a) the result of the dlm operation is
1335  * an error, (b) the locking operation was an unlock operation or
1336  * (c) if there is a "noexp" flagged request anywhere in the queue
1337  *
1338  * Returns: 1 if freezing should occur, 0 otherwise
1339  */
1340
1341 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1342 {
1343         const struct gfs2_holder *gh;
1344
1345         if (gl->gl_reply & ~LM_OUT_ST_MASK)
1346                 return 0;
1347         if (gl->gl_target == LM_ST_UNLOCKED)
1348                 return 0;
1349
1350         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1351                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1352                         continue;
1353                 if (LM_FLAG_NOEXP & gh->gh_flags)
1354                         return 0;
1355         }
1356
1357         return 1;
1358 }
1359
1360 /**
1361  * gfs2_glock_complete - Callback used by locking
1362  * @gl: Pointer to the glock
1363  * @ret: The return value from the dlm
1364  *
1365  */
1366
1367 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1368 {
1369         struct lm_lockstruct *ls = &gl->gl_sbd->sd_lockstruct;
1370
1371         gl->gl_reply = ret;
1372
1373         if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_flags))) {
1374                 spin_lock(&gl->gl_spin);
1375                 if (gfs2_should_freeze(gl)) {
1376                         set_bit(GLF_FROZEN, &gl->gl_flags);
1377                         spin_unlock(&gl->gl_spin);
1378                         return;
1379                 }
1380                 spin_unlock(&gl->gl_spin);
1381         }
1382         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1383         gfs2_glock_hold(gl);
1384         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1385                 gfs2_glock_put(gl);
1386 }
1387
1388
1389 static int gfs2_shrink_glock_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
1390 {
1391         struct gfs2_glock *gl;
1392         int may_demote;
1393         int nr_skipped = 0;
1394         LIST_HEAD(skipped);
1395
1396         if (nr == 0)
1397                 goto out;
1398
1399         if (!(gfp_mask & __GFP_FS))
1400                 return -1;
1401
1402         spin_lock(&lru_lock);
1403         while(nr && !list_empty(&lru_list)) {
1404                 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1405                 list_del_init(&gl->gl_lru);
1406                 atomic_dec(&lru_count);
1407
1408                 /* Test for being demotable */
1409                 if (!test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1410                         gfs2_glock_hold(gl);
1411                         spin_unlock(&lru_lock);
1412                         spin_lock(&gl->gl_spin);
1413                         may_demote = demote_ok(gl);
1414                         if (may_demote) {
1415                                 handle_callback(gl, LM_ST_UNLOCKED, 0);
1416                                 nr--;
1417                         }
1418                         clear_bit(GLF_LOCK, &gl->gl_flags);
1419                         smp_mb__after_clear_bit();
1420                         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1421                                 gfs2_glock_put_nolock(gl);
1422                         spin_unlock(&gl->gl_spin);
1423                         spin_lock(&lru_lock);
1424                         continue;
1425                 }
1426                 nr_skipped++;
1427                 list_add(&gl->gl_lru, &skipped);
1428         }
1429         list_splice(&skipped, &lru_list);
1430         atomic_add(nr_skipped, &lru_count);
1431         spin_unlock(&lru_lock);
1432 out:
1433         return (atomic_read(&lru_count) / 100) * sysctl_vfs_cache_pressure;
1434 }
1435
1436 static struct shrinker glock_shrinker = {
1437         .shrink = gfs2_shrink_glock_memory,
1438         .seeks = DEFAULT_SEEKS,
1439 };
1440
1441 /**
1442  * examine_bucket - Call a function for glock in a hash bucket
1443  * @examiner: the function
1444  * @sdp: the filesystem
1445  * @bucket: the bucket
1446  *
1447  * Returns: 1 if the bucket has entries
1448  */
1449
1450 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1451                           unsigned int hash)
1452 {
1453         struct gfs2_glock *gl, *prev = NULL;
1454         int has_entries = 0;
1455         struct hlist_head *head = &gl_hash_table[hash].hb_list;
1456
1457         read_lock(gl_lock_addr(hash));
1458         /* Can't use hlist_for_each_entry - don't want prefetch here */
1459         if (hlist_empty(head))
1460                 goto out;
1461         gl = list_entry(head->first, struct gfs2_glock, gl_list);
1462         while(1) {
1463                 if (!sdp || gl->gl_sbd == sdp) {
1464                         gfs2_glock_hold(gl);
1465                         read_unlock(gl_lock_addr(hash));
1466                         if (prev)
1467                                 gfs2_glock_put(prev);
1468                         prev = gl;
1469                         examiner(gl);
1470                         has_entries = 1;
1471                         read_lock(gl_lock_addr(hash));
1472                 }
1473                 if (gl->gl_list.next == NULL)
1474                         break;
1475                 gl = list_entry(gl->gl_list.next, struct gfs2_glock, gl_list);
1476         }
1477 out:
1478         read_unlock(gl_lock_addr(hash));
1479         if (prev)
1480                 gfs2_glock_put(prev);
1481         cond_resched();
1482         return has_entries;
1483 }
1484
1485
1486 /**
1487  * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1488  * @gl: The glock to thaw
1489  *
1490  * N.B. When we freeze a glock, we leave a ref to the glock outstanding,
1491  * so this has to result in the ref count being dropped by one.
