fs/gfs2/glock.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
   4  *
   5  * This copyrighted material is made available to anyone wishing to use,
   6  * modify, copy, or redistribute it subject to the terms and conditions
   7  * of the GNU General Public License version 2.
   8  */
   9
  10 #include <linux/sched.h>
  11 #include <linux/slab.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/buffer_head.h>
  14 #include <linux/delay.h>
  15 #include <linux/sort.h>
  16 #include <linux/jhash.h>
  17 #include <linux/kallsyms.h>
  18 #include <linux/gfs2_ondisk.h>
  19 #include <linux/list.h>
  20 #include <linux/wait.h>
  21 #include <linux/module.h>
  22 #include <asm/uaccess.h>
  23 #include <linux/seq_file.h>
  24 #include <linux/debugfs.h>
  25 #include <linux/kthread.h>
  26 #include <linux/freezer.h>
  27 #include <linux/workqueue.h>
  28 #include <linux/jiffies.h>
  29 #include <linux/rcupdate.h>
  30 #include <linux/rculist_bl.h>
  31 #include <linux/bit_spinlock.h>
  32 #include <linux/percpu.h>
  33 #include <linux/list_sort.h>
  34
  35 #include "gfs2.h"
  36 #include "incore.h"
  37 #include "glock.h"
  38 #include "glops.h"
  39 #include "inode.h"
  40 #include "lops.h"
  41 #include "meta_io.h"
  42 #include "quota.h"
  43 #include "super.h"
  44 #include "util.h"
  45 #include "bmap.h"
  46 #define CREATE_TRACE_POINTS
  47 #include "trace_gfs2.h"
  48
  49 struct gfs2_glock_iter {
  50         int hash;                       /* hash bucket index           */
  51         unsigned nhash;                 /* Index within current bucket */
  52         struct gfs2_sbd *sdp;           /* incore superblock           */
  53         struct gfs2_glock *gl;          /* current glock struct        */
  54         loff_t last_pos;                /* last position               */
  55 };
  56
  57 typedef void (*glock_examiner) (struct gfs2_glock * gl);
  58
  59 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
  60
  61 static struct dentry *gfs2_root;
  62 static struct workqueue_struct *glock_workqueue;
  63 struct workqueue_struct *gfs2_delete_workqueue;
  64 static LIST_HEAD(lru_list);
  65 static atomic_t lru_count = ATOMIC_INIT(0);
  66 static DEFINE_SPINLOCK(lru_lock);
  67
  68 #define GFS2_GL_HASH_SHIFT      15
  69 #define GFS2_GL_HASH_SIZE       (1 << GFS2_GL_HASH_SHIFT)
  70 #define GFS2_GL_HASH_MASK       (GFS2_GL_HASH_SIZE - 1)
  71
  72 static struct hlist_bl_head gl_hash_table[GFS2_GL_HASH_SIZE];
  73 static struct dentry *gfs2_root;
  74
  75 /**
  76  * gl_hash() - Turn glock number into hash bucket number
  77  * @lock: The glock number
  78  *
  79  * Returns: The number of the corresponding hash bucket
  80  */
  81
  82 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
  83                             const struct lm_lockname *name)
  84 {
  85         unsigned int h;
  86
  87         h = jhash(&name->ln_number, sizeof(u64), 0);
  88         h = jhash(&name->ln_type, sizeof(unsigned int), h);
  89         h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
  90         h &= GFS2_GL_HASH_MASK;
  91
  92         return h;
  93 }
  94
  95 static inline void spin_lock_bucket(unsigned int hash)
  96 {
  97         hlist_bl_lock(&gl_hash_table[hash]);
  98 }
  99
 100 static inline void spin_unlock_bucket(unsigned int hash)
 101 {
 102         hlist_bl_unlock(&gl_hash_table[hash]);
 103 }
 104
 105 static void gfs2_glock_dealloc(struct rcu_head *rcu)
 106 {
 107         struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
 108
 109         if (gl->gl_ops->go_flags & GLOF_ASPACE) {
 110                 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
 111         } else {
 112                 kfree(gl->gl_lksb.sb_lvbptr);
 113                 kmem_cache_free(gfs2_glock_cachep, gl);
 114         }
 115 }
 116
 117 void gfs2_glock_free(struct gfs2_glock *gl)
 118 {
 119         struct gfs2_sbd *sdp = gl->gl_sbd;
 120
 121         call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
 122         if (atomic_dec_and_test(&sdp->sd_glock_disposal))
 123                 wake_up(&sdp->sd_glock_wait);
 124 }
 125
 126 /**
 127  * gfs2_glock_hold() - increment reference count on glock
 128  * @gl: The glock to hold
 129  *
 130  */
 131
 132 void gfs2_glock_hold(struct gfs2_glock *gl)
 133 {
 134         GLOCK_BUG_ON(gl, atomic_read(&gl->gl_ref) == 0);
 135         atomic_inc(&gl->gl_ref);
 136 }
 137
 138 /**
 139  * demote_ok - Check to see if it's ok to unlock a glock
 140  * @gl: the glock
 141  *
 142  * Returns: 1 if it's ok
 143  */
 144
 145 static int demote_ok(const struct gfs2_glock *gl)
 146 {
 147         const struct gfs2_glock_operations *glops = gl->gl_ops;
 148
 149         if (gl->gl_state == LM_ST_UNLOCKED)
 150                 return 0;
 151         if (!list_empty(&gl->gl_holders))
 152                 return 0;
 153         if (glops->go_demote_ok)
 154                 return glops->go_demote_ok(gl);
 155         return 1;
 156 }
 157
 158
 159 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
 160 {
 161         spin_lock(&lru_lock);
 162
 163         if (!list_empty(&gl->gl_lru))
 164                 list_del_init(&gl->gl_lru);
 165         else
 166                 atomic_inc(&lru_count);
 167
 168         list_add_tail(&gl->gl_lru, &lru_list);
 169         set_bit(GLF_LRU, &gl->gl_flags);
 170         spin_unlock(&lru_lock);
 171 }
 172
 173 static void __gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
 174 {
 175         if (!list_empty(&gl->gl_lru)) {
 176                 list_del_init(&gl->gl_lru);
 177                 atomic_dec(&lru_count);
 178                 clear_bit(GLF_LRU, &gl->gl_flags);
 179         }
 180 }
 181
 182 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
 183 {
 184         spin_lock(&lru_lock);
 185         __gfs2_glock_remove_from_lru(gl);
 186         spin_unlock(&lru_lock);
 187 }
 188
 189 /**
 190  * gfs2_glock_put_nolock() - Decrement reference count on glock
 191  * @gl: The glock to put
 192  *
 193  * This function should only be used if the caller has its own reference
 194  * to the glock, in addition to the one it is dropping.
