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                 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1169                 BUG();
1170         }
1171
1172         existing = find_holder_by_owner(&gl->gl_waiters3, gh->gh_owner);
1173         if (existing) {
1174                 print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
1175                 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1176                 BUG();
1177         }
1178
1179         if (gh->gh_flags & LM_FLAG_PRIORITY)
1180                 list_add(&gh->gh_list, &gl->gl_waiters3);
1181         else
1182                 list_add_tail(&gh->gh_list, &gl->gl_waiters3);
1183 }
1184
1185 /**
1186  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1187  * @gh: the holder structure
1188  *
1189  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1190  *
1191  * Returns: 0, GLR_TRYFAILED, or errno on failure
1192  */
1193
1194 int gfs2_glock_nq(struct gfs2_holder *gh)
1195 {
1196         struct gfs2_glock *gl = gh->gh_gl;
1197         struct gfs2_sbd *sdp = gl->gl_sbd;
1198         int error = 0;
1199
1200 restart:
1201         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1202                 set_bit(HIF_ABORTED, &gh->gh_iflags);
1203                 return -EIO;
1204         }
1205
1206         set_bit(HIF_PROMOTE, &gh->gh_iflags);
1207
1208         spin_lock(&gl->gl_spin);
1209         add_to_queue(gh);
1210         run_queue(gl);
1211         spin_unlock(&gl->gl_spin);
1212
1213         if (!(gh->gh_flags & GL_ASYNC)) {
1214                 error = glock_wait_internal(gh);
1215                 if (error == GLR_CANCELED) {
1216                         msleep(100);
1217                         goto restart;
1218                 }
1219         }
1220
1221         clear_bit(GLF_PREFETCH, &gl->gl_flags);
1222
1223         if (error == GLR_TRYFAILED && (gh->gh_flags & GL_DUMP))
1224                 dump_glock(gl);
1225
1226         return error;
1227 }
1228
1229 /**
1230  * gfs2_glock_poll - poll to see if an async request has been completed
1231  * @gh: the holder
1232  *
1233  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1234  */
1235
1236 int gfs2_glock_poll(struct gfs2_holder *gh)
1237 {
1238         struct gfs2_glock *gl = gh->gh_gl;
1239         int ready = 0;
1240
1241         spin_lock(&gl->gl_spin);
1242
1243         if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1244                 ready = 1;
1245         else if (list_empty(&gh->gh_list)) {
1246                 if (gh->gh_error == GLR_CANCELED) {
1247                         spin_unlock(&gl->gl_spin);
1248                         msleep(100);
1249                         if (gfs2_glock_nq(gh))
1250                                 return 1;
1251                         return 0;
1252                 } else
1253                         ready = 1;
1254         }
1255
1256         spin_unlock(&gl->gl_spin);
1257
1258         return ready;
1259 }
1260
1261 /**
1262  * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
1263  * @gh: the holder structure
1264  *
1265  * Returns: 0, GLR_TRYFAILED, or errno on failure
1266  */
1267
1268 int gfs2_glock_wait(struct gfs2_holder *gh)
1269 {
1270         int error;
1271
1272         error = glock_wait_internal(gh);
1273         if (error == GLR_CANCELED) {
1274                 msleep(100);
1275                 gh->gh_flags &= ~GL_ASYNC;
1276                 error = gfs2_glock_nq(gh);
1277         }
1278
1279         return error;
1280 }
1281
1282 /**
1283  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1284  * @gh: the glock holder
1285  *
1286  */
1287
1288 void gfs2_glock_dq(struct gfs2_holder *gh)
1289 {
1290         struct gfs2_glock *gl = gh->gh_gl;
1291         struct gfs2_glock_operations *glops = gl->gl_ops;
1292
1293         if (gh->gh_flags & GL_SYNC)
1294                 set_bit(GLF_SYNC, &gl->gl_flags);
1295
1296         if (gh->gh_flags & GL_NOCACHE)
1297                 handle_callback(gl, LM_ST_UNLOCKED);
1298
1299         gfs2_glmutex_lock(gl);
1300
1301         spin_lock(&gl->gl_spin);
1302         list_del_init(&gh->gh_list);
1303
1304         if (list_empty(&gl->gl_holders)) {
1305                 spin_unlock(&gl->gl_spin);
1306
1307                 if (glops->go_unlock)
1308                         glops->go_unlock(gh);
1309
1310                 if (test_bit(GLF_SYNC, &gl->gl_flags)) {
1311                         if (glops->go_sync)
1312                                 glops->go_sync(gl, DIO_METADATA | DIO_DATA);
