fs/gfs2/glock.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
   4  *
   5  * This copyrighted material is made available to anyone wishing to use,
   6  * modify, copy, or redistribute it subject to the terms and conditions
   7  * of the GNU General Public License version 2.
   8  */
   9
  10 #include <linux/sched.h>
  11 #include <linux/slab.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/buffer_head.h>
  14 #include <linux/delay.h>
  15 #include <linux/sort.h>
  16 #include <linux/jhash.h>
  17 #include <linux/kallsyms.h>
  18 #include <linux/gfs2_ondisk.h>
  19 #include <linux/list.h>
  20 #include <linux/wait.h>
  21 #include <linux/module.h>
  22 #include <asm/uaccess.h>
  23 #include <linux/seq_file.h>
  24 #include <linux/debugfs.h>
  25 #include <linux/kthread.h>
  26 #include <linux/freezer.h>
  27 #include <linux/workqueue.h>
  28 #include <linux/jiffies.h>
  29 #include <linux/rcupdate.h>
  30 #include <linux/rculist_bl.h>
  31 #include <linux/bit_spinlock.h>
  32 #include <linux/percpu.h>
  33
  34 #include "gfs2.h"
  35 #include "incore.h"
  36 #include "glock.h"
  37 #include "glops.h"
  38 #include "inode.h"
  39 #include "lops.h"
  40 #include "meta_io.h"
  41 #include "quota.h"
  42 #include "super.h"
  43 #include "util.h"
  44 #include "bmap.h"
  45 #define CREATE_TRACE_POINTS
  46 #include "trace_gfs2.h"
  47
  48 struct gfs2_glock_iter {
  49         int hash;                       /* hash bucket index           */
  50         unsigned nhash;                 /* Index within current bucket */
  51         struct gfs2_sbd *sdp;           /* incore superblock           */
  52         struct gfs2_glock *gl;          /* current glock struct        */
  53         loff_t last_pos;                /* last position               */
  54 };
  55
  56 typedef void (*glock_examiner) (struct gfs2_glock * gl);
  57
  58 static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl);
  59 #define GLOCK_BUG_ON(gl,x) do { if (unlikely(x)) { __dump_glock(NULL, gl); BUG(); } } while(0)
  60 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
  61
  62 static struct dentry *gfs2_root;
  63 static struct workqueue_struct *glock_workqueue;
  64 struct workqueue_struct *gfs2_delete_workqueue;
  65 static LIST_HEAD(lru_list);
  66 static atomic_t lru_count = ATOMIC_INIT(0);
  67 static DEFINE_SPINLOCK(lru_lock);
  68
  69 #define GFS2_GL_HASH_SHIFT      15
  70 #define GFS2_GL_HASH_SIZE       (1 << GFS2_GL_HASH_SHIFT)
  71 #define GFS2_GL_HASH_MASK       (GFS2_GL_HASH_SIZE - 1)
  72
  73 static struct hlist_bl_head gl_hash_table[GFS2_GL_HASH_SIZE];
  74 static struct dentry *gfs2_root;
  75
  76 /**
  77  * gl_hash() - Turn glock number into hash bucket number
  78  * @lock: The glock number
  79  *
  80  * Returns: The number of the corresponding hash bucket
  81  */
  82
  83 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
  84                             const struct lm_lockname *name)
  85 {
  86         unsigned int h;
  87
  88         h = jhash(&name->ln_number, sizeof(u64), 0);
  89         h = jhash(&name->ln_type, sizeof(unsigned int), h);
  90         h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
  91         h &= GFS2_GL_HASH_MASK;
  92
  93         return h;
  94 }
  95
  96 static inline void spin_lock_bucket(unsigned int hash)
  97 {
  98         hlist_bl_lock(&gl_hash_table[hash]);
  99 }
 100
 101 static inline void spin_unlock_bucket(unsigned int hash)
 102 {
 103         hlist_bl_unlock(&gl_hash_table[hash]);
 104 }
 105
 106 static void gfs2_glock_dealloc(struct rcu_head *rcu)
 107 {
 108         struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
 109
 110         if (gl->gl_ops->go_flags & GLOF_ASPACE)
 111                 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
 112         else
 113                 kmem_cache_free(gfs2_glock_cachep, gl);
 114 }
 115
 116 void gfs2_glock_free(struct gfs2_glock *gl)
 117 {
 118         struct gfs2_sbd *sdp = gl->gl_sbd;
 119
 120         call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
 121         if (atomic_dec_and_test(&sdp->sd_glock_disposal))
 122                 wake_up(&sdp->sd_glock_wait);
 123 }
 124
 125 /**
 126  * gfs2_glock_hold() - increment reference count on glock
 127  * @gl: The glock to hold
 128  *
 129  */
 130
 131 void gfs2_glock_hold(struct gfs2_glock *gl)
 132 {
 133         GLOCK_BUG_ON(gl, atomic_read(&gl->gl_ref) == 0);
 134         atomic_inc(&gl->gl_ref);
 135 }
 136
 137 /**
 138  * demote_ok - Check to see if it's ok to unlock a glock
 139  * @gl: the glock
 140  *
 141  * Returns: 1 if it's ok
 142  */
 143
 144 static int demote_ok(const struct gfs2_glock *gl)
 145 {
 146         const struct gfs2_glock_operations *glops = gl->gl_ops;
 147
 148         if (gl->gl_state == LM_ST_UNLOCKED)
 149                 return 0;
 150         if (!list_empty(&gl->gl_holders))
 151                 return 0;
 152         if (glops->go_demote_ok)
 153                 return glops->go_demote_ok(gl);
 154         return 1;
 155 }
 156
 157
 158 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
 159 {
 160         spin_lock(&lru_lock);
 161
 162         if (!list_empty(&gl->gl_lru))
 163                 list_del_init(&gl->gl_lru);
 164         else
 165                 atomic_inc(&lru_count);
 166
 167         list_add_tail(&gl->gl_lru, &lru_list);
 168         set_bit(GLF_LRU, &gl->gl_flags);
 169         spin_unlock(&lru_lock);
 170 }
 171
 172 static void __gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
 173 {
 174         if (!list_empty(&gl->gl_lru)) {
 175                 list_del_init(&gl->gl_lru);
 176                 atomic_dec(&lru_count);
 177                 clear_bit(GLF_LRU, &gl->gl_flags);
 178         }
 179 }
 180
 181 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
 182 {
 183         spin_lock(&lru_lock);
 184         __gfs2_glock_remove_from_lru(gl);
 185         spin_unlock(&lru_lock);
 186 }
 187
 188 /**
 189  * gfs2_glock_put_nolock() - Decrement reference count on glock
 190  * @gl: The glock to put
 191  *
 192  * This function should only be used if the caller has its own reference
 193  * to the glock, in addition to the one it is dropping.