 195  */
 196
 197 void gfs2_glock_put_nolock(struct gfs2_glock *gl)
 198 {
 199         if (atomic_dec_and_test(&gl->gl_ref))
 200                 GLOCK_BUG_ON(gl, 1);
 201 }
 202
 203 /**
 204  * gfs2_glock_put() - Decrement reference count on glock
 205  * @gl: The glock to put
 206  *
 207  */
 208
 209 void gfs2_glock_put(struct gfs2_glock *gl)
 210 {
 211         struct gfs2_sbd *sdp = gl->gl_sbd;
 212         struct address_space *mapping = gfs2_glock2aspace(gl);
 213
 214         if (atomic_dec_and_lock(&gl->gl_ref, &lru_lock)) {
 215                 __gfs2_glock_remove_from_lru(gl);
 216                 spin_unlock(&lru_lock);
 217                 spin_lock_bucket(gl->gl_hash);
 218                 hlist_bl_del_rcu(&gl->gl_list);
 219                 spin_unlock_bucket(gl->gl_hash);
 220                 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
 221                 GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
 222                 trace_gfs2_glock_put(gl);
 223                 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
 224         }
 225 }
 226
 227 /**
 228  * search_bucket() - Find struct gfs2_glock by lock number
 229  * @bucket: the bucket to search
 230  * @name: The lock name
 231  *
 232  * Returns: NULL, or the struct gfs2_glock with the requested number
 233  */
 234
 235 static struct gfs2_glock *search_bucket(unsigned int hash,
 236                                         const struct gfs2_sbd *sdp,
 237                                         const struct lm_lockname *name)
 238 {
 239         struct gfs2_glock *gl;
 240         struct hlist_bl_node *h;
 241
 242         hlist_bl_for_each_entry_rcu(gl, h, &gl_hash_table[hash], gl_list) {
 243                 if (!lm_name_equal(&gl->gl_name, name))
 244                         continue;
 245                 if (gl->gl_sbd != sdp)
 246                         continue;
 247                 if (atomic_inc_not_zero(&gl->gl_ref))
 248                         return gl;
 249         }
 250
 251         return NULL;
 252 }
 253
 254 /**
 255  * may_grant - check if its ok to grant a new lock
 256  * @gl: The glock
 257  * @gh: The lock request which we wish to grant
 258  *
 259  * Returns: true if its ok to grant the lock
 260  */
 261
 262 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
 263 {
 264         const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
 265         if ((gh->gh_state == LM_ST_EXCLUSIVE ||
 266              gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
 267                 return 0;
 268         if (gl->gl_state == gh->gh_state)
 269                 return 1;
 270         if (gh->gh_flags & GL_EXACT)
 271                 return 0;
 272         if (gl->gl_state == LM_ST_EXCLUSIVE) {
 273                 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
 274                         return 1;
 275                 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
 276                         return 1;
 277         }
 278         if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
 279                 return 1;
 280         return 0;
 281 }
 282
 283 static void gfs2_holder_wake(struct gfs2_holder *gh)
 284 {
 285         clear_bit(HIF_WAIT, &gh->gh_iflags);
 286         smp_mb__after_clear_bit();
 287         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
 288 }
 289
 290 /**
 291  * do_error - Something unexpected has happened during a lock request
 292  *
 293  */
 294
 295 static inline void do_error(struct gfs2_glock *gl, const int ret)
 296 {
 297         struct gfs2_holder *gh, *tmp;
 298
 299         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
 300                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 301                         continue;
 302                 if (ret & LM_OUT_ERROR)
 303                         gh->gh_error = -EIO;
 304                 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
 305                         gh->gh_error = GLR_TRYFAILED;
 306                 else
 307                         continue;
 308                 list_del_init(&gh->gh_list);
 309                 trace_gfs2_glock_queue(gh, 0);
 310                 gfs2_holder_wake(gh);
 311         }
 312 }
 313
 314 /**
 315  * do_promote - promote as many requests as possible on the current queue
 316  * @gl: The glock
 317  *
 318  * Returns: 1 if there is a blocked holder at the head of the list, or 2
 319  *          if a type specific operation is underway.
 320  */
 321
 322 static int do_promote(struct gfs2_glock *gl)
 323 __releases(&gl->gl_spin)
 324 __acquires(&gl->gl_spin)
 325 {
 326         const struct gfs2_glock_operations *glops = gl->gl_ops;
 327         struct gfs2_holder *gh, *tmp;
 328         int ret;
 329
 330 restart:
 331         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
 332                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 333                         continue;
 334                 if (may_grant(gl, gh)) {
 335                         if (gh->gh_list.prev == &gl->gl_holders &&
 336                             glops->go_lock) {
 337                                 spin_unlock(&gl->gl_spin);
 338                                 /* FIXME: eliminate this eventually */
 339                                 ret = glops->go_lock(gh);
 340                                 spin_lock(&gl->gl_spin);
 341                                 if (ret) {
 342                                         if (ret == 1)
 343                                                 return 2;
 344                                         gh->gh_error = ret;
 345                                         list_del_init(&gh->gh_list);
 346                                         trace_gfs2_glock_queue(gh, 0);
 347                                         gfs2_holder_wake(gh);
 348                                         goto restart;
 349                                 }
 350                                 set_bit(HIF_HOLDER, &gh->gh_iflags);
 351                                 trace_gfs2_promote(gh, 1);
 352                                 gfs2_holder_wake(gh);
 353                                 goto restart;
 354                         }
 355                         set_bit(HIF_HOLDER, &gh->gh_iflags);
 356                         trace_gfs2_promote(gh, 0);
 357                         gfs2_holder_wake(gh);
 358                         continue;
 359                 }
 360                 if (gh->gh_list.prev == &gl->gl_holders)
 361                         return 1;
 362                 do_error(gl, 0);
 363                 break;
 364         }
 365         return 0;
 366 }
 367
 368 /**
 369  * find_first_waiter - find the first gh that's waiting for the glock
 370  * @gl: the glock
 371  */
 372
 373 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
 374 {
 375         struct gfs2_holder *gh;
 376
 377         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
 378                 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
 379                         return gh;
 380         }
 381         return NULL;
 382 }
 383
 384 /**
 385  * state_change - record that the glock is now in a different state
 386  * @gl: the glock
 387  * @new_state the new state
 388  *
 389  */
 390
 391 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
 392 {
 393         int held1, held2;
 394
 395         held1 = (gl->gl_state != LM_ST_UNLOCKED);
 396         held2 = (new_state != LM_ST_UNLOCKED);
 397
 398         if (held1 != held2) {
 399                 if (held2)
 400                         gfs2_glock_hold(gl);
 401                 else
 402                         gfs2_glock_put_nolock(gl);
 403         }
 404         if (held1 && held2 && list_empty(&gl->gl_holders))
 405                 clear_bit(GLF_QUEUED, &gl->gl_flags);
 406
 407         if (new_state != gl->gl_target)
 408                 /* shorten our minimum hold time */
 409                 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
 410                                        GL_GLOCK_MIN_HOLD);
 411         gl->gl_state = new_state;
 412         gl->gl_tchange = jiffies;
 413 }
 414
 415 static void gfs2_demote_wake(struct gfs2_glock *gl)
 416 {
 417         gl->gl_demote_state = LM_ST_EXCLUSIVE;
 418         clear_bit(GLF_DEMOTE, &gl->gl_flags);
 419         smp_mb__after_clear_bit();
 420         wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
 421 }
 422
 423 /**
 424  * finish_xmote - The DLM has replied to one of our lock requests
 425  * @gl: The glock
 426  * @ret: The status from the DLM
 427  *
 428  */
 429
 430 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
 431 {
 432         const struct gfs2_glock_operations *glops = gl->gl_ops;
 433         struct gfs2_holder *gh;
 434         unsigned state = ret & LM_OUT_ST_MASK;
 435         int rv;
 436
 437         spin_lock(&gl->gl_spin);
 438         trace_gfs2_glock_state_change(gl, state);
 439         state_change(gl, state);
 440         gh = find_first_waiter(gl);
 441
 442         /* Demote to UN request arrived during demote to SH or DF */
 443         if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
 444             state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
 445                 gl->gl_target = LM_ST_UNLOCKED;
 446
 447         /* Check for state != intended state */
 448         if (unlikely(state != gl->gl_target)) {
 449                 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
 450                         /* move to back of queue and try next entry */
 451                         if (ret & LM_OUT_CANCELED) {
 452                                 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
 453                                         list_move_tail(&gh->gh_list, &gl->gl_holders);
 454                                 gh = find_first_waiter(gl);
 455                                 gl->gl_target = gh->gh_state;
 456                                 goto retry;
 457                         }
 458                         /* Some error or failed "try lock" - report it */
 459                         if ((ret & LM_OUT_ERROR) ||
 460                             (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
 461                                 gl->gl_target = gl->gl_state;
 462                                 do_error(gl, ret);
 463                                 goto out;
 464                         }
 465                 }
 466                 switch(state) {
 467                 /* Unlocked due to conversion deadlock, try again */
 468                 case LM_ST_UNLOCKED:
 469 retry:
 470                         do_xmote(gl, gh, gl->gl_target);
 471                         break;
 472                 /* Conversion fails, unlock and try again */
 473                 case LM_ST_SHARED:
 474                 case LM_ST_DEFERRED:
 475                         do_xmote(gl, gh, LM_ST_UNLOCKED);
 476                         break;
 477                 default: /* Everything else */
 478                         printk(KERN_ERR "GFS2: wanted %u got %u\n", gl->gl_target, state);
 479                         GLOCK_BUG_ON(gl, 1);
 480                 }
 481                 spin_unlock(&gl->gl_spin);
 482                 return;
 483         }
 484
 485         /* Fast path - we got what we asked for */
 486         if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
 487                 gfs2_demote_wake(gl);
 488         if (state != LM_ST_UNLOCKED) {
 489                 if (glops->go_xmote_bh) {
 490                         spin_unlock(&gl->gl_spin);
 491                         rv = glops->go_xmote_bh(gl, gh);
 492                         spin_lock(&gl->gl_spin);
 493                         if (rv) {
 494                                 do_error(gl, rv);
 495                                 goto out;
 496                         }
 497                 }
 498                 rv = do_promote(gl);
 499                 if (rv == 2)
 500                         goto out_locked;
 501         }
 502 out:
 503         clear_bit(GLF_LOCK, &gl->gl_flags);
 504 out_locked:
 505         spin_unlock(&gl->gl_spin);
 506 }
 507
 508 /**
