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         if (!list_empty(&gl->gl_lru)) {
 179                 list_del_init(&gl->gl_lru);
 180                 atomic_dec(&lru_count);
 181                 clear_bit(GLF_LRU, &gl->gl_flags);
 182         }
 183 }
 184
 185 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
 186 {
 187         spin_lock(&lru_lock);
 188         __gfs2_glock_remove_from_lru(gl);
 189         spin_unlock(&lru_lock);
 190 }
 191
 192 /**
 193  * gfs2_glock_put() - Decrement reference count on glock
 194  * @gl: The glock to put
 195  *
 196  */
 197
 198 void gfs2_glock_put(struct gfs2_glock *gl)
 199 {
 200         struct gfs2_sbd *sdp = gl->gl_sbd;
 201         struct address_space *mapping = gfs2_glock2aspace(gl);
 202
 203         if (lockref_put_or_lock(&gl->gl_lockref))
 204                 return;
 205
 206         lockref_mark_dead(&gl->gl_lockref);
 207
 208         spin_lock(&lru_lock);
 209         __gfs2_glock_remove_from_lru(gl);
 210         spin_unlock(&lru_lock);
 211         spin_unlock(&gl->gl_lockref.lock);
 212         spin_lock_bucket(gl->gl_hash);
 213         hlist_bl_del_rcu(&gl->gl_list);
 214         spin_unlock_bucket(gl->gl_hash);
 215         GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
 216         GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
 217         trace_gfs2_glock_put(gl);
 218         sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
 219 }
 220
 221 /**
 222  * search_bucket() - Find struct gfs2_glock by lock number
 223  * @bucket: the bucket to search
 224  * @name: The lock name
 225  *
 226  * Returns: NULL, or the struct gfs2_glock with the requested number
 227  */
 228
 229 static struct gfs2_glock *search_bucket(unsigned int hash,
 230                                         const struct gfs2_sbd *sdp,
 231                                         const struct lm_lockname *name)
 232 {
 233         struct gfs2_glock *gl;
 234         struct hlist_bl_node *h;
 235
 236         hlist_bl_for_each_entry_rcu(gl, h, &gl_hash_table[hash], gl_list) {
 237                 if (!lm_name_equal(&gl->gl_name, name))
 238                         continue;
 239                 if (gl->gl_sbd != sdp)
 240                         continue;
 241                 if (lockref_get_not_dead(&gl->gl_lockref))
 242                         return gl;
 243         }
 244
 245         return NULL;
 246 }
 247
 248 /**
 249  * may_grant - check if its ok to grant a new lock
 250  * @gl: The glock
 251  * @gh: The lock request which we wish to grant
 252  *
 253  * Returns: true if its ok to grant the lock
 254  */
 255
 256 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
 257 {
 258         const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
 259         if ((gh->gh_state == LM_ST_EXCLUSIVE ||
 260              gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
 261                 return 0;
 262         if (gl->gl_state == gh->gh_state)
 263                 return 1;
 264         if (gh->gh_flags & GL_EXACT)
 265                 return 0;
 266         if (gl->gl_state == LM_ST_EXCLUSIVE) {
 267                 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
 268                         return 1;
 269                 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
 270                         return 1;
 271         }
 272         if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
 273                 return 1;
 274         return 0;
 275 }
 276
 277 static void gfs2_holder_wake(struct gfs2_holder *gh)
 278 {
 279         clear_bit(HIF_WAIT, &gh->gh_iflags);
 280         smp_mb__after_atomic();
 281         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
 282 }
 283
 284 /**
 285  * do_error - Something unexpected has happened during a lock request
 286  *
 287  */
 288
 289 static inline void do_error(struct gfs2_glock *gl, const int ret)
 290 {
 291         struct gfs2_holder *gh, *tmp;
 292
 293         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
 294                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 295                         continue;
 296                 if (ret & LM_OUT_ERROR)
 297                         gh->gh_error = -EIO;
 298                 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
 299                         gh->gh_error = GLR_TRYFAILED;
 300                 else
 301                         continue;
 302                 list_del_init(&gh->gh_list);
 303                 trace_gfs2_glock_queue(gh, 0);
 304                 gfs2_holder_wake(gh);
 305         }
 306 }
 307
 308 /**
 309  * do_promote - promote as many requests as possible on the current queue
 310  * @gl: The glock
 311  *
 312  * Returns: 1 if there is a blocked holder at the head of the list, or 2
 313  *          if a type specific operation is underway.
 314  */
 315
 316 static int do_promote(struct gfs2_glock *gl)
 317 __releases(&gl->gl_spin)
 318 __acquires(&gl->gl_spin)
 319 {
 320         const struct gfs2_glock_operations *glops = gl->gl_ops;
 321         struct gfs2_holder *gh, *tmp;
 322         int ret;
 323
 324 restart:
 325         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
 326                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 327                         continue;
 328                 if (may_grant(gl, gh)) {
 329                         if (gh->gh_list.prev == &gl->gl_holders &&
 330                             glops->go_lock) {
 331                                 spin_unlock(&gl->gl_spin);
 332                                 /* FIXME: eliminate this eventually */
 333                                 ret = glops->go_lock(gh);
 334                                 spin_lock(&gl->gl_spin);
 335                                 if (ret) {
 336                                         if (ret == 1)
 337                                                 return 2;
 338                                         gh->gh_error = ret;
 339                                         list_del_init(&gh->gh_list);
 340                                         trace_gfs2_glock_queue(gh, 0);
 341                                         gfs2_holder_wake(gh);
 342                                         goto restart;
 343                                 }
 344                                 set_bit(HIF_HOLDER, &gh->gh_iflags);
 345                                 trace_gfs2_promote(gh, 1);
 346                                 gfs2_holder_wake(gh);
 347                                 goto restart;
 348                         }
 349                         set_bit(HIF_HOLDER, &gh->gh_iflags);
 350                         trace_gfs2_promote(gh, 0);
 351                         gfs2_holder_wake(gh);
 352                         continue;
 353                 }
 354                 if (gh->gh_list.prev == &gl->gl_holders)
 355                         return 1;
 356                 do_error(gl, 0);
 357                 break;
 358         }
 359         return 0;
 360 }
 361
 362 /**
 363  * find_first_waiter - find the first gh that's waiting for the glock
 364  * @gl: the glock
 365  */
 366
 367 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
 368 {
 369         struct gfs2_holder *gh;
 370
 371         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
 372                 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
 373                         return gh;
 374         }
 375         return NULL;
 376 }
 377
 378 /**
 379  * state_change - record that the glock is now in a different state
 380  * @gl: the glock
 381  * @new_state the new state
 382  *
 383  */
 384
 385 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
 386 {
 387         int held1, held2;
 388
 389         held1 = (gl->gl_state != LM_ST_UNLOCKED);
 390         held2 = (new_state != LM_ST_UNLOCKED);
 391
 392         if (held1 != held2) {
 393                 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
 394                 if (held2)
 395                         gl->gl_lockref.count++;
 396                 else
 397                         gl->gl_lockref.count--;
 398         }
 399         if (held1 && held2 && list_empty(&gl->gl_holders))
 400                 clear_bit(GLF_QUEUED, &gl->gl_flags);
