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/kthread.h>
  29 #include <linux/freezer.h>
  30 #include <linux/workqueue.h>
  31 #include <linux/jiffies.h>
  32
  33 #include "gfs2.h"
  34 #include "incore.h"
  35 #include "glock.h"
  36 #include "glops.h"
  37 #include "inode.h"
  38 #include "lm.h"
  39 #include "lops.h"
  40 #include "meta_io.h"
  41 #include "quota.h"
  42 #include "super.h"
  43 #include "util.h"
  44
  45 struct gfs2_gl_hash_bucket {
  46         struct hlist_head hb_list;
  47 };
  48
  49 struct glock_iter {
  50         int hash;                     /* hash bucket index         */
  51         struct gfs2_sbd *sdp;         /* incore superblock         */
  52         struct gfs2_glock *gl;        /* current glock struct      */
  53         struct seq_file *seq;         /* sequence file for debugfs */
  54         char string[512];             /* scratch space             */
  55 };
  56
  57 typedef void (*glock_examiner) (struct gfs2_glock * gl);
  58
  59 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp);
  60 static int dump_glock(struct glock_iter *gi, struct gfs2_glock *gl);
  61 static void gfs2_glock_xmote_th(struct gfs2_glock *gl, struct gfs2_holder *gh);
  62 static void gfs2_glock_drop_th(struct gfs2_glock *gl);
  63 static void run_queue(struct gfs2_glock *gl);
  64
  65 static DECLARE_RWSEM(gfs2_umount_flush_sem);
  66 static struct dentry *gfs2_root;
  67 static struct task_struct *scand_process;
  68 static unsigned int scand_secs = 5;
  69 static struct workqueue_struct *glock_workqueue;
  70
  71 #define GFS2_GL_HASH_SHIFT      15
  72 #define GFS2_GL_HASH_SIZE       (1 << GFS2_GL_HASH_SHIFT)
  73 #define GFS2_GL_HASH_MASK       (GFS2_GL_HASH_SIZE - 1)
  74
  75 static struct gfs2_gl_hash_bucket gl_hash_table[GFS2_GL_HASH_SIZE];
  76 static struct dentry *gfs2_root;
  77
  78 /*
  79  * Despite what you might think, the numbers below are not arbitrary :-)
  80  * They are taken from the ipv4 routing hash code, which is well tested
  81  * and thus should be nearly optimal. Later on we might tweek the numbers
  82  * but for now this should be fine.
  83  *
  84  * The reason for putting the locks in a separate array from the list heads
  85  * is that we can have fewer locks than list heads and save memory. We use
  86  * the same hash function for both, but with a different hash mask.
  87  */
  88 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
  89         defined(CONFIG_PROVE_LOCKING)
  90
  91 #ifdef CONFIG_LOCKDEP
  92 # define GL_HASH_LOCK_SZ        256
  93 #else
  94 # if NR_CPUS >= 32
  95 #  define GL_HASH_LOCK_SZ       4096
  96 # elif NR_CPUS >= 16
  97 #  define GL_HASH_LOCK_SZ       2048
  98 # elif NR_CPUS >= 8
  99 #  define GL_HASH_LOCK_SZ       1024
 100 # elif NR_CPUS >= 4
 101 #  define GL_HASH_LOCK_SZ       512
 102 # else
 103 #  define GL_HASH_LOCK_SZ       256
 104 # endif
 105 #endif
 106
 107 /* We never want more locks than chains */
 108 #if GFS2_GL_HASH_SIZE < GL_HASH_LOCK_SZ
 109 # undef GL_HASH_LOCK_SZ
 110 # define GL_HASH_LOCK_SZ GFS2_GL_HASH_SIZE
 111 #endif
 112
 113 static rwlock_t gl_hash_locks[GL_HASH_LOCK_SZ];
 114
 115 static inline rwlock_t *gl_lock_addr(unsigned int x)
 116 {
 117         return &gl_hash_locks[x & (GL_HASH_LOCK_SZ-1)];
 118 }
 119 #else /* not SMP, so no spinlocks required */
 120 static inline rwlock_t *gl_lock_addr(unsigned int x)
 121 {
 122         return NULL;
 123 }
 124 #endif
 125
 126 /**
 127  * relaxed_state_ok - is a requested lock compatible with the current lock mode?
 128  * @actual: the current state of the lock
 129  * @requested: the lock state that was requested by the caller
 130  * @flags: the modifier flags passed in by the caller
 131  *
 132  * Returns: 1 if the locks are compatible, 0 otherwise
 133  */
 134
 135 static inline int relaxed_state_ok(unsigned int actual, unsigned requested,
 136                                    int flags)
 137 {
 138         if (actual == requested)
 139                 return 1;
 140
 141         if (flags & GL_EXACT)
 142                 return 0;
 143
 144         if (actual == LM_ST_EXCLUSIVE && requested == LM_ST_SHARED)
 145                 return 1;
 146
 147         if (actual != LM_ST_UNLOCKED && (flags & LM_FLAG_ANY))
 148                 return 1;
 149
 150         return 0;
 151 }
 152
 153 /**
 154  * gl_hash() - Turn glock number into hash bucket number
 155  * @lock: The glock number
 156  *
 157  * Returns: The number of the corresponding hash bucket
 158  */
 159
 160 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
 161                             const struct lm_lockname *name)
 162 {
 163         unsigned int h;
 164
 165         h = jhash(&name->ln_number, sizeof(u64), 0);
 166         h = jhash(&name->ln_type, sizeof(unsigned int), h);
 167         h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
 168         h &= GFS2_GL_HASH_MASK;
 169
 170         return h;
 171 }
 172
 173 /**
 174  * glock_free() - Perform a few checks and then release struct gfs2_glock
 175  * @gl: The glock to release
 176  *
 177  * Also calls lock module to release its internal structure for this glock.
 178  *
 179  */
 180
 181 static void glock_free(struct gfs2_glock *gl)
 182 {
 183         struct gfs2_sbd *sdp = gl->gl_sbd;
 184         struct inode *aspace = gl->gl_aspace;
 185
 186         gfs2_lm_put_lock(sdp, gl->gl_lock);
 187
 188         if (aspace)
 189                 gfs2_aspace_put(aspace);
 190
 191         kmem_cache_free(gfs2_glock_cachep, gl);
 192 }
 193
 194 /**
 195  * gfs2_glock_hold() - increment reference count on glock
 196  * @gl: The glock to hold
 197  *
 198  */
 199
 200 void gfs2_glock_hold(struct gfs2_glock *gl)
 201 {
 202         atomic_inc(&gl->gl_ref);
 203 }
 204
 205 /**
 206  * gfs2_glock_put() - Decrement reference count on glock
 207  * @gl: The glock to put
 208  *
 209  */
 210
 211 int gfs2_glock_put(struct gfs2_glock *gl)
 212 {
 213         int rv = 0;
 214         struct gfs2_sbd *sdp = gl->gl_sbd;
 215
 216         write_lock(gl_lock_addr(gl->gl_hash));
 217         if (atomic_dec_and_test(&gl->gl_ref)) {
 218                 hlist_del(&gl->gl_list);
 219                 write_unlock(gl_lock_addr(gl->gl_hash));
 220                 BUG_ON(spin_is_locked(&gl->gl_spin));
 221                 gfs2_assert(sdp, gl->gl_state == LM_ST_UNLOCKED);
 222                 gfs2_assert(sdp, list_empty(&gl->gl_reclaim));
 223                 gfs2_assert(sdp, list_empty(&gl->gl_holders));
 224                 gfs2_assert(sdp, list_empty(&gl->gl_waiters1));
 225                 gfs2_assert(sdp, list_empty(&gl->gl_waiters3));
 226                 glock_free(gl);
 227                 rv = 1;
 228                 goto out;
 229         }
 230         write_unlock(gl_lock_addr(gl->gl_hash));
 231 out:
 232         return rv;
 233 }
 234
 235 /**
 236  * search_bucket() - Find struct gfs2_glock by lock number
 237  * @bucket: the bucket to search
 238  * @name: The lock name
 239  *
 240  * Returns: NULL, or the struct gfs2_glock with the requested number
 241  */
 242
 243 static struct gfs2_glock *search_bucket(unsigned int hash,
 244                                         const struct gfs2_sbd *sdp,
 245                                         const struct lm_lockname *name)
 246 {
 247         struct gfs2_glock *gl;
 248         struct hlist_node *h;
 249
 250         hlist_for_each_entry(gl, h, &gl_hash_table[hash].hb_list, gl_list) {
 251                 if (!lm_name_equal(&gl->gl_name, name))
 252                         continue;
 253                 if (gl->gl_sbd != sdp)
 254                         continue;
 255
 256                 atomic_inc(&gl->gl_ref);
 257
 258                 return gl;
 259         }
 260
 261         return NULL;
 262 }
 263
 264 /**
 265  * gfs2_glock_find() - Find glock by lock number
 266  * @sdp: The GFS2 superblock
 267  * @name: The lock name
 268  *
 269  * Returns: NULL, or the struct gfs2_glock with the requested number
 270  */
 271
 272 static struct gfs2_glock *gfs2_glock_find(const struct gfs2_sbd *sdp,
 273                                           const struct lm_lockname *name)
 274 {
 275         unsigned int hash = gl_hash(sdp, name);
 276         struct gfs2_glock *gl;
 277
 278         read_lock(gl_lock_addr(hash));
 279         gl = search_bucket(hash, sdp, name);
 280         read_unlock(gl_lock_addr(hash));
 281
 282         return gl;
 283 }
 284
 285 static void glock_work_func(struct work_struct *work)
 286 {
 287         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
 288
 289         spin_lock(&gl->gl_spin);
 290         if (test_and_clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags))
 291                 set_bit(GLF_DEMOTE, &gl->gl_flags);
 292         run_queue(gl);
 293         spin_unlock(&gl->gl_spin);
 294         gfs2_glock_put(gl);
 295 }
 296
 297 /**
 298  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
 299  * @sdp: The GFS2 superblock
 300  * @number: the lock number
 301  * @glops: The glock_operations to use
 302  * @create: If 0, don't create the glock if it doesn't exist
 303  * @glp: the glock is returned here
 304  *
 305  * This does not lock a glock, just finds/creates structures for one.
