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 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/uaccess.h>
  22
  23 #include "gfs2.h"
  24 #include "lm_interface.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 /*  Must be kept in sync with the beginning of struct gfs2_glock  */
  37 struct glock_plug {
  38         struct list_head gl_list;
  39         unsigned long gl_flags;
  40 };
  41
  42 struct greedy {
  43         struct gfs2_holder gr_gh;
  44         struct work_struct gr_work;
  45 };
  46
  47 typedef void (*glock_examiner) (struct gfs2_glock * gl);
  48
  49 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp);
  50 static int dump_glock(struct gfs2_glock *gl);
  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_gl_hash_bucket *bucket = gl->gl_bucket;
 154         int rv = 0;
 155
 156         write_lock(&bucket->hb_lock);
 157         if (kref_put(&gl->gl_ref, kill_glock)) {
 158                 list_del_init(&gl->gl_list);
 159                 write_unlock(&bucket->hb_lock);
 160                 BUG_ON(spin_is_locked(&gl->gl_spin));
 161                 glock_free(gl);
 162                 rv = 1;
 163                 goto out;
 164         }
 165         write_unlock(&bucket->hb_lock);
 166 out:
 167         return rv;
 168 }
 169
 170 /**
 171  * queue_empty - check to see if a glock's queue is empty
 172  * @gl: the glock
 173  * @head: the head of the queue to check
 174  *
 175  * This function protects the list in the event that a process already
 176  * has a holder on the list and is adding a second holder for itself.
 177  * The glmutex lock is what generally prevents processes from working
 178  * on the same glock at once, but the special case of adding a second
 179  * holder for yourself ("recursive" locking) doesn't involve locking
 180  * glmutex, making the spin lock necessary.
 181  *
 182  * Returns: 1 if the queue is empty
 183  */
 184
 185 static inline int queue_empty(struct gfs2_glock *gl, struct list_head *head)
 186 {
 187         int empty;
 188         spin_lock(&gl->gl_spin);
 189         empty = list_empty(head);
 190         spin_unlock(&gl->gl_spin);
 191         return empty;
 192 }
 193
 194 /**
 195  * search_bucket() - Find struct gfs2_glock by lock number
 196  * @bucket: the bucket to search
 197  * @name: The lock name
 198  *
 199  * Returns: NULL, or the struct gfs2_glock with the requested number
 200  */
 201
 202 static struct gfs2_glock *search_bucket(struct gfs2_gl_hash_bucket *bucket,
 203                                         struct lm_lockname *name)
 204 {
 205         struct gfs2_glock *gl;
 206
 207         list_for_each_entry(gl, &bucket->hb_list, gl_list) {
 208                 if (test_bit(GLF_PLUG, &gl->gl_flags))
 209                         continue;
 210                 if (!lm_name_equal(&gl->gl_name, name))
 211                         continue;
 212
 213                 kref_get(&gl->gl_ref);
 214
 215                 return gl;
 216         }
 217
 218         return NULL;
 219 }
 220
 221 /**
 222  * gfs2_glock_find() - Find glock by lock number
 223  * @sdp: The GFS2 superblock
 224  * @name: The lock name
 225  *
 226  * Returns: NULL, or the struct gfs2_glock with the requested number
 227  */
 228
 229 static struct gfs2_glock *gfs2_glock_find(struct gfs2_sbd *sdp,
 230                                           struct lm_lockname *name)
 231 {
 232         struct gfs2_gl_hash_bucket *bucket = &sdp->sd_gl_hash[gl_hash(name)];
 233         struct gfs2_glock *gl;
 234
 235         read_lock(&bucket->hb_lock);
 236         gl = search_bucket(bucket, name);
 237         read_unlock(&bucket->hb_lock);
 238
 239         return gl;
 240 }
 241
 242 /**
 243  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
 244  * @sdp: The GFS2 superblock
 245  * @number: the lock number
 246  * @glops: The glock_operations to use
 247  * @create: If 0, don't create the glock if it doesn't exist
 248  * @glp: the glock is returned here
 249  *
 250  * This does not lock a glock, just finds/creates structures for one.
 251  *
 252  * Returns: errno
 253  */
 254
 255 int gfs2_glock_get(struct gfs2_sbd *sdp, uint64_t number,
 256                    struct gfs2_glock_operations *glops, int create,
 257                    struct gfs2_glock **glp)
 258 {
 259         struct lm_lockname name;
 260         struct gfs2_glock *gl, *tmp;
 261         struct gfs2_gl_hash_bucket *bucket;
 262         int error;
 263
 264         name.ln_number = number;
 265         name.ln_type = glops->go_type;
 266         bucket = &sdp->sd_gl_hash[gl_hash(&name)];
 267
 268         read_lock(&bucket->hb_lock);
 269         gl = search_bucket(bucket, &name);
 270         read_unlock(&bucket->hb_lock);
 271
 272         if (gl || !create) {
 273                 *glp = gl;
 274                 return 0;
 275         }
 276
 277         gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
 278         if (!gl)
 279                 return -ENOMEM;
 280
 281         memset(gl, 0, sizeof(struct gfs2_glock));
 282
 283         INIT_LIST_HEAD(&gl->gl_list);
 284         gl->gl_name = name;
 285         kref_init(&gl->gl_ref);
 286
 287         spin_lock_init(&gl->gl_spin);
 288
 289         gl->gl_state = LM_ST_UNLOCKED;
 290         gl->gl_owner = NULL;
 291         gl->gl_ip = 0;
 292         INIT_LIST_HEAD(&gl->gl_holders);
 293         INIT_LIST_HEAD(&gl->gl_waiters1);
 294         INIT_LIST_HEAD(&gl->gl_waiters2);
 295         INIT_LIST_HEAD(&gl->gl_waiters3);
 296
 297         gl->gl_ops = glops;
 298
 299         gl->gl_bucket = bucket;
 300         INIT_LIST_HEAD(&gl->gl_reclaim);
 301
 302         gl->gl_sbd = sdp;
 303
 304         lops_init_le(&gl->gl_le, &gfs2_glock_lops);
 305         INIT_LIST_HEAD(&gl->gl_ail_list);
 306
 307         /* If this glock protects actual on-disk data or metadata blocks,
 308            create a VFS inode to manage the pages/buffers holding them. */
 309         if (glops == &gfs2_inode_glops ||
 310             glops == &gfs2_rgrp_glops) {
 311                 gl->gl_aspace = gfs2_aspace_get(sdp);
 312                 if (!gl->gl_aspace) {
 313                         error = -ENOMEM;
 314                         goto fail;
 315                 }
 316         }
 317
 318         error = gfs2_lm_get_lock(sdp, &name, &gl->gl_lock);
 319         if (error)
 320                 goto fail_aspace;
 321
 322         write_lock(&bucket->hb_lock);
 323         tmp = search_bucket(bucket, &name);
 324         if (tmp) {
 325                 write_unlock(&bucket->hb_lock);
 326                 glock_free(gl);
 327                 gl = tmp;
 328         } else {
 329                 list_add_tail(&gl->gl_list, &bucket->hb_list);
 330                 write_unlock(&bucket->hb_lock);
 331         }
 332
 333         *glp = gl;
 334
 335         return 0;
 336
 337  fail_aspace:
 338         if (gl->gl_aspace)
 339                 gfs2_aspace_put(gl->gl_aspace);
 340
 341  fail:
 342         kmem_cache_free(gfs2_glock_cachep, gl);
 343
 344         return error;
 345 }
 346
 347 /**
 348  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
 349  * @gl: the glock
 350  * @state: the state we're requesting
 351  * @flags: the modifier flags
 352  * @gh: the holder structure
 353  *
 354  */
 355
 356 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
 357                       struct gfs2_holder *gh)
 358 {
 359         INIT_LIST_HEAD(&gh->gh_list);
 360         gh->gh_gl = gl;
 361         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 362         gh->gh_owner = current;
 363         gh->gh_state = state;
 364         gh->gh_flags = flags;
 365         gh->gh_error = 0;
 366         gh->gh_iflags = 0;
 367         init_completion(&gh->gh_wait);
 368
 369         if (gh->gh_state == LM_ST_EXCLUSIVE)
 370                 gh->gh_flags |= GL_LOCAL_EXCL;
 371
 372         gfs2_glock_hold(gl);
 373 }
 374
 375 /**
 376  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
 377  * @state: the state we're requesting
 378  * @flags: the modifier flags
 379  * @gh: the holder structure
 380  *
 381  * Don't mess with the glock.
