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 work_struct 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(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 void gfs2_glock_inode_squish(struct inode *inode)
 789 {
 790         struct gfs2_holder gh;
 791         struct gfs2_glock *gl = GFS2_I(inode)->i_gl;
 792         gfs2_holder_init(gl, LM_ST_UNLOCKED, 0, &gh);
 793         set_bit(HIF_DEMOTE, &gh.gh_iflags);
 794         spin_lock(&gl->gl_spin);
 795         gfs2_assert(inode->i_sb->s_fs_info, list_empty(&gl->gl_holders));
 796         list_add_tail(&gh.gh_list, &gl->gl_waiters2);
 797         run_queue(gl);
 798         spin_unlock(&gl->gl_spin);
 799         wait_for_completion(&gh.gh_wait);
 800         gfs2_holder_uninit(&gh);
 801 }
 802
 803 /**
 804  * state_change - record that the glock is now in a different state
 805  * @gl: the glock
 806  * @new_state the new state
 807  *
 808  */
 809
 810 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
 811 {
 812         int held1, held2;
 813
 814         held1 = (gl->gl_state != LM_ST_UNLOCKED);
 815         held2 = (new_state != LM_ST_UNLOCKED);
 816
 817         if (held1 != held2) {
 818                 if (held2)
 819                         gfs2_glock_hold(gl);
 820                 else
 821                         gfs2_glock_put(gl);
 822         }
 823
 824         gl->gl_state = new_state;
 825 }
 826
 827 /**
 828  * xmote_bh - Called after the lock module is done acquiring a lock
 829  * @gl: The glock in question
 830  * @ret: the int returned from the lock module
 831  *
 832  */
 833
 834 static void xmote_bh(struct gfs2_glock *gl, unsigned int ret)
 835 {
 836         struct gfs2_sbd *sdp = gl->gl_sbd;
 837         const struct gfs2_glock_operations *glops = gl->gl_ops;
 838         struct gfs2_holder *gh = gl->gl_req_gh;
 839         int prev_state = gl->gl_state;
 840         int op_done = 1;
 841
 842         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
 843         gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
 844         gfs2_assert_warn(sdp, !(ret & LM_OUT_ASYNC));
 845
 846         state_change(gl, ret & LM_OUT_ST_MASK);
 847
 848         if (prev_state != LM_ST_UNLOCKED && !(ret & LM_OUT_CACHEABLE)) {
 849                 if (glops->go_inval)
 850                         glops->go_inval(gl, DIO_METADATA);
 851         } else if (gl->gl_state == LM_ST_DEFERRED) {
 852                 /* We might not want to do this here.
 853                    Look at moving to the inode glops. */
 854                 if (glops->go_inval)
 855                         glops->go_inval(gl, 0);
 856         }
 857
 858         /*  Deal with each possible exit condition  */
 859
 860         if (!gh)
 861                 gl->gl_stamp = jiffies;
 862         else if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
 863                 spin_lock(&gl->gl_spin);
 864                 list_del_init(&gh->gh_list);
 865                 gh->gh_error = -EIO;
 866                 spin_unlock(&gl->gl_spin);
 867         } else if (test_bit(HIF_DEMOTE, &gh->gh_iflags)) {
 868                 spin_lock(&gl->gl_spin);
 869                 list_del_init(&gh->gh_list);
 870                 if (gl->gl_state == gh->gh_state ||
 871                     gl->gl_state == LM_ST_UNLOCKED) {
 872                         gh->gh_error = 0;
 873                 } else {
 874                         if (gfs2_assert_warn(sdp, gh->gh_flags &
 875                                         (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) == -1)
 876                                 fs_warn(sdp, "ret = 0x%.8X\n", ret);
 877                         gh->gh_error = GLR_TRYFAILED;
 878                 }
 879                 spin_unlock(&gl->gl_spin);
 880
 881                 if (ret & LM_OUT_CANCELED)
 882                         handle_callback(gl, LM_ST_UNLOCKED);
 883
 884         } else if (ret & LM_OUT_CANCELED) {
 885                 spin_lock(&gl->gl_spin);
 886                 list_del_init(&gh->gh_list);
 887                 gh->gh_error = GLR_CANCELED;
 888                 spin_unlock(&gl->gl_spin);
 889
 890         } else if (relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
 891                 spin_lock(&gl->gl_spin);
 892                 list_move_tail(&gh->gh_list, &gl->gl_holders);
 893                 gh->gh_error = 0;
 894                 set_bit(HIF_HOLDER, &gh->gh_iflags);
 895                 spin_unlock(&gl->gl_spin);
 896
 897                 set_bit(HIF_FIRST, &gh->gh_iflags);
 898
 899                 op_done = 0;
 900
 901         } else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
 902                 spin_lock(&gl->gl_spin);
 903                 list_del_init(&gh->gh_list);
 904                 gh->gh_error = GLR_TRYFAILED;
 905                 spin_unlock(&gl->gl_spin);
 906
 907         } else {
 908                 if (gfs2_assert_withdraw(sdp, 0) == -1)
 909                         fs_err(sdp, "ret = 0x%.8X\n", ret);
 910         }
 911
 912         if (glops->go_xmote_bh)
 913                 glops->go_xmote_bh(gl);
 914
 915         if (op_done) {