 401
 402         if (new_state != gl->gl_target)
 403                 /* shorten our minimum hold time */
 404                 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
 405                                        GL_GLOCK_MIN_HOLD);
 406         gl->gl_state = new_state;
 407         gl->gl_tchange = jiffies;
 408 }
 409
 410 static void gfs2_demote_wake(struct gfs2_glock *gl)
 411 {
 412         gl->gl_demote_state = LM_ST_EXCLUSIVE;
 413         clear_bit(GLF_DEMOTE, &gl->gl_flags);
 414         smp_mb__after_atomic();
 415         wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
 416 }
 417
 418 /**
 419  * finish_xmote - The DLM has replied to one of our lock requests
 420  * @gl: The glock
 421  * @ret: The status from the DLM
 422  *
 423  */
 424
 425 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
 426 {
 427         const struct gfs2_glock_operations *glops = gl->gl_ops;
 428         struct gfs2_holder *gh;
 429         unsigned state = ret & LM_OUT_ST_MASK;
 430         int rv;
 431
 432         spin_lock(&gl->gl_spin);
 433         trace_gfs2_glock_state_change(gl, state);
 434         state_change(gl, state);
 435         gh = find_first_waiter(gl);
 436
 437         /* Demote to UN request arrived during demote to SH or DF */
 438         if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
 439             state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
 440                 gl->gl_target = LM_ST_UNLOCKED;
 441
 442         /* Check for state != intended state */
 443         if (unlikely(state != gl->gl_target)) {
 444                 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
 445                         /* move to back of queue and try next entry */
 446                         if (ret & LM_OUT_CANCELED) {
 447                                 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
 448                                         list_move_tail(&gh->gh_list, &gl->gl_holders);
 449                                 gh = find_first_waiter(gl);
 450                                 gl->gl_target = gh->gh_state;
 451                                 goto retry;
 452                         }
 453                         /* Some error or failed "try lock" - report it */
 454                         if ((ret & LM_OUT_ERROR) ||
 455                             (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
 456                                 gl->gl_target = gl->gl_state;
 457                                 do_error(gl, ret);
 458                                 goto out;
 459                         }
 460                 }
 461                 switch(state) {
 462                 /* Unlocked due to conversion deadlock, try again */
 463                 case LM_ST_UNLOCKED:
 464 retry:
 465                         do_xmote(gl, gh, gl->gl_target);
 466                         break;
 467                 /* Conversion fails, unlock and try again */
 468                 case LM_ST_SHARED:
 469                 case LM_ST_DEFERRED:
 470                         do_xmote(gl, gh, LM_ST_UNLOCKED);
 471                         break;
 472                 default: /* Everything else */
 473                         pr_err("wanted %u got %u\n", gl->gl_target, state);
 474                         GLOCK_BUG_ON(gl, 1);
 475                 }
 476                 spin_unlock(&gl->gl_spin);
 477                 return;
 478         }
 479
 480         /* Fast path - we got what we asked for */
 481         if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
 482                 gfs2_demote_wake(gl);
 483         if (state != LM_ST_UNLOCKED) {
 484                 if (glops->go_xmote_bh) {
 485                         spin_unlock(&gl->gl_spin);
 486                         rv = glops->go_xmote_bh(gl, gh);
 487                         spin_lock(&gl->gl_spin);
 488                         if (rv) {
 489                                 do_error(gl, rv);
 490                                 goto out;
 491                         }
 492                 }
 493                 rv = do_promote(gl);
 494                 if (rv == 2)
 495                         goto out_locked;
 496         }
 497 out:
 498         clear_bit(GLF_LOCK, &gl->gl_flags);
 499 out_locked:
 500         spin_unlock(&gl->gl_spin);
 501 }
 502
 503 /**
 504  * do_xmote - Calls the DLM to change the state of a lock
 505  * @gl: The lock state
 506  * @gh: The holder (only for promotes)
 507  * @target: The target lock state
 508  *
 509  */
 510
 511 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
 512 __releases(&gl->gl_spin)
 513 __acquires(&gl->gl_spin)
 514 {
 515         const struct gfs2_glock_operations *glops = gl->gl_ops;
 516         struct gfs2_sbd *sdp = gl->gl_sbd;
 517         unsigned int lck_flags = gh ? gh->gh_flags : 0;
 518         int ret;
 519
 520         lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
 521                       LM_FLAG_PRIORITY);
 522         GLOCK_BUG_ON(gl, gl->gl_state == target);
 523         GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
 524         if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
 525             glops->go_inval) {
 526                 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 527                 do_error(gl, 0); /* Fail queued try locks */
 528         }
 529         gl->gl_req = target;
 530         set_bit(GLF_BLOCKING, &gl->gl_flags);
 531         if ((gl->gl_req == LM_ST_UNLOCKED) ||
 532             (gl->gl_state == LM_ST_EXCLUSIVE) ||
 533             (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
 534                 clear_bit(GLF_BLOCKING, &gl->gl_flags);
 535         spin_unlock(&gl->gl_spin);
 536         if (glops->go_sync)
 537                 glops->go_sync(gl);
 538         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 539                 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
 540         clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 541
 542         gfs2_glock_hold(gl);
 543         if (sdp->sd_lockstruct.ls_ops->lm_lock) {
 544                 /* lock_dlm */
 545                 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
 546                 if (ret) {
 547                         pr_err("lm_lock ret %d\n", ret);
 548                         GLOCK_BUG_ON(gl, 1);
 549                 }
 550         } else { /* lock_nolock */
 551                 finish_xmote(gl, target);
 552                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 553                         gfs2_glock_put(gl);
 554         }
 555
 556         spin_lock(&gl->gl_spin);
 557 }
 558
 559 /**
 560  * find_first_holder - find the first "holder" gh
 561  * @gl: the glock
 562  */
 563
 564 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
 565 {
 566         struct gfs2_holder *gh;
 567
 568         if (!list_empty(&gl->gl_holders)) {
 569                 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
 570                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 571                         return gh;
 572         }
 573         return NULL;
 574 }
 575
 576 /**
 577  * run_queue - do all outstanding tasks related to a glock
 578  * @gl: The glock in question
 579  * @nonblock: True if we must not block in run_queue
 580  *
 581  */
 582
 583 static void run_queue(struct gfs2_glock *gl, const int nonblock)
 584 __releases(&gl->gl_spin)
 585 __acquires(&gl->gl_spin)
 586 {
 587         struct gfs2_holder *gh = NULL;
 588         int ret;
 589
 590         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
 591                 return;
 592
 593         GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
 594
 595         if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
 596             gl->gl_demote_state != gl->gl_state) {
 597                 if (find_first_holder(gl))
 598                         goto out_unlock;
 599                 if (nonblock)
 600                         goto out_sched;
 601                 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
 602                 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
 603                 gl->gl_target = gl->gl_demote_state;
 604         } else {