 194  */
 195
 196 void gfs2_glock_put_nolock(struct gfs2_glock *gl)
 197 {
 198         if (atomic_dec_and_test(&gl->gl_ref))
 199                 GLOCK_BUG_ON(gl, 1);
 200 }
 201
 202 /**
 203  * gfs2_glock_put() - Decrement reference count on glock
 204  * @gl: The glock to put
 205  *
 206  */
 207
 208 void gfs2_glock_put(struct gfs2_glock *gl)
 209 {
 210         struct gfs2_sbd *sdp = gl->gl_sbd;
 211         struct address_space *mapping = gfs2_glock2aspace(gl);
 212
 213         if (atomic_dec_and_lock(&gl->gl_ref, &lru_lock)) {
 214                 __gfs2_glock_remove_from_lru(gl);
 215                 spin_unlock(&lru_lock);
 216                 spin_lock_bucket(gl->gl_hash);
 217                 hlist_bl_del_rcu(&gl->gl_list);
 218                 spin_unlock_bucket(gl->gl_hash);
 219                 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
 220                 GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
 221                 trace_gfs2_glock_put(gl);
 222                 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
 223         }
 224 }
 225
 226 /**
 227  * search_bucket() - Find struct gfs2_glock by lock number
 228  * @bucket: the bucket to search
 229  * @name: The lock name
 230  *
 231  * Returns: NULL, or the struct gfs2_glock with the requested number
 232  */
 233
 234 static struct gfs2_glock *search_bucket(unsigned int hash,
 235                                         const struct gfs2_sbd *sdp,
 236                                         const struct lm_lockname *name)
 237 {
 238         struct gfs2_glock *gl;
 239         struct hlist_bl_node *h;
 240
 241         hlist_bl_for_each_entry_rcu(gl, h, &gl_hash_table[hash], gl_list) {
 242                 if (!lm_name_equal(&gl->gl_name, name))
 243                         continue;
 244                 if (gl->gl_sbd != sdp)
 245                         continue;
 246                 if (atomic_inc_not_zero(&gl->gl_ref))
 247                         return gl;
 248         }
 249
 250         return NULL;
 251 }
 252
 253 /**
 254  * may_grant - check if its ok to grant a new lock
 255  * @gl: The glock
 256  * @gh: The lock request which we wish to grant
 257  *
 258  * Returns: true if its ok to grant the lock
 259  */
 260
 261 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
 262 {
 263         const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
 264         if ((gh->gh_state == LM_ST_EXCLUSIVE ||
 265              gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
 266                 return 0;
 267         if (gl->gl_state == gh->gh_state)
 268                 return 1;
 269         if (gh->gh_flags & GL_EXACT)
 270                 return 0;
 271         if (gl->gl_state == LM_ST_EXCLUSIVE) {
 272                 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
 273                         return 1;
 274                 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
 275                         return 1;
 276         }
 277         if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
 278                 return 1;
 279         return 0;
 280 }
 281
 282 static void gfs2_holder_wake(struct gfs2_holder *gh)
 283 {
 284         clear_bit(HIF_WAIT, &gh->gh_iflags);
 285         smp_mb__after_clear_bit();
 286         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
 287 }
 288
 289 /**
 290  * do_error - Something unexpected has happened during a lock request
 291  *
 292  */
 293
 294 static inline void do_error(struct gfs2_glock *gl, const int ret)
 295 {
 296         struct gfs2_holder *gh, *tmp;
 297
 298         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
 299                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 300                         continue;
 301                 if (ret & LM_OUT_ERROR)
 302                         gh->gh_error = -EIO;
 303                 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
 304                         gh->gh_error = GLR_TRYFAILED;
 305                 else
 306                         continue;
 307                 list_del_init(&gh->gh_list);
 308                 trace_gfs2_glock_queue(gh, 0);
 309                 gfs2_holder_wake(gh);
 310         }
 311 }
 312
 313 /**
 314  * do_promote - promote as many requests as possible on the current queue
 315  * @gl: The glock
 316  *
 317  * Returns: 1 if there is a blocked holder at the head of the list, or 2
 318  *          if a type specific operation is underway.
 319  */
 320
 321 static int do_promote(struct gfs2_glock *gl)
 322 __releases(&gl->gl_spin)
 323 __acquires(&gl->gl_spin)
 324 {
 325         const struct gfs2_glock_operations *glops = gl->gl_ops;
 326         struct gfs2_holder *gh, *tmp;
 327         int ret;
 328
 329 restart:
 330         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
 331                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 332                         continue;
 333                 if (may_grant(gl, gh)) {
 334                         if (gh->gh_list.prev == &gl->gl_holders &&
 335                             glops->go_lock) {
 336                                 spin_unlock(&gl->gl_spin);
 337                                 /* FIXME: eliminate this eventually */
 338                                 ret = glops->go_lock(gh);
 339                                 spin_lock(&gl->gl_spin);
 340                                 if (ret) {
 341                                         if (ret == 1)
 342                                                 return 2;
 343                                         gh->gh_error = ret;
 344                                         list_del_init(&gh->gh_list);
 345                                         trace_gfs2_glock_queue(gh, 0);
 346                                         gfs2_holder_wake(gh);
 347                                         goto restart;
 348                                 }
 349                                 set_bit(HIF_HOLDER, &gh->gh_iflags);
 350                                 trace_gfs2_promote(gh, 1);
 351                                 gfs2_holder_wake(gh);
 352                                 goto restart;
 353                         }
 354                         set_bit(HIF_HOLDER, &gh->gh_iflags);
 355                         trace_gfs2_promote(gh, 0);
 356                         gfs2_holder_wake(gh);
 357                         continue;
 358                 }
 359                 if (gh->gh_list.prev == &gl->gl_holders)
 360                         return 1;
 361                 do_error(gl, 0);
 362                 break;
 363         }
 364         return 0;
 365 }
 366
 367 /**
 368  * find_first_waiter - find the first gh that's waiting for the glock
 369  * @gl: the glock
 370  */
 371
 372 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
 373 {
 374         struct gfs2_holder *gh;
 375
 376         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
 377                 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
 378                         return gh;
 379         }
 380         return NULL;
 381 }
 382
 383 /**
 384  * state_change - record that the glock is now in a different state
 385  * @gl: the glock
 386  * @new_state the new state
 387  *
 388  */
 389
 390 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
 391 {
 392         int held1, held2;
 393
 394         held1 = (gl->gl_state != LM_ST_UNLOCKED);
 395         held2 = (new_state != LM_ST_UNLOCKED);
 396
 397         if (held1 != held2) {
 398                 if (held2)
 399                         gfs2_glock_hold(gl);
 400                 else
 401                         gfs2_glock_put_nolock(gl);
 402         }
 403         if (held1 && held2 && list_empty(&gl->gl_holders))
 404                 clear_bit(GLF_QUEUED, &gl->gl_flags);
 405
 406         if (new_state != gl->gl_target)
 407                 /* shorten our minimum hold time */
 408                 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
 409                                        GL_GLOCK_MIN_HOLD);
 410         gl->gl_state = new_state;
 411         gl->gl_tchange = jiffies;
 412 }
 413
 414 static void gfs2_demote_wake(struct gfs2_glock *gl)
 415 {
 416         gl->gl_demote_state = LM_ST_EXCLUSIVE;
 417         clear_bit(GLF_DEMOTE, &gl->gl_flags);
 418         smp_mb__after_clear_bit();
 419         wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
 420 }
 421
 422 /**
 423  * finish_xmote - The DLM has replied to one of our lock requests
 424  * @gl: The glock
 425  * @ret: The status from the DLM
 426  *
 427  */
 428
 429 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
 430 {
 431         const struct gfs2_glock_operations *glops = gl->gl_ops;
 432         struct gfs2_holder *gh;
 433         unsigned state = ret & LM_OUT_ST_MASK;
 434         int rv;
 435
 436         spin_lock(&gl->gl_spin);
 437         trace_gfs2_glock_state_change(gl, state);
 438         state_change(gl, state);
 439         gh = find_first_waiter(gl);
 440
 441         /* Demote to UN request arrived during demote to SH or DF */
 442         if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
 443             state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
 444                 gl->gl_target = LM_ST_UNLOCKED;
 445
 446         /* Check for state != intended state */
 447         if (unlikely(state != gl->gl_target)) {
 448                 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
 449                         /* move to back of queue and try next entry */
 450                         if (ret & LM_OUT_CANCELED) {
 451                                 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
 452                                         list_move_tail(&gh->gh_list, &gl->gl_holders);
 453                                 gh = find_first_waiter(gl);