 916                 spin_lock(&gl->gl_spin);
 917                 gl->gl_req_gh = NULL;
 918                 gl->gl_req_bh = NULL;
 919                 clear_bit(GLF_LOCK, &gl->gl_flags);
 920                 run_queue(gl);
 921                 spin_unlock(&gl->gl_spin);
 922         }
 923
 924         gfs2_glock_put(gl);
 925
 926         if (gh) {
 927                 if (test_bit(HIF_DEALLOC, &gh->gh_iflags))
 928                         gfs2_holder_put(gh);
 929                 else
 930                         complete(&gh->gh_wait);
 931         }
 932 }
 933
 934 /**
 935  * gfs2_glock_xmote_th - Call into the lock module to acquire or change a glock
 936  * @gl: The glock in question
 937  * @state: the requested state
 938  * @flags: modifier flags to the lock call
 939  *
 940  */
 941
 942 void gfs2_glock_xmote_th(struct gfs2_glock *gl, unsigned int state, int flags)
 943 {
 944         struct gfs2_sbd *sdp = gl->gl_sbd;
 945         const struct gfs2_glock_operations *glops = gl->gl_ops;
 946         int lck_flags = flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB |
 947                                  LM_FLAG_NOEXP | LM_FLAG_ANY |
 948                                  LM_FLAG_PRIORITY);
 949         unsigned int lck_ret;
 950
 951         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
 952         gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
 953         gfs2_assert_warn(sdp, state != LM_ST_UNLOCKED);
 954         gfs2_assert_warn(sdp, state != gl->gl_state);
 955
 956         if (gl->gl_state == LM_ST_EXCLUSIVE && glops->go_sync)
 957                 glops->go_sync(gl);
 958
 959         gfs2_glock_hold(gl);
 960         gl->gl_req_bh = xmote_bh;
 961
 962         lck_ret = gfs2_lm_lock(sdp, gl->gl_lock, gl->gl_state, state, lck_flags);
 963
 964         if (gfs2_assert_withdraw(sdp, !(lck_ret & LM_OUT_ERROR)))
 965                 return;
 966
 967         if (lck_ret & LM_OUT_ASYNC)
 968                 gfs2_assert_warn(sdp, lck_ret == LM_OUT_ASYNC);
 969         else
 970                 xmote_bh(gl, lck_ret);
 971 }
 972
 973 /**
 974  * drop_bh - Called after a lock module unlock completes
 975  * @gl: the glock
 976  * @ret: the return status
 977  *
 978  * Doesn't wake up the process waiting on the struct gfs2_holder (if any)
 979  * Doesn't drop the reference on the glock the top half took out
 980  *
 981  */
 982
 983 static void drop_bh(struct gfs2_glock *gl, unsigned int ret)
 984 {
 985         struct gfs2_sbd *sdp = gl->gl_sbd;
 986         const struct gfs2_glock_operations *glops = gl->gl_ops;
 987         struct gfs2_holder *gh = gl->gl_req_gh;
 988
 989         clear_bit(GLF_PREFETCH, &gl->gl_flags);
 990
 991         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
 992         gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
 993         gfs2_assert_warn(sdp, !ret);
 994
 995         state_change(gl, LM_ST_UNLOCKED);
 996
 997         if (glops->go_inval)
 998                 glops->go_inval(gl, DIO_METADATA);
 999
1000         if (gh) {
1001                 spin_lock(&gl->gl_spin);
1002                 list_del_init(&gh->gh_list);
1003                 gh->gh_error = 0;
1004                 spin_unlock(&gl->gl_spin);
1005         }
1006
1007         if (glops->go_drop_bh)
1008                 glops->go_drop_bh(gl);
1009
1010         spin_lock(&gl->gl_spin);
1011         gl->gl_req_gh = NULL;
1012         gl->gl_req_bh = NULL;
1013         clear_bit(GLF_LOCK, &gl->gl_flags);
1014         run_queue(gl);
1015         spin_unlock(&gl->gl_spin);
1016
1017         gfs2_glock_put(gl);
1018
1019         if (gh) {
1020                 if (test_bit(HIF_DEALLOC, &gh->gh_iflags))
1021                         gfs2_holder_put(gh);
1022                 else
1023                         complete(&gh->gh_wait);
1024         }
1025 }
1026
1027 /**
1028  * gfs2_glock_drop_th - call into the lock module to unlock a lock
1029  * @gl: the glock
1030  *
1031  */
1032
1033 void gfs2_glock_drop_th(struct gfs2_glock *gl)
1034 {
1035         struct gfs2_sbd *sdp = gl->gl_sbd;
1036         const struct gfs2_glock_operations *glops = gl->gl_ops;
1037         unsigned int ret;
1038
1039         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1040         gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
1041         gfs2_assert_warn(sdp, gl->gl_state != LM_ST_UNLOCKED);
1042
1043         if (gl->gl_state == LM_ST_EXCLUSIVE && glops->go_sync)
1044                 glops->go_sync(gl);
1045
1046         gfs2_glock_hold(gl);
1047         gl->gl_req_bh = drop_bh;
1048
1049         ret = gfs2_lm_unlock(sdp, gl->gl_lock, gl->gl_state);
1050
1051         if (gfs2_assert_withdraw(sdp, !(ret & LM_OUT_ERROR)))
1052                 return;
1053
1054         if (!ret)
1055                 drop_bh(gl, ret);
1056         else
1057                 gfs2_assert_warn(sdp, ret == LM_OUT_ASYNC);
1058 }
1059
1060 /**
1061  * do_cancels - cancel requests for locks stuck waiting on an expire flag
1062  * @gh: the LM_FLAG_PRIORITY holder waiting to acquire the lock
1063  *
1064  * Don't cancel GL_NOCANCEL requests.
