fs/gfs2/glock.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2006 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/completion.h>
  14 #include <linux/buffer_head.h>
  15 #include <linux/delay.h>
  16 #include <linux/sort.h>
  17 #include <linux/jhash.h>
  18 #include <linux/kallsyms.h>
  19 #include <linux/gfs2_ondisk.h>
  20 #include <linux/list.h>
  21 #include <linux/lm_interface.h>
  22 #include <linux/wait.h>
  23 #include <linux/module.h>
  24 #include <linux/rwsem.h>
  25 #include <asm/uaccess.h>
  26 #include <linux/seq_file.h>
  27 #include <linux/debugfs.h>
  28 #include <linux/module.h>
  29 #include <linux/kallsyms.h>
  30
  31 #include "gfs2.h"
  32 #include "incore.h"
  33 #include "glock.h"
  34 #include "glops.h"
  35 #include "inode.h"
  36 #include "lm.h"
  37 #include "lops.h"
  38 #include "meta_io.h"
  39 #include "quota.h"
  40 #include "super.h"
  41 #include "util.h"
  42
  43 struct gfs2_gl_hash_bucket {
  44         struct hlist_head hb_list;
  45 };
  46
  47 struct glock_iter {
  48         int hash;                     /* hash bucket index         */
  49         struct gfs2_sbd *sdp;         /* incore superblock         */
  50         struct gfs2_glock *gl;        /* current glock struct      */
  51         struct hlist_head *hb_list;   /* current hash bucket ptr   */
  52         struct seq_file *seq;         /* sequence file for debugfs */
  53         char string[512];             /* scratch space             */
  54 };
  55
  56 typedef void (*glock_examiner) (struct gfs2_glock * gl);
  57
  58 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp);
  59 static int dump_glock(struct glock_iter *gi, struct gfs2_glock *gl);
  60 static void gfs2_glock_xmote_th(struct gfs2_glock *gl, struct gfs2_holder *gh);
  61 static void gfs2_glock_drop_th(struct gfs2_glock *gl);
  62 static DECLARE_RWSEM(gfs2_umount_flush_sem);
  63 static struct dentry *gfs2_root;
  64
  65 #define GFS2_GL_HASH_SHIFT      15
  66 #define GFS2_GL_HASH_SIZE       (1 << GFS2_GL_HASH_SHIFT)
  67 #define GFS2_GL_HASH_MASK       (GFS2_GL_HASH_SIZE - 1)
  68
  69 static struct gfs2_gl_hash_bucket gl_hash_table[GFS2_GL_HASH_SIZE];
  70 static struct dentry *gfs2_root;
  71
  72 /*
  73  * Despite what you might think, the numbers below are not arbitrary :-)
  74  * They are taken from the ipv4 routing hash code, which is well tested
  75  * and thus should be nearly optimal. Later on we might tweek the numbers
  76  * but for now this should be fine.
  77  *
  78  * The reason for putting the locks in a separate array from the list heads
  79  * is that we can have fewer locks than list heads and save memory. We use
  80  * the same hash function for both, but with a different hash mask.
  81  */
  82 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
  83         defined(CONFIG_PROVE_LOCKING)
  84
  85 #ifdef CONFIG_LOCKDEP
  86 # define GL_HASH_LOCK_SZ        256
  87 #else
  88 # if NR_CPUS >= 32
  89 #  define GL_HASH_LOCK_SZ       4096
  90 # elif NR_CPUS >= 16
  91 #  define GL_HASH_LOCK_SZ       2048
  92 # elif NR_CPUS >= 8
  93 #  define GL_HASH_LOCK_SZ       1024
  94 # elif NR_CPUS >= 4
  95 #  define GL_HASH_LOCK_SZ       512
  96 # else
  97 #  define GL_HASH_LOCK_SZ       256
  98 # endif
  99 #endif
 100
 101 /* We never want more locks than chains */
 102 #if GFS2_GL_HASH_SIZE < GL_HASH_LOCK_SZ
 103 # undef GL_HASH_LOCK_SZ
 104 # define GL_HASH_LOCK_SZ GFS2_GL_HASH_SIZE
 105 #endif
 106
 107 static rwlock_t gl_hash_locks[GL_HASH_LOCK_SZ];
 108
 109 static inline rwlock_t *gl_lock_addr(unsigned int x)
 110 {
 111         return &gl_hash_locks[x & (GL_HASH_LOCK_SZ-1)];
 112 }
 113 #else /* not SMP, so no spinlocks required */
 114 static inline rwlock_t *gl_lock_addr(unsigned int x)
 115 {
 116         return NULL;
 117 }
 118 #endif
 119
 120 /**
 121  * relaxed_state_ok - is a requested lock compatible with the current lock mode?
 122  * @actual: the current state of the lock
 123  * @requested: the lock state that was requested by the caller
 124  * @flags: the modifier flags passed in by the caller
 125  *
 126  * Returns: 1 if the locks are compatible, 0 otherwise
 127  */
 128
 129 static inline int relaxed_state_ok(unsigned int actual, unsigned requested,
 130                                    int flags)
 131 {
 132         if (actual == requested)
 133                 return 1;
 134
 135         if (flags & GL_EXACT)
 136                 return 0;
 137
 138         if (actual == LM_ST_EXCLUSIVE && requested == LM_ST_SHARED)
 139                 return 1;
 140
 141         if (actual != LM_ST_UNLOCKED && (flags & LM_FLAG_ANY))
 142                 return 1;
 143
 144         return 0;
 145 }
 146
 147 /**
 148  * gl_hash() - Turn glock number into hash bucket number
 149  * @lock: The glock number
 150  *
 151  * Returns: The number of the corresponding hash bucket
 152  */
 153
 154 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
 155                             const struct lm_lockname *name)
 156 {
 157         unsigned int h;
 158
 159         h = jhash(&name->ln_number, sizeof(u64), 0);
 160         h = jhash(&name->ln_type, sizeof(unsigned int), h);
 161         h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
 162         h &= GFS2_GL_HASH_MASK;
 163
 164         return h;
 165 }
 166
 167 /**
 168  * glock_free() - Perform a few checks and then release struct gfs2_glock
 169  * @gl: The glock to release
 170  *
 171  * Also calls lock module to release its internal structure for this glock.
 172  *
 173  */
 174
 175 static void glock_free(struct gfs2_glock *gl)
 176 {
 177         struct gfs2_sbd *sdp = gl->gl_sbd;
 178         struct inode *aspace = gl->gl_aspace;
 179
 180         gfs2_lm_put_lock(sdp, gl->gl_lock);
 181
 182         if (aspace)
 183                 gfs2_aspace_put(aspace);
 184
 185         kmem_cache_free(gfs2_glock_cachep, gl);
 186 }
 187
 188 /**
 189  * gfs2_glock_hold() - increment reference count on glock
 190  * @gl: The glock to hold
 191  *
 192  */
 193
 194 void gfs2_glock_hold(struct gfs2_glock *gl)
 195 {
 196         atomic_inc(&gl->gl_ref);
 197 }
 198
 199 /**
 200  * gfs2_glock_put() - Decrement reference count on glock
 201  * @gl: The glock to put
 202  *
 203  */
 204
 205 int gfs2_glock_put(struct gfs2_glock *gl)
 206 {
 207         int rv = 0;
 208         struct gfs2_sbd *sdp = gl->gl_sbd;
 209
 210         write_lock(gl_lock_addr(gl->gl_hash));
 211         if (atomic_dec_and_test(&gl->gl_ref)) {
 212                 hlist_del(&gl->gl_list);
 213                 write_unlock(gl_lock_addr(gl->gl_hash));
 214                 BUG_ON(spin_is_locked(&gl->gl_spin));
 215                 gfs2_assert(sdp, gl->gl_state == LM_ST_UNLOCKED);
 216                 gfs2_assert(sdp, list_empty(&gl->gl_reclaim));
 217                 gfs2_assert(sdp, list_empty(&gl->gl_holders));
 218                 gfs2_assert(sdp, list_empty(&gl->gl_waiters1));
 219                 gfs2_assert(sdp, list_empty(&gl->gl_waiters3));
 220                 glock_free(gl);
 221                 rv = 1;
 222                 goto out;
 223         }
 224         write_unlock(gl_lock_addr(gl->gl_hash));
 225 out:
 226         return rv;
 227 }
 228
 229 /**
 230  * search_bucket() - Find struct gfs2_glock by lock number
 231  * @bucket: the bucket to search
 232  * @name: The lock name
 233  *
 234  * Returns: NULL, or the struct gfs2_glock with the requested number
 235  */
 236
 237 static struct gfs2_glock *search_bucket(unsigned int hash,
 238                                         const struct gfs2_sbd *sdp,
 239                                         const struct lm_lockname *name)
 240 {
 241         struct gfs2_glock *gl;
 242         struct hlist_node *h;
 243
 244         hlist_for_each_entry(gl, h, &gl_hash_table[hash].hb_list, gl_list) {
 245                 if (!lm_name_equal(&gl->gl_name, name))
 246                         continue;
 247                 if (gl->gl_sbd != sdp)
 248                         continue;
 249
 250                 atomic_inc(&gl->gl_ref);
 251
 252                 return gl;
 253         }
 254
 255         return NULL;
 256 }
 257
 258 /**
 259  * gfs2_glock_find() - Find glock by lock number
 260  * @sdp: The GFS2 superblock
 261  * @name: The lock name
 262  *
 263  * Returns: NULL, or the struct gfs2_glock with the requested number
 264  */
 265
 266 static struct gfs2_glock *gfs2_glock_find(const struct gfs2_sbd *sdp,
 267                                           const struct lm_lockname *name)
 268 {
 269         unsigned int hash = gl_hash(sdp, name);
 270         struct gfs2_glock *gl;
 271
 272         read_lock(gl_lock_addr(hash));
 273         gl = search_bucket(hash, sdp, name);
 274         read_unlock(gl_lock_addr(hash));
 275
 276         return gl;
 277 }
 278
 279 /**
 280  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
 281  * @sdp: The GFS2 superblock
 282  * @number: the lock number
 283  * @glops: The glock_operations to use
 284  * @create: If 0, don't create the glock if it doesn't exist
 285  * @glp: the glock is returned here
 286  *
 287  * This does not lock a glock, just finds/creates structures for one.
