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 <linux/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 #include <linux/rhashtable.h>
  38
  39 #include "gfs2.h"
  40 #include "incore.h"
  41 #include "glock.h"
  42 #include "glops.h"
  43 #include "inode.h"
  44 #include "lops.h"
  45 #include "meta_io.h"
  46 #include "quota.h"
  47 #include "super.h"
  48 #include "util.h"
  49 #include "bmap.h"
  50 #define CREATE_TRACE_POINTS
  51 #include "trace_gfs2.h"
  52
  53 struct gfs2_glock_iter {
  54         struct gfs2_sbd *sdp;           /* incore superblock           */
  55         struct rhashtable_iter hti;     /* rhashtable iterator         */
  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       BIT(GFS2_GL_HASH_SHIFT)
  73
  74 static struct rhashtable_params ht_parms = {
  75         .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4,
  76         .key_len = sizeof(struct lm_lockname),
  77         .key_offset = offsetof(struct gfs2_glock, gl_name),
  78         .head_offset = offsetof(struct gfs2_glock, gl_node),
  79 };
  80
  81 static struct rhashtable gl_hash_table;
  82
  83 void gfs2_glock_free(struct gfs2_glock *gl)
  84 {
  85         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
  86
  87         if (gl->gl_ops->go_flags & GLOF_ASPACE) {
  88                 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
  89         } else {
  90                 kfree(gl->gl_lksb.sb_lvbptr);
  91                 kmem_cache_free(gfs2_glock_cachep, gl);
  92         }
  93         if (atomic_dec_and_test(&sdp->sd_glock_disposal))
  94                 wake_up(&sdp->sd_glock_wait);
  95 }
  96
  97 /**
  98  * gfs2_glock_hold() - increment reference count on glock
  99  * @gl: The glock to hold
 100  *
 101  */
 102
 103 static void gfs2_glock_hold(struct gfs2_glock *gl)
 104 {
 105         GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
 106         lockref_get(&gl->gl_lockref);
 107 }
 108
 109 /**
 110  * demote_ok - Check to see if it's ok to unlock a glock
 111  * @gl: the glock
 112  *
 113  * Returns: 1 if it's ok
 114  */
 115
 116 static int demote_ok(const struct gfs2_glock *gl)
 117 {
 118         const struct gfs2_glock_operations *glops = gl->gl_ops;
 119
 120         if (gl->gl_state == LM_ST_UNLOCKED)
 121                 return 0;
 122         if (!list_empty(&gl->gl_holders))
 123                 return 0;
 124         if (glops->go_demote_ok)
 125                 return glops->go_demote_ok(gl);
 126         return 1;
 127 }
 128
 129
 130 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
 131 {
 132         spin_lock(&lru_lock);
 133
 134         if (!list_empty(&gl->gl_lru))
 135                 list_del_init(&gl->gl_lru);
 136         else
 137                 atomic_inc(&lru_count);
 138
 139         list_add_tail(&gl->gl_lru, &lru_list);
 140         set_bit(GLF_LRU, &gl->gl_flags);
 141         spin_unlock(&lru_lock);
 142 }
 143
 144 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
 145 {
 146         spin_lock(&lru_lock);
 147         if (!list_empty(&gl->gl_lru)) {
 148                 list_del_init(&gl->gl_lru);
 149                 atomic_dec(&lru_count);
 150                 clear_bit(GLF_LRU, &gl->gl_flags);
 151         }
 152         spin_unlock(&lru_lock);
 153 }
 154
 155 /**
 156  * gfs2_glock_put() - Decrement reference count on glock
 157  * @gl: The glock to put
 158  *
 159  */
 160
 161 void gfs2_glock_put(struct gfs2_glock *gl)
 162 {
 163         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 164         struct address_space *mapping = gfs2_glock2aspace(gl);
 165
 166         if (lockref_put_or_lock(&gl->gl_lockref))
 167                 return;
 168
 169         lockref_mark_dead(&gl->gl_lockref);
 170
 171         gfs2_glock_remove_from_lru(gl);
 172         spin_unlock(&gl->gl_lockref.lock);
 173         rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
 174         GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
 175         GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
 176         trace_gfs2_glock_put(gl);
 177         sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
 178 }
 179
 180 /**
 181  * may_grant - check if its ok to grant a new lock
 182  * @gl: The glock
 183  * @gh: The lock request which we wish to grant
 184  *
 185  * Returns: true if its ok to grant the lock
 186  */
 187
 188 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
 189 {
 190         const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
 191         if ((gh->gh_state == LM_ST_EXCLUSIVE ||
 192              gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
 193                 return 0;
 194         if (gl->gl_state == gh->gh_state)
 195                 return 1;
 196         if (gh->gh_flags & GL_EXACT)
 197                 return 0;
 198         if (gl->gl_state == LM_ST_EXCLUSIVE) {
 199                 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
 200                         return 1;
 201                 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
 202                         return 1;
 203         }
 204         if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
 205                 return 1;
 206         return 0;
 207 }
 208
 209 static void gfs2_holder_wake(struct gfs2_holder *gh)
 210 {
 211         clear_bit(HIF_WAIT, &gh->gh_iflags);
 212         smp_mb__after_atomic();
 213         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
 214 }
 215
 216 /**
 217  * do_error - Something unexpected has happened during a lock request
 218  *
 219  */
 220
 221 static void do_error(struct gfs2_glock *gl, const int ret)
 222 {
 223         struct gfs2_holder *gh, *tmp;
 224
 225         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
 226                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 227                         continue;
 228                 if (ret & LM_OUT_ERROR)
 229                         gh->gh_error = -EIO;
 230                 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
 231                         gh->gh_error = GLR_TRYFAILED;
 232                 else
 233                         continue;
 234                 list_del_init(&gh->gh_list);
 235                 trace_gfs2_glock_queue(gh, 0);
 236                 gfs2_holder_wake(gh);
 237         }
 238 }
 239
 240 /**
 241  * do_promote - promote as many requests as possible on the current queue
 242  * @gl: The glock
 243  *
 244  * Returns: 1 if there is a blocked holder at the head of the list, or 2
 245  *          if a type specific operation is underway.