 454                                 gl->gl_target = gh->gh_state;
 455                                 goto retry;
 456                         }
 457                         /* Some error or failed "try lock" - report it */
 458                         if ((ret & LM_OUT_ERROR) ||
 459                             (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
 460                                 gl->gl_target = gl->gl_state;
 461                                 do_error(gl, ret);
 462                                 goto out;
 463                         }
 464                 }
 465                 switch(state) {
 466                 /* Unlocked due to conversion deadlock, try again */
 467                 case LM_ST_UNLOCKED:
 468 retry:
 469                         do_xmote(gl, gh, gl->gl_target);
 470                         break;
 471                 /* Conversion fails, unlock and try again */
 472                 case LM_ST_SHARED:
 473                 case LM_ST_DEFERRED:
 474                         do_xmote(gl, gh, LM_ST_UNLOCKED);
 475                         break;
 476                 default: /* Everything else */
 477                         printk(KERN_ERR "GFS2: wanted %u got %u\n", gl->gl_target, state);
 478                         GLOCK_BUG_ON(gl, 1);
 479                 }
 480                 spin_unlock(&gl->gl_spin);
 481                 return;
 482         }
 483
 484         /* Fast path - we got what we asked for */
 485         if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
 486                 gfs2_demote_wake(gl);
 487         if (state != LM_ST_UNLOCKED) {
 488                 if (glops->go_xmote_bh) {
 489                         spin_unlock(&gl->gl_spin);
 490                         rv = glops->go_xmote_bh(gl, gh);
 491                         spin_lock(&gl->gl_spin);
 492                         if (rv) {
 493                                 do_error(gl, rv);
 494                                 goto out;
 495                         }
 496                 }
 497                 rv = do_promote(gl);
 498                 if (rv == 2)
 499                         goto out_locked;
 500         }
 501 out:
 502         clear_bit(GLF_LOCK, &gl->gl_flags);
 503 out_locked:
 504         spin_unlock(&gl->gl_spin);
 505 }
 506
 507 /**
 508  * do_xmote - Calls the DLM to change the state of a lock
 509  * @gl: The lock state
 510  * @gh: The holder (only for promotes)
 511  * @target: The target lock state
 512  *
 513  */
 514
 515 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
 516 __releases(&gl->gl_spin)
 517 __acquires(&gl->gl_spin)
 518 {
 519         const struct gfs2_glock_operations *glops = gl->gl_ops;
 520         struct gfs2_sbd *sdp = gl->gl_sbd;
 521         unsigned int lck_flags = gh ? gh->gh_flags : 0;
 522         int ret;
 523
 524         lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
 525                       LM_FLAG_PRIORITY);
 526         GLOCK_BUG_ON(gl, gl->gl_state == target);
 527         GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
 528         if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
 529             glops->go_inval) {
 530                 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 531                 do_error(gl, 0); /* Fail queued try locks */
 532         }
 533         gl->gl_req = target;
 534         set_bit(GLF_BLOCKING, &gl->gl_flags);
 535         if ((gl->gl_req == LM_ST_UNLOCKED) ||
 536             (gl->gl_state == LM_ST_EXCLUSIVE) ||
 537             (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
 538                 clear_bit(GLF_BLOCKING, &gl->gl_flags);
 539         spin_unlock(&gl->gl_spin);
 540         if (glops->go_xmote_th)
 541                 glops->go_xmote_th(gl);
 542         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 543                 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
 544         clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 545
 546         gfs2_glock_hold(gl);
 547         if (sdp->sd_lockstruct.ls_ops->lm_lock) {
 548                 /* lock_dlm */
 549                 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
 550                 GLOCK_BUG_ON(gl, ret);
 551         } else { /* lock_nolock */
 552                 finish_xmote(gl, target);
 553                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 554                         gfs2_glock_put(gl);
 555         }
 556
 557         spin_lock(&gl->gl_spin);
 558 }
 559
 560 /**
 561  * find_first_holder - find the first "holder" gh
 562  * @gl: the glock
 563  */
 564
 565 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
 566 {
 567         struct gfs2_holder *gh;
 568
 569         if (!list_empty(&gl->gl_holders)) {
 570                 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
 571                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 572                         return gh;
 573         }
 574         return NULL;
 575 }
 576
 577 /**
 578  * run_queue - do all outstanding tasks related to a glock
 579  * @gl: The glock in question
 580  * @nonblock: True if we must not block in run_queue
 581  *
 582  */
 583
 584 static void run_queue(struct gfs2_glock *gl, const int nonblock)
 585 __releases(&gl->gl_spin)
 586 __acquires(&gl->gl_spin)
 587 {
 588         struct gfs2_holder *gh = NULL;
 589         int ret;
 590
 591         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
 592                 return;
 593
 594         GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
 595
 596         if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
 597             gl->gl_demote_state != gl->gl_state) {
 598                 if (find_first_holder(gl))
 599                         goto out_unlock;
 600                 if (nonblock)
 601                         goto out_sched;
 602                 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
 603                 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
 604                 gl->gl_target = gl->gl_demote_state;
 605         } else {
 606                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
 607                         gfs2_demote_wake(gl);
 608                 ret = do_promote(gl);
 609                 if (ret == 0)
 610                         goto out_unlock;
 611                 if (ret == 2)
 612                         goto out;
 613                 gh = find_first_waiter(gl);
 614                 gl->gl_target = gh->gh_state;
 615                 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
 616                         do_error(gl, 0); /* Fail queued try locks */
 617         }
 618         do_xmote(gl, gh, gl->gl_target);
 619 out:
 620         return;
 621
 622 out_sched:
 623         clear_bit(GLF_LOCK, &gl->gl_flags);
 624         smp_mb__after_clear_bit();
 625         gfs2_glock_hold(gl);
 626         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 627                 gfs2_glock_put_nolock(gl);
 628         return;
 629
 630 out_unlock:
 631         clear_bit(GLF_LOCK, &gl->gl_flags);
 632         smp_mb__after_clear_bit();
 633         return;
 634 }
 635
 636 static void delete_work_func(struct work_struct *work)
 637 {
 638         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
 639         struct gfs2_sbd *sdp = gl->gl_sbd;
 640         struct gfs2_inode *ip;
 641         struct inode *inode;
 642         u64 no_addr = gl->gl_name.ln_number;
 643
 644         ip = gl->gl_object;
 645         /* Note: Unsafe to dereference ip as we don't hold right refs/locks */
 646
 647         if (ip)
 648                 inode = gfs2_ilookup(sdp->sd_vfs, no_addr, 1);
 649         else
 650                 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
 651         if (inode && !IS_ERR(inode)) {
 652                 d_prune_aliases(inode);
 653                 iput(inode);
 654         }
 655         gfs2_glock_put(gl);
 656 }
 657
 658 static void glock_work_func(struct work_struct *work)
 659 {
 660         unsigned long delay = 0;
 661         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
 662         int drop_ref = 0;
 663
 664         if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
 665                 finish_xmote(gl, gl->gl_reply);
 666                 drop_ref = 1;
 667         }
 668         spin_lock(&gl->gl_spin);
 669         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
 670             gl->gl_state != LM_ST_UNLOCKED &&
 671             gl->gl_demote_state != LM_ST_EXCLUSIVE) {
 672                 unsigned long holdtime, now = jiffies;
 673
 674                 holdtime = gl->gl_tchange + gl->gl_hold_time;
 675                 if (time_before(now, holdtime))
 676                         delay = holdtime - now;
 677
 678                 if (!delay) {
 679                         clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
 680                         set_bit(GLF_DEMOTE, &gl->gl_flags);
 681                 }
 682         }
 683         run_queue(gl, 0);
 684         spin_unlock(&gl->gl_spin);
 685         if (!delay)
 686                 gfs2_glock_put(gl);
 687         else {
 688                 if (gl->gl_name.ln_type != LM_TYPE_INODE)
 689                         delay = 0;
 690                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
 691                         gfs2_glock_put(gl);
 692         }
 693         if (drop_ref)
 694                 gfs2_glock_put(gl);
 695 }
 696
 697 /**
 698  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
 699  * @sdp: The GFS2 superblock
 700  * @number: the lock number
 701  * @glops: The glock_operations to use
 702  * @create: If 0, don't create the glock if it doesn't exist
 703  * @glp: the glock is returned here
 704  *
 705  * This does not lock a glock, just finds/creates structures for one.
