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 <asm/uaccess.h>
  23
  24 #include "gfs2.h"
  25 #include "incore.h"
  26 #include "glock.h"
  27 #include "glops.h"
  28 #include "inode.h"
  29 #include "lm.h"
  30 #include "lops.h"
  31 #include "meta_io.h"
  32 #include "quota.h"
  33 #include "super.h"
  34 #include "util.h"
  35
  36 struct greedy {
  37         struct gfs2_holder gr_gh;
  38         struct delayed_work gr_work;
  39 };
  40
  41 struct gfs2_gl_hash_bucket {
  42         struct hlist_head hb_list;
  43 };
  44
  45 typedef void (*glock_examiner) (struct gfs2_glock * gl);
  46
  47 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp);
  48 static int dump_glock(struct gfs2_glock *gl);
  49 static int dump_inode(struct gfs2_inode *ip);
  50
  51 #define GFS2_GL_HASH_SHIFT      15
  52 #define GFS2_GL_HASH_SIZE       (1 << GFS2_GL_HASH_SHIFT)
  53 #define GFS2_GL_HASH_MASK       (GFS2_GL_HASH_SIZE - 1)
  54
  55 static struct gfs2_gl_hash_bucket gl_hash_table[GFS2_GL_HASH_SIZE];
  56
  57 /*
  58  * Despite what you might think, the numbers below are not arbitrary :-)
  59  * They are taken from the ipv4 routing hash code, which is well tested
  60  * and thus should be nearly optimal. Later on we might tweek the numbers
  61  * but for now this should be fine.
  62  *
  63  * The reason for putting the locks in a separate array from the list heads
  64  * is that we can have fewer locks than list heads and save memory. We use
  65  * the same hash function for both, but with a different hash mask.
  66  */
  67 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
  68         defined(CONFIG_PROVE_LOCKING)
  69
  70 #ifdef CONFIG_LOCKDEP
  71 # define GL_HASH_LOCK_SZ        256
  72 #else
  73 # if NR_CPUS >= 32
  74 #  define GL_HASH_LOCK_SZ       4096
  75 # elif NR_CPUS >= 16
  76 #  define GL_HASH_LOCK_SZ       2048
  77 # elif NR_CPUS >= 8
  78 #  define GL_HASH_LOCK_SZ       1024
  79 # elif NR_CPUS >= 4
  80 #  define GL_HASH_LOCK_SZ       512
  81 # else
  82 #  define GL_HASH_LOCK_SZ       256
  83 # endif
  84 #endif
  85
  86 /* We never want more locks than chains */
  87 #if GFS2_GL_HASH_SIZE < GL_HASH_LOCK_SZ
  88 # undef GL_HASH_LOCK_SZ
  89 # define GL_HASH_LOCK_SZ GFS2_GL_HASH_SIZE
  90 #endif
  91
  92 static rwlock_t gl_hash_locks[GL_HASH_LOCK_SZ];
  93
  94 static inline rwlock_t *gl_lock_addr(unsigned int x)
  95 {
  96         return &gl_hash_locks[x & (GL_HASH_LOCK_SZ-1)];
  97 }
  98 #else /* not SMP, so no spinlocks required */
  99 static inline rwlock_t *gl_lock_addr(unsigned int x)
 100 {
 101         return NULL;
 102 }
 103 #endif
 104
 105 /**
 106  * relaxed_state_ok - is a requested lock compatible with the current lock mode?
 107  * @actual: the current state of the lock
 108  * @requested: the lock state that was requested by the caller
 109  * @flags: the modifier flags passed in by the caller
 110  *
 111  * Returns: 1 if the locks are compatible, 0 otherwise
 112  */
 113
 114 static inline int relaxed_state_ok(unsigned int actual, unsigned requested,
 115                                    int flags)
 116 {
 117         if (actual == requested)
 118                 return 1;
 119
 120         if (flags & GL_EXACT)
 121                 return 0;
 122
 123         if (actual == LM_ST_EXCLUSIVE && requested == LM_ST_SHARED)
 124                 return 1;
 125
 126         if (actual != LM_ST_UNLOCKED && (flags & LM_FLAG_ANY))
 127                 return 1;
 128
 129         return 0;
 130 }
 131
 132 /**
 133  * gl_hash() - Turn glock number into hash bucket number
 134  * @lock: The glock number
 135  *
 136  * Returns: The number of the corresponding hash bucket
 137  */
 138
 139 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
 140                             const struct lm_lockname *name)
 141 {
 142         unsigned int h;
 143
 144         h = jhash(&name->ln_number, sizeof(u64), 0);
 145         h = jhash(&name->ln_type, sizeof(unsigned int), h);
 146         h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
 147         h &= GFS2_GL_HASH_MASK;
 148
 149         return h;
 150 }
 151
 152 /**
 153  * glock_free() - Perform a few checks and then release struct gfs2_glock
 154  * @gl: The glock to release
 155  *
 156  * Also calls lock module to release its internal structure for this glock.
 157  *
 158  */
 159
 160 static void glock_free(struct gfs2_glock *gl)
 161 {
 162         struct gfs2_sbd *sdp = gl->gl_sbd;
 163         struct inode *aspace = gl->gl_aspace;
 164
 165         gfs2_lm_put_lock(sdp, gl->gl_lock);
 166
 167         if (aspace)
 168                 gfs2_aspace_put(aspace);
 169
 170         kmem_cache_free(gfs2_glock_cachep, gl);
 171 }
 172
 173 /**
 174  * gfs2_glock_hold() - increment reference count on glock
 175  * @gl: The glock to hold
 176  *
 177  */
 178
 179 void gfs2_glock_hold(struct gfs2_glock *gl)
 180 {
 181         atomic_inc(&gl->gl_ref);
 182 }
 183
 184 /**
 185  * gfs2_glock_put() - Decrement reference count on glock
 186  * @gl: The glock to put
 187  *
 188  */
 189
 190 int gfs2_glock_put(struct gfs2_glock *gl)
 191 {
 192         int rv = 0;
 193         struct gfs2_sbd *sdp = gl->gl_sbd;
 194
 195         write_lock(gl_lock_addr(gl->gl_hash));
 196         if (atomic_dec_and_test(&gl->gl_ref)) {
 197                 hlist_del(&gl->gl_list);
 198                 write_unlock(gl_lock_addr(gl->gl_hash));
 199                 BUG_ON(spin_is_locked(&gl->gl_spin));
 200                 gfs2_assert(sdp, gl->gl_state == LM_ST_UNLOCKED);
 201                 gfs2_assert(sdp, list_empty(&gl->gl_reclaim));
 202                 gfs2_assert(sdp, list_empty(&gl->gl_holders));
 203                 gfs2_assert(sdp, list_empty(&gl->gl_waiters1));
 204                 gfs2_assert(sdp, list_empty(&gl->gl_waiters2));
 205                 gfs2_assert(sdp, list_empty(&gl->gl_waiters3));
 206                 glock_free(gl);
 207                 rv = 1;
 208                 goto out;
 209         }
 210         write_unlock(gl_lock_addr(gl->gl_hash));
 211 out:
 212         return rv;
 213 }
 214
 215 /**
 216  * queue_empty - check to see if a glock's queue is empty
 217  * @gl: the glock
 218  * @head: the head of the queue to check
 219  *
 220  * This function protects the list in the event that a process already
 221  * has a holder on the list and is adding a second holder for itself.
 222  * The glmutex lock is what generally prevents processes from working
 223  * on the same glock at once, but the special case of adding a second
 224  * holder for yourself ("recursive" locking) doesn't involve locking
 225  * glmutex, making the spin lock necessary.
 226  *
 227  * Returns: 1 if the queue is empty
 228  */
 229
 230 static inline int queue_empty(struct gfs2_glock *gl, struct list_head *head)
 231 {
 232         int empty;
 233         spin_lock(&gl->gl_spin);
 234         empty = list_empty(head);
 235         spin_unlock(&gl->gl_spin);
 236         return empty;
 237 }
 238
 239 /**
 240  * search_bucket() - Find struct gfs2_glock by lock number
 241  * @bucket: the bucket to search
 242  * @name: The lock name
 243  *
 244  * Returns: NULL, or the struct gfs2_glock with the requested number
 245  */
 246
 247 static struct gfs2_glock *search_bucket(unsigned int hash,
 248                                         const struct gfs2_sbd *sdp,
 249                                         const struct lm_lockname *name)
 250 {
 251         struct gfs2_glock *gl;
 252         struct hlist_node *h;
 253
 254         hlist_for_each_entry(gl, h, &gl_hash_table[hash].hb_list, gl_list) {
 255                 if (!lm_name_equal(&gl->gl_name, name))
 256                         continue;
 257                 if (gl->gl_sbd != sdp)
 258                         continue;
 259
 260                 atomic_inc(&gl->gl_ref);
 261
 262                 return gl;
 263         }
 264
 265         return NULL;
 266 }
 267
 268 /**
 269  * gfs2_glock_find() - Find glock by lock number
 270  * @sdp: The GFS2 superblock
 271  * @name: The lock name
 272  *
 273  * Returns: NULL, or the struct gfs2_glock with the requested number
 274  */
 275
 276 static struct gfs2_glock *gfs2_glock_find(const struct gfs2_sbd *sdp,
 277                                           const struct lm_lockname *name)
 278 {
 279         unsigned int hash = gl_hash(sdp, name);
 280         struct gfs2_glock *gl;
 281
 282         read_lock(gl_lock_addr(hash));
 283         gl = search_bucket(hash, sdp, name);
 284         read_unlock(gl_lock_addr(hash));
 285
 286         return gl;
 287 }
 288
 289 /**
 290  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
 291  * @sdp: The GFS2 superblock
 292  * @number: the lock number
 293  * @glops: The glock_operations to use
 294  * @create: If 0, don't create the glock if it doesn't exist
 295  * @glp: the glock is returned here
 296  *
 297  * This does not lock a glock, just finds/creates structures for one.