 246  */
 247
 248 static int do_promote(struct gfs2_glock *gl)
 249 __releases(&gl->gl_lockref.lock)
 250 __acquires(&gl->gl_lockref.lock)
 251 {
 252         const struct gfs2_glock_operations *glops = gl->gl_ops;
 253         struct gfs2_holder *gh, *tmp;
 254         int ret;
 255
 256 restart:
 257         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
 258                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 259                         continue;
 260                 if (may_grant(gl, gh)) {
 261                         if (gh->gh_list.prev == &gl->gl_holders &&
 262                             glops->go_lock) {
 263                                 spin_unlock(&gl->gl_lockref.lock);
 264                                 /* FIXME: eliminate this eventually */
 265                                 ret = glops->go_lock(gh);
 266                                 spin_lock(&gl->gl_lockref.lock);
 267                                 if (ret) {
 268                                         if (ret == 1)
 269                                                 return 2;
 270                                         gh->gh_error = ret;
 271                                         list_del_init(&gh->gh_list);
 272                                         trace_gfs2_glock_queue(gh, 0);
 273                                         gfs2_holder_wake(gh);
 274                                         goto restart;
 275                                 }
 276                                 set_bit(HIF_HOLDER, &gh->gh_iflags);
 277                                 trace_gfs2_promote(gh, 1);
 278                                 gfs2_holder_wake(gh);
 279                                 goto restart;
 280                         }
 281                         set_bit(HIF_HOLDER, &gh->gh_iflags);
 282                         trace_gfs2_promote(gh, 0);
 283                         gfs2_holder_wake(gh);
 284                         continue;
 285                 }
 286                 if (gh->gh_list.prev == &gl->gl_holders)
 287                         return 1;
 288                 do_error(gl, 0);
 289                 break;
 290         }
 291         return 0;
 292 }
 293
 294 /**
 295  * find_first_waiter - find the first gh that's waiting for the glock
 296  * @gl: the glock
 297  */
 298
 299 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
 300 {
 301         struct gfs2_holder *gh;
 302
 303         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
 304                 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
 305                         return gh;
 306         }
 307         return NULL;
 308 }
 309
 310 /**
 311  * state_change - record that the glock is now in a different state
 312  * @gl: the glock
 313  * @new_state the new state
 314  *
 315  */
 316
 317 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
 318 {
 319         int held1, held2;
 320
 321         held1 = (gl->gl_state != LM_ST_UNLOCKED);
 322         held2 = (new_state != LM_ST_UNLOCKED);
 323
 324         if (held1 != held2) {
 325                 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
 326                 if (held2)
 327                         gl->gl_lockref.count++;
 328                 else
 329                         gl->gl_lockref.count--;
 330         }
 331         if (held1 && held2 && list_empty(&gl->gl_holders))
 332                 clear_bit(GLF_QUEUED, &gl->gl_flags);
 333
 334         if (new_state != gl->gl_target)
 335                 /* shorten our minimum hold time */
 336                 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
 337                                        GL_GLOCK_MIN_HOLD);
 338         gl->gl_state = new_state;
 339         gl->gl_tchange = jiffies;
 340 }
 341
 342 static void gfs2_demote_wake(struct gfs2_glock *gl)
 343 {
 344         gl->gl_demote_state = LM_ST_EXCLUSIVE;
 345         clear_bit(GLF_DEMOTE, &gl->gl_flags);
 346         smp_mb__after_atomic();
 347         wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
 348 }
 349
 350 /**
 351  * finish_xmote - The DLM has replied to one of our lock requests
 352  * @gl: The glock
 353  * @ret: The status from the DLM
 354  *
 355  */
 356
 357 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
 358 {
 359         const struct gfs2_glock_operations *glops = gl->gl_ops;
 360         struct gfs2_holder *gh;
 361         unsigned state = ret & LM_OUT_ST_MASK;
 362         int rv;
 363
 364         spin_lock(&gl->gl_lockref.lock);
 365         trace_gfs2_glock_state_change(gl, state);
 366         state_change(gl, state);
 367         gh = find_first_waiter(gl);
 368
 369         /* Demote to UN request arrived during demote to SH or DF */
 370         if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
 371             state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
 372                 gl->gl_target = LM_ST_UNLOCKED;
 373
 374         /* Check for state != intended state */
 375         if (unlikely(state != gl->gl_target)) {
 376                 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
 377                         /* move to back of queue and try next entry */
 378                         if (ret & LM_OUT_CANCELED) {
 379                                 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
