fs/gfs2/glock.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2005 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 v.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/kref.h>
  19 #include <linux/kallsyms.h>
  20 #include <linux/gfs2_ondisk.h>
  21 #include <asm/semaphore.h>
  22 #include <asm/uaccess.h>
  23
  24 #include "gfs2.h"
  25 #include "lm_interface.h"
  26 #include "incore.h"
  27 #include "glock.h"
  28 #include "glops.h"
  29 #include "inode.h"
  30 #include "lm.h"
  31 #include "lops.h"
  32 #include "meta_io.h"
  33 #include "quota.h"
  34 #include "super.h"
  35 #include "util.h"
  36
  37 /*  Must be kept in sync with the beginning of struct gfs2_glock  */
  38 struct glock_plug {
  39         struct list_head gl_list;
  40         unsigned long gl_flags;
  41 };
  42
  43 struct greedy {
  44         struct gfs2_holder gr_gh;
  45         struct work_struct gr_work;
  46 };
  47
  48 typedef void (*glock_examiner) (struct gfs2_glock * gl);
  49
  50 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp);
  51
  52 /**
  53  * relaxed_state_ok - is a requested lock compatible with the current lock mode?
  54  * @actual: the current state of the lock
  55  * @requested: the lock state that was requested by the caller
  56  * @flags: the modifier flags passed in by the caller
  57  *
  58  * Returns: 1 if the locks are compatible, 0 otherwise
  59  */
  60
  61 static inline int relaxed_state_ok(unsigned int actual, unsigned requested,
  62                                    int flags)
  63 {
  64         if (actual == requested)
  65                 return 1;
  66
  67         if (flags & GL_EXACT)
  68                 return 0;
  69
  70         if (actual == LM_ST_EXCLUSIVE && requested == LM_ST_SHARED)
  71                 return 1;
  72
  73         if (actual != LM_ST_UNLOCKED && (flags & LM_FLAG_ANY))
  74                 return 1;
  75
  76         return 0;
  77 }
  78
  79 /**
  80  * gl_hash() - Turn glock number into hash bucket number
  81  * @lock: The glock number
  82  *
  83  * Returns: The number of the corresponding hash bucket
  84  */
  85
  86 static unsigned int gl_hash(struct lm_lockname *name)
  87 {
  88         unsigned int h;
  89
  90         h = jhash(&name->ln_number, sizeof(uint64_t), 0);
  91         h = jhash(&name->ln_type, sizeof(unsigned int), h);
  92         h &= GFS2_GL_HASH_MASK;
  93
  94         return h;
  95 }
  96
  97 /**
  98  * glock_free() - Perform a few checks and then release struct gfs2_glock
  99  * @gl: The glock to release
 100  *
 101  * Also calls lock module to release its internal structure for this glock.
 102  *
 103  */
 104
 105 static void glock_free(struct gfs2_glock *gl)
 106 {
 107         struct gfs2_sbd *sdp = gl->gl_sbd;
 108         struct inode *aspace = gl->gl_aspace;
 109
 110         gfs2_lm_put_lock(sdp, gl->gl_lock);
 111
 112         if (aspace)
 113                 gfs2_aspace_put(aspace);
 114
 115         kmem_cache_free(gfs2_glock_cachep, gl);
 116 }
 117
 118 /**
 119  * gfs2_glock_hold() - increment reference count on glock
 120  * @gl: The glock to hold
 121  *
 122  */
 123
 124 void gfs2_glock_hold(struct gfs2_glock *gl)
 125 {
 126         kref_get(&gl->gl_ref);
 127 }
 128
 129 /* All work is done after the return from kref_put() so we
 130    can release the write_lock before the free. */
 131
 132 static void kill_glock(struct kref *kref)
 133 {
 134         struct gfs2_glock *gl = container_of(kref, struct gfs2_glock, gl_ref);
 135         struct gfs2_sbd *sdp = gl->gl_sbd;
 136
 137         gfs2_assert(sdp, gl->gl_state == LM_ST_UNLOCKED);
 138         gfs2_assert(sdp, list_empty(&gl->gl_reclaim));
 139         gfs2_assert(sdp, list_empty(&gl->gl_holders));
 140         gfs2_assert(sdp, list_empty(&gl->gl_waiters1));
 141         gfs2_assert(sdp, list_empty(&gl->gl_waiters2));
 142         gfs2_assert(sdp, list_empty(&gl->gl_waiters3));
 143 }
 144
 145 /**
 146  * gfs2_glock_put() - Decrement reference count on glock
 147  * @gl: The glock to put
 148  *
 149  */
 150
 151 int gfs2_glock_put(struct gfs2_glock *gl)
 152 {
 153         struct gfs2_sbd *sdp = gl->gl_sbd;
 154         struct gfs2_gl_hash_bucket *bucket = gl->gl_bucket;
 155         int rv = 0;
 156
 157         mutex_lock(&sdp->sd_invalidate_inodes_mutex);
 158
 159         write_lock(&bucket->hb_lock);
 160         if (kref_put(&gl->gl_ref, kill_glock)) {
 161                 list_del_init(&gl->gl_list);
 162                 write_unlock(&bucket->hb_lock);
 163                 BUG_ON(spin_is_locked(&gl->gl_spin));
 164                 glock_free(gl);
 165                 rv = 1;
 166                 goto out;
 167         }
 168         write_unlock(&bucket->hb_lock);
 169  out:
 170         mutex_unlock(&sdp->sd_invalidate_inodes_mutex);
 171         return rv;
 172 }
 173
 174 /**
 175  * queue_empty - check to see if a glock's queue is empty
 176  * @gl: the glock
 177  * @head: the head of the queue to check
 178  *
 179  * This function protects the list in the event that a process already
 180  * has a holder on the list and is adding a second holder for itself.
 181  * The glmutex lock is what generally prevents processes from working
 182  * on the same glock at once, but the special case of adding a second
 183  * holder for yourself ("recursive" locking) doesn't involve locking
 184  * glmutex, making the spin lock necessary.
 185  *
 186  * Returns: 1 if the queue is empty
 187  */
 188
 189 static inline int queue_empty(struct gfs2_glock *gl, struct list_head *head)
 190 {
 191         int empty;
 192         spin_lock(&gl->gl_spin);
 193         empty = list_empty(head);
 194         spin_unlock(&gl->gl_spin);
 195         return empty;
 196 }
 197
 198 /**
 199  * search_bucket() - Find struct gfs2_glock by lock number
 200  * @bucket: the bucket to search
 201  * @name: The lock name
 202  *
 203  * Returns: NULL, or the struct gfs2_glock with the requested number
 204  */
 205
 206 static struct gfs2_glock *search_bucket(struct gfs2_gl_hash_bucket *bucket,
 207                                         struct lm_lockname *name)
 208 {
 209         struct gfs2_glock *gl;
 210
 211         list_for_each_entry(gl, &bucket->hb_list, gl_list) {
 212                 if (test_bit(GLF_PLUG, &gl->gl_flags))
 213                         continue;
 214                 if (!lm_name_equal(&gl->gl_name, name))
 215                         continue;
 216
 217                 kref_get(&gl->gl_ref);
 218
 219                 return gl;
 220         }
 221
 222         return NULL;
 223 }
 224
 225 /**
 226  * gfs2_glock_find() - Find glock by lock number
 227  * @sdp: The GFS2 superblock
 228  * @name: The lock name
 229  *
 230  * Returns: NULL, or the struct gfs2_glock with the requested number
 231  */
 232
 233 static struct gfs2_glock *gfs2_glock_find(struct gfs2_sbd *sdp,
 234                                           struct lm_lockname *name)
 235 {
 236         struct gfs2_gl_hash_bucket *bucket = &sdp->sd_gl_hash[gl_hash(name)];
 237         struct gfs2_glock *gl;
 238
 239         read_lock(&bucket->hb_lock);
 240         gl = search_bucket(bucket, name);
 241         read_unlock(&bucket->hb_lock);
 242
 243         return gl;
 244 }
 245
 246 /**
 247  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
 248  * @sdp: The GFS2 superblock
 249  * @number: the lock number
 250  * @glops: The glock_operations to use
 251  * @create: If 0, don't create the glock if it doesn't exist
 252  * @glp: the glock is returned here
 253  *
 254  * This does not lock a glock, just finds/creates structures for one.