 509  * do_xmote - Calls the DLM to change the state of a lock
 510  * @gl: The lock state
 511  * @gh: The holder (only for promotes)
 512  * @target: The target lock state
 513  *
 514  */
 515
 516 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
 517 __releases(&gl->gl_spin)
 518 __acquires(&gl->gl_spin)
 519 {
 520         const struct gfs2_glock_operations *glops = gl->gl_ops;
 521         struct gfs2_sbd *sdp = gl->gl_sbd;
 522         unsigned int lck_flags = gh ? gh->gh_flags : 0;
 523         int ret;
 524
 525         lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
 526                       LM_FLAG_PRIORITY);
 527         GLOCK_BUG_ON(gl, gl->gl_state == target);
 528         GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
 529         if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
 530             glops->go_inval) {
 531                 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 532                 do_error(gl, 0); /* Fail queued try locks */
 533         }
 534         gl->gl_req = target;
 535         set_bit(GLF_BLOCKING, &gl->gl_flags);
 536         if ((gl->gl_req == LM_ST_UNLOCKED) ||
 537             (gl->gl_state == LM_ST_EXCLUSIVE) ||
 538             (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
 539                 clear_bit(GLF_BLOCKING, &gl->gl_flags);
 540         spin_unlock(&gl->gl_spin);
 541         if (glops->go_sync)
 542                 glops->go_sync(gl);
 543         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 544                 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
 545         clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 546
 547         gfs2_glock_hold(gl);
 548         if (sdp->sd_lockstruct.ls_ops->lm_lock) {
 549                 /* lock_dlm */
 550                 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
 551                 if (ret) {
 552                         printk(KERN_ERR "GFS2: lm_lock ret %d\n", ret);
 553                         GLOCK_BUG_ON(gl, 1);
 554                 }
 555         } else { /* lock_nolock */
 556                 finish_xmote(gl, target);
 557                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 558                         gfs2_glock_put(gl);
 559         }
 560
 561         spin_lock(&gl->gl_spin);
 562 }
 563
 564 /**
 565  * find_first_holder - find the first "holder" gh
 566  * @gl: the glock
 567  */
 568
 569 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
 570 {
 571         struct gfs2_holder *gh;
 572
 573         if (!list_empty(&gl->gl_holders)) {
 574                 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
 575                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 576                         return gh;
 577         }
 578         return NULL;
 579 }
 580
 581 /**
 582  * run_queue - do all outstanding tasks related to a glock
 583  * @gl: The glock in question
 584  * @nonblock: True if we must not block in run_queue
 585  *
 586  */
 587
 588 static void run_queue(struct gfs2_glock *gl, const int nonblock)
 589 __releases(&gl->gl_spin)
 590 __acquires(&gl->gl_spin)
 591 {
 592         struct gfs2_holder *gh = NULL;
 593         int ret;
 594
 595         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
 596                 return;
 597
 598         GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
 599
 600         if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
 601             gl->gl_demote_state != gl->gl_state) {
 602                 if (find_first_holder(gl))
 603                         goto out_unlock;
 604                 if (nonblock)
 605                         goto out_sched;
 606                 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
 607                 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
 608                 gl->gl_target = gl->gl_demote_state;
 609         } else {
 610                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
 611                         gfs2_demote_wake(gl);
 612                 ret = do_promote(gl);
 613                 if (ret == 0)
 614                         goto out_unlock;
 615                 if (ret == 2)
 616                         goto out;
 617                 gh = find_first_waiter(gl);
 618                 gl->gl_target = gh->gh_state;
 619                 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
 620                         do_error(gl, 0); /* Fail queued try locks */
 621         }
 622         do_xmote(gl, gh, gl->gl_target);
 623 out:
 624         return;
 625
 626 out_sched:
 627         clear_bit(GLF_LOCK, &gl->gl_flags);
 628         smp_mb__after_clear_bit();
 629         gfs2_glock_hold(gl);
 630         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 631                 gfs2_glock_put_nolock(gl);
 632         return;
 633
 634 out_unlock:
 635         clear_bit(GLF_LOCK, &gl->gl_flags);
 636         smp_mb__after_clear_bit();
 637         return;
 638 }
 639
 640 static void delete_work_func(struct work_struct *work)
 641 {
 642         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
 643         struct gfs2_sbd *sdp = gl->gl_sbd;
 644         struct gfs2_inode *ip;
 645         struct inode *inode;
 646         u64 no_addr = gl->gl_name.ln_number;
 647
 648         ip = gl->gl_object;
 649         /* Note: Unsafe to dereference ip as we don't hold right refs/locks */
 650
 651         if (ip)
 652                 inode = gfs2_ilookup(sdp->sd_vfs, no_addr, 1);
 653         else
 654                 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
 655         if (inode && !IS_ERR(inode)) {
 656                 d_prune_aliases(inode);
 657                 iput(inode);
 658         }
 659         gfs2_glock_put(gl);
 660 }
 661
 662 static void glock_work_func(struct work_struct *work)
 663 {
 664         unsigned long delay = 0;
 665         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
 666         int drop_ref = 0;
 667
 668         if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
 669                 finish_xmote(gl, gl->gl_reply);
 670                 drop_ref = 1;
 671         }
 672         spin_lock(&gl->gl_spin);
 673         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
 674             gl->gl_state != LM_ST_UNLOCKED &&
 675             gl->gl_demote_state != LM_ST_EXCLUSIVE) {
 676                 unsigned long holdtime, now = jiffies;
 677
 678                 holdtime = gl->gl_tchange + gl->gl_hold_time;
 679                 if (time_before(now, holdtime))
 680                         delay = holdtime - now;
 681
 682                 if (!delay) {
 683                         clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
 684                         set_bit(GLF_DEMOTE, &gl->gl_flags);
 685                 }
 686         }
 687         run_queue(gl, 0);
 688         spin_unlock(&gl->gl_spin);
 689         if (!delay)
 690                 gfs2_glock_put(gl);
 691         else {
 692                 if (gl->gl_name.ln_type != LM_TYPE_INODE)
 693                         delay = 0;
 694                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
 695                         gfs2_glock_put(gl);
 696         }
 697         if (drop_ref)
 698                 gfs2_glock_put(gl);
 699 }
 700
 701 /**
 702  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
 703  * @sdp: The GFS2 superblock
 704  * @number: the lock number
 705  * @glops: The glock_operations to use
 706  * @create: If 0, don't create the glock if it doesn't exist
 707  * @glp: the glock is returned here
 708  *
 709  * This does not lock a glock, just finds/creates structures for one.
 710  *
 711  * Returns: errno
 712  */
 713
 714 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
 715                    const struct gfs2_glock_operations *glops, int create,
 716                    struct gfs2_glock **glp)
 717 {
 718         struct super_block *s = sdp->sd_vfs;
 719         struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
 720         struct gfs2_glock *gl, *tmp;
 721         unsigned int hash = gl_hash(sdp, &name);
 722         struct address_space *mapping;
 723         struct kmem_cache *cachep;
 724
 725         rcu_read_lock();
 726         gl = search_bucket(hash, sdp, &name);
 727         rcu_read_unlock();
 728
 729         *glp = gl;
 730         if (gl)
 731                 return 0;
 732         if (!create)
 733                 return -ENOENT;
 734
 735         if (glops->go_flags & GLOF_ASPACE)
 736                 cachep = gfs2_glock_aspace_cachep;
 737         else
 738                 cachep = gfs2_glock_cachep;
 739         gl = kmem_cache_alloc(cachep, GFP_KERNEL);
 740         if (!gl)
 741                 return -ENOMEM;
 742
 743         memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
 744
 745         if (glops->go_flags & GLOF_LVB) {
 746                 gl->gl_lksb.sb_lvbptr = kzalloc(GFS2_MIN_LVB_SIZE, GFP_KERNEL);
 747                 if (!gl->gl_lksb.sb_lvbptr) {
 748                         kmem_cache_free(cachep, gl);
 749                         return -ENOMEM;
 750                 }
 751         }
 752
 753         atomic_inc(&sdp->sd_glock_disposal);
 754         gl->gl_sbd = sdp;
 755         gl->gl_flags = 0;
 756         gl->gl_name = name;
 757         atomic_set(&gl->gl_ref, 1);
 758         gl->gl_state = LM_ST_UNLOCKED;
 759         gl->gl_target = LM_ST_UNLOCKED;
 760         gl->gl_demote_state = LM_ST_EXCLUSIVE;
 761         gl->gl_hash = hash;
 762         gl->gl_ops = glops;
 763         gl->gl_dstamp = ktime_set(0, 0);
 764         preempt_disable();
 765         /* We use the global stats to estimate the initial per-glock stats */
 766         gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
 767         preempt_enable();
 768         gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
 769         gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
 770         gl->gl_tchange = jiffies;
 771         gl->gl_object = NULL;
 772         gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
 773         INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
 774         INIT_WORK(&gl->gl_delete, delete_work_func);
 775
 776         mapping = gfs2_glock2aspace(gl);
 777         if (mapping) {
 778                 mapping->a_ops = &gfs2_meta_aops;
 779                 mapping->host = s->s_bdev->bd_inode;
 780                 mapping->flags = 0;
 781                 mapping_set_gfp_mask(mapping, GFP_NOFS);
 782                 mapping->private_data = NULL;
 783                 mapping->backing_dev_info = s->s_bdi;
 784                 mapping->writeback_index = 0;
 785         }
 786
 787         spin_lock_bucket(hash);
 788         tmp = search_bucket(hash, sdp, &name);
 789         if (tmp) {
 790                 spin_unlock_bucket(hash);
 791                 kfree(gl->gl_lksb.sb_lvbptr);
 792                 kmem_cache_free(cachep, gl);
 793                 atomic_dec(&sdp->sd_glock_disposal);
 794                 gl = tmp;
 795         } else {
 796                 hlist_bl_add_head_rcu(&gl->gl_list, &gl_hash_table[hash]);
 797                 spin_unlock_bucket(hash);
 798         }
 799
 800         *glp = gl;
 801
 802         return 0;
 803 }
 804
 805 /**
 806  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
 807  * @gl: the glock
 808  * @state: the state we're requesting
 809  * @flags: the modifier flags
 810  * @gh: the holder structure
 811  *
 812  */
 813
 814 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
 815                       struct gfs2_holder *gh)
 816 {
 817         INIT_LIST_HEAD(&gh->gh_list);
 818         gh->gh_gl = gl;
 819         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 820         gh->gh_owner_pid = get_pid(task_pid(current));
 821         gh->gh_state = state;
 822         gh->gh_flags = flags;
 823         gh->gh_error = 0;
 824         gh->gh_iflags = 0;
 825         gfs2_glock_hold(gl);
 826 }
 827
 828 /**
 829  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
 830  * @state: the state we're requesting
 831  * @flags: the modifier flags
 832  * @gh: the holder structure
 833  *
 834  * Don't mess with the glock.