1313                 }
1314
1315                 gl->gl_stamp = jiffies;
1316
1317                 spin_lock(&gl->gl_spin);
1318         }
1319
1320         clear_bit(GLF_LOCK, &gl->gl_flags);
1321         run_queue(gl);
1322         spin_unlock(&gl->gl_spin);
1323 }
1324
1325 /**
1326  * gfs2_glock_prefetch - Try to prefetch a glock
1327  * @gl: the glock
1328  * @state: the state to prefetch in
1329  * @flags: flags passed to go_xmote_th()
1330  *
1331  */
1332
1333 static void gfs2_glock_prefetch(struct gfs2_glock *gl, unsigned int state,
1334                                 int flags)
1335 {
1336         struct gfs2_glock_operations *glops = gl->gl_ops;
1337
1338         spin_lock(&gl->gl_spin);
1339
1340         if (test_bit(GLF_LOCK, &gl->gl_flags) ||
1341             !list_empty(&gl->gl_holders) ||
1342             !list_empty(&gl->gl_waiters1) ||
1343             !list_empty(&gl->gl_waiters2) ||
1344             !list_empty(&gl->gl_waiters3) ||
1345             relaxed_state_ok(gl->gl_state, state, flags)) {
1346                 spin_unlock(&gl->gl_spin);
1347                 return;
1348         }
1349
1350         set_bit(GLF_PREFETCH, &gl->gl_flags);
1351         set_bit(GLF_LOCK, &gl->gl_flags);
1352         spin_unlock(&gl->gl_spin);
1353
1354         glops->go_xmote_th(gl, state, flags);
1355 }
1356
1357 static void greedy_work(void *data)
1358 {
1359         struct greedy *gr = data;
1360         struct gfs2_holder *gh = &gr->gr_gh;
1361         struct gfs2_glock *gl = gh->gh_gl;
1362         struct gfs2_glock_operations *glops = gl->gl_ops;
1363
1364         clear_bit(GLF_SKIP_WAITERS2, &gl->gl_flags);
1365
1366         if (glops->go_greedy)
1367                 glops->go_greedy(gl);
1368
1369         spin_lock(&gl->gl_spin);
1370
1371         if (list_empty(&gl->gl_waiters2)) {
1372                 clear_bit(GLF_GREEDY, &gl->gl_flags);
1373                 spin_unlock(&gl->gl_spin);
1374                 gfs2_holder_uninit(gh);
1375                 kfree(gr);
1376         } else {
1377                 gfs2_glock_hold(gl);
1378                 list_add_tail(&gh->gh_list, &gl->gl_waiters2);
1379                 run_queue(gl);
1380                 spin_unlock(&gl->gl_spin);
1381                 gfs2_glock_put(gl);
1382         }
1383 }
1384
1385 /**
1386  * gfs2_glock_be_greedy -
1387  * @gl:
1388  * @time:
1389  *
1390  * Returns: 0 if go_greedy will be called, 1 otherwise
1391  */
1392
1393 int gfs2_glock_be_greedy(struct gfs2_glock *gl, unsigned int time)
1394 {
1395         struct greedy *gr;
1396         struct gfs2_holder *gh;
1397
1398         if (!time || gl->gl_sbd->sd_args.ar_localcaching ||
1399             test_and_set_bit(GLF_GREEDY, &gl->gl_flags))
1400                 return 1;
1401
1402         gr = kmalloc(sizeof(struct greedy), GFP_KERNEL);
1403         if (!gr) {
1404                 clear_bit(GLF_GREEDY, &gl->gl_flags);
1405                 return 1;
1406         }
1407         gh = &gr->gr_gh;
1408
1409         gfs2_holder_init(gl, 0, 0, gh);
1410         set_bit(HIF_GREEDY, &gh->gh_iflags);
1411         INIT_WORK(&gr->gr_work, greedy_work, gr);
1412
1413         set_bit(GLF_SKIP_WAITERS2, &gl->gl_flags);
1414         schedule_delayed_work(&gr->gr_work, time);
1415
1416         return 0;
1417 }
1418
1419 /**
1420  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1421  * @gh: the holder structure
1422  *
1423  */
1424
1425 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1426 {
1427         gfs2_glock_dq(gh);
1428         gfs2_holder_uninit(gh);
1429 }
1430
1431 /**
1432  * gfs2_glock_nq_num - acquire a glock based on lock number
1433  * @sdp: the filesystem
1434  * @number: the lock number
1435  * @glops: the glock operations for the type of glock
1436  * @state: the state to acquire the glock in
1437  * @flags: modifier flags for the aquisition
1438  * @gh: the struct gfs2_holder
1439  *
1440  * Returns: errno
1441  */
1442
1443 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, uint64_t number,
1444                       struct gfs2_glock_operations *glops, unsigned int state,
1445                       int flags, struct gfs2_holder *gh)
1446 {
1447         struct gfs2_glock *gl;
1448         int error;
1449
1450         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1451         if (!error) {
1452                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1453                 gfs2_glock_put(gl);
1454         }
1455
1456         return error;
1457 }
1458
1459 /**