 255  *
 256  * Returns: errno
 257  */
 258
 259 int gfs2_glock_get(struct gfs2_sbd *sdp, uint64_t number,
 260                    struct gfs2_glock_operations *glops, int create,
 261                    struct gfs2_glock **glp)
 262 {
 263         struct lm_lockname name;
 264         struct gfs2_glock *gl, *tmp;
 265         struct gfs2_gl_hash_bucket *bucket;
 266         int error;
 267
 268         name.ln_number = number;
 269         name.ln_type = glops->go_type;
 270         bucket = &sdp->sd_gl_hash[gl_hash(&name)];
 271
 272         read_lock(&bucket->hb_lock);
 273         gl = search_bucket(bucket, &name);
 274         read_unlock(&bucket->hb_lock);
 275
 276         if (gl || !create) {
 277                 *glp = gl;
 278                 return 0;
 279         }
 280
 281         gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
 282         if (!gl)
 283                 return -ENOMEM;
 284
 285         memset(gl, 0, sizeof(struct gfs2_glock));
 286
 287         INIT_LIST_HEAD(&gl->gl_list);
 288         gl->gl_name = name;
 289         kref_init(&gl->gl_ref);
 290
 291         spin_lock_init(&gl->gl_spin);
 292
 293         gl->gl_state = LM_ST_UNLOCKED;
 294         INIT_LIST_HEAD(&gl->gl_holders);
 295         INIT_LIST_HEAD(&gl->gl_waiters1);
 296         INIT_LIST_HEAD(&gl->gl_waiters2);
 297         INIT_LIST_HEAD(&gl->gl_waiters3);
 298
 299         gl->gl_ops = glops;
 300
 301         gl->gl_bucket = bucket;
 302         INIT_LIST_HEAD(&gl->gl_reclaim);
 303
 304         gl->gl_sbd = sdp;
 305
 306         lops_init_le(&gl->gl_le, &gfs2_glock_lops);
 307         INIT_LIST_HEAD(&gl->gl_ail_list);
 308
 309         /* If this glock protects actual on-disk data or metadata blocks,
 310            create a VFS inode to manage the pages/buffers holding them. */
 311         if (glops == &gfs2_inode_glops ||
 312             glops == &gfs2_rgrp_glops ||
 313             glops == &gfs2_meta_glops) {
 314                 gl->gl_aspace = gfs2_aspace_get(sdp);
 315                 if (!gl->gl_aspace) {
 316                         error = -ENOMEM;
 317                         goto fail;
 318                 }
 319         }
 320
 321         error = gfs2_lm_get_lock(sdp, &name, &gl->gl_lock);
 322         if (error)
 323                 goto fail_aspace;
 324
 325         write_lock(&bucket->hb_lock);
 326         tmp = search_bucket(bucket, &name);
 327         if (tmp) {
 328                 write_unlock(&bucket->hb_lock);
 329                 glock_free(gl);
 330                 gl = tmp;
 331         } else {
 332                 list_add_tail(&gl->gl_list, &bucket->hb_list);
 333                 write_unlock(&bucket->hb_lock);
 334         }
 335
 336         *glp = gl;
 337
 338         return 0;
 339
 340  fail_aspace:
 341         if (gl->gl_aspace)
 342                 gfs2_aspace_put(gl->gl_aspace);
 343
 344  fail:
 345         kmem_cache_free(gfs2_glock_cachep, gl);
 346
 347         return error;
 348 }
 349
 350 /**
 351  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
 352  * @gl: the glock
 353  * @state: the state we're requesting
 354  * @flags: the modifier flags
 355  * @gh: the holder structure
 356  *
 357  */
 358
 359 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
 360                       struct gfs2_holder *gh)
 361 {
 362         INIT_LIST_HEAD(&gh->gh_list);
 363         gh->gh_gl = gl;
 364         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 365         gh->gh_owner = current;
 366         gh->gh_state = state;
 367         gh->gh_flags = flags;
 368         gh->gh_error = 0;
 369         gh->gh_iflags = 0;
 370         init_completion(&gh->gh_wait);
 371
 372         if (gh->gh_state == LM_ST_EXCLUSIVE)
 373                 gh->gh_flags |= GL_LOCAL_EXCL;
 374
 375         gfs2_glock_hold(gl);
 376 }
 377
 378 /**
 379  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
 380  * @state: the state we're requesting
 381  * @flags: the modifier flags
 382  * @gh: the holder structure
 383  *
 384  * Don't mess with the glock.
 385  *
 386  */
 387
 388 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
 389 {
 390         gh->gh_state = state;
 391         gh->gh_flags = flags;
 392         if (gh->gh_state == LM_ST_EXCLUSIVE)
 393                 gh->gh_flags |= GL_LOCAL_EXCL;
 394
 395         gh->gh_iflags &= 1 << HIF_ALLOCED;
 396         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 397 }
 398
 399 /**
 400  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
 401  * @gh: the holder structure
 402  *
 403  */
 404
 405 void gfs2_holder_uninit(struct gfs2_holder *gh)
 406 {
 407         gfs2_glock_put(gh->gh_gl);
 408         gh->gh_gl = NULL;
 409         gh->gh_ip = 0;
 410 }
 411
 412 /**
 413  * gfs2_holder_get - get a struct gfs2_holder structure
 414  * @gl: the glock
 415  * @state: the state we're requesting
 416  * @flags: the modifier flags
 417  * @gfp_flags: __GFP_NOFAIL
 418  *
 419  * Figure out how big an impact this function has.  Either:
 420  * 1) Replace it with a cache of structures hanging off the struct gfs2_sbd
 421  * 2) Leave it like it is
 422  *
 423  * Returns: the holder structure, NULL on ENOMEM
 424  */
 425
 426 static struct gfs2_holder *gfs2_holder_get(struct gfs2_glock *gl,
 427                                            unsigned int state,
 428                                            int flags, gfp_t gfp_flags)
 429 {
 430         struct gfs2_holder *gh;
 431
 432         gh = kmalloc(sizeof(struct gfs2_holder), gfp_flags);
 433         if (!gh)
 434                 return NULL;
 435
 436         gfs2_holder_init(gl, state, flags, gh);
 437         set_bit(HIF_ALLOCED, &gh->gh_iflags);
 438         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 439         return gh;
 440 }
 441
 442 /**
 443  * gfs2_holder_put - get rid of a struct gfs2_holder structure
 444  * @gh: the holder structure
 445  *
 446  */
 447
 448 static void gfs2_holder_put(struct gfs2_holder *gh)
 449 {
 450         gfs2_holder_uninit(gh);
 451         kfree(gh);
 452 }
 453
 454 /**
 455  * rq_mutex - process a mutex request in the queue
 456  * @gh: the glock holder
 457  *
 458  * Returns: 1 if the queue is blocked
 459  */
 460
 461 static int rq_mutex(struct gfs2_holder *gh)
 462 {
 463         struct gfs2_glock *gl = gh->gh_gl;
 464
 465         list_del_init(&gh->gh_list);
 466         /*  gh->gh_error never examined.  */
 467         set_bit(GLF_LOCK, &gl->gl_flags);
 468         complete(&gh->gh_wait);
 469
 470         return 1;
 471 }
 472
 473 /**
 474  * rq_promote - process a promote request in the queue
 475  * @gh: the glock holder
 476  *
 477  * Acquire a new inter-node lock, or change a lock state to more restrictive.
 478  *
 479  * Returns: 1 if the queue is blocked
 480  */
 481
 482 static int rq_promote(struct gfs2_holder *gh)
 483 {
 484         struct gfs2_glock *gl = gh->gh_gl;
 485         struct gfs2_sbd *sdp = gl->gl_sbd;
 486         struct gfs2_glock_operations *glops = gl->gl_ops;
 487
 488         if (!relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
 489                 if (list_empty(&gl->gl_holders)) {
 490                         gl->gl_req_gh = gh;
 491                         set_bit(GLF_LOCK, &gl->gl_flags);
 492                         spin_unlock(&gl->gl_spin);
 493
 494                         if (atomic_read(&sdp->sd_reclaim_count) >
 495                             gfs2_tune_get(sdp, gt_reclaim_limit) &&
 496                             !(gh->gh_flags & LM_FLAG_PRIORITY)) {
 497                                 gfs2_reclaim_glock(sdp);
 498                                 gfs2_reclaim_glock(sdp);
 499                         }
 500
 501                         glops->go_xmote_th(gl, gh->gh_state,
 502                                            gh->gh_flags);
 503
 504                         spin_lock(&gl->gl_spin);
 505                 }
 506                 return 1;
 507         }
 508
 509         if (list_empty(&gl->gl_holders)) {
 510                 set_bit(HIF_FIRST, &gh->gh_iflags);
 511                 set_bit(GLF_LOCK, &gl->gl_flags);
 512         } else {
 513                 struct gfs2_holder *next_gh;
 514                 if (gh->gh_flags & GL_LOCAL_EXCL)
 515                         return 1;
 516                 next_gh = list_entry(gl->gl_holders.next, struct gfs2_holder,
 517                                      gh_list);
 518                 if (next_gh->gh_flags & GL_LOCAL_EXCL)
 519                          return 1;
 520         }
 521
 522         list_move_tail(&gh->gh_list, &gl->gl_holders);
 523         gh->gh_error = 0;
 524         set_bit(HIF_HOLDER, &gh->gh_iflags);
 525
 526         complete(&gh->gh_wait);
 527
 528         return 0;
 529 }
 530
 531 /**
 532  * rq_demote - process a demote request in the queue
 533  * @gh: the glock holder
 534  *
 535  * Returns: 1 if the queue is blocked
 536  */
 537
 538 static int rq_demote(struct gfs2_holder *gh)
 539 {
 540         struct gfs2_glock *gl = gh->gh_gl;
 541         struct gfs2_glock_operations *glops = gl->gl_ops;
 542
 543         if (!list_empty(&gl->gl_holders))
 544                 return 1;
 545
 546         if (gl->gl_state == gh->gh_state || gl->gl_state == LM_ST_UNLOCKED) {
 547                 list_del_init(&gh->gh_list);
 548                 gh->gh_error = 0;
 549                 spin_unlock(&gl->gl_spin);
 550                 if (test_bit(HIF_DEALLOC, &gh->gh_iflags))
 551                         gfs2_holder_put(gh);
 552                 else
 553                         complete(&gh->gh_wait);
 554                 spin_lock(&gl->gl_spin);
 555         } else {
 556                 gl->gl_req_gh = gh;
 557                 set_bit(GLF_LOCK, &gl->gl_flags);
 558                 spin_unlock(&gl->gl_spin);
 559
 560                 if (gh->gh_state == LM_ST_UNLOCKED ||
 561                     gl->gl_state != LM_ST_EXCLUSIVE)
 562                         glops->go_drop_th(gl);
 563                 else
 564                         glops->go_xmote_th(gl, gh->gh_state, gh->gh_flags);
 565
 566                 spin_lock(&gl->gl_spin);
 567         }
 568
 569         return 0;
 570 }
 571
 572 /**
 573  * rq_greedy - process a queued request to drop greedy status
 574  * @gh: the glock holder
 575  *
 576  * Returns: 1 if the queue is blocked
 577  */
 578
 579 static int rq_greedy(struct gfs2_holder *gh)
 580 {
 581         struct gfs2_glock *gl = gh->gh_gl;
 582
 583         list_del_init(&gh->gh_list);
 584         /*  gh->gh_error never examined.  */
 585         clear_bit(GLF_GREEDY, &gl->gl_flags);
 586         spin_unlock(&gl->gl_spin);
 587
 588         gfs2_holder_uninit(gh);
 589         kfree(container_of(gh, struct greedy, gr_gh));
 590
 591         spin_lock(&gl->gl_spin);
 592
 593         return 0;
 594 }
 595
 596 /**
 597  * run_queue - process holder structures on a glock
 598  * @gl: the glock
 599  *
 600  */
 601 static void run_queue(struct gfs2_glock *gl)
 602 {
 603         struct gfs2_holder *gh;
 604         int blocked = 1;
 605
 606         for (;;) {
 607                 if (test_bit(GLF_LOCK, &gl->gl_flags))
 608                         break;
 609
 610                 if (!list_empty(&gl->gl_waiters1)) {
 611                         gh = list_entry(gl->gl_waiters1.next,
 612                                         struct gfs2_holder, gh_list);
 613
 614                         if (test_bit(HIF_MUTEX, &gh->gh_iflags))
 615                                 blocked = rq_mutex(gh);
 616                         else
 617                                 gfs2_assert_warn(gl->gl_sbd, 0);
 618
 619                 } else if (!list_empty(&gl->gl_waiters2) &&
 620                            !test_bit(GLF_SKIP_WAITERS2, &gl->gl_flags)) {
 621                         gh = list_entry(gl->gl_waiters2.next,
 622                                         struct gfs2_holder, gh_list);
 623
 624                         if (test_bit(HIF_DEMOTE, &gh->gh_iflags))
 625                                 blocked = rq_demote(gh);
 626                         else if (test_bit(HIF_GREEDY, &gh->gh_iflags))
 627                                 blocked = rq_greedy(gh);
 628                         else
 629                                 gfs2_assert_warn(gl->gl_sbd, 0);
 630
 631                 } else if (!list_empty(&gl->gl_waiters3)) {
 632                         gh = list_entry(gl->gl_waiters3.next,
 633                                         struct gfs2_holder, gh_list);
 634
 635                         if (test_bit(HIF_PROMOTE, &gh->gh_iflags))
 636                                 blocked = rq_promote(gh);
 637                         else
 638                                 gfs2_assert_warn(gl->gl_sbd, 0);
 639
 640                 } else
 641                         break;
 642
 643                 if (blocked)
 644                         break;
 645         }
 646 }
 647
 648 /**
 649  * gfs2_glmutex_lock - acquire a local lock on a glock
 650  * @gl: the glock
 651  *
 652  * Gives caller exclusive access to manipulate a glock structure.