 835  *
 836  */
 837
 838 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
 839 {
 840         gh->gh_state = state;
 841         gh->gh_flags = flags;
 842         gh->gh_iflags = 0;
 843         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 844         if (gh->gh_owner_pid)
 845                 put_pid(gh->gh_owner_pid);
 846         gh->gh_owner_pid = get_pid(task_pid(current));
 847 }
 848
 849 /**
 850  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
 851  * @gh: the holder structure
 852  *
 853  */
 854
 855 void gfs2_holder_uninit(struct gfs2_holder *gh)
 856 {
 857         put_pid(gh->gh_owner_pid);
 858         gfs2_glock_put(gh->gh_gl);
 859         gh->gh_gl = NULL;
 860         gh->gh_ip = 0;
 861 }
 862
 863 /**
 864  * gfs2_glock_holder_wait
 865  * @word: unused
 866  *
 867  * This function and gfs2_glock_demote_wait both show up in the WCHAN
 868  * field. Thus I've separated these otherwise identical functions in
 869  * order to be more informative to the user.
 870  */
 871
 872 static int gfs2_glock_holder_wait(void *word)
 873 {
 874         schedule();
 875         return 0;
 876 }
 877
 878 static int gfs2_glock_demote_wait(void *word)
 879 {
 880         schedule();
 881         return 0;
 882 }
 883
 884 /**
 885  * gfs2_glock_wait - wait on a glock acquisition
 886  * @gh: the glock holder
 887  *
 888  * Returns: 0 on success
 889  */
 890
 891 int gfs2_glock_wait(struct gfs2_holder *gh)
 892 {
 893         unsigned long time1 = jiffies;
 894
 895         might_sleep();
 896         wait_on_bit(&gh->gh_iflags, HIF_WAIT, gfs2_glock_holder_wait, TASK_UNINTERRUPTIBLE);
 897         if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
 898                 /* Lengthen the minimum hold time. */
 899                 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
 900                                               GL_GLOCK_HOLD_INCR,
 901                                               GL_GLOCK_MAX_HOLD);
 902         return gh->gh_error;
 903 }
 904
 905 /**
 906  * handle_callback - process a demote request
 907  * @gl: the glock
 908  * @state: the state the caller wants us to change to
 909  *
 910  * There are only two requests that we are going to see in actual
 911  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
 912  */
 913
 914 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
 915                             unsigned long delay, bool remote)
 916 {
 917         int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
 918
 919         set_bit(bit, &gl->gl_flags);
 920         if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
 921                 gl->gl_demote_state = state;
 922                 gl->gl_demote_time = jiffies;
 923         } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
 924                         gl->gl_demote_state != state) {
 925                 gl->gl_demote_state = LM_ST_UNLOCKED;
 926         }
 927         if (gl->gl_ops->go_callback)
 928                 gl->gl_ops->go_callback(gl, remote);
 929         trace_gfs2_demote_rq(gl, remote);
 930 }
 931
 932 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
 933 {
 934         struct va_format vaf;
 935         va_list args;
 936
 937         va_start(args, fmt);
 938
 939         if (seq) {
 940                 seq_vprintf(seq, fmt, args);
 941         } else {
 942                 vaf.fmt = fmt;
 943                 vaf.va = &args;
 944
 945                 printk(KERN_ERR " %pV", &vaf);
 946         }
 947
 948         va_end(args);
 949 }
 950
 951 /**
 952  * add_to_queue - Add a holder to the wait queue (but look for recursion)
 953  * @gh: the holder structure to add
 954  *
 955  * Eventually we should move the recursive locking trap to a
 956  * debugging option or something like that. This is the fast
 957  * path and needs to have the minimum number of distractions.
 958  *
 959  */
 960
 961 static inline void add_to_queue(struct gfs2_holder *gh)
 962 __releases(&gl->gl_spin)
 963 __acquires(&gl->gl_spin)
 964 {
 965         struct gfs2_glock *gl = gh->gh_gl;
 966         struct gfs2_sbd *sdp = gl->gl_sbd;
 967         struct list_head *insert_pt = NULL;
 968         struct gfs2_holder *gh2;
 969         int try_futile = 0;
 970
 971         BUG_ON(gh->gh_owner_pid == NULL);
 972         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
 973                 BUG();
 974
 975         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
 976                 if (test_bit(GLF_LOCK, &gl->gl_flags))
 977                         try_futile = !may_grant(gl, gh);
 978                 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 979                         goto fail;
 980         }
 981
 982         list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
 983                 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
 984                     (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
 985                         goto trap_recursive;
 986                 if (try_futile &&
 987                     !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
 988 fail:
 989                         gh->gh_error = GLR_TRYFAILED;
 990                         gfs2_holder_wake(gh);
 991                         return;
 992                 }
 993                 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
 994                         continue;
 995                 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
 996                         insert_pt = &gh2->gh_list;
 997         }
 998         set_bit(GLF_QUEUED, &gl->gl_flags);
 999         trace_gfs2_glock_queue(gh, 1);
1000         gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1001         gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1002         if (likely(insert_pt == NULL)) {
1003                 list_add_tail(&gh->gh_list, &gl->gl_holders);
1004                 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1005                         goto do_cancel;
1006                 return;
1007         }
1008         list_add_tail(&gh->gh_list, insert_pt);
1009 do_cancel:
1010         gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
1011         if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1012                 spin_unlock(&gl->gl_spin);
1013                 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1014                         sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1015                 spin_lock(&gl->gl_spin);
1016         }
1017         return;
1018
1019 trap_recursive:
1020         printk(KERN_ERR "original: %pSR\n", (void *)gh2->gh_ip);
1021         printk(KERN_ERR "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1022         printk(KERN_ERR "lock type: %d req lock state : %d\n",
1023                gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1024         printk(KERN_ERR "new: %pSR\n", (void *)gh->gh_ip);
1025         printk(KERN_ERR "pid: %d\n", pid_nr(gh->gh_owner_pid));
1026         printk(KERN_ERR "lock type: %d req lock state : %d\n",
1027                gh->gh_gl->gl_name.ln_type, gh->gh_state);
1028         gfs2_dump_glock(NULL, gl);
1029         BUG();
1030 }
1031
1032 /**
1033  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1034  * @gh: the holder structure
1035  *
1036  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1037  *
1038  * Returns: 0, GLR_TRYFAILED, or errno on failure
1039  */
1040
1041 int gfs2_glock_nq(struct gfs2_holder *gh)
1042 {
1043         struct gfs2_glock *gl = gh->gh_gl;
1044         struct gfs2_sbd *sdp = gl->gl_sbd;
1045         int error = 0;
1046
1047         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1048                 return -EIO;
1049
1050         if (test_bit(GLF_LRU, &gl->gl_flags))
1051                 gfs2_glock_remove_from_lru(gl);
1052
1053         spin_lock(&gl->gl_spin);
1054         add_to_queue(gh);
1055         if ((LM_FLAG_NOEXP & gh->gh_flags) &&
1056             test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1057                 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1058         run_queue(gl, 1);
1059         spin_unlock(&gl->gl_spin);
1060
1061         if (!(gh->gh_flags & GL_ASYNC))
1062                 error = gfs2_glock_wait(gh);
1063
1064         return error;
1065 }
1066
1067 /**
1068  * gfs2_glock_poll - poll to see if an async request has been completed
1069  * @gh: the holder
1070  *
1071  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1072  */
1073
1074 int gfs2_glock_poll(struct gfs2_holder *gh)
1075 {
1076         return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1077 }
1078
1079 /**
1080  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1081  * @gh: the glock holder
1082  *