 380                                         list_move_tail(&gh->gh_list, &gl->gl_holders);
 381                                 gh = find_first_waiter(gl);
 382                                 gl->gl_target = gh->gh_state;
 383                                 goto retry;
 384                         }
 385                         /* Some error or failed "try lock" - report it */
 386                         if ((ret & LM_OUT_ERROR) ||
 387                             (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
 388                                 gl->gl_target = gl->gl_state;
 389                                 do_error(gl, ret);
 390                                 goto out;
 391                         }
 392                 }
 393                 switch(state) {
 394                 /* Unlocked due to conversion deadlock, try again */
 395                 case LM_ST_UNLOCKED:
 396 retry:
 397                         do_xmote(gl, gh, gl->gl_target);
 398                         break;
 399                 /* Conversion fails, unlock and try again */
 400                 case LM_ST_SHARED:
 401                 case LM_ST_DEFERRED:
 402                         do_xmote(gl, gh, LM_ST_UNLOCKED);
 403                         break;
 404                 default: /* Everything else */
 405                         pr_err("wanted %u got %u\n", gl->gl_target, state);
 406                         GLOCK_BUG_ON(gl, 1);
 407                 }
 408                 spin_unlock(&gl->gl_lockref.lock);
 409                 return;
 410         }
 411
 412         /* Fast path - we got what we asked for */
 413         if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
 414                 gfs2_demote_wake(gl);
 415         if (state != LM_ST_UNLOCKED) {
 416                 if (glops->go_xmote_bh) {
 417                         spin_unlock(&gl->gl_lockref.lock);
 418                         rv = glops->go_xmote_bh(gl, gh);
 419                         spin_lock(&gl->gl_lockref.lock);
 420                         if (rv) {
 421                                 do_error(gl, rv);
 422                                 goto out;
 423                         }
 424                 }
 425                 rv = do_promote(gl);
 426                 if (rv == 2)
 427                         goto out_locked;
 428         }
 429 out:
 430         clear_bit(GLF_LOCK, &gl->gl_flags);
 431 out_locked:
 432         spin_unlock(&gl->gl_lockref.lock);
 433 }
 434
 435 /**
 436  * do_xmote - Calls the DLM to change the state of a lock
 437  * @gl: The lock state
 438  * @gh: The holder (only for promotes)
 439  * @target: The target lock state
 440  *
 441  */
 442
 443 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
 444 __releases(&gl->gl_lockref.lock)
 445 __acquires(&gl->gl_lockref.lock)
 446 {
 447         const struct gfs2_glock_operations *glops = gl->gl_ops;
 448         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 449         unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
 450         int ret;
 451
 452         lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
 453                       LM_FLAG_PRIORITY);
 454         GLOCK_BUG_ON(gl, gl->gl_state == target);
 455         GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
 456         if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
 457             glops->go_inval) {
 458                 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 459                 do_error(gl, 0); /* Fail queued try locks */
 460         }
 461         gl->gl_req = target;
 462         set_bit(GLF_BLOCKING, &gl->gl_flags);
 463         if ((gl->gl_req == LM_ST_UNLOCKED) ||
 464             (gl->gl_state == LM_ST_EXCLUSIVE) ||
 465             (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
 466                 clear_bit(GLF_BLOCKING, &gl->gl_flags);
 467         spin_unlock(&gl->gl_lockref.lock);
 468         if (glops->go_sync)
 469                 glops->go_sync(gl);
 470         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 471                 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
 472         clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 473
 474         gfs2_glock_hold(gl);
 475         if (sdp->sd_lockstruct.ls_ops->lm_lock) {
 476                 /* lock_dlm */
 477                 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
 478                 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
 479                     target == LM_ST_UNLOCKED &&
 480                     test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
 481                         finish_xmote(gl, target);
 482                         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 483                                 gfs2_glock_put(gl);
 484                 }
 485                 else if (ret) {
 486                         pr_err("lm_lock ret %d\n", ret);
 487                         GLOCK_BUG_ON(gl, 1);
 488                 }
 489         } else { /* lock_nolock */
 490                 finish_xmote(gl, target);
 491                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 492                         gfs2_glock_put(gl);
 493         }
 494
 495         spin_lock(&gl->gl_lockref.lock);
 496 }
 497
 498 /**
 499  * find_first_holder - find the first "holder" gh
 500  * @gl: the glock
 501  */
 502