 653  */
 654
 655 void gfs2_glmutex_lock(struct gfs2_glock *gl)
 656 {
 657         struct gfs2_holder gh;
 658
 659         gfs2_holder_init(gl, 0, 0, &gh);
 660         set_bit(HIF_MUTEX, &gh.gh_iflags);
 661
 662         spin_lock(&gl->gl_spin);
 663         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
 664                 list_add_tail(&gh.gh_list, &gl->gl_waiters1);
 665         else
 666                 complete(&gh.gh_wait);
 667         spin_unlock(&gl->gl_spin);
 668
 669         wait_for_completion(&gh.gh_wait);
 670         gfs2_holder_uninit(&gh);
 671 }
 672
 673 /**
 674  * gfs2_glmutex_trylock - try to acquire a local lock on a glock
 675  * @gl: the glock
 676  *
 677  * Returns: 1 if the glock is acquired
 678  */
 679
 680 static int gfs2_glmutex_trylock(struct gfs2_glock *gl)
 681 {
 682         int acquired = 1;
 683
 684         spin_lock(&gl->gl_spin);
 685         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
 686                 acquired = 0;
 687         spin_unlock(&gl->gl_spin);
 688
 689         return acquired;
 690 }
 691
 692 /**
 693  * gfs2_glmutex_unlock - release a local lock on a glock
 694  * @gl: the glock
 695  *
 696  */
 697
 698 void gfs2_glmutex_unlock(struct gfs2_glock *gl)
 699 {
 700         spin_lock(&gl->gl_spin);
 701         clear_bit(GLF_LOCK, &gl->gl_flags);
 702         run_queue(gl);
 703         BUG_ON(!spin_is_locked(&gl->gl_spin));
 704         spin_unlock(&gl->gl_spin);
 705 }
 706
 707 /**
 708  * handle_callback - add a demote request to a lock's queue
 709  * @gl: the glock
 710  * @state: the state the caller wants us to change to
 711  *
 712  */
 713
 714 static void handle_callback(struct gfs2_glock *gl, unsigned int state)
 715 {
 716         struct gfs2_holder *gh, *new_gh = NULL;
 717
 718  restart:
 719         spin_lock(&gl->gl_spin);
 720
 721         list_for_each_entry(gh, &gl->gl_waiters2, gh_list) {
 722                 if (test_bit(HIF_DEMOTE, &gh->gh_iflags) &&
 723                     gl->gl_req_gh != gh) {
 724                         if (gh->gh_state != state)
 725                                 gh->gh_state = LM_ST_UNLOCKED;
 726                         goto out;
 727                 }
 728         }
 729
 730         if (new_gh) {
 731                 list_add_tail(&new_gh->gh_list, &gl->gl_waiters2);
 732                 new_gh = NULL;
 733         } else {
 734                 spin_unlock(&gl->gl_spin);
 735
 736                 new_gh = gfs2_holder_get(gl, state, LM_FLAG_TRY,
 737                                          GFP_KERNEL | __GFP_NOFAIL),
 738                 set_bit(HIF_DEMOTE, &new_gh->gh_iflags);
 739                 set_bit(HIF_DEALLOC, &new_gh->gh_iflags);
 740
 741                 goto restart;
 742         }
 743
 744  out:
 745         spin_unlock(&gl->gl_spin);
 746
 747         if (new_gh)
 748                 gfs2_holder_put(new_gh);
 749 }
 750
 751 /**
 752  * state_change - record that the glock is now in a different state
 753  * @gl: the glock
 754  * @new_state the new state
 755  *
 756  */
 757
 758 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
 759 {
 760         int held1, held2;
 761
 762         held1 = (gl->gl_state != LM_ST_UNLOCKED);
 763         held2 = (new_state != LM_ST_UNLOCKED);
 764
 765         if (held1 != held2) {
 766                 if (held2)
 767                         gfs2_glock_hold(gl);
 768                 else
 769                         gfs2_glock_put(gl);
 770         }
 771
 772         gl->gl_state = new_state;
 773 }
 774
 775 /**
 776  * xmote_bh - Called after the lock module is done acquiring a lock
 777  * @gl: The glock in question
 778  * @ret: the int returned from the lock module
 779  *
 780  */
 781
 782 static void xmote_bh(struct gfs2_glock *gl, unsigned int ret)
 783 {
 784         struct gfs2_sbd *sdp = gl->gl_sbd;
 785         struct gfs2_glock_operations *glops = gl->gl_ops;
 786         struct gfs2_holder *gh = gl->gl_req_gh;
 787         int prev_state = gl->gl_state;
 788         int op_done = 1;
 789
 790         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
 791         gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
 792         gfs2_assert_warn(sdp, !(ret & LM_OUT_ASYNC));
 793
 794         state_change(gl, ret & LM_OUT_ST_MASK);
 795
 796         if (prev_state != LM_ST_UNLOCKED && !(ret & LM_OUT_CACHEABLE)) {
 797                 if (glops->go_inval)
 798                         glops->go_inval(gl, DIO_METADATA | DIO_DATA);
 799         } else if (gl->gl_state == LM_ST_DEFERRED) {
 800                 /* We might not want to do this here.
