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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  11
  12 #include <linux/sched.h>
  13 #include <linux/slab.h>
  14 #include <linux/spinlock.h>
  15 #include <linux/buffer_head.h>
  16 #include <linux/delay.h>
  17 #include <linux/sort.h>
  18 #include <linux/jhash.h>
  19 #include <linux/kallsyms.h>
  20 #include <linux/gfs2_ondisk.h>
  21 #include <linux/list.h>
  22 #include <linux/wait.h>
  23 #include <linux/module.h>
  24 #include <asm/uaccess.h>
  25 #include <linux/seq_file.h>
  26 #include <linux/debugfs.h>
  27 #include <linux/kthread.h>
  28 #include <linux/freezer.h>
  29 #include <linux/workqueue.h>
  30 #include <linux/jiffies.h>
  31 #include <linux/rcupdate.h>
  32 #include <linux/rculist_bl.h>
  33 #include <linux/bit_spinlock.h>
  34 #include <linux/percpu.h>
  35 #include <linux/list_sort.h>
  36 #include <linux/lockref.h>
  37
  38 #include "gfs2.h"
  39 #include "incore.h"
  40 #include "glock.h"
  41 #include "glops.h"
  42 #include "inode.h"
  43 #include "lops.h"
  44 #include "meta_io.h"
  45 #include "quota.h"
  46 #include "super.h"
  47 #include "util.h"
  48 #include "bmap.h"
  49 #define CREATE_TRACE_POINTS
  50 #include "trace_gfs2.h"
  51
  52 struct gfs2_glock_iter {
  53         int hash;                       /* hash bucket index           */
  54         unsigned nhash;                 /* Index within current bucket */
  55         struct gfs2_sbd *sdp;           /* incore superblock           */
  56         struct gfs2_glock *gl;          /* current glock struct        */
  57         loff_t last_pos;                /* last position               */
  58 };
  59
  60 typedef void (*glock_examiner) (struct gfs2_glock * gl);
  61
  62 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
  63
  64 static struct dentry *gfs2_root;
  65 static struct workqueue_struct *glock_workqueue;
  66 struct workqueue_struct *gfs2_delete_workqueue;
  67 static LIST_HEAD(lru_list);
  68 static atomic_t lru_count = ATOMIC_INIT(0);
  69 static DEFINE_SPINLOCK(lru_lock);
  70
  71 #define GFS2_GL_HASH_SHIFT      15
  72 #define GFS2_GL_HASH_SIZE       (1 << GFS2_GL_HASH_SHIFT)
  73 #define GFS2_GL_HASH_MASK       (GFS2_GL_HASH_SIZE - 1)
  74
  75 static struct hlist_bl_head gl_hash_table[GFS2_GL_HASH_SIZE];
  76 static struct dentry *gfs2_root;
  77
  78 /**
  79  * gl_hash() - Turn glock number into hash bucket number
  80  * @lock: The glock number
  81  *
  82  * Returns: The number of the corresponding hash bucket
  83  */
  84
  85 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
  86                             const struct lm_lockname *name)
  87 {
  88         unsigned int h;
  89
  90         h = jhash(&name->ln_number, sizeof(u64), 0);
  91         h = jhash(&name->ln_type, sizeof(unsigned int), h);
  92         h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
  93         h &= GFS2_GL_HASH_MASK;
  94
  95         return h;
  96 }
  97
  98 static inline void spin_lock_bucket(unsigned int hash)
  99 {
 100         hlist_bl_lock(&gl_hash_table[hash]);
 101 }
 102
 103 static inline void spin_unlock_bucket(unsigned int hash)
 104 {
 105         hlist_bl_unlock(&gl_hash_table[hash]);
 106 }
 107
 108 static void gfs2_glock_dealloc(struct rcu_head *rcu)
 109 {
 110         struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
 111
 112         if (gl->gl_ops->go_flags & GLOF_ASPACE) {
 113                 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
 114         } else {
 115                 kfree(gl->gl_lksb.sb_lvbptr);
 116                 kmem_cache_free(gfs2_glock_cachep, gl);
 117         }
 118 }
 119
 120 void gfs2_glock_free(struct gfs2_glock *gl)
 121 {
 122         struct gfs2_sbd *sdp = gl->gl_sbd;
 123
 124         call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
 125         if (atomic_dec_and_test(&sdp->sd_glock_disposal))
 126                 wake_up(&sdp->sd_glock_wait);
 127 }
 128
 129 /**
 130  * gfs2_glock_hold() - increment reference count on glock
 131  * @gl: The glock to hold
 132  *
 133  */
 134
 135 static void gfs2_glock_hold(struct gfs2_glock *gl)
 136 {
 137         GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
 138         lockref_get(&gl->gl_lockref);
 139 }
 140
 141 /**
 142  * demote_ok - Check to see if it's ok to unlock a glock
 143  * @gl: the glock
 144  *
 145  * Returns: 1 if it's ok
 146  */
 147
 148 static int demote_ok(const struct gfs2_glock *gl)
 149 {
 150         const struct gfs2_glock_operations *glops = gl->gl_ops;
 151
 152         if (gl->gl_state == LM_ST_UNLOCKED)
 153                 return 0;
 154         if (!list_empty(&gl->gl_holders))
 155                 return 0;
 156         if (glops->go_demote_ok)
 157                 return glops->go_demote_ok(gl);
 158         return 1;
 159 }
 160
 161
 162 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
 163 {
 164         spin_lock(&lru_lock);
 165
 166         if (!list_empty(&gl->gl_lru))
 167                 list_del_init(&gl->gl_lru);
 168         else
 169                 atomic_inc(&lru_count);
 170
 171         list_add_tail(&gl->gl_lru, &lru_list);
 172         set_bit(GLF_LRU, &gl->gl_flags);
 173         spin_unlock(&lru_lock);
 174 }
 175
 176 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
 177 {
 178         spin_lock(&lru_lock);
 179         if (!list_empty(&gl->gl_lru)) {
 180                 list_del_init(&gl->gl_lru);
 181                 atomic_dec(&lru_count);
 182                 clear_bit(GLF_LRU, &gl->gl_flags);
 183         }
 184         spin_unlock(&lru_lock);
 185 }
 186
 187 /**
 188  * gfs2_glock_put() - Decrement reference count on glock
 189  * @gl: The glock to put
 190  *
 191  */
 192
 193 void gfs2_glock_put(struct gfs2_glock *gl)
 194 {
 195         struct gfs2_sbd *sdp = gl->gl_sbd;
 196         struct address_space *mapping = gfs2_glock2aspace(gl);
 197
 198         if (lockref_put_or_lock(&gl->gl_lockref))
 199                 return;
 200
 201         lockref_mark_dead(&gl->gl_lockref);
 202
 203         gfs2_glock_remove_from_lru(gl);
 204         spin_unlock(&gl->gl_lockref.lock);
 205         spin_lock_bucket(gl->gl_hash);
 206         hlist_bl_del_rcu(&gl->gl_list);
 207         spin_unlock_bucket(gl->gl_hash);
 208         GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
 209         GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
 210         trace_gfs2_glock_put(gl);
 211         sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
 212 }
 213
 214 /**
 215  * search_bucket() - Find struct gfs2_glock by lock number
 216  * @bucket: the bucket to search
 217  * @name: The lock name
 218  *
 219  * Returns: NULL, or the struct gfs2_glock with the requested number
 220  */
 221
 222 static struct gfs2_glock *search_bucket(unsigned int hash,
 223                                         const struct gfs2_sbd *sdp,
 224                                         const struct lm_lockname *name)
 225 {
 226         struct gfs2_glock *gl;
 227         struct hlist_bl_node *h;
 228
 229         hlist_bl_for_each_entry_rcu(gl, h, &gl_hash_table[hash], gl_list) {
 230                 if (!lm_name_equal(&gl->gl_name, name))
 231                         continue;
 232                 if (gl->gl_sbd != sdp)
 233                         continue;
 234                 if (lockref_get_not_dead(&gl->gl_lockref))
 235                         return gl;
 236         }
 237
 238         return NULL;
 239 }
 240
 241 /**
 242  * may_grant - check if its ok to grant a new lock
 243  * @gl: The glock
 244  * @gh: The lock request which we wish to grant
 245  *
 246  * Returns: true if its ok to grant the lock
 247  */
 248
 249 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
 250 {
 251         const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
 252         if ((gh->gh_state == LM_ST_EXCLUSIVE ||
 253              gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
 254                 return 0;
 255         if (gl->gl_state == gh->gh_state)
 256                 return 1;
 257         if (gh->gh_flags & GL_EXACT)
 258                 return 0;
 259         if (gl->gl_state == LM_ST_EXCLUSIVE) {
 260                 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
 261                         return 1;
 262                 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
 263                         return 1;
 264         }
 265         if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
 266                 return 1;
 267         return 0;
 268 }
 269
 270 static void gfs2_holder_wake(struct gfs2_holder *gh)
 271 {
 272         clear_bit(HIF_WAIT, &gh->gh_iflags);
 273         smp_mb__after_atomic();
 274         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
 275 }
 276
 277 /**
 278  * do_error - Something unexpected has happened during a lock request
 279  *
 280  */
 281
 282 static inline void do_error(struct gfs2_glock *gl, const int ret)
 283 {
 284         struct gfs2_holder *gh, *tmp;
 285
 286         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
 287                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 288                         continue;
 289                 if (ret & LM_OUT_ERROR)
 290                         gh->gh_error = -EIO;
 291                 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
 292                         gh->gh_error = GLR_TRYFAILED;
 293                 else
 294                         continue;
 295                 list_del_init(&gh->gh_list);
 296                 trace_gfs2_glock_queue(gh, 0);
 297                 gfs2_holder_wake(gh);
 298         }
 299 }
 300
 301 /**
 302  * do_promote - promote as many requests as possible on the current queue
 303  * @gl: The glock
 304  *
 305  * Returns: 1 if there is a blocked holder at the head of the list, or 2
 306  *          if a type specific operation is underway.