1460  * glock_compare - Compare two struct gfs2_glock structures for sorting
1461  * @arg_a: the first structure
1462  * @arg_b: the second structure
1463  *
1464  */
1465
1466 static int glock_compare(const void *arg_a, const void *arg_b)
1467 {
1468         struct gfs2_holder *gh_a = *(struct gfs2_holder **)arg_a;
1469         struct gfs2_holder *gh_b = *(struct gfs2_holder **)arg_b;
1470         struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1471         struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1472         int ret = 0;
1473
1474         if (a->ln_number > b->ln_number)
1475                 ret = 1;
1476         else if (a->ln_number < b->ln_number)
1477                 ret = -1;
1478         else {
1479                 if (gh_a->gh_state == LM_ST_SHARED &&
1480                     gh_b->gh_state == LM_ST_EXCLUSIVE)
1481                         ret = 1;
1482                 else if (!(gh_a->gh_flags & GL_LOCAL_EXCL) &&
1483                          (gh_b->gh_flags & GL_LOCAL_EXCL))
1484                         ret = 1;
1485         }
1486
1487         return ret;
1488 }
1489
1490 /**
1491  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1492  * @num_gh: the number of structures
1493  * @ghs: an array of struct gfs2_holder structures
1494  *
1495  * Returns: 0 on success (all glocks acquired),
1496  *          errno on failure (no glocks acquired)
1497  */
1498
1499 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1500                      struct gfs2_holder **p)
1501 {
1502         unsigned int x;
1503         int error = 0;
1504
1505         for (x = 0; x < num_gh; x++)
1506                 p[x] = &ghs[x];
1507
1508         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1509
1510         for (x = 0; x < num_gh; x++) {
1511                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1512
1513                 error = gfs2_glock_nq(p[x]);
1514                 if (error) {
1515                         while (x--)
1516                                 gfs2_glock_dq(p[x]);
1517                         break;
1518                 }
1519         }
1520
1521         return error;
1522 }
1523
1524 /**
1525  * gfs2_glock_nq_m - acquire multiple glocks
1526  * @num_gh: the number of structures
1527  * @ghs: an array of struct gfs2_holder structures
1528  *
1529  * Figure out how big an impact this function has.  Either:
1530  * 1) Replace this code with code that calls gfs2_glock_prefetch()
1531  * 2) Forget async stuff and just call nq_m_sync()
1532  * 3) Leave it like it is
1533  *
1534  * Returns: 0 on success (all glocks acquired),
1535  *          errno on failure (no glocks acquired)
1536  */
1537
1538 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1539 {
1540         int *e;
1541         unsigned int x;
1542         int borked = 0, serious = 0;
1543         int error = 0;
1544
1545         if (!num_gh)
1546                 return 0;
1547
1548         if (num_gh == 1) {
1549                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1550                 return gfs2_glock_nq(ghs);
1551         }
1552
1553         e = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL);
1554         if (!e)
1555                 return -ENOMEM;
1556
1557         for (x = 0; x < num_gh; x++) {
1558                 ghs[x].gh_flags |= LM_FLAG_TRY | GL_ASYNC;
1559                 error = gfs2_glock_nq(&ghs[x]);
1560                 if (error) {
1561                         borked = 1;
1562                         serious = error;
1563                         num_gh = x;
1564                         break;
1565                 }
1566         }
1567
1568         for (x = 0; x < num_gh; x++) {
1569                 error = e[x] = glock_wait_internal(&ghs[x]);
1570                 if (error) {
1571                         borked = 1;
1572                         if (error != GLR_TRYFAILED && error != GLR_CANCELED)
1573                                 serious = error;
1574                 }
1575         }
1576
1577         if (!borked) {
1578                 kfree(e);
1579                 return 0;
1580         }
1581
1582         for (x = 0; x < num_gh; x++)
1583                 if (!e[x])
1584                         gfs2_glock_dq(&ghs[x]);
1585
1586         if (serious)
1587                 error = serious;
1588         else {
1589                 for (x = 0; x < num_gh; x++)
1590                         gfs2_holder_reinit(ghs[x].gh_state, ghs[x].gh_flags,
1591                                           &ghs[x]);