 801                    Look at moving to the inode glops. */
 802                 if (glops->go_inval)
 803                         glops->go_inval(gl, DIO_DATA);
 804         }
 805
 806         /*  Deal with each possible exit condition  */
 807
 808         if (!gh)
 809                 gl->gl_stamp = jiffies;
 810
 811         else if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
 812                 spin_lock(&gl->gl_spin);
 813                 list_del_init(&gh->gh_list);
 814                 gh->gh_error = -EIO;
 815                 spin_unlock(&gl->gl_spin);
 816
 817         } else if (test_bit(HIF_DEMOTE, &gh->gh_iflags)) {
 818                 spin_lock(&gl->gl_spin);
 819                 list_del_init(&gh->gh_list);
 820                 if (gl->gl_state == gh->gh_state ||
 821                     gl->gl_state == LM_ST_UNLOCKED)
 822                         gh->gh_error = 0;
 823                 else {
 824                         if (gfs2_assert_warn(sdp, gh->gh_flags &
 825                                         (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) == -1)
 826                                 fs_warn(sdp, "ret = 0x%.8X\n", ret);
 827                         gh->gh_error = GLR_TRYFAILED;
 828                 }
 829                 spin_unlock(&gl->gl_spin);
 830
 831                 if (ret & LM_OUT_CANCELED)
 832                         handle_callback(gl, LM_ST_UNLOCKED); /* Lame */
 833
 834         } else if (ret & LM_OUT_CANCELED) {
 835                 spin_lock(&gl->gl_spin);
 836                 list_del_init(&gh->gh_list);
 837                 gh->gh_error = GLR_CANCELED;
 838                 spin_unlock(&gl->gl_spin);
 839
 840         } else if (relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
 841                 spin_lock(&gl->gl_spin);
 842                 list_move_tail(&gh->gh_list, &gl->gl_holders);
 843                 gh->gh_error = 0;
 844                 set_bit(HIF_HOLDER, &gh->gh_iflags);
 845                 spin_unlock(&gl->gl_spin);
 846
 847                 set_bit(HIF_FIRST, &gh->gh_iflags);
 848
 849                 op_done = 0;
 850
 851         } else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
 852                 spin_lock(&gl->gl_spin);
 853                 list_del_init(&gh->gh_list);
 854                 gh->gh_error = GLR_TRYFAILED;
 855                 spin_unlock(&gl->gl_spin);
 856
 857         } else {
 858                 if (gfs2_assert_withdraw(sdp, 0) == -1)
 859                         fs_err(sdp, "ret = 0x%.8X\n", ret);
 860         }
 861
 862         if (glops->go_xmote_bh)
 863                 glops->go_xmote_bh(gl);
 864
 865         if (op_done) {
 866                 spin_lock(&gl->gl_spin);
 867                 gl->gl_req_gh = NULL;
 868                 gl->gl_req_bh = NULL;
 869                 clear_bit(GLF_LOCK, &gl->gl_flags);
 870                 run_queue(gl);
 871                 spin_unlock(&gl->gl_spin);
 872         }
 873
 874         gfs2_glock_put(gl);
 875
 876         if (gh) {
 877                 if (test_bit(HIF_DEALLOC, &gh->gh_iflags))
 878                         gfs2_holder_put(gh);
 879                 else
 880                         complete(&gh->gh_wait);
 881         }
 882 }
 883
 884 /**
 885  * gfs2_glock_xmote_th - Call into the lock module to acquire or change a glock
 886  * @gl: The glock in question
 887  * @state: the requested state
 888  * @flags: modifier flags to the lock call
 889  *
 890  */
 891
 892 void gfs2_glock_xmote_th(struct gfs2_glock *gl, unsigned int state, int flags)
 893 {
 894         struct gfs2_sbd *sdp = gl->gl_sbd;
 895         struct gfs2_glock_operations *glops = gl->gl_ops;
 896         int lck_flags = flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB |
 897                                  LM_FLAG_NOEXP | LM_FLAG_ANY |
 898                                  LM_FLAG_PRIORITY);
 899         unsigned int lck_ret;
 900
 901         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
 902         gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
 903         gfs2_assert_warn(sdp, state != LM_ST_UNLOCKED);
 904         gfs2_assert_warn(sdp, state != gl->gl_state);
 905
 906         if (gl->gl_state == LM_ST_EXCLUSIVE) {
 907                 if (glops->go_sync)
 908                         glops->go_sync(gl,
 909                                        DIO_METADATA | DIO_DATA | DIO_RELEASE);
 910         }
 911
 912         gfs2_glock_hold(gl);
 913         gl->gl_req_bh = xmote_bh;
 914
 915         lck_ret = gfs2_lm_lock(sdp, gl->gl_lock, gl->gl_state, state,
 916                                lck_flags);
 917
 918         if (gfs2_assert_withdraw(sdp, !(lck_ret & LM_OUT_ERROR)))
 919                 return;
 920
 921         if (lck_ret & LM_OUT_ASYNC)
 922                 gfs2_assert_warn(sdp, lck_ret == LM_OUT_ASYNC);
 923         else
 924                 xmote_bh(gl, lck_ret);
 925 }
 926
 927 /**
 928  * drop_bh - Called after a lock module unlock completes
 929  * @gl: the glock
 930  * @ret: the return status
 931  *
 932  * Doesn't wake up the process waiting on the struct gfs2_holder (if any)
 933  * Doesn't drop the reference on the glock the top half took out
 934  *
 935  */
 936
 937 static void drop_bh(struct gfs2_glock *gl, unsigned int ret)
 938 {
 939         struct gfs2_sbd *sdp = gl->gl_sbd;
 940         struct gfs2_glock_operations *glops = gl->gl_ops;
 941         struct gfs2_holder *gh = gl->gl_req_gh;
 942
 943         clear_bit(GLF_PREFETCH, &gl->gl_flags);
 944
 945         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
 946         gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
 947         gfs2_assert_warn(sdp, !ret);
 948
 949         state_change(gl, LM_ST_UNLOCKED);
 950
 951         if (glops->go_inval)
 952                 glops->go_inval(gl, DIO_METADATA | DIO_DATA);
 953
 954         if (gh) {
 955                 spin_lock(&gl->gl_spin);
 956                 list_del_init(&gh->gh_list);
 957                 gh->gh_error = 0;
 958                 spin_unlock(&gl->gl_spin);
 959         }
 960
 961         if (glops->go_drop_bh)
 962                 glops->go_drop_bh(gl);
 963
 964         spin_lock(&gl->gl_spin);
 965         gl->gl_req_gh = NULL;
 966         gl->gl_req_bh = NULL;
 967         clear_bit(GLF_LOCK, &gl->gl_flags);
 968         run_queue(gl);
 969         spin_unlock(&gl->gl_spin);
 970
 971         gfs2_glock_put(gl);
 972
 973         if (gh) {
 974                 if (test_bit(HIF_DEALLOC, &gh->gh_iflags))
 975                         gfs2_holder_put(gh);
 976                 else
 977                         complete(&gh->gh_wait);
 978         }
 979 }
 980
 981 /**
 982  * gfs2_glock_drop_th - call into the lock module to unlock a lock
 983  * @gl: the glock
 984  *
 985  */
 986
 987 void gfs2_glock_drop_th(struct gfs2_glock *gl)
 988 {
 989         struct gfs2_sbd *sdp = gl->gl_sbd;
 990         struct gfs2_glock_operations *glops = gl->gl_ops;
 991         unsigned int ret;
 992
 993         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
 994         gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
 995         gfs2_assert_warn(sdp, gl->gl_state != LM_ST_UNLOCKED);
 996
 997         if (gl->gl_state == LM_ST_EXCLUSIVE) {
 998                 if (glops->go_sync)
 999                         glops->go_sync(gl,
1000                                        DIO_METADATA | DIO_DATA | DIO_RELEASE);
1001         }
1002
1003         gfs2_glock_hold(gl);
1004         gl->gl_req_bh = drop_bh;
1005
1006         ret = gfs2_lm_unlock(sdp, gl->gl_lock, gl->gl_state);
1007
1008         if (gfs2_assert_withdraw(sdp, !(ret & LM_OUT_ERROR)))
1009                 return;
1010
1011         if (!ret)
1012                 drop_bh(gl, ret);
1013         else
1014                 gfs2_assert_warn(sdp, ret == LM_OUT_ASYNC);
1015 }
1016
1017 /**
1018  * do_cancels - cancel requests for locks stuck waiting on an expire flag
1019  * @gh: the LM_FLAG_PRIORITY holder waiting to acquire the lock
1020  *
1021  * Don't cancel GL_NOCANCEL requests.
1022  */
1023
1024 static void do_cancels(struct gfs2_holder *gh)
1025 {
1026         struct gfs2_glock *gl = gh->gh_gl;
1027
1028         spin_lock(&gl->gl_spin);
1029
1030         while (gl->gl_req_gh != gh &&
1031                !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1032                !list_empty(&gh->gh_list)) {
1033                 if (gl->gl_req_bh &&
1034                     !(gl->gl_req_gh &&
1035                       (gl->gl_req_gh->gh_flags & GL_NOCANCEL))) {
1036                         spin_unlock(&gl->gl_spin);
1037                         gfs2_lm_cancel(gl->gl_sbd, gl->gl_lock);
1038                         msleep(100);
1039                         spin_lock(&gl->gl_spin);
1040                 } else {
1041                         spin_unlock(&gl->gl_spin);
1042                         msleep(100);
1043                         spin_lock(&gl->gl_spin);
1044                 }
1045         }
1046
1047         spin_unlock(&gl->gl_spin);
1048 }
1049
1050 /**
1051  * glock_wait_internal - wait on a glock acquisition
1052  * @gh: the glock holder
1053  *
1054  * Returns: 0 on success
1055  */
1056
1057 static int glock_wait_internal(struct gfs2_holder *gh)
1058 {
1059         struct gfs2_glock *gl = gh->gh_gl;
1060         struct gfs2_sbd *sdp = gl->gl_sbd;
1061         struct gfs2_glock_operations *glops = gl->gl_ops;
1062
1063         if (test_bit(HIF_ABORTED, &gh->gh_iflags))
1064                 return -EIO;
1065
1066         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1067                 spin_lock(&gl->gl_spin);
1068                 if (gl->gl_req_gh != gh &&
1069                     !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1070                     !list_empty(&gh->gh_list)) {
1071                         list_del_init(&gh->gh_list);
1072                         gh->gh_error = GLR_TRYFAILED;
1073                         run_queue(gl);
1074                         spin_unlock(&gl->gl_spin);
1075                         return gh->gh_error;