 298  *
 299  * Returns: errno
 300  */
 301
 302 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
 303                    const struct gfs2_glock_operations *glops, int create,
 304                    struct gfs2_glock **glp)
 305 {
 306         struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
 307         struct gfs2_glock *gl, *tmp;
 308         unsigned int hash = gl_hash(sdp, &name);
 309         int error;
 310
 311         read_lock(gl_lock_addr(hash));
 312         gl = search_bucket(hash, sdp, &name);
 313         read_unlock(gl_lock_addr(hash));
 314
 315         if (gl || !create) {
 316                 *glp = gl;
 317                 return 0;
 318         }
 319
 320         gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
 321         if (!gl)
 322                 return -ENOMEM;
 323
 324         gl->gl_flags = 0;
 325         gl->gl_name = name;
 326         atomic_set(&gl->gl_ref, 1);
 327         gl->gl_state = LM_ST_UNLOCKED;
 328         gl->gl_hash = hash;
 329         gl->gl_owner = NULL;
 330         gl->gl_ip = 0;
 331         gl->gl_ops = glops;
 332         gl->gl_req_gh = NULL;
 333         gl->gl_req_bh = NULL;
 334         gl->gl_vn = 0;
 335         gl->gl_stamp = jiffies;
 336         gl->gl_object = NULL;
 337         gl->gl_sbd = sdp;
 338         gl->gl_aspace = NULL;
 339         lops_init_le(&gl->gl_le, &gfs2_glock_lops);
 340
 341         /* If this glock protects actual on-disk data or metadata blocks,
 342            create a VFS inode to manage the pages/buffers holding them. */
 343         if (glops == &gfs2_inode_glops || glops == &gfs2_rgrp_glops) {
 344                 gl->gl_aspace = gfs2_aspace_get(sdp);
 345                 if (!gl->gl_aspace) {
 346                         error = -ENOMEM;
 347                         goto fail;
 348                 }
 349         }
 350
 351         error = gfs2_lm_get_lock(sdp, &name, &gl->gl_lock);
 352         if (error)
 353                 goto fail_aspace;
 354
 355         write_lock(gl_lock_addr(hash));
 356         tmp = search_bucket(hash, sdp, &name);
 357         if (tmp) {
 358                 write_unlock(gl_lock_addr(hash));
 359                 glock_free(gl);
 360                 gl = tmp;
 361         } else {
 362                 hlist_add_head(&gl->gl_list, &gl_hash_table[hash].hb_list);
 363                 write_unlock(gl_lock_addr(hash));
 364         }
 365
 366         *glp = gl;
 367
 368         return 0;
 369
 370 fail_aspace:
 371         if (gl->gl_aspace)
 372                 gfs2_aspace_put(gl->gl_aspace);
 373 fail:
 374         kmem_cache_free(gfs2_glock_cachep, gl);
 375         return error;
 376 }
 377
 378 /**
 379  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
 380  * @gl: the glock
 381  * @state: the state we're requesting
 382  * @flags: the modifier flags
 383  * @gh: the holder structure
 384  *
 385  */
 386
 387 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
 388                       struct gfs2_holder *gh)
 389 {
 390         INIT_LIST_HEAD(&gh->gh_list);
 391         gh->gh_gl = gl;
 392         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 393         gh->gh_owner = current;
 394         gh->gh_state = state;
 395         gh->gh_flags = flags;
 396         gh->gh_error = 0;
 397         gh->gh_iflags = 0;
 398         init_completion(&gh->gh_wait);
 399
 400         if (gh->gh_state == LM_ST_EXCLUSIVE)
 401                 gh->gh_flags |= GL_LOCAL_EXCL;
 402
 403         gfs2_glock_hold(gl);
 404 }
 405
 406 /**
 407  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
 408  * @state: the state we're requesting
 409  * @flags: the modifier flags
 410  * @gh: the holder structure
 411  *
 412  * Don't mess with the glock.
 413  *
 414  */
 415
 416 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
 417 {
 418         gh->gh_state = state;
 419         gh->gh_flags = flags;
 420         if (gh->gh_state == LM_ST_EXCLUSIVE)
 421                 gh->gh_flags |= GL_LOCAL_EXCL;
 422
 423         gh->gh_iflags &= 1 << HIF_ALLOCED;
 424         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 425 }
 426
 427 /**
 428  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
 429  * @gh: the holder structure
 430  *
 431  */
 432
 433 void gfs2_holder_uninit(struct gfs2_holder *gh)
 434 {
 435         gfs2_glock_put(gh->gh_gl);
 436         gh->gh_gl = NULL;
 437         gh->gh_ip = 0;
 438 }
 439
 440 /**
 441  * gfs2_holder_get - get a struct gfs2_holder structure
 442  * @gl: the glock
 443  * @state: the state we're requesting
 444  * @flags: the modifier flags
 445  * @gfp_flags:
 446  *
 447  * Figure out how big an impact this function has.  Either:
 448  * 1) Replace it with a cache of structures hanging off the struct gfs2_sbd
 449  * 2) Leave it like it is
 450  *
 451  * Returns: the holder structure, NULL on ENOMEM
 452  */
 453
 454 static struct gfs2_holder *gfs2_holder_get(struct gfs2_glock *gl,
 455                                            unsigned int state,
 456                                            int flags, gfp_t gfp_flags)
 457 {
 458         struct gfs2_holder *gh;
 459
 460         gh = kmalloc(sizeof(struct gfs2_holder), gfp_flags);
 461         if (!gh)
 462                 return NULL;
 463
 464         gfs2_holder_init(gl, state, flags, gh);
 465         set_bit(HIF_ALLOCED, &gh->gh_iflags);
 466         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 467         return gh;
 468 }
 469
 470 /**
 471  * gfs2_holder_put - get rid of a struct gfs2_holder structure
 472  * @gh: the holder structure
 473  *
 474  */
 475
 476 static void gfs2_holder_put(struct gfs2_holder *gh)
 477 {
 478         gfs2_holder_uninit(gh);
 479         kfree(gh);
 480 }
 481
 482 /**
 483  * rq_mutex - process a mutex request in the queue
 484  * @gh: the glock holder
 485  *
 486  * Returns: 1 if the queue is blocked
 487  */
 488
 489 static int rq_mutex(struct gfs2_holder *gh)
 490 {
 491         struct gfs2_glock *gl = gh->gh_gl;
 492
 493         list_del_init(&gh->gh_list);
 494         /*  gh->gh_error never examined.  */
 495         set_bit(GLF_LOCK, &gl->gl_flags);
 496         complete(&gh->gh_wait);
 497
 498         return 1;
 499 }
 500
 501 /**
 502  * rq_promote - process a promote request in the queue
 503  * @gh: the glock holder
 504  *
 505  * Acquire a new inter-node lock, or change a lock state to more restrictive.