1592                 error = nq_m_sync(num_gh, ghs, (struct gfs2_holder **)e);
1593         }
1594
1595         kfree(e);
1596
1597         return error;
1598 }
1599
1600 /**
1601  * gfs2_glock_dq_m - release multiple glocks
1602  * @num_gh: the number of structures
1603  * @ghs: an array of struct gfs2_holder structures
1604  *
1605  */
1606
1607 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1608 {
1609         unsigned int x;
1610
1611         for (x = 0; x < num_gh; x++)
1612                 gfs2_glock_dq(&ghs[x]);
1613 }
1614
1615 /**
1616  * gfs2_glock_dq_uninit_m - release multiple glocks
1617  * @num_gh: the number of structures
1618  * @ghs: an array of struct gfs2_holder structures
1619  *
1620  */
1621
1622 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1623 {
1624         unsigned int x;
1625
1626         for (x = 0; x < num_gh; x++)
1627                 gfs2_glock_dq_uninit(&ghs[x]);
1628 }
1629
1630 /**
1631  * gfs2_glock_prefetch_num - prefetch a glock based on lock number
1632  * @sdp: the filesystem
1633  * @number: the lock number
1634  * @glops: the glock operations for the type of glock
1635  * @state: the state to acquire the glock in
1636  * @flags: modifier flags for the aquisition
1637  *
1638  * Returns: errno
1639  */
1640
1641 void gfs2_glock_prefetch_num(struct gfs2_sbd *sdp, uint64_t number,
1642                              struct gfs2_glock_operations *glops,
1643                              unsigned int state, int flags)
1644 {
1645         struct gfs2_glock *gl;
1646         int error;
1647
1648         if (atomic_read(&sdp->sd_reclaim_count) <
1649             gfs2_tune_get(sdp, gt_reclaim_limit)) {
1650                 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1651                 if (!error) {
1652                         gfs2_glock_prefetch(gl, state, flags);
1653                         gfs2_glock_put(gl);
1654                 }
1655         }
1656 }
1657
1658 /**
1659  * gfs2_lvb_hold - attach a LVB from a glock
1660  * @gl: The glock in question
1661  *
1662  */
1663
1664 int gfs2_lvb_hold(struct gfs2_glock *gl)
1665 {
1666         int error;
1667
1668         gfs2_glmutex_lock(gl);
1669
1670         if (!atomic_read(&gl->gl_lvb_count)) {
1671                 error = gfs2_lm_hold_lvb(gl->gl_sbd, gl->gl_lock, &gl->gl_lvb);
1672                 if (error) {
1673                         gfs2_glmutex_unlock(gl);
1674                         return error;
1675                 }
1676                 gfs2_glock_hold(gl);
1677         }
1678         atomic_inc(&gl->gl_lvb_count);
1679
1680         gfs2_glmutex_unlock(gl);
1681
1682         return 0;
1683 }
1684
1685 /**
1686  * gfs2_lvb_unhold - detach a LVB from a glock
1687  * @gl: The glock in question
1688  *
1689  */
1690
1691 void gfs2_lvb_unhold(struct gfs2_glock *gl)
1692 {
1693         gfs2_glock_hold(gl);
1694         gfs2_glmutex_lock(gl);
1695
1696         gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count) > 0);
1697         if (atomic_dec_and_test(&gl->gl_lvb_count)) {
1698                 gfs2_lm_unhold_lvb(gl->gl_sbd, gl->gl_lock, gl->gl_lvb);
1699                 gl->gl_lvb = NULL;
1700                 gfs2_glock_put(gl);
1701         }
1702
1703         gfs2_glmutex_unlock(gl);
1704         gfs2_glock_put(gl);
1705 }
1706
1707 #if 0
1708 void gfs2_lvb_sync(struct gfs2_glock *gl)
1709 {
1710         gfs2_glmutex_lock(gl);
1711
1712         gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count));
1713         if (!gfs2_assert_warn(gl->gl_sbd, gfs2_glock_is_held_excl(gl)))
1714                 gfs2_lm_sync_lvb(gl->gl_sbd, gl->gl_lock, gl->gl_lvb);
1715
1716         gfs2_glmutex_unlock(gl);
1717 }
1718 #endif  /*  0  */
1719
1720 static void blocking_cb(struct gfs2_sbd *sdp, struct lm_lockname *name,
1721                         unsigned int state)
1722 {
1723         struct gfs2_glock *gl;
1724
1725         gl = gfs2_glock_find(sdp, name);
1726         if (!gl)
1727                 return;
1728
1729         if (gl->gl_ops->go_callback)
1730                 gl->gl_ops->go_callback(gl, state);
1731         handle_callback(gl, state);
1732
1733         spin_lock(&gl->gl_spin);
1734         run_queue(gl);
1735         spin_unlock(&gl->gl_spin);
1736
1737         gfs2_glock_put(gl);
1738 }
1739
1740 /**
1741  * gfs2_glock_cb - Callback used by locking module
1742  * @fsdata: Pointer to the superblock
1743  * @type: Type of callback
1744  * @data: Type dependent data pointer
1745  *
1746  * Called by the locking module when it wants to tell us something.