 288  *
 289  * Returns: errno
 290  */
 291
 292 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
 293                    const struct gfs2_glock_operations *glops, int create,
 294                    struct gfs2_glock **glp)
 295 {
 296         struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
 297         struct gfs2_glock *gl, *tmp;
 298         unsigned int hash = gl_hash(sdp, &name);
 299         int error;
 300
 301         read_lock(gl_lock_addr(hash));
 302         gl = search_bucket(hash, sdp, &name);
 303         read_unlock(gl_lock_addr(hash));
 304
 305         if (gl || !create) {
 306                 *glp = gl;
 307                 return 0;
 308         }
 309
 310         gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
 311         if (!gl)
 312                 return -ENOMEM;
 313
 314         gl->gl_flags = 0;
 315         gl->gl_name = name;
 316         atomic_set(&gl->gl_ref, 1);
 317         gl->gl_state = LM_ST_UNLOCKED;
 318         gl->gl_hash = hash;
 319         gl->gl_owner_pid = 0;
 320         gl->gl_ip = 0;
 321         gl->gl_ops = glops;
 322         gl->gl_req_gh = NULL;
 323         gl->gl_req_bh = NULL;
 324         gl->gl_vn = 0;
 325         gl->gl_stamp = jiffies;
 326         gl->gl_object = NULL;
 327         gl->gl_sbd = sdp;
 328         gl->gl_aspace = NULL;
 329         lops_init_le(&gl->gl_le, &gfs2_glock_lops);
 330
 331         /* If this glock protects actual on-disk data or metadata blocks,
 332            create a VFS inode to manage the pages/buffers holding them. */
 333         if (glops == &gfs2_inode_glops || glops == &gfs2_rgrp_glops) {
 334                 gl->gl_aspace = gfs2_aspace_get(sdp);
 335                 if (!gl->gl_aspace) {
 336                         error = -ENOMEM;
 337                         goto fail;
 338                 }
 339         }
 340
 341         error = gfs2_lm_get_lock(sdp, &name, &gl->gl_lock);
 342         if (error)
 343                 goto fail_aspace;
 344
 345         write_lock(gl_lock_addr(hash));
 346         tmp = search_bucket(hash, sdp, &name);
 347         if (tmp) {
 348                 write_unlock(gl_lock_addr(hash));
 349                 glock_free(gl);
 350                 gl = tmp;
 351         } else {
 352                 hlist_add_head(&gl->gl_list, &gl_hash_table[hash].hb_list);
 353                 write_unlock(gl_lock_addr(hash));
 354         }
 355
 356         *glp = gl;
 357
 358         return 0;
 359
 360 fail_aspace:
 361         if (gl->gl_aspace)
 362                 gfs2_aspace_put(gl->gl_aspace);
 363 fail:
 364         kmem_cache_free(gfs2_glock_cachep, gl);
 365         return error;
 366 }
 367
 368 /**
 369  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
 370  * @gl: the glock
 371  * @state: the state we're requesting
 372  * @flags: the modifier flags
 373  * @gh: the holder structure
 374  *
 375  */
 376
 377 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
 378                       struct gfs2_holder *gh)
 379 {
 380         INIT_LIST_HEAD(&gh->gh_list);
 381         gh->gh_gl = gl;
 382         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 383         gh->gh_owner_pid = current->pid;
 384         gh->gh_state = state;
 385         gh->gh_flags = flags;
 386         gh->gh_error = 0;
 387         gh->gh_iflags = 0;
 388         gfs2_glock_hold(gl);
 389 }
 390
 391 /**
 392  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
 393  * @state: the state we're requesting
 394  * @flags: the modifier flags
 395  * @gh: the holder structure
 396  *
 397  * Don't mess with the glock.
 398  *
 399  */
 400
 401 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
 402 {
 403         gh->gh_state = state;
 404         gh->gh_flags = flags;
 405         gh->gh_iflags = 0;
 406         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 407 }
 408
 409 /**
 410  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
 411  * @gh: the holder structure
 412  *
 413  */
 414
 415 void gfs2_holder_uninit(struct gfs2_holder *gh)
 416 {
 417         gfs2_glock_put(gh->gh_gl);
 418         gh->gh_gl = NULL;
 419         gh->gh_ip = 0;
 420 }
 421
 422 static void gfs2_holder_wake(struct gfs2_holder *gh)
 423 {
 424         clear_bit(HIF_WAIT, &gh->gh_iflags);
 425         smp_mb();
 426         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
 427 }
 428
 429 static int holder_wait(void *word)
 430 {
 431         schedule();
 432         return 0;
 433 }
 434
 435 static void wait_on_holder(struct gfs2_holder *gh)
 436 {
 437         might_sleep();
 438         wait_on_bit(&gh->gh_iflags, HIF_WAIT, holder_wait, TASK_UNINTERRUPTIBLE);
 439 }
 440
 441 /**
 442  * rq_mutex - process a mutex request in the queue
 443  * @gh: the glock holder
 444  *
 445  * Returns: 1 if the queue is blocked
 446  */
 447
 448 static int rq_mutex(struct gfs2_holder *gh)
 449 {
 450         struct gfs2_glock *gl = gh->gh_gl;
 451
 452         list_del_init(&gh->gh_list);
 453         /*  gh->gh_error never examined.  */
 454         set_bit(GLF_LOCK, &gl->gl_flags);
 455         clear_bit(HIF_WAIT, &gh->gh_iflags);
 456         smp_mb();
 457         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
 458
 459         return 1;
 460 }
 461
 462 /**
 463  * rq_promote - process a promote request in the queue
 464  * @gh: the glock holder
 465  *
 466  * Acquire a new inter-node lock, or change a lock state to more restrictive.
 467  *
 468  * Returns: 1 if the queue is blocked
 469  */
 470
 471 static int rq_promote(struct gfs2_holder *gh)
 472 {
 473         struct gfs2_glock *gl = gh->gh_gl;
 474         struct gfs2_sbd *sdp = gl->gl_sbd;
 475
 476         if (!relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
 477                 if (list_empty(&gl->gl_holders)) {
 478                         gl->gl_req_gh = gh;
 479                         set_bit(GLF_LOCK, &gl->gl_flags);
 480                         spin_unlock(&gl->gl_spin);
 481
 482                         if (atomic_read(&sdp->sd_reclaim_count) >
 483                             gfs2_tune_get(sdp, gt_reclaim_limit) &&
 484                             !(gh->gh_flags & LM_FLAG_PRIORITY)) {
 485                                 gfs2_reclaim_glock(sdp);
 486                                 gfs2_reclaim_glock(sdp);
 487                         }
 488
 489                         gfs2_glock_xmote_th(gh->gh_gl, gh);
 490                         spin_lock(&gl->gl_spin);
 491                 }
 492                 return 1;
 493         }
 494
 495         if (list_empty(&gl->gl_holders)) {
 496                 set_bit(HIF_FIRST, &gh->gh_iflags);
 497                 set_bit(GLF_LOCK, &gl->gl_flags);
 498         } else {
 499                 struct gfs2_holder *next_gh;
 500                 if (gh->gh_state == LM_ST_EXCLUSIVE)
 501                         return 1;
 502                 next_gh = list_entry(gl->gl_holders.next, struct gfs2_holder,
 503                                      gh_list);
 504                 if (next_gh->gh_state == LM_ST_EXCLUSIVE)
 505                          return 1;
 506         }
 507
 508         list_move_tail(&gh->gh_list, &gl->gl_holders);
 509         gh->gh_error = 0;
 510         set_bit(HIF_HOLDER, &gh->gh_iflags);
 511
 512         gfs2_holder_wake(gh);
 513
 514         return 0;
 515 }
 516
 517 /**
 518  * rq_demote - process a demote request in the queue
 519  * @gh: the glock holder
 520  *
 521  * Returns: 1 if the queue is blocked
 522  */
 523
 524 static int rq_demote(struct gfs2_glock *gl)
 525 {
 526         if (!list_empty(&gl->gl_holders))
 527                 return 1;
 528
 529         if (gl->gl_state == gl->gl_demote_state ||
 530             gl->gl_state == LM_ST_UNLOCKED) {
 531                 clear_bit(GLF_DEMOTE, &gl->gl_flags);
 532                 return 0;
 533         }
 534         set_bit(GLF_LOCK, &gl->gl_flags);
 535         spin_unlock(&gl->gl_spin);
 536         if (gl->gl_demote_state == LM_ST_UNLOCKED ||
 537             gl->gl_state != LM_ST_EXCLUSIVE)
 538                 gfs2_glock_drop_th(gl);
 539         else
 540                 gfs2_glock_xmote_th(gl, NULL);
 541         spin_lock(&gl->gl_spin);
 542
 543         return 0;
 544 }
 545
 546 /**
 547  * run_queue - process holder structures on a glock
 548  * @gl: the glock
 549  *
 550  */
 551 static void run_queue(struct gfs2_glock *gl)
 552 {
 553         struct gfs2_holder *gh;
 554         int blocked = 1;
 555
 556         for (;;) {
 557                 if (test_bit(GLF_LOCK, &gl->gl_flags))
 558                         break;
 559
 560                 if (!list_empty(&gl->gl_waiters1)) {
 561                         gh = list_entry(gl->gl_waiters1.next,
 562                                         struct gfs2_holder, gh_list);
 563
 564                         if (test_bit(HIF_MUTEX, &gh->gh_iflags))
 565                                 blocked = rq_mutex(gh);
 566                         else
 567                                 gfs2_assert_warn(gl->gl_sbd, 0);
 568
 569                 } else if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
 570                         blocked = rq_demote(gl);
 571                 } else if (!list_empty(&gl->gl_waiters3)) {
 572                         gh = list_entry(gl->gl_waiters3.next,
 573                                         struct gfs2_holder, gh_list);
 574
 575                         if (test_bit(HIF_PROMOTE, &gh->gh_iflags))
 576                                 blocked = rq_promote(gh);
 577                         else
 578                                 gfs2_assert_warn(gl->gl_sbd, 0);
 579
 580                 } else
 581                         break;
 582
 583                 if (blocked)
 584                         break;
 585         }
 586 }
 587
 588 /**
 589  * gfs2_glmutex_lock - acquire a local lock on a glock
 590  * @gl: the glock
 591  *
 592  * Gives caller exclusive access to manipulate a glock structure.