 506  *
 507  * Returns: 1 if the queue is blocked
 508  */
 509
 510 static int rq_promote(struct gfs2_holder *gh)
 511 {
 512         struct gfs2_glock *gl = gh->gh_gl;
 513         struct gfs2_sbd *sdp = gl->gl_sbd;
 514         const struct gfs2_glock_operations *glops = gl->gl_ops;
 515
 516         if (!relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
 517                 if (list_empty(&gl->gl_holders)) {
 518                         gl->gl_req_gh = gh;
 519                         set_bit(GLF_LOCK, &gl->gl_flags);
 520                         spin_unlock(&gl->gl_spin);
 521
 522                         if (atomic_read(&sdp->sd_reclaim_count) >
 523                             gfs2_tune_get(sdp, gt_reclaim_limit) &&
 524                             !(gh->gh_flags & LM_FLAG_PRIORITY)) {
 525                                 gfs2_reclaim_glock(sdp);
 526                                 gfs2_reclaim_glock(sdp);
 527                         }
 528
 529                         glops->go_xmote_th(gl, gh->gh_state, gh->gh_flags);
 530                         spin_lock(&gl->gl_spin);
 531                 }
 532                 return 1;
 533         }
 534
 535         if (list_empty(&gl->gl_holders)) {
 536                 set_bit(HIF_FIRST, &gh->gh_iflags);
 537                 set_bit(GLF_LOCK, &gl->gl_flags);
 538         } else {
 539                 struct gfs2_holder *next_gh;
 540                 if (gh->gh_flags & GL_LOCAL_EXCL)
 541                         return 1;
 542                 next_gh = list_entry(gl->gl_holders.next, struct gfs2_holder,
 543                                      gh_list);
 544                 if (next_gh->gh_flags & GL_LOCAL_EXCL)
 545                          return 1;
 546         }
 547
 548         list_move_tail(&gh->gh_list, &gl->gl_holders);
 549         gh->gh_error = 0;
 550         set_bit(HIF_HOLDER, &gh->gh_iflags);
 551
 552         complete(&gh->gh_wait);
 553
 554         return 0;
 555 }
 556
 557 /**
 558  * rq_demote - process a demote request in the queue
 559  * @gh: the glock holder
 560  *
 561  * Returns: 1 if the queue is blocked
 562  */
 563
 564 static int rq_demote(struct gfs2_holder *gh)
 565 {
 566         struct gfs2_glock *gl = gh->gh_gl;
 567         const struct gfs2_glock_operations *glops = gl->gl_ops;
 568
 569         if (!list_empty(&gl->gl_holders))
 570                 return 1;
 571
 572         if (gl->gl_state == gh->gh_state || gl->gl_state == LM_ST_UNLOCKED) {
 573                 list_del_init(&gh->gh_list);
 574                 gh->gh_error = 0;
 575                 spin_unlock(&gl->gl_spin);
 576                 if (test_bit(HIF_DEALLOC, &gh->gh_iflags))
 577                         gfs2_holder_put(gh);
 578                 else
 579                         complete(&gh->gh_wait);
 580                 spin_lock(&gl->gl_spin);
 581         } else {
 582                 gl->gl_req_gh = gh;
 583                 set_bit(GLF_LOCK, &gl->gl_flags);
 584                 spin_unlock(&gl->gl_spin);
 585
 586                 if (gh->gh_state == LM_ST_UNLOCKED ||
 587                     gl->gl_state != LM_ST_EXCLUSIVE)
 588                         glops->go_drop_th(gl);
 589                 else
 590                         glops->go_xmote_th(gl, gh->gh_state, gh->gh_flags);
 591
 592                 spin_lock(&gl->gl_spin);
 593         }
 594
 595         return 0;
 596 }
 597
 598 /**
 599  * rq_greedy - process a queued request to drop greedy status
 600  * @gh: the glock holder
 601  *
 602  * Returns: 1 if the queue is blocked
 603  */
 604
 605 static int rq_greedy(struct gfs2_holder *gh)
 606 {
 607         struct gfs2_glock *gl = gh->gh_gl;
 608
 609         list_del_init(&gh->gh_list);
 610         /*  gh->gh_error never examined.  */
 611         clear_bit(GLF_GREEDY, &gl->gl_flags);
 612         spin_unlock(&gl->gl_spin);
 613
 614         gfs2_holder_uninit(gh);
 615         kfree(container_of(gh, struct greedy, gr_gh));
 616
 617         spin_lock(&gl->gl_spin);
 618
 619         return 0;
 620 }
 621
 622 /**
 623  * run_queue - process holder structures on a glock
 624  * @gl: the glock
 625  *
 626  */
 627 static void run_queue(struct gfs2_glock *gl)
 628 {
 629         struct gfs2_holder *gh;
 630         int blocked = 1;
 631
 632         for (;;) {
 633                 if (test_bit(GLF_LOCK, &gl->gl_flags))
 634                         break;
 635
 636                 if (!list_empty(&gl->gl_waiters1)) {
 637                         gh = list_entry(gl->gl_waiters1.next,
 638                                         struct gfs2_holder, gh_list);
 639
 640                         if (test_bit(HIF_MUTEX, &gh->gh_iflags))
 641                                 blocked = rq_mutex(gh);
 642                         else
 643                                 gfs2_assert_warn(gl->gl_sbd, 0);
 644
 645                 } else if (!list_empty(&gl->gl_waiters2) &&
 646                            !test_bit(GLF_SKIP_WAITERS2, &gl->gl_flags)) {
 647                         gh = list_entry(gl->gl_waiters2.next,
 648                                         struct gfs2_holder, gh_list);
 649
 650                         if (test_bit(HIF_DEMOTE, &gh->gh_iflags))
 651                                 blocked = rq_demote(gh);
 652                         else if (test_bit(HIF_GREEDY, &gh->gh_iflags))
 653                                 blocked = rq_greedy(gh);
 654                         else
 655                                 gfs2_assert_warn(gl->gl_sbd, 0);
 656
 657                 } else if (!list_empty(&gl->gl_waiters3)) {
 658                         gh = list_entry(gl->gl_waiters3.next,
 659                                         struct gfs2_holder, gh_list);
 660
 661                         if (test_bit(HIF_PROMOTE, &gh->gh_iflags))
 662                                 blocked = rq_promote(gh);
 663                         else
 664                                 gfs2_assert_warn(gl->gl_sbd, 0);
 665
 666                 } else
 667                         break;
 668
 669                 if (blocked)
 670                         break;
 671         }
 672 }
 673
 674 /**
 675  * gfs2_glmutex_lock - acquire a local lock on a glock
 676  * @gl: the glock
 677  *
 678  * Gives caller exclusive access to manipulate a glock structure.
 679  */
 680
 681 static void gfs2_glmutex_lock(struct gfs2_glock *gl)
 682 {
 683         struct gfs2_holder gh;
 684
 685         gfs2_holder_init(gl, 0, 0, &gh);
 686         set_bit(HIF_MUTEX, &gh.gh_iflags);
 687
 688         spin_lock(&gl->gl_spin);
 689         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
 690                 list_add_tail(&gh.gh_list, &gl->gl_waiters1);
 691         } else {
 692                 gl->gl_owner = current;
 693                 gl->gl_ip = (unsigned long)__builtin_return_address(0);
 694                 complete(&gh.gh_wait);
 695         }
 696         spin_unlock(&gl->gl_spin);
 697
 698         wait_for_completion(&gh.gh_wait);
 699         gfs2_holder_uninit(&gh);
 700 }
 701
 702 /**
 703  * gfs2_glmutex_trylock - try to acquire a local lock on a glock
 704  * @gl: the glock
 705  *
 706  * Returns: 1 if the glock is acquired
 707  */
 708
 709 static int gfs2_glmutex_trylock(struct gfs2_glock *gl)
 710 {
 711         int acquired = 1;
 712
 713         spin_lock(&gl->gl_spin);
 714         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
 715                 acquired = 0;
 716         } else {
 717                 gl->gl_owner = current;
 718                 gl->gl_ip = (unsigned long)__builtin_return_address(0);
 719         }
 720         spin_unlock(&gl->gl_spin);
 721
 722         return acquired;
 723 }
 724
 725 /**
 726  * gfs2_glmutex_unlock - release a local lock on a glock
 727  * @gl: the glock
 728  *
 729  */
 730
 731 static void gfs2_glmutex_unlock(struct gfs2_glock *gl)
 732 {
 733         spin_lock(&gl->gl_spin);
 734         clear_bit(GLF_LOCK, &gl->gl_flags);
 735         gl->gl_owner = NULL;
 736         gl->gl_ip = 0;
 737         run_queue(gl);
 738         BUG_ON(!spin_is_locked(&gl->gl_spin));
 739         spin_unlock(&gl->gl_spin);
 740 }
 741
 742 /**
 743  * handle_callback - add a demote request to a lock's queue
 744  * @gl: the glock
 745  * @state: the state the caller wants us to change to
 746  *
 747  * Note: This may fail sliently if we are out of memory.
 748  */
 749
 750 static void handle_callback(struct gfs2_glock *gl, unsigned int state)
 751 {
 752         struct gfs2_holder *gh, *new_gh = NULL;
 753
 754 restart:
 755         spin_lock(&gl->gl_spin);
 756
 757         list_for_each_entry(gh, &gl->gl_waiters2, gh_list) {
 758                 if (test_bit(HIF_DEMOTE, &gh->gh_iflags) &&
 759                     gl->gl_req_gh != gh) {
 760                         if (gh->gh_state != state)
 761                                 gh->gh_state = LM_ST_UNLOCKED;
 762                         goto out;
 763                 }
 764         }
 765
 766         if (new_gh) {
 767                 list_add_tail(&new_gh->gh_list, &gl->gl_waiters2);
 768                 new_gh = NULL;
 769         } else {
 770                 spin_unlock(&gl->gl_spin);
 771
 772                 new_gh = gfs2_holder_get(gl, state, LM_FLAG_TRY, GFP_NOFS);
 773                 if (!new_gh)
 774                         return;
 775                 set_bit(HIF_DEMOTE, &new_gh->gh_iflags);
 776                 set_bit(HIF_DEALLOC, &new_gh->gh_iflags);
 777
 778                 goto restart;
 779         }
 780
 781 out:
 782         spin_unlock(&gl->gl_spin);
 783
 784         if (new_gh)
 785                 gfs2_holder_put(new_gh);
 786 }
 787
 788 /**
 789  * state_change - record that the glock is now in a different state
 790  * @gl: the glock
 791  * @new_state the new state
 792  *
 793  */
 794
 795 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
 796 {
 797         int held1, held2;
 798
 799         held1 = (gl->gl_state != LM_ST_UNLOCKED);
 800         held2 = (new_state != LM_ST_UNLOCKED);
 801
 802         if (held1 != held2) {
 803                 if (held2)
 804                         gfs2_glock_hold(gl);
 805                 else
 806                         gfs2_glock_put(gl);
 807         }
 808
 809         gl->gl_state = new_state;
 810 }
 811
 812 /**
 813  * xmote_bh - Called after the lock module is done acquiring a lock
 814  * @gl: The glock in question
 815  * @ret: the int returned from the lock module
 816  *
 817  */
 818
 819 static void xmote_bh(struct gfs2_glock *gl, unsigned int ret)
 820 {
 821         struct gfs2_sbd *sdp = gl->gl_sbd;
 822         const struct gfs2_glock_operations *glops = gl->gl_ops;
 823         struct gfs2_holder *gh = gl->gl_req_gh;
 824         int prev_state = gl->gl_state;
 825         int op_done = 1;
 826
 827         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
 828         gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
 829         gfs2_assert_warn(sdp, !(ret & LM_OUT_ASYNC));
 830
 831         state_change(gl, ret & LM_OUT_ST_MASK);
 832
 833         if (prev_state != LM_ST_UNLOCKED && !(ret & LM_OUT_CACHEABLE)) {
 834                 if (glops->go_inval)
 835                         glops->go_inval(gl, DIO_METADATA);
 836         } else if (gl->gl_state == LM_ST_DEFERRED) {
 837                 /* We might not want to do this here.