 307  */
 308
 309 static int do_promote(struct gfs2_glock *gl)
 310 __releases(&gl->gl_spin)
 311 __acquires(&gl->gl_spin)
 312 {
 313         const struct gfs2_glock_operations *glops = gl->gl_ops;
 314         struct gfs2_holder *gh, *tmp;
 315         int ret;
 316
 317 restart:
 318         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
 319                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 320                         continue;
 321                 if (may_grant(gl, gh)) {
 322                         if (gh->gh_list.prev == &gl->gl_holders &&
 323                             glops->go_lock) {
 324                                 spin_unlock(&gl->gl_spin);
 325                                 /* FIXME: eliminate this eventually */
 326                                 ret = glops->go_lock(gh);
 327                                 spin_lock(&gl->gl_spin);
 328                                 if (ret) {
 329                                         if (ret == 1)
 330                                                 return 2;
 331                                         gh->gh_error = ret;
 332                                         list_del_init(&gh->gh_list);
 333                                         trace_gfs2_glock_queue(gh, 0);
 334                                         gfs2_holder_wake(gh);
 335                                         goto restart;
 336                                 }
 337                                 set_bit(HIF_HOLDER, &gh->gh_iflags);
 338                                 trace_gfs2_promote(gh, 1);
 339                                 gfs2_holder_wake(gh);
 340                                 goto restart;
 341                         }
 342                         set_bit(HIF_HOLDER, &gh->gh_iflags);
 343                         trace_gfs2_promote(gh, 0);
 344                         gfs2_holder_wake(gh);
 345                         continue;
 346                 }
 347                 if (gh->gh_list.prev == &gl->gl_holders)
 348                         return 1;
 349                 do_error(gl, 0);
 350                 break;
 351         }
 352         return 0;
 353 }
 354
 355 /**
 356  * find_first_waiter - find the first gh that's waiting for the glock
 357  * @gl: the glock
 358  */
 359
 360 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
 361 {
 362         struct gfs2_holder *gh;
 363
 364         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
 365                 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
 366                         return gh;
 367         }
 368         return NULL;
 369 }
 370
 371 /**
 372  * state_change - record that the glock is now in a different state
 373  * @gl: the glock
 374  * @new_state the new state
 375  *
 376  */
 377
 378 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
 379 {
 380         int held1, held2;
 381
 382         held1 = (gl->gl_state != LM_ST_UNLOCKED);
 383         held2 = (new_state != LM_ST_UNLOCKED);
 384
 385         if (held1 != held2) {
 386                 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
 387                 if (held2)
 388                         gl->gl_lockref.count++;
 389                 else
 390                         gl->gl_lockref.count--;
 391         }
 392         if (held1 && held2 && list_empty(&gl->gl_holders))
 393                 clear_bit(GLF_QUEUED, &gl->gl_flags);
 394
 395         if (new_state != gl->gl_target)
 396                 /* shorten our minimum hold time */
 397                 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
 398                                        GL_GLOCK_MIN_HOLD);
 399         gl->gl_state = new_state;
 400         gl->gl_tchange = jiffies;
 401 }
 402
 403 static void gfs2_demote_wake(struct gfs2_glock *gl)
 404 {
 405         gl->gl_demote_state = LM_ST_EXCLUSIVE;
 406         clear_bit(GLF_DEMOTE, &gl->gl_flags);
 407         smp_mb__after_atomic();
 408         wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
 409 }
 410
 411 /**
 412  * finish_xmote - The DLM has replied to one of our lock requests
 413  * @gl: The glock
 414  * @ret: The status from the DLM
 415  *
 416  */
 417
 418 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
 419 {
 420         const struct gfs2_glock_operations *glops = gl->gl_ops;
 421         struct gfs2_holder *gh;
 422         unsigned state = ret & LM_OUT_ST_MASK;
 423         int rv;
 424
 425         spin_lock(&gl->gl_spin);
 426         trace_gfs2_glock_state_change(gl, state);
 427         state_change(gl, state);
 428         gh = find_first_waiter(gl);
 429
 430         /* Demote to UN request arrived during demote to SH or DF */
 431         if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
 432             state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
 433                 gl->gl_target = LM_ST_UNLOCKED;
 434
 435         /* Check for state != intended state */
 436         if (unlikely(state != gl->gl_target)) {
 437                 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
 438                         /* move to back of queue and try next entry */
 439                         if (ret & LM_OUT_CANCELED) {
 440                                 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
 441                                         list_move_tail(&gh->gh_list, &gl->gl_holders);
 442                                 gh = find_first_waiter(gl);
 443                                 gl->gl_target = gh->gh_state;
 444                                 goto retry;
 445                         }
 446                         /* Some error or failed "try lock" - report it */
 447                         if ((ret & LM_OUT_ERROR) ||
 448                             (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
 449                                 gl->gl_target = gl->gl_state;
 450                                 do_error(gl, ret);
 451                                 goto out;
 452                         }
 453                 }
 454                 switch(state) {
 455                 /* Unlocked due to conversion deadlock, try again */
 456                 case LM_ST_UNLOCKED:
 457 retry:
 458                         do_xmote(gl, gh, gl->gl_target);
 459                         break;
 460                 /* Conversion fails, unlock and try again */
 461                 case LM_ST_SHARED:
 462                 case LM_ST_DEFERRED:
 463                         do_xmote(gl, gh, LM_ST_UNLOCKED);
 464                         break;
 465                 default: /* Everything else */
 466                         pr_err("wanted %u got %u\n", gl->gl_target, state);
 467                         GLOCK_BUG_ON(gl, 1);
 468                 }
 469                 spin_unlock(&gl->gl_spin);
 470                 return;
 471         }
 472
 473         /* Fast path - we got what we asked for */
 474         if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
 475                 gfs2_demote_wake(gl);
 476         if (state != LM_ST_UNLOCKED) {
 477                 if (glops->go_xmote_bh) {
 478                         spin_unlock(&gl->gl_spin);
 479                         rv = glops->go_xmote_bh(gl, gh);
 480                         spin_lock(&gl->gl_spin);
 481                         if (rv) {
 482                                 do_error(gl, rv);
 483                                 goto out;
 484                         }
 485                 }
 486                 rv = do_promote(gl);
 487                 if (rv == 2)
 488                         goto out_locked;
 489         }
 490 out:
 491         clear_bit(GLF_LOCK, &gl->gl_flags);
 492 out_locked:
 493         spin_unlock(&gl->gl_spin);
 494 }
 495
 496 /**
 497  * do_xmote - Calls the DLM to change the state of a lock
 498  * @gl: The lock state
 499  * @gh: The holder (only for promotes)
 500  * @target: The target lock state
 501  *
 502  */
 503
 504 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
 505 __releases(&gl->gl_spin)
 506 __acquires(&gl->gl_spin)
 507 {
 508         const struct gfs2_glock_operations *glops = gl->gl_ops;
 509         struct gfs2_sbd *sdp = gl->gl_sbd;
 510         unsigned int lck_flags = gh ? gh->gh_flags : 0;
 511         int ret;
 512
 513         lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
 514                       LM_FLAG_PRIORITY);
 515         GLOCK_BUG_ON(gl, gl->gl_state == target);
 516         GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
 517         if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
 518             glops->go_inval) {
 519                 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 520                 do_error(gl, 0); /* Fail queued try locks */
 521         }
 522         gl->gl_req = target;
 523         set_bit(GLF_BLOCKING, &gl->gl_flags);
 524         if ((gl->gl_req == LM_ST_UNLOCKED) ||
 525             (gl->gl_state == LM_ST_EXCLUSIVE) ||
 526             (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
 527                 clear_bit(GLF_BLOCKING, &gl->gl_flags);
 528         spin_unlock(&gl->gl_spin);
 529         if (glops->go_sync)
 530                 glops->go_sync(gl);
 531         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 532                 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
 533         clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 534
 535         gfs2_glock_hold(gl);
 536         if (sdp->sd_lockstruct.ls_ops->lm_lock) {
 537                 /* lock_dlm */
 538                 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
 539                 if (ret) {
 540                         pr_err("lm_lock ret %d\n", ret);
 541                         GLOCK_BUG_ON(gl, 1);
 542                 }
 543         } else { /* lock_nolock */
 544                 finish_xmote(gl, target);
 545                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 546                         gfs2_glock_put(gl);