 605                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
 606                         gfs2_demote_wake(gl);
 607                 ret = do_promote(gl);
 608                 if (ret == 0)
 609                         goto out_unlock;
 610                 if (ret == 2)
 611                         goto out;
 612                 gh = find_first_waiter(gl);
 613                 gl->gl_target = gh->gh_state;
 614                 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
 615                         do_error(gl, 0); /* Fail queued try locks */
 616         }
 617         do_xmote(gl, gh, gl->gl_target);
 618 out:
 619         return;
 620
 621 out_sched:
 622         clear_bit(GLF_LOCK, &gl->gl_flags);
 623         smp_mb__after_atomic();
 624         gl->gl_lockref.count++;
 625         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 626                 gl->gl_lockref.count--;
 627         return;
 628
 629 out_unlock:
 630         clear_bit(GLF_LOCK, &gl->gl_flags);
 631         smp_mb__after_atomic();
 632         return;
 633 }
 634
 635 static void delete_work_func(struct work_struct *work)
 636 {
 637         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
 638         struct gfs2_sbd *sdp = gl->gl_sbd;
 639         struct gfs2_inode *ip;
 640         struct inode *inode;
 641         u64 no_addr = gl->gl_name.ln_number;
 642
 643         ip = gl->gl_object;
 644         /* Note: Unsafe to dereference ip as we don't hold right refs/locks */
 645
 646         if (ip)
 647                 inode = gfs2_ilookup(sdp->sd_vfs, no_addr, 1);
 648         else
 649                 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
 650         if (inode && !IS_ERR(inode)) {
 651                 d_prune_aliases(inode);
 652                 iput(inode);
 653         }
 654         gfs2_glock_put(gl);
 655 }
 656
 657 static void glock_work_func(struct work_struct *work)
 658 {
 659         unsigned long delay = 0;
 660         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
 661         int drop_ref = 0;
 662
 663         if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
 664                 finish_xmote(gl, gl->gl_reply);
 665                 drop_ref = 1;
 666         }
 667         spin_lock(&gl->gl_spin);
 668         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
 669             gl->gl_state != LM_ST_UNLOCKED &&
 670             gl->gl_demote_state != LM_ST_EXCLUSIVE) {
 671                 unsigned long holdtime, now = jiffies;
 672
 673                 holdtime = gl->gl_tchange + gl->gl_hold_time;
 674                 if (time_before(now, holdtime))
 675                         delay = holdtime - now;
 676
 677                 if (!delay) {
 678                         clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
 679                         set_bit(GLF_DEMOTE, &gl->gl_flags);
 680                 }
 681         }
 682         run_queue(gl, 0);
 683         spin_unlock(&gl->gl_spin);
 684         if (!delay)
 685                 gfs2_glock_put(gl);
 686         else {
 687                 if (gl->gl_name.ln_type != LM_TYPE_INODE)
 688                         delay = 0;
 689                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
 690                         gfs2_glock_put(gl);
 691         }
 692         if (drop_ref)
 693                 gfs2_glock_put(gl);
 694 }
 695
 696 /**
 697  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
 698  * @sdp: The GFS2 superblock
 699  * @number: the lock number
 700  * @glops: The glock_operations to use
 701  * @create: If 0, don't create the glock if it doesn't exist
 702  * @glp: the glock is returned here
 703  *
 704  * This does not lock a glock, just finds/creates structures for one.
 705  *
 706  * Returns: errno
 707  */
 708
 709 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
 710                    const struct gfs2_glock_operations *glops, int create,
 711                    struct gfs2_glock **glp)
 712 {
 713         struct super_block *s = sdp->sd_vfs;
 714         struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
 715         struct gfs2_glock *gl, *tmp;
 716         unsigned int hash = gl_hash(sdp, &name);
 717         struct address_space *mapping;
 718         struct kmem_cache *cachep;
 719
 720         rcu_read_lock();
 721         gl = search_bucket(hash, sdp, &name);
 722         rcu_read_unlock();
 723
 724         *glp = gl;
 725         if (gl)
 726                 return 0;
 727         if (!create)
 728                 return -ENOENT;
 729
 730         if (glops->go_flags & GLOF_ASPACE)
 731                 cachep = gfs2_glock_aspace_cachep;
 732         else
 733                 cachep = gfs2_glock_cachep;
 734         gl = kmem_cache_alloc(cachep, GFP_KERNEL);
 735         if (!gl)
 736                 return -ENOMEM;
 737
 738         memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
 739
 740         if (glops->go_flags & GLOF_LVB) {
 741                 gl->gl_lksb.sb_lvbptr = kzalloc(GFS2_MIN_LVB_SIZE, GFP_KERNEL);
 742                 if (!gl->gl_lksb.sb_lvbptr) {
 743                         kmem_cache_free(cachep, gl);
 744                         return -ENOMEM;
 745                 }
 746         }
 747
 748         atomic_inc(&sdp->sd_glock_disposal);
 749         gl->gl_sbd = sdp;
 750         gl->gl_flags = 0;
 751         gl->gl_name = name;
 752         gl->gl_lockref.count = 1;
 753         gl->gl_state = LM_ST_UNLOCKED;
 754         gl->gl_target = LM_ST_UNLOCKED;
 755         gl->gl_demote_state = LM_ST_EXCLUSIVE;
 756         gl->gl_hash = hash;
 757         gl->gl_ops = glops;
 758         gl->gl_dstamp = ktime_set(0, 0);
 759         preempt_disable();
 760         /* We use the global stats to estimate the initial per-glock stats */
 761         gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
 762         preempt_enable();
 763         gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
 764         gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
 765         gl->gl_tchange = jiffies;
 766         gl->gl_object = NULL;
 767         gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
 768         INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
 769         INIT_WORK(&gl->gl_delete, delete_work_func);
 770
 771         mapping = gfs2_glock2aspace(gl);
 772         if (mapping) {
 773                 mapping->a_ops = &gfs2_meta_aops;
 774                 mapping->host = s->s_bdev->bd_inode;
 775                 mapping->flags = 0;
 776                 mapping_set_gfp_mask(mapping, GFP_NOFS);
 777                 mapping->private_data = NULL;
 778                 mapping->backing_dev_info = s->s_bdi;
 779                 mapping->writeback_index = 0;
 780         }
 781
 782         spin_lock_bucket(hash);
 783         tmp = search_bucket(hash, sdp, &name);
 784         if (tmp) {
 785                 spin_unlock_bucket(hash);
 786                 kfree(gl->gl_lksb.sb_lvbptr);
 787                 kmem_cache_free(cachep, gl);
 788                 atomic_dec(&sdp->sd_glock_disposal);
 789                 gl = tmp;
 790         } else {
 791                 hlist_bl_add_head_rcu(&gl->gl_list, &gl_hash_table[hash]);
 792                 spin_unlock_bucket(hash);
 793         }
 794
 795         *glp = gl;
 796
 797         return 0;
 798 }
 799
 800 /**
 801  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
 802  * @gl: the glock
 803  * @state: the state we're requesting
 804  * @flags: the modifier flags
 805  * @gh: the holder structure
 806  *
 807  */
 808
 809 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
 810                       struct gfs2_holder *gh)
 811 {
 812         INIT_LIST_HEAD(&gh->gh_list);
 813         gh->gh_gl = gl;
 814         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 815         gh->gh_owner_pid = get_pid(task_pid(current));
 816         gh->gh_state = state;
 817         gh->gh_flags = flags;
 818         gh->gh_error = 0;
 819         gh->gh_iflags = 0;
 820         gfs2_glock_hold(gl);
 821 }
 822
 823 /**
 824  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
 825  * @state: the state we're requesting
 826  * @flags: the modifier flags
 827  * @gh: the holder structure
 828  *
 829  * Don't mess with the glock.