 503 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
 504 {
 505         struct gfs2_holder *gh;
 506
 507         if (!list_empty(&gl->gl_holders)) {
 508                 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
 509                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 510                         return gh;
 511         }
 512         return NULL;
 513 }
 514
 515 /**
 516  * run_queue - do all outstanding tasks related to a glock
 517  * @gl: The glock in question
 518  * @nonblock: True if we must not block in run_queue
 519  *
 520  */
 521
 522 static void run_queue(struct gfs2_glock *gl, const int nonblock)
 523 __releases(&gl->gl_lockref.lock)
 524 __acquires(&gl->gl_lockref.lock)
 525 {
 526         struct gfs2_holder *gh = NULL;
 527         int ret;
 528
 529         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
 530                 return;
 531
 532         GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
 533
 534         if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
 535             gl->gl_demote_state != gl->gl_state) {
 536                 if (find_first_holder(gl))
 537                         goto out_unlock;
 538                 if (nonblock)
 539                         goto out_sched;
 540                 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
 541                 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
 542                 gl->gl_target = gl->gl_demote_state;
 543         } else {
 544                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
 545                         gfs2_demote_wake(gl);
 546                 ret = do_promote(gl);
 547                 if (ret == 0)
 548                         goto out_unlock;
 549                 if (ret == 2)
 550                         goto out;
 551                 gh = find_first_waiter(gl);
 552                 gl->gl_target = gh->gh_state;
 553                 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
 554                         do_error(gl, 0); /* Fail queued try locks */
 555         }
 556         do_xmote(gl, gh, gl->gl_target);
 557 out:
 558         return;
 559
 560 out_sched:
 561         clear_bit(GLF_LOCK, &gl->gl_flags);
 562         smp_mb__after_atomic();
 563         gl->gl_lockref.count++;
 564         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 565                 gl->gl_lockref.count--;
 566         return;
 567
 568 out_unlock:
 569         clear_bit(GLF_LOCK, &gl->gl_flags);
 570         smp_mb__after_atomic();
 571         return;
 572 }
 573
 574 static void delete_work_func(struct work_struct *work)
 575 {
 576         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
 577         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 578         struct inode *inode;
 579         u64 no_addr = gl->gl_name.ln_number;
 580
 581         /* If someone's using this glock to create a new dinode, the block must
 582            have been freed by another node, then re-used, in which case our
 583            iopen callback is too late after the fact. Ignore it. */
 584         if (test_bit(GLF_INODE_CREATING, &gl->gl_flags))
 585                 goto out;
 586
 587         inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
 588         if (inode && !IS_ERR(inode)) {
 589                 d_prune_aliases(inode);
 590                 iput(inode);
 591         }
 592 out:
 593         gfs2_glock_put(gl);
 594 }
 595
 596 static void glock_work_func(struct work_struct *work)
 597 {
 598         unsigned long delay = 0;
 599         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
 600         int drop_ref = 0;
 601
 602         if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
 603                 finish_xmote(gl, gl->gl_reply);
 604                 drop_ref = 1;
 605         }
 606         spin_lock(&gl->gl_lockref.lock);
 607         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
 608             gl->gl_state != LM_ST_UNLOCKED &&
 609             gl->gl_demote_state != LM_ST_EXCLUSIVE) {
 610                 unsigned long holdtime, now = jiffies;
 611
 612                 holdtime = gl->gl_tchange + gl->gl_hold_time;
 613                 if (time_before(now, holdtime))
 614                         delay = holdtime - now;
 615
 616                 if (!delay) {
 617                         clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
 618                         set_bit(GLF_DEMOTE, &gl->gl_flags);
 619                 }
 620         }
 621         run_queue(gl, 0);
 622         spin_unlock(&gl->gl_lockref.lock);
 623         if (!delay)
 624                 gfs2_glock_put(gl);
 625         else {
 626                 if (gl->gl_name.ln_type != LM_TYPE_INODE)
 627                         delay = 0;
 628                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
 629                         gfs2_glock_put(gl);
 630         }
 631         if (drop_ref)
 632                 gfs2_glock_put(gl);
 633 }
 634
 635 /**
 636  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
 637  * @sdp: The GFS2 superblock
 638  * @number: the lock number
 639  * @glops: The glock_operations to use
 640  * @create: If 0, don't create the glock if it doesn't exist
 641  * @glp: the glock is returned here
 642  *
 643  * This does not lock a glock, just finds/creates structures for one.