1065  */
1066
1067 static void do_cancels(struct gfs2_holder *gh)
1068 {
1069         struct gfs2_glock *gl = gh->gh_gl;
1070
1071         spin_lock(&gl->gl_spin);
1072
1073         while (gl->gl_req_gh != gh &&
1074                !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1075                !list_empty(&gh->gh_list)) {
1076                 if (gl->gl_req_bh && !(gl->gl_req_gh &&
1077                                      (gl->gl_req_gh->gh_flags & GL_NOCANCEL))) {
1078                         spin_unlock(&gl->gl_spin);
1079                         gfs2_lm_cancel(gl->gl_sbd, gl->gl_lock);
1080                         msleep(100);
1081                         spin_lock(&gl->gl_spin);
1082                 } else {
1083                         spin_unlock(&gl->gl_spin);
1084                         msleep(100);
1085                         spin_lock(&gl->gl_spin);
1086                 }
1087         }
1088
1089         spin_unlock(&gl->gl_spin);
1090 }
1091
1092 /**
1093  * glock_wait_internal - wait on a glock acquisition
1094  * @gh: the glock holder
1095  *
1096  * Returns: 0 on success
1097  */
1098
1099 static int glock_wait_internal(struct gfs2_holder *gh)
1100 {
1101         struct gfs2_glock *gl = gh->gh_gl;
1102         struct gfs2_sbd *sdp = gl->gl_sbd;
1103         const struct gfs2_glock_operations *glops = gl->gl_ops;
1104
1105         if (test_bit(HIF_ABORTED, &gh->gh_iflags))
1106                 return -EIO;
1107
1108         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1109                 spin_lock(&gl->gl_spin);
1110                 if (gl->gl_req_gh != gh &&
1111                     !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1112                     !list_empty(&gh->gh_list)) {
1113                         list_del_init(&gh->gh_list);
1114                         gh->gh_error = GLR_TRYFAILED;
1115                         run_queue(gl);
1116                         spin_unlock(&gl->gl_spin);
1117                         return gh->gh_error;
1118                 }
1119                 spin_unlock(&gl->gl_spin);
1120         }
1121
1122         if (gh->gh_flags & LM_FLAG_PRIORITY)
1123                 do_cancels(gh);
1124
1125         wait_for_completion(&gh->gh_wait);
1126
1127         if (gh->gh_error)
1128                 return gh->gh_error;
1129
1130         gfs2_assert_withdraw(sdp, test_bit(HIF_HOLDER, &gh->gh_iflags));
1131         gfs2_assert_withdraw(sdp, relaxed_state_ok(gl->gl_state, gh->gh_state,
1132                                                    gh->gh_flags));
1133
1134         if (test_bit(HIF_FIRST, &gh->gh_iflags)) {
1135                 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1136
1137                 if (glops->go_lock) {
1138                         gh->gh_error = glops->go_lock(gh);
1139                         if (gh->gh_error) {
1140                                 spin_lock(&gl->gl_spin);
1141                                 list_del_init(&gh->gh_list);
1142                                 spin_unlock(&gl->gl_spin);
1143                         }
1144                 }
1145
1146                 spin_lock(&gl->gl_spin);
1147                 gl->gl_req_gh = NULL;
1148                 gl->gl_req_bh = NULL;
1149                 clear_bit(GLF_LOCK, &gl->gl_flags);
1150                 run_queue(gl);
1151                 spin_unlock(&gl->gl_spin);
1152         }
1153
1154         return gh->gh_error;
1155 }
1156
1157 static inline struct gfs2_holder *
1158 find_holder_by_owner(struct list_head *head, struct task_struct *owner)
1159 {
1160         struct gfs2_holder *gh;
1161
1162         list_for_each_entry(gh, head, gh_list) {
1163                 if (gh->gh_owner == owner)
1164                         return gh;
1165         }
1166
1167         return NULL;
1168 }
1169
1170 /**
1171  * add_to_queue - Add a holder to the wait queue (but look for recursion)
1172  * @gh: the holder structure to add
1173  *
1174  */
1175
1176 static void add_to_queue(struct gfs2_holder *gh)
1177 {
1178         struct gfs2_glock *gl = gh->gh_gl;
1179         struct gfs2_holder *existing;
1180
1181         BUG_ON(!gh->gh_owner);
1182
1183         existing = find_holder_by_owner(&gl->gl_holders, gh->gh_owner);
1184         if (existing) {
1185                 print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
1186                 printk(KERN_INFO "pid : %d\n", existing->gh_owner->pid);
1187                 printk(KERN_INFO "lock type : %d lock state : %d\n",
1188                                 existing->gh_gl->gl_name.ln_type, existing->gh_gl->gl_state);
1189                 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1190                 printk(KERN_INFO "pid : %d\n", gh->gh_owner->pid);
1191                 printk(KERN_INFO "lock type : %d lock state : %d\n",
1192                                 gl->gl_name.ln_type, gl->gl_state);
1193                 BUG();
1194         }
1195
1196         existing = find_holder_by_owner(&gl->gl_waiters3, gh->gh_owner);
1197         if (existing) {
1198                 print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
1199                 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1200                 BUG();
1201         }
1202
1203         if (gh->gh_flags & LM_FLAG_PRIORITY)
1204                 list_add(&gh->gh_list, &gl->gl_waiters3);
1205         else
1206                 list_add_tail(&gh->gh_list, &gl->gl_waiters3);
1207 }
1208
1209 /**
1210  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1211  * @gh: the holder structure
1212  *
1213  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1214  *
1215  * Returns: 0, GLR_TRYFAILED, or errno on failure
1216  */
1217
1218 int gfs2_glock_nq(struct gfs2_holder *gh)
1219 {
1220         struct gfs2_glock *gl = gh->gh_gl;
1221         struct gfs2_sbd *sdp = gl->gl_sbd;
1222         int error = 0;
1223
1224 restart:
1225         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1226                 set_bit(HIF_ABORTED, &gh->gh_iflags);
1227                 return -EIO;
1228         }
1229
1230         set_bit(HIF_PROMOTE, &gh->gh_iflags);
1231
1232         spin_lock(&gl->gl_spin);
1233         add_to_queue(gh);
1234         run_queue(gl);
1235         spin_unlock(&gl->gl_spin);
1236
1237         if (!(gh->gh_flags & GL_ASYNC)) {
1238                 error = glock_wait_internal(gh);
1239                 if (error == GLR_CANCELED) {
1240                         msleep(100);
1241                         goto restart;
1242                 }
1243         }
1244
1245         clear_bit(GLF_PREFETCH, &gl->gl_flags);
1246
1247         return error;
1248 }
1249
1250 /**
1251  * gfs2_glock_poll - poll to see if an async request has been completed
1252  * @gh: the holder
1253  *
1254  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1255  */
1256
1257 int gfs2_glock_poll(struct gfs2_holder *gh)
1258 {
1259         struct gfs2_glock *gl = gh->gh_gl;
1260         int ready = 0;
1261
1262         spin_lock(&gl->gl_spin);
1263
1264         if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1265                 ready = 1;
1266         else if (list_empty(&gh->gh_list)) {
1267                 if (gh->gh_error == GLR_CANCELED) {
