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 /**
 545  * do_xmote - Calls the DLM to change the state of a lock
 546  * @gl: The lock state
 547  * @gh: The holder (only for promotes)
 548  * @target: The target lock state
 549  *
 550  */
 551
 552 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
 553 __releases(&gl->gl_spin)
 554 __acquires(&gl->gl_spin)
 555 {
 556         const struct gfs2_glock_operations *glops = gl->gl_ops;
 557         struct gfs2_sbd *sdp = gl->gl_sbd;
 558         unsigned int lck_flags = gh ? gh->gh_flags : 0;
 559         int ret;
 560
 561         lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
 562                       LM_FLAG_PRIORITY);
 563         GLOCK_BUG_ON(gl, gl->gl_state == target);
 564         GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
 565         if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
 566             glops->go_inval) {
 567                 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 568                 do_error(gl, 0); /* Fail queued try locks */
 569         }
 570         gl->gl_req = target;
 571         spin_unlock(&gl->gl_spin);
 572         if (glops->go_xmote_th)
 573                 glops->go_xmote_th(gl);
 574         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 575                 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
 576         clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 577
 578         gfs2_glock_hold(gl);
 579         if (target != LM_ST_UNLOCKED && (gl->gl_state == LM_ST_SHARED ||
 580             gl->gl_state == LM_ST_DEFERRED) &&
 581             !(lck_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
 582                 lck_flags |= LM_FLAG_TRY_1CB;
 583
 584         if (sdp->sd_lockstruct.ls_ops->lm_lock) {
 585                 /* lock_dlm */
 586                 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
 587                 GLOCK_BUG_ON(gl, ret);
 588         } else { /* lock_nolock */
 589                 finish_xmote(gl, target);
 590                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 591                         gfs2_glock_put(gl);
 592         }
 593
 594         spin_lock(&gl->gl_spin);
 595 }
 596
 597 /**
 598  * find_first_holder - find the first "holder" gh
 599  * @gl: the glock
 600  */
 601
 602 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
 603 {
 604         struct gfs2_holder *gh;
 605
 606         if (!list_empty(&gl->gl_holders)) {
 607                 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
 608                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 609                         return gh;
 610         }
 611         return NULL;
 612 }
 613
 614 /**
 615  * run_queue - do all outstanding tasks related to a glock
 616  * @gl: The glock in question
 617  * @nonblock: True if we must not block in run_queue
 618  *
 619  */
 620
 621 static void run_queue(struct gfs2_glock *gl, const int nonblock)
 622 __releases(&gl->gl_spin)
 623 __acquires(&gl->gl_spin)
 624 {
 625         struct gfs2_holder *gh = NULL;
 626         int ret;
 627
 628         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
 629                 return;
 630
 631         GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
 632
 633         if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
 634             gl->gl_demote_state != gl->gl_state) {
 635                 if (find_first_holder(gl))
 636                         goto out_unlock;
 637                 if (nonblock)
 638                         goto out_sched;
 639                 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
 640                 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
 641                 gl->gl_target = gl->gl_demote_state;
 642         } else {
 643                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
 644                         gfs2_demote_wake(gl);
 645                 ret = do_promote(gl);
 646                 if (ret == 0)
 647                         goto out_unlock;
 648                 if (ret == 2)
 649                         goto out;
 650                 gh = find_first_waiter(gl);
 651                 gl->gl_target = gh->gh_state;
 652                 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
 653                         do_error(gl, 0); /* Fail queued try locks */
 654         }
 655         do_xmote(gl, gh, gl->gl_target);
 656 out:
 657         return;
 658
 659 out_sched:
 660         clear_bit(GLF_LOCK, &gl->gl_flags);
 661         smp_mb__after_clear_bit();
 662         gfs2_glock_hold(gl);
 663         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 664                 gfs2_glock_put_nolock(gl);
 665         return;
 666
 667 out_unlock:
 668         clear_bit(GLF_LOCK, &gl->gl_flags);
 669         smp_mb__after_clear_bit();
 670         return;
 671 }
 672
 673 static void delete_work_func(struct work_struct *work)
 674 {
 675         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
 676         struct gfs2_sbd *sdp = gl->gl_sbd;
 677         struct gfs2_inode *ip;
 678         struct inode *inode;
 679         u64 no_addr = gl->gl_name.ln_number;
 680
 681         ip = gl->gl_object;
 682         /* Note: Unsafe to dereference ip as we don't hold right refs/locks */
 683
 684         if (ip)
 685                 inode = gfs2_ilookup(sdp->sd_vfs, no_addr);
 686         else
 687                 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
 688         if (inode && !IS_ERR(inode)) {
 689                 d_prune_aliases(inode);
 690                 iput(inode);
 691         }
 692         gfs2_glock_put(gl);
 693 }
 694
 695 static void glock_work_func(struct work_struct *work)
 696 {
 697         unsigned long delay = 0;
 698         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
 699         int drop_ref = 0;
 700
 701         if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
 702                 finish_xmote(gl, gl->gl_reply);
 703                 drop_ref = 1;
 704         }
 705         spin_lock(&gl->gl_spin);
 706         if (test_and_clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
 707             gl->gl_state != LM_ST_UNLOCKED &&
 708             gl->gl_demote_state != LM_ST_EXCLUSIVE) {
 709                 unsigned long holdtime, now = jiffies;
 710                 holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
 711                 if (time_before(now, holdtime))
 712                         delay = holdtime - now;
 713                 set_bit(delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE, &gl->gl_flags);
 714         }
 715         run_queue(gl, 0);
 716         spin_unlock(&gl->gl_spin);
 717         if (!delay ||
 718             queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
 719                 gfs2_glock_put(gl);
 720         if (drop_ref)
 721                 gfs2_glock_put(gl);
 722 }
 723
 724 /**
 725  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
 726  * @sdp: The GFS2 superblock
 727  * @number: the lock number
 728  * @glops: The glock_operations to use
 729  * @create: If 0, don't create the glock if it doesn't exist
 730  * @glp: the glock is returned here
 731  *
 732  * This does not lock a glock, just finds/creates structures for one.
