fs/gfs2/glock.c

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