 706  *
 707  * Returns: errno
 708  */
 709
 710 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
 711                    const struct gfs2_glock_operations *glops, int create,
 712                    struct gfs2_glock **glp)
 713 {
 714         struct super_block *s = sdp->sd_vfs;
 715         struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
 716         struct gfs2_glock *gl, *tmp;
 717         unsigned int hash = gl_hash(sdp, &name);
 718         struct address_space *mapping;
 719         struct kmem_cache *cachep;
 720
 721         rcu_read_lock();
 722         gl = search_bucket(hash, sdp, &name);
 723         rcu_read_unlock();
 724
 725         *glp = gl;
 726         if (gl)
 727                 return 0;
 728         if (!create)
 729                 return -ENOENT;
 730
 731         if (glops->go_flags & GLOF_ASPACE)
 732                 cachep = gfs2_glock_aspace_cachep;
 733         else
 734                 cachep = gfs2_glock_cachep;
 735         gl = kmem_cache_alloc(cachep, GFP_KERNEL);
 736         if (!gl)
 737                 return -ENOMEM;
 738
 739         atomic_inc(&sdp->sd_glock_disposal);
 740         gl->gl_sbd = sdp;
 741         gl->gl_flags = 0;
 742         gl->gl_name = name;
 743         atomic_set(&gl->gl_ref, 1);
 744         gl->gl_state = LM_ST_UNLOCKED;
 745         gl->gl_target = LM_ST_UNLOCKED;
 746         gl->gl_demote_state = LM_ST_EXCLUSIVE;
 747         gl->gl_hash = hash;
 748         gl->gl_ops = glops;
 749         gl->gl_dstamp = ktime_set(0, 0);
 750         preempt_disable();
 751         /* We use the global stats to estimate the initial per-glock stats */
 752         gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
 753         preempt_enable();
 754         gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
 755         gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
 756         memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
 757         memset(gl->gl_lvb, 0, 32 * sizeof(char));
 758         gl->gl_lksb.sb_lvbptr = gl->gl_lvb;
 759         gl->gl_tchange = jiffies;
 760         gl->gl_object = NULL;
 761         gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
 762         INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
 763         INIT_WORK(&gl->gl_delete, delete_work_func);
 764
 765         mapping = gfs2_glock2aspace(gl);
 766         if (mapping) {
 767                 mapping->a_ops = &gfs2_meta_aops;
 768                 mapping->host = s->s_bdev->bd_inode;
 769                 mapping->flags = 0;
 770                 mapping_set_gfp_mask(mapping, GFP_NOFS);
 771                 mapping->assoc_mapping = NULL;
 772                 mapping->backing_dev_info = s->s_bdi;
 773                 mapping->writeback_index = 0;
 774         }
 775
 776         spin_lock_bucket(hash);
 777         tmp = search_bucket(hash, sdp, &name);
 778         if (tmp) {
 779                 spin_unlock_bucket(hash);
 780                 kmem_cache_free(cachep, gl);
 781                 atomic_dec(&sdp->sd_glock_disposal);
 782                 gl = tmp;
 783         } else {
 784                 hlist_bl_add_head_rcu(&gl->gl_list, &gl_hash_table[hash]);
 785                 spin_unlock_bucket(hash);
 786         }
 787
 788         *glp = gl;
 789
 790         return 0;
 791 }
 792
 793 /**
 794  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
 795  * @gl: the glock
 796  * @state: the state we're requesting
 797  * @flags: the modifier flags
 798  * @gh: the holder structure
 799  *
 800  */
 801
 802 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
 803                       struct gfs2_holder *gh)
 804 {
 805         INIT_LIST_HEAD(&gh->gh_list);
 806         gh->gh_gl = gl;
 807         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 808         gh->gh_owner_pid = get_pid(task_pid(current));
 809         gh->gh_state = state;
 810         gh->gh_flags = flags;
 811         gh->gh_error = 0;
 812         gh->gh_iflags = 0;
 813         gfs2_glock_hold(gl);
 814 }
 815
 816 /**
 817  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
 818  * @state: the state we're requesting
 819  * @flags: the modifier flags
 820  * @gh: the holder structure
 821  *
 822  * Don't mess with the glock.
 823  *
 824  */
 825
 826 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
 827 {
 828         gh->gh_state = state;
 829         gh->gh_flags = flags;
 830         gh->gh_iflags = 0;
 831         gh->gh_ip = (unsigned long)__builtin_return_address(0);
 832         if (gh->gh_owner_pid)
 833                 put_pid(gh->gh_owner_pid);
 834         gh->gh_owner_pid = get_pid(task_pid(current));
 835 }
 836
 837 /**
 838  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
 839  * @gh: the holder structure
 840  *
 841  */
 842
 843 void gfs2_holder_uninit(struct gfs2_holder *gh)
 844 {
 845         put_pid(gh->gh_owner_pid);
 846         gfs2_glock_put(gh->gh_gl);
 847         gh->gh_gl = NULL;
 848         gh->gh_ip = 0;
 849 }
 850
 851 /**
 852  * gfs2_glock_holder_wait
 853  * @word: unused
 854  *
 855  * This function and gfs2_glock_demote_wait both show up in the WCHAN
 856  * field. Thus I've separated these otherwise identical functions in
 857  * order to be more informative to the user.
 858  */
 859
 860 static int gfs2_glock_holder_wait(void *word)
 861 {
 862         schedule();
 863         return 0;
 864 }
 865
 866 static int gfs2_glock_demote_wait(void *word)
 867 {
 868         schedule();
 869         return 0;
 870 }
 871
 872 /**
 873  * gfs2_glock_wait - wait on a glock acquisition
 874  * @gh: the glock holder
 875  *
 876  * Returns: 0 on success
 877  */
 878
 879 int gfs2_glock_wait(struct gfs2_holder *gh)
 880 {
 881         unsigned long time1 = jiffies;
 882
 883         might_sleep();
 884         wait_on_bit(&gh->gh_iflags, HIF_WAIT, gfs2_glock_holder_wait, TASK_UNINTERRUPTIBLE);
 885         if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
 886                 /* Lengthen the minimum hold time. */
 887                 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
 888                                               GL_GLOCK_HOLD_INCR,
 889                                               GL_GLOCK_MAX_HOLD);
 890         return gh->gh_error;
 891 }
 892
 893 /**
 894  * handle_callback - process a demote request
 895  * @gl: the glock
 896  * @state: the state the caller wants us to change to
 897  *
 898  * There are only two requests that we are going to see in actual
 899  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
 900  */
 901
 902 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
 903                             unsigned long delay)
 904 {
 905         int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
 906
 907         set_bit(bit, &gl->gl_flags);
 908         if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
 909                 gl->gl_demote_state = state;
 910                 gl->gl_demote_time = jiffies;
 911         } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
 912                         gl->gl_demote_state != state) {
 913                 gl->gl_demote_state = LM_ST_UNLOCKED;
 914         }
 915         if (gl->gl_ops->go_callback)
 916                 gl->gl_ops->go_callback(gl);
 917         trace_gfs2_demote_rq(gl);
 918 }
 919
 920 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
 921 {
 922         struct va_format vaf;
 923         va_list args;
 924
 925         va_start(args, fmt);
 926
 927         if (seq) {
 928                 seq_vprintf(seq, fmt, args);
 929         } else {
 930                 vaf.fmt = fmt;
 931                 vaf.va = &args;
 932
 933                 printk(KERN_ERR " %pV", &vaf);
 934         }
 935
 936         va_end(args);
 937 }
 938
 939 /**
 940  * add_to_queue - Add a holder to the wait queue (but look for recursion)
 941  * @gh: the holder structure to add
 942  *
 943  * Eventually we should move the recursive locking trap to a
 944  * debugging option or something like that. This is the fast
 945  * path and needs to have the minimum number of distractions.