 547         }
 548
 549         spin_lock(&gl->gl_spin);
 550 }
 551
 552 /**
 553  * find_first_holder - find the first "holder" gh
 554  * @gl: the glock
 555  */
 556
 557 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
 558 {
 559         struct gfs2_holder *gh;
 560
 561         if (!list_empty(&gl->gl_holders)) {
 562                 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
 563                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 564                         return gh;
 565         }
 566         return NULL;
 567 }
 568
 569 /**
 570  * run_queue - do all outstanding tasks related to a glock
 571  * @gl: The glock in question
 572  * @nonblock: True if we must not block in run_queue
 573  *
 574  */
 575
 576 static void run_queue(struct gfs2_glock *gl, const int nonblock)
 577 __releases(&gl->gl_spin)
 578 __acquires(&gl->gl_spin)
 579 {
 580         struct gfs2_holder *gh = NULL;
 581         int ret;
 582
 583         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
 584                 return;
 585
 586         GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
 587
 588         if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
 589             gl->gl_demote_state != gl->gl_state) {
 590                 if (find_first_holder(gl))
 591                         goto out_unlock;
 592                 if (nonblock)
 593                         goto out_sched;
 594                 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
 595                 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
 596                 gl->gl_target = gl->gl_demote_state;
 597         } else {
 598                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
 599                         gfs2_demote_wake(gl);
 600                 ret = do_promote(gl);
 601                 if (ret == 0)
 602                         goto out_unlock;
 603                 if (ret == 2)
 604                         goto out;
 605                 gh = find_first_waiter(gl);
 606                 gl->gl_target = gh->gh_state;
 607                 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
 608                         do_error(gl, 0); /* Fail queued try locks */
 609         }
 610         do_xmote(gl, gh, gl->gl_target);
 611 out:
 612         return;
 613
 614 out_sched:
 615         clear_bit(GLF_LOCK, &gl->gl_flags);
 616         smp_mb__after_atomic();
 617         gl->gl_lockref.count++;
 618         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 619                 gl->gl_lockref.count--;
 620         return;
 621
 622 out_unlock:
 623         clear_bit(GLF_LOCK, &gl->gl_flags);
 624         smp_mb__after_atomic();
 625         return;
 626 }
 627
 628 static void delete_work_func(struct work_struct *work)
 629 {
 630         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
 631         struct gfs2_sbd *sdp = gl->gl_sbd;
 632         struct gfs2_inode *ip;
 633         struct inode *inode;
 634         u64 no_addr = gl->gl_name.ln_number;
 635
 636         ip = gl->gl_object;
 637         /* Note: Unsafe to dereference ip as we don't hold right refs/locks */
 638
 639         if (ip)
 640                 inode = gfs2_ilookup(sdp->sd_vfs, no_addr, 1);
 641         else
 642                 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
 643         if (inode && !IS_ERR(inode)) {
 644                 d_prune_aliases(inode);
 645                 iput(inode);
 646         }
 647         gfs2_glock_put(gl);
 648 }
 649
 650 static void glock_work_func(struct work_struct *work)
 651 {
 652         unsigned long delay = 0;
 653         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
 654         int drop_ref = 0;
 655
 656         if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
 657                 finish_xmote(gl, gl->gl_reply);
 658                 drop_ref = 1;
 659         }
 660         spin_lock(&gl->gl_spin);
 661         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
 662             gl->gl_state != LM_ST_UNLOCKED &&
 663             gl->gl_demote_state != LM_ST_EXCLUSIVE) {
 664                 unsigned long holdtime, now = jiffies;
 665
 666                 holdtime = gl->gl_tchange + gl->gl_hold_time;
 667                 if (time_before(now, holdtime))
 668                         delay = holdtime - now;
 669
 670                 if (!delay) {
 671                         clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
 672                         set_bit(GLF_DEMOTE, &gl->gl_flags);
 673                 }
 674         }
 675         run_queue(gl, 0);
 676         spin_unlock(&gl->gl_spin);
 677         if (!delay)
 678                 gfs2_glock_put(gl);
 679         else {
 680                 if (gl->gl_name.ln_type != LM_TYPE_INODE)
 681                         delay = 0;
 682                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
 683                         gfs2_glock_put(gl);
 684         }
 685         if (drop_ref)
 686                 gfs2_glock_put(gl);
 687 }
 688
 689 /**
 690  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
 691  * @sdp: The GFS2 superblock
 692  * @number: the lock number
 693  * @glops: The glock_operations to use
 694  * @create: If 0, don't create the glock if it doesn't exist
 695  * @glp: the glock is returned here
 696  *
 697  * This does not lock a glock, just finds/creates structures for one.
 698  *
 699  * Returns: errno
 700  */
 701
 702 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
 703                    const struct gfs2_glock_operations *glops, int create,
 704                    struct gfs2_glock **glp)
 705 {
 706         struct super_block *s = sdp->sd_vfs;
 707         struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
 708         struct gfs2_glock *gl, *tmp;
 709         unsigned int hash = gl_hash(sdp, &name);
 710         struct address_space *mapping;
 711         struct kmem_cache *cachep;
 712
 713         rcu_read_lock();
 714         gl = search_bucket(hash, sdp, &name);
 715         rcu_read_unlock();
 716
 717         *glp = gl;
 718         if (gl)
 719                 return 0;
 720         if (!create)
 721                 return -ENOENT;
 722
 723         if (glops->go_flags & GLOF_ASPACE)
 724                 cachep = gfs2_glock_aspace_cachep;
 725         else
 726                 cachep = gfs2_glock_cachep;
 727         gl = kmem_cache_alloc(cachep, GFP_NOFS);
 728         if (!gl)
 729                 return -ENOMEM;
 730
 731         memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
 732
 733         if (glops->go_flags & GLOF_LVB) {
 734                 gl->gl_lksb.sb_lvbptr = kzalloc(GFS2_MIN_LVB_SIZE, GFP_NOFS);
 735                 if (!gl->gl_lksb.sb_lvbptr) {
 736                         kmem_cache_free(cachep, gl);
 737                         return -ENOMEM;
 738                 }
 739         }
 740
 741         atomic_inc(&sdp->sd_glock_disposal);
 742         gl->gl_sbd = sdp;
 743         gl->gl_flags = 0;
 744         gl->gl_name = name;
 745         gl->gl_lockref.count = 1;
 746         gl->gl_state = LM_ST_UNLOCKED;
 747         gl->gl_target = LM_ST_UNLOCKED;
 748         gl->gl_demote_state = LM_ST_EXCLUSIVE;
 749         gl->gl_hash = hash;
 750         gl->gl_ops = glops;
 751         gl->gl_dstamp = ktime_set(0, 0);
 752         preempt_disable();
 753         /* We use the global stats to estimate the initial per-glock stats */
 754         gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
 755         preempt_enable();
 756         gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
 757         gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
 758         gl->gl_tchange = jiffies;
 759         gl->gl_object = NULL;
 760         gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
 761         INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
 762         INIT_WORK(&gl->gl_delete, delete_work_func);
 763
 764         mapping = gfs2_glock2aspace(gl);
 765         if (mapping) {
 766                 mapping->a_ops = &gfs2_meta_aops;
 767                 mapping->host = s->s_bdev->bd_inode;
 768                 mapping->flags = 0;
 769                 mapping_set_gfp_mask(mapping, GFP_NOFS);
 770                 mapping->private_data = NULL;
 771                 mapping->writeback_index = 0;
 772         }
 773
 774         spin_lock_bucket(hash);
 775         tmp = search_bucket(hash, sdp, &name);
 776         if (tmp) {
 777                 spin_unlock_bucket(hash);
 778                 kfree(gl->gl_lksb.sb_lvbptr);
 779                 kmem_cache_free(cachep, gl);
 780                 atomic_dec(&sdp->sd_glock_disposal);
 781                 gl = tmp;
 782         } else {
 783                 hlist_bl_add_head_rcu(&gl->gl_list, &gl_hash_table[hash]);
 784                 spin_unlock_bucket(hash);
 785         }
 786
 787         *glp = gl;
 788
 789         return 0;
 790 }
 791
 792 /**
 793  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
 794  * @gl: the glock
 795  * @state: the state we're requesting
 796  * @flags: the modifier flags
 797  * @gh: the holder structure
 798  *
 799  */
 800
 801 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
 802                       struct gfs2_holder *gh)
 803 {
 804         INIT_LIST_HEAD(&gh->gh_list);
 805         gh->gh_gl = gl;
 806         gh->gh_ip = _RET_IP_;
 807         gh->gh_owner_pid = get_pid(task_pid(current));
 808         gh->gh_state = state;
 809         gh->gh_flags = flags;
 810         gh->gh_error = 0;
 811         gh->gh_iflags = 0;
 812         gfs2_glock_hold(gl);
 813 }
 814
 815 /**
 816  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
 817  * @state: the state we're requesting
 818  * @flags: the modifier flags
 819  * @gh: the holder structure
 820  *
 821  * Don't mess with the glock.