 644  *
 645  * Returns: errno
 646  */
 647
 648 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
 649                    const struct gfs2_glock_operations *glops, int create,
 650                    struct gfs2_glock **glp)
 651 {
 652         struct super_block *s = sdp->sd_vfs;
 653         struct lm_lockname name = { .ln_number = number,
 654                                     .ln_type = glops->go_type,
 655                                     .ln_sbd = sdp };
 656         struct gfs2_glock *gl, *tmp = NULL;
 657         struct address_space *mapping;
 658         struct kmem_cache *cachep;
 659         int ret, tries = 0;
 660
 661         gl = rhashtable_lookup_fast(&gl_hash_table, &name, ht_parms);
 662         if (gl && !lockref_get_not_dead(&gl->gl_lockref))
 663                 gl = NULL;
 664
 665         *glp = gl;
 666         if (gl)
 667                 return 0;
 668         if (!create)
 669                 return -ENOENT;
 670
 671         if (glops->go_flags & GLOF_ASPACE)
 672                 cachep = gfs2_glock_aspace_cachep;
 673         else
 674                 cachep = gfs2_glock_cachep;
 675         gl = kmem_cache_alloc(cachep, GFP_NOFS);
 676         if (!gl)
 677                 return -ENOMEM;
 678
 679         memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
 680
 681         if (glops->go_flags & GLOF_LVB) {
 682                 gl->gl_lksb.sb_lvbptr = kzalloc(GFS2_MIN_LVB_SIZE, GFP_NOFS);
 683                 if (!gl->gl_lksb.sb_lvbptr) {
 684                         kmem_cache_free(cachep, gl);
 685                         return -ENOMEM;
 686                 }
 687         }
 688
 689         atomic_inc(&sdp->sd_glock_disposal);
 690         gl->gl_node.next = NULL;
 691         gl->gl_flags = 0;
 692         gl->gl_name = name;
 693         gl->gl_lockref.count = 1;
 694         gl->gl_state = LM_ST_UNLOCKED;
 695         gl->gl_target = LM_ST_UNLOCKED;
 696         gl->gl_demote_state = LM_ST_EXCLUSIVE;
 697         gl->gl_ops = glops;
 698         gl->gl_dstamp = 0;
 699         preempt_disable();
 700         /* We use the global stats to estimate the initial per-glock stats */
 701         gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
 702         preempt_enable();
 703         gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
 704         gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
 705         gl->gl_tchange = jiffies;
 706         gl->gl_object = NULL;
 707         gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
 708         INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
 709         INIT_WORK(&gl->gl_delete, delete_work_func);
 710
 711         mapping = gfs2_glock2aspace(gl);
 712         if (mapping) {
 713                 mapping->a_ops = &gfs2_meta_aops;
 714                 mapping->host = s->s_bdev->bd_inode;
 715                 mapping->flags = 0;
 716                 mapping_set_gfp_mask(mapping, GFP_NOFS);
 717                 mapping->private_data = NULL;
 718                 mapping->writeback_index = 0;
 719         }
 720
 721 again:
 722         ret = rhashtable_lookup_insert_fast(&gl_hash_table, &gl->gl_node,
 723                                             ht_parms);
 724         if (ret == 0) {
 725                 *glp = gl;
 726                 return 0;
 727         }
 728
 729         if (ret == -EEXIST) {
 730                 ret = 0;
 731                 tmp = rhashtable_lookup_fast(&gl_hash_table, &name, ht_parms);
 732                 if (tmp == NULL || !lockref_get_not_dead(&tmp->gl_lockref)) {
 733                         if (++tries < 100) {
 734                                 cond_resched();
 735                                 goto again;
 736                         }
 737                         tmp = NULL;
 738                         ret = -ENOMEM;
 739                 }
 740         } else {
 741                 WARN_ON_ONCE(ret);
 742         }
 743         kfree(gl->gl_lksb.sb_lvbptr);
 744         kmem_cache_free(cachep, gl);
 745         atomic_dec(&sdp->sd_glock_disposal);
 746         *glp = tmp;
 747
 748         return ret;
 749 }
 750
 751 /**
 752  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
 753  * @gl: the glock
 754  * @state: the state we're requesting
 755  * @flags: the modifier flags
 756  * @gh: the holder structure
 757  *
 758  */
 759
 760 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
 761                       struct gfs2_holder *gh)
 762 {
 763         INIT_LIST_HEAD(&gh->gh_list);
 764         gh->gh_gl = gl;
 765         gh->gh_ip = _RET_IP_;
 766         gh->gh_owner_pid = get_pid(task_pid(current));
 767         gh->gh_state = state;
 768         gh->gh_flags = flags;
 769         gh->gh_error = 0;
 770         gh->gh_iflags = 0;
 771         gfs2_glock_hold(gl);
 772 }
 773
 774 /**
 775  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
 776  * @state: the state we're requesting
 777  * @flags: the modifier flags
 778  * @gh: the holder structure
 779  *
 780  * Don't mess with the glock.