1083  */
1084
1085 void gfs2_glock_dq(struct gfs2_holder *gh)
1086 {
1087         struct gfs2_glock *gl = gh->gh_gl;
1088         const struct gfs2_glock_operations *glops = gl->gl_ops;
1089         unsigned delay = 0;
1090         int fast_path = 0;
1091
1092         spin_lock(&gl->gl_spin);
1093         if (gh->gh_flags & GL_NOCACHE)
1094                 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1095
1096         list_del_init(&gh->gh_list);
1097         if (find_first_holder(gl) == NULL) {
1098                 if (glops->go_unlock) {
1099                         GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1100                         spin_unlock(&gl->gl_spin);
1101                         glops->go_unlock(gh);
1102                         spin_lock(&gl->gl_spin);
1103                         clear_bit(GLF_LOCK, &gl->gl_flags);
1104                 }
1105                 if (list_empty(&gl->gl_holders) &&
1106                     !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1107                     !test_bit(GLF_DEMOTE, &gl->gl_flags))
1108                         fast_path = 1;
1109         }
1110         if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1111                 gfs2_glock_add_to_lru(gl);
1112
1113         trace_gfs2_glock_queue(gh, 0);
1114         spin_unlock(&gl->gl_spin);
1115         if (likely(fast_path))
1116                 return;
1117
1118         gfs2_glock_hold(gl);
1119         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1120             !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1121             gl->gl_name.ln_type == LM_TYPE_INODE)
1122                 delay = gl->gl_hold_time;
1123         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1124                 gfs2_glock_put(gl);
1125 }
1126
1127 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1128 {
1129         struct gfs2_glock *gl = gh->gh_gl;
1130         gfs2_glock_dq(gh);
1131         might_sleep();
1132         wait_on_bit(&gl->gl_flags, GLF_DEMOTE, gfs2_glock_demote_wait, TASK_UNINTERRUPTIBLE);
1133 }
1134
1135 /**
1136  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1137  * @gh: the holder structure
1138  *
1139  */
1140
1141 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1142 {
1143         gfs2_glock_dq(gh);
1144         gfs2_holder_uninit(gh);
1145 }
1146
1147 /**
1148  * gfs2_glock_nq_num - acquire a glock based on lock number
1149  * @sdp: the filesystem
1150  * @number: the lock number
1151  * @glops: the glock operations for the type of glock
1152  * @state: the state to acquire the glock in
1153  * @flags: modifier flags for the acquisition
1154  * @gh: the struct gfs2_holder
1155  *
1156  * Returns: errno
1157  */
1158
1159 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1160                       const struct gfs2_glock_operations *glops,
1161                       unsigned int state, int flags, struct gfs2_holder *gh)
1162 {
1163         struct gfs2_glock *gl;
1164         int error;
1165
1166         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1167         if (!error) {
1168                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1169                 gfs2_glock_put(gl);
1170         }
1171
1172         return error;
1173 }
1174
1175 /**
1176  * glock_compare - Compare two struct gfs2_glock structures for sorting
1177  * @arg_a: the first structure
1178  * @arg_b: the second structure
1179  *
1180  */
1181
1182 static int glock_compare(const void *arg_a, const void *arg_b)
1183 {
1184         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1185         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1186         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1187         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1188
1189         if (a->ln_number > b->ln_number)
1190                 return 1;
1191         if (a->ln_number < b->ln_number)
1192                 return -1;
1193         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1194         return 0;
1195 }
1196
1197 /**
1198  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1199  * @num_gh: the number of structures
1200  * @ghs: an array of struct gfs2_holder structures
1201  *
1202  * Returns: 0 on success (all glocks acquired),
1203  *          errno on failure (no glocks acquired)
1204  */
1205
1206 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1207                      struct gfs2_holder **p)
1208 {
1209         unsigned int x;
1210         int error = 0;
1211
1212         for (x = 0; x < num_gh; x++)
1213                 p[x] = &ghs[x];
1214
1215         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1216
1217         for (x = 0; x < num_gh; x++) {
1218                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1219
1220                 error = gfs2_glock_nq(p[x]);
1221                 if (error) {
1222                         while (x--)
1223                                 gfs2_glock_dq(p[x]);
1224                         break;
1225                 }
1226         }
1227
1228         return error;
1229 }
1230
1231 /**
1232  * gfs2_glock_nq_m - acquire multiple glocks
1233  * @num_gh: the number of structures
1234  * @ghs: an array of struct gfs2_holder structures
1235  *
1236  *
1237  * Returns: 0 on success (all glocks acquired),
1238  *          errno on failure (no glocks acquired)
1239  */
1240
1241 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1242 {
1243         struct gfs2_holder *tmp[4];
1244         struct gfs2_holder **pph = tmp;
1245         int error = 0;
1246
1247         switch(num_gh) {
1248         case 0:
1249                 return 0;
1250         case 1:
1251                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1252                 return gfs2_glock_nq(ghs);
1253         default:
1254                 if (num_gh <= 4)
1255                         break;
1256                 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1257                 if (!pph)
1258                         return -ENOMEM;
1259         }
1260
1261         error = nq_m_sync(num_gh, ghs, pph);
1262
1263         if (pph != tmp)
1264                 kfree(pph);
1265
1266         return error;
1267 }
1268
1269 /**
1270  * gfs2_glock_dq_m - release multiple glocks
1271  * @num_gh: the number of structures
1272  * @ghs: an array of struct gfs2_holder structures
1273  *
1274  */
1275
1276 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1277 {
1278         while (num_gh--)
1279                 gfs2_glock_dq(&ghs[num_gh]);
1280 }
1281
1282 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1283 {
1284         unsigned long delay = 0;
1285         unsigned long holdtime;
1286         unsigned long now = jiffies;
1287
1288         gfs2_glock_hold(gl);
1289         holdtime = gl->gl_tchange + gl->gl_hold_time;
1290         if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1291             gl->gl_name.ln_type == LM_TYPE_INODE) {
1292                 if (time_before(now, holdtime))
1293                         delay = holdtime - now;
1294                 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1295                         delay = gl->gl_hold_time;
1296         }
1297
1298         spin_lock(&gl->gl_spin);
1299         handle_callback(gl, state, delay, true);
1300         spin_unlock(&gl->gl_spin);
1301         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1302                 gfs2_glock_put(gl);
1303 }
1304
1305 /**
1306  * gfs2_should_freeze - Figure out if glock should be frozen
1307  * @gl: The glock in question
1308  *
1309  * Glocks are not frozen if (a) the result of the dlm operation is
1310  * an error, (b) the locking operation was an unlock operation or
1311  * (c) if there is a "noexp" flagged request anywhere in the queue
1312  *
1313  * Returns: 1 if freezing should occur, 0 otherwise
1314  */
1315
1316 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1317 {
1318         const struct gfs2_holder *gh;
1319
1320         if (gl->gl_reply & ~LM_OUT_ST_MASK)
1321                 return 0;
1322         if (gl->gl_target == LM_ST_UNLOCKED)
1323                 return 0;
1324
1325         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1326                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1327                         continue;
1328                 if (LM_FLAG_NOEXP & gh->gh_flags)
1329                         return 0;
1330         }
1331
1332         return 1;
1333 }
1334
1335 /**
1336  * gfs2_glock_complete - Callback used by locking
1337  * @gl: Pointer to the glock
1338  * @ret: The return value from the dlm
1339  *
1340  * The gl_reply field is under the gl_spin lock so that it is ok
1341  * to use a bitfield shared with other glock state fields.