1492  */
1493
1494 static void thaw_glock(struct gfs2_glock *gl)
1495 {
1496         if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1497                 return;
1498         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1499         gfs2_glock_hold(gl);
1500         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1501                 gfs2_glock_put(gl);
1502 }
1503
1504 /**
1505  * clear_glock - look at a glock and see if we can free it from glock cache
1506  * @gl: the glock to look at
1507  *
1508  */
1509
1510 static void clear_glock(struct gfs2_glock *gl)
1511 {
1512         spin_lock(&lru_lock);
1513         if (!list_empty(&gl->gl_lru)) {
1514                 list_del_init(&gl->gl_lru);
1515                 atomic_dec(&lru_count);
1516         }
1517         spin_unlock(&lru_lock);
1518
1519         spin_lock(&gl->gl_spin);
1520         if (gl->gl_state != LM_ST_UNLOCKED)
1521                 handle_callback(gl, LM_ST_UNLOCKED, 0);
1522         spin_unlock(&gl->gl_spin);
1523         gfs2_glock_hold(gl);
1524         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1525                 gfs2_glock_put(gl);
1526 }
1527
1528 /**
1529  * gfs2_glock_thaw - Thaw any frozen glocks
1530  * @sdp: The super block
1531  *
1532  */
1533
1534 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1535 {
1536         unsigned x;
1537
1538         for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1539                 examine_bucket(thaw_glock, sdp, x);
1540 }
1541
1542 /**
1543  * gfs2_gl_hash_clear - Empty out the glock hash table
1544  * @sdp: the filesystem
1545  * @wait: wait until it's all gone
1546  *
1547  * Called when unmounting the filesystem.
1548  */
1549
1550 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1551 {
1552         unsigned int x;
1553
1554         for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1555                 examine_bucket(clear_glock, sdp, x);
1556         flush_workqueue(glock_workqueue);
1557         wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
1558         gfs2_dump_lockstate(sdp);
1559 }
1560
1561 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1562 {
1563         struct gfs2_glock *gl = ip->i_gl;
1564         int ret;
1565
1566         ret = gfs2_truncatei_resume(ip);
1567         gfs2_assert_withdraw(gl->gl_sbd, ret == 0);
1568
1569         spin_lock(&gl->gl_spin);
1570         clear_bit(GLF_LOCK, &gl->gl_flags);
1571         run_queue(gl, 1);
1572         spin_unlock(&gl->gl_spin);
1573 }
1574
1575 static const char *state2str(unsigned state)
1576 {
1577         switch(state) {
1578         case LM_ST_UNLOCKED:
1579                 return "UN";
1580         case LM_ST_SHARED:
1581                 return "SH";
1582         case LM_ST_DEFERRED:
1583                 return "DF";
1584         case LM_ST_EXCLUSIVE:
1585                 return "EX";
1586         }
1587         return "??";
1588 }
1589
1590 static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1591 {
1592         char *p = buf;
1593         if (flags & LM_FLAG_TRY)
1594                 *p++ = 't';
1595         if (flags & LM_FLAG_TRY_1CB)
1596                 *p++ = 'T';
1597         if (flags & LM_FLAG_NOEXP)
1598                 *p++ = 'e';
1599         if (flags & LM_FLAG_ANY)
1600                 *p++ = 'A';
1601         if (flags & LM_FLAG_PRIORITY)