1076                 }
1077                 spin_unlock(&gl->gl_spin);
1078         }
1079
1080         if (gh->gh_flags & LM_FLAG_PRIORITY)
1081                 do_cancels(gh);
1082
1083         wait_for_completion(&gh->gh_wait);
1084
1085         if (gh->gh_error)
1086                 return gh->gh_error;
1087
1088         gfs2_assert_withdraw(sdp, test_bit(HIF_HOLDER, &gh->gh_iflags));
1089         gfs2_assert_withdraw(sdp, relaxed_state_ok(gl->gl_state,
1090                                                    gh->gh_state,
1091                                                    gh->gh_flags));
1092
1093         if (test_bit(HIF_FIRST, &gh->gh_iflags)) {
1094                 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1095
1096                 if (glops->go_lock) {
1097                         gh->gh_error = glops->go_lock(gh);
1098                         if (gh->gh_error) {
1099                                 spin_lock(&gl->gl_spin);
1100                                 list_del_init(&gh->gh_list);
1101                                 spin_unlock(&gl->gl_spin);
1102                         }
1103                 }
1104
1105                 spin_lock(&gl->gl_spin);
1106                 gl->gl_req_gh = NULL;
1107                 gl->gl_req_bh = NULL;
1108                 clear_bit(GLF_LOCK, &gl->gl_flags);
1109                 run_queue(gl);
1110                 spin_unlock(&gl->gl_spin);
1111         }
1112
1113         return gh->gh_error;
1114 }
1115
1116 static inline struct gfs2_holder *
1117 find_holder_by_owner(struct list_head *head, struct task_struct *owner)
1118 {
1119         struct gfs2_holder *gh;
1120
1121         list_for_each_entry(gh, head, gh_list) {
1122                 if (gh->gh_owner == owner)
1123                         return gh;
1124         }
1125
1126         return NULL;
1127 }
1128
1129 /**
1130  * add_to_queue - Add a holder to the wait queue (but look for recursion)
1131  * @gh: the holder structure to add
1132  *
1133  */
1134
1135 static void add_to_queue(struct gfs2_holder *gh)
1136 {
1137         struct gfs2_glock *gl = gh->gh_gl;
1138         struct gfs2_holder *existing;
1139
1140         BUG_ON(!gh->gh_owner);
1141
1142         existing = find_holder_by_owner(&gl->gl_holders, gh->gh_owner);
1143         if (existing) {
1144                 print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
1145                 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1146                 BUG();
1147         }
1148
1149         existing = find_holder_by_owner(&gl->gl_waiters3, gh->gh_owner);
1150         if (existing) {
1151                 print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
1152                 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1153                 BUG();
1154         }
1155
1156         if (gh->gh_flags & LM_FLAG_PRIORITY)
1157                 list_add(&gh->gh_list, &gl->gl_waiters3);
1158         else
1159                 list_add_tail(&gh->gh_list, &gl->gl_waiters3);
1160 }
1161
1162 /**
1163  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1164  * @gh: the holder structure
1165  *
1166  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1167  *
1168  * Returns: 0, GLR_TRYFAILED, or errno on failure
1169  */
1170
1171 int gfs2_glock_nq(struct gfs2_holder *gh)
1172 {
1173         struct gfs2_glock *gl = gh->gh_gl;
1174         struct gfs2_sbd *sdp = gl->gl_sbd;
1175         int error = 0;
1176
1177  restart:
1178         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1179                 set_bit(HIF_ABORTED, &gh->gh_iflags);
1180                 return -EIO;
1181         }
1182
1183         set_bit(HIF_PROMOTE, &gh->gh_iflags);
1184
1185         spin_lock(&gl->gl_spin);
1186         add_to_queue(gh);
1187         run_queue(gl);
1188         spin_unlock(&gl->gl_spin);
1189
1190         if (!(gh->gh_flags & GL_ASYNC)) {
1191                 error = glock_wait_internal(gh);
1192                 if (error == GLR_CANCELED) {
1193                         msleep(100);
1194                         goto restart;
1195                 }
1196         }
1197
1198         clear_bit(GLF_PREFETCH, &gl->gl_flags);
1199
1200         return error;
1201 }
1202
1203 /**
1204  * gfs2_glock_poll - poll to see if an async request has been completed
1205  * @gh: the holder
1206  *
1207  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1208  */
1209
1210 int gfs2_glock_poll(struct gfs2_holder *gh)
1211 {
1212         struct gfs2_glock *gl = gh->gh_gl;
1213         int ready = 0;
1214
1215         spin_lock(&gl->gl_spin);
1216
1217         if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1218                 ready = 1;
1219         else if (list_empty(&gh->gh_list)) {
1220                 if (gh->gh_error == GLR_CANCELED) {
1221                         spin_unlock(&gl->gl_spin);
1222                         msleep(100);
1223                         if (gfs2_glock_nq(gh))
1224                                 return 1;
1225                         return 0;
1226                 } else
1227                         ready = 1;
1228         }
1229
1230         spin_unlock(&gl->gl_spin);
1231
1232         return ready;
1233 }
1234
1235 /**
1236  * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
1237  * @gh: the holder structure
1238  *
1239  * Returns: 0, GLR_TRYFAILED, or errno on failure
1240  */
1241
1242 int gfs2_glock_wait(struct gfs2_holder *gh)
1243 {
1244         int error;
1245
1246         error = glock_wait_internal(gh);
1247         if (error == GLR_CANCELED) {
1248                 msleep(100);
1249                 gh->gh_flags &= ~GL_ASYNC;
1250                 error = gfs2_glock_nq(gh);
1251         }
1252
1253         return error;
1254 }
1255
1256 /**
1257  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1258  * @gh: the glock holder
1259  *
1260  */
1261
1262 void gfs2_glock_dq(struct gfs2_holder *gh)
1263 {
1264         struct gfs2_glock *gl = gh->gh_gl;
1265         struct gfs2_glock_operations *glops = gl->gl_ops;
1266
1267         if (gh->gh_flags & GL_SYNC)
1268                 set_bit(GLF_SYNC, &gl->gl_flags);
1269
1270         if (gh->gh_flags & GL_NOCACHE)
1271                 handle_callback(gl, LM_ST_UNLOCKED);
1272
1273         gfs2_glmutex_lock(gl);
1274
1275         spin_lock(&gl->gl_spin);
1276         list_del_init(&gh->gh_list);
1277
1278         if (list_empty(&gl->gl_holders)) {
1279                 spin_unlock(&gl->gl_spin);
1280
1281                 if (glops->go_unlock)
1282                         glops->go_unlock(gh);
1283
1284                 if (test_bit(GLF_SYNC, &gl->gl_flags)) {
1285                         if (glops->go_sync)
1286                                 glops->go_sync(gl, DIO_METADATA | DIO_DATA);
1287                 }
1288
1289                 gl->gl_stamp = jiffies;
1290
1291                 spin_lock(&gl->gl_spin);
1292         }
1293
1294         clear_bit(GLF_LOCK, &gl->gl_flags);
1295         run_queue(gl);
1296         spin_unlock(&gl->gl_spin);
1297 }
1298
1299 /**
1300  * gfs2_glock_prefetch - Try to prefetch a glock
1301  * @gl: the glock
1302  * @state: the state to prefetch in
1303  * @flags: flags passed to go_xmote_th()
1304  *
1305  */
1306
1307 static void gfs2_glock_prefetch(struct gfs2_glock *gl, unsigned int state,
1308                                 int flags)
1309 {
1310         struct gfs2_glock_operations *glops = gl->gl_ops;
1311
1312         spin_lock(&gl->gl_spin);
1313
1314         if (test_bit(GLF_LOCK, &gl->gl_flags) ||
1315             !list_empty(&gl->gl_holders) ||
1316             !list_empty(&gl->gl_waiters1) ||
1317             !list_empty(&gl->gl_waiters2) ||
1318             !list_empty(&gl->gl_waiters3) ||
1319             relaxed_state_ok(gl->gl_state, state, flags)) {
1320                 spin_unlock(&gl->gl_spin);
1321                 return;
1322         }
1323
1324         set_bit(GLF_PREFETCH, &gl->gl_flags);
1325         set_bit(GLF_LOCK, &gl->gl_flags);
1326         spin_unlock(&gl->gl_spin);
1327
1328         glops->go_xmote_th(gl, state, flags);
1329 }
1330
1331 static void greedy_work(void *data)
1332 {
1333         struct greedy *gr = data;
1334         struct gfs2_holder *gh = &gr->gr_gh;
1335         struct gfs2_glock *gl = gh->gh_gl;
1336         struct gfs2_glock_operations *glops = gl->gl_ops;
1337
1338         clear_bit(GLF_SKIP_WAITERS2, &gl->gl_flags);
1339
1340         if (glops->go_greedy)
1341                 glops->go_greedy(gl);
1342
1343         spin_lock(&gl->gl_spin);
1344
1345         if (list_empty(&gl->gl_waiters2)) {
1346                 clear_bit(GLF_GREEDY, &gl->gl_flags);
1347                 spin_unlock(&gl->gl_spin);
1348                 gfs2_holder_uninit(gh);
1349                 kfree(gr);
1350         } else {
1351                 gfs2_glock_hold(gl);
1352                 list_add_tail(&gh->gh_list, &gl->gl_waiters2);
1353                 run_queue(gl);
1354                 spin_unlock(&gl->gl_spin);
1355                 gfs2_glock_put(gl);
1356         }
1357 }
1358
1359 /**
1360  * gfs2_glock_be_greedy -
1361  * @gl:
1362  * @time:
1363  *
1364  * Returns: 0 if go_greedy will be called, 1 otherwise
1365  */
1366
1367 int gfs2_glock_be_greedy(struct gfs2_glock *gl, unsigned int time)
1368 {
1369         struct greedy *gr;
1370         struct gfs2_holder *gh;
1371
1372         if (!time ||
1373             gl->gl_sbd->sd_args.ar_localcaching ||
1374             test_and_set_bit(GLF_GREEDY, &gl->gl_flags))
1375                 return 1;
1376
1377         gr = kmalloc(sizeof(struct greedy), GFP_KERNEL);
1378         if (!gr) {
1379                 clear_bit(GLF_GREEDY, &gl->gl_flags);
1380                 return 1;
1381         }
1382         gh = &gr->gr_gh;
1383
1384         gfs2_holder_init(gl, 0, 0, gh);
1385         set_bit(HIF_GREEDY, &gh->gh_iflags);
1386         INIT_WORK(&gr->gr_work, greedy_work, gr);
1387
1388         set_bit(GLF_SKIP_WAITERS2, &gl->gl_flags);
1389         schedule_delayed_work(&gr->gr_work, time);
1390
1391         return 0;
1392 }
1393
1394 /**
1395  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1396  * @gh: the holder structure