 822  *
 823  */
 824
 825 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
 826 {
 827         gh->gh_state = state;
 828         gh->gh_flags = flags;
 829         gh->gh_iflags = 0;
 830         gh->gh_ip = _RET_IP_;
 831         put_pid(gh->gh_owner_pid);
 832         gh->gh_owner_pid = get_pid(task_pid(current));
 833 }
 834
 835 /**
 836  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
 837  * @gh: the holder structure
 838  *
 839  */
 840
 841 void gfs2_holder_uninit(struct gfs2_holder *gh)
 842 {
 843         put_pid(gh->gh_owner_pid);
 844         gfs2_glock_put(gh->gh_gl);
 845         gh->gh_gl = NULL;
 846         gh->gh_ip = 0;
 847 }
 848
 849 /**
 850  * gfs2_glock_wait - wait on a glock acquisition
 851  * @gh: the glock holder
 852  *
 853  * Returns: 0 on success
 854  */
 855
 856 int gfs2_glock_wait(struct gfs2_holder *gh)
 857 {
 858         unsigned long time1 = jiffies;
 859
 860         might_sleep();
 861         wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
 862         if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
 863                 /* Lengthen the minimum hold time. */
 864                 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
 865                                               GL_GLOCK_HOLD_INCR,
 866                                               GL_GLOCK_MAX_HOLD);
 867         return gh->gh_error;
 868 }
 869
 870 /**
 871  * handle_callback - process a demote request
 872  * @gl: the glock
 873  * @state: the state the caller wants us to change to
 874  *
 875  * There are only two requests that we are going to see in actual
 876  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
 877  */
 878
 879 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
 880                             unsigned long delay, bool remote)
 881 {
 882         int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
 883
 884         set_bit(bit, &gl->gl_flags);
 885         if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
 886                 gl->gl_demote_state = state;
 887                 gl->gl_demote_time = jiffies;
 888         } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
 889                         gl->gl_demote_state != state) {
 890                 gl->gl_demote_state = LM_ST_UNLOCKED;
 891         }
 892         if (gl->gl_ops->go_callback)
 893                 gl->gl_ops->go_callback(gl, remote);
 894         trace_gfs2_demote_rq(gl, remote);
 895 }
 896
 897 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
 898 {
 899         struct va_format vaf;
 900         va_list args;
 901
 902         va_start(args, fmt);
 903
 904         if (seq) {
 905                 seq_vprintf(seq, fmt, args);
 906         } else {
 907                 vaf.fmt = fmt;
 908                 vaf.va = &args;
 909
 910                 pr_err("%pV", &vaf);
 911         }
 912
 913         va_end(args);
 914 }
 915
 916 /**
 917  * add_to_queue - Add a holder to the wait queue (but look for recursion)
 918  * @gh: the holder structure to add
 919  *
 920  * Eventually we should move the recursive locking trap to a
 921  * debugging option or something like that. This is the fast
 922  * path and needs to have the minimum number of distractions.
 923  *
 924  */
 925
 926 static inline void add_to_queue(struct gfs2_holder *gh)
 927 __releases(&gl->gl_spin)
 928 __acquires(&gl->gl_spin)
 929 {
 930         struct gfs2_glock *gl = gh->gh_gl;
 931         struct gfs2_sbd *sdp = gl->gl_sbd;
 932         struct list_head *insert_pt = NULL;
 933         struct gfs2_holder *gh2;
 934         int try_futile = 0;
 935
 936         BUG_ON(gh->gh_owner_pid == NULL);
 937         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
 938                 BUG();
 939
 940         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
 941                 if (test_bit(GLF_LOCK, &gl->gl_flags))
 942                         try_futile = !may_grant(gl, gh);
 943                 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 944                         goto fail;
 945         }
 946
 947         list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
 948                 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
 949                     (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
 950                         goto trap_recursive;
 951                 if (try_futile &&
 952                     !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
 953 fail:
 954                         gh->gh_error = GLR_TRYFAILED;
 955                         gfs2_holder_wake(gh);
 956                         return;
 957                 }
 958                 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
 959                         continue;
 960                 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
 961                         insert_pt = &gh2->gh_list;
 962         }
 963         set_bit(GLF_QUEUED, &gl->gl_flags);
 964         trace_gfs2_glock_queue(gh, 1);
 965         gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
 966         gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
 967         if (likely(insert_pt == NULL)) {
 968                 list_add_tail(&gh->gh_list, &gl->gl_holders);
 969                 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
 970                         goto do_cancel;
 971                 return;
 972         }
 973         list_add_tail(&gh->gh_list, insert_pt);
 974 do_cancel:
 975         gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
 976         if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
 977                 spin_unlock(&gl->gl_spin);
 978                 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
 979                         sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
 980                 spin_lock(&gl->gl_spin);
 981         }
 982         return;
 983
 984 trap_recursive:
 985         pr_err("original: %pSR\n", (void *)gh2->gh_ip);
 986         pr_err("pid: %d\n", pid_nr(gh2->gh_owner_pid));
 987         pr_err("lock type: %d req lock state : %d\n",
 988                gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
 989         pr_err("new: %pSR\n", (void *)gh->gh_ip);
 990         pr_err("pid: %d\n", pid_nr(gh->gh_owner_pid));
 991         pr_err("lock type: %d req lock state : %d\n",
 992                gh->gh_gl->gl_name.ln_type, gh->gh_state);
 993         gfs2_dump_glock(NULL, gl);
 994         BUG();
 995 }
 996
 997 /**
 998  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
 999  * @gh: the holder structure
1000  *
1001  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1002  *
1003  * Returns: 0, GLR_TRYFAILED, or errno on failure
1004  */
1005
1006 int gfs2_glock_nq(struct gfs2_holder *gh)
1007 {
1008         struct gfs2_glock *gl = gh->gh_gl;
1009         struct gfs2_sbd *sdp = gl->gl_sbd;
1010         int error = 0;
1011
1012         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1013                 return -EIO;
1014
1015         if (test_bit(GLF_LRU, &gl->gl_flags))
1016                 gfs2_glock_remove_from_lru(gl);
1017
1018         spin_lock(&gl->gl_spin);
1019         add_to_queue(gh);
1020         if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1021                      test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1022                 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1023                 gl->gl_lockref.count++;
1024                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1025                         gl->gl_lockref.count--;
1026         }
1027         run_queue(gl, 1);
1028         spin_unlock(&gl->gl_spin);
1029
1030         if (!(gh->gh_flags & GL_ASYNC))
1031                 error = gfs2_glock_wait(gh);
1032
1033         return error;
1034 }
1035
1036 /**
1037  * gfs2_glock_poll - poll to see if an async request has been completed
1038  * @gh: the holder
1039  *
1040  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1041  */
1042
1043 int gfs2_glock_poll(struct gfs2_holder *gh)
1044 {
1045         return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1046 }
1047
1048 /**
1049  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1050  * @gh: the glock holder
1051  *
1052  */
1053
1054 void gfs2_glock_dq(struct gfs2_holder *gh)
1055 {
1056         struct gfs2_glock *gl = gh->gh_gl;
1057         const struct gfs2_glock_operations *glops = gl->gl_ops;
1058         unsigned delay = 0;
1059         int fast_path = 0;
1060
1061         spin_lock(&gl->gl_spin);
1062         if (gh->gh_flags & GL_NOCACHE)
1063                 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1064
1065         list_del_init(&gh->gh_list);
1066         if (find_first_holder(gl) == NULL) {
1067                 if (glops->go_unlock) {
1068                         GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1069                         spin_unlock(&gl->gl_spin);
1070                         glops->go_unlock(gh);
1071                         spin_lock(&gl->gl_spin);
1072                         clear_bit(GLF_LOCK, &gl->gl_flags);
1073                 }
1074                 if (list_empty(&gl->gl_holders) &&
1075                     !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1076                     !test_bit(GLF_DEMOTE, &gl->gl_flags))