 382  *
 383  */
 384
 385 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
 386 {
 387         gh->gh_state = state;
 388         gh->gh_flags = flags;
 389         if (gh->gh_state == LM_ST_EXCLUSIVE)
 390                 gh->gh_flags |= GL_LOCAL_EXCL;
 391
 392         gh->gh_iflags &= 1 << HIF_ALLOCED;
 393         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 394 }
 395
 396 /**
 397  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
 398  * @gh: the holder structure
 399  *
 400  */
 401
 402 void gfs2_holder_uninit(struct gfs2_holder *gh)
 403 {
 404         gfs2_glock_put(gh->gh_gl);
 405         gh->gh_gl = NULL;
 406         gh->gh_ip = 0;
 407 }
 408
 409 /**
 410  * gfs2_holder_get - get a struct gfs2_holder structure
 411  * @gl: the glock
 412  * @state: the state we're requesting
 413  * @flags: the modifier flags
 414  * @gfp_flags:
 415  *
 416  * Figure out how big an impact this function has.  Either:
 417  * 1) Replace it with a cache of structures hanging off the struct gfs2_sbd
 418  * 2) Leave it like it is
 419  *
 420  * Returns: the holder structure, NULL on ENOMEM
 421  */
 422
 423 static struct gfs2_holder *gfs2_holder_get(struct gfs2_glock *gl,
 424                                            unsigned int state,
 425                                            int flags, gfp_t gfp_flags)
 426 {
 427         struct gfs2_holder *gh;
 428
 429         gh = kmalloc(sizeof(struct gfs2_holder), gfp_flags);
 430         if (!gh)
 431                 return NULL;
 432
 433         gfs2_holder_init(gl, state, flags, gh);
 434         set_bit(HIF_ALLOCED, &gh->gh_iflags);
 435         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 436         return gh;
 437 }
 438
 439 /**
 440  * gfs2_holder_put - get rid of a struct gfs2_holder structure
 441  * @gh: the holder structure
 442  *
 443  */
 444
 445 static void gfs2_holder_put(struct gfs2_holder *gh)
 446 {
 447         gfs2_holder_uninit(gh);
 448         kfree(gh);
 449 }
 450
 451 /**
 452  * rq_mutex - process a mutex request in the queue
 453  * @gh: the glock holder
 454  *
 455  * Returns: 1 if the queue is blocked
 456  */
 457
 458 static int rq_mutex(struct gfs2_holder *gh)
 459 {
 460         struct gfs2_glock *gl = gh->gh_gl;
 461
 462         list_del_init(&gh->gh_list);
 463         /*  gh->gh_error never examined.  */
 464         set_bit(GLF_LOCK, &gl->gl_flags);
 465         complete(&gh->gh_wait);
 466
 467         return 1;
 468 }
 469
 470 /**
 471  * rq_promote - process a promote request in the queue
 472  * @gh: the glock holder
 473  *
 474  * Acquire a new inter-node lock, or change a lock state to more restrictive.
 475  *
 476  * Returns: 1 if the queue is blocked
 477  */
 478
 479 static int rq_promote(struct gfs2_holder *gh)
 480 {
 481         struct gfs2_glock *gl = gh->gh_gl;
 482         struct gfs2_sbd *sdp = gl->gl_sbd;
 483         struct gfs2_glock_operations *glops = gl->gl_ops;
 484
 485         if (!relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
 486                 if (list_empty(&gl->gl_holders)) {
 487                         gl->gl_req_gh = gh;
 488                         set_bit(GLF_LOCK, &gl->gl_flags);
 489                         spin_unlock(&gl->gl_spin);
 490
 491                         if (atomic_read(&sdp->sd_reclaim_count) >
 492                             gfs2_tune_get(sdp, gt_reclaim_limit) &&
 493                             !(gh->gh_flags & LM_FLAG_PRIORITY)) {
 494                                 gfs2_reclaim_glock(sdp);
 495                                 gfs2_reclaim_glock(sdp);
 496                         }
 497
 498                         glops->go_xmote_th(gl, gh->gh_state,
 499                                            gh->gh_flags);
 500
 501                         spin_lock(&gl->gl_spin);
 502                 }
 503                 return 1;
 504         }
 505
 506         if (list_empty(&gl->gl_holders)) {
 507                 set_bit(HIF_FIRST, &gh->gh_iflags);
 508                 set_bit(GLF_LOCK, &gl->gl_flags);
 509         } else {
 510                 struct gfs2_holder *next_gh;
 511                 if (gh->gh_flags & GL_LOCAL_EXCL)
 512                         return 1;
 513                 next_gh = list_entry(gl->gl_holders.next, struct gfs2_holder,
 514                                      gh_list);
 515                 if (next_gh->gh_flags & GL_LOCAL_EXCL)
 516                          return 1;
 517         }
 518
 519         list_move_tail(&gh->gh_list, &gl->gl_holders);
 520         gh->gh_error = 0;
 521         set_bit(HIF_HOLDER, &gh->gh_iflags);
 522
 523         complete(&gh->gh_wait);
 524
 525         return 0;
 526 }
 527
 528 /**
 529  * rq_demote - process a demote request in the queue
 530  * @gh: the glock holder
 531  *
 532  * Returns: 1 if the queue is blocked
 533  */
 534
 535 static int rq_demote(struct gfs2_holder *gh)
 536 {
 537         struct gfs2_glock *gl = gh->gh_gl;
 538         struct gfs2_glock_operations *glops = gl->gl_ops;
 539
 540         if (!list_empty(&gl->gl_holders))
 541                 return 1;
 542
 543         if (gl->gl_state == gh->gh_state || gl->gl_state == LM_ST_UNLOCKED) {
 544                 list_del_init(&gh->gh_list);
 545                 gh->gh_error = 0;
 546                 spin_unlock(&gl->gl_spin);
 547                 if (test_bit(HIF_DEALLOC, &gh->gh_iflags))
 548                         gfs2_holder_put(gh);
 549                 else
 550                         complete(&gh->gh_wait);
 551                 spin_lock(&gl->gl_spin);
 552         } else {
 553                 gl->gl_req_gh = gh;
 554                 set_bit(GLF_LOCK, &gl->gl_flags);
 555                 spin_unlock(&gl->gl_spin);
 556
 557                 if (gh->gh_state == LM_ST_UNLOCKED ||
 558                     gl->gl_state != LM_ST_EXCLUSIVE)
 559                         glops->go_drop_th(gl);
 560                 else
 561                         glops->go_xmote_th(gl, gh->gh_state, gh->gh_flags);
 562
 563                 spin_lock(&gl->gl_spin);
 564         }
 565
 566         return 0;
 567 }
 568
 569 /**
 570  * rq_greedy - process a queued request to drop greedy status
 571  * @gh: the glock holder
 572  *
 573  * Returns: 1 if the queue is blocked
 574  */
 575
 576 static int rq_greedy(struct gfs2_holder *gh)
 577 {
 578         struct gfs2_glock *gl = gh->gh_gl;
 579
 580         list_del_init(&gh->gh_list);
 581         /*  gh->gh_error never examined.  */
 582         clear_bit(GLF_GREEDY, &gl->gl_flags);
 583         spin_unlock(&gl->gl_spin);
 584
 585         gfs2_holder_uninit(gh);
 586         kfree(container_of(gh, struct greedy, gr_gh));
 587
 588         spin_lock(&gl->gl_spin);
 589
 590         return 0;
 591 }
 592
 593 /**
 594  * run_queue - process holder structures on a glock
 595  * @gl: the glock
 596  *
 597  */
 598 static void run_queue(struct gfs2_glock *gl)
 599 {
 600         struct gfs2_holder *gh;
 601         int blocked = 1;
 602
 603         for (;;) {
 604                 if (test_bit(GLF_LOCK, &gl->gl_flags))
 605                         break;
 606
 607                 if (!list_empty(&gl->gl_waiters1)) {
 608                         gh = list_entry(gl->gl_waiters1.next,
 609                                         struct gfs2_holder, gh_list);
 610
 611                         if (test_bit(HIF_MUTEX, &gh->gh_iflags))
 612                                 blocked = rq_mutex(gh);
 613                         else
 614                                 gfs2_assert_warn(gl->gl_sbd, 0);
 615
 616                 } else if (!list_empty(&gl->gl_waiters2) &&
 617                            !test_bit(GLF_SKIP_WAITERS2, &gl->gl_flags)) {
 618                         gh = list_entry(gl->gl_waiters2.next,
 619                                         struct gfs2_holder, gh_list);
 620
 621                         if (test_bit(HIF_DEMOTE, &gh->gh_iflags))
 622                                 blocked = rq_demote(gh);
 623                         else if (test_bit(HIF_GREEDY, &gh->gh_iflags))
 624                                 blocked = rq_greedy(gh);
 625                         else
 626                                 gfs2_assert_warn(gl->gl_sbd, 0);
 627
 628                 } else if (!list_empty(&gl->gl_waiters3)) {
 629                         gh = list_entry(gl->gl_waiters3.next,
 630                                         struct gfs2_holder, gh_list);
 631
 632                         if (test_bit(HIF_PROMOTE, &gh->gh_iflags))
 633                                 blocked = rq_promote(gh);
 634                         else
 635                                 gfs2_assert_warn(gl->gl_sbd, 0);
 636
 637                 } else
 638                         break;
 639
 640                 if (blocked)
 641                         break;
 642         }
 643 }
 644
 645 /**
 646  * gfs2_glmutex_lock - acquire a local lock on a glock
 647  * @gl: the glock
 648  *
 649  * Gives caller exclusive access to manipulate a glock structure.