 946  *
 947  */
 948
 949 static inline void add_to_queue(struct gfs2_holder *gh)
 950 __releases(&gl->gl_spin)
 951 __acquires(&gl->gl_spin)
 952 {
 953         struct gfs2_glock *gl = gh->gh_gl;
 954         struct gfs2_sbd *sdp = gl->gl_sbd;
 955         struct list_head *insert_pt = NULL;
 956         struct gfs2_holder *gh2;
 957         int try_futile = 0;
 958
 959         BUG_ON(gh->gh_owner_pid == NULL);
 960         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
 961                 BUG();
 962
 963         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
 964                 if (test_bit(GLF_LOCK, &gl->gl_flags))
 965                         try_futile = !may_grant(gl, gh);
 966                 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 967                         goto fail;
 968         }
 969
 970         list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
 971                 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
 972                     (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
 973                         goto trap_recursive;
 974                 if (try_futile &&
 975                     !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
 976 fail:
 977                         gh->gh_error = GLR_TRYFAILED;
 978                         gfs2_holder_wake(gh);
 979                         return;
 980                 }
 981                 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
 982                         continue;
 983                 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
 984                         insert_pt = &gh2->gh_list;
 985         }
 986         set_bit(GLF_QUEUED, &gl->gl_flags);
 987         trace_gfs2_glock_queue(gh, 1);
 988         gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
 989         gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
 990         if (likely(insert_pt == NULL)) {
 991                 list_add_tail(&gh->gh_list, &gl->gl_holders);
 992                 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
 993                         goto do_cancel;
 994                 return;
 995         }
 996         list_add_tail(&gh->gh_list, insert_pt);
 997 do_cancel:
 998         gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
 999         if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1000                 spin_unlock(&gl->gl_spin);
1001                 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1002                         sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1003                 spin_lock(&gl->gl_spin);
1004         }
1005         return;
1006
1007 trap_recursive:
1008         print_symbol(KERN_ERR "original: %s\n", gh2->gh_ip);
1009         printk(KERN_ERR "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1010         printk(KERN_ERR "lock type: %d req lock state : %d\n",
1011                gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1012         print_symbol(KERN_ERR "new: %s\n", gh->gh_ip);
1013         printk(KERN_ERR "pid: %d\n", pid_nr(gh->gh_owner_pid));
1014         printk(KERN_ERR "lock type: %d req lock state : %d\n",
1015                gh->gh_gl->gl_name.ln_type, gh->gh_state);
1016         __dump_glock(NULL, gl);
1017         BUG();
1018 }
1019
1020 /**
1021  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1022  * @gh: the holder structure
1023  *
1024  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1025  *
1026  * Returns: 0, GLR_TRYFAILED, or errno on failure
1027  */
1028
1029 int gfs2_glock_nq(struct gfs2_holder *gh)
1030 {
1031         struct gfs2_glock *gl = gh->gh_gl;
1032         struct gfs2_sbd *sdp = gl->gl_sbd;
1033         int error = 0;
1034
1035         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1036                 return -EIO;
1037
1038         if (test_bit(GLF_LRU, &gl->gl_flags))
1039                 gfs2_glock_remove_from_lru(gl);
1040
1041         spin_lock(&gl->gl_spin);
1042         add_to_queue(gh);
1043         if ((LM_FLAG_NOEXP & gh->gh_flags) &&
1044             test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1045                 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1046         run_queue(gl, 1);
1047         spin_unlock(&gl->gl_spin);
1048
1049         if (!(gh->gh_flags & GL_ASYNC))
1050                 error = gfs2_glock_wait(gh);
1051
1052         return error;
1053 }
1054
1055 /**
1056  * gfs2_glock_poll - poll to see if an async request has been completed
1057  * @gh: the holder
1058  *
1059  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1060  */
1061
1062 int gfs2_glock_poll(struct gfs2_holder *gh)
1063 {
1064         return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1065 }
1066
1067 /**
1068  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1069  * @gh: the glock holder
1070  *
1071  */
1072
1073 void gfs2_glock_dq(struct gfs2_holder *gh)
1074 {
1075         struct gfs2_glock *gl = gh->gh_gl;
1076         const struct gfs2_glock_operations *glops = gl->gl_ops;
1077         unsigned delay = 0;
1078         int fast_path = 0;
1079
1080         spin_lock(&gl->gl_spin);
1081         if (gh->gh_flags & GL_NOCACHE)
1082                 handle_callback(gl, LM_ST_UNLOCKED, 0);
1083
1084         list_del_init(&gh->gh_list);
1085         if (find_first_holder(gl) == NULL) {
1086                 if (glops->go_unlock) {
1087                         GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1088                         spin_unlock(&gl->gl_spin);
1089                         glops->go_unlock(gh);
1090                         spin_lock(&gl->gl_spin);
1091                         clear_bit(GLF_LOCK, &gl->gl_flags);
1092                 }
1093                 if (list_empty(&gl->gl_holders) &&
1094                     !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1095                     !test_bit(GLF_DEMOTE, &gl->gl_flags))
1096                         fast_path = 1;
1097         }
1098         if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1099                 gfs2_glock_add_to_lru(gl);
1100
1101         trace_gfs2_glock_queue(gh, 0);
1102         spin_unlock(&gl->gl_spin);
1103         if (likely(fast_path))
1104                 return;
1105
1106         gfs2_glock_hold(gl);
1107         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1108             !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1109             gl->gl_name.ln_type == LM_TYPE_INODE)
1110                 delay = gl->gl_hold_time;
1111         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1112                 gfs2_glock_put(gl);
1113 }
1114
1115 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1116 {
1117         struct gfs2_glock *gl = gh->gh_gl;
1118         gfs2_glock_dq(gh);
1119         might_sleep();
1120         wait_on_bit(&gl->gl_flags, GLF_DEMOTE, gfs2_glock_demote_wait, TASK_UNINTERRUPTIBLE);
1121 }
1122
1123 /**
1124  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1125  * @gh: the holder structure
1126  *
1127  */
1128
1129 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1130 {
1131         gfs2_glock_dq(gh);
1132         gfs2_holder_uninit(gh);
1133 }
1134
1135 /**
1136  * gfs2_glock_nq_num - acquire a glock based on lock number
1137  * @sdp: the filesystem
1138  * @number: the lock number
1139  * @glops: the glock operations for the type of glock
1140  * @state: the state to acquire the glock in
1141  * @flags: modifier flags for the acquisition
1142  * @gh: the struct gfs2_holder
1143  *
1144  * Returns: errno
1145  */
1146
1147 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1148                       const struct gfs2_glock_operations *glops,
1149                       unsigned int state, int flags, struct gfs2_holder *gh)
1150 {
1151         struct gfs2_glock *gl;
1152         int error;
1153
1154         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1155         if (!error) {
1156                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1157                 gfs2_glock_put(gl);
1158         }
1159
1160         return error;
1161 }
1162
1163 /**
1164  * glock_compare - Compare two struct gfs2_glock structures for sorting
1165  * @arg_a: the first structure
1166  * @arg_b: the second structure
1167  *
1168  */
1169
1170 static int glock_compare(const void *arg_a, const void *arg_b)
1171 {
1172         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1173         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1174         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1175         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1176
1177         if (a->ln_number > b->ln_number)
1178                 return 1;
1179         if (a->ln_number < b->ln_number)
1180                 return -1;
1181         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1182         return 0;
1183 }
1184
1185 /**
1186  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1187  * @num_gh: the number of structures
1188  * @ghs: an array of struct gfs2_holder structures
1189  *
1190  * Returns: 0 on success (all glocks acquired),
1191  *          errno on failure (no glocks acquired)
1192  */
1193
1194 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1195                      struct gfs2_holder **p)
1196 {
1197         unsigned int x;
1198         int error = 0;
1199
1200         for (x = 0; x < num_gh; x++)
1201                 p[x] = &ghs[x];
1202
1203         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1204
1205         for (x = 0; x < num_gh; x++) {
1206                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1207
1208                 error = gfs2_glock_nq(p[x]);
1209                 if (error) {
1210                         while (x--)
1211                                 gfs2_glock_dq(p[x]);
1212                         break;
1213                 }
1214         }
1215
1216         return error;
1217 }
1218
1219 /**
1220  * gfs2_glock_nq_m - acquire multiple glocks
1221  * @num_gh: the number of structures
1222  * @ghs: an array of struct gfs2_holder structures
1223  *
1224  *
1225  * Returns: 0 on success (all glocks acquired),
1226  *          errno on failure (no glocks acquired)
1227  */
1228
1229 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1230 {
1231         struct gfs2_holder *tmp[4];
1232         struct gfs2_holder **pph = tmp;
1233         int error = 0;
1234
1235         switch(num_gh) {
1236         case 0:
1237                 return 0;
1238         case 1:
1239                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1240                 return gfs2_glock_nq(ghs);
1241         default:
1242                 if (num_gh <= 4)
1243                         break;
1244                 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1245                 if (!pph)
1246                         return -ENOMEM;
1247         }
1248
1249         error = nq_m_sync(num_gh, ghs, pph);
1250
1251         if (pph != tmp)
1252                 kfree(pph);
1253
1254         return error;
1255 }
1256
1257 /**
1258  * gfs2_glock_dq_m - release multiple glocks
1259  * @num_gh: the number of structures
1260  * @ghs: an array of struct gfs2_holder structures
1261  *
1262  */
1263