 306  *
 307  * Returns: errno
 308  */
 309
 310 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
 311                    const struct gfs2_glock_operations *glops, int create,
 312                    struct gfs2_glock **glp)
 313 {
 314         struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
 315         struct gfs2_glock *gl, *tmp;
 316         unsigned int hash = gl_hash(sdp, &name);
 317         int error;
 318
 319         read_lock(gl_lock_addr(hash));
 320         gl = search_bucket(hash, sdp, &name);
 321         read_unlock(gl_lock_addr(hash));
 322
 323         if (gl || !create) {
 324                 *glp = gl;
 325                 return 0;
 326         }
 327
 328         gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
 329         if (!gl)
 330                 return -ENOMEM;
 331
 332         gl->gl_flags = 0;
 333         gl->gl_name = name;
 334         atomic_set(&gl->gl_ref, 1);
 335         gl->gl_state = LM_ST_UNLOCKED;
 336         gl->gl_demote_state = LM_ST_EXCLUSIVE;
 337         gl->gl_hash = hash;
 338         gl->gl_owner_pid = 0;
 339         gl->gl_ip = 0;
 340         gl->gl_ops = glops;
 341         gl->gl_req_gh = NULL;
 342         gl->gl_req_bh = NULL;
 343         gl->gl_vn = 0;
 344         gl->gl_stamp = jiffies;
 345         gl->gl_tchange = jiffies;
 346         gl->gl_object = NULL;
 347         gl->gl_sbd = sdp;
 348         gl->gl_aspace = NULL;
 349         lops_init_le(&gl->gl_le, &gfs2_glock_lops);
 350         INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
 351
 352         /* If this glock protects actual on-disk data or metadata blocks,
 353            create a VFS inode to manage the pages/buffers holding them. */
 354         if (glops == &gfs2_inode_glops || glops == &gfs2_rgrp_glops) {
 355                 gl->gl_aspace = gfs2_aspace_get(sdp);
 356                 if (!gl->gl_aspace) {
 357                         error = -ENOMEM;
 358                         goto fail;
 359                 }
 360         }
 361
 362         error = gfs2_lm_get_lock(sdp, &name, &gl->gl_lock);
 363         if (error)
 364                 goto fail_aspace;
 365
 366         write_lock(gl_lock_addr(hash));
 367         tmp = search_bucket(hash, sdp, &name);
 368         if (tmp) {
 369                 write_unlock(gl_lock_addr(hash));
 370                 glock_free(gl);
 371                 gl = tmp;
 372         } else {
 373                 hlist_add_head(&gl->gl_list, &gl_hash_table[hash].hb_list);
 374                 write_unlock(gl_lock_addr(hash));
 375         }
 376
 377         *glp = gl;
 378
 379         return 0;
 380
 381 fail_aspace:
 382         if (gl->gl_aspace)
 383                 gfs2_aspace_put(gl->gl_aspace);
 384 fail:
 385         kmem_cache_free(gfs2_glock_cachep, gl);
 386         return error;
 387 }
 388
 389 /**
 390  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
 391  * @gl: the glock
 392  * @state: the state we're requesting
 393  * @flags: the modifier flags
 394  * @gh: the holder structure
 395  *
 396  */
 397
 398 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
 399                       struct gfs2_holder *gh)
 400 {
 401         INIT_LIST_HEAD(&gh->gh_list);
 402         gh->gh_gl = gl;
 403         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 404         gh->gh_owner_pid = current->pid;
 405         gh->gh_state = state;
 406         gh->gh_flags = flags;
 407         gh->gh_error = 0;
 408         gh->gh_iflags = 0;
 409         gfs2_glock_hold(gl);
 410 }
 411
 412 /**
 413  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
 414  * @state: the state we're requesting
 415  * @flags: the modifier flags
 416  * @gh: the holder structure
 417  *
 418  * Don't mess with the glock.
 419  *
 420  */
 421
 422 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
 423 {
 424         gh->gh_state = state;
 425         gh->gh_flags = flags;
 426         gh->gh_iflags = 0;
 427         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 428 }
 429
 430 /**
 431  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
 432  * @gh: the holder structure
 433  *
 434  */
 435
 436 void gfs2_holder_uninit(struct gfs2_holder *gh)
 437 {
 438         gfs2_glock_put(gh->gh_gl);
 439         gh->gh_gl = NULL;
 440         gh->gh_ip = 0;
 441 }
 442
 443 static void gfs2_holder_wake(struct gfs2_holder *gh)
 444 {
 445         clear_bit(HIF_WAIT, &gh->gh_iflags);
 446         smp_mb__after_clear_bit();
 447         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
 448 }
 449
 450 static int just_schedule(void *word)
 451 {
 452         schedule();
 453         return 0;
 454 }
 455
 456 static void wait_on_holder(struct gfs2_holder *gh)
 457 {
 458         might_sleep();
 459         wait_on_bit(&gh->gh_iflags, HIF_WAIT, just_schedule, TASK_UNINTERRUPTIBLE);
 460 }
 461
 462 static void gfs2_demote_wake(struct gfs2_glock *gl)
 463 {
 464         BUG_ON(!spin_is_locked(&gl->gl_spin));
 465         gl->gl_demote_state = LM_ST_EXCLUSIVE;
 466         clear_bit(GLF_DEMOTE, &gl->gl_flags);
 467         smp_mb__after_clear_bit();
 468         wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
 469 }
 470
 471 static void wait_on_demote(struct gfs2_glock *gl)
 472 {
 473         might_sleep();
 474         wait_on_bit(&gl->gl_flags, GLF_DEMOTE, just_schedule, TASK_UNINTERRUPTIBLE);
 475 }
 476
 477 /**
 478  * rq_mutex - process a mutex request in the queue
 479  * @gh: the glock holder
 480  *
 481  * Returns: 1 if the queue is blocked
 482  */
 483
 484 static int rq_mutex(struct gfs2_holder *gh)
 485 {
 486         struct gfs2_glock *gl = gh->gh_gl;
 487
 488         list_del_init(&gh->gh_list);
 489         /*  gh->gh_error never examined.  */
 490         set_bit(GLF_LOCK, &gl->gl_flags);
 491         clear_bit(HIF_WAIT, &gh->gh_iflags);
 492         smp_mb();
 493         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
 494
 495         return 1;
 496 }
 497
 498 /**
 499  * rq_promote - process a promote request in the queue
 500  * @gh: the glock holder
 501  *
 502  * Acquire a new inter-node lock, or change a lock state to more restrictive.
 503  *
 504  * Returns: 1 if the queue is blocked
 505  */
 506
 507 static int rq_promote(struct gfs2_holder *gh)
 508 {
 509         struct gfs2_glock *gl = gh->gh_gl;
 510
 511         if (!relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
 512                 if (list_empty(&gl->gl_holders)) {
 513                         gl->gl_req_gh = gh;
 514                         set_bit(GLF_LOCK, &gl->gl_flags);
 515                         spin_unlock(&gl->gl_spin);
 516                         gfs2_glock_xmote_th(gh->gh_gl, gh);
 517                         spin_lock(&gl->gl_spin);
 518                 }
 519                 return 1;
 520         }
 521
 522         if (list_empty(&gl->gl_holders)) {
 523                 set_bit(HIF_FIRST, &gh->gh_iflags);
 524                 set_bit(GLF_LOCK, &gl->gl_flags);
 525         } else {
 526                 struct gfs2_holder *next_gh;
 527                 if (gh->gh_state == LM_ST_EXCLUSIVE)
 528                         return 1;
 529                 next_gh = list_entry(gl->gl_holders.next, struct gfs2_holder,
 530                                      gh_list);
 531                 if (next_gh->gh_state == LM_ST_EXCLUSIVE)
 532                          return 1;
 533         }
 534
 535         list_move_tail(&gh->gh_list, &gl->gl_holders);
 536         gh->gh_error = 0;
 537         set_bit(HIF_HOLDER, &gh->gh_iflags);
 538
 539         gfs2_holder_wake(gh);
 540
 541         return 0;
 542 }
 543
 544 /**
 545  * rq_demote - process a demote request in the queue
 546  * @gh: the glock holder
 547  *
 548  * Returns: 1 if the queue is blocked
 549  */
 550
 551 static int rq_demote(struct gfs2_glock *gl)
 552 {
 553         if (!list_empty(&gl->gl_holders))
 554                 return 1;
 555
 556         if (gl->gl_state == gl->gl_demote_state ||
 557             gl->gl_state == LM_ST_UNLOCKED) {
 558                 gfs2_demote_wake(gl);
 559                 return 0;
 560         }
 561
 562         set_bit(GLF_LOCK, &gl->gl_flags);
 563         set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
 564
 565         if (gl->gl_demote_state == LM_ST_UNLOCKED ||
 566             gl->gl_state != LM_ST_EXCLUSIVE) {
 567                 spin_unlock(&gl->gl_spin);
 568                 gfs2_glock_drop_th(gl);
 569         } else {
 570                 spin_unlock(&gl->gl_spin);
 571                 gfs2_glock_xmote_th(gl, NULL);
 572         }
 573
 574         spin_lock(&gl->gl_spin);
 575         clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
 576
 577         return 0;
 578 }
 579
 580 /**
 581  * run_queue - process holder structures on a glock
 582  * @gl: the glock
 583  *
 584  */
 585 static void run_queue(struct gfs2_glock *gl)
 586 {
 587         struct gfs2_holder *gh;
 588         int blocked = 1;
 589
 590         for (;;) {
 591                 if (test_bit(GLF_LOCK, &gl->gl_flags))
 592                         break;
 593
 594                 if (!list_empty(&gl->gl_waiters1)) {
 595                         gh = list_entry(gl->gl_waiters1.next,
 596                                         struct gfs2_holder, gh_list);
 597
 598                         if (test_bit(HIF_MUTEX, &gh->gh_iflags))
 599                                 blocked = rq_mutex(gh);
 600                         else
 601                                 gfs2_assert_warn(gl->gl_sbd, 0);
 602
 603                 } else if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
 604                         blocked = rq_demote(gl);
 605                         if (gl->gl_waiters2 && !blocked) {
 606                                 set_bit(GLF_DEMOTE, &gl->gl_flags);
 607                                 gl->gl_demote_state = LM_ST_UNLOCKED;
 608                         }
 609                         gl->gl_waiters2 = 0;
 610                 } else if (!list_empty(&gl->gl_waiters3)) {
 611                         gh = list_entry(gl->gl_waiters3.next,
 612                                         struct gfs2_holder, gh_list);
 613
 614                         if (test_bit(HIF_PROMOTE, &gh->gh_iflags))
 615                                 blocked = rq_promote(gh);
 616                         else
 617                                 gfs2_assert_warn(gl->gl_sbd, 0);
 618
 619                 } else
 620                         break;
 621
 622                 if (blocked)
 623                         break;
 624         }
 625 }
 626
 627 /**
 628  * gfs2_glmutex_lock - acquire a local lock on a glock
 629  * @gl: the glock
 630  *
 631  * Gives caller exclusive access to manipulate a glock structure.