 593  */
 594
 595 static void gfs2_glmutex_lock(struct gfs2_glock *gl)
 596 {
 597         struct gfs2_holder gh;
 598
 599         gfs2_holder_init(gl, 0, 0, &gh);
 600         set_bit(HIF_MUTEX, &gh.gh_iflags);
 601         if (test_and_set_bit(HIF_WAIT, &gh.gh_iflags))
 602                 BUG();
 603
 604         spin_lock(&gl->gl_spin);
 605         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
 606                 list_add_tail(&gh.gh_list, &gl->gl_waiters1);
 607         } else {
 608                 gl->gl_owner_pid = current->pid;
 609                 gl->gl_ip = (unsigned long)__builtin_return_address(0);
 610                 clear_bit(HIF_WAIT, &gh.gh_iflags);
 611                 smp_mb();
 612                 wake_up_bit(&gh.gh_iflags, HIF_WAIT);
 613         }
 614         spin_unlock(&gl->gl_spin);
 615
 616         wait_on_holder(&gh);
 617         gfs2_holder_uninit(&gh);
 618 }
 619
 620 /**
 621  * gfs2_glmutex_trylock - try to acquire a local lock on a glock
 622  * @gl: the glock
 623  *
 624  * Returns: 1 if the glock is acquired
 625  */
 626
 627 static int gfs2_glmutex_trylock(struct gfs2_glock *gl)
 628 {
 629         int acquired = 1;
 630
 631         spin_lock(&gl->gl_spin);
 632         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
 633                 acquired = 0;
 634         } else {
 635                 gl->gl_owner_pid = current->pid;
 636                 gl->gl_ip = (unsigned long)__builtin_return_address(0);
 637         }
 638         spin_unlock(&gl->gl_spin);
 639
 640         return acquired;
 641 }
 642
 643 /**
 644  * gfs2_glmutex_unlock - release a local lock on a glock
 645  * @gl: the glock
 646  *
 647  */
 648
 649 static void gfs2_glmutex_unlock(struct gfs2_glock *gl)
 650 {
 651         spin_lock(&gl->gl_spin);
 652         clear_bit(GLF_LOCK, &gl->gl_flags);
 653         gl->gl_owner_pid = 0;
 654         gl->gl_ip = 0;
 655         run_queue(gl);
 656         BUG_ON(!spin_is_locked(&gl->gl_spin));
 657         spin_unlock(&gl->gl_spin);
 658 }
 659
 660 /**
 661  * handle_callback - process a demote request
 662  * @gl: the glock
 663  * @state: the state the caller wants us to change to
 664  *
 665  * There are only two requests that we are going to see in actual
 666  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
 667  */
 668
 669 static void handle_callback(struct gfs2_glock *gl, unsigned int state)
 670 {
 671         spin_lock(&gl->gl_spin);
 672         if (test_and_set_bit(GLF_DEMOTE, &gl->gl_flags) == 0) {
 673                 gl->gl_demote_state = state;
 674                 gl->gl_demote_time = jiffies;
 675         } else if (gl->gl_demote_state != LM_ST_UNLOCKED) {
 676                 gl->gl_demote_state = state;
 677         }
 678         spin_unlock(&gl->gl_spin);
 679 }
 680
 681 /**
 682  * state_change - record that the glock is now in a different state
 683  * @gl: the glock
 684  * @new_state the new state
 685  *
 686  */
 687
 688 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
 689 {
 690         int held1, held2;
 691
 692         held1 = (gl->gl_state != LM_ST_UNLOCKED);
 693         held2 = (new_state != LM_ST_UNLOCKED);
 694
 695         if (held1 != held2) {
 696                 if (held2)
 697                         gfs2_glock_hold(gl);
 698                 else
 699                         gfs2_glock_put(gl);
 700         }
 701
 702         gl->gl_state = new_state;
 703 }
 704
 705 /**
 706  * xmote_bh - Called after the lock module is done acquiring a lock
 707  * @gl: The glock in question
 708  * @ret: the int returned from the lock module
 709  *
 710  */
 711
 712 static void xmote_bh(struct gfs2_glock *gl, unsigned int ret)
 713 {
 714         struct gfs2_sbd *sdp = gl->gl_sbd;
 715         const struct gfs2_glock_operations *glops = gl->gl_ops;
 716         struct gfs2_holder *gh = gl->gl_req_gh;
 717         int prev_state = gl->gl_state;
 718         int op_done = 1;
 719
 720         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
 721         gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
 722         gfs2_assert_warn(sdp, !(ret & LM_OUT_ASYNC));
 723
 724         state_change(gl, ret & LM_OUT_ST_MASK);
 725
 726         if (prev_state != LM_ST_UNLOCKED && !(ret & LM_OUT_CACHEABLE)) {
 727                 if (glops->go_inval)
 728                         glops->go_inval(gl, DIO_METADATA);
 729         } else if (gl->gl_state == LM_ST_DEFERRED) {
 730                 /* We might not want to do this here.