 830  *
 831  */
 832
 833 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
 834 {
 835         gh->gh_state = state;
 836         gh->gh_flags = flags;
 837         gh->gh_iflags = 0;
 838         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 839         if (gh->gh_owner_pid)
 840                 put_pid(gh->gh_owner_pid);
 841         gh->gh_owner_pid = get_pid(task_pid(current));
 842 }
 843
 844 /**
 845  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
 846  * @gh: the holder structure
 847  *
 848  */
 849
 850 void gfs2_holder_uninit(struct gfs2_holder *gh)
 851 {
 852         put_pid(gh->gh_owner_pid);
 853         gfs2_glock_put(gh->gh_gl);
 854         gh->gh_gl = NULL;
 855         gh->gh_ip = 0;
 856 }
 857
 858 /**
 859  * gfs2_glock_holder_wait
 860  * @word: unused
 861  *
 862  * This function and gfs2_glock_demote_wait both show up in the WCHAN
 863  * field. Thus I've separated these otherwise identical functions in
 864  * order to be more informative to the user.
 865  */
 866
 867 static int gfs2_glock_holder_wait(void *word)
 868 {
 869         schedule();
 870         return 0;
 871 }
 872
 873 static int gfs2_glock_demote_wait(void *word)
 874 {
 875         schedule();
 876         return 0;
 877 }
 878
 879 /**
 880  * gfs2_glock_wait - wait on a glock acquisition
 881  * @gh: the glock holder
 882  *
 883  * Returns: 0 on success
 884  */
 885
 886 int gfs2_glock_wait(struct gfs2_holder *gh)
 887 {
 888         unsigned long time1 = jiffies;
 889
 890         might_sleep();
 891         wait_on_bit(&gh->gh_iflags, HIF_WAIT, gfs2_glock_holder_wait, TASK_UNINTERRUPTIBLE);
 892         if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
 893                 /* Lengthen the minimum hold time. */
 894                 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
 895                                               GL_GLOCK_HOLD_INCR,
 896                                               GL_GLOCK_MAX_HOLD);
 897         return gh->gh_error;
 898 }
 899
 900 /**
 901  * handle_callback - process a demote request
 902  * @gl: the glock
 903  * @state: the state the caller wants us to change to
 904  *
 905  * There are only two requests that we are going to see in actual
 906  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
 907  */
 908
 909 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
 910                             unsigned long delay, bool remote)
 911 {
 912         int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
 913
 914         set_bit(bit, &gl->gl_flags);
 915         if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
 916                 gl->gl_demote_state = state;
 917                 gl->gl_demote_time = jiffies;
 918         } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
 919                         gl->gl_demote_state != state) {
 920                 gl->gl_demote_state = LM_ST_UNLOCKED;
 921         }
 922         if (gl->gl_ops->go_callback)
 923                 gl->gl_ops->go_callback(gl, remote);
 924         trace_gfs2_demote_rq(gl, remote);
 925 }
 926
 927 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
 928 {
 929         struct va_format vaf;
 930         va_list args;
 931
 932         va_start(args, fmt);
 933
 934         if (seq) {
 935                 seq_vprintf(seq, fmt, args);
 936         } else {
 937                 vaf.fmt = fmt;
 938                 vaf.va = &args;
 939
 940                 pr_err("%pV", &vaf);
 941         }
 942
 943         va_end(args);
 944 }
 945
 946 /**
 947  * add_to_queue - Add a holder to the wait queue (but look for recursion)
 948  * @gh: the holder structure to add
 949  *
 950  * Eventually we should move the recursive locking trap to a
 951  * debugging option or something like that. This is the fast
 952  * path and needs to have the minimum number of distractions.
 953  *
 954  */
 955
 956 static inline void add_to_queue(struct gfs2_holder *gh)
 957 __releases(&gl->gl_spin)
 958 __acquires(&gl->gl_spin)
 959 {
 960         struct gfs2_glock *gl = gh->gh_gl;
 961         struct gfs2_sbd *sdp = gl->gl_sbd;
 962         struct list_head *insert_pt = NULL;
 963         struct gfs2_holder *gh2;
 964         int try_futile = 0;
 965
 966         BUG_ON(gh->gh_owner_pid == NULL);
 967         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
 968                 BUG();
 969
 970         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
 971                 if (test_bit(GLF_LOCK, &gl->gl_flags))
 972                         try_futile = !may_grant(gl, gh);
 973                 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 974                         goto fail;
 975         }
 976
 977         list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
 978                 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
 979                     (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
 980                         goto trap_recursive;
 981                 if (try_futile &&
 982                     !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
 983 fail:
 984                         gh->gh_error = GLR_TRYFAILED;
 985                         gfs2_holder_wake(gh);
 986                         return;
 987                 }
 988                 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
 989                         continue;
 990                 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
 991                         insert_pt = &gh2->gh_list;
 992         }
 993         set_bit(GLF_QUEUED, &gl->gl_flags);
 994         trace_gfs2_glock_queue(gh, 1);
 995         gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
 996         gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
 997         if (likely(insert_pt == NULL)) {
 998                 list_add_tail(&gh->gh_list, &gl->gl_holders);
 999                 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1000                         goto do_cancel;
1001                 return;
1002         }
1003         list_add_tail(&gh->gh_list, insert_pt);
1004 do_cancel:
1005         gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
1006         if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1007                 spin_unlock(&gl->gl_spin);
1008                 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1009                         sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1010                 spin_lock(&gl->gl_spin);
1011         }
1012         return;
1013
1014 trap_recursive:
1015         pr_err("original: %pSR\n", (void *)gh2->gh_ip);
1016         pr_err("pid: %d\n", pid_nr(gh2->gh_owner_pid));
1017         pr_err("lock type: %d req lock state : %d\n",
1018                gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1019         pr_err("new: %pSR\n", (void *)gh->gh_ip);
1020         pr_err("pid: %d\n", pid_nr(gh->gh_owner_pid));
1021         pr_err("lock type: %d req lock state : %d\n",
1022                gh->gh_gl->gl_name.ln_type, gh->gh_state);
1023         gfs2_dump_glock(NULL, gl);
1024         BUG();
1025 }
1026
1027 /**
1028  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1029  * @gh: the holder structure
1030  *
1031  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1032  *
1033  * Returns: 0, GLR_TRYFAILED, or errno on failure
1034  */
1035
1036 int gfs2_glock_nq(struct gfs2_holder *gh)
1037 {
1038         struct gfs2_glock *gl = gh->gh_gl;
1039         struct gfs2_sbd *sdp = gl->gl_sbd;
1040         int error = 0;
1041
1042         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1043                 return -EIO;
1044
1045         if (test_bit(GLF_LRU, &gl->gl_flags))
1046                 gfs2_glock_remove_from_lru(gl);
1047
1048         spin_lock(&gl->gl_spin);
1049         add_to_queue(gh);
1050         if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1051                      test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1052                 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1053                 gl->gl_lockref.count++;
1054                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1055                         gl->gl_lockref.count--;
1056         }
1057         run_queue(gl, 1);
1058         spin_unlock(&gl->gl_spin);
1059
1060         if (!(gh->gh_flags & GL_ASYNC))
1061                 error = gfs2_glock_wait(gh);
1062
1063         return error;
1064 }
1065
1066 /**
1067  * gfs2_glock_poll - poll to see if an async request has been completed
1068  * @gh: the holder
1069  *
1070  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1071  */
1072
1073 int gfs2_glock_poll(struct gfs2_holder *gh)
1074 {
1075         return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1076 }
1077
1078 /**
1079  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1080  * @gh: the glock holder