 632  */
 633
 634 static void gfs2_glmutex_lock(struct gfs2_glock *gl)
 635 {
 636         struct gfs2_holder gh;
 637
 638         gfs2_holder_init(gl, 0, 0, &gh);
 639         set_bit(HIF_MUTEX, &gh.gh_iflags);
 640         if (test_and_set_bit(HIF_WAIT, &gh.gh_iflags))
 641                 BUG();
 642
 643         spin_lock(&gl->gl_spin);
 644         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
 645                 list_add_tail(&gh.gh_list, &gl->gl_waiters1);
 646         } else {
 647                 gl->gl_owner_pid = current->pid;
 648                 gl->gl_ip = (unsigned long)__builtin_return_address(0);
 649                 clear_bit(HIF_WAIT, &gh.gh_iflags);
 650                 smp_mb();
 651                 wake_up_bit(&gh.gh_iflags, HIF_WAIT);
 652         }
 653         spin_unlock(&gl->gl_spin);
 654
 655         wait_on_holder(&gh);
 656         gfs2_holder_uninit(&gh);
 657 }
 658
 659 /**
 660  * gfs2_glmutex_trylock - try to acquire a local lock on a glock
 661  * @gl: the glock
 662  *
 663  * Returns: 1 if the glock is acquired
 664  */
 665
 666 static int gfs2_glmutex_trylock(struct gfs2_glock *gl)
 667 {
 668         int acquired = 1;
 669
 670         spin_lock(&gl->gl_spin);
 671         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
 672                 acquired = 0;
 673         } else {
 674                 gl->gl_owner_pid = current->pid;
 675                 gl->gl_ip = (unsigned long)__builtin_return_address(0);
 676         }
 677         spin_unlock(&gl->gl_spin);
 678
 679         return acquired;
 680 }
 681
 682 /**
 683  * gfs2_glmutex_unlock - release a local lock on a glock
 684  * @gl: the glock
 685  *
 686  */
 687
 688 static void gfs2_glmutex_unlock(struct gfs2_glock *gl)
 689 {
 690         spin_lock(&gl->gl_spin);
 691         clear_bit(GLF_LOCK, &gl->gl_flags);
 692         gl->gl_owner_pid = 0;
 693         gl->gl_ip = 0;
 694         run_queue(gl);
 695         BUG_ON(!spin_is_locked(&gl->gl_spin));
 696         spin_unlock(&gl->gl_spin);
 697 }
 698
 699 /**
 700  * handle_callback - process a demote request
 701  * @gl: the glock
 702  * @state: the state the caller wants us to change to
 703  *
 704  * There are only two requests that we are going to see in actual
 705  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
 706  */
 707
 708 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
 709                             int remote, unsigned long delay)
 710 {
 711         int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
 712
 713         spin_lock(&gl->gl_spin);
 714         set_bit(bit, &gl->gl_flags);
 715         if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
 716                 gl->gl_demote_state = state;
 717                 gl->gl_demote_time = jiffies;
 718                 if (remote && gl->gl_ops->go_type == LM_TYPE_IOPEN &&
 719                     gl->gl_object) {
 720                         gfs2_glock_schedule_for_reclaim(gl);
 721                         spin_unlock(&gl->gl_spin);
 722                         return;
 723                 }
 724         } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
 725                         gl->gl_demote_state != state) {
 726                 if (test_bit(GLF_DEMOTE_IN_PROGRESS,  &gl->gl_flags))
 727                         gl->gl_waiters2 = 1;
 728                 else
 729                         gl->gl_demote_state = LM_ST_UNLOCKED;
 730         }
 731         spin_unlock(&gl->gl_spin);
 732 }
 733
 734 /**
 735  * state_change - record that the glock is now in a different state
 736  * @gl: the glock
 737  * @new_state the new state
 738  *
 739  */
 740
 741 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
 742 {
 743         int held1, held2;
 744
 745         held1 = (gl->gl_state != LM_ST_UNLOCKED);
 746         held2 = (new_state != LM_ST_UNLOCKED);
 747
 748         if (held1 != held2) {
 749                 if (held2)
 750                         gfs2_glock_hold(gl);
 751                 else
 752                         gfs2_glock_put(gl);
 753         }
 754
 755         gl->gl_state = new_state;
 756         gl->gl_tchange = jiffies;
 757 }
 758
 759 /**
 760  * xmote_bh - Called after the lock module is done acquiring a lock
 761  * @gl: The glock in question
 762  * @ret: the int returned from the lock module
 763  *
 764  */
 765
 766 static void xmote_bh(struct gfs2_glock *gl, unsigned int ret)
 767 {
 768         struct gfs2_sbd *sdp = gl->gl_sbd;
 769         const struct gfs2_glock_operations *glops = gl->gl_ops;
 770         struct gfs2_holder *gh = gl->gl_req_gh;
 771         int prev_state = gl->gl_state;
 772         int op_done = 1;
 773
 774         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
 775         gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
 776         gfs2_assert_warn(sdp, !(ret & LM_OUT_ASYNC));
 777
 778         state_change(gl, ret & LM_OUT_ST_MASK);
 779
 780         if (prev_state != LM_ST_UNLOCKED && !(ret & LM_OUT_CACHEABLE)) {
 781                 if (glops->go_inval)
 782                         glops->go_inval(gl, DIO_METADATA);
 783         } else if (gl->gl_state == LM_ST_DEFERRED) {
 784                 /* We might not want to do this here.