 781  *
 782  */
 783
 784 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
 785 {
 786         gh->gh_state = state;
 787         gh->gh_flags = flags;
 788         gh->gh_iflags = 0;
 789         gh->gh_ip = _RET_IP_;
 790         put_pid(gh->gh_owner_pid);
 791         gh->gh_owner_pid = get_pid(task_pid(current));
 792 }
 793
 794 /**
 795  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
 796  * @gh: the holder structure
 797  *
 798  */
 799
 800 void gfs2_holder_uninit(struct gfs2_holder *gh)
 801 {
 802         put_pid(gh->gh_owner_pid);
 803         gfs2_glock_put(gh->gh_gl);
 804         gfs2_holder_mark_uninitialized(gh);
 805         gh->gh_ip = 0;
 806 }
 807
 808 /**
 809  * gfs2_glock_wait - wait on a glock acquisition
 810  * @gh: the glock holder
 811  *
 812  * Returns: 0 on success
 813  */
 814
 815 int gfs2_glock_wait(struct gfs2_holder *gh)
 816 {
 817         unsigned long time1 = jiffies;
 818
 819         might_sleep();
 820         wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
 821         if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
 822                 /* Lengthen the minimum hold time. */
 823                 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
 824                                               GL_GLOCK_HOLD_INCR,
 825                                               GL_GLOCK_MAX_HOLD);
 826         return gh->gh_error;
 827 }
 828
 829 /**
 830  * handle_callback - process a demote request
 831  * @gl: the glock
 832  * @state: the state the caller wants us to change to
 833  *
 834  * There are only two requests that we are going to see in actual
 835  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
 836  */
 837
 838 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
 839                             unsigned long delay, bool remote)
 840 {
 841         int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
 842
 843         set_bit(bit, &gl->gl_flags);
 844         if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
 845                 gl->gl_demote_state = state;
 846                 gl->gl_demote_time = jiffies;
 847         } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
 848                         gl->gl_demote_state != state) {
 849                 gl->gl_demote_state = LM_ST_UNLOCKED;
 850         }
 851         if (gl->gl_ops->go_callback)
 852                 gl->gl_ops->go_callback(gl, remote);
 853         trace_gfs2_demote_rq(gl, remote);
 854 }
 855
 856 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
 857 {
 858         struct va_format vaf;
 859         va_list args;
 860
 861         va_start(args, fmt);
 862
 863         if (seq) {
 864                 seq_vprintf(seq, fmt, args);
 865         } else {
 866                 vaf.fmt = fmt;
 867                 vaf.va = &args;
 868
 869                 pr_err("%pV", &vaf);
 870         }
 871
 872         va_end(args);
 873 }
 874
 875 /**
 876  * add_to_queue - Add a holder to the wait queue (but look for recursion)
 877  * @gh: the holder structure to add
 878  *
 879  * Eventually we should move the recursive locking trap to a
 880  * debugging option or something like that. This is the fast
 881  * path and needs to have the minimum number of distractions.