1268                         spin_unlock(&gl->gl_spin);
1269                         msleep(100);
1270                         if (gfs2_glock_nq(gh))
1271                                 return 1;
1272                         return 0;
1273                 } else
1274                         ready = 1;
1275         }
1276
1277         spin_unlock(&gl->gl_spin);
1278
1279         return ready;
1280 }
1281
1282 /**
1283  * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
1284  * @gh: the holder structure
1285  *
1286  * Returns: 0, GLR_TRYFAILED, or errno on failure
1287  */
1288
1289 int gfs2_glock_wait(struct gfs2_holder *gh)
1290 {
1291         int error;
1292
1293         error = glock_wait_internal(gh);
1294         if (error == GLR_CANCELED) {
1295                 msleep(100);
1296                 gh->gh_flags &= ~GL_ASYNC;
1297                 error = gfs2_glock_nq(gh);
1298         }
1299
1300         return error;
1301 }
1302
1303 /**
1304  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1305  * @gh: the glock holder
1306  *
1307  */
1308
1309 void gfs2_glock_dq(struct gfs2_holder *gh)
1310 {
1311         struct gfs2_glock *gl = gh->gh_gl;
1312         const struct gfs2_glock_operations *glops = gl->gl_ops;
1313
1314         if (gh->gh_flags & GL_NOCACHE)
1315                 handle_callback(gl, LM_ST_UNLOCKED);
1316
1317         gfs2_glmutex_lock(gl);
1318
1319         spin_lock(&gl->gl_spin);
1320         list_del_init(&gh->gh_list);
1321
1322         if (list_empty(&gl->gl_holders)) {
1323                 spin_unlock(&gl->gl_spin);
1324
1325                 if (glops->go_unlock)
1326                         glops->go_unlock(gh);
1327
1328                 gl->gl_stamp = jiffies;
1329
1330                 spin_lock(&gl->gl_spin);
1331         }
1332
1333         clear_bit(GLF_LOCK, &gl->gl_flags);
1334         run_queue(gl);
1335         spin_unlock(&gl->gl_spin);
1336 }
1337
1338 /**
1339  * gfs2_glock_prefetch - Try to prefetch a glock
1340  * @gl: the glock
1341  * @state: the state to prefetch in
1342  * @flags: flags passed to go_xmote_th()
1343  *
1344  */
1345
1346 static void gfs2_glock_prefetch(struct gfs2_glock *gl, unsigned int state,
1347                                 int flags)
1348 {
1349         const struct gfs2_glock_operations *glops = gl->gl_ops;
1350
1351         spin_lock(&gl->gl_spin);
1352
1353         if (test_bit(GLF_LOCK, &gl->gl_flags) || !list_empty(&gl->gl_holders) ||
1354             !list_empty(&gl->gl_waiters1) || !list_empty(&gl->gl_waiters2) ||
1355             !list_empty(&gl->gl_waiters3) ||
1356             relaxed_state_ok(gl->gl_state, state, flags)) {
1357                 spin_unlock(&gl->gl_spin);
1358                 return;
1359         }
1360
1361         set_bit(GLF_PREFETCH, &gl->gl_flags);
1362         set_bit(GLF_LOCK, &gl->gl_flags);
1363         spin_unlock(&gl->gl_spin);
1364
1365         glops->go_xmote_th(gl, state, flags);
1366 }
1367
1368 static void greedy_work(void *data)
1369 {
1370         struct greedy *gr = data;
1371         struct gfs2_holder *gh = &gr->gr_gh;
1372         struct gfs2_glock *gl = gh->gh_gl;
1373         const struct gfs2_glock_operations *glops = gl->gl_ops;
1374
1375         clear_bit(GLF_SKIP_WAITERS2, &gl->gl_flags);
1376
1377         if (glops->go_greedy)
1378                 glops->go_greedy(gl);
1379
1380         spin_lock(&gl->gl_spin);
1381
1382         if (list_empty(&gl->gl_waiters2)) {
1383                 clear_bit(GLF_GREEDY, &gl->gl_flags);
1384                 spin_unlock(&gl->gl_spin);
1385                 gfs2_holder_uninit(gh);
1386                 kfree(gr);
1387         } else {
1388                 gfs2_glock_hold(gl);
1389                 list_add_tail(&gh->gh_list, &gl->gl_waiters2);
1390                 run_queue(gl);
1391                 spin_unlock(&gl->gl_spin);
1392                 gfs2_glock_put(gl);
1393         }
1394 }
1395
1396 /**
1397  * gfs2_glock_be_greedy -
1398  * @gl:
1399  * @time:
1400  *
1401  * Returns: 0 if go_greedy will be called, 1 otherwise
1402  */
1403
1404 int gfs2_glock_be_greedy(struct gfs2_glock *gl, unsigned int time)
1405 {
1406         struct greedy *gr;
1407         struct gfs2_holder *gh;
1408
1409         if (!time || gl->gl_sbd->sd_args.ar_localcaching ||
1410             test_and_set_bit(GLF_GREEDY, &gl->gl_flags))
1411                 return 1;
1412
1413         gr = kmalloc(sizeof(struct greedy), GFP_KERNEL);
1414         if (!gr) {
1415                 clear_bit(GLF_GREEDY, &gl->gl_flags);
1416                 return 1;
1417         }
1418         gh = &gr->gr_gh;
1419
1420         gfs2_holder_init(gl, 0, 0, gh);
1421         set_bit(HIF_GREEDY, &gh->gh_iflags);
1422         INIT_WORK(&gr->gr_work, greedy_work, gr);
1423
1424         set_bit(GLF_SKIP_WAITERS2, &gl->gl_flags);
1425         schedule_delayed_work(&gr->gr_work, time);
1426
1427         return 0;
1428 }
1429
1430 /**
1431  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1432  * @gh: the holder structure
1433  *
1434  */
1435
1436 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1437 {
1438         gfs2_glock_dq(gh);
1439         gfs2_holder_uninit(gh);
1440 }
1441
1442 /**
1443  * gfs2_glock_nq_num - acquire a glock based on lock number
1444  * @sdp: the filesystem
1445  * @number: the lock number
1446  * @glops: the glock operations for the type of glock
1447  * @state: the state to acquire the glock in
1448  * @flags: modifier flags for the aquisition
1449  * @gh: the struct gfs2_holder
1450  *
1451  * Returns: errno
1452  */
1453
1454 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1455                       const struct gfs2_glock_operations *glops,
1456                       unsigned int state, int flags, struct gfs2_holder *gh)
1457 {
1458         struct gfs2_glock *gl;
1459         int error;
1460
1461         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1462         if (!error) {
1463                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1464                 gfs2_glock_put(gl);
1465         }
1466
1467         return error;
1468 }
1469
1470 /**
1471  * glock_compare - Compare two struct gfs2_glock structures for sorting
1472  * @arg_a: the first structure
1473  * @arg_b: the second structure
1474  *
1475  */
1476
1477 static int glock_compare(const void *arg_a, const void *arg_b)
1478 {
1479         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1480         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1481         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1482         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1483
1484         if (a->ln_number > b->ln_number)
1485                 return 1;
1486         if (a->ln_number < b->ln_number)
1487                 return -1;
1488         if (gh_a->gh_state == LM_ST_SHARED && gh_b->gh_state == LM_ST_EXCLUSIVE)
1489                 return 1;
1490         if (!(gh_a->gh_flags & GL_LOCAL_EXCL) && (gh_b->gh_flags & GL_LOCAL_EXCL))
1491                 return 1;
1492         return 0;
1493 }
1494
1495 /**
1496  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1497  * @num_gh: the number of structures