 838                    Look at moving to the inode glops. */
 839                 if (glops->go_inval)
 840                         glops->go_inval(gl, 0);
 841         }
 842
 843         /*  Deal with each possible exit condition  */
 844
 845         if (!gh)
 846                 gl->gl_stamp = jiffies;
 847         else if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
 848                 spin_lock(&gl->gl_spin);
 849                 list_del_init(&gh->gh_list);
 850                 gh->gh_error = -EIO;
 851                 spin_unlock(&gl->gl_spin);
 852         } else if (test_bit(HIF_DEMOTE, &gh->gh_iflags)) {
 853                 spin_lock(&gl->gl_spin);
 854                 list_del_init(&gh->gh_list);
 855                 if (gl->gl_state == gh->gh_state ||
 856                     gl->gl_state == LM_ST_UNLOCKED) {
 857                         gh->gh_error = 0;
 858                 } else {
 859                         if (gfs2_assert_warn(sdp, gh->gh_flags &
 860                                         (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) == -1)
 861                                 fs_warn(sdp, "ret = 0x%.8X\n", ret);
 862                         gh->gh_error = GLR_TRYFAILED;
 863                 }
 864                 spin_unlock(&gl->gl_spin);
 865
 866                 if (ret & LM_OUT_CANCELED)
 867                         handle_callback(gl, LM_ST_UNLOCKED);
 868
 869         } else if (ret & LM_OUT_CANCELED) {
 870                 spin_lock(&gl->gl_spin);
 871                 list_del_init(&gh->gh_list);
 872                 gh->gh_error = GLR_CANCELED;
 873                 spin_unlock(&gl->gl_spin);
 874
 875         } else if (relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
 876                 spin_lock(&gl->gl_spin);
 877                 list_move_tail(&gh->gh_list, &gl->gl_holders);
 878                 gh->gh_error = 0;
 879                 set_bit(HIF_HOLDER, &gh->gh_iflags);
 880                 spin_unlock(&gl->gl_spin);
 881
 882                 set_bit(HIF_FIRST, &gh->gh_iflags);
 883
 884                 op_done = 0;
 885
 886         } else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
 887                 spin_lock(&gl->gl_spin);
 888                 list_del_init(&gh->gh_list);
 889                 gh->gh_error = GLR_TRYFAILED;
 890                 spin_unlock(&gl->gl_spin);
 891
 892         } else {
 893                 if (gfs2_assert_withdraw(sdp, 0) == -1)
 894                         fs_err(sdp, "ret = 0x%.8X\n", ret);
 895         }
 896
 897         if (glops->go_xmote_bh)
 898                 glops->go_xmote_bh(gl);
 899
 900         if (op_done) {
 901                 spin_lock(&gl->gl_spin);
 902                 gl->gl_req_gh = NULL;
 903                 gl->gl_req_bh = NULL;
 904                 clear_bit(GLF_LOCK, &gl->gl_flags);
 905                 run_queue(gl);
 906                 spin_unlock(&gl->gl_spin);
 907         }
 908
 909         gfs2_glock_put(gl);
 910
 911         if (gh) {
 912                 if (test_bit(HIF_DEALLOC, &gh->gh_iflags))
 913                         gfs2_holder_put(gh);
 914                 else
 915                         complete(&gh->gh_wait);
 916         }
 917 }
 918
 919 /**
 920  * gfs2_glock_xmote_th - Call into the lock module to acquire or change a glock
 921  * @gl: The glock in question
 922  * @state: the requested state
 923  * @flags: modifier flags to the lock call
 924  *
 925  */
 926
 927 void gfs2_glock_xmote_th(struct gfs2_glock *gl, unsigned int state, int flags)
 928 {
 929         struct gfs2_sbd *sdp = gl->gl_sbd;
 930         const struct gfs2_glock_operations *glops = gl->gl_ops;
 931         int lck_flags = flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB |
 932                                  LM_FLAG_NOEXP | LM_FLAG_ANY |
 933                                  LM_FLAG_PRIORITY);
 934         unsigned int lck_ret;
 935
 936         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
 937         gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
 938         gfs2_assert_warn(sdp, state != LM_ST_UNLOCKED);
 939         gfs2_assert_warn(sdp, state != gl->gl_state);
 940
 941         if (gl->gl_state == LM_ST_EXCLUSIVE && glops->go_sync)
 942                 glops->go_sync(gl);
 943
 944         gfs2_glock_hold(gl);
 945         gl->gl_req_bh = xmote_bh;
 946
 947         lck_ret = gfs2_lm_lock(sdp, gl->gl_lock, gl->gl_state, state, lck_flags);
 948
 949         if (gfs2_assert_withdraw(sdp, !(lck_ret & LM_OUT_ERROR)))
 950                 return;
 951
 952         if (lck_ret & LM_OUT_ASYNC)
 953                 gfs2_assert_warn(sdp, lck_ret == LM_OUT_ASYNC);
 954         else
 955                 xmote_bh(gl, lck_ret);
 956 }
 957
 958 /**
 959  * drop_bh - Called after a lock module unlock completes
 960  * @gl: the glock
 961  * @ret: the return status
 962  *
 963  * Doesn't wake up the process waiting on the struct gfs2_holder (if any)
 964  * Doesn't drop the reference on the glock the top half took out
 965  *
 966  */
 967
 968 static void drop_bh(struct gfs2_glock *gl, unsigned int ret)
 969 {
 970         struct gfs2_sbd *sdp = gl->gl_sbd;
 971         const struct gfs2_glock_operations *glops = gl->gl_ops;
 972         struct gfs2_holder *gh = gl->gl_req_gh;
 973
 974         clear_bit(GLF_PREFETCH, &gl->gl_flags);
 975
 976         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
 977         gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
 978         gfs2_assert_warn(sdp, !ret);
 979
 980         state_change(gl, LM_ST_UNLOCKED);
 981
 982         if (glops->go_inval)
 983                 glops->go_inval(gl, DIO_METADATA);
 984
 985         if (gh) {
 986                 spin_lock(&gl->gl_spin);
 987                 list_del_init(&gh->gh_list);
 988                 gh->gh_error = 0;
 989                 spin_unlock(&gl->gl_spin);
 990         }
 991
 992         if (glops->go_drop_bh)
 993                 glops->go_drop_bh(gl);
 994
 995         spin_lock(&gl->gl_spin);
 996         gl->gl_req_gh = NULL;
 997         gl->gl_req_bh = NULL;
 998         clear_bit(GLF_LOCK, &gl->gl_flags);
 999         run_queue(gl);
1000         spin_unlock(&gl->gl_spin);
1001
1002         gfs2_glock_put(gl);
1003
1004         if (gh) {
1005                 if (test_bit(HIF_DEALLOC, &gh->gh_iflags))
1006                         gfs2_holder_put(gh);
1007                 else
1008                         complete(&gh->gh_wait);
1009         }
1010 }
1011
1012 /**
1013  * gfs2_glock_drop_th - call into the lock module to unlock a lock
1014  * @gl: the glock
1015  *
1016  */
1017
1018 void gfs2_glock_drop_th(struct gfs2_glock *gl)
1019 {
1020         struct gfs2_sbd *sdp = gl->gl_sbd;
1021         const struct gfs2_glock_operations *glops = gl->gl_ops;
1022         unsigned int ret;
1023
1024         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1025         gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
1026         gfs2_assert_warn(sdp, gl->gl_state != LM_ST_UNLOCKED);
1027
1028         if (gl->gl_state == LM_ST_EXCLUSIVE && glops->go_sync)
1029                 glops->go_sync(gl);
1030
1031         gfs2_glock_hold(gl);
1032         gl->gl_req_bh = drop_bh;
1033
1034         ret = gfs2_lm_unlock(sdp, gl->gl_lock, gl->gl_state);
1035
1036         if (gfs2_assert_withdraw(sdp, !(ret & LM_OUT_ERROR)))
1037                 return;
1038
1039         if (!ret)
1040                 drop_bh(gl, ret);
1041         else
1042                 gfs2_assert_warn(sdp, ret == LM_OUT_ASYNC);
1043 }
1044
1045 /**
1046  * do_cancels - cancel requests for locks stuck waiting on an expire flag
1047  * @gh: the LM_FLAG_PRIORITY holder waiting to acquire the lock
1048  *
1049  * Don't cancel GL_NOCANCEL requests.