 733  *
 734  * Returns: errno
 735  */
 736
 737 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
 738                    const struct gfs2_glock_operations *glops, int create,
 739                    struct gfs2_glock **glp)
 740 {
 741         struct super_block *s = sdp->sd_vfs;
 742         struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
 743         struct gfs2_glock *gl, *tmp;
 744         unsigned int hash = gl_hash(sdp, &name);
 745         struct address_space *mapping;
 746
 747         read_lock(gl_lock_addr(hash));
 748         gl = search_bucket(hash, sdp, &name);
 749         read_unlock(gl_lock_addr(hash));
 750
 751         *glp = gl;
 752         if (gl)
 753                 return 0;
 754         if (!create)
 755                 return -ENOENT;
 756
 757         if (glops->go_flags & GLOF_ASPACE)
 758                 gl = kmem_cache_alloc(gfs2_glock_aspace_cachep, GFP_KERNEL);
 759         else
 760                 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
 761         if (!gl)
 762                 return -ENOMEM;
 763
 764         atomic_inc(&sdp->sd_glock_disposal);
 765         gl->gl_flags = 0;
 766         gl->gl_name = name;
 767         atomic_set(&gl->gl_ref, 1);
 768         gl->gl_state = LM_ST_UNLOCKED;
 769         gl->gl_target = LM_ST_UNLOCKED;
 770         gl->gl_demote_state = LM_ST_EXCLUSIVE;
 771         gl->gl_hash = hash;
 772         gl->gl_ops = glops;
 773         snprintf(gl->gl_strname, GDLM_STRNAME_BYTES, "%8x%16llx", name.ln_type, (unsigned long long)number);
 774         memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
 775         gl->gl_lksb.sb_lvbptr = gl->gl_lvb;
 776         gl->gl_tchange = jiffies;
 777         gl->gl_object = NULL;
 778         gl->gl_sbd = sdp;
 779         INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
 780         INIT_WORK(&gl->gl_delete, delete_work_func);
 781
 782         mapping = gfs2_glock2aspace(gl);
 783         if (mapping) {
 784                 mapping->a_ops = &gfs2_meta_aops;
 785                 mapping->host = s->s_bdev->bd_inode;
 786                 mapping->flags = 0;
 787                 mapping_set_gfp_mask(mapping, GFP_NOFS);
 788                 mapping->assoc_mapping = NULL;
 789                 mapping->backing_dev_info = s->s_bdi;
 790                 mapping->writeback_index = 0;
 791         }
 792
 793         write_lock(gl_lock_addr(hash));
 794         tmp = search_bucket(hash, sdp, &name);
 795         if (tmp) {
 796                 write_unlock(gl_lock_addr(hash));
 797                 glock_free(gl);
 798                 gl = tmp;
 799         } else {
 800                 hlist_add_head(&gl->gl_list, &gl_hash_table[hash].hb_list);
 801                 write_unlock(gl_lock_addr(hash));
 802         }
 803
 804         *glp = gl;
 805
 806         return 0;
 807 }
 808
 809 /**
 810  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
 811  * @gl: the glock
 812  * @state: the state we're requesting
 813  * @flags: the modifier flags
 814  * @gh: the holder structure
 815  *
 816  */
 817
 818 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
 819                       struct gfs2_holder *gh)
 820 {
 821         INIT_LIST_HEAD(&gh->gh_list);
 822         gh->gh_gl = gl;
 823         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 824         gh->gh_owner_pid = get_pid(task_pid(current));
 825         gh->gh_state = state;
 826         gh->gh_flags = flags;
 827         gh->gh_error = 0;
 828         gh->gh_iflags = 0;
 829         gfs2_glock_hold(gl);
 830 }
 831
 832 /**
 833  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
 834  * @state: the state we're requesting
 835  * @flags: the modifier flags
 836  * @gh: the holder structure
 837  *
 838  * Don't mess with the glock.
 839  *
 840  */
 841
 842 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
 843 {
 844         gh->gh_state = state;
 845         gh->gh_flags = flags;
 846         gh->gh_iflags = 0;
 847         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 848         if (gh->gh_owner_pid)
 849                 put_pid(gh->gh_owner_pid);
 850         gh->gh_owner_pid = get_pid(task_pid(current));
 851 }
 852
 853 /**
 854  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
 855  * @gh: the holder structure
 856  *
 857  */
 858
 859 void gfs2_holder_uninit(struct gfs2_holder *gh)
 860 {
 861         put_pid(gh->gh_owner_pid);
 862         gfs2_glock_put(gh->gh_gl);
 863         gh->gh_gl = NULL;
 864         gh->gh_ip = 0;
 865 }
 866
 867 /**
 868  * gfs2_glock_holder_wait
 869  * @word: unused
 870  *
 871  * This function and gfs2_glock_demote_wait both show up in the WCHAN
 872  * field. Thus I've separated these otherwise identical functions in
 873  * order to be more informative to the user.