1747  * Either we need to drop a lock, one of our ASYNC requests completed, or
1748  * a journal from another client needs to be recovered.
1749  */
1750
1751 void gfs2_glock_cb(lm_fsdata_t *fsdata, unsigned int type, void *data)
1752 {
1753         struct gfs2_sbd *sdp = (struct gfs2_sbd *)fsdata;
1754
1755         switch (type) {
1756         case LM_CB_NEED_E:
1757                 blocking_cb(sdp, data, LM_ST_UNLOCKED);
1758                 return;
1759
1760         case LM_CB_NEED_D:
1761                 blocking_cb(sdp, data, LM_ST_DEFERRED);
1762                 return;
1763
1764         case LM_CB_NEED_S:
1765                 blocking_cb(sdp, data, LM_ST_SHARED);
1766                 return;
1767
1768         case LM_CB_ASYNC: {
1769                 struct lm_async_cb *async = data;
1770                 struct gfs2_glock *gl;
1771
1772                 gl = gfs2_glock_find(sdp, &async->lc_name);
1773                 if (gfs2_assert_warn(sdp, gl))
1774                         return;
1775                 if (!gfs2_assert_warn(sdp, gl->gl_req_bh))
1776                         gl->gl_req_bh(gl, async->lc_ret);
1777                 gfs2_glock_put(gl);
1778                 return;
1779         }
1780
1781         case LM_CB_NEED_RECOVERY:
1782                 gfs2_jdesc_make_dirty(sdp, *(unsigned int *)data);
1783                 if (sdp->sd_recoverd_process)
1784                         wake_up_process(sdp->sd_recoverd_process);
1785                 return;
1786
1787         case LM_CB_DROPLOCKS:
1788                 gfs2_gl_hash_clear(sdp, NO_WAIT);
1789                 gfs2_quota_scan(sdp);
1790                 return;
1791
1792         default:
1793                 gfs2_assert_warn(sdp, 0);
1794                 return;
1795         }
1796 }
1797
1798 /**
1799  * gfs2_iopen_go_callback - Try to kick the inode/vnode associated with an
1800  *                          iopen glock from memory
1801  * @io_gl: the iopen glock
1802  * @state: the state into which the glock should be put
1803  *
1804  */
1805
1806 void gfs2_iopen_go_callback(struct gfs2_glock *io_gl, unsigned int state)
1807 {
1808
1809         if (state != LM_ST_UNLOCKED)
1810                 return;
1811         /* FIXME: remove this? */
1812 }
1813
1814 /**
1815  * demote_ok - Check to see if it's ok to unlock a glock
1816  * @gl: the glock
1817  *
1818  * Returns: 1 if it's ok
1819  */
1820
1821 static int demote_ok(struct gfs2_glock *gl)
1822 {
1823         struct gfs2_sbd *sdp = gl->gl_sbd;
1824         struct gfs2_glock_operations *glops = gl->gl_ops;
1825         int demote = 1;
1826
1827         if (test_bit(GLF_STICKY, &gl->gl_flags))
1828                 demote = 0;
1829         else if (test_bit(GLF_PREFETCH, &gl->gl_flags))
1830                 demote = time_after_eq(jiffies,
1831                                     gl->gl_stamp +
1832                                     gfs2_tune_get(sdp, gt_prefetch_secs) * HZ);
1833         else if (glops->go_demote_ok)
1834                 demote = glops->go_demote_ok(gl);
1835
1836         return demote;
1837 }
1838
1839 /**
1840  * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1841  * @gl: the glock
1842  *
1843  */
1844
1845 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
1846 {
1847         struct gfs2_sbd *sdp = gl->gl_sbd;
1848
1849         spin_lock(&sdp->sd_reclaim_lock);
1850         if (list_empty(&gl->gl_reclaim)) {
1851                 gfs2_glock_hold(gl);
1852                 list_add(&gl->gl_reclaim, &sdp->sd_reclaim_list);
1853                 atomic_inc(&sdp->sd_reclaim_count);
1854         }
1855         spin_unlock(&sdp->sd_reclaim_lock);
1856
1857         wake_up(&sdp->sd_reclaim_wq);
1858 }
1859
1860 /**
1861  * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1862  * @sdp: the filesystem
1863  *
1864  * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1865  * different glock and we notice that there are a lot of glocks in the
1866  * reclaim list.