1050  */
1051
1052 static void do_cancels(struct gfs2_holder *gh)
1053 {
1054         struct gfs2_glock *gl = gh->gh_gl;
1055
1056         spin_lock(&gl->gl_spin);
1057
1058         while (gl->gl_req_gh != gh &&
1059                !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1060                !list_empty(&gh->gh_list)) {
1061                 if (gl->gl_req_bh && !(gl->gl_req_gh &&
1062                                      (gl->gl_req_gh->gh_flags & GL_NOCANCEL))) {
1063                         spin_unlock(&gl->gl_spin);
1064                         gfs2_lm_cancel(gl->gl_sbd, gl->gl_lock);
1065                         msleep(100);
1066                         spin_lock(&gl->gl_spin);
1067                 } else {
1068                         spin_unlock(&gl->gl_spin);
1069                         msleep(100);
1070                         spin_lock(&gl->gl_spin);
1071                 }
1072         }
1073
1074         spin_unlock(&gl->gl_spin);
1075 }
1076
1077 /**
1078  * glock_wait_internal - wait on a glock acquisition
1079  * @gh: the glock holder
1080  *
1081  * Returns: 0 on success
1082  */
1083
1084 static int glock_wait_internal(struct gfs2_holder *gh)
1085 {
1086         struct gfs2_glock *gl = gh->gh_gl;
1087         struct gfs2_sbd *sdp = gl->gl_sbd;
1088         const struct gfs2_glock_operations *glops = gl->gl_ops;
1089
1090         if (test_bit(HIF_ABORTED, &gh->gh_iflags))
1091                 return -EIO;
1092
1093         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1094                 spin_lock(&gl->gl_spin);
1095                 if (gl->gl_req_gh != gh &&
1096                     !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1097                     !list_empty(&gh->gh_list)) {
1098                         list_del_init(&gh->gh_list);
1099                         gh->gh_error = GLR_TRYFAILED;
1100                         run_queue(gl);
1101                         spin_unlock(&gl->gl_spin);
1102                         return gh->gh_error;
1103                 }
1104                 spin_unlock(&gl->gl_spin);
1105         }
1106
1107         if (gh->gh_flags & LM_FLAG_PRIORITY)
1108                 do_cancels(gh);
1109
1110         wait_for_completion(&gh->gh_wait);
1111
1112         if (gh->gh_error)
1113                 return gh->gh_error;
1114
1115         gfs2_assert_withdraw(sdp, test_bit(HIF_HOLDER, &gh->gh_iflags));
1116         gfs2_assert_withdraw(sdp, relaxed_state_ok(gl->gl_state, gh->gh_state,
1117                                                    gh->gh_flags));
1118
1119         if (test_bit(HIF_FIRST, &gh->gh_iflags)) {
1120                 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1121
1122                 if (glops->go_lock) {
1123                         gh->gh_error = glops->go_lock(gh);
1124                         if (gh->gh_error) {
1125                                 spin_lock(&gl->gl_spin);
1126                                 list_del_init(&gh->gh_list);
1127                                 spin_unlock(&gl->gl_spin);
1128                         }
1129                 }
1130
1131                 spin_lock(&gl->gl_spin);
1132                 gl->gl_req_gh = NULL;
1133                 gl->gl_req_bh = NULL;
1134                 clear_bit(GLF_LOCK, &gl->gl_flags);
1135                 run_queue(gl);
1136                 spin_unlock(&gl->gl_spin);
1137         }
1138
1139         return gh->gh_error;
1140 }
1141
1142 static inline struct gfs2_holder *
1143 find_holder_by_owner(struct list_head *head, struct task_struct *owner)
1144 {
1145         struct gfs2_holder *gh;
1146
1147         list_for_each_entry(gh, head, gh_list) {
1148                 if (gh->gh_owner == owner)
1149                         return gh;
1150         }
1151
1152         return NULL;
1153 }
1154
1155 /**
1156  * add_to_queue - Add a holder to the wait queue (but look for recursion)
1157  * @gh: the holder structure to add
1158  *
1159  */
1160
1161 static void add_to_queue(struct gfs2_holder *gh)
1162 {
1163         struct gfs2_glock *gl = gh->gh_gl;
1164         struct gfs2_holder *existing;
1165
1166         BUG_ON(!gh->gh_owner);
1167
1168         existing = find_holder_by_owner(&gl->gl_holders, gh->gh_owner);
1169         if (existing) {
1170                 print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
1171                 printk(KERN_INFO "pid : %d\n", existing->gh_owner->pid);
1172                 printk(KERN_INFO "lock type : %d lock state : %d\n",
1173                                 existing->gh_gl->gl_name.ln_type, existing->gh_gl->gl_state);
1174                 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1175                 printk(KERN_INFO "pid : %d\n", gh->gh_owner->pid);
1176                 printk(KERN_INFO "lock type : %d lock state : %d\n",
1177                                 gl->gl_name.ln_type, gl->gl_state);
1178                 BUG();
1179         }
1180
1181         existing = find_holder_by_owner(&gl->gl_waiters3, gh->gh_owner);
1182         if (existing) {
1183                 print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
1184                 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1185                 BUG();
1186         }
1187
1188         if (gh->gh_flags & LM_FLAG_PRIORITY)
1189                 list_add(&gh->gh_list, &gl->gl_waiters3);
1190         else
1191                 list_add_tail(&gh->gh_list, &gl->gl_waiters3);
1192 }
1193
1194 /**
1195  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1196  * @gh: the holder structure
1197  *
1198  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1199  *
1200  * Returns: 0, GLR_TRYFAILED, or errno on failure
1201  */
1202
1203 int gfs2_glock_nq(struct gfs2_holder *gh)
1204 {
1205         struct gfs2_glock *gl = gh->gh_gl;
1206         struct gfs2_sbd *sdp = gl->gl_sbd;
1207         int error = 0;
1208
1209 restart:
1210         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1211                 set_bit(HIF_ABORTED, &gh->gh_iflags);
1212                 return -EIO;
1213         }
1214
1215         set_bit(HIF_PROMOTE, &gh->gh_iflags);
1216
1217         spin_lock(&gl->gl_spin);
1218         add_to_queue(gh);
1219         run_queue(gl);
1220         spin_unlock(&gl->gl_spin);
1221
1222         if (!(gh->gh_flags & GL_ASYNC)) {
1223                 error = glock_wait_internal(gh);
1224                 if (error == GLR_CANCELED) {
1225                         msleep(100);
1226                         goto restart;
1227                 }
1228         }
1229
1230         clear_bit(GLF_PREFETCH, &gl->gl_flags);
1231
1232         return error;
1233 }
1234
1235 /**
1236  * gfs2_glock_poll - poll to see if an async request has been completed
1237  * @gh: the holder
1238  *
1239  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1240  */
1241
1242 int gfs2_glock_poll(struct gfs2_holder *gh)
1243 {
1244         struct gfs2_glock *gl = gh->gh_gl;
1245         int ready = 0;
1246
1247         spin_lock(&gl->gl_spin);
1248
1249         if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1250                 ready = 1;
1251         else if (list_empty(&gh->gh_list)) {
1252                 if (gh->gh_error == GLR_CANCELED) {
1253                         spin_unlock(&gl->gl_spin);
1254                         msleep(100);
1255                         if (gfs2_glock_nq(gh))
1256                                 return 1;
1257                         return 0;
1258                 } else
1259                         ready = 1;
1260         }
1261
1262         spin_unlock(&gl->gl_spin);
1263
1264         return ready;
1265 }
1266
1267 /**
1268  * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
1269  * @gh: the holder structure
1270  *
1271  * Returns: 0, GLR_TRYFAILED, or errno on failure
1272  */
1273
1274 int gfs2_glock_wait(struct gfs2_holder *gh)
1275 {
1276         int error;
1277
1278         error = glock_wait_internal(gh);
1279         if (error == GLR_CANCELED) {
1280                 msleep(100);
1281                 gh->gh_flags &= ~GL_ASYNC;
1282                 error = gfs2_glock_nq(gh);
1283         }
1284
1285         return error;
1286 }
1287
1288 /**
1289  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1290  * @gh: the glock holder
1291  *
1292  */
1293
1294 void gfs2_glock_dq(struct gfs2_holder *gh)
1295 {
1296         struct gfs2_glock *gl = gh->gh_gl;
1297         const struct gfs2_glock_operations *glops = gl->gl_ops;
1298
1299         if (gh->gh_flags & GL_NOCACHE)
1300                 handle_callback(gl, LM_ST_UNLOCKED);
1301
1302         gfs2_glmutex_lock(gl);
1303
1304         spin_lock(&gl->gl_spin);
1305         list_del_init(&gh->gh_list);
1306
1307         if (list_empty(&gl->gl_holders)) {
1308                 spin_unlock(&gl->gl_spin);
1309
1310                 if (glops->go_unlock)
1311                         glops->go_unlock(gh);
1312
1313                 gl->gl_stamp = jiffies;
1314
1315                 spin_lock(&gl->gl_spin);
1316         }
1317
1318         clear_bit(GLF_LOCK, &gl->gl_flags);
1319         run_queue(gl);
1320         spin_unlock(&gl->gl_spin);
1321 }
1322
1323 /**
1324  * gfs2_glock_prefetch - Try to prefetch a glock
1325  * @gl: the glock
1326  * @state: the state to prefetch in
1327  * @flags: flags passed to go_xmote_th()
1328  *
1329  */
1330
1331 static void gfs2_glock_prefetch(struct gfs2_glock *gl, unsigned int state,
1332                                 int flags)
1333 {
1334         const struct gfs2_glock_operations *glops = gl->gl_ops;
1335
1336         spin_lock(&gl->gl_spin);
1337
1338         if (test_bit(GLF_LOCK, &gl->gl_flags) || !list_empty(&gl->gl_holders) ||
1339             !list_empty(&gl->gl_waiters1) || !list_empty(&gl->gl_waiters2) ||
1340             !list_empty(&gl->gl_waiters3) ||
1341             relaxed_state_ok(gl->gl_state, state, flags)) {
1342                 spin_unlock(&gl->gl_spin);
1343                 return;
1344         }
1345
1346         set_bit(GLF_PREFETCH, &gl->gl_flags);
1347         set_bit(GLF_LOCK, &gl->gl_flags);
1348         spin_unlock(&gl->gl_spin);
1349
1350         glops->go_xmote_th(gl, state, flags);
1351 }
1352
1353 static void greedy_work(struct work_struct *work)
1354 {
1355         struct greedy *gr = container_of(work, struct greedy, gr_work.work);
1356         struct gfs2_holder *gh = &gr->gr_gh;
1357         struct gfs2_glock *gl = gh->gh_gl;
1358         const struct gfs2_glock_operations *glops = gl->gl_ops;
1359
1360         clear_bit(GLF_SKIP_WAITERS2, &gl->gl_flags);
1361
1362         if (glops->go_greedy)
1363                 glops->go_greedy(gl);
1364
1365         spin_lock(&gl->gl_spin);
1366
1367         if (list_empty(&gl->gl_waiters2)) {
1368                 clear_bit(GLF_GREEDY, &gl->gl_flags);
1369                 spin_unlock(&gl->gl_spin);
1370                 gfs2_holder_uninit(gh);
1371                 kfree(gr);
1372         } else {