 874  */
 875
 876 static int gfs2_glock_holder_wait(void *word)
 877 {
 878         schedule();
 879         return 0;
 880 }
 881
 882 static int gfs2_glock_demote_wait(void *word)
 883 {
 884         schedule();
 885         return 0;
 886 }
 887
 888 static void wait_on_holder(struct gfs2_holder *gh)
 889 {
 890         might_sleep();
 891         wait_on_bit(&gh->gh_iflags, HIF_WAIT, gfs2_glock_holder_wait, TASK_UNINTERRUPTIBLE);
 892 }
 893
 894 static void wait_on_demote(struct gfs2_glock *gl)
 895 {
 896         might_sleep();
 897         wait_on_bit(&gl->gl_flags, GLF_DEMOTE, gfs2_glock_demote_wait, TASK_UNINTERRUPTIBLE);
 898 }
 899
 900 /**
 901  * handle_callback - process a demote request
 902  * @gl: the glock
 903  * @state: the state the caller wants us to change to
 904  *
 905  * There are only two requests that we are going to see in actual
 906  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
 907  */
 908
 909 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
 910                             unsigned long delay)
 911 {
 912         int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
 913
 914         set_bit(bit, &gl->gl_flags);
 915         if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
 916                 gl->gl_demote_state = state;
 917                 gl->gl_demote_time = jiffies;
 918         } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
 919                         gl->gl_demote_state != state) {
 920                 gl->gl_demote_state = LM_ST_UNLOCKED;
 921         }
 922         if (gl->gl_ops->go_callback)
 923                 gl->gl_ops->go_callback(gl);
 924         trace_gfs2_demote_rq(gl);
 925 }
 926
 927 /**
 928  * gfs2_glock_wait - wait on a glock acquisition
 929  * @gh: the glock holder
 930  *
 931  * Returns: 0 on success
 932  */
 933
 934 int gfs2_glock_wait(struct gfs2_holder *gh)
 935 {
 936         wait_on_holder(gh);
 937         return gh->gh_error;
 938 }
 939
 940 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
 941 {
 942         struct va_format vaf;
 943         va_list args;
 944
 945         va_start(args, fmt);
 946
 947         if (seq) {
 948                 struct gfs2_glock_iter *gi = seq->private;
 949                 vsprintf(gi->string, fmt, args);
 950                 seq_printf(seq, gi->string);
 951         } else {
 952                 vaf.fmt = fmt;
 953                 vaf.va = &args;
 954
 955                 printk(KERN_ERR " %pV", &vaf);
 956         }
 957
 958         va_end(args);
 959 }
 960
 961 /**
 962  * add_to_queue - Add a holder to the wait queue (but look for recursion)
 963  * @gh: the holder structure to add
 964  *
 965  * Eventually we should move the recursive locking trap to a
 966  * debugging option or something like that. This is the fast
 967  * path and needs to have the minimum number of distractions.
 968  *
 969  */
 970
 971 static inline void add_to_queue(struct gfs2_holder *gh)
 972 __releases(&gl->gl_spin)
 973 __acquires(&gl->gl_spin)
 974 {
 975         struct gfs2_glock *gl = gh->gh_gl;
 976         struct gfs2_sbd *sdp = gl->gl_sbd;
 977         struct list_head *insert_pt = NULL;
 978         struct gfs2_holder *gh2;
 979         int try_lock = 0;
 980
 981         BUG_ON(gh->gh_owner_pid == NULL);
 982         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
 983                 BUG();
 984
 985         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
 986                 if (test_bit(GLF_LOCK, &gl->gl_flags))
 987                         try_lock = 1;
 988                 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 989                         goto fail;
 990         }
 991
 992         list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
 993                 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
 994                     (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
 995                         goto trap_recursive;
 996                 if (try_lock &&
 997                     !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) &&
 998                     !may_grant(gl, gh)) {
 999 fail:
1000                         gh->gh_error = GLR_TRYFAILED;
1001                         gfs2_holder_wake(gh);
1002                         return;
1003                 }
1004                 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1005                         continue;
1006                 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1007                         insert_pt = &gh2->gh_list;
1008         }
1009         set_bit(GLF_QUEUED, &gl->gl_flags);
1010         if (likely(insert_pt == NULL)) {
1011                 list_add_tail(&gh->gh_list, &gl->gl_holders);
1012                 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1013                         goto do_cancel;
1014                 return;
1015         }
1016         trace_gfs2_glock_queue(gh, 1);
1017         list_add_tail(&gh->gh_list, insert_pt);
1018 do_cancel:
1019         gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
1020         if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1021                 spin_unlock(&gl->gl_spin);
1022                 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1023                         sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1024                 spin_lock(&gl->gl_spin);
1025         }
1026         return;
1027
1028 trap_recursive:
1029         print_symbol(KERN_ERR "original: %s\n", gh2->gh_ip);
1030         printk(KERN_ERR "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1031         printk(KERN_ERR "lock type: %d req lock state : %d\n",
1032                gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1033         print_symbol(KERN_ERR "new: %s\n", gh->gh_ip);
1034         printk(KERN_ERR "pid: %d\n", pid_nr(gh->gh_owner_pid));
1035         printk(KERN_ERR "lock type: %d req lock state : %d\n",