 650  */
 651
 652 static void gfs2_glmutex_lock(struct gfs2_glock *gl)
 653 {
 654         struct gfs2_holder gh;
 655
 656         gfs2_holder_init(gl, 0, 0, &gh);
 657         set_bit(HIF_MUTEX, &gh.gh_iflags);
 658
 659         spin_lock(&gl->gl_spin);
 660         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
 661                 list_add_tail(&gh.gh_list, &gl->gl_waiters1);
 662         else {
 663                 gl->gl_owner = current;
 664                 gl->gl_ip = (unsigned long)__builtin_return_address(0);
 665                 complete(&gh.gh_wait);
 666         }
 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         else {
 688                 gl->gl_owner = current;
 689                 gl->gl_ip = (unsigned long)__builtin_return_address(0);
 690         }
 691         spin_unlock(&gl->gl_spin);
 692
 693         return acquired;
 694 }
 695
 696 /**
 697  * gfs2_glmutex_unlock - release a local lock on a glock
 698  * @gl: the glock
 699  *
 700  */
 701
 702 static void gfs2_glmutex_unlock(struct gfs2_glock *gl)
 703 {
 704         spin_lock(&gl->gl_spin);
 705         clear_bit(GLF_LOCK, &gl->gl_flags);
 706         gl->gl_owner = NULL;
 707         gl->gl_ip = 0;
 708         run_queue(gl);
 709         BUG_ON(!spin_is_locked(&gl->gl_spin));
 710         spin_unlock(&gl->gl_spin);
 711 }
 712
 713 /**
 714  * handle_callback - add a demote request to a lock's queue
 715  * @gl: the glock
 716  * @state: the state the caller wants us to change to
 717  *
 718  * Note: This may fail sliently if we are out of memory.
 719  */
 720
 721 static void handle_callback(struct gfs2_glock *gl, unsigned int state)
 722 {
 723         struct gfs2_holder *gh, *new_gh = NULL;
 724
 725 restart:
 726         spin_lock(&gl->gl_spin);
 727
 728         list_for_each_entry(gh, &gl->gl_waiters2, gh_list) {
 729                 if (test_bit(HIF_DEMOTE, &gh->gh_iflags) &&
 730                     gl->gl_req_gh != gh) {
 731                         if (gh->gh_state != state)
 732                                 gh->gh_state = LM_ST_UNLOCKED;
 733                         goto out;
 734                 }
 735         }
 736
 737         if (new_gh) {
 738                 list_add_tail(&new_gh->gh_list, &gl->gl_waiters2);
 739                 new_gh = NULL;
 740         } else {
 741                 spin_unlock(&gl->gl_spin);
 742
 743                 new_gh = gfs2_holder_get(gl, state, LM_FLAG_TRY, GFP_KERNEL);
 744                 if (!new_gh)
 745                         return;
 746                 set_bit(HIF_DEMOTE, &new_gh->gh_iflags);
 747                 set_bit(HIF_DEALLOC, &new_gh->gh_iflags);
 748
 749                 goto restart;
 750         }
 751
 752 out:
 753         spin_unlock(&gl->gl_spin);
 754
 755         if (new_gh)
 756                 gfs2_holder_put(new_gh);
 757 }
 758
 759 void gfs2_glock_inode_squish(struct inode *inode)
 760 {
 761         struct gfs2_holder gh;
 762         struct gfs2_glock *gl = GFS2_I(inode)->i_gl;
 763         gfs2_holder_init(gl, LM_ST_UNLOCKED, 0, &gh);
 764         set_bit(HIF_DEMOTE, &gh.gh_iflags);
 765         spin_lock(&gl->gl_spin);
 766         gfs2_assert(inode->i_sb->s_fs_info, list_empty(&gl->gl_holders));
 767         list_add_tail(&gh.gh_list, &gl->gl_waiters2);
 768         run_queue(gl);
 769         spin_unlock(&gl->gl_spin);
 770         wait_for_completion(&gh.gh_wait);
 771         gfs2_holder_uninit(&gh);
 772 }
 773
 774 /**
 775  * state_change - record that the glock is now in a different state
 776  * @gl: the glock
 777  * @new_state the new state
 778  *
 779  */
 780
 781 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
 782 {
 783         int held1, held2;
 784
 785         held1 = (gl->gl_state != LM_ST_UNLOCKED);
 786         held2 = (new_state != LM_ST_UNLOCKED);
 787
 788         if (held1 != held2) {
 789                 if (held2)
 790                         gfs2_glock_hold(gl);
 791                 else
 792                         gfs2_glock_put(gl);
 793         }
 794
 795         gl->gl_state = new_state;
 796 }
 797
 798 /**
 799  * xmote_bh - Called after the lock module is done acquiring a lock
 800  * @gl: The glock in question
 801  * @ret: the int returned from the lock module
 802  *
 803  */
 804
 805 static void xmote_bh(struct gfs2_glock *gl, unsigned int ret)
 806 {
 807         struct gfs2_sbd *sdp = gl->gl_sbd;
 808         struct gfs2_glock_operations *glops = gl->gl_ops;
 809         struct gfs2_holder *gh = gl->gl_req_gh;
 810         int prev_state = gl->gl_state;
 811         int op_done = 1;
 812
 813         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
 814         gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
 815         gfs2_assert_warn(sdp, !(ret & LM_OUT_ASYNC));
 816
 817         state_change(gl, ret & LM_OUT_ST_MASK);
 818
 819         if (prev_state != LM_ST_UNLOCKED && !(ret & LM_OUT_CACHEABLE)) {
 820                 if (glops->go_inval)
 821                         glops->go_inval(gl, DIO_METADATA | DIO_DATA);
 822         } else if (gl->gl_state == LM_ST_DEFERRED) {
 823                 /* We might not want to do this here.
 824                    Look at moving to the inode glops. */
 825                 if (glops->go_inval)
 826                         glops->go_inval(gl, DIO_DATA);
 827         }
 828
 829         /*  Deal with each possible exit condition  */
 830
 831         if (!gh)
 832                 gl->gl_stamp = jiffies;
 833
 834         else if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
 835                 spin_lock(&gl->gl_spin);
 836                 list_del_init(&gh->gh_list);
 837                 gh->gh_error = -EIO;
 838                 spin_unlock(&gl->gl_spin);
 839
 840         } else if (test_bit(HIF_DEMOTE, &gh->gh_iflags)) {
 841                 spin_lock(&gl->gl_spin);
 842                 list_del_init(&gh->gh_list);
 843                 if (gl->gl_state == gh->gh_state ||
 844                     gl->gl_state == LM_ST_UNLOCKED)
 845                         gh->gh_error = 0;
 846                 else {
 847                         if (gfs2_assert_warn(sdp, gh->gh_flags &
 848                                         (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) == -1)
 849                                 fs_warn(sdp, "ret = 0x%.8X\n", ret);
 850                         gh->gh_error = GLR_TRYFAILED;
 851                 }
 852                 spin_unlock(&gl->gl_spin);
 853
 854                 if (ret & LM_OUT_CANCELED)
 855                         handle_callback(gl, LM_ST_UNLOCKED); /* Lame */
 856
 857         } else if (ret & LM_OUT_CANCELED) {
 858                 spin_lock(&gl->gl_spin);
 859                 list_del_init(&gh->gh_list);
 860                 gh->gh_error = GLR_CANCELED;
 861                 spin_unlock(&gl->gl_spin);
 862
 863         } else if (relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
 864                 spin_lock(&gl->gl_spin);
 865                 list_move_tail(&gh->gh_list, &gl->gl_holders);
 866                 gh->gh_error = 0;
 867                 set_bit(HIF_HOLDER, &gh->gh_iflags);
 868                 spin_unlock(&gl->gl_spin);
 869
 870                 set_bit(HIF_FIRST, &gh->gh_iflags);
 871
 872                 op_done = 0;
 873
 874         } else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
 875                 spin_lock(&gl->gl_spin);
 876                 list_del_init(&gh->gh_list);
 877                 gh->gh_error = GLR_TRYFAILED;
 878                 spin_unlock(&gl->gl_spin);
 879
 880         } else {
 881                 if (gfs2_assert_withdraw(sdp, 0) == -1)
 882                         fs_err(sdp, "ret = 0x%.8X\n", ret);
 883         }
 884
 885         if (glops->go_xmote_bh)
 886                 glops->go_xmote_bh(gl);
 887
 888         if (op_done) {
 889                 spin_lock(&gl->gl_spin);
 890                 gl->gl_req_gh = NULL;
 891                 gl->gl_req_bh = NULL;
 892                 clear_bit(GLF_LOCK, &gl->gl_flags);
 893                 run_queue(gl);
 894                 spin_unlock(&gl->gl_spin);
 895         }
 896
 897         gfs2_glock_put(gl);
 898
 899         if (gh) {
 900                 if (test_bit(HIF_DEALLOC, &gh->gh_iflags))
 901                         gfs2_holder_put(gh);
 902                 else
 903                         complete(&gh->gh_wait);
 904         }
 905 }
 906
 907 /**
 908  * gfs2_glock_xmote_th - Call into the lock module to acquire or change a glock
 909  * @gl: The glock in question
 910  * @state: the requested state
 911  * @flags: modifier flags to the lock call
 912  *
 913  */
 914
 915 void gfs2_glock_xmote_th(struct gfs2_glock *gl, unsigned int state, int flags)
 916 {
 917         struct gfs2_sbd *sdp = gl->gl_sbd;
 918         struct gfs2_glock_operations *glops = gl->gl_ops;
 919         int lck_flags = flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB |
 920                                  LM_FLAG_NOEXP | LM_FLAG_ANY |