1373                 gfs2_glock_hold(gl);
1374                 list_add_tail(&gh->gh_list, &gl->gl_waiters2);
1375                 run_queue(gl);
1376                 spin_unlock(&gl->gl_spin);
1377                 gfs2_glock_put(gl);
1378         }
1379 }
1380
1381 /**
1382  * gfs2_glock_be_greedy -
1383  * @gl:
1384  * @time:
1385  *
1386  * Returns: 0 if go_greedy will be called, 1 otherwise
1387  */
1388
1389 int gfs2_glock_be_greedy(struct gfs2_glock *gl, unsigned int time)
1390 {
1391         struct greedy *gr;
1392         struct gfs2_holder *gh;
1393
1394         if (!time || gl->gl_sbd->sd_args.ar_localcaching ||
1395             test_and_set_bit(GLF_GREEDY, &gl->gl_flags))
1396                 return 1;
1397
1398         gr = kmalloc(sizeof(struct greedy), GFP_KERNEL);
1399         if (!gr) {
1400                 clear_bit(GLF_GREEDY, &gl->gl_flags);
1401                 return 1;
1402         }
1403         gh = &gr->gr_gh;
1404
1405         gfs2_holder_init(gl, 0, 0, gh);
1406         set_bit(HIF_GREEDY, &gh->gh_iflags);
1407         INIT_DELAYED_WORK(&gr->gr_work, greedy_work);
1408
1409         set_bit(GLF_SKIP_WAITERS2, &gl->gl_flags);
1410         schedule_delayed_work(&gr->gr_work, time);
1411
1412         return 0;
1413 }
1414
1415 /**
1416  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1417  * @gh: the holder structure
1418  *
1419  */
1420
1421 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1422 {
1423         gfs2_glock_dq(gh);
1424         gfs2_holder_uninit(gh);
1425 }
1426
1427 /**
1428  * gfs2_glock_nq_num - acquire a glock based on lock number
1429  * @sdp: the filesystem
1430  * @number: the lock number
1431  * @glops: the glock operations for the type of glock
1432  * @state: the state to acquire the glock in
1433  * @flags: modifier flags for the aquisition
1434  * @gh: the struct gfs2_holder
1435  *
1436  * Returns: errno
1437  */
1438
1439 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1440                       const struct gfs2_glock_operations *glops,
1441                       unsigned int state, int flags, struct gfs2_holder *gh)
1442 {
1443         struct gfs2_glock *gl;
1444         int error;
1445
1446         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1447         if (!error) {
1448                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1449                 gfs2_glock_put(gl);
1450         }
1451
1452         return error;
1453 }
1454
1455 /**
1456  * glock_compare - Compare two struct gfs2_glock structures for sorting
1457  * @arg_a: the first structure
1458  * @arg_b: the second structure
1459  *
1460  */
1461
1462 static int glock_compare(const void *arg_a, const void *arg_b)
1463 {
1464         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1465         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1466         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1467         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1468
1469         if (a->ln_number > b->ln_number)
1470                 return 1;
1471         if (a->ln_number < b->ln_number)
1472                 return -1;
1473         if (gh_a->gh_state == LM_ST_SHARED && gh_b->gh_state == LM_ST_EXCLUSIVE)
1474                 return 1;
1475         if (!(gh_a->gh_flags & GL_LOCAL_EXCL) && (gh_b->gh_flags & GL_LOCAL_EXCL))
1476                 return 1;
1477         return 0;
1478 }
1479
1480 /**
1481  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1482  * @num_gh: the number of structures
1483  * @ghs: an array of struct gfs2_holder structures
1484  *
1485  * Returns: 0 on success (all glocks acquired),
1486  *          errno on failure (no glocks acquired)
1487  */
1488
1489 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1490                      struct gfs2_holder **p)
1491 {
1492         unsigned int x;
1493         int error = 0;
1494
1495         for (x = 0; x < num_gh; x++)
1496                 p[x] = &ghs[x];
1497
1498         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1499
1500         for (x = 0; x < num_gh; x++) {
1501                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1502
1503                 error = gfs2_glock_nq(p[x]);
1504                 if (error) {
1505                         while (x--)
1506                                 gfs2_glock_dq(p[x]);
1507                         break;
1508                 }
1509         }
1510
1511         return error;
1512 }
1513
1514 /**
1515  * gfs2_glock_nq_m - acquire multiple glocks
1516  * @num_gh: the number of structures
1517  * @ghs: an array of struct gfs2_holder structures
1518  *
1519  * Figure out how big an impact this function has.  Either:
1520  * 1) Replace this code with code that calls gfs2_glock_prefetch()
1521  * 2) Forget async stuff and just call nq_m_sync()
1522  * 3) Leave it like it is
1523  *
1524  * Returns: 0 on success (all glocks acquired),
1525  *          errno on failure (no glocks acquired)
1526  */
1527
1528 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1529 {
1530         int *e;
1531         unsigned int x;
1532         int borked = 0, serious = 0;
1533         int error = 0;
1534
1535         if (!num_gh)
1536                 return 0;
1537
1538         if (num_gh == 1) {
1539                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1540                 return gfs2_glock_nq(ghs);
1541         }
1542
1543         e = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL);
1544         if (!e)
1545                 return -ENOMEM;
1546
1547         for (x = 0; x < num_gh; x++) {
1548                 ghs[x].gh_flags |= LM_FLAG_TRY | GL_ASYNC;
1549                 error = gfs2_glock_nq(&ghs[x]);
1550                 if (error) {
1551                         borked = 1;
1552                         serious = error;
1553                         num_gh = x;
1554                         break;
1555                 }
1556         }
1557
1558         for (x = 0; x < num_gh; x++) {
1559                 error = e[x] = glock_wait_internal(&ghs[x]);
1560                 if (error) {
1561                         borked = 1;
1562                         if (error != GLR_TRYFAILED && error != GLR_CANCELED)
1563                                 serious = error;
1564                 }
1565         }
1566
1567         if (!borked) {
1568                 kfree(e);
1569                 return 0;
1570         }
1571
1572         for (x = 0; x < num_gh; x++)
1573                 if (!e[x])
1574                         gfs2_glock_dq(&ghs[x]);
1575
1576         if (serious)
1577                 error = serious;
1578         else {
1579                 for (x = 0; x < num_gh; x++)
1580                         gfs2_holder_reinit(ghs[x].gh_state, ghs[x].gh_flags,
1581                                           &ghs[x]);
1582                 error = nq_m_sync(num_gh, ghs, (struct gfs2_holder **)e);
1583         }
1584
1585         kfree(e);
1586
1587         return error;
1588 }
1589
1590 /**
1591  * gfs2_glock_dq_m - release multiple glocks
1592  * @num_gh: the number of structures
1593  * @ghs: an array of struct gfs2_holder structures
1594  *
1595  */
1596
1597 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1598 {
1599         unsigned int x;
1600
1601         for (x = 0; x < num_gh; x++)
1602                 gfs2_glock_dq(&ghs[x]);
1603 }
1604
1605 /**
1606  * gfs2_glock_dq_uninit_m - release multiple glocks
1607  * @num_gh: the number of structures
1608  * @ghs: an array of struct gfs2_holder structures
1609  *
1610  */
1611
1612 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1613 {
1614         unsigned int x;
1615
1616         for (x = 0; x < num_gh; x++)
1617                 gfs2_glock_dq_uninit(&ghs[x]);
1618 }
1619
1620 /**
1621  * gfs2_glock_prefetch_num - prefetch a glock based on lock number
1622  * @sdp: the filesystem
1623  * @number: the lock number
1624  * @glops: the glock operations for the type of glock
1625  * @state: the state to acquire the glock in
1626  * @flags: modifier flags for the aquisition
1627  *
1628  * Returns: errno
1629  */
1630
1631 void gfs2_glock_prefetch_num(struct gfs2_sbd *sdp, u64 number,
1632                              const struct gfs2_glock_operations *glops,
1633                              unsigned int state, int flags)
1634 {
1635         struct gfs2_glock *gl;
1636         int error;
1637
1638         if (atomic_read(&sdp->sd_reclaim_count) <
1639             gfs2_tune_get(sdp, gt_reclaim_limit)) {
1640                 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1641                 if (!error) {
1642                         gfs2_glock_prefetch(gl, state, flags);
1643                         gfs2_glock_put(gl);
1644                 }
1645         }
1646 }
1647
1648 /**
1649  * gfs2_lvb_hold - attach a LVB from a glock
1650  * @gl: The glock in question
1651  *
1652  */
1653
1654 int gfs2_lvb_hold(struct gfs2_glock *gl)
1655 {
1656         int error;
1657
1658         gfs2_glmutex_lock(gl);
1659
1660         if (!atomic_read(&gl->gl_lvb_count)) {
1661                 error = gfs2_lm_hold_lvb(gl->gl_sbd, gl->gl_lock, &gl->gl_lvb);
1662                 if (error) {
1663                         gfs2_glmutex_unlock(gl);
1664                         return error;
1665                 }
1666                 gfs2_glock_hold(gl);
1667         }
1668         atomic_inc(&gl->gl_lvb_count);
1669
1670         gfs2_glmutex_unlock(gl);
1671
1672         return 0;
1673 }
1674
1675 /**
1676  * gfs2_lvb_unhold - detach a LVB from a glock
1677  * @gl: The glock in question
1678  *
1679  */
1680
1681 void gfs2_lvb_unhold(struct gfs2_glock *gl)
1682 {
1683         gfs2_glock_hold(gl);
1684         gfs2_glmutex_lock(gl);
1685
1686         gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count) > 0);
1687         if (atomic_dec_and_test(&gl->gl_lvb_count)) {
1688                 gfs2_lm_unhold_lvb(gl->gl_sbd, gl->gl_lock, gl->gl_lvb);
1689                 gl->gl_lvb = NULL;
1690                 gfs2_glock_put(gl);
1691         }
1692
1693         gfs2_glmutex_unlock(gl);
1694         gfs2_glock_put(gl);
1695 }
1696
1697 static void blocking_cb(struct gfs2_sbd *sdp, struct lm_lockname *name,
1698                         unsigned int state)
1699 {
1700         struct gfs2_glock *gl;
1701
1702         gl = gfs2_glock_find(sdp, name);
1703         if (!gl)
1704                 return;
1705
1706         if (gl->gl_ops->go_callback)
1707                 gl->gl_ops->go_callback(gl, state);
1708         handle_callback(gl, state);
1709
1710         spin_lock(&gl->gl_spin);
1711         run_queue(gl);
1712         spin_unlock(&gl->gl_spin);
1713
1714         gfs2_glock_put(gl);
1715 }
1716
1717 /**
1718  * gfs2_glock_cb - Callback used by locking module
1719  * @sdp: Pointer to the superblock
1720  * @type: Type of callback
1721  * @data: Type dependent data pointer
1722  *
1723  * Called by the locking module when it wants to tell us something.