 731                    Look at moving to the inode glops. */
 732                 if (glops->go_inval)
 733                         glops->go_inval(gl, 0);
 734         }
 735
 736         /*  Deal with each possible exit condition  */
 737
 738         if (!gh) {
 739                 gl->gl_stamp = jiffies;
 740                 if (ret & LM_OUT_CANCELED)
 741                         op_done = 0;
 742                 else
 743                         clear_bit(GLF_DEMOTE, &gl->gl_flags);
 744         } else {
 745                 spin_lock(&gl->gl_spin);
 746                 list_del_init(&gh->gh_list);
 747                 gh->gh_error = -EIO;
 748                 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 749                         goto out;
 750                 gh->gh_error = GLR_CANCELED;
 751                 if (ret & LM_OUT_CANCELED)
 752                         goto out;
 753                 if (relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
 754                         list_add_tail(&gh->gh_list, &gl->gl_holders);
 755                         gh->gh_error = 0;
 756                         set_bit(HIF_HOLDER, &gh->gh_iflags);
 757                         set_bit(HIF_FIRST, &gh->gh_iflags);
 758                         op_done = 0;
 759                         goto out;
 760                 }
 761                 gh->gh_error = GLR_TRYFAILED;
 762                 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
 763                         goto out;
 764                 gh->gh_error = -EINVAL;
 765                 if (gfs2_assert_withdraw(sdp, 0) == -1)
 766                         fs_err(sdp, "ret = 0x%.8X\n", ret);
 767 out:
 768                 spin_unlock(&gl->gl_spin);
 769         }
 770
 771         if (glops->go_xmote_bh)
 772                 glops->go_xmote_bh(gl);
 773
 774         if (op_done) {
 775                 spin_lock(&gl->gl_spin);
 776                 gl->gl_req_gh = NULL;
 777                 gl->gl_req_bh = NULL;
 778                 clear_bit(GLF_LOCK, &gl->gl_flags);
 779                 run_queue(gl);
 780                 spin_unlock(&gl->gl_spin);
 781         }
 782
 783         gfs2_glock_put(gl);
 784
 785         if (gh)
 786                 gfs2_holder_wake(gh);
 787 }
 788
 789 /**
 790  * gfs2_glock_xmote_th - Call into the lock module to acquire or change a glock
 791  * @gl: The glock in question
 792  * @state: the requested state
 793  * @flags: modifier flags to the lock call
 794  *
 795  */
 796
 797 void gfs2_glock_xmote_th(struct gfs2_glock *gl, struct gfs2_holder *gh)
 798 {
 799         struct gfs2_sbd *sdp = gl->gl_sbd;
 800         int flags = gh ? gh->gh_flags : 0;
 801         unsigned state = gh ? gh->gh_state : gl->gl_demote_state;
 802         const struct gfs2_glock_operations *glops = gl->gl_ops;
 803         int lck_flags = flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB |
 804                                  LM_FLAG_NOEXP | LM_FLAG_ANY |
 805                                  LM_FLAG_PRIORITY);
 806         unsigned int lck_ret;
 807
 808         if (glops->go_xmote_th)
 809                 glops->go_xmote_th(gl);
 810
 811         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
 812         gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
 813         gfs2_assert_warn(sdp, state != LM_ST_UNLOCKED);
 814         gfs2_assert_warn(sdp, state != gl->gl_state);
 815
 816         gfs2_glock_hold(gl);
 817         gl->gl_req_bh = xmote_bh;
 818
 819         lck_ret = gfs2_lm_lock(sdp, gl->gl_lock, gl->gl_state, state, lck_flags);
 820
 821         if (gfs2_assert_withdraw(sdp, !(lck_ret & LM_OUT_ERROR)))
 822                 return;
 823
 824         if (lck_ret & LM_OUT_ASYNC)
 825                 gfs2_assert_warn(sdp, lck_ret == LM_OUT_ASYNC);
 826         else
 827                 xmote_bh(gl, lck_ret);
 828 }
 829
 830 /**
 831  * drop_bh - Called after a lock module unlock completes
 832  * @gl: the glock
 833  * @ret: the return status
 834  *
 835  * Doesn't wake up the process waiting on the struct gfs2_holder (if any)
 836  * Doesn't drop the reference on the glock the top half took out
 837  *
 838  */
 839
 840 static void drop_bh(struct gfs2_glock *gl, unsigned int ret)
 841 {
 842         struct gfs2_sbd *sdp = gl->gl_sbd;
 843         const struct gfs2_glock_operations *glops = gl->gl_ops;
 844         struct gfs2_holder *gh = gl->gl_req_gh;
 845
 846         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
 847         gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
 848         gfs2_assert_warn(sdp, !ret);
 849
 850         state_change(gl, LM_ST_UNLOCKED);
 851         clear_bit(GLF_DEMOTE, &gl->gl_flags);
 852
 853         if (glops->go_inval)
 854                 glops->go_inval(gl, DIO_METADATA);
 855
 856         if (gh) {
 857                 spin_lock(&gl->gl_spin);
 858                 list_del_init(&gh->gh_list);
 859                 gh->gh_error = 0;
 860                 spin_unlock(&gl->gl_spin);
 861         }
 862
 863         spin_lock(&gl->gl_spin);
 864         gl->gl_req_gh = NULL;
 865         gl->gl_req_bh = NULL;
 866         clear_bit(GLF_LOCK, &gl->gl_flags);
 867         run_queue(gl);
 868         spin_unlock(&gl->gl_spin);
 869
 870         gfs2_glock_put(gl);
 871
 872         if (gh)
 873                 gfs2_holder_wake(gh);
 874 }
 875
 876 /**
 877  * gfs2_glock_drop_th - call into the lock module to unlock a lock
 878  * @gl: the glock
 879  *
 880  */
 881
 882 static void gfs2_glock_drop_th(struct gfs2_glock *gl)
 883 {
 884         struct gfs2_sbd *sdp = gl->gl_sbd;
 885         const struct gfs2_glock_operations *glops = gl->gl_ops;
 886         unsigned int ret;
 887
 888         if (glops->go_drop_th)
 889                 glops->go_drop_th(gl);
 890
 891         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
 892         gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
 893         gfs2_assert_warn(sdp, gl->gl_state != LM_ST_UNLOCKED);
 894
 895         gfs2_glock_hold(gl);
 896         gl->gl_req_bh = drop_bh;
 897
 898         ret = gfs2_lm_unlock(sdp, gl->gl_lock, gl->gl_state);
 899
 900         if (gfs2_assert_withdraw(sdp, !(ret & LM_OUT_ERROR)))
 901                 return;
 902
 903         if (!ret)
 904                 drop_bh(gl, ret);
 905         else
 906                 gfs2_assert_warn(sdp, ret == LM_OUT_ASYNC);
 907 }
 908
 909 /**
 910  * do_cancels - cancel requests for locks stuck waiting on an expire flag
 911  * @gh: the LM_FLAG_PRIORITY holder waiting to acquire the lock
 912  *
 913  * Don't cancel GL_NOCANCEL requests.
 914  */
 915
 916 static void do_cancels(struct gfs2_holder *gh)
 917 {
 918         struct gfs2_glock *gl = gh->gh_gl;
 919
 920         spin_lock(&gl->gl_spin);
 921
 922         while (gl->gl_req_gh != gh &&
 923                !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
 924                !list_empty(&gh->gh_list)) {
 925                 if (gl->gl_req_bh && !(gl->gl_req_gh &&
 926                                      (gl->gl_req_gh->gh_flags & GL_NOCANCEL))) {
 927                         spin_unlock(&gl->gl_spin);
 928                         gfs2_lm_cancel(gl->gl_sbd, gl->gl_lock);
 929                         msleep(100);
 930                         spin_lock(&gl->gl_spin);
 931                 } else {
 932                         spin_unlock(&gl->gl_spin);
 933                         msleep(100);
 934                         spin_lock(&gl->gl_spin);
 935                 }
 936         }
 937
 938         spin_unlock(&gl->gl_spin);
 939 }
 940
 941 /**
 942  * glock_wait_internal - wait on a glock acquisition
 943  * @gh: the glock holder
 944  *
 945  * Returns: 0 on success
 946  */
 947
 948 static int glock_wait_internal(struct gfs2_holder *gh)
 949 {
 950         struct gfs2_glock *gl = gh->gh_gl;
 951         struct gfs2_sbd *sdp = gl->gl_sbd;
 952         const struct gfs2_glock_operations *glops = gl->gl_ops;
 953
 954         if (test_bit(HIF_ABORTED, &gh->gh_iflags))
 955                 return -EIO;
 956
 957         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
 958                 spin_lock(&gl->gl_spin);
 959                 if (gl->gl_req_gh != gh &&
 960                     !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
 961                     !list_empty(&gh->gh_list)) {
 962                         list_del_init(&gh->gh_list);
 963                         gh->gh_error = GLR_TRYFAILED;
 964                         run_queue(gl);
 965                         spin_unlock(&gl->gl_spin);
 966                         return gh->gh_error;
 967                 }
 968                 spin_unlock(&gl->gl_spin);
 969         }
 970
 971         if (gh->gh_flags & LM_FLAG_PRIORITY)
 972                 do_cancels(gh);
 973
 974         wait_on_holder(gh);
 975         if (gh->gh_error)
 976                 return gh->gh_error;
 977
 978         gfs2_assert_withdraw(sdp, test_bit(HIF_HOLDER, &gh->gh_iflags));
 979         gfs2_assert_withdraw(sdp, relaxed_state_ok(gl->gl_state, gh->gh_state,
 980                                                    gh->gh_flags));
 981
 982         if (test_bit(HIF_FIRST, &gh->gh_iflags)) {
 983                 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
 984
 985                 if (glops->go_lock) {
 986                         gh->gh_error = glops->go_lock(gh);
 987                         if (gh->gh_error) {
 988                                 spin_lock(&gl->gl_spin);
 989                                 list_del_init(&gh->gh_list);
 990                                 spin_unlock(&gl->gl_spin);
 991                         }
 992                 }
 993
 994                 spin_lock(&gl->gl_spin);
 995                 gl->gl_req_gh = NULL;
 996                 gl->gl_req_bh = NULL;
 997                 clear_bit(GLF_LOCK, &gl->gl_flags);
 998                 run_queue(gl);
 999                 spin_unlock(&gl->gl_spin);
1000         }
1001
1002         return gh->gh_error;
1003 }
1004
1005 static inline struct gfs2_holder *
1006 find_holder_by_owner(struct list_head *head, pid_t pid)
1007 {
1008         struct gfs2_holder *gh;
1009
1010         list_for_each_entry(gh, head, gh_list) {
1011                 if (gh->gh_owner_pid == pid)
1012                         return gh;
1013         }
1014
1015         return NULL;
1016 }
1017
1018 static void print_dbg(struct glock_iter *gi, const char *fmt, ...)