 921                                  LM_FLAG_PRIORITY);
 922         unsigned int lck_ret;
 923
 924         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
 925         gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
 926         gfs2_assert_warn(sdp, state != LM_ST_UNLOCKED);
 927         gfs2_assert_warn(sdp, state != gl->gl_state);
 928
 929         if (gl->gl_state == LM_ST_EXCLUSIVE) {
 930                 if (glops->go_sync)
 931                         glops->go_sync(gl,
 932                                        DIO_METADATA | DIO_DATA | DIO_RELEASE);
 933         }
 934
 935         gfs2_glock_hold(gl);
 936         gl->gl_req_bh = xmote_bh;
 937
 938         lck_ret = gfs2_lm_lock(sdp, gl->gl_lock, gl->gl_state, state,
 939                                lck_flags);
 940
 941         if (gfs2_assert_withdraw(sdp, !(lck_ret & LM_OUT_ERROR)))
 942                 return;
 943
 944         if (lck_ret & LM_OUT_ASYNC)
 945                 gfs2_assert_warn(sdp, lck_ret == LM_OUT_ASYNC);
 946         else
 947                 xmote_bh(gl, lck_ret);
 948 }
 949
 950 /**
 951  * drop_bh - Called after a lock module unlock completes
 952  * @gl: the glock
 953  * @ret: the return status
 954  *
 955  * Doesn't wake up the process waiting on the struct gfs2_holder (if any)
 956  * Doesn't drop the reference on the glock the top half took out
 957  *
 958  */
 959
 960 static void drop_bh(struct gfs2_glock *gl, unsigned int ret)
 961 {
 962         struct gfs2_sbd *sdp = gl->gl_sbd;
 963         struct gfs2_glock_operations *glops = gl->gl_ops;
 964         struct gfs2_holder *gh = gl->gl_req_gh;
 965
 966         clear_bit(GLF_PREFETCH, &gl->gl_flags);
 967
 968         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
 969         gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
 970         gfs2_assert_warn(sdp, !ret);
 971
 972         state_change(gl, LM_ST_UNLOCKED);
 973
 974         if (glops->go_inval)
 975                 glops->go_inval(gl, DIO_METADATA | DIO_DATA);
 976
 977         if (gh) {
 978                 spin_lock(&gl->gl_spin);
 979                 list_del_init(&gh->gh_list);
 980                 gh->gh_error = 0;
 981                 spin_unlock(&gl->gl_spin);
 982         }
 983
 984         if (glops->go_drop_bh)
 985                 glops->go_drop_bh(gl);
 986
 987         spin_lock(&gl->gl_spin);
 988         gl->gl_req_gh = NULL;
 989         gl->gl_req_bh = NULL;
 990         clear_bit(GLF_LOCK, &gl->gl_flags);
 991         run_queue(gl);
 992         spin_unlock(&gl->gl_spin);
 993
 994         gfs2_glock_put(gl);
 995
 996         if (gh) {
 997                 if (test_bit(HIF_DEALLOC, &gh->gh_iflags))
 998                         gfs2_holder_put(gh);
 999                 else
1000                         complete(&gh->gh_wait);
1001         }
1002 }
1003
1004 /**
1005  * gfs2_glock_drop_th - call into the lock module to unlock a lock
1006  * @gl: the glock
1007  *
1008  */
1009
1010 void gfs2_glock_drop_th(struct gfs2_glock *gl)
1011 {
1012         struct gfs2_sbd *sdp = gl->gl_sbd;
1013         struct gfs2_glock_operations *glops = gl->gl_ops;
1014         unsigned int ret;
1015
1016         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1017         gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
1018         gfs2_assert_warn(sdp, gl->gl_state != LM_ST_UNLOCKED);
1019
1020         if (gl->gl_state == LM_ST_EXCLUSIVE) {
1021                 if (glops->go_sync)
1022                         glops->go_sync(gl,
1023                                        DIO_METADATA | DIO_DATA | DIO_RELEASE);
1024         }
1025
1026         gfs2_glock_hold(gl);
1027         gl->gl_req_bh = drop_bh;
1028
1029         ret = gfs2_lm_unlock(sdp, gl->gl_lock, gl->gl_state);
1030
1031         if (gfs2_assert_withdraw(sdp, !(ret & LM_OUT_ERROR)))
1032                 return;
1033
1034         if (!ret)
1035                 drop_bh(gl, ret);
1036         else
1037                 gfs2_assert_warn(sdp, ret == LM_OUT_ASYNC);
1038 }
1039
1040 /**
1041  * do_cancels - cancel requests for locks stuck waiting on an expire flag
1042  * @gh: the LM_FLAG_PRIORITY holder waiting to acquire the lock
1043  *
1044  * Don't cancel GL_NOCANCEL requests.
1045  */
1046
1047 static void do_cancels(struct gfs2_holder *gh)
1048 {
1049         struct gfs2_glock *gl = gh->gh_gl;
1050
1051         spin_lock(&gl->gl_spin);
1052
1053         while (gl->gl_req_gh != gh &&
1054                !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1055                !list_empty(&gh->gh_list)) {
1056                 if (gl->gl_req_bh &&
1057                     !(gl->gl_req_gh &&
1058                       (gl->gl_req_gh->gh_flags & GL_NOCANCEL))) {
1059                         spin_unlock(&gl->gl_spin);
1060                         gfs2_lm_cancel(gl->gl_sbd, gl->gl_lock);
1061                         msleep(100);
1062                         spin_lock(&gl->gl_spin);
1063                 } else {
1064                         spin_unlock(&gl->gl_spin);
1065                         msleep(100);
1066                         spin_lock(&gl->gl_spin);
1067                 }
1068         }
1069
1070         spin_unlock(&gl->gl_spin);
1071 }
1072
1073 /**
1074  * glock_wait_internal - wait on a glock acquisition
1075  * @gh: the glock holder
1076  *
1077  * Returns: 0 on success
1078  */
1079
1080 static int glock_wait_internal(struct gfs2_holder *gh)
1081 {
1082         struct gfs2_glock *gl = gh->gh_gl;
1083         struct gfs2_sbd *sdp = gl->gl_sbd;
1084         struct gfs2_glock_operations *glops = gl->gl_ops;
1085
1086         if (test_bit(HIF_ABORTED, &gh->gh_iflags))
1087                 return -EIO;
1088
1089         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1090                 spin_lock(&gl->gl_spin);
1091                 if (gl->gl_req_gh != gh &&
1092                     !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1093                     !list_empty(&gh->gh_list)) {
1094                         list_del_init(&gh->gh_list);
1095                         gh->gh_error = GLR_TRYFAILED;
1096                         run_queue(gl);
1097                         spin_unlock(&gl->gl_spin);
1098                         return gh->gh_error;
1099                 }
1100                 spin_unlock(&gl->gl_spin);
1101         }
1102
1103         if (gh->gh_flags & LM_FLAG_PRIORITY)
1104                 do_cancels(gh);
1105
1106         wait_for_completion(&gh->gh_wait);
1107
1108         if (gh->gh_error)
1109                 return gh->gh_error;
1110
1111         gfs2_assert_withdraw(sdp, test_bit(HIF_HOLDER, &gh->gh_iflags));
1112         gfs2_assert_withdraw(sdp, relaxed_state_ok(gl->gl_state,
1113                                                    gh->gh_state,
1114                                                    gh->gh_flags));
1115
1116         if (test_bit(HIF_FIRST, &gh->gh_iflags)) {
1117                 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1118
1119                 if (glops->go_lock) {
1120                         gh->gh_error = glops->go_lock(gh);
1121                         if (gh->gh_error) {
1122                                 spin_lock(&gl->gl_spin);
1123                                 list_del_init(&gh->gh_list);
1124                                 spin_unlock(&gl->gl_spin);
1125                         }
1126                 }
1127
1128                 spin_lock(&gl->gl_spin);
1129                 gl->gl_req_gh = NULL;
1130                 gl->gl_req_bh = NULL;
1131                 clear_bit(GLF_LOCK, &gl->gl_flags);
1132                 run_queue(gl);
1133                 spin_unlock(&gl->gl_spin);
1134         }
1135
1136         return gh->gh_error;
1137 }
1138
1139 static inline struct gfs2_holder *
1140 find_holder_by_owner(struct list_head *head, struct task_struct *owner)
1141 {
1142         struct gfs2_holder *gh;
1143
1144         list_for_each_entry(gh, head, gh_list) {
1145                 if (gh->gh_owner == owner)
1146                         return gh;
1147         }
1148
1149         return NULL;
1150 }
1151
1152 /**
1153  * add_to_queue - Add a holder to the wait queue (but look for recursion)
1154  * @gh: the holder structure to add
1155  *
1156  */
1157
1158 static void add_to_queue(struct gfs2_holder *gh)
1159 {
1160         struct gfs2_glock *gl = gh->gh_gl;
1161         struct gfs2_holder *existing;
1162
1163         BUG_ON(!gh->gh_owner);
1164
1165         existing = find_holder_by_owner(&gl->gl_holders, gh->gh_owner);
1166         if (existing) {
1167                 print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
1168                 printk(KERN_INFO "pid : %d\n", existing->gh_owner->pid);
1169                 printk(KERN_INFO "lock type : %d lock state : %d\n",
1170                                 existing->gh_gl->gl_name.ln_type, existing->gh_gl->gl_state);
1171                 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1172                 printk(KERN_INFO "pid : %d\n", gh->gh_owner->pid);