1036                gh->gh_gl->gl_name.ln_type, gh->gh_state);
1037         __dump_glock(NULL, gl);
1038         BUG();
1039 }
1040
1041 /**
1042  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1043  * @gh: the holder structure
1044  *
1045  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1046  *
1047  * Returns: 0, GLR_TRYFAILED, or errno on failure
1048  */
1049
1050 int gfs2_glock_nq(struct gfs2_holder *gh)
1051 {
1052         struct gfs2_glock *gl = gh->gh_gl;
1053         struct gfs2_sbd *sdp = gl->gl_sbd;
1054         int error = 0;
1055
1056         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1057                 return -EIO;
1058
1059         spin_lock(&gl->gl_spin);
1060         add_to_queue(gh);
1061         if ((LM_FLAG_NOEXP & gh->gh_flags) &&
1062             test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1063                 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1064         run_queue(gl, 1);
1065         spin_unlock(&gl->gl_spin);
1066
1067         if (!(gh->gh_flags & GL_ASYNC))
1068                 error = gfs2_glock_wait(gh);
1069
1070         return error;
1071 }
1072
1073 /**
1074  * gfs2_glock_poll - poll to see if an async request has been completed
1075  * @gh: the holder
1076  *
1077  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1078  */
1079
1080 int gfs2_glock_poll(struct gfs2_holder *gh)
1081 {
1082         return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1083 }
1084
1085 /**
1086  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1087  * @gh: the glock holder
1088  *
1089  */
1090
1091 void gfs2_glock_dq(struct gfs2_holder *gh)
1092 {
1093         struct gfs2_glock *gl = gh->gh_gl;
1094         const struct gfs2_glock_operations *glops = gl->gl_ops;
1095         unsigned delay = 0;
1096         int fast_path = 0;
1097
1098         spin_lock(&gl->gl_spin);
1099         if (gh->gh_flags & GL_NOCACHE)
1100                 handle_callback(gl, LM_ST_UNLOCKED, 0);
1101
1102         list_del_init(&gh->gh_list);
1103         if (find_first_holder(gl) == NULL) {
1104                 if (glops->go_unlock) {
1105                         GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1106                         spin_unlock(&gl->gl_spin);
1107                         glops->go_unlock(gh);
1108                         spin_lock(&gl->gl_spin);
1109                         clear_bit(GLF_LOCK, &gl->gl_flags);
1110                 }
1111                 if (list_empty(&gl->gl_holders) &&
1112                     !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1113                     !test_bit(GLF_DEMOTE, &gl->gl_flags))
1114                         fast_path = 1;
1115         }
1116         trace_gfs2_glock_queue(gh, 0);
1117         spin_unlock(&gl->gl_spin);
1118         if (likely(fast_path))
1119                 return;
1120
1121         gfs2_glock_hold(gl);
1122         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1123             !test_bit(GLF_DEMOTE, &gl->gl_flags))
1124                 delay = gl->gl_ops->go_min_hold_time;
1125         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1126                 gfs2_glock_put(gl);
1127 }
1128
1129 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1130 {
1131         struct gfs2_glock *gl = gh->gh_gl;
1132         gfs2_glock_dq(gh);
1133         wait_on_demote(gl);
1134 }
1135
1136 /**
1137  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1138  * @gh: the holder structure
1139  *
1140  */
1141
1142 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1143 {
1144         gfs2_glock_dq(gh);
1145         gfs2_holder_uninit(gh);
1146 }
1147
1148 /**
1149  * gfs2_glock_nq_num - acquire a glock based on lock number
1150  * @sdp: the filesystem
1151  * @number: the lock number
1152  * @glops: the glock operations for the type of glock
1153  * @state: the state to acquire the glock in
1154  * @flags: modifier flags for the aquisition
1155  * @gh: the struct gfs2_holder
1156  *
1157  * Returns: errno
1158  */
1159
1160 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1161                       const struct gfs2_glock_operations *glops,
1162                       unsigned int state, int flags, struct gfs2_holder *gh)
1163 {
1164         struct gfs2_glock *gl;
1165         int error;
1166
1167         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1168         if (!error) {
1169                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1170                 gfs2_glock_put(gl);
1171         }
1172
1173         return error;
1174 }
1175
1176 /**
1177  * glock_compare - Compare two struct gfs2_glock structures for sorting
1178  * @arg_a: the first structure
1179  * @arg_b: the second structure
1180  *
1181  */
1182
1183 static int glock_compare(const void *arg_a, const void *arg_b)
1184 {
1185         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1186         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1187         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1188         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1189
1190         if (a->ln_number > b->ln_number)
1191                 return 1;
1192         if (a->ln_number < b->ln_number)
1193                 return -1;
1194         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1195         return 0;
1196 }
1197
1198 /**
1199  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1200  * @num_gh: the number of structures
1201  * @ghs: an array of struct gfs2_holder structures
1202  *
1203  * Returns: 0 on success (all glocks acquired),
1204  *          errno on failure (no glocks acquired)
1205  */
1206
1207 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1208                      struct gfs2_holder **p)
1209 {
1210         unsigned int x;
1211         int error = 0;
1212
1213         for (x = 0; x < num_gh; x++)
1214                 p[x] = &ghs[x];
1215
1216         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1217
1218         for (x = 0; x < num_gh; x++) {