1397  *
1398  */
1399
1400 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1401 {
1402         gfs2_glock_dq(gh);
1403         gfs2_holder_uninit(gh);
1404 }
1405
1406 /**
1407  * gfs2_glock_nq_num - acquire a glock based on lock number
1408  * @sdp: the filesystem
1409  * @number: the lock number
1410  * @glops: the glock operations for the type of glock
1411  * @state: the state to acquire the glock in
1412  * @flags: modifier flags for the aquisition
1413  * @gh: the struct gfs2_holder
1414  *
1415  * Returns: errno
1416  */
1417
1418 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, uint64_t number,
1419                       struct gfs2_glock_operations *glops, unsigned int state,
1420                       int flags, struct gfs2_holder *gh)
1421 {
1422         struct gfs2_glock *gl;
1423         int error;
1424
1425         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1426         if (!error) {
1427                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1428                 gfs2_glock_put(gl);
1429         }
1430
1431         return error;
1432 }
1433
1434 /**
1435  * glock_compare - Compare two struct gfs2_glock structures for sorting
1436  * @arg_a: the first structure
1437  * @arg_b: the second structure
1438  *
1439  */
1440
1441 static int glock_compare(const void *arg_a, const void *arg_b)
1442 {
1443         struct gfs2_holder *gh_a = *(struct gfs2_holder **)arg_a;
1444         struct gfs2_holder *gh_b = *(struct gfs2_holder **)arg_b;
1445         struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1446         struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1447         int ret = 0;
1448
1449         if (a->ln_number > b->ln_number)
1450                 ret = 1;
1451         else if (a->ln_number < b->ln_number)
1452                 ret = -1;
1453         else {
1454                 if (gh_a->gh_state == LM_ST_SHARED &&
1455                     gh_b->gh_state == LM_ST_EXCLUSIVE)
1456                         ret = 1;
1457                 else if (!(gh_a->gh_flags & GL_LOCAL_EXCL) &&
1458                          (gh_b->gh_flags & GL_LOCAL_EXCL))
1459                         ret = 1;
1460         }
1461
1462         return ret;
1463 }
1464
1465 /**
1466  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1467  * @num_gh: the number of structures
1468  * @ghs: an array of struct gfs2_holder structures
1469  *
1470  * Returns: 0 on success (all glocks acquired),
1471  *          errno on failure (no glocks acquired)
1472  */
1473
1474 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1475                      struct gfs2_holder **p)
1476 {
1477         unsigned int x;
1478         int error = 0;
1479
1480         for (x = 0; x < num_gh; x++)
1481                 p[x] = &ghs[x];
1482
1483         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1484
1485         for (x = 0; x < num_gh; x++) {
1486                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1487
1488                 error = gfs2_glock_nq(p[x]);
1489                 if (error) {
1490                         while (x--)
1491                                 gfs2_glock_dq(p[x]);
1492                         break;
1493                 }
1494         }
1495
1496         return error;
1497 }
1498
1499 /**
1500  * gfs2_glock_nq_m - acquire multiple glocks
1501  * @num_gh: the number of structures
1502  * @ghs: an array of struct gfs2_holder structures
1503  *
1504  * Figure out how big an impact this function has.  Either:
1505  * 1) Replace this code with code that calls gfs2_glock_prefetch()
1506  * 2) Forget async stuff and just call nq_m_sync()
1507  * 3) Leave it like it is
1508  *
1509  * Returns: 0 on success (all glocks acquired),
1510  *          errno on failure (no glocks acquired)
1511  */
1512
1513 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1514 {
1515         int *e;
1516         unsigned int x;
1517         int borked = 0, serious = 0;
1518         int error = 0;
1519
1520         if (!num_gh)
1521                 return 0;
1522
1523         if (num_gh == 1) {
1524                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1525                 return gfs2_glock_nq(ghs);
1526         }
1527
1528         e = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL);
1529         if (!e)
1530                 return -ENOMEM;
1531
1532         for (x = 0; x < num_gh; x++) {
1533                 ghs[x].gh_flags |= LM_FLAG_TRY | GL_ASYNC;
1534                 error = gfs2_glock_nq(&ghs[x]);
1535                 if (error) {
1536                         borked = 1;
1537                         serious = error;
1538                         num_gh = x;
1539                         break;
1540                 }
1541         }
1542
1543         for (x = 0; x < num_gh; x++) {
1544                 error = e[x] = glock_wait_internal(&ghs[x]);
1545                 if (error) {
1546                         borked = 1;
1547                         if (error != GLR_TRYFAILED && error != GLR_CANCELED)
1548                                 serious = error;
1549                 }
1550         }
1551
1552         if (!borked) {
1553                 kfree(e);
1554                 return 0;
1555         }
1556
1557         for (x = 0; x < num_gh; x++)
1558                 if (!e[x])
1559                         gfs2_glock_dq(&ghs[x]);
1560
1561         if (serious)
1562                 error = serious;
1563         else {
1564                 for (x = 0; x < num_gh; x++)
1565                         gfs2_holder_reinit(ghs[x].gh_state, ghs[x].gh_flags,
1566                                           &ghs[x]);
1567                 error = nq_m_sync(num_gh, ghs, (struct gfs2_holder **)e);
1568         }
1569
1570         kfree(e);
1571
1572         return error;
1573 }
1574
1575 /**
1576  * gfs2_glock_dq_m - release multiple glocks
1577  * @num_gh: the number of structures
1578  * @ghs: an array of struct gfs2_holder structures
1579  *
1580  */
1581
1582 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1583 {
1584         unsigned int x;
1585
1586         for (x = 0; x < num_gh; x++)
1587                 gfs2_glock_dq(&ghs[x]);
1588 }
1589
1590 /**
1591  * gfs2_glock_dq_uninit_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_uninit_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_uninit(&ghs[x]);
1603 }
1604
1605 /**
1606  * gfs2_glock_prefetch_num - prefetch a glock based on lock number
1607  * @sdp: the filesystem
1608  * @number: the lock number
1609  * @glops: the glock operations for the type of glock
1610  * @state: the state to acquire the glock in
1611  * @flags: modifier flags for the aquisition
1612  *
1613  * Returns: errno
1614  */
1615
1616 void gfs2_glock_prefetch_num(struct gfs2_sbd *sdp, uint64_t number,
1617                              struct gfs2_glock_operations *glops,
1618                              unsigned int state, int flags)
1619 {
1620         struct gfs2_glock *gl;
1621         int error;
1622
1623         if (atomic_read(&sdp->sd_reclaim_count) <
1624             gfs2_tune_get(sdp, gt_reclaim_limit)) {
1625                 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1626                 if (!error) {
1627                         gfs2_glock_prefetch(gl, state, flags);
1628                         gfs2_glock_put(gl);
1629                 }
1630         }
1631 }
1632
1633 /**
1634  * gfs2_lvb_hold - attach a LVB from a glock
1635  * @gl: The glock in question
1636  *
1637  */
1638
1639 int gfs2_lvb_hold(struct gfs2_glock *gl)
1640 {
1641         int error;
1642
1643         gfs2_glmutex_lock(gl);
1644
1645         if (!atomic_read(&gl->gl_lvb_count)) {
1646                 error = gfs2_lm_hold_lvb(gl->gl_sbd, gl->gl_lock, &gl->gl_lvb);
1647                 if (error) {
1648                         gfs2_glmutex_unlock(gl);
1649                         return error;
1650                 }
1651                 gfs2_glock_hold(gl);
1652         }
1653         atomic_inc(&gl->gl_lvb_count);
1654
1655         gfs2_glmutex_unlock(gl);
1656
1657         return 0;
1658 }
1659
1660 /**
1661  * gfs2_lvb_unhold - detach a LVB from a glock
1662  * @gl: The glock in question
1663  *
1664  */
1665
1666 void gfs2_lvb_unhold(struct gfs2_glock *gl)
1667 {
1668         gfs2_glock_hold(gl);
1669         gfs2_glmutex_lock(gl);
1670
1671         gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count) > 0);
1672         if (atomic_dec_and_test(&gl->gl_lvb_count)) {
1673                 gfs2_lm_unhold_lvb(gl->gl_sbd, gl->gl_lock, gl->gl_lvb);
1674                 gl->gl_lvb = NULL;
1675                 gfs2_glock_put(gl);
1676         }
1677
1678         gfs2_glmutex_unlock(gl);
1679         gfs2_glock_put(gl);
1680 }
1681
1682 #if 0
1683 void gfs2_lvb_sync(struct gfs2_glock *gl)
1684 {
1685         gfs2_glmutex_lock(gl);
1686
1687         gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count));
1688         if (!gfs2_assert_warn(gl->gl_sbd, gfs2_glock_is_held_excl(gl)))
1689                 gfs2_lm_sync_lvb(gl->gl_sbd, gl->gl_lock, gl->gl_lvb);
1690
1691         gfs2_glmutex_unlock(gl);
1692 }
1693 #endif  /*  0  */
1694
1695 static void blocking_cb(struct gfs2_sbd *sdp, struct lm_lockname *name,
1696                         unsigned int state)
1697 {
1698         struct gfs2_glock *gl;
1699
1700         gl = gfs2_glock_find(sdp, name);
1701         if (!gl)
1702                 return;
1703
1704         if (gl->gl_ops->go_callback)
1705                 gl->gl_ops->go_callback(gl, state);
1706         handle_callback(gl, state);
1707
1708         spin_lock(&gl->gl_spin);
1709         run_queue(gl);
1710         spin_unlock(&gl->gl_spin);
1711
1712         gfs2_glock_put(gl);
1713 }
1714
1715 /**
1716  * gfs2_glock_cb - Callback used by locking module
1717  * @fsdata: Pointer to the superblock
1718  * @type: Type of callback
1719  * @data: Type dependent data pointer
1720  *
1721  * Called by the locking module when it wants to tell us something.
1722  * Either we need to drop a lock, one of our ASYNC requests completed, or
1723  * a journal from another client needs to be recovered.