1019 {
1020         va_list args;
1021
1022         va_start(args, fmt);
1023         if (gi) {
1024                 vsprintf(gi->string, fmt, args);
1025                 seq_printf(gi->seq, gi->string);
1026         }
1027         else
1028                 vprintk(fmt, args);
1029         va_end(args);
1030 }
1031
1032 /**
1033  * add_to_queue - Add a holder to the wait queue (but look for recursion)
1034  * @gh: the holder structure to add
1035  *
1036  */
1037
1038 static void add_to_queue(struct gfs2_holder *gh)
1039 {
1040         struct gfs2_glock *gl = gh->gh_gl;
1041         struct gfs2_holder *existing;
1042
1043         BUG_ON(!gh->gh_owner_pid);
1044         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1045                 BUG();
1046
1047         existing = find_holder_by_owner(&gl->gl_holders, gh->gh_owner_pid);
1048         if (existing) {
1049                 print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
1050                 printk(KERN_INFO "pid : %d\n", existing->gh_owner_pid);
1051                 printk(KERN_INFO "lock type : %d lock state : %d\n",
1052                                 existing->gh_gl->gl_name.ln_type, existing->gh_gl->gl_state);
1053                 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1054                 printk(KERN_INFO "pid : %d\n", gh->gh_owner_pid);
1055                 printk(KERN_INFO "lock type : %d lock state : %d\n",
1056                                 gl->gl_name.ln_type, gl->gl_state);
1057                 BUG();
1058         }
1059
1060         existing = find_holder_by_owner(&gl->gl_waiters3, gh->gh_owner_pid);
1061         if (existing) {
1062                 print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
1063                 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1064                 BUG();
1065         }
1066
1067         if (gh->gh_flags & LM_FLAG_PRIORITY)
1068                 list_add(&gh->gh_list, &gl->gl_waiters3);
1069         else
1070                 list_add_tail(&gh->gh_list, &gl->gl_waiters3);
1071 }
1072
1073 /**
1074  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1075  * @gh: the holder structure
1076  *
1077  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1078  *
1079  * Returns: 0, GLR_TRYFAILED, or errno on failure
1080  */
1081
1082 int gfs2_glock_nq(struct gfs2_holder *gh)
1083 {
1084         struct gfs2_glock *gl = gh->gh_gl;
1085         struct gfs2_sbd *sdp = gl->gl_sbd;
1086         int error = 0;
1087
1088 restart:
1089         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1090                 set_bit(HIF_ABORTED, &gh->gh_iflags);
1091                 return -EIO;
1092         }
1093
1094         set_bit(HIF_PROMOTE, &gh->gh_iflags);
1095
1096         spin_lock(&gl->gl_spin);
1097         add_to_queue(gh);
1098         run_queue(gl);
1099         spin_unlock(&gl->gl_spin);
1100
1101         if (!(gh->gh_flags & GL_ASYNC)) {
1102                 error = glock_wait_internal(gh);
1103                 if (error == GLR_CANCELED) {
1104                         msleep(100);
1105                         goto restart;
1106                 }
1107         }
1108
1109         return error;
1110 }
1111
1112 /**
1113  * gfs2_glock_poll - poll to see if an async request has been completed
1114  * @gh: the holder
1115  *
1116  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1117  */
1118
1119 int gfs2_glock_poll(struct gfs2_holder *gh)
1120 {
1121         struct gfs2_glock *gl = gh->gh_gl;
1122         int ready = 0;
1123
1124         spin_lock(&gl->gl_spin);
1125
1126         if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1127                 ready = 1;
1128         else if (list_empty(&gh->gh_list)) {
1129                 if (gh->gh_error == GLR_CANCELED) {
1130                         spin_unlock(&gl->gl_spin);
1131                         msleep(100);
1132                         if (gfs2_glock_nq(gh))
1133                                 return 1;
1134                         return 0;
1135                 } else
1136                         ready = 1;
1137         }
1138
1139         spin_unlock(&gl->gl_spin);
1140
1141         return ready;
1142 }
1143
1144 /**
1145  * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
1146  * @gh: the holder structure
1147  *
1148  * Returns: 0, GLR_TRYFAILED, or errno on failure
1149  */
1150
1151 int gfs2_glock_wait(struct gfs2_holder *gh)
1152 {
1153         int error;
1154
1155         error = glock_wait_internal(gh);
1156         if (error == GLR_CANCELED) {
1157                 msleep(100);
1158                 gh->gh_flags &= ~GL_ASYNC;
1159                 error = gfs2_glock_nq(gh);
1160         }
1161
1162         return error;
1163 }
1164
1165 /**
1166  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1167  * @gh: the glock holder
1168  *
1169  */
1170
1171 void gfs2_glock_dq(struct gfs2_holder *gh)
1172 {
1173         struct gfs2_glock *gl = gh->gh_gl;
1174         const struct gfs2_glock_operations *glops = gl->gl_ops;
1175
1176         if (gh->gh_flags & GL_NOCACHE)
1177                 handle_callback(gl, LM_ST_UNLOCKED);
1178
1179         gfs2_glmutex_lock(gl);
1180
1181         spin_lock(&gl->gl_spin);
1182         list_del_init(&gh->gh_list);
1183
1184         if (list_empty(&gl->gl_holders)) {
1185                 spin_unlock(&gl->gl_spin);
1186
1187                 if (glops->go_unlock)
1188                         glops->go_unlock(gh);
1189
1190                 spin_lock(&gl->gl_spin);
1191                 gl->gl_stamp = jiffies;
1192         }
1193
1194         clear_bit(GLF_LOCK, &gl->gl_flags);
1195         run_queue(gl);
1196         spin_unlock(&gl->gl_spin);
1197 }
1198
1199 /**
1200  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1201  * @gh: the holder structure
1202  *
1203  */
1204
1205 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1206 {
1207         gfs2_glock_dq(gh);
1208         gfs2_holder_uninit(gh);
1209 }
1210
1211 /**
1212  * gfs2_glock_nq_num - acquire a glock based on lock number
1213  * @sdp: the filesystem
1214  * @number: the lock number
1215  * @glops: the glock operations for the type of glock
1216  * @state: the state to acquire the glock in
1217  * @flags: modifier flags for the aquisition
1218  * @gh: the struct gfs2_holder
1219  *
1220  * Returns: errno
1221  */
1222
1223 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1224                       const struct gfs2_glock_operations *glops,
1225                       unsigned int state, int flags, struct gfs2_holder *gh)
1226 {
1227         struct gfs2_glock *gl;
1228         int error;
1229
1230         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1231         if (!error) {
1232                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1233                 gfs2_glock_put(gl);
1234         }
1235
1236         return error;
1237 }
1238
1239 /**
1240  * glock_compare - Compare two struct gfs2_glock structures for sorting
1241  * @arg_a: the first structure
1242  * @arg_b: the second structure
1243  *
1244  */
1245
1246 static int glock_compare(const void *arg_a, const void *arg_b)
1247 {
1248         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1249         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1250         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1251         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1252
1253         if (a->ln_number > b->ln_number)
1254                 return 1;
1255         if (a->ln_number < b->ln_number)
1256                 return -1;
1257         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1258         return 0;
1259 }
1260
1261 /**
1262  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1263  * @num_gh: the number of structures
1264  * @ghs: an array of struct gfs2_holder structures
1265  *
1266  * Returns: 0 on success (all glocks acquired),
1267  *          errno on failure (no glocks acquired)
1268  */
1269
1270 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1271                      struct gfs2_holder **p)
1272 {
1273         unsigned int x;
1274         int error = 0;
1275
1276         for (x = 0; x < num_gh; x++)
1277                 p[x] = &ghs[x];
1278
1279         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1280
1281         for (x = 0; x < num_gh; x++) {
1282                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1283
1284                 error = gfs2_glock_nq(p[x]);
1285                 if (error) {
1286                         while (x--)
1287                                 gfs2_glock_dq(p[x]);
1288                         break;
1289                 }
1290         }
1291
1292         return error;