1077                         fast_path = 1;
1078         }
1079         if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl) &&
1080             (glops->go_flags & GLOF_LRU))
1081                 gfs2_glock_add_to_lru(gl);
1082
1083         trace_gfs2_glock_queue(gh, 0);
1084         spin_unlock(&gl->gl_spin);
1085         if (likely(fast_path))
1086                 return;
1087
1088         gfs2_glock_hold(gl);
1089         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1090             !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1091             gl->gl_name.ln_type == LM_TYPE_INODE)
1092                 delay = gl->gl_hold_time;
1093         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1094                 gfs2_glock_put(gl);
1095 }
1096
1097 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1098 {
1099         struct gfs2_glock *gl = gh->gh_gl;
1100         gfs2_glock_dq(gh);
1101         might_sleep();
1102         wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1103 }
1104
1105 /**
1106  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1107  * @gh: the holder structure
1108  *
1109  */
1110
1111 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1112 {
1113         gfs2_glock_dq(gh);
1114         gfs2_holder_uninit(gh);
1115 }
1116
1117 /**
1118  * gfs2_glock_nq_num - acquire a glock based on lock number
1119  * @sdp: the filesystem
1120  * @number: the lock number
1121  * @glops: the glock operations for the type of glock
1122  * @state: the state to acquire the glock in
1123  * @flags: modifier flags for the acquisition
1124  * @gh: the struct gfs2_holder
1125  *
1126  * Returns: errno
1127  */
1128
1129 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1130                       const struct gfs2_glock_operations *glops,
1131                       unsigned int state, int flags, struct gfs2_holder *gh)
1132 {
1133         struct gfs2_glock *gl;
1134         int error;
1135
1136         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1137         if (!error) {
1138                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1139                 gfs2_glock_put(gl);
1140         }
1141
1142         return error;
1143 }
1144
1145 /**
1146  * glock_compare - Compare two struct gfs2_glock structures for sorting
1147  * @arg_a: the first structure
1148  * @arg_b: the second structure
1149  *
1150  */
1151
1152 static int glock_compare(const void *arg_a, const void *arg_b)
1153 {
1154         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1155         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1156         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1157         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1158
1159         if (a->ln_number > b->ln_number)
1160                 return 1;
1161         if (a->ln_number < b->ln_number)
1162                 return -1;
1163         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1164         return 0;
1165 }
1166
1167 /**
1168  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1169  * @num_gh: the number of structures
1170  * @ghs: an array of struct gfs2_holder structures
1171  *
1172  * Returns: 0 on success (all glocks acquired),
1173  *          errno on failure (no glocks acquired)
1174  */
1175
1176 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1177                      struct gfs2_holder **p)
1178 {
1179         unsigned int x;
1180         int error = 0;
1181
1182         for (x = 0; x < num_gh; x++)
1183                 p[x] = &ghs[x];
1184
1185         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1186
1187         for (x = 0; x < num_gh; x++) {
1188                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1189
1190                 error = gfs2_glock_nq(p[x]);
1191                 if (error) {
1192                         while (x--)
1193                                 gfs2_glock_dq(p[x]);
1194                         break;
1195                 }
1196         }
1197
1198         return error;
1199 }
1200
1201 /**
1202  * gfs2_glock_nq_m - acquire multiple glocks
1203  * @num_gh: the number of structures
1204  * @ghs: an array of struct gfs2_holder structures
1205  *
1206  *
1207  * Returns: 0 on success (all glocks acquired),
1208  *          errno on failure (no glocks acquired)
1209  */
1210
1211 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1212 {
1213         struct gfs2_holder *tmp[4];
1214         struct gfs2_holder **pph = tmp;
1215         int error = 0;
1216
1217         switch(num_gh) {
1218         case 0:
1219                 return 0;
1220         case 1:
1221                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1222                 return gfs2_glock_nq(ghs);
1223         default:
1224                 if (num_gh <= 4)
1225                         break;
1226                 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1227                 if (!pph)
1228                         return -ENOMEM;
1229         }
1230
1231         error = nq_m_sync(num_gh, ghs, pph);
1232
1233         if (pph != tmp)
1234                 kfree(pph);
1235
1236         return error;
1237 }
1238
1239 /**
1240  * gfs2_glock_dq_m - release multiple glocks
1241  * @num_gh: the number of structures
1242  * @ghs: an array of struct gfs2_holder structures
1243  *
1244  */
1245
1246 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1247 {
1248         while (num_gh--)
1249                 gfs2_glock_dq(&ghs[num_gh]);
1250 }
1251
1252 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1253 {
1254         unsigned long delay = 0;
1255         unsigned long holdtime;
1256         unsigned long now = jiffies;
1257
1258         gfs2_glock_hold(gl);
1259         holdtime = gl->gl_tchange + gl->gl_hold_time;
1260         if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1261             gl->gl_name.ln_type == LM_TYPE_INODE) {
1262                 if (time_before(now, holdtime))
1263                         delay = holdtime - now;
1264                 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1265                         delay = gl->gl_hold_time;
1266         }
1267
1268         spin_lock(&gl->gl_spin);
1269         handle_callback(gl, state, delay, true);
1270         spin_unlock(&gl->gl_spin);
1271         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1272                 gfs2_glock_put(gl);
1273 }
1274
1275 /**
1276  * gfs2_should_freeze - Figure out if glock should be frozen
1277  * @gl: The glock in question
1278  *
1279  * Glocks are not frozen if (a) the result of the dlm operation is
1280  * an error, (b) the locking operation was an unlock operation or
1281  * (c) if there is a "noexp" flagged request anywhere in the queue
1282  *
1283  * Returns: 1 if freezing should occur, 0 otherwise
1284  */
1285
1286 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1287 {
1288         const struct gfs2_holder *gh;
1289
1290         if (gl->gl_reply & ~LM_OUT_ST_MASK)
1291                 return 0;
1292         if (gl->gl_target == LM_ST_UNLOCKED)
1293                 return 0;
1294
1295         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1296                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1297                         continue;
1298                 if (LM_FLAG_NOEXP & gh->gh_flags)
1299                         return 0;
1300         }
1301
1302         return 1;
1303 }
1304
1305 /**
1306  * gfs2_glock_complete - Callback used by locking
1307  * @gl: Pointer to the glock
1308  * @ret: The return value from the dlm
1309  *
1310  * The gl_reply field is under the gl_spin lock so that it is ok
1311  * to use a bitfield shared with other glock state fields.
1312  */
1313
1314 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1315 {
1316         struct lm_lockstruct *ls = &gl->gl_sbd->sd_lockstruct;
1317
1318         spin_lock(&gl->gl_spin);
1319         gl->gl_reply = ret;
1320
1321         if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1322                 if (gfs2_should_freeze(gl)) {
1323                         set_bit(GLF_FROZEN, &gl->gl_flags);
1324                         spin_unlock(&gl->gl_spin);
1325                         return;
1326                 }
1327         }
1328
1329         gl->gl_lockref.count++;
1330         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1331         spin_unlock(&gl->gl_spin);
1332
1333         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1334                 gfs2_glock_put(gl);
1335 }
1336
1337 static int glock_cmp(void *priv, struct list_head *a, struct list_head *b)
1338 {
1339         struct gfs2_glock *gla, *glb;
1340
1341         gla = list_entry(a, struct gfs2_glock, gl_lru);
1342         glb = list_entry(b, struct gfs2_glock, gl_lru);
1343
1344         if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1345                 return 1;
1346         if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1347                 return -1;
1348
1349         return 0;
1350 }
1351
1352 /**
1353  * gfs2_dispose_glock_lru - Demote a list of glocks
1354  * @list: The list to dispose of
1355  *
1356  * Disposing of glocks may involve disk accesses, so that here we sort
1357  * the glocks by number (i.e. disk location of the inodes) so that if
1358  * there are any such accesses, they'll be sent in order (mostly).