1498  * @ghs: an array of struct gfs2_holder structures
1499  *
1500  * Returns: 0 on success (all glocks acquired),
1501  *          errno on failure (no glocks acquired)
1502  */
1503
1504 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1505                      struct gfs2_holder **p)
1506 {
1507         unsigned int x;
1508         int error = 0;
1509
1510         for (x = 0; x < num_gh; x++)
1511                 p[x] = &ghs[x];
1512
1513         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1514
1515         for (x = 0; x < num_gh; x++) {
1516                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1517
1518                 error = gfs2_glock_nq(p[x]);
1519                 if (error) {
1520                         while (x--)
1521                                 gfs2_glock_dq(p[x]);
1522                         break;
1523                 }
1524         }
1525
1526         return error;
1527 }
1528
1529 /**
1530  * gfs2_glock_nq_m - acquire multiple glocks
1531  * @num_gh: the number of structures
1532  * @ghs: an array of struct gfs2_holder structures
1533  *
1534  * Figure out how big an impact this function has.  Either:
1535  * 1) Replace this code with code that calls gfs2_glock_prefetch()
1536  * 2) Forget async stuff and just call nq_m_sync()
1537  * 3) Leave it like it is
1538  *
1539  * Returns: 0 on success (all glocks acquired),
1540  *          errno on failure (no glocks acquired)
1541  */
1542
1543 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1544 {
1545         int *e;
1546         unsigned int x;
1547         int borked = 0, serious = 0;
1548         int error = 0;
1549
1550         if (!num_gh)
1551                 return 0;
1552
1553         if (num_gh == 1) {
1554                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1555                 return gfs2_glock_nq(ghs);
1556         }
1557
1558         e = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL);
1559         if (!e)
1560                 return -ENOMEM;
1561
1562         for (x = 0; x < num_gh; x++) {
1563                 ghs[x].gh_flags |= LM_FLAG_TRY | GL_ASYNC;
1564                 error = gfs2_glock_nq(&ghs[x]);
1565                 if (error) {
1566                         borked = 1;
1567                         serious = error;
1568                         num_gh = x;
1569                         break;
1570                 }
1571         }
1572
1573         for (x = 0; x < num_gh; x++) {
1574                 error = e[x] = glock_wait_internal(&ghs[x]);
1575                 if (error) {
1576                         borked = 1;
1577                         if (error != GLR_TRYFAILED && error != GLR_CANCELED)
1578                                 serious = error;
1579                 }
1580         }
1581
1582         if (!borked) {
1583                 kfree(e);
1584                 return 0;
1585         }
1586
1587         for (x = 0; x < num_gh; x++)
1588                 if (!e[x])
1589                         gfs2_glock_dq(&ghs[x]);
1590
1591         if (serious)
1592                 error = serious;
1593         else {
1594                 for (x = 0; x < num_gh; x++)
1595                         gfs2_holder_reinit(ghs[x].gh_state, ghs[x].gh_flags,
1596                                           &ghs[x]);
1597                 error = nq_m_sync(num_gh, ghs, (struct gfs2_holder **)e);
1598         }
1599
1600         kfree(e);
1601
1602         return error;
1603 }
1604
1605 /**
1606  * gfs2_glock_dq_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_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(&ghs[x]);
1618 }
1619
1620 /**
1621  * gfs2_glock_dq_uninit_m - release multiple glocks
1622  * @num_gh: the number of structures
1623  * @ghs: an array of struct gfs2_holder structures
1624  *
1625  */
1626
1627 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1628 {
1629         unsigned int x;
1630
1631         for (x = 0; x < num_gh; x++)
1632                 gfs2_glock_dq_uninit(&ghs[x]);
1633 }
1634
1635 /**
1636  * gfs2_glock_prefetch_num - prefetch a glock based on lock number
1637  * @sdp: the filesystem
1638  * @number: the lock number
1639  * @glops: the glock operations for the type of glock
1640  * @state: the state to acquire the glock in
1641  * @flags: modifier flags for the aquisition
1642  *
1643  * Returns: errno
1644  */
1645
1646 void gfs2_glock_prefetch_num(struct gfs2_sbd *sdp, u64 number,
1647                              const struct gfs2_glock_operations *glops,
1648                              unsigned int state, int flags)
1649 {
1650         struct gfs2_glock *gl;
1651         int error;
1652
1653         if (atomic_read(&sdp->sd_reclaim_count) <
1654             gfs2_tune_get(sdp, gt_reclaim_limit)) {
1655                 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1656                 if (!error) {
1657                         gfs2_glock_prefetch(gl, state, flags);
1658                         gfs2_glock_put(gl);
1659                 }
1660         }
1661 }
1662
1663 /**
1664  * gfs2_lvb_hold - attach a LVB from a glock
1665  * @gl: The glock in question
1666  *
1667  */
1668
1669 int gfs2_lvb_hold(struct gfs2_glock *gl)
1670 {
1671         int error;
1672
1673         gfs2_glmutex_lock(gl);
1674
1675         if (!atomic_read(&gl->gl_lvb_count)) {
1676                 error = gfs2_lm_hold_lvb(gl->gl_sbd, gl->gl_lock, &gl->gl_lvb);
1677                 if (error) {
1678                         gfs2_glmutex_unlock(gl);
1679                         return error;
1680                 }
1681                 gfs2_glock_hold(gl);
1682         }
1683         atomic_inc(&gl->gl_lvb_count);
1684
1685         gfs2_glmutex_unlock(gl);
1686
1687         return 0;
1688 }
1689
1690 /**
1691  * gfs2_lvb_unhold - detach a LVB from a glock
1692  * @gl: The glock in question
1693  *
1694  */
1695
1696 void gfs2_lvb_unhold(struct gfs2_glock *gl)
1697 {
1698         gfs2_glock_hold(gl);
1699         gfs2_glmutex_lock(gl);
1700
1701         gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count) > 0);
1702         if (atomic_dec_and_test(&gl->gl_lvb_count)) {
1703                 gfs2_lm_unhold_lvb(gl->gl_sbd, gl->gl_lock, gl->gl_lvb);
1704                 gl->gl_lvb = NULL;
1705                 gfs2_glock_put(gl);
1706         }
1707
1708         gfs2_glmutex_unlock(gl);
1709         gfs2_glock_put(gl);
1710 }
1711
1712 static void blocking_cb(struct gfs2_sbd *sdp, struct lm_lockname *name,
1713                         unsigned int state)
1714 {
1715         struct gfs2_glock *gl;
1716
1717         gl = gfs2_glock_find(sdp, name);
1718         if (!gl)
1719                 return;
1720
1721         if (gl->gl_ops->go_callback)
1722                 gl->gl_ops->go_callback(gl, state);
1723         handle_callback(gl, state);
1724
1725         spin_lock(&gl->gl_spin);
1726         run_queue(gl);
1727         spin_unlock(&gl->gl_spin);
1728
1729         gfs2_glock_put(gl);
1730 }
1731
1732 /**
1733  * gfs2_glock_cb - Callback used by locking module
1734  * @sdp: Pointer to the superblock
1735  * @type: Type of callback
1736  * @data: Type dependent data pointer
1737  *
1738  * Called by the locking module when it wants to tell us something.