1602                 *p++ = 'p';
1603         if (flags & GL_ASYNC)
1604                 *p++ = 'a';
1605         if (flags & GL_EXACT)
1606                 *p++ = 'E';
1607         if (flags & GL_NOCACHE)
1608                 *p++ = 'c';
1609         if (test_bit(HIF_HOLDER, &iflags))
1610                 *p++ = 'H';
1611         if (test_bit(HIF_WAIT, &iflags))
1612                 *p++ = 'W';
1613         if (test_bit(HIF_FIRST, &iflags))
1614                 *p++ = 'F';
1615         *p = 0;
1616         return buf;
1617 }
1618
1619 /**
1620  * dump_holder - print information about a glock holder
1621  * @seq: the seq_file struct
1622  * @gh: the glock holder
1623  *
1624  * Returns: 0 on success, -ENOBUFS when we run out of space
1625  */
1626
1627 static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1628 {
1629         struct task_struct *gh_owner = NULL;
1630         char buffer[KSYM_SYMBOL_LEN];
1631         char flags_buf[32];
1632
1633         sprint_symbol(buffer, gh->gh_ip);
1634         if (gh->gh_owner_pid)
1635                 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1636         gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %s\n",
1637                   state2str(gh->gh_state),
1638                   hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1639                   gh->gh_error,
1640                   gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1641                   gh_owner ? gh_owner->comm : "(ended)", buffer);
1642         return 0;
1643 }
1644
1645 static const char *gflags2str(char *buf, const unsigned long *gflags)
1646 {
1647         char *p = buf;
1648         if (test_bit(GLF_LOCK, gflags))
1649                 *p++ = 'l';
1650         if (test_bit(GLF_DEMOTE, gflags))
1651                 *p++ = 'D';
1652         if (test_bit(GLF_PENDING_DEMOTE, gflags))
1653                 *p++ = 'd';
1654         if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1655                 *p++ = 'p';
1656         if (test_bit(GLF_DIRTY, gflags))
1657                 *p++ = 'y';
1658         if (test_bit(GLF_LFLUSH, gflags))
1659                 *p++ = 'f';
1660         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1661                 *p++ = 'i';
1662         if (test_bit(GLF_REPLY_PENDING, gflags))
1663                 *p++ = 'r';
1664         if (test_bit(GLF_INITIAL, gflags))
1665                 *p++ = 'I';
1666         if (test_bit(GLF_FROZEN, gflags))
1667                 *p++ = 'F';
1668         if (test_bit(GLF_QUEUED, gflags))
1669                 *p++ = 'q';
1670         *p = 0;
1671         return buf;
1672 }
1673
1674 /**
1675  * __dump_glock - print information about a glock
1676  * @seq: The seq_file struct
1677  * @gl: the glock
1678  *
1679  * The file format is as follows:
1680  * One line per object, capital letters are used to indicate objects
1681  * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1682  * other objects are indented by a single space and follow the glock to
1683  * which they are related. Fields are indicated by lower case letters
1684  * followed by a colon and the field value, except for strings which are in
1685  * [] so that its possible to see if they are composed of spaces for
1686  * example. The field's are n = number (id of the object), f = flags,
1687  * t = type, s = state, r = refcount, e = error, p = pid.