1724  */
1725
1726 void gfs2_glock_cb(lm_fsdata_t *fsdata, unsigned int type, void *data)
1727 {
1728         struct gfs2_sbd *sdp = (struct gfs2_sbd *)fsdata;
1729
1730         switch (type) {
1731         case LM_CB_NEED_E:
1732                 blocking_cb(sdp, data, LM_ST_UNLOCKED);
1733                 return;
1734
1735         case LM_CB_NEED_D:
1736                 blocking_cb(sdp, data, LM_ST_DEFERRED);
1737                 return;
1738
1739         case LM_CB_NEED_S:
1740                 blocking_cb(sdp, data, LM_ST_SHARED);
1741                 return;
1742
1743         case LM_CB_ASYNC: {
1744                 struct lm_async_cb *async = data;
1745                 struct gfs2_glock *gl;
1746
1747                 gl = gfs2_glock_find(sdp, &async->lc_name);
1748                 if (gfs2_assert_warn(sdp, gl))
1749                         return;
1750                 if (!gfs2_assert_warn(sdp, gl->gl_req_bh))
1751                         gl->gl_req_bh(gl, async->lc_ret);
1752                 gfs2_glock_put(gl);
1753                 return;
1754         }
1755
1756         case LM_CB_NEED_RECOVERY:
1757                 gfs2_jdesc_make_dirty(sdp, *(unsigned int *)data);
1758                 if (sdp->sd_recoverd_process)
1759                         wake_up_process(sdp->sd_recoverd_process);
1760                 return;
1761
1762         case LM_CB_DROPLOCKS:
1763                 gfs2_gl_hash_clear(sdp, NO_WAIT);
1764                 gfs2_quota_scan(sdp);
1765                 return;
1766
1767         default:
1768                 gfs2_assert_warn(sdp, 0);
1769                 return;
1770         }
1771 }
1772
1773 /**
1774  * gfs2_try_toss_inode - try to remove a particular inode struct from cache
1775  * sdp: the filesystem
1776  * inum: the inode number
1777  *
1778  */
1779
1780 void gfs2_try_toss_inode(struct gfs2_sbd *sdp, struct gfs2_inum *inum)
1781 {
1782         struct gfs2_glock *gl;
1783         struct gfs2_inode *ip;
1784         int error;
1785
1786         error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_inode_glops,
1787                                NO_CREATE, &gl);
1788         if (error || !gl)
1789                 return;
1790
1791         if (!gfs2_glmutex_trylock(gl))
1792                 goto out;
1793
1794         ip = gl->gl_object;
1795         if (!ip)
1796                 goto out_unlock;
1797
1798         if (atomic_read(&ip->i_count))
1799                 goto out_unlock;
1800
1801         gfs2_inode_destroy(ip, 1);
1802
1803  out_unlock:
1804         gfs2_glmutex_unlock(gl);
1805
1806  out:
1807         gfs2_glock_put(gl);
1808 }
1809
1810 /**
1811  * gfs2_iopen_go_callback - Try to kick the inode/vnode associated with an
1812  *                          iopen glock from memory
1813  * @io_gl: the iopen glock
1814  * @state: the state into which the glock should be put
1815  *
1816  */
1817
1818 void gfs2_iopen_go_callback(struct gfs2_glock *io_gl, unsigned int state)
1819 {
1820         struct gfs2_glock *i_gl;
1821
1822         if (state != LM_ST_UNLOCKED)
1823                 return;
1824
1825         spin_lock(&io_gl->gl_spin);
1826         i_gl = io_gl->gl_object;
1827         if (i_gl) {
1828                 gfs2_glock_hold(i_gl);
1829                 spin_unlock(&io_gl->gl_spin);
1830         } else {
1831                 spin_unlock(&io_gl->gl_spin);
1832                 return;
1833         }
1834
1835         if (gfs2_glmutex_trylock(i_gl)) {
1836                 struct gfs2_inode *ip = i_gl->gl_object;
1837                 if (ip) {
1838                         gfs2_try_toss_vnode(ip);
1839                         gfs2_glmutex_unlock(i_gl);
1840                         gfs2_glock_schedule_for_reclaim(i_gl);
1841                         goto out;
1842                 }
1843                 gfs2_glmutex_unlock(i_gl);
1844         }
1845
1846  out:
1847         gfs2_glock_put(i_gl);
1848 }
1849
1850 /**
1851  * demote_ok - Check to see if it's ok to unlock a glock
1852  * @gl: the glock
1853  *
1854  * Returns: 1 if it's ok
1855  */
1856
1857 static int demote_ok(struct gfs2_glock *gl)
1858 {
1859         struct gfs2_sbd *sdp = gl->gl_sbd;
1860         struct gfs2_glock_operations *glops = gl->gl_ops;
1861         int demote = 1;
1862
1863         if (test_bit(GLF_STICKY, &gl->gl_flags))
1864                 demote = 0;
1865         else if (test_bit(GLF_PREFETCH, &gl->gl_flags))
1866                 demote = time_after_eq(jiffies,
1867                                     gl->gl_stamp +
1868                                     gfs2_tune_get(sdp, gt_prefetch_secs) * HZ);
1869         else if (glops->go_demote_ok)
1870                 demote = glops->go_demote_ok(gl);
1871
1872         return demote;
1873 }
1874
1875 /**
1876  * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1877  * @gl: the glock
1878  *
1879  */
1880
1881 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
1882 {
1883         struct gfs2_sbd *sdp = gl->gl_sbd;
1884
1885         spin_lock(&sdp->sd_reclaim_lock);
1886         if (list_empty(&gl->gl_reclaim)) {
1887                 gfs2_glock_hold(gl);
1888                 list_add(&gl->gl_reclaim, &sdp->sd_reclaim_list);
1889                 atomic_inc(&sdp->sd_reclaim_count);
1890         }
1891         spin_unlock(&sdp->sd_reclaim_lock);
1892
1893         wake_up(&sdp->sd_reclaim_wq);
1894 }
1895
1896 /**
1897  * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1898  * @sdp: the filesystem
1899  *
1900  * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1901  * different glock and we notice that there are a lot of glocks in the
1902  * reclaim list.
1903  *
1904  */
1905
1906 void gfs2_reclaim_glock(struct gfs2_sbd *sdp)
1907 {
1908         struct gfs2_glock *gl;
1909
1910         spin_lock(&sdp->sd_reclaim_lock);
1911         if (list_empty(&sdp->sd_reclaim_list)) {
1912                 spin_unlock(&sdp->sd_reclaim_lock);
1913                 return;
1914         }
1915         gl = list_entry(sdp->sd_reclaim_list.next,
1916                         struct gfs2_glock, gl_reclaim);
1917         list_del_init(&gl->gl_reclaim);
1918         spin_unlock(&sdp->sd_reclaim_lock);
1919
1920         atomic_dec(&sdp->sd_reclaim_count);
1921         atomic_inc(&sdp->sd_reclaimed);
1922
1923         if (gfs2_glmutex_trylock(gl)) {
1924                 if (gl->gl_ops == &gfs2_inode_glops) {
1925                         struct gfs2_inode *ip = gl->gl_object;
1926                         if (ip && !atomic_read(&ip->i_count))
1927                                 gfs2_inode_destroy(ip, 1);
1928                 }
1929                 if (queue_empty(gl, &gl->gl_holders) &&
1930                     gl->gl_state != LM_ST_UNLOCKED &&
1931                     demote_ok(gl))
1932                         handle_callback(gl, LM_ST_UNLOCKED);
1933                 gfs2_glmutex_unlock(gl);
1934         }
1935
1936         gfs2_glock_put(gl);
1937 }
1938
1939 /**
1940  * examine_bucket - Call a function for glock in a hash bucket
1941  * @examiner: the function
1942  * @sdp: the filesystem
1943  * @bucket: the bucket
1944  *
1945  * Returns: 1 if the bucket has entries
1946  */
1947
1948 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1949                           struct gfs2_gl_hash_bucket *bucket)
1950 {
1951         struct glock_plug plug;
1952         struct list_head *tmp;
1953         struct gfs2_glock *gl;
1954         int entries;
1955
1956         /* Add "plug" to end of bucket list, work back up list from there */
1957         memset(&plug.gl_flags, 0, sizeof(unsigned long));
1958         set_bit(GLF_PLUG, &plug.gl_flags);
1959
1960         write_lock(&bucket->hb_lock);
1961         list_add(&plug.gl_list, &bucket->hb_list);
1962         write_unlock(&bucket->hb_lock);
1963
1964         for (;;) {
1965                 write_lock(&bucket->hb_lock);
1966
1967                 for (;;) {
1968                         tmp = plug.gl_list.next;
1969
1970                         if (tmp == &bucket->hb_list) {
1971                                 list_del(&plug.gl_list);
1972                                 entries = !list_empty(&bucket->hb_list);
1973                                 write_unlock(&bucket->hb_lock);
1974                                 return entries;
1975                         }
1976                         gl = list_entry(tmp, struct gfs2_glock, gl_list);
1977
1978                         /* Move plug up list */
1979                         list_move(&plug.gl_list, &gl->gl_list);
1980
1981                         if (test_bit(GLF_PLUG, &gl->gl_flags))
1982                                 continue;
1983
1984                         /* examiner() must glock_put() */
1985                         gfs2_glock_hold(gl);
1986
1987                         break;
1988                 }
1989
1990                 write_unlock(&bucket->hb_lock);
1991
1992                 examiner(gl);
1993         }
1994 }
1995
1996 /**
1997  * scan_glock - look at a glock and see if we can reclaim it
1998  * @gl: the glock to look at
1999  *
2000  */
2001
2002 static void scan_glock(struct gfs2_glock *gl)
2003 {
2004         if (gfs2_glmutex_trylock(gl)) {
2005                 if (gl->gl_ops == &gfs2_inode_glops) {
2006                         struct gfs2_inode *ip = gl->gl_object;
2007                         if (ip && !atomic_read(&ip->i_count))
2008                                 goto out_schedule;
2009                 }
2010                 if (queue_empty(gl, &gl->gl_holders) &&
2011                     gl->gl_state != LM_ST_UNLOCKED &&
2012                     demote_ok(gl))
2013                         goto out_schedule;
2014
2015                 gfs2_glmutex_unlock(gl);
2016         }
2017
2018         gfs2_glock_put(gl);
2019
2020         return;
2021
2022  out_schedule:
2023         gfs2_glmutex_unlock(gl);
2024         gfs2_glock_schedule_for_reclaim(gl);
2025         gfs2_glock_put(gl);
2026 }
2027
2028 /**
2029  * gfs2_scand_internal - Look for glocks and inodes to toss from memory
2030  * @sdp: the filesystem
2031  *
2032  */
2033
2034 void gfs2_scand_internal(struct gfs2_sbd *sdp)
2035 {
2036         unsigned int x;
2037
2038         for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
2039                 examine_bucket(scan_glock, sdp, &sdp->sd_gl_hash[x]);
2040                 cond_resched();
2041         }
2042 }
2043
2044 /**
2045  * clear_glock - look at a glock and see if we can free it from glock cache
2046  * @gl: the glock to look at
2047  *
2048  */
2049
2050 static void clear_glock(struct gfs2_glock *gl)
2051 {
2052         struct gfs2_sbd *sdp = gl->gl_sbd;
2053         int released;
2054
2055         spin_lock(&sdp->sd_reclaim_lock);
2056         if (!list_empty(&gl->gl_reclaim)) {
2057                 list_del_init(&gl->gl_reclaim);
2058                 atomic_dec(&sdp->sd_reclaim_count);
2059                 spin_unlock(&sdp->sd_reclaim_lock);
2060                 released = gfs2_glock_put(gl);
2061                 gfs2_assert(sdp, !released);
2062         } else {
2063                 spin_unlock(&sdp->sd_reclaim_lock);
2064         }
2065
2066         if (gfs2_glmutex_trylock(gl)) {
2067                 if (gl->gl_ops == &gfs2_inode_glops) {
2068                         struct gfs2_inode *ip = gl->gl_object;
2069                         if (ip && !atomic_read(&ip->i_count))
2070                                 gfs2_inode_destroy(ip, 1);
2071                 }
2072                 if (queue_empty(gl, &gl->gl_holders) &&
2073                     gl->gl_state != LM_ST_UNLOCKED)
2074                         handle_callback(gl, LM_ST_UNLOCKED);
2075
2076                 gfs2_glmutex_unlock(gl);
2077         }
2078
2079         gfs2_glock_put(gl);
2080 }
2081
2082 /**
2083  * gfs2_gl_hash_clear - Empty out the glock hash table
2084  * @sdp: the filesystem
2085  * @wait: wait until it's all gone
2086  *
2087  * Called when unmounting the filesystem, or when inter-node lock manager
2088  * requests DROPLOCKS because it is running out of capacity.