1081  *
1082  */
1083
1084 void gfs2_glock_dq(struct gfs2_holder *gh)
1085 {
1086         struct gfs2_glock *gl = gh->gh_gl;
1087         const struct gfs2_glock_operations *glops = gl->gl_ops;
1088         unsigned delay = 0;
1089         int fast_path = 0;
1090
1091         spin_lock(&gl->gl_spin);
1092         if (gh->gh_flags & GL_NOCACHE)
1093                 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1094
1095         list_del_init(&gh->gh_list);
1096         if (find_first_holder(gl) == NULL) {
1097                 if (glops->go_unlock) {
1098                         GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1099                         spin_unlock(&gl->gl_spin);
1100                         glops->go_unlock(gh);
1101                         spin_lock(&gl->gl_spin);
1102                         clear_bit(GLF_LOCK, &gl->gl_flags);
1103                 }
1104                 if (list_empty(&gl->gl_holders) &&
1105                     !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1106                     !test_bit(GLF_DEMOTE, &gl->gl_flags))
1107                         fast_path = 1;
1108         }
1109         if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1110                 gfs2_glock_add_to_lru(gl);
1111
1112         trace_gfs2_glock_queue(gh, 0);
1113         spin_unlock(&gl->gl_spin);
1114         if (likely(fast_path))
1115                 return;
1116
1117         gfs2_glock_hold(gl);
1118         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1119             !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1120             gl->gl_name.ln_type == LM_TYPE_INODE)
1121                 delay = gl->gl_hold_time;
1122         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1123                 gfs2_glock_put(gl);
1124 }
1125
1126 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1127 {
1128         struct gfs2_glock *gl = gh->gh_gl;
1129         gfs2_glock_dq(gh);
1130         might_sleep();
1131         if (!find_first_holder(gl))
1132                 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, gfs2_glock_demote_wait,
1133                             TASK_UNINTERRUPTIBLE);
1134 }
1135
1136 /**
1137  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1138  * @gh: the holder structure
1139  *
1140  */
1141
1142 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1143 {
1144         gfs2_glock_dq(gh);
1145         gfs2_holder_uninit(gh);
1146 }
1147
1148 /**
1149  * gfs2_glock_nq_num - acquire a glock based on lock number
1150  * @sdp: the filesystem
1151  * @number: the lock number
1152  * @glops: the glock operations for the type of glock
1153  * @state: the state to acquire the glock in
1154  * @flags: modifier flags for the acquisition
1155  * @gh: the struct gfs2_holder
1156  *
1157  * Returns: errno
1158  */
1159
1160 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1161                       const struct gfs2_glock_operations *glops,
1162                       unsigned int state, int flags, struct gfs2_holder *gh)
1163 {
1164         struct gfs2_glock *gl;
1165         int error;
1166
1167         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1168         if (!error) {
1169                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1170                 gfs2_glock_put(gl);
1171         }
1172
1173         return error;
1174 }
1175
1176 /**
1177  * glock_compare - Compare two struct gfs2_glock structures for sorting
1178  * @arg_a: the first structure
1179  * @arg_b: the second structure
1180  *
1181  */
1182
1183 static int glock_compare(const void *arg_a, const void *arg_b)
1184 {
1185         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1186         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1187         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1188         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1189
1190         if (a->ln_number > b->ln_number)
1191                 return 1;
1192         if (a->ln_number < b->ln_number)
1193                 return -1;
1194         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1195         return 0;
1196 }
1197
1198 /**
1199  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1200  * @num_gh: the number of structures
1201  * @ghs: an array of struct gfs2_holder structures
1202  *
1203  * Returns: 0 on success (all glocks acquired),
1204  *          errno on failure (no glocks acquired)
1205  */
1206
1207 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1208                      struct gfs2_holder **p)
1209 {
1210         unsigned int x;
1211         int error = 0;
1212
1213         for (x = 0; x < num_gh; x++)
1214                 p[x] = &ghs[x];
1215
1216         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1217
1218         for (x = 0; x < num_gh; x++) {
1219                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1220
1221                 error = gfs2_glock_nq(p[x]);
1222                 if (error) {
1223                         while (x--)
1224                                 gfs2_glock_dq(p[x]);
1225                         break;
1226                 }
1227         }
1228
1229         return error;
1230 }
1231
1232 /**
1233  * gfs2_glock_nq_m - acquire multiple glocks
1234  * @num_gh: the number of structures
1235  * @ghs: an array of struct gfs2_holder structures
1236  *
1237  *
1238  * Returns: 0 on success (all glocks acquired),
1239  *          errno on failure (no glocks acquired)
1240  */
1241
1242 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1243 {
1244         struct gfs2_holder *tmp[4];
1245         struct gfs2_holder **pph = tmp;
1246         int error = 0;
1247
1248         switch(num_gh) {
1249         case 0:
1250                 return 0;
1251         case 1:
1252                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1253                 return gfs2_glock_nq(ghs);
1254         default:
1255                 if (num_gh <= 4)
1256                         break;
1257                 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1258                 if (!pph)
1259                         return -ENOMEM;
1260         }
1261
1262         error = nq_m_sync(num_gh, ghs, pph);
1263
1264         if (pph != tmp)
1265                 kfree(pph);
1266
1267         return error;
1268 }
1269
1270 /**
1271  * gfs2_glock_dq_m - release multiple glocks
1272  * @num_gh: the number of structures
1273  * @ghs: an array of struct gfs2_holder structures
1274  *
1275  */
1276
1277 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1278 {
1279         while (num_gh--)
1280                 gfs2_glock_dq(&ghs[num_gh]);
1281 }
1282
1283 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1284 {
1285         unsigned long delay = 0;
1286         unsigned long holdtime;
1287         unsigned long now = jiffies;
1288
1289         gfs2_glock_hold(gl);
1290         holdtime = gl->gl_tchange + gl->gl_hold_time;
1291         if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1292             gl->gl_name.ln_type == LM_TYPE_INODE) {
1293                 if (time_before(now, holdtime))
1294                         delay = holdtime - now;
1295                 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1296                         delay = gl->gl_hold_time;
1297         }
1298
1299         spin_lock(&gl->gl_spin);
1300         handle_callback(gl, state, delay, true);
1301         spin_unlock(&gl->gl_spin);
1302         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1303                 gfs2_glock_put(gl);
1304 }
1305
1306 /**
1307  * gfs2_should_freeze - Figure out if glock should be frozen
1308  * @gl: The glock in question
1309  *
1310  * Glocks are not frozen if (a) the result of the dlm operation is
1311  * an error, (b) the locking operation was an unlock operation or
1312  * (c) if there is a "noexp" flagged request anywhere in the queue
1313  *
1314  * Returns: 1 if freezing should occur, 0 otherwise
1315  */
1316
1317 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1318 {
1319         const struct gfs2_holder *gh;
1320
1321         if (gl->gl_reply & ~LM_OUT_ST_MASK)
1322                 return 0;
1323         if (gl->gl_target == LM_ST_UNLOCKED)
1324                 return 0;
1325
1326         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1327                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1328                         continue;
1329                 if (LM_FLAG_NOEXP & gh->gh_flags)
1330                         return 0;
1331         }
1332
1333         return 1;
1334 }
1335
1336 /**
1337  * gfs2_glock_complete - Callback used by locking
1338  * @gl: Pointer to the glock
1339  * @ret: The return value from the dlm
1340  *
1341  * The gl_reply field is under the gl_spin lock so that it is ok
1342  * to use a bitfield shared with other glock state fields.