1739  * Either we need to drop a lock, one of our ASYNC requests completed, or
1740  * a journal from another client needs to be recovered.
1741  */
1742
1743 void gfs2_glock_cb(void *cb_data, unsigned int type, void *data)
1744 {
1745         struct gfs2_sbd *sdp = cb_data;
1746
1747         switch (type) {
1748         case LM_CB_NEED_E:
1749                 blocking_cb(sdp, data, LM_ST_UNLOCKED);
1750                 return;
1751
1752         case LM_CB_NEED_D:
1753                 blocking_cb(sdp, data, LM_ST_DEFERRED);
1754                 return;
1755
1756         case LM_CB_NEED_S:
1757                 blocking_cb(sdp, data, LM_ST_SHARED);
1758                 return;
1759
1760         case LM_CB_ASYNC: {
1761                 struct lm_async_cb *async = data;
1762                 struct gfs2_glock *gl;
1763
1764                 gl = gfs2_glock_find(sdp, &async->lc_name);
1765                 if (gfs2_assert_warn(sdp, gl))
1766                         return;
1767                 if (!gfs2_assert_warn(sdp, gl->gl_req_bh))
1768                         gl->gl_req_bh(gl, async->lc_ret);
1769                 gfs2_glock_put(gl);
1770                 return;
1771         }
1772
1773         case LM_CB_NEED_RECOVERY:
1774                 gfs2_jdesc_make_dirty(sdp, *(unsigned int *)data);
1775                 if (sdp->sd_recoverd_process)
1776                         wake_up_process(sdp->sd_recoverd_process);
1777                 return;
1778
1779         case LM_CB_DROPLOCKS:
1780                 gfs2_gl_hash_clear(sdp, NO_WAIT);
1781                 gfs2_quota_scan(sdp);
1782                 return;
1783
1784         default:
1785                 gfs2_assert_warn(sdp, 0);
1786                 return;
1787         }
1788 }
1789
1790 /**
1791  * demote_ok - Check to see if it's ok to unlock a glock
1792  * @gl: the glock
1793  *
1794  * Returns: 1 if it's ok
1795  */
1796
1797 static int demote_ok(struct gfs2_glock *gl)
1798 {
1799         struct gfs2_sbd *sdp = gl->gl_sbd;
1800         const struct gfs2_glock_operations *glops = gl->gl_ops;
1801         int demote = 1;
1802
1803         if (test_bit(GLF_STICKY, &gl->gl_flags))
1804                 demote = 0;
1805         else if (test_bit(GLF_PREFETCH, &gl->gl_flags))
1806                 demote = time_after_eq(jiffies, gl->gl_stamp +
1807                                     gfs2_tune_get(sdp, gt_prefetch_secs) * HZ);
1808         else if (glops->go_demote_ok)
1809                 demote = glops->go_demote_ok(gl);
1810
1811         return demote;
1812 }
1813
1814 /**
1815  * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1816  * @gl: the glock
1817  *
1818  */
1819
1820 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
1821 {
1822         struct gfs2_sbd *sdp = gl->gl_sbd;
1823
1824         spin_lock(&sdp->sd_reclaim_lock);
1825         if (list_empty(&gl->gl_reclaim)) {
1826                 gfs2_glock_hold(gl);
1827                 list_add(&gl->gl_reclaim, &sdp->sd_reclaim_list);
1828                 atomic_inc(&sdp->sd_reclaim_count);
1829         }
1830         spin_unlock(&sdp->sd_reclaim_lock);
1831
1832         wake_up(&sdp->sd_reclaim_wq);
1833 }
1834
1835 /**
1836  * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1837  * @sdp: the filesystem
1838  *
1839  * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1840  * different glock and we notice that there are a lot of glocks in the
1841  * reclaim list.