1293 }
1294
1295 /**
1296  * gfs2_glock_nq_m - acquire multiple glocks
1297  * @num_gh: the number of structures
1298  * @ghs: an array of struct gfs2_holder structures
1299  *
1300  * Figure out how big an impact this function has.  Either:
1301  * 1) Replace this code with code that calls gfs2_glock_prefetch()
1302  * 2) Forget async stuff and just call nq_m_sync()
1303  * 3) Leave it like it is
1304  *
1305  * Returns: 0 on success (all glocks acquired),
1306  *          errno on failure (no glocks acquired)
1307  */
1308
1309 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1310 {
1311         int *e;
1312         unsigned int x;
1313         int borked = 0, serious = 0;
1314         int error = 0;
1315
1316         if (!num_gh)
1317                 return 0;
1318
1319         if (num_gh == 1) {
1320                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1321                 return gfs2_glock_nq(ghs);
1322         }
1323
1324         e = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL);
1325         if (!e)
1326                 return -ENOMEM;
1327
1328         for (x = 0; x < num_gh; x++) {
1329                 ghs[x].gh_flags |= LM_FLAG_TRY | GL_ASYNC;
1330                 error = gfs2_glock_nq(&ghs[x]);
1331                 if (error) {
1332                         borked = 1;
1333                         serious = error;
1334                         num_gh = x;
1335                         break;
1336                 }
1337         }
1338
1339         for (x = 0; x < num_gh; x++) {
1340                 error = e[x] = glock_wait_internal(&ghs[x]);
1341                 if (error) {
1342                         borked = 1;
1343                         if (error != GLR_TRYFAILED && error != GLR_CANCELED)
1344                                 serious = error;
1345                 }
1346         }
1347
1348         if (!borked) {
1349                 kfree(e);
1350                 return 0;
1351         }
1352
1353         for (x = 0; x < num_gh; x++)
1354                 if (!e[x])
1355                         gfs2_glock_dq(&ghs[x]);
1356
1357         if (serious)
1358                 error = serious;
1359         else {
1360                 for (x = 0; x < num_gh; x++)
1361                         gfs2_holder_reinit(ghs[x].gh_state, ghs[x].gh_flags,
1362                                           &ghs[x]);
1363                 error = nq_m_sync(num_gh, ghs, (struct gfs2_holder **)e);
1364         }
1365
1366         kfree(e);
1367
1368         return error;
1369 }
1370
1371 /**
1372  * gfs2_glock_dq_m - release multiple glocks
1373  * @num_gh: the number of structures
1374  * @ghs: an array of struct gfs2_holder structures
1375  *
1376  */
1377
1378 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1379 {
1380         unsigned int x;
1381
1382         for (x = 0; x < num_gh; x++)
1383                 gfs2_glock_dq(&ghs[x]);
1384 }
1385
1386 /**
1387  * gfs2_glock_dq_uninit_m - release multiple glocks
1388  * @num_gh: the number of structures
1389  * @ghs: an array of struct gfs2_holder structures
1390  *
1391  */
1392
1393 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1394 {
1395         unsigned int x;
1396
1397         for (x = 0; x < num_gh; x++)
1398                 gfs2_glock_dq_uninit(&ghs[x]);
1399 }
1400
1401 /**
1402  * gfs2_lvb_hold - attach a LVB from a glock
1403  * @gl: The glock in question
1404  *
1405  */
1406
1407 int gfs2_lvb_hold(struct gfs2_glock *gl)
1408 {
1409         int error;
1410
1411         gfs2_glmutex_lock(gl);
1412
1413         if (!atomic_read(&gl->gl_lvb_count)) {
1414                 error = gfs2_lm_hold_lvb(gl->gl_sbd, gl->gl_lock, &gl->gl_lvb);
1415                 if (error) {
1416                         gfs2_glmutex_unlock(gl);
1417                         return error;
1418                 }
1419                 gfs2_glock_hold(gl);
1420         }
1421         atomic_inc(&gl->gl_lvb_count);
1422
1423         gfs2_glmutex_unlock(gl);
1424
1425         return 0;
1426 }
1427
1428 /**
1429  * gfs2_lvb_unhold - detach a LVB from a glock
1430  * @gl: The glock in question
1431  *
1432  */
1433
1434 void gfs2_lvb_unhold(struct gfs2_glock *gl)
1435 {
1436         gfs2_glock_hold(gl);
1437         gfs2_glmutex_lock(gl);
1438
1439         gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count) > 0);
1440         if (atomic_dec_and_test(&gl->gl_lvb_count)) {
1441                 gfs2_lm_unhold_lvb(gl->gl_sbd, gl->gl_lock, gl->gl_lvb);
1442                 gl->gl_lvb = NULL;
1443                 gfs2_glock_put(gl);
1444         }
1445
1446         gfs2_glmutex_unlock(gl);
1447         gfs2_glock_put(gl);
1448 }
1449
1450 static void blocking_cb(struct gfs2_sbd *sdp, struct lm_lockname *name,
1451                         unsigned int state)
1452 {
1453         struct gfs2_glock *gl;
1454
1455         gl = gfs2_glock_find(sdp, name);
1456         if (!gl)
1457                 return;
1458
1459         handle_callback(gl, state);
1460
1461         spin_lock(&gl->gl_spin);
1462         run_queue(gl);
1463         spin_unlock(&gl->gl_spin);
1464
1465         gfs2_glock_put(gl);
1466 }
1467
1468 /**
1469  * gfs2_glock_cb - Callback used by locking module
1470  * @sdp: Pointer to the superblock
1471  * @type: Type of callback
1472  * @data: Type dependent data pointer
1473  *
1474  * Called by the locking module when it wants to tell us something.
1475  * Either we need to drop a lock, one of our ASYNC requests completed, or
1476  * a journal from another client needs to be recovered.
1477  */
1478
1479 void gfs2_glock_cb(void *cb_data, unsigned int type, void *data)
1480 {
1481         struct gfs2_sbd *sdp = cb_data;
1482
1483         switch (type) {
1484         case LM_CB_NEED_E:
1485                 blocking_cb(sdp, data, LM_ST_UNLOCKED);
1486                 return;
1487
1488         case LM_CB_NEED_D:
1489                 blocking_cb(sdp, data, LM_ST_DEFERRED);
1490                 return;
1491
1492         case LM_CB_NEED_S:
1493                 blocking_cb(sdp, data, LM_ST_SHARED);
1494                 return;
1495
1496         case LM_CB_ASYNC: {
1497                 struct lm_async_cb *async = data;
1498                 struct gfs2_glock *gl;
1499
1500                 down_read(&gfs2_umount_flush_sem);
1501                 gl = gfs2_glock_find(sdp, &async->lc_name);
1502                 if (gfs2_assert_warn(sdp, gl))
1503                         return;
1504                 if (!gfs2_assert_warn(sdp, gl->gl_req_bh))
1505                         gl->gl_req_bh(gl, async->lc_ret);
1506                 gfs2_glock_put(gl);
1507                 up_read(&gfs2_umount_flush_sem);
1508                 return;
1509         }
1510
1511         case LM_CB_NEED_RECOVERY:
1512                 gfs2_jdesc_make_dirty(sdp, *(unsigned int *)data);
1513                 if (sdp->sd_recoverd_process)
1514                         wake_up_process(sdp->sd_recoverd_process);
1515                 return;
1516
1517         case LM_CB_DROPLOCKS:
1518                 gfs2_gl_hash_clear(sdp, NO_WAIT);
1519                 gfs2_quota_scan(sdp);
1520                 return;
1521
1522         default:
1523                 gfs2_assert_warn(sdp, 0);
1524                 return;
1525         }
1526 }
1527
1528 /**
1529  * demote_ok - Check to see if it's ok to unlock a glock
1530  * @gl: the glock
1531  *
1532  * Returns: 1 if it's ok
1533  */
1534
1535 static int demote_ok(struct gfs2_glock *gl)
1536 {
1537         const struct gfs2_glock_operations *glops = gl->gl_ops;
1538         int demote = 1;
1539
1540         if (test_bit(GLF_STICKY, &gl->gl_flags))
1541                 demote = 0;
1542         else if (glops->go_demote_ok)
1543                 demote = glops->go_demote_ok(gl);
1544
1545         return demote;
1546 }
1547
1548 /**
1549  * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1550  * @gl: the glock
1551  *
1552  */
1553
1554 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
1555 {
1556         struct gfs2_sbd *sdp = gl->gl_sbd;
1557
1558         spin_lock(&sdp->sd_reclaim_lock);
1559         if (list_empty(&gl->gl_reclaim)) {
1560                 gfs2_glock_hold(gl);
1561                 list_add(&gl->gl_reclaim, &sdp->sd_reclaim_list);
1562                 atomic_inc(&sdp->sd_reclaim_count);
1563         }
1564         spin_unlock(&sdp->sd_reclaim_lock);
1565
1566         wake_up(&sdp->sd_reclaim_wq);
1567 }
1568
1569 /**
1570  * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1571  * @sdp: the filesystem
1572  *
1573  * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1574  * different glock and we notice that there are a lot of glocks in the
1575  * reclaim list.