1219                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1220
1221                 error = gfs2_glock_nq(p[x]);
1222                 if (error) {
1223                         while (x--)
1224                                 gfs2_glock_dq(p[x]);
1225                         break;
1226                 }
1227         }
1228
1229         return error;
1230 }
1231
1232 /**
1233  * gfs2_glock_nq_m - acquire multiple glocks
1234  * @num_gh: the number of structures
1235  * @ghs: an array of struct gfs2_holder structures
1236  *
1237  *
1238  * Returns: 0 on success (all glocks acquired),
1239  *          errno on failure (no glocks acquired)
1240  */
1241
1242 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1243 {
1244         struct gfs2_holder *tmp[4];
1245         struct gfs2_holder **pph = tmp;
1246         int error = 0;
1247
1248         switch(num_gh) {
1249         case 0:
1250                 return 0;
1251         case 1:
1252                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1253                 return gfs2_glock_nq(ghs);
1254         default:
1255                 if (num_gh <= 4)
1256                         break;
1257                 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1258                 if (!pph)
1259                         return -ENOMEM;
1260         }
1261
1262         error = nq_m_sync(num_gh, ghs, pph);
1263
1264         if (pph != tmp)
1265                 kfree(pph);
1266
1267         return error;
1268 }
1269
1270 /**
1271  * gfs2_glock_dq_m - release multiple glocks
1272  * @num_gh: the number of structures
1273  * @ghs: an array of struct gfs2_holder structures
1274  *
1275  */
1276
1277 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1278 {
1279         unsigned int x;
1280
1281         for (x = 0; x < num_gh; x++)
1282                 gfs2_glock_dq(&ghs[x]);
1283 }
1284
1285 /**
1286  * gfs2_glock_dq_uninit_m - release multiple glocks
1287  * @num_gh: the number of structures
1288  * @ghs: an array of struct gfs2_holder structures
1289  *
1290  */
1291
1292 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1293 {
1294         unsigned int x;
1295
1296         for (x = 0; x < num_gh; x++)
1297                 gfs2_glock_dq_uninit(&ghs[x]);
1298 }
1299
1300 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1301 {
1302         unsigned long delay = 0;
1303         unsigned long holdtime;
1304         unsigned long now = jiffies;
1305
1306         gfs2_glock_hold(gl);
1307         holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
1308         if (test_bit(GLF_QUEUED, &gl->gl_flags)) {
1309                 if (time_before(now, holdtime))
1310                         delay = holdtime - now;
1311                 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1312                         delay = gl->gl_ops->go_min_hold_time;
1313         }
1314
1315         spin_lock(&gl->gl_spin);
1316         handle_callback(gl, state, delay);
1317         spin_unlock(&gl->gl_spin);
1318         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1319                 gfs2_glock_put(gl);
1320 }
1321
1322 /**
1323  * gfs2_should_freeze - Figure out if glock should be frozen
1324  * @gl: The glock in question
1325  *
1326  * Glocks are not frozen if (a) the result of the dlm operation is
1327  * an error, (b) the locking operation was an unlock operation or
1328  * (c) if there is a "noexp" flagged request anywhere in the queue
1329  *
1330  * Returns: 1 if freezing should occur, 0 otherwise
1331  */
1332
1333 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1334 {
1335         const struct gfs2_holder *gh;
1336
1337         if (gl->gl_reply & ~LM_OUT_ST_MASK)
1338                 return 0;
1339         if (gl->gl_target == LM_ST_UNLOCKED)
1340                 return 0;
1341
1342         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1343                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1344                         continue;
1345                 if (LM_FLAG_NOEXP & gh->gh_flags)
1346                         return 0;
1347         }
1348
1349         return 1;
1350 }
1351
1352 /**
1353  * gfs2_glock_complete - Callback used by locking
1354  * @gl: Pointer to the glock
1355  * @ret: The return value from the dlm
1356  *
1357  * The gl_reply field is under the gl_spin lock so that it is ok
1358  * to use a bitfield shared with other glock state fields.
1359  */
1360
1361 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1362 {
1363         struct lm_lockstruct *ls = &gl->gl_sbd->sd_lockstruct;
1364
1365         spin_lock(&gl->gl_spin);
1366         gl->gl_reply = ret;
1367
1368         if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_flags))) {
1369                 if (gfs2_should_freeze(gl)) {
1370                         set_bit(GLF_FROZEN, &gl->gl_flags);
1371                         spin_unlock(&gl->gl_spin);
1372                         return;
1373                 }
1374         }
1375
1376         spin_unlock(&gl->gl_spin);
1377         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1378         smp_wmb();
1379         gfs2_glock_hold(gl);
1380         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1381                 gfs2_glock_put(gl);
1382 }
1383
1384
1385 static int gfs2_shrink_glock_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
1386 {
1387         struct gfs2_glock *gl;
1388         int may_demote;
1389         int nr_skipped = 0;
1390         LIST_HEAD(skipped);
1391
1392         if (nr == 0)
1393                 goto out;
1394
1395         if (!(gfp_mask & __GFP_FS))
1396                 return -1;
1397
1398         spin_lock(&lru_lock);
1399         while(nr && !list_empty(&lru_list)) {
1400                 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1401                 list_del_init(&gl->gl_lru);
1402                 atomic_dec(&lru_count);
1403
1404                 /* Test for being demotable */