1264 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1265 {
1266         while (num_gh--)
1267                 gfs2_glock_dq(&ghs[num_gh]);
1268 }
1269
1270 /**
1271  * gfs2_glock_dq_uninit_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_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1278 {
1279         while (num_gh--)
1280                 gfs2_glock_dq_uninit(&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);
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         spin_unlock(&gl->gl_spin);
1361         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1362         smp_wmb();
1363         gfs2_glock_hold(gl);
1364         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1365                 gfs2_glock_put(gl);
1366 }
1367
1368
1369 static int gfs2_shrink_glock_memory(struct shrinker *shrink,
1370                                     struct shrink_control *sc)
1371 {
1372         struct gfs2_glock *gl;
1373         int may_demote;
1374         int nr_skipped = 0;
1375         int nr = sc->nr_to_scan;
1376         gfp_t gfp_mask = sc->gfp_mask;
1377         LIST_HEAD(skipped);
1378
1379         if (nr == 0)
1380                 goto out;
1381
1382         if (!(gfp_mask & __GFP_FS))
1383                 return -1;
1384
1385         spin_lock(&lru_lock);
1386         while(nr && !list_empty(&lru_list)) {
1387                 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1388                 list_del_init(&gl->gl_lru);
1389                 clear_bit(GLF_LRU, &gl->gl_flags);
1390                 atomic_dec(&lru_count);
1391
1392                 /* Test for being demotable */
1393                 if (!test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1394                         gfs2_glock_hold(gl);
1395                         spin_unlock(&lru_lock);
1396                         spin_lock(&gl->gl_spin);
1397                         may_demote = demote_ok(gl);
1398                         if (may_demote) {
1399                                 handle_callback(gl, LM_ST_UNLOCKED, 0);
1400                                 nr--;
1401                         }
1402                         clear_bit(GLF_LOCK, &gl->gl_flags);
1403                         smp_mb__after_clear_bit();
1404                         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1405                                 gfs2_glock_put_nolock(gl);
1406                         spin_unlock(&gl->gl_spin);
1407                         spin_lock(&lru_lock);
1408                         continue;
1409                 }
1410                 nr_skipped++;
1411                 list_add(&gl->gl_lru, &skipped);
1412                 set_bit(GLF_LRU, &gl->gl_flags);
1413         }
1414         list_splice(&skipped, &lru_list);
1415         atomic_add(nr_skipped, &lru_count);
1416         spin_unlock(&lru_lock);
1417 out:
1418         return (atomic_read(&lru_count) / 100) * sysctl_vfs_cache_pressure;
1419 }
1420
1421 static struct shrinker glock_shrinker = {
1422         .shrink = gfs2_shrink_glock_memory,
1423         .seeks = DEFAULT_SEEKS,
1424 };
1425
1426 /**
1427  * examine_bucket - Call a function for glock in a hash bucket
1428  * @examiner: the function
1429  * @sdp: the filesystem
1430  * @bucket: the bucket
1431  *
1432  */
1433
1434 static void examine_bucket(glock_examiner examiner, const struct gfs2_sbd *sdp,
1435                           unsigned int hash)
1436 {
1437         struct gfs2_glock *gl;
1438         struct hlist_bl_head *head = &gl_hash_table[hash];
1439         struct hlist_bl_node *pos;
1440
1441         rcu_read_lock();
1442         hlist_bl_for_each_entry_rcu(gl, pos, head, gl_list) {
1443                 if ((gl->gl_sbd == sdp) && atomic_read(&gl->gl_ref))
1444                         examiner(gl);
1445         }
1446         rcu_read_unlock();
1447         cond_resched();
1448 }
1449
1450 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1451 {
1452         unsigned x;
1453
1454         for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1455                 examine_bucket(examiner, sdp, x);
1456 }
1457
1458
1459 /**
1460  * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1461  * @gl: The glock to thaw
1462  *
1463  * N.B. When we freeze a glock, we leave a ref to the glock outstanding,
1464  * so this has to result in the ref count being dropped by one.
1465  */
1466
1467 static void thaw_glock(struct gfs2_glock *gl)
1468 {
1469         if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1470                 return;
1471         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1472         gfs2_glock_hold(gl);
1473         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1474                 gfs2_glock_put(gl);
1475 }
1476
1477 /**
1478  * clear_glock - look at a glock and see if we can free it from glock cache
1479  * @gl: the glock to look at
1480  *
1481  */
1482
1483 static void clear_glock(struct gfs2_glock *gl)
1484 {
1485         gfs2_glock_remove_from_lru(gl);
1486
1487         spin_lock(&gl->gl_spin);
1488         if (gl->gl_state != LM_ST_UNLOCKED)
1489                 handle_callback(gl, LM_ST_UNLOCKED, 0);
1490         spin_unlock(&gl->gl_spin);
1491         gfs2_glock_hold(gl);
1492         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1493                 gfs2_glock_put(gl);
1494 }
1495
1496 /**
1497  * gfs2_glock_thaw - Thaw any frozen glocks
1498  * @sdp: The super block
1499  *
1500  */
1501
1502 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1503 {
1504         glock_hash_walk(thaw_glock, sdp);
1505 }
1506
1507 static int dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1508 {
1509         int ret;
1510         spin_lock(&gl->gl_spin);
1511         ret = __dump_glock(seq, gl);
1512         spin_unlock(&gl->gl_spin);
1513         return ret;
1514 }
1515
1516 static void dump_glock_func(struct gfs2_glock *gl)
1517 {
1518         dump_glock(NULL, gl);
1519 }
1520
1521 /**
1522  * gfs2_gl_hash_clear - Empty out the glock hash table
1523  * @sdp: the filesystem
1524  * @wait: wait until it's all gone
1525  *
1526  * Called when unmounting the filesystem.
1527  */
1528
1529 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1530 {
1531         glock_hash_walk(clear_glock, sdp);
1532         flush_workqueue(glock_workqueue);
1533         wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
1534         glock_hash_walk(dump_glock_func, sdp);
1535 }
1536
1537 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1538 {
1539         struct gfs2_glock *gl = ip->i_gl;
1540         int ret;
1541
1542         ret = gfs2_truncatei_resume(ip);
1543         gfs2_assert_withdraw(gl->gl_sbd, ret == 0);
1544
1545         spin_lock(&gl->gl_spin);
1546         clear_bit(GLF_LOCK, &gl->gl_flags);
1547         run_queue(gl, 1);
1548         spin_unlock(&gl->gl_spin);
1549 }
1550
1551 static const char *state2str(unsigned state)
1552 {
1553         switch(state) {
1554         case LM_ST_UNLOCKED:
1555                 return "UN";
1556         case LM_ST_SHARED:
1557                 return "SH";
1558         case LM_ST_DEFERRED:
1559                 return "DF";
1560         case LM_ST_EXCLUSIVE:
1561                 return "EX";
1562         }
1563         return "??";
1564 }
1565
1566 static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1567 {
1568         char *p = buf;
1569         if (flags & LM_FLAG_TRY)
1570                 *p++ = 't';
1571         if (flags & LM_FLAG_TRY_1CB)
1572                 *p++ = 'T';
1573         if (flags & LM_FLAG_NOEXP)
1574                 *p++ = 'e';
1575         if (flags & LM_FLAG_ANY)
1576                 *p++ = 'A';
1577         if (flags & LM_FLAG_PRIORITY)
1578                 *p++ = 'p';
1579         if (flags & GL_ASYNC)
1580                 *p++ = 'a';
1581         if (flags & GL_EXACT)
1582                 *p++ = 'E';
1583         if (flags & GL_NOCACHE)
1584                 *p++ = 'c';
1585         if (test_bit(HIF_HOLDER, &iflags))
1586                 *p++ = 'H';
1587         if (test_bit(HIF_WAIT, &iflags))
1588                 *p++ = 'W';
1589         if (test_bit(HIF_FIRST, &iflags))
1590                 *p++ = 'F';
1591         *p = 0;
1592         return buf;
1593 }
1594
1595 /**
1596  * dump_holder - print information about a glock holder
1597  * @seq: the seq_file struct
1598  * @gh: the glock holder
1599  *
1600  * Returns: 0 on success, -ENOBUFS when we run out of space
1601  */
1602
1603 static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1604 {
1605         struct task_struct *gh_owner = NULL;
1606         char flags_buf[32];
1607
1608         if (gh->gh_owner_pid)
1609                 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1610         gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1611                        state2str(gh->gh_state),
1612                        hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1613                        gh->gh_error,
1614                        gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1615                        gh_owner ? gh_owner->comm : "(ended)",
1616                        (void *)gh->gh_ip);
1617         return 0;
1618 }
1619
1620 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1621 {
1622         const unsigned long *gflags = &gl->gl_flags;
1623         char *p = buf;
1624
1625         if (test_bit(GLF_LOCK, gflags))
1626                 *p++ = 'l';
1627         if (test_bit(GLF_DEMOTE, gflags))
1628                 *p++ = 'D';
1629         if (test_bit(GLF_PENDING_DEMOTE, gflags))
1630                 *p++ = 'd';
1631         if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1632                 *p++ = 'p';
1633         if (test_bit(GLF_DIRTY, gflags))
1634                 *p++ = 'y';
1635         if (test_bit(GLF_LFLUSH, gflags))
1636                 *p++ = 'f';
1637         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1638                 *p++ = 'i';
1639         if (test_bit(GLF_REPLY_PENDING, gflags))
1640                 *p++ = 'r';
1641         if (test_bit(GLF_INITIAL, gflags))
1642                 *p++ = 'I';
1643         if (test_bit(GLF_FROZEN, gflags))
1644                 *p++ = 'F';
1645         if (test_bit(GLF_QUEUED, gflags))
1646                 *p++ = 'q';
1647         if (test_bit(GLF_LRU, gflags))
1648                 *p++ = 'L';
1649         if (gl->gl_object)
1650                 *p++ = 'o';
1651         if (test_bit(GLF_BLOCKING, gflags))
1652                 *p++ = 'b';
1653         *p = 0;
1654         return buf;
1655 }
1656
1657 /**
1658  * __dump_glock - print information about a glock
1659  * @seq: The seq_file struct
1660  * @gl: the glock
1661  *
1662  * The file format is as follows:
1663  * One line per object, capital letters are used to indicate objects
1664  * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1665  * other objects are indented by a single space and follow the glock to
1666  * which they are related. Fields are indicated by lower case letters
1667  * followed by a colon and the field value, except for strings which are in
1668  * [] so that its possible to see if they are composed of spaces for
1669  * example. The field's are n = number (id of the object), f = flags,
1670  * t = type, s = state, r = refcount, e = error, p = pid.