1576  *
1577  */
1578
1579 void gfs2_reclaim_glock(struct gfs2_sbd *sdp)
1580 {
1581         struct gfs2_glock *gl;
1582
1583         spin_lock(&sdp->sd_reclaim_lock);
1584         if (list_empty(&sdp->sd_reclaim_list)) {
1585                 spin_unlock(&sdp->sd_reclaim_lock);
1586                 return;
1587         }
1588         gl = list_entry(sdp->sd_reclaim_list.next,
1589                         struct gfs2_glock, gl_reclaim);
1590         list_del_init(&gl->gl_reclaim);
1591         spin_unlock(&sdp->sd_reclaim_lock);
1592
1593         atomic_dec(&sdp->sd_reclaim_count);
1594         atomic_inc(&sdp->sd_reclaimed);
1595
1596         if (gfs2_glmutex_trylock(gl)) {
1597                 if (list_empty(&gl->gl_holders) &&
1598                     gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1599                         handle_callback(gl, LM_ST_UNLOCKED);
1600                 gfs2_glmutex_unlock(gl);
1601         }
1602
1603         gfs2_glock_put(gl);
1604 }
1605
1606 /**
1607  * examine_bucket - Call a function for glock in a hash bucket
1608  * @examiner: the function
1609  * @sdp: the filesystem
1610  * @bucket: the bucket
1611  *
1612  * Returns: 1 if the bucket has entries
1613  */
1614
1615 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1616                           unsigned int hash)
1617 {
1618         struct gfs2_glock *gl, *prev = NULL;
1619         int has_entries = 0;
1620         struct hlist_head *head = &gl_hash_table[hash].hb_list;
1621
1622         read_lock(gl_lock_addr(hash));
1623         /* Can't use hlist_for_each_entry - don't want prefetch here */
1624         if (hlist_empty(head))
1625                 goto out;
1626         gl = list_entry(head->first, struct gfs2_glock, gl_list);
1627         while(1) {
1628                 if (gl->gl_sbd == sdp) {
1629                         gfs2_glock_hold(gl);
1630                         read_unlock(gl_lock_addr(hash));
1631                         if (prev)
1632                                 gfs2_glock_put(prev);
1633                         prev = gl;
1634                         examiner(gl);
1635                         has_entries = 1;
1636                         read_lock(gl_lock_addr(hash));
1637                 }
1638                 if (gl->gl_list.next == NULL)
1639                         break;
1640                 gl = list_entry(gl->gl_list.next, struct gfs2_glock, gl_list);
1641         }
1642 out:
1643         read_unlock(gl_lock_addr(hash));
1644         if (prev)
1645                 gfs2_glock_put(prev);
1646         return has_entries;
1647 }
1648
1649 /**
1650  * scan_glock - look at a glock and see if we can reclaim it
1651  * @gl: the glock to look at
1652  *
1653  */
1654
1655 static void scan_glock(struct gfs2_glock *gl)
1656 {
1657         if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object)
1658                 return;
1659
1660         if (gfs2_glmutex_trylock(gl)) {
1661                 if (list_empty(&gl->gl_holders) &&
1662                     gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1663                         goto out_schedule;
1664                 gfs2_glmutex_unlock(gl);
1665         }
1666         return;
1667
1668 out_schedule:
1669         gfs2_glmutex_unlock(gl);
1670         gfs2_glock_schedule_for_reclaim(gl);
1671 }
1672
1673 /**
1674  * gfs2_scand_internal - Look for glocks and inodes to toss from memory
1675  * @sdp: the filesystem
1676  *
1677  */
1678
1679 void gfs2_scand_internal(struct gfs2_sbd *sdp)
1680 {
1681         unsigned int x;
1682
1683         for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1684                 examine_bucket(scan_glock, sdp, x);
1685 }
1686
1687 /**
1688  * clear_glock - look at a glock and see if we can free it from glock cache
1689  * @gl: the glock to look at
1690  *
1691  */
1692
1693 static void clear_glock(struct gfs2_glock *gl)
1694 {
1695         struct gfs2_sbd *sdp = gl->gl_sbd;
1696         int released;
1697
1698         spin_lock(&sdp->sd_reclaim_lock);
1699         if (!list_empty(&gl->gl_reclaim)) {
1700                 list_del_init(&gl->gl_reclaim);
1701                 atomic_dec(&sdp->sd_reclaim_count);
1702                 spin_unlock(&sdp->sd_reclaim_lock);
1703                 released = gfs2_glock_put(gl);
1704                 gfs2_assert(sdp, !released);
1705         } else {
1706                 spin_unlock(&sdp->sd_reclaim_lock);
1707         }
1708
1709         if (gfs2_glmutex_trylock(gl)) {
1710                 if (list_empty(&gl->gl_holders) &&
1711                     gl->gl_state != LM_ST_UNLOCKED)
1712                         handle_callback(gl, LM_ST_UNLOCKED);
1713                 gfs2_glmutex_unlock(gl);
1714         }
1715 }
1716
1717 /**
1718  * gfs2_gl_hash_clear - Empty out the glock hash table
1719  * @sdp: the filesystem
1720  * @wait: wait until it's all gone
1721  *
1722  * Called when unmounting the filesystem, or when inter-node lock manager
1723  * requests DROPLOCKS because it is running out of capacity.
1724  */
1725
1726 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp, int wait)
1727 {
1728         unsigned long t;
1729         unsigned int x;
1730         int cont;
1731
1732         t = jiffies;
1733
1734         for (;;) {
1735                 cont = 0;
1736                 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1737                         if (examine_bucket(clear_glock, sdp, x))
1738                                 cont = 1;
1739                 }
1740
1741                 if (!wait || !cont)
1742                         break;
1743
1744                 if (time_after_eq(jiffies,
1745                                   t + gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
1746                         fs_warn(sdp, "Unmount seems to be stalled. "
1747                                      "Dumping lock state...\n");
1748                         gfs2_dump_lockstate(sdp);
1749                         t = jiffies;
1750                 }
1751
1752                 down_write(&gfs2_umount_flush_sem);
1753                 invalidate_inodes(sdp->sd_vfs);
1754                 up_write(&gfs2_umount_flush_sem);
1755                 msleep(10);
1756         }
1757 }
1758
1759 /*
1760  *  Diagnostic routines to help debug distributed deadlock
1761  */
1762
1763 static void gfs2_print_symbol(struct glock_iter *gi, const char *fmt,
1764                               unsigned long address)
1765 {
1766 /* when sprint_symbol becomes available in the new kernel, replace this */
1767 /* function with:
1768         char buffer[KSYM_SYMBOL_LEN];
1769
1770         sprint_symbol(buffer, address);
1771         print_dbg(gi, fmt, buffer);
1772 */
1773         if (gi)
1774                 print_dbg(gi, fmt, address);
1775         else
1776                 print_symbol(fmt, address);
1777 }
1778
1779 /**
1780  * dump_holder - print information about a glock holder
1781  * @str: a string naming the type of holder
1782  * @gh: the glock holder
1783  *
1784  * Returns: 0 on success, -ENOBUFS when we run out of space
1785  */
1786
1787 static int dump_holder(struct glock_iter *gi, char *str,
1788                        struct gfs2_holder *gh)
1789 {
1790         unsigned int x;
1791         struct task_struct *gh_owner;
1792
1793         print_dbg(gi, "  %s\n", str);
1794         if (gh->gh_owner_pid) {
1795                 print_dbg(gi, "    owner = %ld ", (long)gh->gh_owner_pid);
1796                 gh_owner = find_task_by_pid(gh->gh_owner_pid);
1797                 if (gh_owner)
1798                         print_dbg(gi, "(%s)\n", gh_owner->comm);
1799                 else
1800                         print_dbg(gi, "(ended)\n");
1801         } else
1802                 print_dbg(gi, "    owner = -1\n");
1803         print_dbg(gi, "    gh_state = %u\n", gh->gh_state);
1804         print_dbg(gi, "    gh_flags =");
1805         for (x = 0; x < 32; x++)
1806                 if (gh->gh_flags & (1 << x))
1807                         print_dbg(gi, " %u", x);
1808         print_dbg(gi, " \n");
1809         print_dbg(gi, "    error = %d\n", gh->gh_error);
1810         print_dbg(gi, "    gh_iflags =");
1811         for (x = 0; x < 32; x++)
1812                 if (test_bit(x, &gh->gh_iflags))
1813                         print_dbg(gi, " %u", x);
1814         print_dbg(gi, " \n");
1815         gfs2_print_symbol(gi, "    initialized at: %s\n", gh->gh_ip);
1816
1817         return 0;
1818 }
1819
1820 /**
1821  * dump_inode - print information about an inode
1822  * @ip: the inode
1823  *
1824  * Returns: 0 on success, -ENOBUFS when we run out of space
1825  */
1826
1827 static int dump_inode(struct glock_iter *gi, struct gfs2_inode *ip)
1828 {
1829         unsigned int x;
1830
1831         print_dbg(gi, "  Inode:\n");
1832         print_dbg(gi, "    num = %llu/%llu\n",
1833                     ip->i_num.no_formal_ino, ip->i_num.no_addr);
1834         print_dbg(gi, "    type = %u\n", IF2DT(ip->i_inode.i_mode));
1835         print_dbg(gi, "    i_flags =");
1836         for (x = 0; x < 32; x++)
1837                 if (test_bit(x, &ip->i_flags))
1838                         print_dbg(gi, " %u", x);
1839         print_dbg(gi, " \n");
1840         return 0;
1841 }
1842
1843 /**
1844  * dump_glock - print information about a glock
1845  * @gl: the glock
1846  * @count: where we are in the buffer
1847  *
1848  * Returns: 0 on success, -ENOBUFS when we run out of space
1849  */
1850
1851 static int dump_glock(struct glock_iter *gi, struct gfs2_glock *gl)
1852 {
1853         struct gfs2_holder *gh;
1854         unsigned int x;
1855         int error = -ENOBUFS;
1856         struct task_struct *gl_owner;
1857
1858         spin_lock(&gl->gl_spin);
1859
1860         print_dbg(gi, "Glock 0x%p (%u, %llu)\n", gl, gl->gl_name.ln_type,
1861                    (unsigned long long)gl->gl_name.ln_number);
1862         print_dbg(gi, "  gl_flags =");
1863         for (x = 0; x < 32; x++) {
1864                 if (test_bit(x, &gl->gl_flags))
1865                         print_dbg(gi, " %u", x);
1866         }
1867         if (!test_bit(GLF_LOCK, &gl->gl_flags))
1868                 print_dbg(gi, " (unlocked)");
1869         print_dbg(gi, " \n");
1870         print_dbg(gi, "  gl_ref = %d\n", atomic_read(&gl->gl_ref));
1871         print_dbg(gi, "  gl_state = %u\n", gl->gl_state);
1872         if (gl->gl_owner_pid) {
1873                 gl_owner = find_task_by_pid(gl->gl_owner_pid);
1874                 if (gl_owner)
1875                         print_dbg(gi, "  gl_owner = pid %d (%s)\n",
1876                                   gl->gl_owner_pid, gl_owner->comm);
1877                 else
1878                         print_dbg(gi, "  gl_owner = %d (ended)\n",
1879                                   gl->gl_owner_pid);
1880         } else
1881                 print_dbg(gi, "  gl_owner = -1\n");
1882         print_dbg(gi, "  gl_ip = %lu\n", gl->gl_ip);
1883         print_dbg(gi, "  req_gh = %s\n", (gl->gl_req_gh) ? "yes" : "no");
1884         print_dbg(gi, "  req_bh = %s\n", (gl->gl_req_bh) ? "yes" : "no");
1885         print_dbg(gi, "  lvb_count = %d\n", atomic_read(&gl->gl_lvb_count));
1886         print_dbg(gi, "  object = %s\n", (gl->gl_object) ? "yes" : "no");
1887         print_dbg(gi, "  le = %s\n",
1888                    (list_empty(&gl->gl_le.le_list)) ? "no" : "yes");
1889         print_dbg(gi, "  reclaim = %s\n",
1890                    (list_empty(&gl->gl_reclaim)) ? "no" : "yes");
1891         if (gl->gl_aspace)
1892                 print_dbg(gi, "  aspace = 0x%p nrpages = %lu\n", gl->gl_aspace,
1893                            gl->gl_aspace->i_mapping->nrpages);
1894         else
1895                 print_dbg(gi, "  aspace = no\n");
1896         print_dbg(gi, "  ail = %d\n", atomic_read(&gl->gl_ail_count));
1897         if (gl->gl_req_gh) {
1898                 error = dump_holder(gi, "Request", gl->gl_req_gh);
1899                 if (error)
1900                         goto out;
1901         }
1902         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1903                 error = dump_holder(gi, "Holder", gh);
1904                 if (error)
1905                         goto out;
1906         }
1907         list_for_each_entry(gh, &gl->gl_waiters1, gh_list) {
1908                 error = dump_holder(gi, "Waiter1", gh);
1909                 if (error)
1910                         goto out;
1911         }
1912         list_for_each_entry(gh, &gl->gl_waiters3, gh_list) {
1913                 error = dump_holder(gi, "Waiter3", gh);
1914                 if (error)
1915                         goto out;
1916         }
1917         if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
1918                 print_dbg(gi, "  Demotion req to state %u (%llu uS ago)\n",
1919                           gl->gl_demote_state,
1920                           (u64)(jiffies - gl->gl_demote_time)*(1000000/HZ));
1921         }
1922         if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object) {
1923                 if (!test_bit(GLF_LOCK, &gl->gl_flags) &&
1924                         list_empty(&gl->gl_holders)) {
1925                         error = dump_inode(gi, gl->gl_object);
1926                         if (error)
1927                                 goto out;
1928                 } else {
1929                         error = -ENOBUFS;
1930                         print_dbg(gi, "  Inode: busy\n");
1931                 }
1932         }
1933
1934         error = 0;
1935
1936 out:
1937         spin_unlock(&gl->gl_spin);
1938         return error;
1939 }
1940
1941 /**
1942  * gfs2_dump_lockstate - print out the current lockstate
1943  * @sdp: the filesystem
1944  * @ub: the buffer to copy the information into
1945  *
1946  * If @ub is NULL, dump the lockstate to the console.