1342  */
1343
1344 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1345 {
1346         struct lm_lockstruct *ls = &gl->gl_sbd->sd_lockstruct;
1347
1348         spin_lock(&gl->gl_spin);
1349         gl->gl_reply = ret;
1350
1351         if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1352                 if (gfs2_should_freeze(gl)) {
1353                         set_bit(GLF_FROZEN, &gl->gl_flags);
1354                         spin_unlock(&gl->gl_spin);
1355                         return;
1356                 }
1357         }
1358
1359         spin_unlock(&gl->gl_spin);
1360         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1361         smp_wmb();
1362         gfs2_glock_hold(gl);
1363         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1364                 gfs2_glock_put(gl);
1365 }
1366
1367 static int glock_cmp(void *priv, struct list_head *a, struct list_head *b)
1368 {
1369         struct gfs2_glock *gla, *glb;
1370
1371         gla = list_entry(a, struct gfs2_glock, gl_lru);
1372         glb = list_entry(b, struct gfs2_glock, gl_lru);
1373
1374         if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1375                 return 1;
1376         if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1377                 return -1;
1378
1379         return 0;
1380 }
1381
1382 /**
1383  * gfs2_dispose_glock_lru - Demote a list of glocks
1384  * @list: The list to dispose of
1385  *
1386  * Disposing of glocks may involve disk accesses, so that here we sort
1387  * the glocks by number (i.e. disk location of the inodes) so that if
1388  * there are any such accesses, they'll be sent in order (mostly).
1389  *
1390  * Must be called under the lru_lock, but may drop and retake this
1391  * lock. While the lru_lock is dropped, entries may vanish from the
1392  * list, but no new entries will appear on the list (since it is
1393  * private)
1394  */
1395
1396 static void gfs2_dispose_glock_lru(struct list_head *list)
1397 __releases(&lru_lock)
1398 __acquires(&lru_lock)
1399 {
1400         struct gfs2_glock *gl;
1401
1402         list_sort(NULL, list, glock_cmp);
1403
1404         while(!list_empty(list)) {
1405                 gl = list_entry(list->next, struct gfs2_glock, gl_lru);
1406                 list_del_init(&gl->gl_lru);
1407                 clear_bit(GLF_LRU, &gl->gl_flags);
1408                 gfs2_glock_hold(gl);
1409                 spin_unlock(&lru_lock);
1410                 spin_lock(&gl->gl_spin);
1411                 if (demote_ok(gl))
1412                         handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1413                 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1414                 smp_mb__after_clear_bit();
1415                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1416                         gfs2_glock_put_nolock(gl);
1417                 spin_unlock(&gl->gl_spin);
1418                 spin_lock(&lru_lock);
1419         }
1420 }
1421
1422 /**
1423  * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1424  * @nr: The number of entries to scan
1425  *
1426  * This function selects the entries on the LRU which are able to
1427  * be demoted, and then kicks off the process by calling
1428  * gfs2_dispose_glock_lru() above.
1429  */
1430
1431 static void gfs2_scan_glock_lru(int nr)
1432 {
1433         struct gfs2_glock *gl;
1434         LIST_HEAD(skipped);
1435         LIST_HEAD(dispose);
1436
1437         spin_lock(&lru_lock);
1438         while(nr && !list_empty(&lru_list)) {
1439                 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1440
1441                 /* Test for being demotable */
1442                 if (!test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1443                         list_move(&gl->gl_lru, &dispose);
1444                         atomic_dec(&lru_count);
1445                         nr--;
1446                         continue;
1447                 }
1448
1449                 list_move(&gl->gl_lru, &skipped);
1450         }
1451         list_splice(&skipped, &lru_list);
1452         if (!list_empty(&dispose))
1453                 gfs2_dispose_glock_lru(&dispose);
1454         spin_unlock(&lru_lock);
1455 }
1456
1457 static int gfs2_shrink_glock_memory(struct shrinker *shrink,
1458                                     struct shrink_control *sc)
1459 {
1460         if (sc->nr_to_scan) {
1461                 if (!(sc->gfp_mask & __GFP_FS))
1462                         return -1;
1463                 gfs2_scan_glock_lru(sc->nr_to_scan);
1464         }
1465
1466         return (atomic_read(&lru_count) / 100) * sysctl_vfs_cache_pressure;
1467 }
1468
1469 static struct shrinker glock_shrinker = {
1470         .shrink = gfs2_shrink_glock_memory,
1471         .seeks = DEFAULT_SEEKS,
1472 };
1473
1474 /**
1475  * examine_bucket - Call a function for glock in a hash bucket
1476  * @examiner: the function
1477  * @sdp: the filesystem
1478  * @bucket: the bucket
1479  *
1480  */
1481
1482 static void examine_bucket(glock_examiner examiner, const struct gfs2_sbd *sdp,
1483                           unsigned int hash)
1484 {
1485         struct gfs2_glock *gl;
1486         struct hlist_bl_head *head = &gl_hash_table[hash];
1487         struct hlist_bl_node *pos;
1488
1489         rcu_read_lock();
1490         hlist_bl_for_each_entry_rcu(gl, pos, head, gl_list) {
1491                 if ((gl->gl_sbd == sdp) && atomic_read(&gl->gl_ref))
1492                         examiner(gl);
1493         }
1494         rcu_read_unlock();
1495         cond_resched();
1496 }
1497
1498 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1499 {
1500         unsigned x;
1501
1502         for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1503                 examine_bucket(examiner, sdp, x);
1504 }
1505
1506
1507 /**
1508  * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1509  * @gl: The glock to thaw
1510  *
1511  * N.B. When we freeze a glock, we leave a ref to the glock outstanding,
1512  * so this has to result in the ref count being dropped by one.
1513  */
1514
1515 static void thaw_glock(struct gfs2_glock *gl)
1516 {
1517         if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1518                 return;
1519         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1520         gfs2_glock_hold(gl);
1521         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1522                 gfs2_glock_put(gl);
1523 }
1524
1525 /**
1526  * clear_glock - look at a glock and see if we can free it from glock cache
1527  * @gl: the glock to look at
1528  *
1529  */
1530
1531 static void clear_glock(struct gfs2_glock *gl)
1532 {
1533         gfs2_glock_remove_from_lru(gl);
1534
1535         spin_lock(&gl->gl_spin);
1536         if (gl->gl_state != LM_ST_UNLOCKED)
1537                 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1538         spin_unlock(&gl->gl_spin);
1539         gfs2_glock_hold(gl);
1540         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1541                 gfs2_glock_put(gl);
1542 }
1543
1544 /**
1545  * gfs2_glock_thaw - Thaw any frozen glocks
1546  * @sdp: The super block
1547  *
1548  */
1549
1550 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1551 {
1552         glock_hash_walk(thaw_glock, sdp);
1553 }
1554
1555 static int dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1556 {
1557         int ret;
1558         spin_lock(&gl->gl_spin);
1559         ret = gfs2_dump_glock(seq, gl);
1560         spin_unlock(&gl->gl_spin);
1561         return ret;
1562 }
1563
1564 static void dump_glock_func(struct gfs2_glock *gl)
1565 {
1566         dump_glock(NULL, gl);
1567 }
1568
1569 /**
1570  * gfs2_gl_hash_clear - Empty out the glock hash table
1571  * @sdp: the filesystem
1572  * @wait: wait until it's all gone
1573  *
1574  * Called when unmounting the filesystem.