2089  */
2090
2091 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp, int wait)
2092 {
2093         unsigned long t;
2094         unsigned int x;
2095         int cont;
2096
2097         t = jiffies;
2098
2099         for (;;) {
2100                 cont = 0;
2101
2102                 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
2103                         if (examine_bucket(clear_glock, sdp,
2104                                            &sdp->sd_gl_hash[x]))
2105                                 cont = 1;
2106
2107                 if (!wait || !cont)
2108                         break;
2109
2110                 if (time_after_eq(jiffies,
2111                                   t + gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
2112                         fs_warn(sdp, "Unmount seems to be stalled. "
2113                                      "Dumping lock state...\n");
2114                         gfs2_dump_lockstate(sdp);
2115                         t = jiffies;
2116                 }
2117
2118                 /* invalidate_inodes() requires that the sb inodes list
2119                    not change, but an async completion callback for an
2120                    unlock can occur which does glock_put() which
2121                    can call iput() which will change the sb inodes list.
2122                    invalidate_inodes_mutex prevents glock_put()'s during
2123                    an invalidate_inodes() */
2124
2125                 mutex_lock(&sdp->sd_invalidate_inodes_mutex);
2126                 invalidate_inodes(sdp->sd_vfs);
2127                 mutex_unlock(&sdp->sd_invalidate_inodes_mutex);
2128                 msleep(10);
2129         }
2130 }
2131
2132 /*
2133  *  Diagnostic routines to help debug distributed deadlock
2134  */
2135
2136 /**
2137  * dump_holder - print information about a glock holder
2138  * @str: a string naming the type of holder
2139  * @gh: the glock holder
2140  *
2141  * Returns: 0 on success, -ENOBUFS when we run out of space
2142  */
2143
2144 static int dump_holder(char *str, struct gfs2_holder *gh)
2145 {
2146         unsigned int x;
2147         int error = -ENOBUFS;
2148
2149         printk(KERN_INFO "  %s\n", str);
2150         printk(KERN_INFO "    owner = %ld\n",
2151                    (gh->gh_owner) ? (long)gh->gh_owner->pid : -1);
2152         printk(KERN_INFO "    gh_state = %u\n", gh->gh_state);
2153         printk(KERN_INFO "    gh_flags =");
2154         for (x = 0; x < 32; x++)
2155                 if (gh->gh_flags & (1 << x))
2156                         printk(" %u", x);
2157         printk(" \n");
2158         printk(KERN_INFO "    error = %d\n", gh->gh_error);
2159         printk(KERN_INFO "    gh_iflags =");
2160         for (x = 0; x < 32; x++)
2161                 if (test_bit(x, &gh->gh_iflags))
2162                         printk(" %u", x);
2163         printk(" \n");
2164         print_symbol(KERN_INFO "    initialized at: %s\n", gh->gh_ip);
2165
2166         error = 0;
2167
2168         return error;
2169 }
2170
2171 /**
2172  * dump_inode - print information about an inode
2173  * @ip: the inode
2174  *
2175  * Returns: 0 on success, -ENOBUFS when we run out of space
2176  */
2177
2178 static int dump_inode(struct gfs2_inode *ip)
2179 {
2180         unsigned int x;
2181         int error = -ENOBUFS;
2182
2183         printk(KERN_INFO "  Inode:\n");
2184         printk(KERN_INFO "    num = %llu %llu\n",
2185                     ip->i_num.no_formal_ino, ip->i_num.no_addr);
2186         printk(KERN_INFO "    type = %u\n", IF2DT(ip->i_di.di_mode));
2187         printk(KERN_INFO "    i_count = %d\n", atomic_read(&ip->i_count));
2188         printk(KERN_INFO "    i_flags =");
2189         for (x = 0; x < 32; x++)
2190                 if (test_bit(x, &ip->i_flags))
2191                         printk(" %u", x);
2192         printk(" \n");
2193         printk(KERN_INFO "    vnode = %s\n", (ip->i_vnode) ? "yes" : "no");
2194
2195         error = 0;
2196
2197         return error;
2198 }
2199
2200 /**
2201  * dump_glock - print information about a glock
2202  * @gl: the glock
2203  * @count: where we are in the buffer
2204  *
2205  * Returns: 0 on success, -ENOBUFS when we run out of space
2206  */
2207
2208 static int dump_glock(struct gfs2_glock *gl)
2209 {
2210         struct gfs2_holder *gh;
2211         unsigned int x;
2212         int error = -ENOBUFS;
2213
2214         spin_lock(&gl->gl_spin);
2215
2216         printk(KERN_INFO "Glock (%u, %llu)\n",
2217                     gl->gl_name.ln_type,
2218                     gl->gl_name.ln_number);
2219         printk(KERN_INFO "  gl_flags =");
2220         for (x = 0; x < 32; x++)
2221                 if (test_bit(x, &gl->gl_flags))
2222                         printk(" %u", x);
2223         printk(" \n");
2224         printk(KERN_INFO "  gl_ref = %d\n", atomic_read(&gl->gl_ref.refcount));
2225         printk(KERN_INFO "  gl_state = %u\n", gl->gl_state);
2226         printk(KERN_INFO "  req_gh = %s\n", (gl->gl_req_gh) ? "yes" : "no");
2227         printk(KERN_INFO "  req_bh = %s\n", (gl->gl_req_bh) ? "yes" : "no");
2228         printk(KERN_INFO "  lvb_count = %d\n", atomic_read(&gl->gl_lvb_count));
2229         printk(KERN_INFO "  object = %s\n", (gl->gl_object) ? "yes" : "no");
2230         printk(KERN_INFO "  le = %s\n",
2231                    (list_empty(&gl->gl_le.le_list)) ? "no" : "yes");
2232         printk(KERN_INFO "  reclaim = %s\n",
2233                     (list_empty(&gl->gl_reclaim)) ? "no" : "yes");
2234         if (gl->gl_aspace)
2235                 printk(KERN_INFO "  aspace = %lu\n",
2236                             gl->gl_aspace->i_mapping->nrpages);
2237         else
2238                 printk(KERN_INFO "  aspace = no\n");
2239         printk(KERN_INFO "  ail = %d\n", atomic_read(&gl->gl_ail_count));
2240         if (gl->gl_req_gh) {
2241                 error = dump_holder("Request", gl->gl_req_gh);
2242                 if (error)
2243                         goto out;
2244         }
2245         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
2246                 error = dump_holder("Holder", gh);
2247                 if (error)
2248                         goto out;
2249         }
2250         list_for_each_entry(gh, &gl->gl_waiters1, gh_list) {
2251                 error = dump_holder("Waiter1", gh);
2252                 if (error)
2253                         goto out;
2254         }
2255         list_for_each_entry(gh, &gl->gl_waiters2, gh_list) {
2256                 error = dump_holder("Waiter2", gh);
2257                 if (error)
2258                         goto out;
2259         }
2260         list_for_each_entry(gh, &gl->gl_waiters3, gh_list) {
2261                 error = dump_holder("Waiter3", gh);
2262                 if (error)
2263                         goto out;
2264         }
2265         if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object) {
2266                 if (!test_bit(GLF_LOCK, &gl->gl_flags) &&
2267                     list_empty(&gl->gl_holders)) {
2268                         error = dump_inode(gl->gl_object);
2269                         if (error)
2270                                 goto out;
2271                 } else {
2272                         error = -ENOBUFS;
2273                         printk(KERN_INFO "  Inode: busy\n");
2274                 }
2275         }
2276
2277         error = 0;
2278
2279  out:
2280         spin_unlock(&gl->gl_spin);
2281
2282         return error;
2283 }
2284
2285 /**
2286  * gfs2_dump_lockstate - print out the current lockstate
2287  * @sdp: the filesystem
2288  * @ub: the buffer to copy the information into
2289  *
2290  * If @ub is NULL, dump the lockstate to the console.
2291  *
2292  */
2293
2294 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
2295 {
2296         struct gfs2_gl_hash_bucket *bucket;
2297         struct gfs2_glock *gl;
2298         unsigned int x;
2299         int error = 0;
2300
2301         for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
2302                 bucket = &sdp->sd_gl_hash[x];
2303
2304                 read_lock(&bucket->hb_lock);
2305
2306                 list_for_each_entry(gl, &bucket->hb_list, gl_list) {
2307                         if (test_bit(GLF_PLUG, &gl->gl_flags))
2308                                 continue;
2309
2310                         error = dump_glock(gl);
2311                         if (error)
2312                                 break;
2313                 }
2314
2315                 read_unlock(&bucket->hb_lock);
2316
2317                 if (error)
2318                         break;
2319         }
2320
2321
2322         return error;
2323 }
2324