1724  * Either we need to drop a lock, one of our ASYNC requests completed, or
1725  * a journal from another client needs to be recovered.
1726  */
1727
1728 void gfs2_glock_cb(void *cb_data, unsigned int type, void *data)
1729 {
1730         struct gfs2_sbd *sdp = cb_data;
1731
1732         switch (type) {
1733         case LM_CB_NEED_E:
1734                 blocking_cb(sdp, data, LM_ST_UNLOCKED);
1735                 return;
1736
1737         case LM_CB_NEED_D:
1738                 blocking_cb(sdp, data, LM_ST_DEFERRED);
1739                 return;
1740
1741         case LM_CB_NEED_S:
1742                 blocking_cb(sdp, data, LM_ST_SHARED);
1743                 return;
1744
1745         case LM_CB_ASYNC: {
1746                 struct lm_async_cb *async = data;
1747                 struct gfs2_glock *gl;
1748
1749                 gl = gfs2_glock_find(sdp, &async->lc_name);
1750                 if (gfs2_assert_warn(sdp, gl))
1751                         return;
1752                 if (!gfs2_assert_warn(sdp, gl->gl_req_bh))
1753                         gl->gl_req_bh(gl, async->lc_ret);
1754                 gfs2_glock_put(gl);
1755                 return;
1756         }
1757
1758         case LM_CB_NEED_RECOVERY:
1759                 gfs2_jdesc_make_dirty(sdp, *(unsigned int *)data);
1760                 if (sdp->sd_recoverd_process)
1761                         wake_up_process(sdp->sd_recoverd_process);
1762                 return;
1763
1764         case LM_CB_DROPLOCKS:
1765                 gfs2_gl_hash_clear(sdp, NO_WAIT);
1766                 gfs2_quota_scan(sdp);
1767                 return;
1768
1769         default:
1770                 gfs2_assert_warn(sdp, 0);
1771                 return;
1772         }
1773 }
1774
1775 /**
1776  * demote_ok - Check to see if it's ok to unlock a glock
1777  * @gl: the glock
1778  *
1779  * Returns: 1 if it's ok
1780  */
1781
1782 static int demote_ok(struct gfs2_glock *gl)
1783 {
1784         struct gfs2_sbd *sdp = gl->gl_sbd;
1785         const struct gfs2_glock_operations *glops = gl->gl_ops;
1786         int demote = 1;
1787
1788         if (test_bit(GLF_STICKY, &gl->gl_flags))
1789                 demote = 0;
1790         else if (test_bit(GLF_PREFETCH, &gl->gl_flags))
1791                 demote = time_after_eq(jiffies, gl->gl_stamp +
1792                                     gfs2_tune_get(sdp, gt_prefetch_secs) * HZ);
1793         else if (glops->go_demote_ok)
1794                 demote = glops->go_demote_ok(gl);
1795
1796         return demote;
1797 }
1798
1799 /**
1800  * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1801  * @gl: the glock
1802  *
1803  */
1804
1805 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
1806 {
1807         struct gfs2_sbd *sdp = gl->gl_sbd;
1808
1809         spin_lock(&sdp->sd_reclaim_lock);
1810         if (list_empty(&gl->gl_reclaim)) {
1811                 gfs2_glock_hold(gl);
1812                 list_add(&gl->gl_reclaim, &sdp->sd_reclaim_list);
1813                 atomic_inc(&sdp->sd_reclaim_count);
1814         }
1815         spin_unlock(&sdp->sd_reclaim_lock);
1816
1817         wake_up(&sdp->sd_reclaim_wq);
1818 }
1819
1820 /**
1821  * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1822  * @sdp: the filesystem
1823  *
1824  * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1825  * different glock and we notice that there are a lot of glocks in the
1826  * reclaim list.
1827  *
1828  */
1829
1830 void gfs2_reclaim_glock(struct gfs2_sbd *sdp)
1831 {
1832         struct gfs2_glock *gl;
1833
1834         spin_lock(&sdp->sd_reclaim_lock);
1835         if (list_empty(&sdp->sd_reclaim_list)) {
1836                 spin_unlock(&sdp->sd_reclaim_lock);
1837                 return;
1838         }
1839         gl = list_entry(sdp->sd_reclaim_list.next,
1840                         struct gfs2_glock, gl_reclaim);
1841         list_del_init(&gl->gl_reclaim);
1842         spin_unlock(&sdp->sd_reclaim_lock);
1843
1844         atomic_dec(&sdp->sd_reclaim_count);
1845         atomic_inc(&sdp->sd_reclaimed);
1846
1847         if (gfs2_glmutex_trylock(gl)) {
1848                 if (queue_empty(gl, &gl->gl_holders) &&
1849                     gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1850                         handle_callback(gl, LM_ST_UNLOCKED);
1851                 gfs2_glmutex_unlock(gl);
1852         }
1853
1854         gfs2_glock_put(gl);
1855 }
1856
1857 /**
1858  * examine_bucket - Call a function for glock in a hash bucket
1859  * @examiner: the function
1860  * @sdp: the filesystem
1861  * @bucket: the bucket
1862  *
1863  * Returns: 1 if the bucket has entries
1864  */
1865
1866 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1867                           unsigned int hash)
1868 {
1869         struct gfs2_glock *gl, *prev = NULL;
1870         int has_entries = 0;
1871         struct hlist_head *head = &gl_hash_table[hash].hb_list;
1872
1873         read_lock(gl_lock_addr(hash));
1874         /* Can't use hlist_for_each_entry - don't want prefetch here */
1875         if (hlist_empty(head))
1876                 goto out;
1877         gl = list_entry(head->first, struct gfs2_glock, gl_list);
1878         while(1) {
1879                 if (gl->gl_sbd == sdp) {
1880                         gfs2_glock_hold(gl);
1881                         read_unlock(gl_lock_addr(hash));
1882                         if (prev)
1883                                 gfs2_glock_put(prev);
1884                         prev = gl;
1885                         examiner(gl);
1886                         has_entries = 1;
1887                         read_lock(gl_lock_addr(hash));
1888                 }
1889                 if (gl->gl_list.next == NULL)
1890                         break;
1891                 gl = list_entry(gl->gl_list.next, struct gfs2_glock, gl_list);
1892         }
1893 out:
1894         read_unlock(gl_lock_addr(hash));
1895         if (prev)
1896                 gfs2_glock_put(prev);
1897         return has_entries;
1898 }
1899
1900 /**
1901  * scan_glock - look at a glock and see if we can reclaim it
1902  * @gl: the glock to look at
1903  *
1904  */
1905
1906 static void scan_glock(struct gfs2_glock *gl)
1907 {
1908         if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object)
1909                 return;
1910
1911         if (gfs2_glmutex_trylock(gl)) {
1912                 if (queue_empty(gl, &gl->gl_holders) &&
1913                     gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1914                         goto out_schedule;
1915                 gfs2_glmutex_unlock(gl);
1916         }
1917         return;
1918
1919 out_schedule:
1920         gfs2_glmutex_unlock(gl);
1921         gfs2_glock_schedule_for_reclaim(gl);
1922 }
1923
1924 /**
1925  * gfs2_scand_internal - Look for glocks and inodes to toss from memory
1926  * @sdp: the filesystem
1927  *
1928  */
1929
1930 void gfs2_scand_internal(struct gfs2_sbd *sdp)
1931 {
1932         unsigned int x;
1933
1934         for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1935                 examine_bucket(scan_glock, sdp, x);
1936 }
1937
1938 /**
1939  * clear_glock - look at a glock and see if we can free it from glock cache
1940  * @gl: the glock to look at
1941  *
1942  */
1943
1944 static void clear_glock(struct gfs2_glock *gl)
1945 {
1946         struct gfs2_sbd *sdp = gl->gl_sbd;
1947         int released;
1948
1949         spin_lock(&sdp->sd_reclaim_lock);
1950         if (!list_empty(&gl->gl_reclaim)) {
1951                 list_del_init(&gl->gl_reclaim);
1952                 atomic_dec(&sdp->sd_reclaim_count);
1953                 spin_unlock(&sdp->sd_reclaim_lock);
1954                 released = gfs2_glock_put(gl);
1955                 gfs2_assert(sdp, !released);
1956         } else {
1957                 spin_unlock(&sdp->sd_reclaim_lock);
1958         }
1959
1960         if (gfs2_glmutex_trylock(gl)) {
1961                 if (queue_empty(gl, &gl->gl_holders) &&
1962                     gl->gl_state != LM_ST_UNLOCKED)
1963                         handle_callback(gl, LM_ST_UNLOCKED);
1964                 gfs2_glmutex_unlock(gl);
1965         }
1966 }
1967
1968 /**
1969  * gfs2_gl_hash_clear - Empty out the glock hash table
1970  * @sdp: the filesystem
1971  * @wait: wait until it's all gone
1972  *
1973  * Called when unmounting the filesystem, or when inter-node lock manager
1974  * requests DROPLOCKS because it is running out of capacity.