1671  *
1672  * Returns: 0 on success, -ENOBUFS when we run out of space
1673  */
1674
1675 static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1676 {
1677         const struct gfs2_glock_operations *glops = gl->gl_ops;
1678         unsigned long long dtime;
1679         const struct gfs2_holder *gh;
1680         char gflags_buf[32];
1681         int error = 0;
1682
1683         dtime = jiffies - gl->gl_demote_time;
1684         dtime *= 1000000/HZ; /* demote time in uSec */
1685         if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1686                 dtime = 0;
1687         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",
1688                   state2str(gl->gl_state),
1689                   gl->gl_name.ln_type,
1690                   (unsigned long long)gl->gl_name.ln_number,
1691                   gflags2str(gflags_buf, gl),
1692                   state2str(gl->gl_target),
1693                   state2str(gl->gl_demote_state), dtime,
1694                   atomic_read(&gl->gl_ail_count),
1695                   atomic_read(&gl->gl_revokes),
1696                   atomic_read(&gl->gl_ref), gl->gl_hold_time);
1697
1698         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1699                 error = dump_holder(seq, gh);
1700                 if (error)
1701                         goto out;
1702         }
1703         if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1704                 error = glops->go_dump(seq, gl);
1705 out:
1706         return error;
1707 }
1708
1709 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
1710 {
1711         struct gfs2_glock *gl = iter_ptr;
1712
1713         seq_printf(seq, "G: n:%u/%llx rtt:%lld/%lld rttb:%lld/%lld irt:%lld/%lld dcnt: %lld qcnt: %lld\n",
1714                    gl->gl_name.ln_type,
1715                    (unsigned long long)gl->gl_name.ln_number,
1716                    (long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
1717                    (long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
1718                    (long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
1719                    (long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
1720                    (long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
1721                    (long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
1722                    (long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
1723                    (long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
1724         return 0;
1725 }
1726
1727 static const char *gfs2_gltype[] = {
1728         "type",
1729         "reserved",
1730         "nondisk",
1731         "inode",
1732         "rgrp",
1733         "meta",
1734         "iopen",
1735         "flock",
1736         "plock",
1737         "quota",
1738         "journal",
1739 };
1740
1741 static const char *gfs2_stype[] = {
1742         [GFS2_LKS_SRTT]         = "srtt",
1743         [GFS2_LKS_SRTTVAR]      = "srttvar",
1744         [GFS2_LKS_SRTTB]        = "srttb",
1745         [GFS2_LKS_SRTTVARB]     = "srttvarb",
1746         [GFS2_LKS_SIRT]         = "sirt",
1747         [GFS2_LKS_SIRTVAR]      = "sirtvar",
1748         [GFS2_LKS_DCOUNT]       = "dlm",
1749         [GFS2_LKS_QCOUNT]       = "queue",
1750 };
1751
1752 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1753
1754 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
1755 {
1756         struct gfs2_glock_iter *gi = seq->private;
1757         struct gfs2_sbd *sdp = gi->sdp;
1758         unsigned index = gi->hash >> 3;
1759         unsigned subindex = gi->hash & 0x07;
1760         s64 value;
1761         int i;
1762
1763         if (index == 0 && subindex != 0)
1764                 return 0;
1765
1766         seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
1767                    (index == 0) ? "cpu": gfs2_stype[subindex]);
1768
1769         for_each_possible_cpu(i) {
1770                 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
1771                 if (index == 0) {
1772                         value = i;
1773                 } else {
1774                         value = lkstats->lkstats[index - 1].stats[subindex];
1775                 }
1776                 seq_printf(seq, " %15lld", (long long)value);
1777         }
1778         seq_putc(seq, '\n');
1779         return 0;
1780 }
1781
1782 int __init gfs2_glock_init(void)
1783 {
1784         unsigned i;
1785         for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
1786                 INIT_HLIST_BL_HEAD(&gl_hash_table[i]);
1787         }
1788
1789         glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1790                                           WQ_HIGHPRI | WQ_FREEZABLE, 0);
1791         if (IS_ERR(glock_workqueue))
1792                 return PTR_ERR(glock_workqueue);
1793         gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1794                                                 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1795                                                 0);
1796         if (IS_ERR(gfs2_delete_workqueue)) {
1797                 destroy_workqueue(glock_workqueue);
1798                 return PTR_ERR(gfs2_delete_workqueue);
1799         }
1800
1801         register_shrinker(&glock_shrinker);
1802
1803         return 0;
1804 }
1805
1806 void gfs2_glock_exit(void)
1807 {
1808         unregister_shrinker(&glock_shrinker);
1809         destroy_workqueue(glock_workqueue);
1810         destroy_workqueue(gfs2_delete_workqueue);
1811 }
1812
1813 static inline struct gfs2_glock *glock_hash_chain(unsigned hash)
1814 {
1815         return hlist_bl_entry(hlist_bl_first_rcu(&gl_hash_table[hash]),
1816                               struct gfs2_glock, gl_list);
1817 }
1818
1819 static inline struct gfs2_glock *glock_hash_next(struct gfs2_glock *gl)
1820 {
1821         return hlist_bl_entry(rcu_dereference(gl->gl_list.next),
1822                               struct gfs2_glock, gl_list);
1823 }
1824
1825 static int gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1826 {
1827         struct gfs2_glock *gl;
1828
1829         do {
1830                 gl = gi->gl;
1831                 if (gl) {
1832                         gi->gl = glock_hash_next(gl);
1833                         gi->nhash++;
1834                 } else {
1835                         if (gi->hash >= GFS2_GL_HASH_SIZE) {
1836                                 rcu_read_unlock();
1837                                 return 1;
1838                         }
1839                         gi->gl = glock_hash_chain(gi->hash);
1840                         gi->nhash = 0;
1841                 }
1842                 while (gi->gl == NULL) {
1843                         gi->hash++;
1844                         if (gi->hash >= GFS2_GL_HASH_SIZE) {
1845                                 rcu_read_unlock();
1846                                 return 1;
1847                         }
1848                         gi->gl = glock_hash_chain(gi->hash);
1849                         gi->nhash = 0;
1850                 }
1851         /* Skip entries for other sb and dead entries */
1852         } while (gi->sdp != gi->gl->gl_sbd || atomic_read(&gi->gl->gl_ref) == 0);
1853
1854         return 0;
1855 }
1856
1857 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1858 {
1859         struct gfs2_glock_iter *gi = seq->private;
1860         loff_t n = *pos;
1861
1862         if (gi->last_pos <= *pos)
1863                 n = gi->nhash + (*pos - gi->last_pos);
1864         else
1865                 gi->hash = 0;
1866
1867         gi->nhash = 0;
1868         rcu_read_lock();
1869
1870         do {
1871                 if (gfs2_glock_iter_next(gi))
1872                         return NULL;