1359  *
1360  * Must be called under the lru_lock, but may drop and retake this
1361  * lock. While the lru_lock is dropped, entries may vanish from the
1362  * list, but no new entries will appear on the list (since it is
1363  * private)
1364  */
1365
1366 static void gfs2_dispose_glock_lru(struct list_head *list)
1367 __releases(&lru_lock)
1368 __acquires(&lru_lock)
1369 {
1370         struct gfs2_glock *gl;
1371
1372         list_sort(NULL, list, glock_cmp);
1373
1374         while(!list_empty(list)) {
1375                 gl = list_entry(list->next, struct gfs2_glock, gl_lru);
1376                 list_del_init(&gl->gl_lru);
1377                 if (!spin_trylock(&gl->gl_spin)) {
1378 add_back_to_lru:
1379                         list_add(&gl->gl_lru, &lru_list);
1380                         atomic_inc(&lru_count);
1381                         continue;
1382                 }
1383                 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1384                         spin_unlock(&gl->gl_spin);
1385                         goto add_back_to_lru;
1386                 }
1387                 clear_bit(GLF_LRU, &gl->gl_flags);
1388                 gl->gl_lockref.count++;
1389                 if (demote_ok(gl))
1390                         handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1391                 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1392                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1393                         gl->gl_lockref.count--;
1394                 spin_unlock(&gl->gl_spin);
1395                 cond_resched_lock(&lru_lock);
1396         }
1397 }
1398
1399 /**
1400  * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1401  * @nr: The number of entries to scan
1402  *
1403  * This function selects the entries on the LRU which are able to
1404  * be demoted, and then kicks off the process by calling
1405  * gfs2_dispose_glock_lru() above.
1406  */
1407
1408 static long gfs2_scan_glock_lru(int nr)
1409 {
1410         struct gfs2_glock *gl;
1411         LIST_HEAD(skipped);
1412         LIST_HEAD(dispose);
1413         long freed = 0;
1414
1415         spin_lock(&lru_lock);
1416         while ((nr-- >= 0) && !list_empty(&lru_list)) {
1417                 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1418
1419                 /* Test for being demotable */
1420                 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1421                         list_move(&gl->gl_lru, &dispose);
1422                         atomic_dec(&lru_count);
1423                         freed++;
1424                         continue;
1425                 }
1426
1427                 list_move(&gl->gl_lru, &skipped);
1428         }
1429         list_splice(&skipped, &lru_list);
1430         if (!list_empty(&dispose))
1431                 gfs2_dispose_glock_lru(&dispose);
1432         spin_unlock(&lru_lock);
1433
1434         return freed;
1435 }
1436
1437 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1438                                             struct shrink_control *sc)
1439 {
1440         if (!(sc->gfp_mask & __GFP_FS))
1441                 return SHRINK_STOP;
1442         return gfs2_scan_glock_lru(sc->nr_to_scan);
1443 }
1444
1445 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1446                                              struct shrink_control *sc)
1447 {
1448         return vfs_pressure_ratio(atomic_read(&lru_count));
1449 }
1450
1451 static struct shrinker glock_shrinker = {
1452         .seeks = DEFAULT_SEEKS,
1453         .count_objects = gfs2_glock_shrink_count,
1454         .scan_objects = gfs2_glock_shrink_scan,
1455 };
1456
1457 /**
1458  * examine_bucket - Call a function for glock in a hash bucket
1459  * @examiner: the function
1460  * @sdp: the filesystem
1461  * @bucket: the bucket
1462  *
1463  */
1464
1465 static void examine_bucket(glock_examiner examiner, const struct gfs2_sbd *sdp,
1466                           unsigned int hash)
1467 {
1468         struct gfs2_glock *gl;
1469         struct hlist_bl_head *head = &gl_hash_table[hash];
1470         struct hlist_bl_node *pos;
1471
1472         rcu_read_lock();
1473         hlist_bl_for_each_entry_rcu(gl, pos, head, gl_list) {
1474                 if ((gl->gl_sbd == sdp) && lockref_get_not_dead(&gl->gl_lockref))
1475                         examiner(gl);
1476         }
1477         rcu_read_unlock();
1478         cond_resched();
1479 }
1480
1481 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1482 {
1483         unsigned x;
1484
1485         for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1486                 examine_bucket(examiner, sdp, x);
1487 }
1488
1489
1490 /**
1491  * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1492  * @gl: The glock to thaw
1493  *
1494  */
1495
1496 static void thaw_glock(struct gfs2_glock *gl)
1497 {
1498         if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1499                 goto out;
1500         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1501         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0) {
1502 out:
1503                 gfs2_glock_put(gl);
1504         }
1505 }
1506
1507 /**
1508  * clear_glock - look at a glock and see if we can free it from glock cache
1509  * @gl: the glock to look at
1510  *
1511  */
1512
1513 static void clear_glock(struct gfs2_glock *gl)
1514 {
1515         gfs2_glock_remove_from_lru(gl);
1516
1517         spin_lock(&gl->gl_spin);
1518         if (gl->gl_state != LM_ST_UNLOCKED)
1519                 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1520         spin_unlock(&gl->gl_spin);
1521         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1522                 gfs2_glock_put(gl);
1523 }
1524
1525 /**
1526  * gfs2_glock_thaw - Thaw any frozen glocks
1527  * @sdp: The super block
1528  *
1529  */
1530
1531 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1532 {
1533         glock_hash_walk(thaw_glock, sdp);
1534 }
1535
1536 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1537 {
1538         spin_lock(&gl->gl_spin);
1539         gfs2_dump_glock(seq, gl);
1540         spin_unlock(&gl->gl_spin);
1541 }
1542
1543 static void dump_glock_func(struct gfs2_glock *gl)
1544 {
1545         dump_glock(NULL, gl);
1546 }
1547
1548 /**
1549  * gfs2_gl_hash_clear - Empty out the glock hash table
1550  * @sdp: the filesystem
1551  * @wait: wait until it's all gone
1552  *
1553  * Called when unmounting the filesystem.