1688  *
1689  * Returns: 0 on success, -ENOBUFS when we run out of space
1690  */
1691
1692 static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1693 {
1694         const struct gfs2_glock_operations *glops = gl->gl_ops;
1695         unsigned long long dtime;
1696         const struct gfs2_holder *gh;
1697         char gflags_buf[32];
1698         int error = 0;
1699
1700         dtime = jiffies - gl->gl_demote_time;
1701         dtime *= 1000000/HZ; /* demote time in uSec */
1702         if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1703                 dtime = 0;
1704         gfs2_print_dbg(seq, "G:  s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d r:%d\n",
1705                   state2str(gl->gl_state),
1706                   gl->gl_name.ln_type,
1707                   (unsigned long long)gl->gl_name.ln_number,
1708                   gflags2str(gflags_buf, &gl->gl_flags),
1709                   state2str(gl->gl_target),
1710                   state2str(gl->gl_demote_state), dtime,
1711                   atomic_read(&gl->gl_ail_count),
1712                   atomic_read(&gl->gl_ref));
1713
1714         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1715                 error = dump_holder(seq, gh);
1716                 if (error)
1717                         goto out;
1718         }
1719         if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1720                 error = glops->go_dump(seq, gl);
1721 out:
1722         return error;
1723 }
1724
1725 static int dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1726 {
1727         int ret;
1728         spin_lock(&gl->gl_spin);
1729         ret = __dump_glock(seq, gl);
1730         spin_unlock(&gl->gl_spin);
1731         return ret;
1732 }
1733
1734 /**
1735  * gfs2_dump_lockstate - print out the current lockstate
1736  * @sdp: the filesystem
1737  * @ub: the buffer to copy the information into
1738  *
1739  * If @ub is NULL, dump the lockstate to the console.
1740  *
1741  */
1742
1743 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
1744 {
1745         struct gfs2_glock *gl;
1746         struct hlist_node *h;
1747         unsigned int x;
1748         int error = 0;
1749
1750         for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1751
1752                 read_lock(gl_lock_addr(x));
1753
1754                 hlist_for_each_entry(gl, h, &gl_hash_table[x].hb_list, gl_list) {
1755                         if (gl->gl_sbd != sdp)
1756                                 continue;
1757
1758                         error = dump_glock(NULL, gl);
1759                         if (error)
1760                                 break;
1761                 }
1762
1763                 read_unlock(gl_lock_addr(x));
1764
1765                 if (error)
1766                         break;
1767         }
1768
1769
1770         return error;
1771 }
1772
1773
1774 int __init gfs2_glock_init(void)
1775 {
1776         unsigned i;
1777         for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
1778                 INIT_HLIST_HEAD(&gl_hash_table[i].hb_list);
1779         }
1780 #ifdef GL_HASH_LOCK_SZ
1781         for(i = 0; i < GL_HASH_LOCK_SZ; i++) {
1782                 rwlock_init(&gl_hash_locks[i]);
1783         }
1784 #endif
1785
1786         glock_workqueue = alloc_workqueue("glock_workqueue", WQ_RESCUER |
1787                                           WQ_HIGHPRI | WQ_FREEZEABLE, 0);
1788         if (IS_ERR(glock_workqueue))
1789                 return PTR_ERR(glock_workqueue);
1790         gfs2_delete_workqueue = alloc_workqueue("delete_workqueue", WQ_RESCUER |
1791                                                 WQ_FREEZEABLE, 0);
1792         if (IS_ERR(gfs2_delete_workqueue)) {
1793                 destroy_workqueue(glock_workqueue);
1794                 return PTR_ERR(gfs2_delete_workqueue);
1795         }
1796
1797         register_shrinker(&glock_shrinker);
1798
1799         return 0;
1800 }
1801
1802 void gfs2_glock_exit(void)
1803 {
1804         unregister_shrinker(&glock_shrinker);
1805         destroy_workqueue(glock_workqueue);
1806         destroy_workqueue(gfs2_delete_workqueue);
1807 }
1808
1809 static int gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1810 {
1811         struct gfs2_glock *gl;
1812
1813 restart:
1814         read_lock(gl_lock_addr(gi->hash));
1815         gl = gi->gl;
1816         if (gl) {
1817                 gi->gl = hlist_entry(gl->gl_list.next,
1818                                      struct gfs2_glock, gl_list);
1819         } else {
1820                 gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