1873         } while (n--);
1874
1875         gi->last_pos = *pos;
1876         return gi->gl;
1877 }
1878
1879 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1880                                  loff_t *pos)
1881 {
1882         struct gfs2_glock_iter *gi = seq->private;
1883
1884         (*pos)++;
1885         gi->last_pos = *pos;
1886         if (gfs2_glock_iter_next(gi))
1887                 return NULL;
1888
1889         return gi->gl;
1890 }
1891
1892 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1893 {
1894         struct gfs2_glock_iter *gi = seq->private;
1895
1896         if (gi->gl)
1897                 rcu_read_unlock();
1898         gi->gl = NULL;
1899 }
1900
1901 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1902 {
1903         return dump_glock(seq, iter_ptr);
1904 }
1905
1906 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
1907 {
1908         struct gfs2_glock_iter *gi = seq->private;
1909
1910         gi->hash = *pos;
1911         if (*pos >= GFS2_NR_SBSTATS)
1912                 return NULL;
1913         preempt_disable();
1914         return SEQ_START_TOKEN;
1915 }
1916
1917 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
1918                                    loff_t *pos)
1919 {
1920         struct gfs2_glock_iter *gi = seq->private;
1921         (*pos)++;
1922         gi->hash++;
1923         if (gi->hash >= GFS2_NR_SBSTATS) {
1924                 preempt_enable();
1925                 return NULL;
1926         }
1927         return SEQ_START_TOKEN;
1928 }
1929
1930 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
1931 {
1932         preempt_enable();
1933 }
1934
1935 static const struct seq_operations gfs2_glock_seq_ops = {
1936         .start = gfs2_glock_seq_start,
1937         .next  = gfs2_glock_seq_next,
1938         .stop  = gfs2_glock_seq_stop,
1939         .show  = gfs2_glock_seq_show,
1940 };
1941
1942 static const struct seq_operations gfs2_glstats_seq_ops = {
1943         .start = gfs2_glock_seq_start,
1944         .next  = gfs2_glock_seq_next,
1945         .stop  = gfs2_glock_seq_stop,
1946         .show  = gfs2_glstats_seq_show,
1947 };
1948
1949 static const struct seq_operations gfs2_sbstats_seq_ops = {
1950         .start = gfs2_sbstats_seq_start,
1951         .next  = gfs2_sbstats_seq_next,
1952         .stop  = gfs2_sbstats_seq_stop,
1953         .show  = gfs2_sbstats_seq_show,
1954 };
1955
1956 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
1957
1958 static int gfs2_glocks_open(struct inode *inode, struct file *file)
1959 {
1960         int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1961                                    sizeof(struct gfs2_glock_iter));
1962         if (ret == 0) {
1963                 struct seq_file *seq = file->private_data;
1964                 struct gfs2_glock_iter *gi = seq->private;
1965                 gi->sdp = inode->i_private;
1966                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
1967                 if (seq->buf)
1968                         seq->size = GFS2_SEQ_GOODSIZE;
1969         }
1970         return ret;
1971 }
1972
1973 static int gfs2_glstats_open(struct inode *inode, struct file *file)
1974 {
1975         int ret = seq_open_private(file, &gfs2_glstats_seq_ops,
1976                                    sizeof(struct gfs2_glock_iter));
1977         if (ret == 0) {
1978                 struct seq_file *seq = file->private_data;
1979                 struct gfs2_glock_iter *gi = seq->private;
1980                 gi->sdp = inode->i_private;
1981                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
1982                 if (seq->buf)
1983                         seq->size = GFS2_SEQ_GOODSIZE;
1984         }
1985         return ret;
1986 }
1987
1988 static int gfs2_sbstats_open(struct inode *inode, struct file *file)
1989 {
1990         int ret = seq_open_private(file, &gfs2_sbstats_seq_ops,
1991                                    sizeof(struct gfs2_glock_iter));
1992         if (ret == 0) {
1993                 struct seq_file *seq = file->private_data;
1994                 struct gfs2_glock_iter *gi = seq->private;
1995                 gi->sdp = inode->i_private;
1996         }
1997         return ret;
1998 }
1999
2000 static const struct file_operations gfs2_glocks_fops = {
2001         .owner   = THIS_MODULE,
2002         .open    = gfs2_glocks_open,
2003         .read    = seq_read,
2004         .llseek  = seq_lseek,
2005         .release = seq_release_private,
2006 };
2007
2008 static const struct file_operations gfs2_glstats_fops = {
2009         .owner   = THIS_MODULE,
2010         .open    = gfs2_glstats_open,
2011         .read    = seq_read,
2012         .llseek  = seq_lseek,
2013         .release = seq_release_private,
2014 };
2015
2016 static const struct file_operations gfs2_sbstats_fops = {
2017         .owner   = THIS_MODULE,
2018         .open    = gfs2_sbstats_open,
2019         .read    = seq_read,
2020         .llseek  = seq_lseek,
2021         .release = seq_release_private,
2022 };
2023
2024 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2025 {
2026         sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2027         if (!sdp->debugfs_dir)
2028                 return -ENOMEM;
2029         sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
2030                                                          S_IFREG | S_IRUGO,
2031                                                          sdp->debugfs_dir, sdp,
2032                                                          &gfs2_glocks_fops);
2033         if (!sdp->debugfs_dentry_glocks)
2034                 goto fail;
2035
2036         sdp->debugfs_dentry_glstats = debugfs_create_file("glstats",
2037                                                         S_IFREG | S_IRUGO,
2038                                                         sdp->debugfs_dir, sdp,
2039                                                         &gfs2_glstats_fops);
2040         if (!sdp->debugfs_dentry_glstats)
2041                 goto fail;
2042
2043         sdp->debugfs_dentry_sbstats = debugfs_create_file("sbstats",
2044                                                         S_IFREG | S_IRUGO,
2045                                                         sdp->debugfs_dir, sdp,
2046                                                         &gfs2_sbstats_fops);
2047         if (!sdp->debugfs_dentry_sbstats)
2048                 goto fail;
2049
2050         return 0;
2051 fail:
2052         gfs2_delete_debugfs_file(sdp);
2053         return -ENOMEM;
2054 }
2055
2056 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2057 {
2058         if (sdp->debugfs_dir) {
2059                 if (sdp->debugfs_dentry_glocks) {
2060                         debugfs_remove(sdp->debugfs_dentry_glocks);
2061                         sdp->debugfs_dentry_glocks = NULL;
2062                 }
2063                 if (sdp->debugfs_dentry_glstats) {
2064                         debugfs_remove(sdp->debugfs_dentry_glstats);
2065                         sdp->debugfs_dentry_glstats = NULL;
2066                 }
2067                 if (sdp->debugfs_dentry_sbstats) {
2068                         debugfs_remove(sdp->debugfs_dentry_sbstats);
2069                         sdp->debugfs_dentry_sbstats = NULL;
2070                 }
2071                 debugfs_remove(sdp->debugfs_dir);
2072                 sdp->debugfs_dir = NULL;
2073         }
2074 }
2075
2076 int gfs2_register_debugfs(void)
2077 {
2078         gfs2_root = debugfs_create_dir("gfs2", NULL);
2079         return gfs2_root ? 0 : -ENOMEM;
2080 }
2081
2082 void gfs2_unregister_debugfs(void)
2083 {
2084         debugfs_remove(gfs2_root);
2085         gfs2_root = NULL;
2086 }