1554  */
1555
1556 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1557 {
1558         set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1559         flush_workqueue(glock_workqueue);
1560         glock_hash_walk(clear_glock, sdp);
1561         flush_workqueue(glock_workqueue);
1562         wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
1563         glock_hash_walk(dump_glock_func, sdp);
1564 }
1565
1566 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1567 {
1568         struct gfs2_glock *gl = ip->i_gl;
1569         int ret;
1570
1571         ret = gfs2_truncatei_resume(ip);
1572         gfs2_assert_withdraw(gl->gl_sbd, ret == 0);
1573
1574         spin_lock(&gl->gl_spin);
1575         clear_bit(GLF_LOCK, &gl->gl_flags);
1576         run_queue(gl, 1);
1577         spin_unlock(&gl->gl_spin);
1578 }
1579
1580 static const char *state2str(unsigned state)
1581 {
1582         switch(state) {
1583         case LM_ST_UNLOCKED:
1584                 return "UN";
1585         case LM_ST_SHARED:
1586                 return "SH";
1587         case LM_ST_DEFERRED:
1588                 return "DF";
1589         case LM_ST_EXCLUSIVE:
1590                 return "EX";
1591         }
1592         return "??";
1593 }
1594
1595 static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1596 {
1597         char *p = buf;
1598         if (flags & LM_FLAG_TRY)
1599                 *p++ = 't';
1600         if (flags & LM_FLAG_TRY_1CB)
1601                 *p++ = 'T';
1602         if (flags & LM_FLAG_NOEXP)
1603                 *p++ = 'e';
1604         if (flags & LM_FLAG_ANY)
1605                 *p++ = 'A';
1606         if (flags & LM_FLAG_PRIORITY)
1607                 *p++ = 'p';
1608         if (flags & GL_ASYNC)
1609                 *p++ = 'a';
1610         if (flags & GL_EXACT)
1611                 *p++ = 'E';
1612         if (flags & GL_NOCACHE)
1613                 *p++ = 'c';
1614         if (test_bit(HIF_HOLDER, &iflags))
1615                 *p++ = 'H';
1616         if (test_bit(HIF_WAIT, &iflags))
1617                 *p++ = 'W';
1618         if (test_bit(HIF_FIRST, &iflags))
1619                 *p++ = 'F';
1620         *p = 0;
1621         return buf;
1622 }
1623
1624 /**
1625  * dump_holder - print information about a glock holder
1626  * @seq: the seq_file struct
1627  * @gh: the glock holder
1628  *
1629  */
1630
1631 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1632 {
1633         struct task_struct *gh_owner = NULL;
1634         char flags_buf[32];
1635
1636         rcu_read_lock();
1637         if (gh->gh_owner_pid)
1638                 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1639         gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1640                        state2str(gh->gh_state),
1641                        hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1642                        gh->gh_error,
1643                        gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1644                        gh_owner ? gh_owner->comm : "(ended)",
1645                        (void *)gh->gh_ip);
1646         rcu_read_unlock();
1647 }
1648
1649 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1650 {
1651         const unsigned long *gflags = &gl->gl_flags;
1652         char *p = buf;
1653
1654         if (test_bit(GLF_LOCK, gflags))
1655                 *p++ = 'l';
1656         if (test_bit(GLF_DEMOTE, gflags))
1657                 *p++ = 'D';
1658         if (test_bit(GLF_PENDING_DEMOTE, gflags))
1659                 *p++ = 'd';
1660         if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1661                 *p++ = 'p';
1662         if (test_bit(GLF_DIRTY, gflags))
1663                 *p++ = 'y';
1664         if (test_bit(GLF_LFLUSH, gflags))
1665                 *p++ = 'f';
1666         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1667                 *p++ = 'i';
1668         if (test_bit(GLF_REPLY_PENDING, gflags))
1669                 *p++ = 'r';
1670         if (test_bit(GLF_INITIAL, gflags))
1671                 *p++ = 'I';
1672         if (test_bit(GLF_FROZEN, gflags))
1673                 *p++ = 'F';
1674         if (test_bit(GLF_QUEUED, gflags))
1675                 *p++ = 'q';
1676         if (test_bit(GLF_LRU, gflags))
1677                 *p++ = 'L';
1678         if (gl->gl_object)
1679                 *p++ = 'o';
1680         if (test_bit(GLF_BLOCKING, gflags))
1681                 *p++ = 'b';
1682         *p = 0;
1683         return buf;
1684 }
1685
1686 /**
1687  * gfs2_dump_glock - print information about a glock
1688  * @seq: The seq_file struct
1689  * @gl: the glock
1690  *
1691  * The file format is as follows:
1692  * One line per object, capital letters are used to indicate objects
1693  * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1694  * other objects are indented by a single space and follow the glock to
1695  * which they are related. Fields are indicated by lower case letters
1696  * followed by a colon and the field value, except for strings which are in
1697  * [] so that its possible to see if they are composed of spaces for
1698  * example. The field's are n = number (id of the object), f = flags,
1699  * t = type, s = state, r = refcount, e = error, p = pid.
1700  *
1701  */
1702
1703 void gfs2_dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1704 {
1705         const struct gfs2_glock_operations *glops = gl->gl_ops;
1706         unsigned long long dtime;
1707         const struct gfs2_holder *gh;
1708         char gflags_buf[32];
1709
1710         dtime = jiffies - gl->gl_demote_time;
1711         dtime *= 1000000/HZ; /* demote time in uSec */
1712         if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1713                 dtime = 0;
1714         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",
1715                   state2str(gl->gl_state),
1716                   gl->gl_name.ln_type,
1717                   (unsigned long long)gl->gl_name.ln_number,
1718                   gflags2str(gflags_buf, gl),
1719                   state2str(gl->gl_target),
1720                   state2str(gl->gl_demote_state), dtime,
1721                   atomic_read(&gl->gl_ail_count),
1722                   atomic_read(&gl->gl_revokes),
1723                   (int)gl->gl_lockref.count, gl->gl_hold_time);
1724
1725         list_for_each_entry(gh, &gl->gl_holders, gh_list)
1726                 dump_holder(seq, gh);
1727
1728         if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1729                 glops->go_dump(seq, gl);
1730 }
1731
1732 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
1733 {
1734         struct gfs2_glock *gl = iter_ptr;
1735
1736         seq_printf(seq, "G: n:%u/%llx rtt:%lld/%lld rttb:%lld/%lld irt:%lld/%lld dcnt: %lld qcnt: %lld\n",
1737                    gl->gl_name.ln_type,
1738                    (unsigned long long)gl->gl_name.ln_number,
1739                    (long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
1740                    (long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
1741                    (long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
1742                    (long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
1743                    (long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
1744                    (long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
1745                    (long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
1746                    (long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
1747         return 0;
1748 }
1749
1750 static const char *gfs2_gltype[] = {
1751         "type",
1752         "reserved",
1753         "nondisk",
1754         "inode",
1755         "rgrp",
1756         "meta",
1757         "iopen",
1758         "flock",
1759         "plock",
1760         "quota",
1761         "journal",
1762 };
1763
1764 static const char *gfs2_stype[] = {
1765         [GFS2_LKS_SRTT]         = "srtt",
1766         [GFS2_LKS_SRTTVAR]      = "srttvar",
1767         [GFS2_LKS_SRTTB]        = "srttb",
1768         [GFS2_LKS_SRTTVARB]     = "srttvarb",
1769         [GFS2_LKS_SIRT]         = "sirt",
1770         [GFS2_LKS_SIRTVAR]      = "sirtvar",
1771         [GFS2_LKS_DCOUNT]       = "dlm",
1772         [GFS2_LKS_QCOUNT]       = "queue",
1773 };
1774
1775 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1776
1777 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
1778 {
1779         struct gfs2_glock_iter *gi = seq->private;
1780         struct gfs2_sbd *sdp = gi->sdp;
1781         unsigned index = gi->hash >> 3;
1782         unsigned subindex = gi->hash & 0x07;
1783         s64 value;
1784         int i;
1785
1786         if (index == 0 && subindex != 0)
1787                 return 0;
1788
1789         seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
1790                    (index == 0) ? "cpu": gfs2_stype[subindex]);
1791
1792         for_each_possible_cpu(i) {
1793                 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
1794                 if (index == 0) {
1795                         value = i;
1796                 } else {
1797                         value = lkstats->lkstats[index - 1].stats[subindex];
1798                 }
1799                 seq_printf(seq, " %15lld", (long long)value);
1800         }
1801         seq_putc(seq, '\n');
1802         return 0;
1803 }
1804
1805 int __init gfs2_glock_init(void)
1806 {
1807         unsigned i;
1808         for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
1809                 INIT_HLIST_BL_HEAD(&gl_hash_table[i]);
1810         }
1811
1812         glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1813                                           WQ_HIGHPRI | WQ_FREEZABLE, 0);
1814         if (!glock_workqueue)
1815                 return -ENOMEM;
1816         gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1817                                                 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1818                                                 0);
1819         if (!gfs2_delete_workqueue) {
1820                 destroy_workqueue(glock_workqueue);
1821                 return -ENOMEM;
1822         }
1823
1824         register_shrinker(&glock_shrinker);
1825