1405                 if (!test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1406                         gfs2_glock_hold(gl);
1407                         spin_unlock(&lru_lock);
1408                         spin_lock(&gl->gl_spin);
1409                         may_demote = demote_ok(gl);
1410                         if (may_demote) {
1411                                 handle_callback(gl, LM_ST_UNLOCKED, 0);
1412                                 nr--;
1413                         }
1414                         clear_bit(GLF_LOCK, &gl->gl_flags);
1415                         smp_mb__after_clear_bit();
1416                         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1417                                 gfs2_glock_put_nolock(gl);
1418                         spin_unlock(&gl->gl_spin);
1419                         spin_lock(&lru_lock);
1420                         continue;
1421                 }
1422                 nr_skipped++;
1423                 list_add(&gl->gl_lru, &skipped);
1424         }
1425         list_splice(&skipped, &lru_list);
1426         atomic_add(nr_skipped, &lru_count);
1427         spin_unlock(&lru_lock);
1428 out:
1429         return (atomic_read(&lru_count) / 100) * sysctl_vfs_cache_pressure;
1430 }
1431
1432 static struct shrinker glock_shrinker = {
1433         .shrink = gfs2_shrink_glock_memory,
1434         .seeks = DEFAULT_SEEKS,
1435 };
1436
1437 /**
1438  * examine_bucket - Call a function for glock in a hash bucket
1439  * @examiner: the function
1440  * @sdp: the filesystem
1441  * @bucket: the bucket
1442  *
1443  * Returns: 1 if the bucket has entries
1444  */
1445
1446 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1447                           unsigned int hash)
1448 {
1449         struct gfs2_glock *gl, *prev = NULL;
1450         int has_entries = 0;
1451         struct hlist_head *head = &gl_hash_table[hash].hb_list;
1452
1453         read_lock(gl_lock_addr(hash));
1454         /* Can't use hlist_for_each_entry - don't want prefetch here */
1455         if (hlist_empty(head))
1456                 goto out;
1457         gl = list_entry(head->first, struct gfs2_glock, gl_list);
1458         while(1) {
1459                 if (!sdp || gl->gl_sbd == sdp) {
1460                         gfs2_glock_hold(gl);
1461                         read_unlock(gl_lock_addr(hash));
1462                         if (prev)
1463                                 gfs2_glock_put(prev);
1464                         prev = gl;
1465                         examiner(gl);
1466                         has_entries = 1;
1467                         read_lock(gl_lock_addr(hash));
1468                 }
1469                 if (gl->gl_list.next == NULL)
1470                         break;
1471                 gl = list_entry(gl->gl_list.next, struct gfs2_glock, gl_list);
1472         }
1473 out:
1474         read_unlock(gl_lock_addr(hash));
1475         if (prev)
1476                 gfs2_glock_put(prev);
1477         cond_resched();
1478         return has_entries;
1479 }
1480
1481
1482 /**
1483  * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1484  * @gl: The glock to thaw
1485  *
1486  * N.B. When we freeze a glock, we leave a ref to the glock outstanding,
1487  * so this has to result in the ref count being dropped by one.
1488  */
1489
1490 static void thaw_glock(struct gfs2_glock *gl)
1491 {
1492         if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1493                 return;
1494         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1495         gfs2_glock_hold(gl);
1496         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1497                 gfs2_glock_put(gl);
1498 }
1499
1500 /**
1501  * clear_glock - look at a glock and see if we can free it from glock cache
1502  * @gl: the glock to look at
1503  *
1504  */
1505
1506 static void clear_glock(struct gfs2_glock *gl)
1507 {
1508         spin_lock(&lru_lock);
1509         if (!list_empty(&gl->gl_lru)) {
1510                 list_del_init(&gl->gl_lru);
1511                 atomic_dec(&lru_count);
1512         }
1513         spin_unlock(&lru_lock);
1514
1515         spin_lock(&gl->gl_spin);
1516         if (gl->gl_state != LM_ST_UNLOCKED)
1517                 handle_callback(gl, LM_ST_UNLOCKED, 0);
1518         spin_unlock(&gl->gl_spin);
1519         gfs2_glock_hold(gl);
1520         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1521                 gfs2_glock_put(gl);
1522 }
1523
1524 /**
1525  * gfs2_glock_thaw - Thaw any frozen glocks
1526  * @sdp: The super block
1527  *
1528  */
1529
1530 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1531 {
1532         unsigned x;
1533
1534         for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1535                 examine_bucket(thaw_glock, sdp, x);
1536 }
1537
1538 /**
1539  * gfs2_gl_hash_clear - Empty out the glock hash table
1540  * @sdp: the filesystem
1541  * @wait: wait until it's all gone
1542  *
1543  * Called when unmounting the filesystem.
1544  */
1545
1546 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1547 {
1548         unsigned int x;
1549
1550         for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1551                 examine_bucket(clear_glock, sdp, x);
1552         flush_workqueue(glock_workqueue);
1553         wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
1554         gfs2_dump_lockstate(sdp);
1555 }
1556
1557 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1558 {
1559         struct gfs2_glock *gl = ip->i_gl;
1560         int ret;
1561
1562         ret = gfs2_truncatei_resume(ip);
1563         gfs2_assert_withdraw(gl->gl_sbd, ret == 0);
1564
1565         spin_lock(&gl->gl_spin);
1566         clear_bit(GLF_LOCK, &gl->gl_flags);
1567         run_queue(gl, 1);
1568         spin_unlock(&gl->gl_spin);
1569 }
1570
1571 static const char *state2str(unsigned state)
1572 {
1573         switch(state) {
1574         case LM_ST_UNLOCKED:
1575                 return "UN";
1576         case LM_ST_SHARED:
1577                 return "SH";
1578         case LM_ST_DEFERRED:
1579                 return "DF";
1580         case LM_ST_EXCLUSIVE:
1581                 return "EX";
1582         }
1583         return "??";
1584 }
1585
1586 static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1587 {
1588         char *p = buf;
1589         if (flags & LM_FLAG_TRY)
1590                 *p++ = 't';
1591         if (flags & LM_FLAG_TRY_1CB)
1592                 *p++ = 'T';
1593         if (flags & LM_FLAG_NOEXP)
1594                 *p++ = 'e';
1595         if (flags & LM_FLAG_ANY)
1596                 *p++ = 'A';
1597         if (flags & LM_FLAG_PRIORITY)
1598                 *p++ = 'p';
1599         if (flags & GL_ASYNC)
1600                 *p++ = 'a';
1601         if (flags & GL_EXACT)
1602                 *p++ = 'E';
1603         if (flags & GL_NOCACHE)
1604                 *p++ = 'c';
1605         if (test_bit(HIF_HOLDER, &iflags))
1606                 *p++ = 'H';
1607         if (test_bit(HIF_WAIT, &iflags))
1608                 *p++ = 'W';
1609         if (test_bit(HIF_FIRST, &iflags))
1610                 *p++ = 'F';
1611         *p = 0;
1612         return buf;
1613 }
1614
1615 /**
1616  * dump_holder - print information about a glock holder
1617  * @seq: the seq_file struct
1618  * @gh: the glock holder
1619  *
1620  * Returns: 0 on success, -ENOBUFS when we run out of space
1621  */
1622
1623 static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1624 {
1625         struct task_struct *gh_owner = NULL;
1626         char flags_buf[32];
1627
1628         if (gh->gh_owner_pid)
1629                 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1630         gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1631                        state2str(gh->gh_state),
1632                        hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1633                        gh->gh_error,
1634                        gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1635                        gh_owner ? gh_owner->comm : "(ended)",
1636                        (void *)gh->gh_ip);
1637         return 0;
1638 }
1639
1640 static const char *gflags2str(char *buf, const unsigned long *gflags)
1641 {
1642         char *p = buf;
1643         if (test_bit(GLF_LOCK, gflags))
1644                 *p++ = 'l';
1645         if (test_bit(GLF_DEMOTE, gflags))
1646                 *p++ = 'D';
1647         if (test_bit(GLF_PENDING_DEMOTE, gflags))
1648                 *p++ = 'd';
1649         if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1650                 *p++ = 'p';
1651         if (test_bit(GLF_DIRTY, gflags))
1652                 *p++ = 'y';
1653         if (test_bit(GLF_LFLUSH, gflags))
1654                 *p++ = 'f';
1655         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1656                 *p++ = 'i';
1657         if (test_bit(GLF_REPLY_PENDING, gflags))
1658                 *p++ = 'r';
1659         if (test_bit(GLF_INITIAL, gflags))
1660                 *p++ = 'I';
1661         if (test_bit(GLF_FROZEN, gflags))
1662                 *p++ = 'F';
1663         if (test_bit(GLF_QUEUED, gflags))
1664                 *p++ = 'q';
1665         *p = 0;
1666         return buf;
1667 }
1668
1669 /**
1670  * __dump_glock - print information about a glock
1671  * @seq: The seq_file struct
1672  * @gl: the glock
1673  *
1674  * The file format is as follows:
1675  * One line per object, capital letters are used to indicate objects
1676  * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1677  * other objects are indented by a single space and follow the glock to
1678  * which they are related. Fields are indicated by lower case letters
1679  * followed by a colon and the field value, except for strings which are in
1680  * [] so that its possible to see if they are composed of spaces for
1681  * example. The field's are n = number (id of the object), f = flags,
1682  * t = type, s = state, r = refcount, e = error, p = pid.
1683  *
1684  * Returns: 0 on success, -ENOBUFS when we run out of space
1685  */
1686
1687 static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1688 {
1689         const struct gfs2_glock_operations *glops = gl->gl_ops;
1690         unsigned long long dtime;
1691         const struct gfs2_holder *gh;
1692         char gflags_buf[32];
1693         int error = 0;
1694
1695         dtime = jiffies - gl->gl_demote_time;
1696         dtime *= 1000000/HZ; /* demote time in uSec */
1697         if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1698                 dtime = 0;
1699         gfs2_print_dbg(seq, "G:  s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d r:%d\n",
1700                   state2str(gl->gl_state),
1701                   gl->gl_name.ln_type,
1702                   (unsigned long long)gl->gl_name.ln_number,
1703                   gflags2str(gflags_buf, &gl->gl_flags),
1704                   state2str(gl->gl_target),
1705                   state2str(gl->gl_demote_state), dtime,
1706                   atomic_read(&gl->gl_ail_count),
1707                   atomic_read(&gl->gl_ref));
1708
1709         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1710                 error = dump_holder(seq, gh);
1711                 if (error)
1712                         goto out;
1713         }
1714         if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1715                 error = glops->go_dump(seq, gl);
1716 out:
1717         return error;
1718 }
1719
1720 static int dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1721 {
1722         int ret;
1723         spin_lock(&gl->gl_spin);
1724         ret = __dump_glock(seq, gl);
1725         spin_unlock(&gl->gl_spin);
1726         return ret;
1727 }
1728
1729 /**
1730  * gfs2_dump_lockstate - print out the current lockstate
1731  * @sdp: the filesystem
1732  * @ub: the buffer to copy the information into
1733  *
1734  * If @ub is NULL, dump the lockstate to the console.