 882  *
 883  */
 884
 885 static inline void add_to_queue(struct gfs2_holder *gh)
 886 __releases(&gl->gl_lockref.lock)
 887 __acquires(&gl->gl_lockref.lock)
 888 {
 889         struct gfs2_glock *gl = gh->gh_gl;
 890         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 891         struct list_head *insert_pt = NULL;
 892         struct gfs2_holder *gh2;
 893         int try_futile = 0;
 894
 895         BUG_ON(gh->gh_owner_pid == NULL);
 896         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
 897                 BUG();
 898
 899         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
 900                 if (test_bit(GLF_LOCK, &gl->gl_flags))
 901                         try_futile = !may_grant(gl, gh);
 902                 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 903                         goto fail;
 904         }
 905
 906         list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
 907                 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
 908                     (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
 909                         goto trap_recursive;
 910                 if (try_futile &&
 911                     !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
 912 fail:
 913                         gh->gh_error = GLR_TRYFAILED;
 914                         gfs2_holder_wake(gh);
 915                         return;
 916                 }
 917                 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
 918                         continue;
 919                 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
 920                         insert_pt = &gh2->gh_list;
 921         }
 922         set_bit(GLF_QUEUED, &gl->gl_flags);
 923         trace_gfs2_glock_queue(gh, 1);
 924         gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
 925         gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
 926         if (likely(insert_pt == NULL)) {
 927                 list_add_tail(&gh->gh_list, &gl->gl_holders);
 928                 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
 929                         goto do_cancel;
 930                 return;
 931         }
 932         list_add_tail(&gh->gh_list, insert_pt);
 933 do_cancel:
 934         gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
 935         if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
 936                 spin_unlock(&gl->gl_lockref.lock);
 937                 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
 938                         sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
 939                 spin_lock(&gl->gl_lockref.lock);
 940         }
 941         return;
 942
 943 trap_recursive:
 944         pr_err("original: %pSR\n", (void *)gh2->gh_ip);
 945         pr_err("pid: %d\n", pid_nr(gh2->gh_owner_pid));
 946         pr_err("lock type: %d req lock state : %d\n",
 947                gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
 948         pr_err("new: %pSR\n", (void *)gh->gh_ip);
 949         pr_err("pid: %d\n", pid_nr(gh->gh_owner_pid));
 950         pr_err("lock type: %d req lock state : %d\n",
 951                gh->gh_gl->gl_name.ln_type, gh->gh_state);
 952         gfs2_dump_glock(NULL, gl);
 953         BUG();
 954 }
 955
 956 /**
 957  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
 958  * @gh: the holder structure
 959  *
 960  * if (gh->gh_flags & GL_ASYNC), this never returns an error
 961  *
 962  * Returns: 0, GLR_TRYFAILED, or errno on failure
 963  */
 964
 965 int gfs2_glock_nq(struct gfs2_holder *gh)
 966 {
 967         struct gfs2_glock *gl = gh->gh_gl;
 968         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 969         int error = 0;
 970
 971         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 972                 return -EIO;
 973
 974         if (test_bit(GLF_LRU, &gl->gl_flags))
 975                 gfs2_glock_remove_from_lru(gl);
 976
 977         spin_lock(&gl->gl_lockref.lock);
 978         add_to_queue(gh);
 979         if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
 980                      test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
 981                 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
 982                 gl->gl_lockref.count++;
 983                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 984                         gl->gl_lockref.count--;
 985         }
 986         run_queue(gl, 1);
 987         spin_unlock(&gl->gl_lockref.lock);
 988
 989         if (!(gh->gh_flags & GL_ASYNC))
 990                 error = gfs2_glock_wait(gh);
 991
 992         return error;
 993 }
 994
 995 /**
 996  * gfs2_glock_poll - poll to see if an async request has been completed
 997  * @gh: the holder
 998  *
 999  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1000  */
1001
1002 int gfs2_glock_poll(struct gfs2_holder *gh)
1003 {
1004         return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1005 }
1006
1007 /**
1008  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1009  * @gh: the glock holder
1010  *
1011  */
1012
1013 void gfs2_glock_dq(struct gfs2_holder *gh)
1014 {
1015         struct gfs2_glock *gl = gh->gh_gl;
1016         const struct gfs2_glock_operations *glops = gl->gl_ops;
1017         unsigned delay = 0;
1018         int fast_path = 0;
1019
1020         spin_lock(&gl->gl_lockref.lock);
1021         if (gh->gh_flags & GL_NOCACHE)
1022                 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1023
1024         list_del_init(&gh->gh_list);
1025         clear_bit(HIF_HOLDER, &gh->gh_iflags);
1026         if (find_first_holder(gl) == NULL) {
1027                 if (glops->go_unlock) {
1028                         GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1029                         spin_unlock(&gl->gl_lockref.lock);
1030                         glops->go_unlock(gh);
1031                         spin_lock(&gl->gl_lockref.lock);
1032                         clear_bit(GLF_LOCK, &gl->gl_flags);
1033                 }
1034                 if (list_empty(&gl->gl_holders) &&
1035                     !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1036                     !test_bit(GLF_DEMOTE, &gl->gl_flags))
1037                         fast_path = 1;
1038         }
1039         if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl) &&