 785                    Look at moving to the inode glops. */
 786                 if (glops->go_inval)
 787                         glops->go_inval(gl, 0);
 788         }
 789
 790         /*  Deal with each possible exit condition  */
 791
 792         if (!gh) {
 793                 gl->gl_stamp = jiffies;
 794                 if (ret & LM_OUT_CANCELED) {
 795                         op_done = 0;
 796                 } else {
 797                         spin_lock(&gl->gl_spin);
 798                         if (gl->gl_state != gl->gl_demote_state) {
 799                                 gl->gl_req_bh = NULL;
 800                                 spin_unlock(&gl->gl_spin);
 801                                 gfs2_glock_drop_th(gl);
 802                                 gfs2_glock_put(gl);
 803                                 return;
 804                         }
 805                         gfs2_demote_wake(gl);
 806                         spin_unlock(&gl->gl_spin);
 807                 }
 808         } else {
 809                 spin_lock(&gl->gl_spin);
 810                 list_del_init(&gh->gh_list);
 811                 gh->gh_error = -EIO;
 812                 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 813                         goto out;
 814                 gh->gh_error = GLR_CANCELED;
 815                 if (ret & LM_OUT_CANCELED)
 816                         goto out;
 817                 if (relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
 818                         list_add_tail(&gh->gh_list, &gl->gl_holders);
 819                         gh->gh_error = 0;
 820                         set_bit(HIF_HOLDER, &gh->gh_iflags);
 821                         set_bit(HIF_FIRST, &gh->gh_iflags);
 822                         op_done = 0;
 823                         goto out;
 824                 }
 825                 gh->gh_error = GLR_TRYFAILED;
 826                 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
 827                         goto out;
 828                 gh->gh_error = -EINVAL;
 829                 if (gfs2_assert_withdraw(sdp, 0) == -1)
 830                         fs_err(sdp, "ret = 0x%.8X\n", ret);
 831 out:
 832                 spin_unlock(&gl->gl_spin);
 833         }
 834
 835         if (glops->go_xmote_bh)
 836                 glops->go_xmote_bh(gl);
 837
 838         if (op_done) {
 839                 spin_lock(&gl->gl_spin);
 840                 gl->gl_req_gh = NULL;
 841                 gl->gl_req_bh = NULL;
 842                 clear_bit(GLF_LOCK, &gl->gl_flags);
 843                 spin_unlock(&gl->gl_spin);
 844         }
 845
 846         gfs2_glock_put(gl);
 847
 848         if (gh)
 849                 gfs2_holder_wake(gh);
 850 }
 851
 852 /**
 853  * gfs2_glock_xmote_th - Call into the lock module to acquire or change a glock
 854  * @gl: The glock in question
 855  * @state: the requested state
 856  * @flags: modifier flags to the lock call
 857  *
 858  */
 859
 860 static void gfs2_glock_xmote_th(struct gfs2_glock *gl, struct gfs2_holder *gh)
 861 {
 862         struct gfs2_sbd *sdp = gl->gl_sbd;
 863         int flags = gh ? gh->gh_flags : 0;
 864         unsigned state = gh ? gh->gh_state : gl->gl_demote_state;
 865         const struct gfs2_glock_operations *glops = gl->gl_ops;
 866         int lck_flags = flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB |
 867                                  LM_FLAG_NOEXP | LM_FLAG_ANY |
 868                                  LM_FLAG_PRIORITY);
 869         unsigned int lck_ret;
 870
 871         if (glops->go_xmote_th)
 872                 glops->go_xmote_th(gl);
 873
 874         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
 875         gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
 876         gfs2_assert_warn(sdp, state != LM_ST_UNLOCKED);
 877         gfs2_assert_warn(sdp, state != gl->gl_state);
 878
 879         gfs2_glock_hold(gl);
 880         gl->gl_req_bh = xmote_bh;
 881
 882         lck_ret = gfs2_lm_lock(sdp, gl->gl_lock, gl->gl_state, state, lck_flags);
 883
 884         if (gfs2_assert_withdraw(sdp, !(lck_ret & LM_OUT_ERROR)))
 885                 return;
 886
 887         if (lck_ret & LM_OUT_ASYNC)
 888                 gfs2_assert_warn(sdp, lck_ret == LM_OUT_ASYNC);
 889         else
 890                 xmote_bh(gl, lck_ret);
 891 }
 892
 893 /**
 894  * drop_bh - Called after a lock module unlock completes
 895  * @gl: the glock
 896  * @ret: the return status
 897  *
 898  * Doesn't wake up the process waiting on the struct gfs2_holder (if any)
 899  * Doesn't drop the reference on the glock the top half took out
 900  *
 901  */
 902
 903 static void drop_bh(struct gfs2_glock *gl, unsigned int ret)
 904 {
 905         struct gfs2_sbd *sdp = gl->gl_sbd;
 906         const struct gfs2_glock_operations *glops = gl->gl_ops;
 907         struct gfs2_holder *gh = gl->gl_req_gh;
 908
 909         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
 910         gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
 911         gfs2_assert_warn(sdp, !ret);
 912
 913         state_change(gl, LM_ST_UNLOCKED);
 914
 915         if (glops->go_inval)
 916                 glops->go_inval(gl, DIO_METADATA);
 917
 918         if (gh) {
 919                 spin_lock(&gl->gl_spin);
 920                 list_del_init(&gh->gh_list);
 921                 gh->gh_error = 0;
 922                 spin_unlock(&gl->gl_spin);
 923         }
 924
 925         spin_lock(&gl->gl_spin);
 926         gfs2_demote_wake(gl);
 927         gl->gl_req_gh = NULL;
 928         gl->gl_req_bh = NULL;
 929         clear_bit(GLF_LOCK, &gl->gl_flags);
 930         spin_unlock(&gl->gl_spin);
 931
 932         gfs2_glock_put(gl);
 933
 934         if (gh)
 935                 gfs2_holder_wake(gh);
 936 }
 937
 938 /**
 939  * gfs2_glock_drop_th - call into the lock module to unlock a lock
 940  * @gl: the glock
 941  *
 942  */
 943
 944 static void gfs2_glock_drop_th(struct gfs2_glock *gl)
 945 {
 946         struct gfs2_sbd *sdp = gl->gl_sbd;
 947         const struct gfs2_glock_operations *glops = gl->gl_ops;
 948         unsigned int ret;
 949
 950         if (glops->go_xmote_th)
 951                 glops->go_xmote_th(gl);
 952
 953         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
 954         gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
 955         gfs2_assert_warn(sdp, gl->gl_state != LM_ST_UNLOCKED);
 956
 957         gfs2_glock_hold(gl);
 958         gl->gl_req_bh = drop_bh;
 959
 960         ret = gfs2_lm_unlock(sdp, gl->gl_lock, gl->gl_state);
 961
 962         if (gfs2_assert_withdraw(sdp, !(ret & LM_OUT_ERROR)))
 963                 return;
 964
 965         if (!ret)
 966                 drop_bh(gl, ret);
 967         else
 968                 gfs2_assert_warn(sdp, ret == LM_OUT_ASYNC);
 969 }
 970
 971 /**
 972  * do_cancels - cancel requests for locks stuck waiting on an expire flag
 973  * @gh: the LM_FLAG_PRIORITY holder waiting to acquire the lock
 974  *
 975  * Don't cancel GL_NOCANCEL requests.
 976  */
 977
 978 static void do_cancels(struct gfs2_holder *gh)
 979 {
 980         struct gfs2_glock *gl = gh->gh_gl;
 981
 982         spin_lock(&gl->gl_spin);
 983
 984         while (gl->gl_req_gh != gh &&
 985                !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
 986                !list_empty(&gh->gh_list)) {
 987                 if (gl->gl_req_bh && !(gl->gl_req_gh &&
 988                                      (gl->gl_req_gh->gh_flags & GL_NOCANCEL))) {
 989                         spin_unlock(&gl->gl_spin);
 990                         gfs2_lm_cancel(gl->gl_sbd, gl->gl_lock);
 991                         msleep(100);
 992                         spin_lock(&gl->gl_spin);
 993                 } else {
 994                         spin_unlock(&gl->gl_spin);
 995                         msleep(100);
 996                         spin_lock(&gl->gl_spin);
 997                 }
 998         }
 999
1000         spin_unlock(&gl->gl_spin);
1001 }
1002
1003 /**
1004  * glock_wait_internal - wait on a glock acquisition
1005  * @gh: the glock holder
1006  *
1007  * Returns: 0 on success
1008  */
1009
1010 static int glock_wait_internal(struct gfs2_holder *gh)
1011 {
1012         struct gfs2_glock *gl = gh->gh_gl;
1013         struct gfs2_sbd *sdp = gl->gl_sbd;
1014         const struct gfs2_glock_operations *glops = gl->gl_ops;
1015
1016         if (test_bit(HIF_ABORTED, &gh->gh_iflags))
1017                 return -EIO;
1018
1019         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1020                 spin_lock(&gl->gl_spin);
1021                 if (gl->gl_req_gh != gh &&
1022                     !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1023                     !list_empty(&gh->gh_list)) {
1024                         list_del_init(&gh->gh_list);
1025                         gh->gh_error = GLR_TRYFAILED;
1026                         run_queue(gl);
1027                         spin_unlock(&gl->gl_spin);
1028                         return gh->gh_error;
1029                 }
1030                 spin_unlock(&gl->gl_spin);
1031         }
1032
1033         if (gh->gh_flags & LM_FLAG_PRIORITY)
1034                 do_cancels(gh);
1035
1036         wait_on_holder(gh);
1037         if (gh->gh_error)
1038                 return gh->gh_error;
1039
1040         gfs2_assert_withdraw(sdp, test_bit(HIF_HOLDER, &gh->gh_iflags));
1041         gfs2_assert_withdraw(sdp, relaxed_state_ok(gl->gl_state, gh->gh_state,
1042                                                    gh->gh_flags));
1043
1044         if (test_bit(HIF_FIRST, &gh->gh_iflags)) {
1045                 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1046
1047                 if (glops->go_lock) {
1048                         gh->gh_error = glops->go_lock(gh);
1049                         if (gh->gh_error) {
1050                                 spin_lock(&gl->gl_spin);
1051                                 list_del_init(&gh->gh_list);
1052                                 spin_unlock(&gl->gl_spin);
1053                         }
1054                 }
1055
1056                 spin_lock(&gl->gl_spin);
1057                 gl->gl_req_gh = NULL;
1058                 gl->gl_req_bh = NULL;
1059                 clear_bit(GLF_LOCK, &gl->gl_flags);
1060                 run_queue(gl);
1061                 spin_unlock(&gl->gl_spin);
1062         }
1063
1064         return gh->gh_error;
1065 }
1066
1067 static inline struct gfs2_holder *
1068 find_holder_by_owner(struct list_head *head, pid_t pid)
1069 {
1070         struct gfs2_holder *gh;
1071
1072         list_for_each_entry(gh, head, gh_list) {
1073                 if (gh->gh_owner_pid == pid)
1074                         return gh;
1075         }
1076
1077         return NULL;
1078 }
1079
1080 static void print_dbg(struct glock_iter *gi, const char *fmt, ...)