1842  *
1843  */
1844
1845 void gfs2_reclaim_glock(struct gfs2_sbd *sdp)
1846 {
1847         struct gfs2_glock *gl;
1848
1849         spin_lock(&sdp->sd_reclaim_lock);
1850         if (list_empty(&sdp->sd_reclaim_list)) {
1851                 spin_unlock(&sdp->sd_reclaim_lock);
1852                 return;
1853         }
1854         gl = list_entry(sdp->sd_reclaim_list.next,
1855                         struct gfs2_glock, gl_reclaim);
1856         list_del_init(&gl->gl_reclaim);
1857         spin_unlock(&sdp->sd_reclaim_lock);
1858
1859         atomic_dec(&sdp->sd_reclaim_count);
1860         atomic_inc(&sdp->sd_reclaimed);
1861
1862         if (gfs2_glmutex_trylock(gl)) {
1863                 if (queue_empty(gl, &gl->gl_holders) &&
1864                     gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1865                         handle_callback(gl, LM_ST_UNLOCKED);
1866                 gfs2_glmutex_unlock(gl);
1867         }
1868
1869         gfs2_glock_put(gl);
1870 }
1871
1872 /**
1873  * examine_bucket - Call a function for glock in a hash bucket
1874  * @examiner: the function
1875  * @sdp: the filesystem
1876  * @bucket: the bucket
1877  *
1878  * Returns: 1 if the bucket has entries
1879  */
1880
1881 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1882                           unsigned int hash)
1883 {
1884         struct gfs2_glock *gl, *prev = NULL;
1885         int has_entries = 0;
1886         struct hlist_head *head = &gl_hash_table[hash].hb_list;
1887
1888         read_lock(gl_lock_addr(hash));
1889         /* Can't use hlist_for_each_entry - don't want prefetch here */
1890         if (hlist_empty(head))
1891                 goto out;
1892         gl = list_entry(head->first, struct gfs2_glock, gl_list);
1893         while(1) {
1894                 if (gl->gl_sbd == sdp) {
1895                         gfs2_glock_hold(gl);
1896                         read_unlock(gl_lock_addr(hash));
1897                         if (prev)
1898                                 gfs2_glock_put(prev);
1899                         prev = gl;
1900                         examiner(gl);
1901                         has_entries = 1;
1902                         read_lock(gl_lock_addr(hash));
1903                 }
1904                 if (gl->gl_list.next == NULL)
1905                         break;
1906                 gl = list_entry(gl->gl_list.next, struct gfs2_glock, gl_list);
1907         }
1908 out:
1909         read_unlock(gl_lock_addr(hash));
1910         if (prev)
1911                 gfs2_glock_put(prev);
1912         return has_entries;
1913 }
1914
1915 /**
1916  * scan_glock - look at a glock and see if we can reclaim it
1917  * @gl: the glock to look at
1918  *
1919  */
1920
1921 static void scan_glock(struct gfs2_glock *gl)
1922 {
1923         if (gl->gl_ops == &gfs2_inode_glops)
1924                 return;
1925
1926         if (gfs2_glmutex_trylock(gl)) {
1927                 if (queue_empty(gl, &gl->gl_holders) &&
1928                     gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1929                         goto out_schedule;
1930                 gfs2_glmutex_unlock(gl);
1931         }
1932         return;
1933
1934 out_schedule:
1935         gfs2_glmutex_unlock(gl);
1936         gfs2_glock_schedule_for_reclaim(gl);
1937 }
1938
1939 /**
1940  * gfs2_scand_internal - Look for glocks and inodes to toss from memory
1941  * @sdp: the filesystem
1942  *
1943  */
1944
1945 void gfs2_scand_internal(struct gfs2_sbd *sdp)
1946 {
1947         unsigned int x;
1948
1949         for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1950                 examine_bucket(scan_glock, sdp, x);
1951 }
1952
1953 /**
1954  * clear_glock - look at a glock and see if we can free it from glock cache
1955  * @gl: the glock to look at
1956  *
1957  */
1958
1959 static void clear_glock(struct gfs2_glock *gl)
1960 {
1961         struct gfs2_sbd *sdp = gl->gl_sbd;
1962         int released;
1963
1964         spin_lock(&sdp->sd_reclaim_lock);
1965         if (!list_empty(&gl->gl_reclaim)) {
1966                 list_del_init(&gl->gl_reclaim);
1967                 atomic_dec(&sdp->sd_reclaim_count);
1968                 spin_unlock(&sdp->sd_reclaim_lock);
1969                 released = gfs2_glock_put(gl);
1970                 gfs2_assert(sdp, !released);
1971         } else {
1972                 spin_unlock(&sdp->sd_reclaim_lock);
1973         }
1974
1975         if (gfs2_glmutex_trylock(gl)) {
1976                 if (queue_empty(gl, &gl->gl_holders) &&
1977                     gl->gl_state != LM_ST_UNLOCKED)
1978                         handle_callback(gl, LM_ST_UNLOCKED);
1979                 gfs2_glmutex_unlock(gl);
1980         }
1981 }
1982
1983 /**
1984  * gfs2_gl_hash_clear - Empty out the glock hash table
1985  * @sdp: the filesystem
1986  * @wait: wait until it's all gone
1987  *
1988  * Called when unmounting the filesystem, or when inter-node lock manager
1989  * requests DROPLOCKS because it is running out of capacity.
1990  */
1991
1992 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp, int wait)
1993 {
1994         unsigned long t;
1995         unsigned int x;
1996         int cont;
1997
1998         t = jiffies;
1999
2000         for (;;) {
2001                 cont = 0;
2002                 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
2003                         if (examine_bucket(clear_glock, sdp, x))
2004                                 cont = 1;
2005                 }
2006
2007                 if (!wait || !cont)
2008                         break;
2009
2010                 if (time_after_eq(jiffies,
2011                                   t + gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
2012                         fs_warn(sdp, "Unmount seems to be stalled. "
2013                                      "Dumping lock state...\n");
2014                         gfs2_dump_lockstate(sdp);
2015                         t = jiffies;
2016                 }
2017
2018                 invalidate_inodes(sdp->sd_vfs);
2019                 msleep(10);
2020         }
2021 }
2022
2023 /*
2024  *  Diagnostic routines to help debug distributed deadlock
2025  */
2026
2027 /**
2028  * dump_holder - print information about a glock holder
2029  * @str: a string naming the type of holder
2030  * @gh: the glock holder
2031  *
2032  * Returns: 0 on success, -ENOBUFS when we run out of space
2033  */
2034
2035 static int dump_holder(char *str, struct gfs2_holder *gh)
2036 {
2037         unsigned int x;
2038         int error = -ENOBUFS;
2039
2040         printk(KERN_INFO "  %s\n", str);
2041         printk(KERN_INFO "    owner = %ld\n",
2042                    (gh->gh_owner) ? (long)gh->gh_owner->pid : -1);