1173                 printk(KERN_INFO "lock type : %d lock state : %d\n",
1174                                 gl->gl_name.ln_type, gl->gl_state);
1175                 BUG();
1176         }
1177
1178         existing = find_holder_by_owner(&gl->gl_waiters3, gh->gh_owner);
1179         if (existing) {
1180                 print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
1181                 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1182                 BUG();
1183         }
1184
1185         if (gh->gh_flags & LM_FLAG_PRIORITY)
1186                 list_add(&gh->gh_list, &gl->gl_waiters3);
1187         else
1188                 list_add_tail(&gh->gh_list, &gl->gl_waiters3);
1189 }
1190
1191 /**
1192  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1193  * @gh: the holder structure
1194  *
1195  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1196  *
1197  * Returns: 0, GLR_TRYFAILED, or errno on failure
1198  */
1199
1200 int gfs2_glock_nq(struct gfs2_holder *gh)
1201 {
1202         struct gfs2_glock *gl = gh->gh_gl;
1203         struct gfs2_sbd *sdp = gl->gl_sbd;
1204         int error = 0;
1205
1206 restart:
1207         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1208                 set_bit(HIF_ABORTED, &gh->gh_iflags);
1209                 return -EIO;
1210         }
1211
1212         set_bit(HIF_PROMOTE, &gh->gh_iflags);
1213
1214         spin_lock(&gl->gl_spin);
1215         add_to_queue(gh);
1216         run_queue(gl);
1217         spin_unlock(&gl->gl_spin);
1218
1219         if (!(gh->gh_flags & GL_ASYNC)) {
1220                 error = glock_wait_internal(gh);
1221                 if (error == GLR_CANCELED) {
1222                         msleep(100);
1223                         goto restart;
1224                 }
1225         }
1226
1227         clear_bit(GLF_PREFETCH, &gl->gl_flags);
1228
1229         if (error == GLR_TRYFAILED && (gh->gh_flags & GL_DUMP))
1230                 dump_glock(gl);
1231
1232         return error;
1233 }
1234
1235 /**
1236  * gfs2_glock_poll - poll to see if an async request has been completed
1237  * @gh: the holder
1238  *
1239  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1240  */
1241
1242 int gfs2_glock_poll(struct gfs2_holder *gh)
1243 {
1244         struct gfs2_glock *gl = gh->gh_gl;
1245         int ready = 0;
1246
1247         spin_lock(&gl->gl_spin);
1248
1249         if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1250                 ready = 1;
1251         else if (list_empty(&gh->gh_list)) {
1252                 if (gh->gh_error == GLR_CANCELED) {
1253                         spin_unlock(&gl->gl_spin);
1254                         msleep(100);
1255                         if (gfs2_glock_nq(gh))
1256                                 return 1;
1257                         return 0;
1258                 } else
1259                         ready = 1;
1260         }
1261
1262         spin_unlock(&gl->gl_spin);
1263
1264         return ready;
1265 }
1266
1267 /**
1268  * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
1269  * @gh: the holder structure
1270  *
1271  * Returns: 0, GLR_TRYFAILED, or errno on failure
1272  */
1273
1274 int gfs2_glock_wait(struct gfs2_holder *gh)
1275 {
1276         int error;
1277
1278         error = glock_wait_internal(gh);
1279         if (error == GLR_CANCELED) {
1280                 msleep(100);
1281                 gh->gh_flags &= ~GL_ASYNC;
1282                 error = gfs2_glock_nq(gh);
1283         }
1284
1285         return error;
1286 }
1287
1288 /**
1289  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1290  * @gh: the glock holder
1291  *
1292  */
1293
1294 void gfs2_glock_dq(struct gfs2_holder *gh)
1295 {
1296         struct gfs2_glock *gl = gh->gh_gl;
1297         struct gfs2_glock_operations *glops = gl->gl_ops;
1298
1299         if (gh->gh_flags & GL_SYNC)
1300                 set_bit(GLF_SYNC, &gl->gl_flags);
1301
1302         if (gh->gh_flags & GL_NOCACHE)
1303                 handle_callback(gl, LM_ST_UNLOCKED);
1304
1305         gfs2_glmutex_lock(gl);
1306
1307         spin_lock(&gl->gl_spin);
1308         list_del_init(&gh->gh_list);
1309
1310         if (list_empty(&gl->gl_holders)) {
1311                 spin_unlock(&gl->gl_spin);
1312
1313                 if (glops->go_unlock)
1314                         glops->go_unlock(gh);
1315
1316                 if (test_bit(GLF_SYNC, &gl->gl_flags)) {
1317                         if (glops->go_sync)
1318                                 glops->go_sync(gl, DIO_METADATA | DIO_DATA);
1319                 }
1320
1321                 gl->gl_stamp = jiffies;
1322
1323                 spin_lock(&gl->gl_spin);
1324         }
1325
1326         clear_bit(GLF_LOCK, &gl->gl_flags);
1327         run_queue(gl);
1328         spin_unlock(&gl->gl_spin);
1329 }
1330
1331 /**
1332  * gfs2_glock_prefetch - Try to prefetch a glock
1333  * @gl: the glock
1334  * @state: the state to prefetch in
1335  * @flags: flags passed to go_xmote_th()
1336  *
1337  */
1338
1339 static void gfs2_glock_prefetch(struct gfs2_glock *gl, unsigned int state,
1340                                 int flags)
1341 {
1342         struct gfs2_glock_operations *glops = gl->gl_ops;
1343
1344         spin_lock(&gl->gl_spin);
1345
1346         if (test_bit(GLF_LOCK, &gl->gl_flags) ||
1347             !list_empty(&gl->gl_holders) ||
1348             !list_empty(&gl->gl_waiters1) ||
1349             !list_empty(&gl->gl_waiters2) ||
1350             !list_empty(&gl->gl_waiters3) ||
1351             relaxed_state_ok(gl->gl_state, state, flags)) {
1352                 spin_unlock(&gl->gl_spin);
1353                 return;
1354         }
1355
1356         set_bit(GLF_PREFETCH, &gl->gl_flags);
1357         set_bit(GLF_LOCK, &gl->gl_flags);
1358         spin_unlock(&gl->gl_spin);
1359
1360         glops->go_xmote_th(gl, state, flags);
1361 }
1362
1363 static void greedy_work(void *data)
1364 {
1365         struct greedy *gr = data;
1366         struct gfs2_holder *gh = &gr->gr_gh;
1367         struct gfs2_glock *gl = gh->gh_gl;
1368         struct gfs2_glock_operations *glops = gl->gl_ops;
1369
1370         clear_bit(GLF_SKIP_WAITERS2, &gl->gl_flags);
1371
1372         if (glops->go_greedy)
1373                 glops->go_greedy(gl);
1374
1375         spin_lock(&gl->gl_spin);
1376
1377         if (list_empty(&gl->gl_waiters2)) {
1378                 clear_bit(GLF_GREEDY, &gl->gl_flags);
1379                 spin_unlock(&gl->gl_spin);
1380                 gfs2_holder_uninit(gh);
1381                 kfree(gr);
1382         } else {
1383                 gfs2_glock_hold(gl);
1384                 list_add_tail(&gh->gh_list, &gl->gl_waiters2);
1385                 run_queue(gl);
1386                 spin_unlock(&gl->gl_spin);
1387                 gfs2_glock_put(gl);
1388         }
1389 }
1390
1391 /**
1392  * gfs2_glock_be_greedy -
1393  * @gl:
1394  * @time:
1395  *
1396  * Returns: 0 if go_greedy will be called, 1 otherwise
1397  */
1398
1399 int gfs2_glock_be_greedy(struct gfs2_glock *gl, unsigned int time)
1400 {
1401         struct greedy *gr;
1402         struct gfs2_holder *gh;
1403
1404         if (!time || gl->gl_sbd->sd_args.ar_localcaching ||
1405             test_and_set_bit(GLF_GREEDY, &gl->gl_flags))
1406                 return 1;
1407
1408         gr = kmalloc(sizeof(struct greedy), GFP_KERNEL);
1409         if (!gr) {
1410                 clear_bit(GLF_GREEDY, &gl->gl_flags);
1411                 return 1;
1412         }
1413         gh = &gr->gr_gh;
1414
1415         gfs2_holder_init(gl, 0, 0, gh);
1416         set_bit(HIF_GREEDY, &gh->gh_iflags);
1417         INIT_WORK(&gr->gr_work, greedy_work, gr);
1418
1419         set_bit(GLF_SKIP_WAITERS2, &gl->gl_flags);
1420         schedule_delayed_work(&gr->gr_work, time);
1421
1422         return 0;
1423 }
1424
1425 /**
1426  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1427  * @gh: the holder structure
1428  *
1429  */
1430
1431 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1432 {
1433         gfs2_glock_dq(gh);
1434         gfs2_holder_uninit(gh);
1435 }
1436
1437 /**
1438  * gfs2_glock_nq_num - acquire a glock based on lock number
1439  * @sdp: the filesystem
1440  * @number: the lock number
1441  * @glops: the glock operations for the type of glock
1442  * @state: the state to acquire the glock in
1443  * @flags: modifier flags for the aquisition
1444  * @gh: the struct gfs2_holder
1445  *
1446  * Returns: errno
1447  */
1448
1449 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, uint64_t number,
1450                       struct gfs2_glock_operations *glops, unsigned int state,
1451                       int flags, struct gfs2_holder *gh)
1452 {
1453         struct gfs2_glock *gl;
1454         int error;
1455
1456         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1457         if (!error) {
1458                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1459                 gfs2_glock_put(gl);
1460         }
1461
1462         return error;
1463 }
1464
1465 /**