1081 {
1082         va_list args;
1083
1084         va_start(args, fmt);
1085         if (gi) {
1086                 vsprintf(gi->string, fmt, args);
1087                 seq_printf(gi->seq, gi->string);
1088         }
1089         else
1090                 vprintk(fmt, args);
1091         va_end(args);
1092 }
1093
1094 /**
1095  * add_to_queue - Add a holder to the wait queue (but look for recursion)
1096  * @gh: the holder structure to add
1097  *
1098  */
1099
1100 static void add_to_queue(struct gfs2_holder *gh)
1101 {
1102         struct gfs2_glock *gl = gh->gh_gl;
1103         struct gfs2_holder *existing;
1104
1105         BUG_ON(!gh->gh_owner_pid);
1106         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1107                 BUG();
1108
1109         if (!(gh->gh_flags & GL_FLOCK)) {
1110                 existing = find_holder_by_owner(&gl->gl_holders,
1111                                                 gh->gh_owner_pid);
1112                 if (existing) {
1113                         print_symbol(KERN_WARNING "original: %s\n",
1114                                      existing->gh_ip);
1115                         printk(KERN_INFO "pid : %d\n", existing->gh_owner_pid);
1116                         printk(KERN_INFO "lock type : %d lock state : %d\n",
1117                                existing->gh_gl->gl_name.ln_type,
1118                                existing->gh_gl->gl_state);
1119                         print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1120                         printk(KERN_INFO "pid : %d\n", gh->gh_owner_pid);
1121                         printk(KERN_INFO "lock type : %d lock state : %d\n",
1122                                gl->gl_name.ln_type, gl->gl_state);
1123                         BUG();
1124                 }
1125
1126                 existing = find_holder_by_owner(&gl->gl_waiters3,
1127                                                 gh->gh_owner_pid);
1128                 if (existing) {
1129                         print_symbol(KERN_WARNING "original: %s\n",
1130                                      existing->gh_ip);
1131                         print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1132                         BUG();
1133                 }
1134         }
1135
1136         if (gh->gh_flags & LM_FLAG_PRIORITY)
1137                 list_add(&gh->gh_list, &gl->gl_waiters3);
1138         else
1139                 list_add_tail(&gh->gh_list, &gl->gl_waiters3);
1140 }
1141
1142 /**
1143  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1144  * @gh: the holder structure
1145  *
1146  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1147  *
1148  * Returns: 0, GLR_TRYFAILED, or errno on failure
1149  */
1150
1151 int gfs2_glock_nq(struct gfs2_holder *gh)
1152 {
1153         struct gfs2_glock *gl = gh->gh_gl;
1154         struct gfs2_sbd *sdp = gl->gl_sbd;
1155         int error = 0;
1156
1157 restart:
1158         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1159                 set_bit(HIF_ABORTED, &gh->gh_iflags);
1160                 return -EIO;
1161         }
1162
1163         set_bit(HIF_PROMOTE, &gh->gh_iflags);
1164
1165         spin_lock(&gl->gl_spin);
1166         add_to_queue(gh);
1167         run_queue(gl);
1168         spin_unlock(&gl->gl_spin);
1169
1170         if (!(gh->gh_flags & GL_ASYNC)) {
1171                 error = glock_wait_internal(gh);
1172                 if (error == GLR_CANCELED) {
1173                         msleep(100);
1174                         goto restart;
1175                 }
1176         }
1177
1178         return error;
1179 }
1180
1181 /**
1182  * gfs2_glock_poll - poll to see if an async request has been completed
1183  * @gh: the holder
1184  *
1185  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1186  */
1187
1188 int gfs2_glock_poll(struct gfs2_holder *gh)
1189 {
1190         struct gfs2_glock *gl = gh->gh_gl;
1191         int ready = 0;
1192
1193         spin_lock(&gl->gl_spin);
1194
1195         if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1196                 ready = 1;
1197         else if (list_empty(&gh->gh_list)) {
1198                 if (gh->gh_error == GLR_CANCELED) {
1199                         spin_unlock(&gl->gl_spin);
1200                         msleep(100);
1201                         if (gfs2_glock_nq(gh))
1202                                 return 1;
1203                         return 0;
1204                 } else
1205                         ready = 1;
1206         }
1207
1208         spin_unlock(&gl->gl_spin);
1209
1210         return ready;
1211 }
1212
1213 /**
1214  * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
1215  * @gh: the holder structure
1216  *
1217  * Returns: 0, GLR_TRYFAILED, or errno on failure
1218  */
1219
1220 int gfs2_glock_wait(struct gfs2_holder *gh)
1221 {
1222         int error;
1223
1224         error = glock_wait_internal(gh);
1225         if (error == GLR_CANCELED) {
1226                 msleep(100);
1227                 gh->gh_flags &= ~GL_ASYNC;
1228                 error = gfs2_glock_nq(gh);
1229         }
1230
1231         return error;
1232 }
1233
1234 /**
1235  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1236  * @gh: the glock holder
1237  *
1238  */
1239
1240 void gfs2_glock_dq(struct gfs2_holder *gh)
1241 {
1242         struct gfs2_glock *gl = gh->gh_gl;
1243         const struct gfs2_glock_operations *glops = gl->gl_ops;
1244         unsigned delay = 0;
1245
1246         if (gh->gh_flags & GL_NOCACHE)
1247                 handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1248
1249         gfs2_glmutex_lock(gl);
1250
1251         spin_lock(&gl->gl_spin);
1252         list_del_init(&gh->gh_list);
1253
1254         if (list_empty(&gl->gl_holders)) {
1255                 if (glops->go_unlock) {
1256                         spin_unlock(&gl->gl_spin);
1257                         glops->go_unlock(gh);
1258                         spin_lock(&gl->gl_spin);
1259                 }
1260                 gl->gl_stamp = jiffies;
1261         }
1262
1263         clear_bit(GLF_LOCK, &gl->gl_flags);
1264         spin_unlock(&gl->gl_spin);
1265
1266         gfs2_glock_hold(gl);
1267         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1268             !test_bit(GLF_DEMOTE, &gl->gl_flags))
1269                 delay = gl->gl_ops->go_min_hold_time;
1270         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1271                 gfs2_glock_put(gl);
1272 }
1273
1274 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1275 {
1276         struct gfs2_glock *gl = gh->gh_gl;
1277         gfs2_glock_dq(gh);
1278         wait_on_demote(gl);
1279 }
1280
1281 /**
1282  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1283  * @gh: the holder structure
1284  *
1285  */
1286
1287 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1288 {
1289         gfs2_glock_dq(gh);
1290         gfs2_holder_uninit(gh);
1291 }
1292
1293 /**
1294  * gfs2_glock_nq_num - acquire a glock based on lock number
1295  * @sdp: the filesystem
1296  * @number: the lock number
1297  * @glops: the glock operations for the type of glock
1298  * @state: the state to acquire the glock in
1299  * @flags: modifier flags for the aquisition
1300  * @gh: the struct gfs2_holder
1301  *
1302  * Returns: errno
1303  */
1304
1305 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1306                       const struct gfs2_glock_operations *glops,
1307                       unsigned int state, int flags, struct gfs2_holder *gh)
1308 {
1309         struct gfs2_glock *gl;
1310         int error;
1311
1312         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1313         if (!error) {
1314                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1315                 gfs2_glock_put(gl);
1316         }
1317
1318         return error;
1319 }
1320
1321 /**
1322  * glock_compare - Compare two struct gfs2_glock structures for sorting
1323  * @arg_a: the first structure
1324  * @arg_b: the second structure
1325  *
1326  */
1327
1328 static int glock_compare(const void *arg_a, const void *arg_b)
1329 {
1330         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1331         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1332         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1333         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1334
1335         if (a->ln_number > b->ln_number)
1336                 return 1;
1337         if (a->ln_number < b->ln_number)
1338                 return -1;
1339         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1340         return 0;
1341 }
1342
1343 /**
1344  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1345  * @num_gh: the number of structures
1346  * @ghs: an array of struct gfs2_holder structures
1347  *
1348  * Returns: 0 on success (all glocks acquired),
1349  *          errno on failure (no glocks acquired)
1350  */
1351
1352 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1353                      struct gfs2_holder **p)
1354 {
1355         unsigned int x;
1356         int error = 0;
1357
1358         for (x = 0; x < num_gh; x++)
1359                 p[x] = &ghs[x];
1360
1361         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1362
1363         for (x = 0; x < num_gh; x++) {
1364                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1365
1366                 error = gfs2_glock_nq(p[x]);
1367                 if (error) {
1368                         while (x--)
1369                                 gfs2_glock_dq(p[x]);
1370                         break;
1371                 }
1372         }
1373
1374         return error;
1375 }
1376
1377 /**
1378  * gfs2_glock_nq_m - acquire multiple glocks
1379  * @num_gh: the number of structures
1380  * @ghs: an array of struct gfs2_holder structures
1381  *
1382  *
1383  * Returns: 0 on success (all glocks acquired),
1384  *          errno on failure (no glocks acquired)
1385  */
1386
1387 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1388 {
1389         struct gfs2_holder *tmp[4];
1390         struct gfs2_holder **pph = tmp;
1391         int error = 0;
1392
1393         switch(num_gh) {
1394         case 0:
1395                 return 0;
1396         case 1:
1397                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1398                 return gfs2_glock_nq(ghs);
1399         default:
1400                 if (num_gh <= 4)
1401                         break;
1402                 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1403                 if (!pph)
1404                         return -ENOMEM;
1405         }
1406
1407         error = nq_m_sync(num_gh, ghs, pph);
1408
1409         if (pph != tmp)
1410                 kfree(pph);
1411
1412         return error;
1413 }
1414
1415 /**
1416  * gfs2_glock_dq_m - release multiple glocks
1417  * @num_gh: the number of structures
1418  * @ghs: an array of struct gfs2_holder structures
1419  *
1420  */
1421