2043         printk(KERN_INFO "    gh_state = %u\n", gh->gh_state);
2044         printk(KERN_INFO "    gh_flags =");
2045         for (x = 0; x < 32; x++)
2046                 if (gh->gh_flags & (1 << x))
2047                         printk(" %u", x);
2048         printk(" \n");
2049         printk(KERN_INFO "    error = %d\n", gh->gh_error);
2050         printk(KERN_INFO "    gh_iflags =");
2051         for (x = 0; x < 32; x++)
2052                 if (test_bit(x, &gh->gh_iflags))
2053                         printk(" %u", x);
2054         printk(" \n");
2055         print_symbol(KERN_INFO "    initialized at: %s\n", gh->gh_ip);
2056
2057         error = 0;
2058
2059         return error;
2060 }
2061
2062 /**
2063  * dump_inode - print information about an inode
2064  * @ip: the inode
2065  *
2066  * Returns: 0 on success, -ENOBUFS when we run out of space
2067  */
2068
2069 static int dump_inode(struct gfs2_inode *ip)
2070 {
2071         unsigned int x;
2072         int error = -ENOBUFS;
2073
2074         printk(KERN_INFO "  Inode:\n");
2075         printk(KERN_INFO "    num = %llu %llu\n",
2076                     (unsigned long long)ip->i_num.no_formal_ino,
2077                     (unsigned long long)ip->i_num.no_addr);
2078         printk(KERN_INFO "    type = %u\n", IF2DT(ip->i_inode.i_mode));
2079         printk(KERN_INFO "    i_flags =");
2080         for (x = 0; x < 32; x++)
2081                 if (test_bit(x, &ip->i_flags))
2082                         printk(" %u", x);
2083         printk(" \n");
2084
2085         error = 0;
2086
2087         return error;
2088 }
2089
2090 /**
2091  * dump_glock - print information about a glock
2092  * @gl: the glock
2093  * @count: where we are in the buffer
2094  *
2095  * Returns: 0 on success, -ENOBUFS when we run out of space
2096  */
2097
2098 static int dump_glock(struct gfs2_glock *gl)
2099 {
2100         struct gfs2_holder *gh;
2101         unsigned int x;
2102         int error = -ENOBUFS;
2103
2104         spin_lock(&gl->gl_spin);
2105
2106         printk(KERN_INFO "Glock 0x%p (%u, %llu)\n", gl, gl->gl_name.ln_type,
2107                (unsigned long long)gl->gl_name.ln_number);
2108         printk(KERN_INFO "  gl_flags =");
2109         for (x = 0; x < 32; x++) {
2110                 if (test_bit(x, &gl->gl_flags))
2111                         printk(" %u", x);
2112         }
2113         printk(" \n");
2114         printk(KERN_INFO "  gl_ref = %d\n", atomic_read(&gl->gl_ref));
2115         printk(KERN_INFO "  gl_state = %u\n", gl->gl_state);
2116         printk(KERN_INFO "  gl_owner = %s\n", gl->gl_owner->comm);
2117         print_symbol(KERN_INFO "  gl_ip = %s\n", gl->gl_ip);
2118         printk(KERN_INFO "  req_gh = %s\n", (gl->gl_req_gh) ? "yes" : "no");
2119         printk(KERN_INFO "  req_bh = %s\n", (gl->gl_req_bh) ? "yes" : "no");
2120         printk(KERN_INFO "  lvb_count = %d\n", atomic_read(&gl->gl_lvb_count));
2121         printk(KERN_INFO "  object = %s\n", (gl->gl_object) ? "yes" : "no");
2122         printk(KERN_INFO "  le = %s\n",
2123                    (list_empty(&gl->gl_le.le_list)) ? "no" : "yes");
2124         printk(KERN_INFO "  reclaim = %s\n",
2125                     (list_empty(&gl->gl_reclaim)) ? "no" : "yes");
2126         if (gl->gl_aspace)
2127                 printk(KERN_INFO "  aspace = 0x%p nrpages = %lu\n", gl->gl_aspace,
2128                        gl->gl_aspace->i_mapping->nrpages);
2129         else
2130                 printk(KERN_INFO "  aspace = no\n");
2131         printk(KERN_INFO "  ail = %d\n", atomic_read(&gl->gl_ail_count));
2132         if (gl->gl_req_gh) {
2133                 error = dump_holder("Request", gl->gl_req_gh);
2134                 if (error)
2135                         goto out;
2136         }
2137         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
2138                 error = dump_holder("Holder", gh);
2139                 if (error)
2140                         goto out;
2141         }
2142         list_for_each_entry(gh, &gl->gl_waiters1, gh_list) {
2143                 error = dump_holder("Waiter1", gh);
2144                 if (error)
2145                         goto out;
2146         }
2147         list_for_each_entry(gh, &gl->gl_waiters2, gh_list) {
2148                 error = dump_holder("Waiter2", gh);
2149                 if (error)
2150                         goto out;
2151         }
2152         list_for_each_entry(gh, &gl->gl_waiters3, gh_list) {
2153                 error = dump_holder("Waiter3", gh);
2154                 if (error)
2155                         goto out;
2156         }
2157         if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object) {
2158                 if (!test_bit(GLF_LOCK, &gl->gl_flags) &&
2159                     list_empty(&gl->gl_holders)) {
2160                         error = dump_inode(gl->gl_object);
2161                         if (error)
2162                                 goto out;
2163                 } else {
2164                         error = -ENOBUFS;
2165                         printk(KERN_INFO "  Inode: busy\n");
2166                 }
2167         }
2168
2169         error = 0;
2170
2171 out:
2172         spin_unlock(&gl->gl_spin);
2173         return error;
2174 }
2175
2176 /**
2177  * gfs2_dump_lockstate - print out the current lockstate
2178  * @sdp: the filesystem
2179  * @ub: the buffer to copy the information into
2180  *
2181  * If @ub is NULL, dump the lockstate to the console.
2182  *
2183  */
2184
2185 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
2186 {
2187         struct gfs2_glock *gl;
2188         struct hlist_node *h;
2189         unsigned int x;
2190         int error = 0;
2191
2192         for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
2193
2194                 read_lock(gl_lock_addr(x));
2195
2196                 hlist_for_each_entry(gl, h, &gl_hash_table[x].hb_list, gl_list) {
2197                         if (gl->gl_sbd != sdp)
2198                                 continue;
2199
2200                         error = dump_glock(gl);
2201                         if (error)
2202                                 break;
2203                 }
2204
2205                 read_unlock(gl_lock_addr(x));
2206
2207                 if (error)
2208                         break;
2209         }
2210
2211
2212         return error;
2213 }
2214
2215 int __init gfs2_glock_init(void)
2216 {
2217         unsigned i;
2218         for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
2219                 INIT_HLIST_HEAD(&gl_hash_table[i].hb_list);
2220         }
2221 #ifdef GL_HASH_LOCK_SZ
2222         for(i = 0; i < GL_HASH_LOCK_SZ; i++) {
2223                 rwlock_init(&gl_hash_locks[i]);
2224         }
2225 #endif
2226         return 0;
2227 }
2228