1975  */
1976
1977 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp, int wait)
1978 {
1979         unsigned long t;
1980         unsigned int x;
1981         int cont;
1982
1983         t = jiffies;
1984
1985         for (;;) {
1986                 cont = 0;
1987                 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1988                         if (examine_bucket(clear_glock, sdp, x))
1989                                 cont = 1;
1990                 }
1991
1992                 if (!wait || !cont)
1993                         break;
1994
1995                 if (time_after_eq(jiffies,
1996                                   t + gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
1997                         fs_warn(sdp, "Unmount seems to be stalled. "
1998                                      "Dumping lock state...\n");
1999                         gfs2_dump_lockstate(sdp);
2000                         t = jiffies;
2001                 }
2002
2003                 invalidate_inodes(sdp->sd_vfs);
2004                 msleep(10);
2005         }
2006 }
2007
2008 /*
2009  *  Diagnostic routines to help debug distributed deadlock
2010  */
2011
2012 /**
2013  * dump_holder - print information about a glock holder
2014  * @str: a string naming the type of holder
2015  * @gh: the glock holder
2016  *
2017  * Returns: 0 on success, -ENOBUFS when we run out of space
2018  */
2019
2020 static int dump_holder(char *str, struct gfs2_holder *gh)
2021 {
2022         unsigned int x;
2023         int error = -ENOBUFS;
2024
2025         printk(KERN_INFO "  %s\n", str);
2026         printk(KERN_INFO "    owner = %ld\n",
2027                    (gh->gh_owner) ? (long)gh->gh_owner->pid : -1);
2028         printk(KERN_INFO "    gh_state = %u\n", gh->gh_state);
2029         printk(KERN_INFO "    gh_flags =");
2030         for (x = 0; x < 32; x++)
2031                 if (gh->gh_flags & (1 << x))
2032                         printk(" %u", x);
2033         printk(" \n");
2034         printk(KERN_INFO "    error = %d\n", gh->gh_error);
2035         printk(KERN_INFO "    gh_iflags =");
2036         for (x = 0; x < 32; x++)
2037                 if (test_bit(x, &gh->gh_iflags))
2038                         printk(" %u", x);
2039         printk(" \n");
2040         print_symbol(KERN_INFO "    initialized at: %s\n", gh->gh_ip);
2041
2042         error = 0;
2043
2044         return error;
2045 }
2046
2047 /**
2048  * dump_inode - print information about an inode
2049  * @ip: the inode
2050  *
2051  * Returns: 0 on success, -ENOBUFS when we run out of space
2052  */
2053
2054 static int dump_inode(struct gfs2_inode *ip)
2055 {
2056         unsigned int x;
2057         int error = -ENOBUFS;
2058
2059         printk(KERN_INFO "  Inode:\n");
2060         printk(KERN_INFO "    num = %llu %llu\n",
2061                     (unsigned long long)ip->i_num.no_formal_ino,
2062                     (unsigned long long)ip->i_num.no_addr);
2063         printk(KERN_INFO "    type = %u\n", IF2DT(ip->i_inode.i_mode));
2064         printk(KERN_INFO "    i_flags =");
2065         for (x = 0; x < 32; x++)
2066                 if (test_bit(x, &ip->i_flags))
2067                         printk(" %u", x);
2068         printk(" \n");
2069
2070         error = 0;
2071
2072         return error;
2073 }
2074
2075 /**
2076  * dump_glock - print information about a glock
2077  * @gl: the glock
2078  * @count: where we are in the buffer
2079  *
2080  * Returns: 0 on success, -ENOBUFS when we run out of space
2081  */
2082
2083 static int dump_glock(struct gfs2_glock *gl)
2084 {
2085         struct gfs2_holder *gh;
2086         unsigned int x;
2087         int error = -ENOBUFS;
2088
2089         spin_lock(&gl->gl_spin);
2090
2091         printk(KERN_INFO "Glock 0x%p (%u, %llu)\n", gl, gl->gl_name.ln_type,
2092                (unsigned long long)gl->gl_name.ln_number);
2093         printk(KERN_INFO "  gl_flags =");
2094         for (x = 0; x < 32; x++) {
2095                 if (test_bit(x, &gl->gl_flags))
2096                         printk(" %u", x);
2097         }
2098         printk(" \n");
2099         printk(KERN_INFO "  gl_ref = %d\n", atomic_read(&gl->gl_ref));
2100         printk(KERN_INFO "  gl_state = %u\n", gl->gl_state);
2101         printk(KERN_INFO "  gl_owner = %s\n", gl->gl_owner->comm);
2102         print_symbol(KERN_INFO "  gl_ip = %s\n", gl->gl_ip);
2103         printk(KERN_INFO "  req_gh = %s\n", (gl->gl_req_gh) ? "yes" : "no");
2104         printk(KERN_INFO "  req_bh = %s\n", (gl->gl_req_bh) ? "yes" : "no");
2105         printk(KERN_INFO "  lvb_count = %d\n", atomic_read(&gl->gl_lvb_count));
2106         printk(KERN_INFO "  object = %s\n", (gl->gl_object) ? "yes" : "no");
2107         printk(KERN_INFO "  le = %s\n",
2108                    (list_empty(&gl->gl_le.le_list)) ? "no" : "yes");
2109         printk(KERN_INFO "  reclaim = %s\n",
2110                     (list_empty(&gl->gl_reclaim)) ? "no" : "yes");
2111         if (gl->gl_aspace)
2112                 printk(KERN_INFO "  aspace = 0x%p nrpages = %lu\n", gl->gl_aspace,
2113                        gl->gl_aspace->i_mapping->nrpages);
2114         else
2115                 printk(KERN_INFO "  aspace = no\n");
2116         printk(KERN_INFO "  ail = %d\n", atomic_read(&gl->gl_ail_count));
2117         if (gl->gl_req_gh) {
2118                 error = dump_holder("Request", gl->gl_req_gh);
2119                 if (error)
2120                         goto out;
2121         }
2122         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
2123                 error = dump_holder("Holder", gh);
2124                 if (error)
2125                         goto out;
2126         }
2127         list_for_each_entry(gh, &gl->gl_waiters1, gh_list) {
2128                 error = dump_holder("Waiter1", gh);
2129                 if (error)
2130                         goto out;
2131         }
2132         list_for_each_entry(gh, &gl->gl_waiters2, gh_list) {
2133                 error = dump_holder("Waiter2", gh);
2134                 if (error)
2135                         goto out;
2136         }
2137         list_for_each_entry(gh, &gl->gl_waiters3, gh_list) {
2138                 error = dump_holder("Waiter3", gh);
2139                 if (error)
2140                         goto out;
2141         }
2142         if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object) {
2143                 if (!test_bit(GLF_LOCK, &gl->gl_flags) &&
2144                     list_empty(&gl->gl_holders)) {
2145                         error = dump_inode(gl->gl_object);
2146                         if (error)
2147                                 goto out;
2148                 } else {
2149                         error = -ENOBUFS;
2150                         printk(KERN_INFO "  Inode: busy\n");
2151                 }
2152         }
2153
2154         error = 0;
2155
2156 out:
2157         spin_unlock(&gl->gl_spin);
2158         return error;
2159 }
2160
2161 /**
2162  * gfs2_dump_lockstate - print out the current lockstate
2163  * @sdp: the filesystem
2164  * @ub: the buffer to copy the information into
2165  *
2166  * If @ub is NULL, dump the lockstate to the console.
2167  *
2168  */
2169
2170 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
2171 {
2172         struct gfs2_glock *gl;
2173         struct hlist_node *h;
2174         unsigned int x;
2175         int error = 0;
2176
2177         for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
2178
2179                 read_lock(gl_lock_addr(x));
2180
2181                 hlist_for_each_entry(gl, h, &gl_hash_table[x].hb_list, gl_list) {
2182                         if (gl->gl_sbd != sdp)
2183                                 continue;
2184
2185                         error = dump_glock(gl);
2186                         if (error)
2187                                 break;
2188                 }
2189
2190                 read_unlock(gl_lock_addr(x));
2191
2192                 if (error)
2193                         break;
2194         }
2195
2196
2197         return error;
2198 }
2199
2200 int __init gfs2_glock_init(void)
2201 {
2202         unsigned i;
2203         for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
2204                 INIT_HLIST_HEAD(&gl_hash_table[i].hb_list);
2205         }
2206 #ifdef GL_HASH_LOCK_SZ
2207         for(i = 0; i < GL_HASH_LOCK_SZ; i++) {
2208                 rwlock_init(&gl_hash_locks[i]);
2209         }
2210 #endif
2211         return 0;
2212 }
2213