1947  *
1948  */
1949
1950 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
1951 {
1952         struct gfs2_glock *gl;
1953         struct hlist_node *h;
1954         unsigned int x;
1955         int error = 0;
1956
1957         for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1958
1959                 read_lock(gl_lock_addr(x));
1960
1961                 hlist_for_each_entry(gl, h, &gl_hash_table[x].hb_list, gl_list) {
1962                         if (gl->gl_sbd != sdp)
1963                                 continue;
1964
1965                         error = dump_glock(NULL, gl);
1966                         if (error)
1967                                 break;
1968                 }
1969
1970                 read_unlock(gl_lock_addr(x));
1971
1972                 if (error)
1973                         break;
1974         }
1975
1976
1977         return error;
1978 }
1979
1980 int __init gfs2_glock_init(void)
1981 {
1982         unsigned i;
1983         for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
1984                 INIT_HLIST_HEAD(&gl_hash_table[i].hb_list);
1985         }
1986 #ifdef GL_HASH_LOCK_SZ
1987         for(i = 0; i < GL_HASH_LOCK_SZ; i++) {
1988                 rwlock_init(&gl_hash_locks[i]);
1989         }
1990 #endif
1991         return 0;
1992 }
1993
1994 static int gfs2_glock_iter_next(struct glock_iter *gi)
1995 {
1996         while (1) {
1997                 if (!gi->hb_list) {  /* If we don't have a hash bucket yet */
1998                         gi->hb_list = &gl_hash_table[gi->hash].hb_list;
1999                         if (hlist_empty(gi->hb_list)) {
2000                                 gi->hash++;
2001                                 gi->hb_list = NULL;
2002                                 if (gi->hash >= GFS2_GL_HASH_SIZE)
2003                                         return 1;
2004                                 else
2005                                         continue;
2006                         }
2007                         if (!hlist_empty(gi->hb_list)) {
2008                                 gi->gl = list_entry(gi->hb_list->first,
2009                                                     struct gfs2_glock,
2010                                                     gl_list);
2011                         }
2012                 } else {
2013                         if (gi->gl->gl_list.next == NULL) {
2014                                 gi->hash++;
2015                                 gi->hb_list = NULL;
2016                                 continue;
2017                         }
2018                         gi->gl = list_entry(gi->gl->gl_list.next,
2019                                             struct gfs2_glock, gl_list);
2020                 }
2021                 if (gi->gl)
2022                         break;
2023         }
2024         return 0;
2025 }
2026
2027 static void gfs2_glock_iter_free(struct glock_iter *gi)
2028 {
2029         kfree(gi);
2030 }
2031
2032 static struct glock_iter *gfs2_glock_iter_init(struct gfs2_sbd *sdp)
2033 {
2034         struct glock_iter *gi;
2035
2036         gi = kmalloc(sizeof (*gi), GFP_KERNEL);
2037         if (!gi)
2038                 return NULL;
2039
2040         gi->sdp = sdp;
2041         gi->hash = 0;
2042         gi->gl = NULL;
2043         gi->hb_list = NULL;
2044         gi->seq = NULL;
2045         memset(gi->string, 0, sizeof(gi->string));
2046
2047         if (gfs2_glock_iter_next(gi)) {
2048                 gfs2_glock_iter_free(gi);
2049                 return NULL;
2050         }
2051
2052         return gi;
2053 }
2054
2055 static void *gfs2_glock_seq_start(struct seq_file *file, loff_t *pos)
2056 {
2057         struct glock_iter *gi;
2058         loff_t n = *pos;
2059
2060         gi = gfs2_glock_iter_init(file->private);
2061         if (!gi)
2062                 return NULL;
2063
2064         while (n--) {
2065                 if (gfs2_glock_iter_next(gi)) {
2066                         gfs2_glock_iter_free(gi);
2067                         return NULL;
2068                 }
2069         }
2070
2071         return gi;
2072 }
2073
2074 static void *gfs2_glock_seq_next(struct seq_file *file, void *iter_ptr,
2075                                  loff_t *pos)
2076 {
2077         struct glock_iter *gi = iter_ptr;
2078
2079         (*pos)++;
2080
2081         if (gfs2_glock_iter_next(gi)) {
2082                 gfs2_glock_iter_free(gi);
2083                 return NULL;
2084         }
2085
2086         return gi;
2087 }
2088
2089 static void gfs2_glock_seq_stop(struct seq_file *file, void *iter_ptr)
2090 {
2091         /* nothing for now */
2092 }
2093
2094 static int gfs2_glock_seq_show(struct seq_file *file, void *iter_ptr)
2095 {
2096         struct glock_iter *gi = iter_ptr;
2097
2098         gi->seq = file;
2099         dump_glock(gi, gi->gl);
2100
2101         return 0;
2102 }
2103
2104 static struct seq_operations gfs2_glock_seq_ops = {
2105         .start = gfs2_glock_seq_start,
2106         .next  = gfs2_glock_seq_next,
2107         .stop  = gfs2_glock_seq_stop,
2108         .show  = gfs2_glock_seq_show,
2109 };
2110
2111 static int gfs2_debugfs_open(struct inode *inode, struct file *file)
2112 {
2113         struct seq_file *seq;
2114         int ret;
2115
2116         ret = seq_open(file, &gfs2_glock_seq_ops);
2117         if (ret)
2118                 return ret;
2119
2120         seq = file->private_data;
2121         seq->private = inode->i_private;
2122
2123         return 0;
2124 }
2125
2126 static const struct file_operations gfs2_debug_fops = {
2127         .owner   = THIS_MODULE,
2128         .open    = gfs2_debugfs_open,
2129         .read    = seq_read,
2130         .llseek  = seq_lseek,
2131         .release = seq_release
2132 };
2133
2134 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2135 {
2136         sdp->debugfs_dentry = debugfs_create_file(sdp->sd_table_name,
2137                                                   S_IFREG | S_IRUGO,
2138                                                   gfs2_root, sdp,
2139                                                   &gfs2_debug_fops);
2140         if (!sdp->debugfs_dentry)
2141                 return -ENOMEM;
2142
2143         return 0;
2144 }
2145
2146 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2147 {
2148         if (sdp && sdp->debugfs_dentry)
2149                 debugfs_remove(sdp->debugfs_dentry);
2150 }
2151
2152 int gfs2_register_debugfs(void)
2153 {
2154         gfs2_root = debugfs_create_dir("gfs2", NULL);
2155         return gfs2_root ? 0 : -ENOMEM;
2156 }
2157
2158 void gfs2_unregister_debugfs(void)
2159 {
2160         debugfs_remove(gfs2_root);
2161 }