1422 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1423 {
1424         unsigned int x;
1425
1426         for (x = 0; x < num_gh; x++)
1427                 gfs2_glock_dq(&ghs[x]);
1428 }
1429
1430 /**
1431  * gfs2_glock_dq_uninit_m - release multiple glocks
1432  * @num_gh: the number of structures
1433  * @ghs: an array of struct gfs2_holder structures
1434  *
1435  */
1436
1437 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1438 {
1439         unsigned int x;
1440
1441         for (x = 0; x < num_gh; x++)
1442                 gfs2_glock_dq_uninit(&ghs[x]);
1443 }
1444
1445 /**
1446  * gfs2_lvb_hold - attach a LVB from a glock
1447  * @gl: The glock in question
1448  *
1449  */
1450
1451 int gfs2_lvb_hold(struct gfs2_glock *gl)
1452 {
1453         int error;
1454
1455         gfs2_glmutex_lock(gl);
1456
1457         if (!atomic_read(&gl->gl_lvb_count)) {
1458                 error = gfs2_lm_hold_lvb(gl->gl_sbd, gl->gl_lock, &gl->gl_lvb);
1459                 if (error) {
1460                         gfs2_glmutex_unlock(gl);
1461                         return error;
1462                 }
1463                 gfs2_glock_hold(gl);
1464         }
1465         atomic_inc(&gl->gl_lvb_count);
1466
1467         gfs2_glmutex_unlock(gl);
1468
1469         return 0;
1470 }
1471
1472 /**
1473  * gfs2_lvb_unhold - detach a LVB from a glock
1474  * @gl: The glock in question
1475  *
1476  */
1477
1478 void gfs2_lvb_unhold(struct gfs2_glock *gl)
1479 {
1480         gfs2_glock_hold(gl);
1481         gfs2_glmutex_lock(gl);
1482
1483         gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count) > 0);
1484         if (atomic_dec_and_test(&gl->gl_lvb_count)) {
1485                 gfs2_lm_unhold_lvb(gl->gl_sbd, gl->gl_lock, gl->gl_lvb);
1486                 gl->gl_lvb = NULL;
1487                 gfs2_glock_put(gl);
1488         }
1489
1490         gfs2_glmutex_unlock(gl);
1491         gfs2_glock_put(gl);
1492 }
1493
1494 static void blocking_cb(struct gfs2_sbd *sdp, struct lm_lockname *name,
1495                         unsigned int state)
1496 {
1497         struct gfs2_glock *gl;
1498         unsigned long delay = 0;
1499         unsigned long holdtime;
1500         unsigned long now = jiffies;
1501
1502         gl = gfs2_glock_find(sdp, name);
1503         if (!gl)
1504                 return;
1505
1506         holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
1507         if (time_before(now, holdtime))
1508                 delay = holdtime - now;
1509
1510         handle_callback(gl, state, 1, delay);
1511         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1512                 gfs2_glock_put(gl);
1513 }
1514
1515 /**
1516  * gfs2_glock_cb - Callback used by locking module
1517  * @sdp: Pointer to the superblock
1518  * @type: Type of callback
1519  * @data: Type dependent data pointer
1520  *
1521  * Called by the locking module when it wants to tell us something.
1522  * Either we need to drop a lock, one of our ASYNC requests completed, or
1523  * a journal from another client needs to be recovered.
1524  */
1525
1526 void gfs2_glock_cb(void *cb_data, unsigned int type, void *data)
1527 {
1528         struct gfs2_sbd *sdp = cb_data;
1529
1530         switch (type) {
1531         case LM_CB_NEED_E:
1532                 blocking_cb(sdp, data, LM_ST_UNLOCKED);
1533                 return;
1534
1535         case LM_CB_NEED_D:
1536                 blocking_cb(sdp, data, LM_ST_DEFERRED);
1537                 return;
1538
1539         case LM_CB_NEED_S:
1540                 blocking_cb(sdp, data, LM_ST_SHARED);
1541                 return;
1542
1543         case LM_CB_ASYNC: {
1544                 struct lm_async_cb *async = data;
1545                 struct gfs2_glock *gl;
1546
1547                 down_read(&gfs2_umount_flush_sem);
1548                 gl = gfs2_glock_find(sdp, &async->lc_name);
1549                 if (gfs2_assert_warn(sdp, gl))
1550                         return;
1551                 if (!gfs2_assert_warn(sdp, gl->gl_req_bh))
1552                         gl->gl_req_bh(gl, async->lc_ret);
1553                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1554                         gfs2_glock_put(gl);
1555                 up_read(&gfs2_umount_flush_sem);
1556                 return;
1557         }
1558
1559         case LM_CB_NEED_RECOVERY:
1560                 gfs2_jdesc_make_dirty(sdp, *(unsigned int *)data);
1561                 if (sdp->sd_recoverd_process)
1562                         wake_up_process(sdp->sd_recoverd_process);
1563                 return;
1564
1565         case LM_CB_DROPLOCKS:
1566                 gfs2_gl_hash_clear(sdp, NO_WAIT);
1567                 gfs2_quota_scan(sdp);
1568                 return;
1569
1570         default:
1571                 gfs2_assert_warn(sdp, 0);
1572                 return;
1573         }
1574 }
1575
1576 /**
1577  * demote_ok - Check to see if it's ok to unlock a glock
1578  * @gl: the glock
1579  *
1580  * Returns: 1 if it's ok
1581  */
1582
1583 static int demote_ok(struct gfs2_glock *gl)
1584 {
1585         const struct gfs2_glock_operations *glops = gl->gl_ops;
1586         int demote = 1;
1587
1588         if (test_bit(GLF_STICKY, &gl->gl_flags))
1589                 demote = 0;
1590         else if (glops->go_demote_ok)
1591                 demote = glops->go_demote_ok(gl);
1592
1593         return demote;
1594 }
1595
1596 /**
1597  * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1598  * @gl: the glock
1599  *
1600  */
1601
1602 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
1603 {
1604         struct gfs2_sbd *sdp = gl->gl_sbd;
1605
1606         spin_lock(&sdp->sd_reclaim_lock);
1607         if (list_empty(&gl->gl_reclaim)) {
1608                 gfs2_glock_hold(gl);
1609                 list_add(&gl->gl_reclaim, &sdp->sd_reclaim_list);
1610                 atomic_inc(&sdp->sd_reclaim_count);
1611         }
1612         spin_unlock(&sdp->sd_reclaim_lock);
1613
1614         wake_up(&sdp->sd_reclaim_wq);
1615 }
1616
1617 /**
1618  * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1619  * @sdp: the filesystem
1620  *
1621  * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1622  * different glock and we notice that there are a lot of glocks in the
1623  * reclaim list.
1624  *
1625  */
1626
1627 void gfs2_reclaim_glock(struct gfs2_sbd *sdp)
1628 {
1629         struct gfs2_glock *gl;
1630
1631         spin_lock(&sdp->sd_reclaim_lock);
1632         if (list_empty(&sdp->sd_reclaim_list)) {
1633                 spin_unlock(&sdp->sd_reclaim_lock);
1634                 return;
1635         }
1636         gl = list_entry(sdp->sd_reclaim_list.next,
1637                         struct gfs2_glock, gl_reclaim);
1638         list_del_init(&gl->gl_reclaim);
1639         spin_unlock(&sdp->sd_reclaim_lock);
1640
1641         atomic_dec(&sdp->sd_reclaim_count);
1642         atomic_inc(&sdp->sd_reclaimed);
1643
1644         if (gfs2_glmutex_trylock(gl)) {
1645                 if (list_empty(&gl->gl_holders) &&
1646                     gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1647                         handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1648                 gfs2_glmutex_unlock(gl);
1649         }
1650
1651         gfs2_glock_put(gl);
1652 }
1653
1654 /**
1655  * examine_bucket - Call a function for glock in a hash bucket
1656  * @examiner: the function
1657  * @sdp: the filesystem
1658  * @bucket: the bucket
1659  *
1660  * Returns: 1 if the bucket has entries
1661  */
1662
1663 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1664                           unsigned int hash)
1665 {
1666         struct gfs2_glock *gl, *prev = NULL;
1667         int has_entries = 0;
1668         struct hlist_head *head = &gl_hash_table[hash].hb_list;
1669
1670         read_lock(gl_lock_addr(hash));
1671         /* Can't use hlist_for_each_entry - don't want prefetch here */
1672         if (hlist_empty(head))
1673                 goto out;
1674         gl = list_entry(head->first, struct gfs2_glock, gl_list);
1675         while(1) {
1676                 if (!sdp || gl->gl_sbd == sdp) {
1677                         gfs2_glock_hold(gl);
1678                         read_unlock(gl_lock_addr(hash));
1679                         if (prev)
1680                                 gfs2_glock_put(prev);
1681                         prev = gl;
1682                         examiner(gl);
1683                         has_entries = 1;
1684                         read_lock(gl_lock_addr(hash));
1685                 }
1686                 if (gl->gl_list.next == NULL)
1687                         break;
1688                 gl = list_entry(gl->gl_list.next, struct gfs2_glock, gl_list);
1689         }
1690 out:
1691         read_unlock(gl_lock_addr(hash));
1692         if (prev)
1693                 gfs2_glock_put(prev);
1694         cond_resched();
1695         return has_entries;
1696 }
1697
1698 /**
1699  * scan_glock - look at a glock and see if we can reclaim it
1700  * @gl: the glock to look at
1701  *
1702  */
1703
1704 static void scan_glock(struct gfs2_glock *gl)
1705 {
1706         if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object)
1707                 return;
1708
1709         if (gfs2_glmutex_trylock(gl)) {
1710                 if (list_empty(&gl->gl_holders) &&
1711                     gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1712                         goto out_schedule;
1713                 gfs2_glmutex_unlock(gl);
1714         }
1715         return;
1716
1717 out_schedule:
1718         gfs2_glmutex_unlock(gl);
1719         gfs2_glock_schedule_for_reclaim(gl);
1720 }
1721
1722 /**
1723  * clear_glock - look at a glock and see if we can free it from glock cache
1724  * @gl: the glock to look at
1725  *
1726  */
1727
1728 static void clear_glock(struct gfs2_glock *gl)
1729 {
1730         struct gfs2_sbd *sdp = gl->gl_sbd;
1731         int released;
1732
1733         spin_lock(&sdp->sd_reclaim_lock);
1734         if (!list_empty(&gl->gl_reclaim)) {
1735                 list_del_init(&gl->gl_reclaim);
1736                 atomic_dec(&sdp->sd_reclaim_count);
1737                 spin_unlock(&sdp->sd_reclaim_lock);
1738                 released = gfs2_glock_put(gl);
1739                 gfs2_assert(sdp, !released);
1740         } else {
1741                 spin_unlock(&sdp->sd_reclaim_lock);
1742         }
1743
1744         if (gfs2_glmutex_trylock(gl)) {
1745                 if (list_empty(&gl->gl_holders) &&
1746                     gl->gl_state != LM_ST_UNLOCKED)
1747                         handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1748                 gfs2_glmutex_unlock(gl);
1749         }
1750 }
1751
1752 /**
1753  * gfs2_gl_hash_clear - Empty out the glock hash table
1754  * @sdp: the filesystem
1755  * @wait: wait until it's all gone
1756  *
1757  * Called when unmounting the filesystem, or when inter-node lock manager
1758  * requests DROPLOCKS because it is running out of capacity.