1735  *
1736  */
1737
1738 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
1739 {
1740         struct gfs2_glock *gl;
1741         struct hlist_node *h;
1742         unsigned int x;
1743         int error = 0;
1744
1745         for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1746
1747                 read_lock(gl_lock_addr(x));
1748
1749                 hlist_for_each_entry(gl, h, &gl_hash_table[x].hb_list, gl_list) {
1750                         if (gl->gl_sbd != sdp)
1751                                 continue;
1752
1753                         error = dump_glock(NULL, gl);
1754                         if (error)
1755                                 break;
1756                 }
1757
1758                 read_unlock(gl_lock_addr(x));
1759
1760                 if (error)
1761                         break;
1762         }
1763
1764
1765         return error;
1766 }
1767
1768
1769 int __init gfs2_glock_init(void)
1770 {
1771         unsigned i;
1772         for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
1773                 INIT_HLIST_HEAD(&gl_hash_table[i].hb_list);
1774         }
1775 #ifdef GL_HASH_LOCK_SZ
1776         for(i = 0; i < GL_HASH_LOCK_SZ; i++) {
1777                 rwlock_init(&gl_hash_locks[i]);
1778         }
1779 #endif
1780
1781         glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1782                                           WQ_HIGHPRI | WQ_FREEZEABLE, 0);
1783         if (IS_ERR(glock_workqueue))
1784                 return PTR_ERR(glock_workqueue);
1785         gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1786                                                 WQ_MEM_RECLAIM | WQ_FREEZEABLE,
1787                                                 0);
1788         if (IS_ERR(gfs2_delete_workqueue)) {
1789                 destroy_workqueue(glock_workqueue);
1790                 return PTR_ERR(gfs2_delete_workqueue);
1791         }
1792
1793         register_shrinker(&glock_shrinker);
1794
1795         return 0;
1796 }
1797
1798 void gfs2_glock_exit(void)
1799 {
1800         unregister_shrinker(&glock_shrinker);
1801         destroy_workqueue(glock_workqueue);
1802         destroy_workqueue(gfs2_delete_workqueue);
1803 }
1804
1805 static int gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1806 {
1807         struct gfs2_glock *gl;
1808
1809 restart:
1810         read_lock(gl_lock_addr(gi->hash));
1811         gl = gi->gl;
1812         if (gl) {
1813                 gi->gl = hlist_entry(gl->gl_list.next,
1814                                      struct gfs2_glock, gl_list);
1815         } else {
1816                 gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
1817                                      struct gfs2_glock, gl_list);
1818         }
1819         if (gi->gl)
1820                 gfs2_glock_hold(gi->gl);
1821         read_unlock(gl_lock_addr(gi->hash));
1822         if (gl)
1823                 gfs2_glock_put(gl);
1824         while (gi->gl == NULL) {
1825                 gi->hash++;
1826                 if (gi->hash >= GFS2_GL_HASH_SIZE)
1827                         return 1;
1828                 read_lock(gl_lock_addr(gi->hash));
1829                 gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
1830                                      struct gfs2_glock, gl_list);
1831                 if (gi->gl)
1832                         gfs2_glock_hold(gi->gl);
1833                 read_unlock(gl_lock_addr(gi->hash));
1834         }
1835
1836         if (gi->sdp != gi->gl->gl_sbd)
1837                 goto restart;
1838
1839         return 0;
1840 }
1841
1842 static void gfs2_glock_iter_free(struct gfs2_glock_iter *gi)
1843 {
1844         if (gi->gl)
1845                 gfs2_glock_put(gi->gl);
1846         gi->gl = NULL;
1847 }
1848
1849 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1850 {
1851         struct gfs2_glock_iter *gi = seq->private;
1852         loff_t n = *pos;
1853
1854         gi->hash = 0;
1855
1856         do {
1857                 if (gfs2_glock_iter_next(gi)) {
1858                         gfs2_glock_iter_free(gi);
1859                         return NULL;
1860                 }
1861         } while (n--);
1862
1863         return gi->gl;
1864 }
1865
1866 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1867                                  loff_t *pos)
1868 {
1869         struct gfs2_glock_iter *gi = seq->private;
1870
1871         (*pos)++;
1872
1873         if (gfs2_glock_iter_next(gi)) {
1874                 gfs2_glock_iter_free(gi);
1875                 return NULL;
1876         }
1877
1878         return gi->gl;
1879 }
1880
1881 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1882 {
1883         struct gfs2_glock_iter *gi = seq->private;
1884         gfs2_glock_iter_free(gi);
1885 }
1886
1887 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1888 {
1889         return dump_glock(seq, iter_ptr);
1890 }
1891
1892 static const struct seq_operations gfs2_glock_seq_ops = {
1893         .start = gfs2_glock_seq_start,
1894         .next  = gfs2_glock_seq_next,
1895         .stop  = gfs2_glock_seq_stop,
1896         .show  = gfs2_glock_seq_show,
1897 };
1898
1899 static int gfs2_debugfs_open(struct inode *inode, struct file *file)
1900 {
1901         int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1902                                    sizeof(struct gfs2_glock_iter));
1903         if (ret == 0) {
1904                 struct seq_file *seq = file->private_data;
1905                 struct gfs2_glock_iter *gi = seq->private;
1906                 gi->sdp = inode->i_private;
1907         }
1908         return ret;
1909 }
1910
1911 static const struct file_operations gfs2_debug_fops = {
1912         .owner   = THIS_MODULE,
1913         .open    = gfs2_debugfs_open,
1914         .read    = seq_read,
1915         .llseek  = seq_lseek,
1916         .release = seq_release_private,
1917 };
1918
1919 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
1920 {
1921         sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
1922         if (!sdp->debugfs_dir)
1923                 return -ENOMEM;
1924         sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
1925                                                          S_IFREG | S_IRUGO,
1926                                                          sdp->debugfs_dir, sdp,
1927                                                          &gfs2_debug_fops);
1928         if (!sdp->debugfs_dentry_glocks)
1929                 return -ENOMEM;
1930
1931         return 0;
1932 }
1933
1934 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
1935 {
1936         if (sdp && sdp->debugfs_dir) {
1937                 if (sdp->debugfs_dentry_glocks) {
1938                         debugfs_remove(sdp->debugfs_dentry_glocks);
1939                         sdp->debugfs_dentry_glocks = NULL;
1940                 }
1941                 debugfs_remove(sdp->debugfs_dir);
1942                 sdp->debugfs_dir = NULL;
1943         }
1944 }
1945
1946 int gfs2_register_debugfs(void)
1947 {
1948         gfs2_root = debugfs_create_dir("gfs2", NULL);
1949         return gfs2_root ? 0 : -ENOMEM;
1950 }
1951
1952 void gfs2_unregister_debugfs(void)
1953 {
1954         debugfs_remove(gfs2_root);
1955         gfs2_root = NULL;
1956 }