1867  *
1868  */
1869
1870 void gfs2_reclaim_glock(struct gfs2_sbd *sdp)
1871 {
1872         struct gfs2_glock *gl;
1873
1874         spin_lock(&sdp->sd_reclaim_lock);
1875         if (list_empty(&sdp->sd_reclaim_list)) {
1876                 spin_unlock(&sdp->sd_reclaim_lock);
1877                 return;
1878         }
1879         gl = list_entry(sdp->sd_reclaim_list.next,
1880                         struct gfs2_glock, gl_reclaim);
1881         list_del_init(&gl->gl_reclaim);
1882         spin_unlock(&sdp->sd_reclaim_lock);
1883
1884         atomic_dec(&sdp->sd_reclaim_count);
1885         atomic_inc(&sdp->sd_reclaimed);
1886
1887         if (gfs2_glmutex_trylock(gl)) {
1888                 if (queue_empty(gl, &gl->gl_holders) &&
1889                     gl->gl_state != LM_ST_UNLOCKED &&
1890                     demote_ok(gl))
1891                         handle_callback(gl, LM_ST_UNLOCKED);
1892                 gfs2_glmutex_unlock(gl);
1893         }
1894
1895         gfs2_glock_put(gl);
1896 }
1897
1898 /**
1899  * examine_bucket - Call a function for glock in a hash bucket
1900  * @examiner: the function
1901  * @sdp: the filesystem
1902  * @bucket: the bucket
1903  *
1904  * Returns: 1 if the bucket has entries
1905  */
1906
1907 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1908                           struct gfs2_gl_hash_bucket *bucket)
1909 {
1910         struct glock_plug plug;
1911         struct list_head *tmp;
1912         struct gfs2_glock *gl;
1913         int entries;
1914
1915         /* Add "plug" to end of bucket list, work back up list from there */
1916         memset(&plug.gl_flags, 0, sizeof(unsigned long));
1917         set_bit(GLF_PLUG, &plug.gl_flags);
1918
1919         write_lock(&bucket->hb_lock);
1920         list_add(&plug.gl_list, &bucket->hb_list);
1921         write_unlock(&bucket->hb_lock);
1922
1923         for (;;) {
1924                 write_lock(&bucket->hb_lock);
1925
1926                 for (;;) {
1927                         tmp = plug.gl_list.next;
1928
1929                         if (tmp == &bucket->hb_list) {
1930                                 list_del(&plug.gl_list);
1931                                 entries = !list_empty(&bucket->hb_list);
1932                                 write_unlock(&bucket->hb_lock);
1933                                 return entries;
1934                         }
1935                         gl = list_entry(tmp, struct gfs2_glock, gl_list);
1936
1937                         /* Move plug up list */
1938                         list_move(&plug.gl_list, &gl->gl_list);
1939
1940                         if (test_bit(GLF_PLUG, &gl->gl_flags))
1941                                 continue;
1942
1943                         /* examiner() must glock_put() */
1944                         gfs2_glock_hold(gl);
1945
1946                         break;
1947                 }
1948
1949                 write_unlock(&bucket->hb_lock);
1950
1951                 examiner(gl);
1952         }
1953 }
1954
1955 /**
1956  * scan_glock - look at a glock and see if we can reclaim it
1957  * @gl: the glock to look at
1958  *
1959  */
1960
1961 static void scan_glock(struct gfs2_glock *gl)
1962 {
1963         if (gl->gl_ops == &gfs2_inode_glops)
1964                 goto out;
1965
1966         if (gfs2_glmutex_trylock(gl)) {
1967                 if (queue_empty(gl, &gl->gl_holders) &&
1968                     gl->gl_state != LM_ST_UNLOCKED &&
1969                     demote_ok(gl))
1970                         goto out_schedule;
1971                 gfs2_glmutex_unlock(gl);
1972         }
1973 out:
1974         gfs2_glock_put(gl);
1975         return;
1976
1977 out_schedule:
1978         gfs2_glmutex_unlock(gl);
1979         gfs2_glock_schedule_for_reclaim(gl);
1980         gfs2_glock_put(gl);
1981 }
1982
1983 /**
1984  * gfs2_scand_internal - Look for glocks and inodes to toss from memory
1985  * @sdp: the filesystem
1986  *
1987  */
1988
1989 void gfs2_scand_internal(struct gfs2_sbd *sdp)
1990 {
1991         unsigned int x;
1992
1993         for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1994                 examine_bucket(scan_glock, sdp, &sdp->sd_gl_hash[x]);
1995                 cond_resched();
1996         }
1997 }
1998
1999 /**
2000  * clear_glock - look at a glock and see if we can free it from glock cache
2001  * @gl: the glock to look at
2002  *
2003  */
2004
2005 static void clear_glock(struct gfs2_glock *gl)
2006 {
2007         struct gfs2_sbd *sdp = gl->gl_sbd;
2008         int released;
2009
2010         spin_lock(&sdp->sd_reclaim_lock);
2011         if (!list_empty(&gl->gl_reclaim)) {
2012                 list_del_init(&gl->gl_reclaim);
2013                 atomic_dec(&sdp->sd_reclaim_count);
2014                 spin_unlock(&sdp->sd_reclaim_lock);
2015                 released = gfs2_glock_put(gl);
2016                 gfs2_assert(sdp, !released);
2017         } else {
2018                 spin_unlock(&sdp->sd_reclaim_lock);
2019         }
2020
2021         if (gfs2_glmutex_trylock(gl)) {
2022                 if (queue_empty(gl, &gl->gl_holders) &&
2023                     gl->gl_state != LM_ST_UNLOCKED)
2024                         handle_callback(gl, LM_ST_UNLOCKED);
2025
2026                 gfs2_glmutex_unlock(gl);
2027         }
2028
2029         gfs2_glock_put(gl);
2030 }
2031
2032 /**
2033  * gfs2_gl_hash_clear - Empty out the glock hash table
2034  * @sdp: the filesystem
2035  * @wait: wait until it's all gone
2036  *
2037  * Called when unmounting the filesystem, or when inter-node lock manager
2038  * requests DROPLOCKS because it is running out of capacity.