1466  * glock_compare - Compare two struct gfs2_glock structures for sorting
1467  * @arg_a: the first structure
1468  * @arg_b: the second structure
1469  *
1470  */
1471
1472 static int glock_compare(const void *arg_a, const void *arg_b)
1473 {
1474         struct gfs2_holder *gh_a = *(struct gfs2_holder **)arg_a;
1475         struct gfs2_holder *gh_b = *(struct gfs2_holder **)arg_b;
1476         struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1477         struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1478         int ret = 0;
1479
1480         if (a->ln_number > b->ln_number)
1481                 ret = 1;
1482         else if (a->ln_number < b->ln_number)
1483                 ret = -1;
1484         else {
1485                 if (gh_a->gh_state == LM_ST_SHARED &&
1486                     gh_b->gh_state == LM_ST_EXCLUSIVE)
1487                         ret = 1;
1488                 else if (!(gh_a->gh_flags & GL_LOCAL_EXCL) &&
1489                          (gh_b->gh_flags & GL_LOCAL_EXCL))
1490                         ret = 1;
1491         }
1492
1493         return ret;
1494 }
1495
1496 /**
1497  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1498  * @num_gh: the number of structures
1499  * @ghs: an array of struct gfs2_holder structures
1500  *
1501  * Returns: 0 on success (all glocks acquired),
1502  *          errno on failure (no glocks acquired)
1503  */
1504
1505 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1506                      struct gfs2_holder **p)
1507 {
1508         unsigned int x;
1509         int error = 0;
1510
1511         for (x = 0; x < num_gh; x++)
1512                 p[x] = &ghs[x];
1513
1514         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1515
1516         for (x = 0; x < num_gh; x++) {
1517                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1518
1519                 error = gfs2_glock_nq(p[x]);
1520                 if (error) {
1521                         while (x--)
1522                                 gfs2_glock_dq(p[x]);
1523                         break;
1524                 }
1525         }
1526
1527         return error;
1528 }
1529
1530 /**
1531  * gfs2_glock_nq_m - acquire multiple glocks
1532  * @num_gh: the number of structures
1533  * @ghs: an array of struct gfs2_holder structures
1534  *
1535  * Figure out how big an impact this function has.  Either:
1536  * 1) Replace this code with code that calls gfs2_glock_prefetch()
1537  * 2) Forget async stuff and just call nq_m_sync()
1538  * 3) Leave it like it is
1539  *
1540  * Returns: 0 on success (all glocks acquired),
1541  *          errno on failure (no glocks acquired)
1542  */
1543
1544 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1545 {
1546         int *e;
1547         unsigned int x;
1548         int borked = 0, serious = 0;
1549         int error = 0;
1550
1551         if (!num_gh)
1552                 return 0;
1553
1554         if (num_gh == 1) {
1555                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1556                 return gfs2_glock_nq(ghs);
1557         }
1558
1559         e = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL);
1560         if (!e)
1561                 return -ENOMEM;
1562
1563         for (x = 0; x < num_gh; x++) {
1564                 ghs[x].gh_flags |= LM_FLAG_TRY | GL_ASYNC;
1565                 error = gfs2_glock_nq(&ghs[x]);
1566                 if (error) {
1567                         borked = 1;
1568                         serious = error;
1569                         num_gh = x;
1570                         break;
1571                 }
1572         }
1573
1574         for (x = 0; x < num_gh; x++) {
1575                 error = e[x] = glock_wait_internal(&ghs[x]);
1576                 if (error) {
1577                         borked = 1;
1578                         if (error != GLR_TRYFAILED && error != GLR_CANCELED)
1579                                 serious = error;
1580                 }
1581         }
1582
1583         if (!borked) {
1584                 kfree(e);
1585                 return 0;
1586         }
1587
1588         for (x = 0; x < num_gh; x++)
1589                 if (!e[x])
1590                         gfs2_glock_dq(&ghs[x]);
1591
1592         if (serious)
1593                 error = serious;
1594         else {
1595                 for (x = 0; x < num_gh; x++)
1596                         gfs2_holder_reinit(ghs[x].gh_state, ghs[x].gh_flags,
1597                                           &ghs[x]);
1598                 error = nq_m_sync(num_gh, ghs, (struct gfs2_holder **)e);
1599         }
1600
1601         kfree(e);
1602
1603         return error;
1604 }
1605
1606 /**
1607  * gfs2_glock_dq_m - release multiple glocks
1608  * @num_gh: the number of structures
1609  * @ghs: an array of struct gfs2_holder structures
1610  *
1611  */
1612
1613 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1614 {
1615         unsigned int x;
1616
1617         for (x = 0; x < num_gh; x++)
1618                 gfs2_glock_dq(&ghs[x]);
1619 }
1620
1621 /**
1622  * gfs2_glock_dq_uninit_m - release multiple glocks
1623  * @num_gh: the number of structures
1624  * @ghs: an array of struct gfs2_holder structures
1625  *
1626  */
1627
1628 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1629 {
1630         unsigned int x;
1631
1632         for (x = 0; x < num_gh; x++)
1633                 gfs2_glock_dq_uninit(&ghs[x]);
1634 }
1635
1636 /**
1637  * gfs2_glock_prefetch_num - prefetch a glock based on lock number
1638  * @sdp: the filesystem
1639  * @number: the lock number
1640  * @glops: the glock operations for the type of glock
1641  * @state: the state to acquire the glock in
1642  * @flags: modifier flags for the aquisition
1643  *
1644  * Returns: errno
1645  */
1646
1647 void gfs2_glock_prefetch_num(struct gfs2_sbd *sdp, uint64_t number,
1648                              struct gfs2_glock_operations *glops,
1649                              unsigned int state, int flags)
1650 {
1651         struct gfs2_glock *gl;
1652         int error;
1653
1654         if (atomic_read(&sdp->sd_reclaim_count) <
1655             gfs2_tune_get(sdp, gt_reclaim_limit)) {
1656                 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1657                 if (!error) {
1658                         gfs2_glock_prefetch(gl, state, flags);
1659                         gfs2_glock_put(gl);
1660                 }
1661         }
1662 }
1663
1664 /**
1665  * gfs2_lvb_hold - attach a LVB from a glock
1666  * @gl: The glock in question
1667  *
1668  */
1669
1670 int gfs2_lvb_hold(struct gfs2_glock *gl)
1671 {
1672         int error;
1673
1674         gfs2_glmutex_lock(gl);
1675
1676         if (!atomic_read(&gl->gl_lvb_count)) {
1677                 error = gfs2_lm_hold_lvb(gl->gl_sbd, gl->gl_lock, &gl->gl_lvb);
1678                 if (error) {
1679                         gfs2_glmutex_unlock(gl);
1680                         return error;
1681                 }
1682                 gfs2_glock_hold(gl);
1683         }
1684         atomic_inc(&gl->gl_lvb_count);
1685
1686         gfs2_glmutex_unlock(gl);
1687
1688         return 0;
1689 }
1690
1691 /**
1692  * gfs2_lvb_unhold - detach a LVB from a glock
1693  * @gl: The glock in question
1694  *
1695  */
1696
1697 void gfs2_lvb_unhold(struct gfs2_glock *gl)
1698 {
1699         gfs2_glock_hold(gl);
1700         gfs2_glmutex_lock(gl);
1701
1702         gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count) > 0);
1703         if (atomic_dec_and_test(&gl->gl_lvb_count)) {
1704                 gfs2_lm_unhold_lvb(gl->gl_sbd, gl->gl_lock, gl->gl_lvb);
1705                 gl->gl_lvb = NULL;
1706                 gfs2_glock_put(gl);
1707         }
1708
1709         gfs2_glmutex_unlock(gl);
1710         gfs2_glock_put(gl);
1711 }
1712
1713 #if 0
1714 void gfs2_lvb_sync(struct gfs2_glock *gl)
1715 {
1716         gfs2_glmutex_lock(gl);
1717
1718         gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count));
1719         if (!gfs2_assert_warn(gl->gl_sbd, gfs2_glock_is_held_excl(gl)))
1720                 gfs2_lm_sync_lvb(gl->gl_sbd, gl->gl_lock, gl->gl_lvb);
1721
1722         gfs2_glmutex_unlock(gl);
1723 }
1724 #endif  /*  0  */
1725
1726 static void blocking_cb(struct gfs2_sbd *sdp, struct lm_lockname *name,
1727                         unsigned int state)
1728 {
1729         struct gfs2_glock *gl;
1730
1731         gl = gfs2_glock_find(sdp, name);
1732         if (!gl)
1733                 return;
1734
1735         if (gl->gl_ops->go_callback)
1736                 gl->gl_ops->go_callback(gl, state);
1737         handle_callback(gl, state);
1738
1739         spin_lock(&gl->gl_spin);
1740         run_queue(gl);
1741         spin_unlock(&gl->gl_spin);
1742
1743         gfs2_glock_put(gl);
1744 }
1745
1746 /**
1747  * gfs2_glock_cb - Callback used by locking module
1748  * @fsdata: Pointer to the superblock
1749  * @type: Type of callback
1750  * @data: Type dependent data pointer
1751  *
1752  * Called by the locking module when it wants to tell us something.