1759  */
1760
1761 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp, int wait)
1762 {
1763         unsigned long t;
1764         unsigned int x;
1765         int cont;
1766
1767         t = jiffies;
1768
1769         for (;;) {
1770                 cont = 0;
1771                 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1772                         if (examine_bucket(clear_glock, sdp, x))
1773                                 cont = 1;
1774                 }
1775
1776                 if (!wait || !cont)
1777                         break;
1778
1779                 if (time_after_eq(jiffies,
1780                                   t + gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
1781                         fs_warn(sdp, "Unmount seems to be stalled. "
1782                                      "Dumping lock state...\n");
1783                         gfs2_dump_lockstate(sdp);
1784                         t = jiffies;
1785                 }
1786
1787                 down_write(&gfs2_umount_flush_sem);
1788                 invalidate_inodes(sdp->sd_vfs);
1789                 up_write(&gfs2_umount_flush_sem);
1790                 msleep(10);
1791         }
1792 }
1793
1794 /*
1795  *  Diagnostic routines to help debug distributed deadlock
1796  */
1797
1798 static void gfs2_print_symbol(struct glock_iter *gi, const char *fmt,
1799                               unsigned long address)
1800 {
1801         char buffer[KSYM_SYMBOL_LEN];
1802
1803         sprint_symbol(buffer, address);
1804         print_dbg(gi, fmt, buffer);
1805 }
1806
1807 /**
1808  * dump_holder - print information about a glock holder
1809  * @str: a string naming the type of holder
1810  * @gh: the glock holder
1811  *
1812  * Returns: 0 on success, -ENOBUFS when we run out of space
1813  */
1814
1815 static int dump_holder(struct glock_iter *gi, char *str,
1816                        struct gfs2_holder *gh)
1817 {
1818         unsigned int x;
1819         struct task_struct *gh_owner;
1820
1821         print_dbg(gi, "  %s\n", str);
1822         if (gh->gh_owner_pid) {
1823                 print_dbg(gi, "    owner = %ld ", (long)gh->gh_owner_pid);
1824                 gh_owner = find_task_by_pid(gh->gh_owner_pid);
1825                 if (gh_owner)
1826                         print_dbg(gi, "(%s)\n", gh_owner->comm);
1827                 else
1828                         print_dbg(gi, "(ended)\n");
1829         } else
1830                 print_dbg(gi, "    owner = -1\n");
1831         print_dbg(gi, "    gh_state = %u\n", gh->gh_state);
1832         print_dbg(gi, "    gh_flags =");
1833         for (x = 0; x < 32; x++)
1834                 if (gh->gh_flags & (1 << x))
1835                         print_dbg(gi, " %u", x);
1836         print_dbg(gi, " \n");
1837         print_dbg(gi, "    error = %d\n", gh->gh_error);
1838         print_dbg(gi, "    gh_iflags =");
1839         for (x = 0; x < 32; x++)
1840                 if (test_bit(x, &gh->gh_iflags))
1841                         print_dbg(gi, " %u", x);
1842         print_dbg(gi, " \n");
1843         gfs2_print_symbol(gi, "    initialized at: %s\n", gh->gh_ip);
1844
1845         return 0;
1846 }
1847
1848 /**
1849  * dump_inode - print information about an inode
1850  * @ip: the inode
1851  *
1852  * Returns: 0 on success, -ENOBUFS when we run out of space
1853  */
1854
1855 static int dump_inode(struct glock_iter *gi, struct gfs2_inode *ip)
1856 {
1857         unsigned int x;
1858
1859         print_dbg(gi, "  Inode:\n");
1860         print_dbg(gi, "    num = %llu/%llu\n",
1861                   (unsigned long long)ip->i_no_formal_ino,
1862                   (unsigned long long)ip->i_no_addr);
1863         print_dbg(gi, "    type = %u\n", IF2DT(ip->i_inode.i_mode));
1864         print_dbg(gi, "    i_flags =");
1865         for (x = 0; x < 32; x++)
1866                 if (test_bit(x, &ip->i_flags))
1867                         print_dbg(gi, " %u", x);
1868         print_dbg(gi, " \n");
1869         return 0;
1870 }
1871
1872 /**
1873  * dump_glock - print information about a glock
1874  * @gl: the glock
1875  * @count: where we are in the buffer
1876  *
1877  * Returns: 0 on success, -ENOBUFS when we run out of space
1878  */
1879
1880 static int dump_glock(struct glock_iter *gi, struct gfs2_glock *gl)
1881 {
1882         struct gfs2_holder *gh;
1883         unsigned int x;
1884         int error = -ENOBUFS;
1885         struct task_struct *gl_owner;
1886
1887         spin_lock(&gl->gl_spin);
1888
1889         print_dbg(gi, "Glock 0x%p (%u, 0x%llx)\n", gl, gl->gl_name.ln_type,
1890                    (unsigned long long)gl->gl_name.ln_number);
1891         print_dbg(gi, "  gl_flags =");
1892         for (x = 0; x < 32; x++) {
1893                 if (test_bit(x, &gl->gl_flags))
1894                         print_dbg(gi, " %u", x);
1895         }
1896         if (!test_bit(GLF_LOCK, &gl->gl_flags))
1897                 print_dbg(gi, " (unlocked)");
1898         print_dbg(gi, " \n");
1899         print_dbg(gi, "  gl_ref = %d\n", atomic_read(&gl->gl_ref));
1900         print_dbg(gi, "  gl_state = %u\n", gl->gl_state);
1901         if (gl->gl_owner_pid) {
1902                 gl_owner = find_task_by_pid(gl->gl_owner_pid);
1903                 if (gl_owner)
1904                         print_dbg(gi, "  gl_owner = pid %d (%s)\n",
1905                                   gl->gl_owner_pid, gl_owner->comm);
1906                 else
1907                         print_dbg(gi, "  gl_owner = %d (ended)\n",
1908                                   gl->gl_owner_pid);
1909         } else
1910                 print_dbg(gi, "  gl_owner = -1\n");
1911         print_dbg(gi, "  gl_ip = %lu\n", gl->gl_ip);
1912         print_dbg(gi, "  req_gh = %s\n", (gl->gl_req_gh) ? "yes" : "no");
1913         print_dbg(gi, "  req_bh = %s\n", (gl->gl_req_bh) ? "yes" : "no");
1914         print_dbg(gi, "  lvb_count = %d\n", atomic_read(&gl->gl_lvb_count));
1915         print_dbg(gi, "  object = %s\n", (gl->gl_object) ? "yes" : "no");
1916         print_dbg(gi, "  le = %s\n",
1917                    (list_empty(&gl->gl_le.le_list)) ? "no" : "yes");
1918         print_dbg(gi, "  reclaim = %s\n",
1919                    (list_empty(&gl->gl_reclaim)) ? "no" : "yes");
1920         if (gl->gl_aspace)
1921                 print_dbg(gi, "  aspace = 0x%p nrpages = %lu\n", gl->gl_aspace,
1922                            gl->gl_aspace->i_mapping->nrpages);
1923         else
1924                 print_dbg(gi, "  aspace = no\n");
1925         print_dbg(gi, "  ail = %d\n", atomic_read(&gl->gl_ail_count));
1926         if (gl->gl_req_gh) {
1927                 error = dump_holder(gi, "Request", gl->gl_req_gh);
1928                 if (error)
1929                         goto out;
1930         }
1931         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1932                 error = dump_holder(gi, "Holder", gh);
1933                 if (error)
1934                         goto out;
1935         }
1936         list_for_each_entry(gh, &gl->gl_waiters1, gh_list) {
1937                 error = dump_holder(gi, "Waiter1", gh);
1938                 if (error)
1939                         goto out;
1940         }
1941         list_for_each_entry(gh, &gl->gl_waiters3, gh_list) {
1942                 error = dump_holder(gi, "Waiter3", gh);
1943                 if (error)
1944                         goto out;
1945         }
1946         if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
1947                 print_dbg(gi, "  Demotion req to state %u (%llu uS ago)\n",
1948                           gl->gl_demote_state, (unsigned long long)
1949                           (jiffies - gl->gl_demote_time)*(1000000/HZ));
1950         }
1951         if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object) {
1952                 if (!test_bit(GLF_LOCK, &gl->gl_flags) &&
1953                         list_empty(&gl->gl_holders)) {
1954                         error = dump_inode(gi, gl->gl_object);
1955                         if (error)
1956                                 goto out;
1957                 } else {
1958                         error = -ENOBUFS;
1959                         print_dbg(gi, "  Inode: busy\n");
1960                 }
1961         }
1962
1963         error = 0;
1964
1965 out:
1966         spin_unlock(&gl->gl_spin);
1967         return error;
1968 }
1969
1970 /**
1971  * gfs2_dump_lockstate - print out the current lockstate
1972  * @sdp: the filesystem
1973  * @ub: the buffer to copy the information into
1974  *
1975  * If @ub is NULL, dump the lockstate to the console.