1821                                      struct gfs2_glock, gl_list);
1822         }
1823         if (gi->gl)
1824                 gfs2_glock_hold(gi->gl);
1825         read_unlock(gl_lock_addr(gi->hash));
1826         if (gl)
1827                 gfs2_glock_put(gl);
1828         while (gi->gl == NULL) {
1829                 gi->hash++;
1830                 if (gi->hash >= GFS2_GL_HASH_SIZE)
1831                         return 1;
1832                 read_lock(gl_lock_addr(gi->hash));
1833                 gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
1834                                      struct gfs2_glock, gl_list);
1835                 if (gi->gl)
1836                         gfs2_glock_hold(gi->gl);
1837                 read_unlock(gl_lock_addr(gi->hash));
1838         }
1839
1840         if (gi->sdp != gi->gl->gl_sbd)
1841                 goto restart;
1842
1843         return 0;
1844 }
1845
1846 static void gfs2_glock_iter_free(struct gfs2_glock_iter *gi)
1847 {
1848         if (gi->gl)
1849                 gfs2_glock_put(gi->gl);
1850         gi->gl = NULL;
1851 }
1852
1853 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1854 {
1855         struct gfs2_glock_iter *gi = seq->private;
1856         loff_t n = *pos;
1857
1858         gi->hash = 0;
1859
1860         do {
1861                 if (gfs2_glock_iter_next(gi)) {
1862                         gfs2_glock_iter_free(gi);
1863                         return NULL;
1864                 }
1865         } while (n--);
1866
1867         return gi->gl;
1868 }
1869
1870 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1871                                  loff_t *pos)
1872 {
1873         struct gfs2_glock_iter *gi = seq->private;
1874
1875         (*pos)++;
1876
1877         if (gfs2_glock_iter_next(gi)) {
1878                 gfs2_glock_iter_free(gi);
1879                 return NULL;
1880         }
1881
1882         return gi->gl;
1883 }
1884
1885 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1886 {
1887         struct gfs2_glock_iter *gi = seq->private;
1888         gfs2_glock_iter_free(gi);
1889 }
1890
1891 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1892 {
1893         return dump_glock(seq, iter_ptr);
1894 }
1895
1896 static const struct seq_operations gfs2_glock_seq_ops = {
1897         .start = gfs2_glock_seq_start,
1898         .next  = gfs2_glock_seq_next,
1899         .stop  = gfs2_glock_seq_stop,
1900         .show  = gfs2_glock_seq_show,
1901 };
1902
1903 static int gfs2_debugfs_open(struct inode *inode, struct file *file)
1904 {
1905         int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1906                                    sizeof(struct gfs2_glock_iter));
1907         if (ret == 0) {
1908                 struct seq_file *seq = file->private_data;
1909                 struct gfs2_glock_iter *gi = seq->private;
1910                 gi->sdp = inode->i_private;
1911         }
1912         return ret;
1913 }
1914
1915 static const struct file_operations gfs2_debug_fops = {
1916         .owner   = THIS_MODULE,
1917         .open    = gfs2_debugfs_open,
1918         .read    = seq_read,
1919         .llseek  = seq_lseek,
1920         .release = seq_release_private,
1921 };
1922
1923 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
1924 {
1925         sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
1926         if (!sdp->debugfs_dir)
1927                 return -ENOMEM;
1928         sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
1929                                                          S_IFREG | S_IRUGO,
1930                                                          sdp->debugfs_dir, sdp,
1931                                                          &gfs2_debug_fops);
1932         if (!sdp->debugfs_dentry_glocks)
1933                 return -ENOMEM;
1934
1935         return 0;
1936 }
1937
1938 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
1939 {
1940         if (sdp && sdp->debugfs_dir) {
1941                 if (sdp->debugfs_dentry_glocks) {
1942                         debugfs_remove(sdp->debugfs_dentry_glocks);
1943                         sdp->debugfs_dentry_glocks = NULL;
1944                 }
1945                 debugfs_remove(sdp->debugfs_dir);
1946                 sdp->debugfs_dir = NULL;
1947         }
1948 }
1949
1950 int gfs2_register_debugfs(void)
1951 {
1952         gfs2_root = debugfs_create_dir("gfs2", NULL);
1953         return gfs2_root ? 0 : -ENOMEM;
1954 }
1955
1956 void gfs2_unregister_debugfs(void)
1957 {
1958         debugfs_remove(gfs2_root);
1959         gfs2_root = NULL;
1960 }