1753  * Either we need to drop a lock, one of our ASYNC requests completed, or
1754  * a journal from another client needs to be recovered.
1755  */
1756
1757 void gfs2_glock_cb(lm_fsdata_t *fsdata, unsigned int type, void *data)
1758 {
1759         struct gfs2_sbd *sdp = (struct gfs2_sbd *)fsdata;
1760
1761         switch (type) {
1762         case LM_CB_NEED_E:
1763                 blocking_cb(sdp, data, LM_ST_UNLOCKED);
1764                 return;
1765
1766         case LM_CB_NEED_D:
1767                 blocking_cb(sdp, data, LM_ST_DEFERRED);
1768                 return;
1769
1770         case LM_CB_NEED_S:
1771                 blocking_cb(sdp, data, LM_ST_SHARED);
1772                 return;
1773
1774         case LM_CB_ASYNC: {
1775                 struct lm_async_cb *async = data;
1776                 struct gfs2_glock *gl;
1777
1778                 gl = gfs2_glock_find(sdp, &async->lc_name);
1779                 if (gfs2_assert_warn(sdp, gl))
1780                         return;
1781                 if (!gfs2_assert_warn(sdp, gl->gl_req_bh))
1782                         gl->gl_req_bh(gl, async->lc_ret);
1783                 gfs2_glock_put(gl);
1784                 return;
1785         }
1786
1787         case LM_CB_NEED_RECOVERY:
1788                 gfs2_jdesc_make_dirty(sdp, *(unsigned int *)data);
1789                 if (sdp->sd_recoverd_process)
1790                         wake_up_process(sdp->sd_recoverd_process);
1791                 return;
1792
1793         case LM_CB_DROPLOCKS:
1794                 gfs2_gl_hash_clear(sdp, NO_WAIT);
1795                 gfs2_quota_scan(sdp);
1796                 return;
1797
1798         default:
1799                 gfs2_assert_warn(sdp, 0);
1800                 return;
1801         }
1802 }
1803
1804 /**
1805  * gfs2_iopen_go_callback - Try to kick the inode/vnode associated with an
1806  *                          iopen glock from memory
1807  * @io_gl: the iopen glock
1808  * @state: the state into which the glock should be put
1809  *
1810  */
1811
1812 void gfs2_iopen_go_callback(struct gfs2_glock *io_gl, unsigned int state)
1813 {
1814
1815         if (state != LM_ST_UNLOCKED)
1816                 return;
1817         /* FIXME: remove this? */
1818 }
1819
1820 /**
1821  * demote_ok - Check to see if it's ok to unlock a glock
1822  * @gl: the glock
1823  *
1824  * Returns: 1 if it's ok
1825  */
1826
1827 static int demote_ok(struct gfs2_glock *gl)
1828 {
1829         struct gfs2_sbd *sdp = gl->gl_sbd;
1830         struct gfs2_glock_operations *glops = gl->gl_ops;
1831         int demote = 1;
1832
1833         if (test_bit(GLF_STICKY, &gl->gl_flags))
1834                 demote = 0;
1835         else if (test_bit(GLF_PREFETCH, &gl->gl_flags))
1836                 demote = time_after_eq(jiffies,
1837                                     gl->gl_stamp +
1838                                     gfs2_tune_get(sdp, gt_prefetch_secs) * HZ);
1839         else if (glops->go_demote_ok)
1840                 demote = glops->go_demote_ok(gl);
1841
1842         return demote;
1843 }
1844
1845 /**
1846  * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1847  * @gl: the glock
1848  *
1849  */
1850
1851 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
1852 {
1853         struct gfs2_sbd *sdp = gl->gl_sbd;
1854
1855         spin_lock(&sdp->sd_reclaim_lock);
1856         if (list_empty(&gl->gl_reclaim)) {
1857                 gfs2_glock_hold(gl);
1858                 list_add(&gl->gl_reclaim, &sdp->sd_reclaim_list);
1859                 atomic_inc(&sdp->sd_reclaim_count);
1860         }
1861         spin_unlock(&sdp->sd_reclaim_lock);
1862
1863         wake_up(&sdp->sd_reclaim_wq);
1864 }
1865
1866 /**
1867  * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1868  * @sdp: the filesystem
1869  *
1870  * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1871  * different glock and we notice that there are a lot of glocks in the
1872  * reclaim list.
1873  *
1874  */
1875
1876 void gfs2_reclaim_glock(struct gfs2_sbd *sdp)
1877 {
1878         struct gfs2_glock *gl;
1879
1880         spin_lock(&sdp->sd_reclaim_lock);
1881         if (list_empty(&sdp->sd_reclaim_list)) {
1882                 spin_unlock(&sdp->sd_reclaim_lock);
1883                 return;
1884         }
1885         gl = list_entry(sdp->sd_reclaim_list.next,
1886                         struct gfs2_glock, gl_reclaim);
1887         list_del_init(&gl->gl_reclaim);
1888         spin_unlock(&sdp->sd_reclaim_lock);
1889
1890         atomic_dec(&sdp->sd_reclaim_count);
1891         atomic_inc(&sdp->sd_reclaimed);
1892
1893         if (gfs2_glmutex_trylock(gl)) {
1894                 if (queue_empty(gl, &gl->gl_holders) &&
1895                     gl->gl_state != LM_ST_UNLOCKED &&
1896                     demote_ok(gl))
1897                         handle_callback(gl, LM_ST_UNLOCKED);
1898                 gfs2_glmutex_unlock(gl);
1899         }
1900
1901         gfs2_glock_put(gl);
1902 }
1903
1904 /**
1905  * examine_bucket - Call a function for glock in a hash bucket
1906  * @examiner: the function
1907  * @sdp: the filesystem
1908  * @bucket: the bucket
1909  *
1910  * Returns: 1 if the bucket has entries
1911  */
1912
1913 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1914                           struct gfs2_gl_hash_bucket *bucket)
1915 {
1916         struct glock_plug plug;
1917         struct list_head *tmp;
1918         struct gfs2_glock *gl;
1919         int entries;
1920
1921         /* Add "plug" to end of bucket list, work back up list from there */
1922         memset(&plug.gl_flags, 0, sizeof(unsigned long));
1923         set_bit(GLF_PLUG, &plug.gl_flags);
1924
1925         write_lock(&bucket->hb_lock);
1926         list_add(&plug.gl_list, &bucket->hb_list);
1927         write_unlock(&bucket->hb_lock);
1928
1929         for (;;) {
1930                 write_lock(&bucket->hb_lock);
1931
1932                 for (;;) {
1933                         tmp = plug.gl_list.next;
1934
1935                         if (tmp == &bucket->hb_list) {
1936                                 list_del(&plug.gl_list);
1937                                 entries = !list_empty(&bucket->hb_list);
1938                                 write_unlock(&bucket->hb_lock);
1939                                 return entries;
1940                         }
1941                         gl = list_entry(tmp, struct gfs2_glock, gl_list);
1942
1943                         /* Move plug up list */
1944                         list_move(&plug.gl_list, &gl->gl_list);
1945
1946                         if (test_bit(GLF_PLUG, &gl->gl_flags))
1947                                 continue;
1948
1949                         /* examiner() must glock_put() */
1950                         gfs2_glock_hold(gl);
1951
1952                         break;
1953                 }
1954
1955                 write_unlock(&bucket->hb_lock);
1956
1957                 examiner(gl);
1958         }
1959 }
1960
1961 /**
1962  * scan_glock - look at a glock and see if we can reclaim it
1963  * @gl: the glock to look at
1964  *
1965  */
1966
1967 static void scan_glock(struct gfs2_glock *gl)
1968 {
1969         if (gl->gl_ops == &gfs2_inode_glops)
1970                 goto out;
1971
1972         if (gfs2_glmutex_trylock(gl)) {
1973                 if (queue_empty(gl, &gl->gl_holders) &&
1974                     gl->gl_state != LM_ST_UNLOCKED &&
1975                     demote_ok(gl))
1976                         goto out_schedule;
1977                 gfs2_glmutex_unlock(gl);
1978         }
1979 out:
1980         gfs2_glock_put(gl);
1981         return;
1982
1983 out_schedule:
1984         gfs2_glmutex_unlock(gl);
1985         gfs2_glock_schedule_for_reclaim(gl);
1986         gfs2_glock_put(gl);
1987 }
1988
1989 /**
1990  * gfs2_scand_internal - Look for glocks and inodes to toss from memory
1991  * @sdp: the filesystem
1992  *
1993  */
1994
1995 void gfs2_scand_internal(struct gfs2_sbd *sdp)
1996 {
1997         unsigned int x;
1998
1999         for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
2000                 examine_bucket(scan_glock, sdp, &sdp->sd_gl_hash[x]);
2001                 cond_resched();
2002         }
2003 }
2004
2005 /**
2006  * clear_glock - look at a glock and see if we can free it from glock cache
2007  * @gl: the glock to look at
2008  *
2009  */
2010
2011 static void clear_glock(struct gfs2_glock *gl)
2012 {
2013         struct gfs2_sbd *sdp = gl->gl_sbd;
2014         int released;
2015
2016         spin_lock(&sdp->sd_reclaim_lock);
2017         if (!list_empty(&gl->gl_reclaim)) {
2018                 list_del_init(&gl->gl_reclaim);
2019                 atomic_dec(&sdp->sd_reclaim_count);
2020                 spin_unlock(&sdp->sd_reclaim_lock);
2021                 released = gfs2_glock_put(gl);
2022                 gfs2_assert(sdp, !released);
2023         } else {
2024                 spin_unlock(&sdp->sd_reclaim_lock);
2025         }
2026
2027         if (gfs2_glmutex_trylock(gl)) {
2028                 if (queue_empty(gl, &gl->gl_holders) &&
2029                     gl->gl_state != LM_ST_UNLOCKED)
2030                         handle_callback(gl, LM_ST_UNLOCKED);
2031
2032                 gfs2_glmutex_unlock(gl);
2033         }
2034
2035         gfs2_glock_put(gl);
2036 }
2037
2038 /**
2039  * gfs2_gl_hash_clear - Empty out the glock hash table
2040  * @sdp: the filesystem
2041  * @wait: wait until it's all gone
2042  *
2043  * Called when unmounting the filesystem, or when inter-node lock manager
2044  * requests DROPLOCKS because it is running out of capacity.