2039  */
2040
2041 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp, int wait)
2042 {
2043         unsigned long t;
2044         unsigned int x;
2045         int cont;
2046
2047         t = jiffies;
2048
2049         for (;;) {
2050                 cont = 0;
2051
2052                 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
2053                         if (examine_bucket(clear_glock, sdp,
2054                                            &sdp->sd_gl_hash[x]))
2055                                 cont = 1;
2056
2057                 if (!wait || !cont)
2058                         break;
2059
2060                 if (time_after_eq(jiffies,
2061                                   t + gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
2062                         fs_warn(sdp, "Unmount seems to be stalled. "
2063                                      "Dumping lock state...\n");
2064                         gfs2_dump_lockstate(sdp);
2065                         t = jiffies;
2066                 }
2067
2068                 invalidate_inodes(sdp->sd_vfs);
2069                 msleep(10);
2070         }
2071 }
2072
2073 /*
2074  *  Diagnostic routines to help debug distributed deadlock
2075  */
2076
2077 /**
2078  * dump_holder - print information about a glock holder
2079  * @str: a string naming the type of holder
2080  * @gh: the glock holder
2081  *
2082  * Returns: 0 on success, -ENOBUFS when we run out of space
2083  */
2084
2085 static int dump_holder(char *str, struct gfs2_holder *gh)
2086 {
2087         unsigned int x;
2088         int error = -ENOBUFS;
2089
2090         printk(KERN_INFO "  %s\n", str);
2091         printk(KERN_INFO "    owner = %ld\n",
2092                    (gh->gh_owner) ? (long)gh->gh_owner->pid : -1);
2093         printk(KERN_INFO "    gh_state = %u\n", gh->gh_state);
2094         printk(KERN_INFO "    gh_flags =");
2095         for (x = 0; x < 32; x++)
2096                 if (gh->gh_flags & (1 << x))
2097                         printk(" %u", x);
2098         printk(" \n");
2099         printk(KERN_INFO "    error = %d\n", gh->gh_error);
2100         printk(KERN_INFO "    gh_iflags =");
2101         for (x = 0; x < 32; x++)
2102                 if (test_bit(x, &gh->gh_iflags))
2103                         printk(" %u", x);
2104         printk(" \n");
2105         print_symbol(KERN_INFO "    initialized at: %s\n", gh->gh_ip);
2106
2107         error = 0;
2108
2109         return error;
2110 }
2111
2112 /**
2113  * dump_inode - print information about an inode
2114  * @ip: the inode
2115  *
2116  * Returns: 0 on success, -ENOBUFS when we run out of space
2117  */
2118
2119 static int dump_inode(struct gfs2_inode *ip)
2120 {
2121         unsigned int x;
2122         int error = -ENOBUFS;
2123
2124         printk(KERN_INFO "  Inode:\n");
2125         printk(KERN_INFO "    num = %llu %llu\n",
2126                     (unsigned long long)ip->i_num.no_formal_ino,
2127                     (unsigned long long)ip->i_num.no_addr);
2128         printk(KERN_INFO "    type = %u\n", IF2DT(ip->i_di.di_mode));
2129         printk(KERN_INFO "    i_flags =");
2130         for (x = 0; x < 32; x++)
2131                 if (test_bit(x, &ip->i_flags))
2132                         printk(" %u", x);
2133         printk(" \n");
2134
2135         error = 0;
2136
2137         return error;
2138 }
2139
2140 /**
2141  * dump_glock - print information about a glock
2142  * @gl: the glock
2143  * @count: where we are in the buffer
2144  *
2145  * Returns: 0 on success, -ENOBUFS when we run out of space
2146  */
2147
2148 static int dump_glock(struct gfs2_glock *gl)
2149 {
2150         struct gfs2_holder *gh;
2151         unsigned int x;
2152         int error = -ENOBUFS;
2153
2154         spin_lock(&gl->gl_spin);
2155
2156         printk(KERN_INFO "Glock 0x%p (%u, %llu)\n",
2157                gl,
2158                gl->gl_name.ln_type,
2159                (unsigned long long)gl->gl_name.ln_number);
2160         printk(KERN_INFO "  gl_flags =");
2161         for (x = 0; x < 32; x++)
2162                 if (test_bit(x, &gl->gl_flags))