1575  */
1576
1577 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1578 {
1579         set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1580         flush_workqueue(glock_workqueue);
1581         glock_hash_walk(clear_glock, sdp);
1582         flush_workqueue(glock_workqueue);
1583         wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
1584         glock_hash_walk(dump_glock_func, sdp);
1585 }
1586
1587 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1588 {
1589         struct gfs2_glock *gl = ip->i_gl;
1590         int ret;
1591
1592         ret = gfs2_truncatei_resume(ip);
1593         gfs2_assert_withdraw(gl->gl_sbd, ret == 0);
1594
1595         spin_lock(&gl->gl_spin);
1596         clear_bit(GLF_LOCK, &gl->gl_flags);
1597         run_queue(gl, 1);
1598         spin_unlock(&gl->gl_spin);
1599 }
1600
1601 static const char *state2str(unsigned state)
1602 {
1603         switch(state) {
1604         case LM_ST_UNLOCKED:
1605                 return "UN";
1606         case LM_ST_SHARED:
1607                 return "SH";
1608         case LM_ST_DEFERRED:
1609                 return "DF";
1610         case LM_ST_EXCLUSIVE:
1611                 return "EX";
1612         }
1613         return "??";
1614 }
1615
1616 static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1617 {
1618         char *p = buf;
1619         if (flags & LM_FLAG_TRY)
1620                 *p++ = 't';
1621         if (flags & LM_FLAG_TRY_1CB)
1622                 *p++ = 'T';
1623         if (flags & LM_FLAG_NOEXP)
1624                 *p++ = 'e';
1625         if (flags & LM_FLAG_ANY)
1626                 *p++ = 'A';
1627         if (flags & LM_FLAG_PRIORITY)
1628                 *p++ = 'p';
1629         if (flags & GL_ASYNC)
1630                 *p++ = 'a';
1631         if (flags & GL_EXACT)
1632                 *p++ = 'E';
1633         if (flags & GL_NOCACHE)
1634                 *p++ = 'c';
1635         if (test_bit(HIF_HOLDER, &iflags))
1636                 *p++ = 'H';
1637         if (test_bit(HIF_WAIT, &iflags))
1638                 *p++ = 'W';
1639         if (test_bit(HIF_FIRST, &iflags))
1640                 *p++ = 'F';
1641         *p = 0;
1642         return buf;
1643 }
1644
1645 /**
1646  * dump_holder - print information about a glock holder
1647  * @seq: the seq_file struct
1648  * @gh: the glock holder
1649  *
1650  * Returns: 0 on success, -ENOBUFS when we run out of space
1651  */
1652
1653 static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1654 {
1655         struct task_struct *gh_owner = NULL;
1656         char flags_buf[32];
1657
1658         if (gh->gh_owner_pid)
1659                 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1660         gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1661                        state2str(gh->gh_state),
1662                        hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1663                        gh->gh_error,
1664                        gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1665                        gh_owner ? gh_owner->comm : "(ended)",
1666                        (void *)gh->gh_ip);
1667         return 0;
1668 }
1669
1670 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1671 {
1672         const unsigned long *gflags = &gl->gl_flags;
1673         char *p = buf;
1674
1675         if (test_bit(GLF_LOCK, gflags))
1676                 *p++ = 'l';
1677         if (test_bit(GLF_DEMOTE, gflags))
1678                 *p++ = 'D';
1679         if (test_bit(GLF_PENDING_DEMOTE, gflags))
1680                 *p++ = 'd';
1681         if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1682                 *p++ = 'p';
1683         if (test_bit(GLF_DIRTY, gflags))
1684                 *p++ = 'y';
1685         if (test_bit(GLF_LFLUSH, gflags))
1686                 *p++ = 'f';
1687         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1688                 *p++ = 'i';
1689         if (test_bit(GLF_REPLY_PENDING, gflags))
1690                 *p++ = 'r';
1691         if (test_bit(GLF_INITIAL, gflags))
1692                 *p++ = 'I';
1693         if (test_bit(GLF_FROZEN, gflags))
1694                 *p++ = 'F';
1695         if (test_bit(GLF_QUEUED, gflags))
1696                 *p++ = 'q';
1697         if (test_bit(GLF_LRU, gflags))
1698                 *p++ = 'L';
1699         if (gl->gl_object)
1700                 *p++ = 'o';
1701         if (test_bit(GLF_BLOCKING, gflags))
1702                 *p++ = 'b';
1703         *p = 0;
1704         return buf;
1705 }
1706
1707 /**
1708  * gfs2_dump_glock - print information about a glock
1709  * @seq: The seq_file struct
1710  * @gl: the glock
1711  *
1712  * The file format is as follows:
1713  * One line per object, capital letters are used to indicate objects
1714  * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1715  * other objects are indented by a single space and follow the glock to
1716  * which they are related. Fields are indicated by lower case letters
1717  * followed by a colon and the field value, except for strings which are in
1718  * [] so that its possible to see if they are composed of spaces for
1719  * example. The field's are n = number (id of the object), f = flags,
1720  * t = type, s = state, r = refcount, e = error, p = pid.
1721  *
1722  * Returns: 0 on success, -ENOBUFS when we run out of space
1723  */
1724
1725 int gfs2_dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1726 {
1727         const struct gfs2_glock_operations *glops = gl->gl_ops;
1728         unsigned long long dtime;
1729         const struct gfs2_holder *gh;
1730         char gflags_buf[32];
1731         int error = 0;
1732
1733         dtime = jiffies - gl->gl_demote_time;
1734         dtime *= 1000000/HZ; /* demote time in uSec */
1735         if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1736                 dtime = 0;
1737         gfs2_print_dbg(seq, "G:  s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d v:%d r:%d m:%ld\n",
1738                   state2str(gl->gl_state),
1739                   gl->gl_name.ln_type,
1740                   (unsigned long long)gl->gl_name.ln_number,
1741                   gflags2str(gflags_buf, gl),
1742                   state2str(gl->gl_target),
1743                   state2str(gl->gl_demote_state), dtime,
1744                   atomic_read(&gl->gl_ail_count),
1745                   atomic_read(&gl->gl_revokes),
1746                   atomic_read(&gl->gl_ref), gl->gl_hold_time);
1747
1748         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1749                 error = dump_holder(seq, gh);
1750                 if (error)
1751                         goto out;
1752         }
1753         if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1754                 error = glops->go_dump(seq, gl);
1755 out:
1756         return error;
1757 }
1758
1759 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
1760 {
1761         struct gfs2_glock *gl = iter_ptr;
1762
1763         seq_printf(seq, "G: n:%u/%llx rtt:%lld/%lld rttb:%lld/%lld irt:%lld/%lld dcnt: %lld qcnt: %lld\n",
1764                    gl->gl_name.ln_type,
1765                    (unsigned long long)gl->gl_name.ln_number,
1766                    (long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
1767                    (long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
1768                    (long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
1769                    (long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
1770                    (long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
1771                    (long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
1772                    (long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
1773                    (long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
1774         return 0;
1775 }
1776
1777 static const char *gfs2_gltype[] = {
1778         "type",
1779         "reserved",
1780         "nondisk",
1781         "inode",
1782         "rgrp",
1783         "meta",
1784         "iopen",
1785         "flock",
1786         "plock",
1787         "quota",
1788         "journal",
1789 };
1790
1791 static const char *gfs2_stype[] = {
1792         [GFS2_LKS_SRTT]         = "srtt",
1793         [GFS2_LKS_SRTTVAR]      = "srttvar",
1794         [GFS2_LKS_SRTTB]        = "srttb",
1795         [GFS2_LKS_SRTTVARB]     = "srttvarb",
1796         [GFS2_LKS_SIRT]         = "sirt",
1797         [GFS2_LKS_SIRTVAR]      = "sirtvar",
1798         [GFS2_LKS_DCOUNT]       = "dlm",
1799         [GFS2_LKS_QCOUNT]       = "queue",
1800 };
1801
1802 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1803
1804 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
1805 {
1806         struct gfs2_glock_iter *gi = seq->private;
1807         struct gfs2_sbd *sdp = gi->sdp;
1808         unsigned index = gi->hash >> 3;
1809         unsigned subindex = gi->hash & 0x07;
1810         s64 value;
1811         int i;
1812
1813         if (index == 0 && subindex != 0)
1814                 return 0;
1815
1816         seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
1817                    (index == 0) ? "cpu": gfs2_stype[subindex]);
1818
1819         for_each_possible_cpu(i) {
1820                 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
1821                 if (index == 0) {
1822                         value = i;
1823                 } else {
1824                         value = lkstats->lkstats[index - 1].stats[subindex];
1825                 }
1826                 seq_printf(seq, " %15lld", (long long)value);
1827         }
1828         seq_putc(seq, '\n');
1829         return 0;
1830 }
1831
1832 int __init gfs2_glock_init(void)
1833 {
1834         unsigned i;
1835         for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
1836                 INIT_HLIST_BL_HEAD(&gl_hash_table[i]);
1837         }
1838
1839         glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1840                                           WQ_HIGHPRI | WQ_FREEZABLE, 0);
1841         if (IS_ERR(glock_workqueue))
1842                 return PTR_ERR(glock_workqueue);
1843         gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1844                                                 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1845                                                 0);