1343  */
1344
1345 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1346 {
1347         struct lm_lockstruct *ls = &gl->gl_sbd->sd_lockstruct;
1348
1349         spin_lock(&gl->gl_spin);
1350         gl->gl_reply = ret;
1351
1352         if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1353                 if (gfs2_should_freeze(gl)) {
1354                         set_bit(GLF_FROZEN, &gl->gl_flags);
1355                         spin_unlock(&gl->gl_spin);
1356                         return;
1357                 }
1358         }
1359
1360         gl->gl_lockref.count++;
1361         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1362         spin_unlock(&gl->gl_spin);
1363
1364         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1365                 gfs2_glock_put(gl);
1366 }
1367
1368 static int glock_cmp(void *priv, struct list_head *a, struct list_head *b)
1369 {
1370         struct gfs2_glock *gla, *glb;
1371
1372         gla = list_entry(a, struct gfs2_glock, gl_lru);
1373         glb = list_entry(b, struct gfs2_glock, gl_lru);
1374
1375         if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1376                 return 1;
1377         if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1378                 return -1;
1379
1380         return 0;
1381 }
1382
1383 /**
1384  * gfs2_dispose_glock_lru - Demote a list of glocks
1385  * @list: The list to dispose of
1386  *
1387  * Disposing of glocks may involve disk accesses, so that here we sort
1388  * the glocks by number (i.e. disk location of the inodes) so that if
1389  * there are any such accesses, they'll be sent in order (mostly).
1390  *
1391  * Must be called under the lru_lock, but may drop and retake this
1392  * lock. While the lru_lock is dropped, entries may vanish from the
1393  * list, but no new entries will appear on the list (since it is
1394  * private)
1395  */
1396
1397 static void gfs2_dispose_glock_lru(struct list_head *list)
1398 __releases(&lru_lock)
1399 __acquires(&lru_lock)
1400 {
1401         struct gfs2_glock *gl;
1402
1403         list_sort(NULL, list, glock_cmp);
1404
1405         while(!list_empty(list)) {
1406                 gl = list_entry(list->next, struct gfs2_glock, gl_lru);
1407                 list_del_init(&gl->gl_lru);
1408                 if (!spin_trylock(&gl->gl_spin)) {
1409                         list_add(&gl->gl_lru, &lru_list);
1410                         atomic_inc(&lru_count);
1411                         continue;
1412                 }
1413                 clear_bit(GLF_LRU, &gl->gl_flags);
1414                 spin_unlock(&lru_lock);
1415                 gl->gl_lockref.count++;
1416                 if (demote_ok(gl))
1417                         handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1418                 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1419                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1420                         gl->gl_lockref.count--;
1421                 spin_unlock(&gl->gl_spin);
1422                 spin_lock(&lru_lock);
1423         }
1424 }
1425
1426 /**
1427  * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1428  * @nr: The number of entries to scan
1429  *
1430  * This function selects the entries on the LRU which are able to
1431  * be demoted, and then kicks off the process by calling
1432  * gfs2_dispose_glock_lru() above.
1433  */
1434
1435 static long gfs2_scan_glock_lru(int nr)
1436 {
1437         struct gfs2_glock *gl;
1438         LIST_HEAD(skipped);
1439         LIST_HEAD(dispose);
1440         long freed = 0;
1441
1442         spin_lock(&lru_lock);
1443         while ((nr-- >= 0) && !list_empty(&lru_list)) {
1444                 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1445
1446                 /* Test for being demotable */
1447                 if (!test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1448                         list_move(&gl->gl_lru, &dispose);
1449                         atomic_dec(&lru_count);
1450                         freed++;
1451                         continue;
1452                 }
1453
1454                 list_move(&gl->gl_lru, &skipped);
1455         }
1456         list_splice(&skipped, &lru_list);
1457         if (!list_empty(&dispose))
1458                 gfs2_dispose_glock_lru(&dispose);
1459         spin_unlock(&lru_lock);
1460
1461         return freed;
1462 }
1463
1464 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1465                                             struct shrink_control *sc)
1466 {
1467         if (!(sc->gfp_mask & __GFP_FS))
1468                 return SHRINK_STOP;
1469         return gfs2_scan_glock_lru(sc->nr_to_scan);
1470 }
1471
1472 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1473                                              struct shrink_control *sc)
1474 {
1475         return vfs_pressure_ratio(atomic_read(&lru_count));
1476 }
1477
1478 static struct shrinker glock_shrinker = {
1479         .seeks = DEFAULT_SEEKS,
1480         .count_objects = gfs2_glock_shrink_count,
1481         .scan_objects = gfs2_glock_shrink_scan,
1482 };
1483
1484 /**
1485  * examine_bucket - Call a function for glock in a hash bucket
1486  * @examiner: the function
1487  * @sdp: the filesystem
1488  * @bucket: the bucket
1489  *
1490  */
1491
1492 static void examine_bucket(glock_examiner examiner, const struct gfs2_sbd *sdp,
1493                           unsigned int hash)
1494 {
1495         struct gfs2_glock *gl;
1496         struct hlist_bl_head *head = &gl_hash_table[hash];
1497         struct hlist_bl_node *pos;
1498
1499         rcu_read_lock();
1500         hlist_bl_for_each_entry_rcu(gl, pos, head, gl_list) {
1501                 if ((gl->gl_sbd == sdp) && lockref_get_not_dead(&gl->gl_lockref))
1502                         examiner(gl);
1503         }
1504         rcu_read_unlock();
1505         cond_resched();
1506 }
1507
1508 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1509 {
1510         unsigned x;
1511
1512         for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1513                 examine_bucket(examiner, sdp, x);
1514 }
1515
1516
1517 /**
1518  * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1519  * @gl: The glock to thaw
1520  *
1521  */
1522
1523 static void thaw_glock(struct gfs2_glock *gl)
1524 {
1525         if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1526                 goto out;
1527         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1528         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0) {
1529 out:
1530                 gfs2_glock_put(gl);
1531         }
1532 }
1533
1534 /**
1535  * clear_glock - look at a glock and see if we can free it from glock cache
1536  * @gl: the glock to look at
1537  *
1538  */
1539
1540 static void clear_glock(struct gfs2_glock *gl)
1541 {
1542         gfs2_glock_remove_from_lru(gl);
1543
1544         spin_lock(&gl->gl_spin);
1545         if (gl->gl_state != LM_ST_UNLOCKED)
1546                 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1547         spin_unlock(&gl->gl_spin);
1548         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1549                 gfs2_glock_put(gl);
1550 }
1551
1552 /**
1553  * gfs2_glock_thaw - Thaw any frozen glocks
1554  * @sdp: The super block
1555  *
1556  */
1557
1558 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1559 {
1560         glock_hash_walk(thaw_glock, sdp);
1561 }
1562
1563 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1564 {
1565         spin_lock(&gl->gl_spin);
1566         gfs2_dump_glock(seq, gl);
1567         spin_unlock(&gl->gl_spin);
1568 }
1569
1570 static void dump_glock_func(struct gfs2_glock *gl)
1571 {
1572         dump_glock(NULL, gl);
1573 }
1574
1575 /**
1576  * gfs2_gl_hash_clear - Empty out the glock hash table
1577  * @sdp: the filesystem
1578  * @wait: wait until it's all gone
1579  *
1580  * Called when unmounting the filesystem.