2163                         printk(" %u", x);
2164         printk(" \n");
2165         printk(KERN_INFO "  gl_ref = %d\n", atomic_read(&gl->gl_ref.refcount));
2166         printk(KERN_INFO "  gl_state = %u\n", gl->gl_state);
2167         printk(KERN_INFO "  gl_owner = %s\n", gl->gl_owner->comm);
2168         print_symbol(KERN_INFO "  gl_ip = %s\n", gl->gl_ip);
2169         printk(KERN_INFO "  req_gh = %s\n", (gl->gl_req_gh) ? "yes" : "no");
2170         printk(KERN_INFO "  req_bh = %s\n", (gl->gl_req_bh) ? "yes" : "no");
2171         printk(KERN_INFO "  lvb_count = %d\n", atomic_read(&gl->gl_lvb_count));
2172         printk(KERN_INFO "  object = %s\n", (gl->gl_object) ? "yes" : "no");
2173         printk(KERN_INFO "  le = %s\n",
2174                    (list_empty(&gl->gl_le.le_list)) ? "no" : "yes");
2175         printk(KERN_INFO "  reclaim = %s\n",
2176                     (list_empty(&gl->gl_reclaim)) ? "no" : "yes");
2177         if (gl->gl_aspace)
2178                 printk(KERN_INFO "  aspace = 0x%p nrpages = %lu\n",
2179                        gl->gl_aspace,
2180                        gl->gl_aspace->i_mapping->nrpages);
2181         else
2182                 printk(KERN_INFO "  aspace = no\n");
2183         printk(KERN_INFO "  ail = %d\n", atomic_read(&gl->gl_ail_count));
2184         if (gl->gl_req_gh) {
2185                 error = dump_holder("Request", gl->gl_req_gh);
2186                 if (error)
2187                         goto out;
2188         }
2189         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
2190                 error = dump_holder("Holder", gh);
2191                 if (error)
2192                         goto out;
2193         }
2194         list_for_each_entry(gh, &gl->gl_waiters1, gh_list) {
2195                 error = dump_holder("Waiter1", gh);
2196                 if (error)
2197                         goto out;
2198         }
2199         list_for_each_entry(gh, &gl->gl_waiters2, gh_list) {
2200                 error = dump_holder("Waiter2", gh);
2201                 if (error)
2202                         goto out;
2203         }
2204         list_for_each_entry(gh, &gl->gl_waiters3, gh_list) {
2205                 error = dump_holder("Waiter3", gh);
2206                 if (error)
2207                         goto out;
2208         }
2209         if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object) {
2210                 if (!test_bit(GLF_LOCK, &gl->gl_flags) &&
2211                     list_empty(&gl->gl_holders)) {
2212                         error = dump_inode(gl->gl_object);
2213                         if (error)
2214                                 goto out;
2215                 } else {
2216                         error = -ENOBUFS;
2217                         printk(KERN_INFO "  Inode: busy\n");
2218                 }
2219         }
2220
2221         error = 0;
2222
2223  out:
2224         spin_unlock(&gl->gl_spin);
2225
2226         return error;
2227 }
2228
2229 /**
2230  * gfs2_dump_lockstate - print out the current lockstate
2231  * @sdp: the filesystem
2232  * @ub: the buffer to copy the information into
2233  *
2234  * If @ub is NULL, dump the lockstate to the console.
2235  *
2236  */
2237
2238 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
2239 {
2240         struct gfs2_gl_hash_bucket *bucket;
2241         struct gfs2_glock *gl;
2242         unsigned int x;
2243         int error = 0;
2244
2245         for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
2246                 bucket = &sdp->sd_gl_hash[x];
2247
2248                 read_lock(&bucket->hb_lock);
2249
2250                 list_for_each_entry(gl, &bucket->hb_list, gl_list) {
2251                         if (test_bit(GLF_PLUG, &gl->gl_flags))
2252                                 continue;
2253
2254                         error = dump_glock(gl);
2255                         if (error)
2256                                 break;
2257                 }
2258
2259                 read_unlock(&bucket->hb_lock);
2260
2261                 if (error)
2262                         break;
2263         }
2264
2265
2266         return error;
2267 }
2268