2045  */
2046
2047 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp, int wait)
2048 {
2049         unsigned long t;
2050         unsigned int x;
2051         int cont;
2052
2053         t = jiffies;
2054
2055         for (;;) {
2056                 cont = 0;
2057
2058                 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
2059                         if (examine_bucket(clear_glock, sdp,
2060                                            &sdp->sd_gl_hash[x]))
2061                                 cont = 1;
2062
2063                 if (!wait || !cont)
2064                         break;
2065
2066                 if (time_after_eq(jiffies,
2067                                   t + gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
2068                         fs_warn(sdp, "Unmount seems to be stalled. "
2069                                      "Dumping lock state...\n");
2070                         gfs2_dump_lockstate(sdp);
2071                         t = jiffies;
2072                 }
2073
2074                 invalidate_inodes(sdp->sd_vfs);
2075                 msleep(10);
2076         }
2077 }
2078
2079 /*
2080  *  Diagnostic routines to help debug distributed deadlock
2081  */
2082
2083 /**
2084  * dump_holder - print information about a glock holder
2085  * @str: a string naming the type of holder
2086  * @gh: the glock holder
2087  *
2088  * Returns: 0 on success, -ENOBUFS when we run out of space
2089  */
2090
2091 static int dump_holder(char *str, struct gfs2_holder *gh)
2092 {
2093         unsigned int x;
2094         int error = -ENOBUFS;
2095
2096         printk(KERN_INFO "  %s\n", str);
2097         printk(KERN_INFO "    owner = %ld\n",
2098                    (gh->gh_owner) ? (long)gh->gh_owner->pid : -1);
2099         printk(KERN_INFO "    gh_state = %u\n", gh->gh_state);
2100         printk(KERN_INFO "    gh_flags =");
2101         for (x = 0; x < 32; x++)
2102                 if (gh->gh_flags & (1 << x))
2103                         printk(" %u", x);
2104         printk(" \n");
2105         printk(KERN_INFO "    error = %d\n", gh->gh_error);
2106         printk(KERN_INFO "    gh_iflags =");
2107         for (x = 0; x < 32; x++)
2108                 if (test_bit(x, &gh->gh_iflags))
2109                         printk(" %u", x);
2110         printk(" \n");
2111         print_symbol(KERN_INFO "    initialized at: %s\n", gh->gh_ip);
2112
2113         error = 0;
2114
2115         return error;
2116 }
2117
2118 /**
2119  * dump_inode - print information about an inode
2120  * @ip: the inode
2121  *
2122  * Returns: 0 on success, -ENOBUFS when we run out of space
2123  */
2124
2125 static int dump_inode(struct gfs2_inode *ip)
2126 {
2127         unsigned int x;
2128         int error = -ENOBUFS;
2129
2130         printk(KERN_INFO "  Inode:\n");
2131         printk(KERN_INFO "    num = %llu %llu\n",
2132                     (unsigned long long)ip->i_num.no_formal_ino,
2133                     (unsigned long long)ip->i_num.no_addr);
2134         printk(KERN_INFO "    type = %u\n", IF2DT(ip->i_di.di_mode));
2135         printk(KERN_INFO "    i_flags =");
2136         for (x = 0; x < 32; x++)
2137                 if (test_bit(x, &ip->i_flags))
2138                         printk(" %u", x);
2139         printk(" \n");
2140
2141         error = 0;
2142
2143         return error;
2144 }
2145
2146 /**
2147  * dump_glock - print information about a glock
2148  * @gl: the glock
2149  * @count: where we are in the buffer
2150  *
2151  * Returns: 0 on success, -ENOBUFS when we run out of space
2152  */
2153
2154 static int dump_glock(struct gfs2_glock *gl)
2155 {
2156         struct gfs2_holder *gh;
2157         unsigned int x;
2158         int error = -ENOBUFS;
2159
2160         spin_lock(&gl->gl_spin);
2161
2162         printk(KERN_INFO "Glock 0x%p (%u, %llu)\n",
2163                gl,
2164                gl->gl_name.ln_type,
2165                (unsigned long long)gl->gl_name.ln_number);
2166         printk(KERN_INFO "  gl_flags =");
2167         for (x = 0; x < 32; x++)
2168                 if (test_bit(x, &gl->gl_flags))
2169                         printk(" %u", x);
2170         printk(" \n");
2171         printk(KERN_INFO "  gl_ref = %d\n", atomic_read(&gl->gl_ref.refcount));
2172         printk(KERN_INFO "  gl_state = %u\n", gl->gl_state);
2173         printk(KERN_INFO "  gl_owner = %s\n", gl->gl_owner->comm);
2174         print_symbol(KERN_INFO "  gl_ip = %s\n", gl->gl_ip);
2175         printk(KERN_INFO "  req_gh = %s\n", (gl->gl_req_gh) ? "yes" : "no");
2176         printk(KERN_INFO "  req_bh = %s\n", (gl->gl_req_bh) ? "yes" : "no");
2177         printk(KERN_INFO "  lvb_count = %d\n", atomic_read(&gl->gl_lvb_count));
2178         printk(KERN_INFO "  object = %s\n", (gl->gl_object) ? "yes" : "no");
2179         printk(KERN_INFO "  le = %s\n",
2180                    (list_empty(&gl->gl_le.le_list)) ? "no" : "yes");
2181         printk(KERN_INFO "  reclaim = %s\n",
2182                     (list_empty(&gl->gl_reclaim)) ? "no" : "yes");
2183         if (gl->gl_aspace)
2184                 printk(KERN_INFO "  aspace = 0x%p nrpages = %lu\n",
2185                        gl->gl_aspace,
2186                        gl->gl_aspace->i_mapping->nrpages);
2187         else
2188                 printk(KERN_INFO "  aspace = no\n");
2189         printk(KERN_INFO "  ail = %d\n", atomic_read(&gl->gl_ail_count));
2190         if (gl->gl_req_gh) {
2191                 error = dump_holder("Request", gl->gl_req_gh);
2192                 if (error)
2193                         goto out;
2194         }
2195         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
2196                 error = dump_holder("Holder", gh);
2197                 if (error)
2198                         goto out;
2199         }
2200         list_for_each_entry(gh, &gl->gl_waiters1, gh_list) {
2201                 error = dump_holder("Waiter1", gh);
2202                 if (error)
2203                         goto out;
2204         }
2205         list_for_each_entry(gh, &gl->gl_waiters2, gh_list) {
2206                 error = dump_holder("Waiter2", gh);
2207                 if (error)
2208                         goto out;
2209         }
2210         list_for_each_entry(gh, &gl->gl_waiters3, gh_list) {
2211                 error = dump_holder("Waiter3", gh);
2212                 if (error)
2213                         goto out;
2214         }
2215         if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object) {
2216                 if (!test_bit(GLF_LOCK, &gl->gl_flags) &&
2217                     list_empty(&gl->gl_holders)) {
2218                         error = dump_inode(gl->gl_object);
2219                         if (error)
2220                                 goto out;
2221                 } else {
2222                         error = -ENOBUFS;
2223                         printk(KERN_INFO "  Inode: busy\n");
2224                 }
2225         }
2226
2227         error = 0;
2228
2229  out:
2230         spin_unlock(&gl->gl_spin);
2231
2232         return error;
2233 }
2234
2235 /**
2236  * gfs2_dump_lockstate - print out the current lockstate
2237  * @sdp: the filesystem
2238  * @ub: the buffer to copy the information into
2239  *
2240  * If @ub is NULL, dump the lockstate to the console.
2241  *
2242  */
2243
2244 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
2245 {
2246         struct gfs2_gl_hash_bucket *bucket;
2247         struct gfs2_glock *gl;
2248         unsigned int x;
2249         int error = 0;
2250
2251         for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
2252                 bucket = &sdp->sd_gl_hash[x];
2253
2254                 read_lock(&bucket->hb_lock);
2255
2256                 list_for_each_entry(gl, &bucket->hb_list, gl_list) {
2257                         if (test_bit(GLF_PLUG, &gl->gl_flags))
2258                                 continue;
2259
2260                         error = dump_glock(gl);
2261                         if (error)
2262                                 break;
2263                 }
2264
2265                 read_unlock(&bucket->hb_lock);
2266
2267                 if (error)
2268                         break;
2269         }
2270
2271
2272         return error;
2273 }
2274