1581  */
1582
1583 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1584 {
1585         set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1586         flush_workqueue(glock_workqueue);
1587         glock_hash_walk(clear_glock, sdp);
1588         flush_workqueue(glock_workqueue);
1589         wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
1590         glock_hash_walk(dump_glock_func, sdp);
1591 }
1592
1593 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1594 {
1595         struct gfs2_glock *gl = ip->i_gl;
1596         int ret;
1597
1598         ret = gfs2_truncatei_resume(ip);
1599         gfs2_assert_withdraw(gl->gl_sbd, ret == 0);
1600
1601         spin_lock(&gl->gl_spin);
1602         clear_bit(GLF_LOCK, &gl->gl_flags);
1603         run_queue(gl, 1);
1604         spin_unlock(&gl->gl_spin);
1605 }
1606
1607 static const char *state2str(unsigned state)
1608 {
1609         switch(state) {
1610         case LM_ST_UNLOCKED:
1611                 return "UN";
1612         case LM_ST_SHARED:
1613                 return "SH";
1614         case LM_ST_DEFERRED:
1615                 return "DF";
1616         case LM_ST_EXCLUSIVE:
1617                 return "EX";
1618         }
1619         return "??";
1620 }
1621
1622 static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1623 {
1624         char *p = buf;
1625         if (flags & LM_FLAG_TRY)
1626                 *p++ = 't';
1627         if (flags & LM_FLAG_TRY_1CB)
1628                 *p++ = 'T';
1629         if (flags & LM_FLAG_NOEXP)
1630                 *p++ = 'e';
1631         if (flags & LM_FLAG_ANY)
1632                 *p++ = 'A';
1633         if (flags & LM_FLAG_PRIORITY)
1634                 *p++ = 'p';
1635         if (flags & GL_ASYNC)
1636                 *p++ = 'a';
1637         if (flags & GL_EXACT)
1638                 *p++ = 'E';
1639         if (flags & GL_NOCACHE)
1640                 *p++ = 'c';
1641         if (test_bit(HIF_HOLDER, &iflags))
1642                 *p++ = 'H';
1643         if (test_bit(HIF_WAIT, &iflags))
1644                 *p++ = 'W';
1645         if (test_bit(HIF_FIRST, &iflags))
1646                 *p++ = 'F';
1647         *p = 0;
1648         return buf;
1649 }
1650
1651 /**
1652  * dump_holder - print information about a glock holder
1653  * @seq: the seq_file struct
1654  * @gh: the glock holder
1655  *
1656  */
1657
1658 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1659 {
1660         struct task_struct *gh_owner = NULL;
1661         char flags_buf[32];
1662
1663         rcu_read_lock();
1664         if (gh->gh_owner_pid)
1665                 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1666         gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1667                        state2str(gh->gh_state),
1668                        hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1669                        gh->gh_error,
1670                        gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1671                        gh_owner ? gh_owner->comm : "(ended)",
1672                        (void *)gh->gh_ip);
1673         rcu_read_unlock();
1674 }
1675
1676 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1677 {
1678         const unsigned long *gflags = &gl->gl_flags;
1679         char *p = buf;
1680
1681         if (test_bit(GLF_LOCK, gflags))
1682                 *p++ = 'l';
1683         if (test_bit(GLF_DEMOTE, gflags))
1684                 *p++ = 'D';
1685         if (test_bit(GLF_PENDING_DEMOTE, gflags))
1686                 *p++ = 'd';
1687         if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1688                 *p++ = 'p';
1689         if (test_bit(GLF_DIRTY, gflags))
1690                 *p++ = 'y';
1691         if (test_bit(GLF_LFLUSH, gflags))
1692                 *p++ = 'f';
1693         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1694                 *p++ = 'i';
1695         if (test_bit(GLF_REPLY_PENDING, gflags))
1696                 *p++ = 'r';
1697         if (test_bit(GLF_INITIAL, gflags))
1698                 *p++ = 'I';
1699         if (test_bit(GLF_FROZEN, gflags))
1700                 *p++ = 'F';
1701         if (test_bit(GLF_QUEUED, gflags))
1702                 *p++ = 'q';
1703         if (test_bit(GLF_LRU, gflags))
1704                 *p++ = 'L';
1705         if (gl->gl_object)
1706                 *p++ = 'o';
1707         if (test_bit(GLF_BLOCKING, gflags))
1708                 *p++ = 'b';
1709         *p = 0;
1710         return buf;
1711 }
1712
1713 /**
1714  * gfs2_dump_glock - print information about a glock
1715  * @seq: The seq_file struct
1716  * @gl: the glock
1717  *
1718  * The file format is as follows:
1719  * One line per object, capital letters are used to indicate objects
1720  * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1721  * other objects are indented by a single space and follow the glock to
1722  * which they are related. Fields are indicated by lower case letters
1723  * followed by a colon and the field value, except for strings which are in
1724  * [] so that its possible to see if they are composed of spaces for
1725  * example. The field's are n = number (id of the object), f = flags,
1726  * t = type, s = state, r = refcount, e = error, p = pid.