1846         if (IS_ERR(gfs2_delete_workqueue)) {
1847                 destroy_workqueue(glock_workqueue);
1848                 return PTR_ERR(gfs2_delete_workqueue);
1849         }
1850
1851         register_shrinker(&glock_shrinker);
1852
1853         return 0;
1854 }
1855
1856 void gfs2_glock_exit(void)
1857 {
1858         unregister_shrinker(&glock_shrinker);
1859         destroy_workqueue(glock_workqueue);
1860         destroy_workqueue(gfs2_delete_workqueue);
1861 }
1862
1863 static inline struct gfs2_glock *glock_hash_chain(unsigned hash)
1864 {
1865         return hlist_bl_entry(hlist_bl_first_rcu(&gl_hash_table[hash]),
1866                               struct gfs2_glock, gl_list);
1867 }
1868
1869 static inline struct gfs2_glock *glock_hash_next(struct gfs2_glock *gl)
1870 {
1871         return hlist_bl_entry(rcu_dereference(gl->gl_list.next),
1872                               struct gfs2_glock, gl_list);
1873 }
1874
1875 static int gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1876 {
1877         struct gfs2_glock *gl;
1878
1879         do {
1880                 gl = gi->gl;
1881                 if (gl) {
1882                         gi->gl = glock_hash_next(gl);
1883                         gi->nhash++;
1884                 } else {
1885                         if (gi->hash >= GFS2_GL_HASH_SIZE) {
1886                                 rcu_read_unlock();
1887                                 return 1;
1888                         }
1889                         gi->gl = glock_hash_chain(gi->hash);
1890                         gi->nhash = 0;
1891                 }
1892                 while (gi->gl == NULL) {
1893                         gi->hash++;
1894                         if (gi->hash >= GFS2_GL_HASH_SIZE) {
1895                                 rcu_read_unlock();
1896                                 return 1;
1897                         }
1898                         gi->gl = glock_hash_chain(gi->hash);
1899                         gi->nhash = 0;
1900                 }
1901         /* Skip entries for other sb and dead entries */
1902         } while (gi->sdp != gi->gl->gl_sbd || atomic_read(&gi->gl->gl_ref) == 0);
1903
1904         return 0;
1905 }
1906
1907 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1908 {
1909         struct gfs2_glock_iter *gi = seq->private;
1910         loff_t n = *pos;
1911
1912         if (gi->last_pos <= *pos)
1913                 n = gi->nhash + (*pos - gi->last_pos);
1914         else
1915                 gi->hash = 0;
1916
1917         gi->nhash = 0;
1918         rcu_read_lock();
1919
1920         do {
1921                 if (gfs2_glock_iter_next(gi))
1922                         return NULL;
1923         } while (n--);
1924
1925         gi->last_pos = *pos;
1926         return gi->gl;
1927 }
1928
1929 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1930                                  loff_t *pos)
1931 {
1932         struct gfs2_glock_iter *gi = seq->private;
1933
1934         (*pos)++;
1935         gi->last_pos = *pos;
1936         if (gfs2_glock_iter_next(gi))
1937                 return NULL;
1938
1939         return gi->gl;
1940 }
1941
1942 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1943 {
1944         struct gfs2_glock_iter *gi = seq->private;
1945
1946         if (gi->gl)
1947                 rcu_read_unlock();
1948         gi->gl = NULL;
1949 }
1950
1951 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1952 {
1953         return dump_glock(seq, iter_ptr);
1954 }
1955
1956 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
1957 {
1958         struct gfs2_glock_iter *gi = seq->private;
1959
1960         gi->hash = *pos;
1961         if (*pos >= GFS2_NR_SBSTATS)
1962                 return NULL;
1963         preempt_disable();
1964         return SEQ_START_TOKEN;
1965 }
1966
1967 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
1968                                    loff_t *pos)
1969 {
1970         struct gfs2_glock_iter *gi = seq->private;
1971         (*pos)++;
1972         gi->hash++;
1973         if (gi->hash >= GFS2_NR_SBSTATS) {
1974                 preempt_enable();
1975                 return NULL;
1976         }
1977         return SEQ_START_TOKEN;
1978 }
1979
1980 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
1981 {
1982         preempt_enable();
1983 }
1984
1985 static const struct seq_operations gfs2_glock_seq_ops = {
1986         .start = gfs2_glock_seq_start,
1987         .next  = gfs2_glock_seq_next,
1988         .stop  = gfs2_glock_seq_stop,
1989         .show  = gfs2_glock_seq_show,
1990 };
1991
1992 static const struct seq_operations gfs2_glstats_seq_ops = {
1993         .start = gfs2_glock_seq_start,
1994         .next  = gfs2_glock_seq_next,
1995         .stop  = gfs2_glock_seq_stop,
1996         .show  = gfs2_glstats_seq_show,
1997 };
1998
1999 static const struct seq_operations gfs2_sbstats_seq_ops = {
2000         .start = gfs2_sbstats_seq_start,
2001         .next  = gfs2_sbstats_seq_next,
2002         .stop  = gfs2_sbstats_seq_stop,
2003         .show  = gfs2_sbstats_seq_show,
2004 };
2005
2006 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2007
2008 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2009 {
2010         int ret = seq_open_private(file, &gfs2_glock_seq_ops,
2011                                    sizeof(struct gfs2_glock_iter));
2012         if (ret == 0) {
2013                 struct seq_file *seq = file->private_data;
2014                 struct gfs2_glock_iter *gi = seq->private;
2015                 gi->sdp = inode->i_private;
2016                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2017                 if (seq->buf)
2018                         seq->size = GFS2_SEQ_GOODSIZE;
2019         }
2020         return ret;
2021 }
2022
2023 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2024 {
2025         int ret = seq_open_private(file, &gfs2_glstats_seq_ops,
2026                                    sizeof(struct gfs2_glock_iter));
2027         if (ret == 0) {
2028                 struct seq_file *seq = file->private_data;
2029                 struct gfs2_glock_iter *gi = seq->private;
2030                 gi->sdp = inode->i_private;
2031                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2032                 if (seq->buf)
2033                         seq->size = GFS2_SEQ_GOODSIZE;
2034         }
2035         return ret;
2036 }
2037
2038 static int gfs2_sbstats_open(struct inode *inode, struct file *file)
2039 {
2040         int ret = seq_open_private(file, &gfs2_sbstats_seq_ops,
2041                                    sizeof(struct gfs2_glock_iter));
2042         if (ret == 0) {
2043                 struct seq_file *seq = file->private_data;
2044                 struct gfs2_glock_iter *gi = seq->private;
2045                 gi->sdp = inode->i_private;
2046         }
2047         return ret;
2048 }
2049
2050 static const struct file_operations gfs2_glocks_fops = {
2051         .owner   = THIS_MODULE,
2052         .open    = gfs2_glocks_open,
2053         .read    = seq_read,
2054         .llseek  = seq_lseek,
2055         .release = seq_release_private,
2056 };
2057
2058 static const struct file_operations gfs2_glstats_fops = {
2059         .owner   = THIS_MODULE,
2060         .open    = gfs2_glstats_open,
2061         .read    = seq_read,
2062         .llseek  = seq_lseek,
2063         .release = seq_release_private,
2064 };
2065
2066 static const struct file_operations gfs2_sbstats_fops = {
2067         .owner   = THIS_MODULE,
2068         .open    = gfs2_sbstats_open,
2069         .read    = seq_read,
2070         .llseek  = seq_lseek,
2071         .release = seq_release_private,
2072 };
2073
2074 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2075 {
2076         sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2077         if (!sdp->debugfs_dir)
2078                 return -ENOMEM;
2079         sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
2080                                                          S_IFREG | S_IRUGO,
2081                                                          sdp->debugfs_dir, sdp,
2082                                                          &gfs2_glocks_fops);
2083         if (!sdp->debugfs_dentry_glocks)
2084                 goto fail;
2085
2086         sdp->debugfs_dentry_glstats = debugfs_create_file("glstats",
2087                                                         S_IFREG | S_IRUGO,
2088                                                         sdp->debugfs_dir, sdp,
2089                                                         &gfs2_glstats_fops);
2090         if (!sdp->debugfs_dentry_glstats)
2091                 goto fail;
2092
2093         sdp->debugfs_dentry_sbstats = debugfs_create_file("sbstats",
2094                                                         S_IFREG | S_IRUGO,
2095                                                         sdp->debugfs_dir, sdp,
2096                                                         &gfs2_sbstats_fops);
2097         if (!sdp->debugfs_dentry_sbstats)
2098                 goto fail;
2099
2100         return 0;
2101 fail:
2102         gfs2_delete_debugfs_file(sdp);
2103         return -ENOMEM;
2104 }
2105
2106 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2107 {
2108         if (sdp->debugfs_dir) {
2109                 if (sdp->debugfs_dentry_glocks) {
2110                         debugfs_remove(sdp->debugfs_dentry_glocks);
2111                         sdp->debugfs_dentry_glocks = NULL;
2112                 }
2113                 if (sdp->debugfs_dentry_glstats) {
2114                         debugfs_remove(sdp->debugfs_dentry_glstats);
2115                         sdp->debugfs_dentry_glstats = NULL;
2116                 }
2117                 if (sdp->debugfs_dentry_sbstats) {
2118                         debugfs_remove(sdp->debugfs_dentry_sbstats);
2119                         sdp->debugfs_dentry_sbstats = NULL;
2120                 }
2121                 debugfs_remove(sdp->debugfs_dir);
2122                 sdp->debugfs_dir = NULL;
2123         }
2124 }
2125
2126 int gfs2_register_debugfs(void)
2127 {
2128         gfs2_root = debugfs_create_dir("gfs2", NULL);
2129         return gfs2_root ? 0 : -ENOMEM;
2130 }
2131
2132 void gfs2_unregister_debugfs(void)
2133 {
2134         debugfs_remove(gfs2_root);
2135         gfs2_root = NULL;
2136 }