1976  *
1977  */
1978
1979 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
1980 {
1981         struct gfs2_glock *gl;
1982         struct hlist_node *h;
1983         unsigned int x;
1984         int error = 0;
1985
1986         for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1987
1988                 read_lock(gl_lock_addr(x));
1989
1990                 hlist_for_each_entry(gl, h, &gl_hash_table[x].hb_list, gl_list) {
1991                         if (gl->gl_sbd != sdp)
1992                                 continue;
1993
1994                         error = dump_glock(NULL, gl);
1995                         if (error)
1996                                 break;
1997                 }
1998
1999                 read_unlock(gl_lock_addr(x));
2000
2001                 if (error)
2002                         break;
2003         }
2004
2005
2006         return error;
2007 }
2008
2009 /**
2010  * gfs2_scand - Look for cached glocks and inodes to toss from memory
2011  * @sdp: Pointer to GFS2 superblock
2012  *
2013  * One of these daemons runs, finding candidates to add to sd_reclaim_list.
2014  * See gfs2_glockd()
2015  */
2016
2017 static int gfs2_scand(void *data)
2018 {
2019         unsigned x;
2020         unsigned delay;
2021
2022         while (!kthread_should_stop()) {
2023                 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
2024                         examine_bucket(scan_glock, NULL, x);
2025                 if (freezing(current))
2026                         refrigerator();
2027                 delay = scand_secs;
2028                 if (delay < 1)
2029                         delay = 1;
2030                 schedule_timeout_interruptible(delay * HZ);
2031         }
2032
2033         return 0;
2034 }
2035
2036
2037
2038 int __init gfs2_glock_init(void)
2039 {
2040         unsigned i;
2041         for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
2042                 INIT_HLIST_HEAD(&gl_hash_table[i].hb_list);
2043         }
2044 #ifdef GL_HASH_LOCK_SZ
2045         for(i = 0; i < GL_HASH_LOCK_SZ; i++) {
2046                 rwlock_init(&gl_hash_locks[i]);
2047         }
2048 #endif
2049
2050         scand_process = kthread_run(gfs2_scand, NULL, "gfs2_scand");
2051         if (IS_ERR(scand_process))
2052                 return PTR_ERR(scand_process);
2053
2054         glock_workqueue = create_workqueue("glock_workqueue");
2055         if (IS_ERR(glock_workqueue)) {
2056                 kthread_stop(scand_process);
2057                 return PTR_ERR(glock_workqueue);
2058         }
2059
2060         return 0;
2061 }
2062
2063 void gfs2_glock_exit(void)
2064 {
2065         destroy_workqueue(glock_workqueue);
2066         kthread_stop(scand_process);
2067 }
2068
2069 module_param(scand_secs, uint, S_IRUGO|S_IWUSR);
2070 MODULE_PARM_DESC(scand_secs, "The number of seconds between scand runs");
2071
2072 static int gfs2_glock_iter_next(struct glock_iter *gi)
2073 {
2074         struct gfs2_glock *gl;
2075
2076 restart:
2077         read_lock(gl_lock_addr(gi->hash));
2078         gl = gi->gl;
2079         if (gl) {
2080                 gi->gl = hlist_entry(gl->gl_list.next,
2081                                      struct gfs2_glock, gl_list);
2082                 if (gi->gl)
2083                         gfs2_glock_hold(gi->gl);
2084         }
2085         read_unlock(gl_lock_addr(gi->hash));
2086         if (gl)
2087                 gfs2_glock_put(gl);
2088         if (gl && gi->gl == NULL)
2089                 gi->hash++;
2090         while(gi->gl == NULL) {
2091                 if (gi->hash >= GFS2_GL_HASH_SIZE)
2092                         return 1;
2093                 read_lock(gl_lock_addr(gi->hash));
2094                 gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
2095                                      struct gfs2_glock, gl_list);
2096                 if (gi->gl)
2097                         gfs2_glock_hold(gi->gl);
2098                 read_unlock(gl_lock_addr(gi->hash));
2099                 gi->hash++;
2100         }
2101
2102         if (gi->sdp != gi->gl->gl_sbd)
2103                 goto restart;
2104
2105         return 0;
2106 }
2107
2108 static void gfs2_glock_iter_free(struct glock_iter *gi)
2109 {
2110         if (gi->gl)
2111                 gfs2_glock_put(gi->gl);
2112         kfree(gi);
2113 }
2114
2115 static struct glock_iter *gfs2_glock_iter_init(struct gfs2_sbd *sdp)
2116 {
2117         struct glock_iter *gi;
2118
2119         gi = kmalloc(sizeof (*gi), GFP_KERNEL);
2120         if (!gi)
2121                 return NULL;
2122
2123         gi->sdp = sdp;
2124         gi->hash = 0;
2125         gi->seq = NULL;
2126         gi->gl = NULL;
2127         memset(gi->string, 0, sizeof(gi->string));
2128
2129         if (gfs2_glock_iter_next(gi)) {
2130                 gfs2_glock_iter_free(gi);
2131                 return NULL;
2132         }
2133
2134         return gi;
2135 }
2136
2137 static void *gfs2_glock_seq_start(struct seq_file *file, loff_t *pos)
2138 {
2139         struct glock_iter *gi;
2140         loff_t n = *pos;
2141
2142         gi = gfs2_glock_iter_init(file->private);
2143         if (!gi)
2144                 return NULL;
2145
2146         while(n--) {
2147                 if (gfs2_glock_iter_next(gi)) {
2148                         gfs2_glock_iter_free(gi);
2149                         return NULL;
2150                 }
2151         }
2152
2153         return gi;
2154 }
2155
2156 static void *gfs2_glock_seq_next(struct seq_file *file, void *iter_ptr,
2157                                  loff_t *pos)
2158 {
2159         struct glock_iter *gi = iter_ptr;
2160
2161         (*pos)++;
2162
2163         if (gfs2_glock_iter_next(gi)) {
2164                 gfs2_glock_iter_free(gi);
2165                 return NULL;
2166         }
2167
2168         return gi;
2169 }
2170
2171 static void gfs2_glock_seq_stop(struct seq_file *file, void *iter_ptr)
2172 {
2173         struct glock_iter *gi = iter_ptr;
2174         if (gi)
2175                 gfs2_glock_iter_free(gi);
2176 }
2177
2178 static int gfs2_glock_seq_show(struct seq_file *file, void *iter_ptr)
2179 {
2180         struct glock_iter *gi = iter_ptr;
2181
2182         gi->seq = file;
2183         dump_glock(gi, gi->gl);
2184
2185         return 0;
2186 }
2187
2188 static const struct seq_operations gfs2_glock_seq_ops = {
2189         .start = gfs2_glock_seq_start,
2190         .next  = gfs2_glock_seq_next,
2191         .stop  = gfs2_glock_seq_stop,
2192         .show  = gfs2_glock_seq_show,
2193 };
2194
2195 static int gfs2_debugfs_open(struct inode *inode, struct file *file)
2196 {
2197         struct seq_file *seq;
2198         int ret;
2199
2200         ret = seq_open(file, &gfs2_glock_seq_ops);
2201         if (ret)
2202                 return ret;
2203
2204         seq = file->private_data;
2205         seq->private = inode->i_private;
2206
2207         return 0;
2208 }
2209
2210 static const struct file_operations gfs2_debug_fops = {
2211         .owner   = THIS_MODULE,
2212         .open    = gfs2_debugfs_open,
2213         .read    = seq_read,
2214         .llseek  = seq_lseek,
2215         .release = seq_release
2216 };
2217
2218 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2219 {
2220         sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2221         if (!sdp->debugfs_dir)
2222                 return -ENOMEM;
2223         sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
2224                                                          S_IFREG | S_IRUGO,
2225                                                          sdp->debugfs_dir, sdp,
2226                                                          &gfs2_debug_fops);
2227         if (!sdp->debugfs_dentry_glocks)
2228                 return -ENOMEM;
2229
2230         return 0;
2231 }
2232
2233 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2234 {
2235         if (sdp && sdp->debugfs_dir) {
2236                 if (sdp->debugfs_dentry_glocks) {
2237                         debugfs_remove(sdp->debugfs_dentry_glocks);
2238                         sdp->debugfs_dentry_glocks = NULL;
2239                 }
2240                 debugfs_remove(sdp->debugfs_dir);
2241                 sdp->debugfs_dir = NULL;
2242         }
2243 }
2244
2245 int gfs2_register_debugfs(void)
2246 {
2247         gfs2_root = debugfs_create_dir("gfs2", NULL);
2248         return gfs2_root ? 0 : -ENOMEM;
2249 }
2250
2251 void gfs2_unregister_debugfs(void)
2252 {
2253         debugfs_remove(gfs2_root);
2254         gfs2_root = NULL;
2255 }