1727  *
1728  */
1729
1730 void gfs2_dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1731 {
1732         const struct gfs2_glock_operations *glops = gl->gl_ops;
1733         unsigned long long dtime;
1734         const struct gfs2_holder *gh;
1735         char gflags_buf[32];
1736
1737         dtime = jiffies - gl->gl_demote_time;
1738         dtime *= 1000000/HZ; /* demote time in uSec */
1739         if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1740                 dtime = 0;
1741         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",
1742                   state2str(gl->gl_state),
1743                   gl->gl_name.ln_type,
1744                   (unsigned long long)gl->gl_name.ln_number,
1745                   gflags2str(gflags_buf, gl),
1746                   state2str(gl->gl_target),
1747                   state2str(gl->gl_demote_state), dtime,
1748                   atomic_read(&gl->gl_ail_count),
1749                   atomic_read(&gl->gl_revokes),
1750                   (int)gl->gl_lockref.count, gl->gl_hold_time);
1751
1752         list_for_each_entry(gh, &gl->gl_holders, gh_list)
1753                 dump_holder(seq, gh);
1754
1755         if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1756                 glops->go_dump(seq, gl);
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 (!glock_workqueue)
1842                 return -ENOMEM;
1843         gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1844                                                 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1845                                                 0);
1846         if (!gfs2_delete_workqueue) {
1847                 destroy_workqueue(glock_workqueue);
1848                 return -ENOMEM;
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 ||
1903                  __lockref_is_dead(&gi->gl->gl_lockref));
1904
1905         return 0;
1906 }
1907
1908 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1909 {
1910         struct gfs2_glock_iter *gi = seq->private;
1911         loff_t n = *pos;
1912
1913         if (gi->last_pos <= *pos)
1914                 n = gi->nhash + (*pos - gi->last_pos);
1915         else
1916                 gi->hash = 0;
1917
1918         gi->nhash = 0;
1919         rcu_read_lock();
1920
1921         do {
1922                 if (gfs2_glock_iter_next(gi))
1923                         return NULL;
1924         } while (n--);
1925
1926         gi->last_pos = *pos;
1927         return gi->gl;
1928 }
1929
1930 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1931                                  loff_t *pos)
1932 {
1933         struct gfs2_glock_iter *gi = seq->private;
1934
1935         (*pos)++;
1936         gi->last_pos = *pos;
1937         if (gfs2_glock_iter_next(gi))
1938                 return NULL;
1939
1940         return gi->gl;
1941 }
1942
1943 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1944 {
1945         struct gfs2_glock_iter *gi = seq->private;
1946
1947         if (gi->gl)
1948                 rcu_read_unlock();
1949         gi->gl = NULL;
1950 }
1951
1952 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1953 {
1954         dump_glock(seq, iter_ptr);
1955         return 0;
1956 }
1957
1958 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
1959 {
1960         struct gfs2_glock_iter *gi = seq->private;
1961
1962         gi->hash = *pos;
1963         if (*pos >= GFS2_NR_SBSTATS)
1964                 return NULL;
1965         preempt_disable();
1966         return SEQ_START_TOKEN;
1967 }
1968
1969 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
1970                                    loff_t *pos)
1971 {
1972         struct gfs2_glock_iter *gi = seq->private;
1973         (*pos)++;
1974         gi->hash++;
1975         if (gi->hash >= GFS2_NR_SBSTATS) {
1976                 preempt_enable();
1977                 return NULL;
1978         }
1979         return SEQ_START_TOKEN;
1980 }
1981
1982 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
1983 {
1984         preempt_enable();
1985 }
1986
1987 static const struct seq_operations gfs2_glock_seq_ops = {
1988         .start = gfs2_glock_seq_start,
1989         .next  = gfs2_glock_seq_next,
1990         .stop  = gfs2_glock_seq_stop,
1991         .show  = gfs2_glock_seq_show,
1992 };
1993
1994 static const struct seq_operations gfs2_glstats_seq_ops = {
1995         .start = gfs2_glock_seq_start,
1996         .next  = gfs2_glock_seq_next,
1997         .stop  = gfs2_glock_seq_stop,
1998         .show  = gfs2_glstats_seq_show,
1999 };
2000
2001 static const struct seq_operations gfs2_sbstats_seq_ops = {
2002         .start = gfs2_sbstats_seq_start,
2003         .next  = gfs2_sbstats_seq_next,
2004         .stop  = gfs2_sbstats_seq_stop,
2005         .show  = gfs2_sbstats_seq_show,
2006 };
2007
2008 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2009
2010 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2011 {
2012         int ret = seq_open_private(file, &gfs2_glock_seq_ops,
2013                                    sizeof(struct gfs2_glock_iter));
2014         if (ret == 0) {
2015                 struct seq_file *seq = file->private_data;
2016                 struct gfs2_glock_iter *gi = seq->private;
2017                 gi->sdp = inode->i_private;
2018                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2019                 if (seq->buf)
2020                         seq->size = GFS2_SEQ_GOODSIZE;
2021         }
2022         return ret;
2023 }
2024
2025 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2026 {
2027         int ret = seq_open_private(file, &gfs2_glstats_seq_ops,
2028                                    sizeof(struct gfs2_glock_iter));
2029         if (ret == 0) {
2030                 struct seq_file *seq = file->private_data;
2031                 struct gfs2_glock_iter *gi = seq->private;
2032                 gi->sdp = inode->i_private;
2033                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2034                 if (seq->buf)
2035                         seq->size = GFS2_SEQ_GOODSIZE;
2036         }
2037         return ret;
2038 }
2039
2040 static int gfs2_sbstats_open(struct inode *inode, struct file *file)
2041 {
2042         int ret = seq_open_private(file, &gfs2_sbstats_seq_ops,
2043                                    sizeof(struct gfs2_glock_iter));
2044         if (ret == 0) {
2045                 struct seq_file *seq = file->private_data;
2046                 struct gfs2_glock_iter *gi = seq->private;
2047                 gi->sdp = inode->i_private;
2048         }
2049         return ret;
2050 }
2051
2052 static const struct file_operations gfs2_glocks_fops = {
2053         .owner   = THIS_MODULE,
2054         .open    = gfs2_glocks_open,
2055         .read    = seq_read,
2056         .llseek  = seq_lseek,
2057         .release = seq_release_private,
2058 };
2059
2060 static const struct file_operations gfs2_glstats_fops = {
2061         .owner   = THIS_MODULE,
2062         .open    = gfs2_glstats_open,
2063         .read    = seq_read,
2064         .llseek  = seq_lseek,
2065         .release = seq_release_private,
2066 };
2067
2068 static const struct file_operations gfs2_sbstats_fops = {
2069         .owner   = THIS_MODULE,
2070         .open    = gfs2_sbstats_open,
2071         .read    = seq_read,
2072         .llseek  = seq_lseek,
2073         .release = seq_release_private,
2074 };
2075
2076 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2077 {
2078         sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2079         if (!sdp->debugfs_dir)
2080                 return -ENOMEM;
2081         sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
2082                                                          S_IFREG | S_IRUGO,
2083                                                          sdp->debugfs_dir, sdp,
2084                                                          &gfs2_glocks_fops);
2085         if (!sdp->debugfs_dentry_glocks)
2086                 goto fail;
2087
2088         sdp->debugfs_dentry_glstats = debugfs_create_file("glstats",
2089                                                         S_IFREG | S_IRUGO,
2090                                                         sdp->debugfs_dir, sdp,
2091                                                         &gfs2_glstats_fops);
2092         if (!sdp->debugfs_dentry_glstats)
2093                 goto fail;
2094
2095         sdp->debugfs_dentry_sbstats = debugfs_create_file("sbstats",
2096                                                         S_IFREG | S_IRUGO,
2097                                                         sdp->debugfs_dir, sdp,
2098                                                         &gfs2_sbstats_fops);
2099         if (!sdp->debugfs_dentry_sbstats)
2100                 goto fail;
2101
2102         return 0;
2103 fail:
2104         gfs2_delete_debugfs_file(sdp);
2105         return -ENOMEM;
2106 }
2107
2108 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2109 {
2110         if (sdp->debugfs_dir) {
2111                 if (sdp->debugfs_dentry_glocks) {
2112                         debugfs_remove(sdp->debugfs_dentry_glocks);
2113                         sdp->debugfs_dentry_glocks = NULL;
2114                 }
2115                 if (sdp->debugfs_dentry_glstats) {
2116                         debugfs_remove(sdp->debugfs_dentry_glstats);
2117                         sdp->debugfs_dentry_glstats = NULL;
2118                 }
2119                 if (sdp->debugfs_dentry_sbstats) {
2120                         debugfs_remove(sdp->debugfs_dentry_sbstats);
2121                         sdp->debugfs_dentry_sbstats = NULL;
2122                 }
2123                 debugfs_remove(sdp->debugfs_dir);
2124                 sdp->debugfs_dir = NULL;
2125         }
2126 }
2127
2128 int gfs2_register_debugfs(void)
2129 {
2130         gfs2_root = debugfs_create_dir("gfs2", NULL);
2131         return gfs2_root ? 0 : -ENOMEM;
2132 }
2133
2134 void gfs2_unregister_debugfs(void)
2135 {
2136         debugfs_remove(gfs2_root);
2137         gfs2_root = NULL;
2138 }