1826         return 0;
1827 }
1828
1829 void gfs2_glock_exit(void)
1830 {
1831         unregister_shrinker(&glock_shrinker);
1832         destroy_workqueue(glock_workqueue);
1833         destroy_workqueue(gfs2_delete_workqueue);
1834 }
1835
1836 static inline struct gfs2_glock *glock_hash_chain(unsigned hash)
1837 {
1838         return hlist_bl_entry(hlist_bl_first_rcu(&gl_hash_table[hash]),
1839                               struct gfs2_glock, gl_list);
1840 }
1841
1842 static inline struct gfs2_glock *glock_hash_next(struct gfs2_glock *gl)
1843 {
1844         return hlist_bl_entry(rcu_dereference(gl->gl_list.next),
1845                               struct gfs2_glock, gl_list);
1846 }
1847
1848 static int gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1849 {
1850         struct gfs2_glock *gl;
1851
1852         do {
1853                 gl = gi->gl;
1854                 if (gl) {
1855                         gi->gl = glock_hash_next(gl);
1856                         gi->nhash++;
1857                 } else {
1858                         if (gi->hash >= GFS2_GL_HASH_SIZE) {
1859                                 rcu_read_unlock();
1860                                 return 1;
1861                         }
1862                         gi->gl = glock_hash_chain(gi->hash);
1863                         gi->nhash = 0;
1864                 }
1865                 while (gi->gl == NULL) {
1866                         gi->hash++;
1867                         if (gi->hash >= GFS2_GL_HASH_SIZE) {
1868                                 rcu_read_unlock();
1869                                 return 1;
1870                         }
1871                         gi->gl = glock_hash_chain(gi->hash);
1872                         gi->nhash = 0;
1873                 }
1874         /* Skip entries for other sb and dead entries */
1875         } while (gi->sdp != gi->gl->gl_sbd ||
1876                  __lockref_is_dead(&gi->gl->gl_lockref));
1877
1878         return 0;
1879 }
1880
1881 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1882 {
1883         struct gfs2_glock_iter *gi = seq->private;
1884         loff_t n = *pos;
1885
1886         if (gi->last_pos <= *pos)
1887                 n = gi->nhash + (*pos - gi->last_pos);
1888         else
1889                 gi->hash = 0;
1890
1891         gi->nhash = 0;
1892         rcu_read_lock();
1893
1894         do {
1895                 if (gfs2_glock_iter_next(gi))
1896                         return NULL;
1897         } while (n--);
1898
1899         gi->last_pos = *pos;
1900         return gi->gl;
1901 }
1902
1903 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1904                                  loff_t *pos)
1905 {
1906         struct gfs2_glock_iter *gi = seq->private;
1907
1908         (*pos)++;
1909         gi->last_pos = *pos;
1910         if (gfs2_glock_iter_next(gi))
1911                 return NULL;
1912
1913         return gi->gl;
1914 }
1915
1916 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1917 {
1918         struct gfs2_glock_iter *gi = seq->private;
1919
1920         if (gi->gl)
1921                 rcu_read_unlock();
1922         gi->gl = NULL;
1923 }
1924
1925 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1926 {
1927         dump_glock(seq, iter_ptr);
1928         return 0;
1929 }
1930
1931 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
1932 {
1933         struct gfs2_glock_iter *gi = seq->private;
1934
1935         gi->hash = *pos;
1936         if (*pos >= GFS2_NR_SBSTATS)
1937                 return NULL;
1938         preempt_disable();
1939         return SEQ_START_TOKEN;
1940 }
1941
1942 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
1943                                    loff_t *pos)
1944 {
1945         struct gfs2_glock_iter *gi = seq->private;
1946         (*pos)++;
1947         gi->hash++;
1948         if (gi->hash >= GFS2_NR_SBSTATS) {
1949                 preempt_enable();
1950                 return NULL;
1951         }
1952         return SEQ_START_TOKEN;
1953 }
1954
1955 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
1956 {
1957         preempt_enable();
1958 }
1959
1960 static const struct seq_operations gfs2_glock_seq_ops = {
1961         .start = gfs2_glock_seq_start,
1962         .next  = gfs2_glock_seq_next,
1963         .stop  = gfs2_glock_seq_stop,
1964         .show  = gfs2_glock_seq_show,
1965 };
1966
1967 static const struct seq_operations gfs2_glstats_seq_ops = {
1968         .start = gfs2_glock_seq_start,
1969         .next  = gfs2_glock_seq_next,
1970         .stop  = gfs2_glock_seq_stop,
1971         .show  = gfs2_glstats_seq_show,
1972 };
1973
1974 static const struct seq_operations gfs2_sbstats_seq_ops = {
1975         .start = gfs2_sbstats_seq_start,
1976         .next  = gfs2_sbstats_seq_next,
1977         .stop  = gfs2_sbstats_seq_stop,
1978         .show  = gfs2_sbstats_seq_show,
1979 };
1980
1981 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
1982
1983 static int gfs2_glocks_open(struct inode *inode, struct file *file)
1984 {
1985         int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1986                                    sizeof(struct gfs2_glock_iter));
1987         if (ret == 0) {
1988                 struct seq_file *seq = file->private_data;
1989                 struct gfs2_glock_iter *gi = seq->private;
1990                 gi->sdp = inode->i_private;
1991                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
1992                 if (seq->buf)
1993                         seq->size = GFS2_SEQ_GOODSIZE;
1994         }
1995         return ret;
1996 }
1997
1998 static int gfs2_glstats_open(struct inode *inode, struct file *file)
1999 {
2000         int ret = seq_open_private(file, &gfs2_glstats_seq_ops,
2001                                    sizeof(struct gfs2_glock_iter));
2002         if (ret == 0) {
2003                 struct seq_file *seq = file->private_data;
2004                 struct gfs2_glock_iter *gi = seq->private;
2005                 gi->sdp = inode->i_private;
2006                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2007                 if (seq->buf)
2008                         seq->size = GFS2_SEQ_GOODSIZE;
2009         }
2010         return ret;
2011 }
2012
2013 static int gfs2_sbstats_open(struct inode *inode, struct file *file)
2014 {
2015         int ret = seq_open_private(file, &gfs2_sbstats_seq_ops,
2016                                    sizeof(struct gfs2_glock_iter));
2017         if (ret == 0) {
2018                 struct seq_file *seq = file->private_data;
2019                 struct gfs2_glock_iter *gi = seq->private;
2020                 gi->sdp = inode->i_private;
2021         }
2022         return ret;
2023 }
2024
2025 static const struct file_operations gfs2_glocks_fops = {
2026         .owner   = THIS_MODULE,
2027         .open    = gfs2_glocks_open,
2028         .read    = seq_read,
2029         .llseek  = seq_lseek,
2030         .release = seq_release_private,
2031 };
2032
2033 static const struct file_operations gfs2_glstats_fops = {
2034         .owner   = THIS_MODULE,
2035         .open    = gfs2_glstats_open,
2036         .read    = seq_read,
2037         .llseek  = seq_lseek,
2038         .release = seq_release_private,
2039 };
2040
2041 static const struct file_operations gfs2_sbstats_fops = {
2042         .owner   = THIS_MODULE,
2043         .open    = gfs2_sbstats_open,
2044         .read    = seq_read,
2045         .llseek  = seq_lseek,
2046         .release = seq_release_private,
2047 };
2048
2049 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2050 {
2051         struct dentry *dent;
2052
2053         dent = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2054         if (IS_ERR_OR_NULL(dent))
2055                 goto fail;
2056         sdp->debugfs_dir = dent;
2057
2058         dent = debugfs_create_file("glocks",
2059                                    S_IFREG | S_IRUGO,
2060                                    sdp->debugfs_dir, sdp,
2061                                    &gfs2_glocks_fops);
2062         if (IS_ERR_OR_NULL(dent))
2063                 goto fail;
2064         sdp->debugfs_dentry_glocks = dent;
2065
2066         dent = debugfs_create_file("glstats",
2067                                    S_IFREG | S_IRUGO,
2068                                    sdp->debugfs_dir, sdp,
2069                                    &gfs2_glstats_fops);
2070         if (IS_ERR_OR_NULL(dent))
2071                 goto fail;
2072         sdp->debugfs_dentry_glstats = dent;
2073
2074         dent = debugfs_create_file("sbstats",
2075                                    S_IFREG | S_IRUGO,
2076                                    sdp->debugfs_dir, sdp,
2077                                    &gfs2_sbstats_fops);
2078         if (IS_ERR_OR_NULL(dent))
2079                 goto fail;
2080         sdp->debugfs_dentry_sbstats = dent;
2081
2082         return 0;
2083 fail:
2084         gfs2_delete_debugfs_file(sdp);
2085         return dent ? PTR_ERR(dent) : -ENOMEM;
2086 }
2087
2088 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2089 {
2090         if (sdp->debugfs_dir) {
2091                 if (sdp->debugfs_dentry_glocks) {
2092                         debugfs_remove(sdp->debugfs_dentry_glocks);
2093                         sdp->debugfs_dentry_glocks = NULL;
2094                 }
2095                 if (sdp->debugfs_dentry_glstats) {
2096                         debugfs_remove(sdp->debugfs_dentry_glstats);
2097                         sdp->debugfs_dentry_glstats = NULL;
2098                 }
2099                 if (sdp->debugfs_dentry_sbstats) {
2100                         debugfs_remove(sdp->debugfs_dentry_sbstats);
2101                         sdp->debugfs_dentry_sbstats = NULL;
2102                 }
2103                 debugfs_remove(sdp->debugfs_dir);
2104                 sdp->debugfs_dir = NULL;
2105         }
2106 }
2107
2108 int gfs2_register_debugfs(void)
2109 {
2110         gfs2_root = debugfs_create_dir("gfs2", NULL);
2111         if (IS_ERR(gfs2_root))
2112                 return PTR_ERR(gfs2_root);
2113         return gfs2_root ? 0 : -ENOMEM;
2114 }
2115
2116 void gfs2_unregister_debugfs(void)
2117 {
2118         debugfs_remove(gfs2_root);
2119         gfs2_root = NULL;
2120 }