1040             (glops->go_flags & GLOF_LRU))
1041                 gfs2_glock_add_to_lru(gl);
1042
1043         trace_gfs2_glock_queue(gh, 0);
1044         spin_unlock(&gl->gl_lockref.lock);
1045         if (likely(fast_path))
1046                 return;
1047
1048         gfs2_glock_hold(gl);
1049         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1050             !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1051             gl->gl_name.ln_type == LM_TYPE_INODE)
1052                 delay = gl->gl_hold_time;
1053         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1054                 gfs2_glock_put(gl);
1055 }
1056
1057 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1058 {
1059         struct gfs2_glock *gl = gh->gh_gl;
1060         gfs2_glock_dq(gh);
1061         might_sleep();
1062         wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1063 }
1064
1065 /**
1066  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1067  * @gh: the holder structure
1068  *
1069  */
1070
1071 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1072 {
1073         gfs2_glock_dq(gh);
1074         gfs2_holder_uninit(gh);
1075 }
1076
1077 /**
1078  * gfs2_glock_nq_num - acquire a glock based on lock number
1079  * @sdp: the filesystem
1080  * @number: the lock number
1081  * @glops: the glock operations for the type of glock
1082  * @state: the state to acquire the glock in
1083  * @flags: modifier flags for the acquisition
1084  * @gh: the struct gfs2_holder
1085  *
1086  * Returns: errno
1087  */
1088
1089 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1090                       const struct gfs2_glock_operations *glops,
1091                       unsigned int state, u16 flags, struct gfs2_holder *gh)
1092 {
1093         struct gfs2_glock *gl;
1094         int error;
1095
1096         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1097         if (!error) {
1098                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1099                 gfs2_glock_put(gl);
1100         }
1101
1102         return error;
1103 }
1104
1105 /**
1106  * glock_compare - Compare two struct gfs2_glock structures for sorting
1107  * @arg_a: the first structure
1108  * @arg_b: the second structure
1109  *
1110  */
1111
1112 static int glock_compare(const void *arg_a, const void *arg_b)
1113 {
1114         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1115         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1116         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1117         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1118
1119         if (a->ln_number > b->ln_number)
1120                 return 1;
1121         if (a->ln_number < b->ln_number)
1122                 return -1;
1123         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1124         return 0;
1125 }
1126
1127 /**
1128  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1129  * @num_gh: the number of structures
1130  * @ghs: an array of struct gfs2_holder structures
1131  *
1132  * Returns: 0 on success (all glocks acquired),
1133  *          errno on failure (no glocks acquired)
1134  */
1135
1136 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1137                      struct gfs2_holder **p)
1138 {
1139         unsigned int x;
1140         int error = 0;
1141
1142         for (x = 0; x < num_gh; x++)
1143                 p[x] = &ghs[x];
1144
1145         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1146
1147         for (x = 0; x < num_gh; x++) {
1148                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1149
1150                 error = gfs2_glock_nq(p[x]);
1151                 if (error) {
1152                         while (x--)
1153                                 gfs2_glock_dq(p[x]);
1154                         break;
1155                 }
1156         }
1157
1158         return error;
1159 }
1160
1161 /**
1162  * gfs2_glock_nq_m - acquire multiple glocks
1163  * @num_gh: the number of structures
1164  * @ghs: an array of struct gfs2_holder structures
1165  *
1166  *
1167  * Returns: 0 on success (all glocks acquired),
1168  *          errno on failure (no glocks acquired)
1169  */
1170
1171 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1172 {
1173         struct gfs2_holder *tmp[4];
1174         struct gfs2_holder **pph = tmp;
1175         int error = 0;
1176
1177         switch(num_gh) {
1178         case 0:
1179                 return 0;
1180         case 1:
1181                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1182                 return gfs2_glock_nq(ghs);
1183         default:
1184                 if (num_gh <= 4)
1185                         break;
1186                 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1187                 if (!pph)
1188                         return -ENOMEM;
1189         }
1190
1191         error = nq_m_sync(num_gh, ghs, pph);
1192
1193         if (pph != tmp)
1194                 kfree(pph);
1195
1196         return error;
1197 }
1198
1199 /**
1200  * gfs2_glock_dq_m - release multiple glocks
1201  * @num_gh: the number of structures
1202  * @ghs: an array of struct gfs2_holder structures
1203  *
1204  */
1205
1206 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1207 {
1208         while (num_gh--)
1209                 gfs2_glock_dq(&ghs[num_gh]);
1210 }
1211
1212 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1213 {
1214         unsigned long delay = 0;
1215         unsigned long holdtime;
1216         unsigned long now = jiffies;
1217
1218         gfs2_glock_hold(gl);
1219         holdtime = gl->gl_tchange + gl->gl_hold_time;
1220         if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1221             gl->gl_name.ln_type == LM_TYPE_INODE) {
1222                 if (time_before(now, holdtime))
1223                         delay = holdtime - now;
1224                 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1225                         delay = gl->gl_hold_time;
1226         }
1227
1228         spin_lock(&gl->gl_lockref.lock);
1229         handle_callback(gl, state, delay, true);
1230         spin_unlock(&gl->gl_lockref.lock);
1231         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1232                 gfs2_glock_put(gl);
1233 }
1234
1235 /**
1236  * gfs2_should_freeze - Figure out if glock should be frozen
1237  * @gl: The glock in question
1238  *
1239  * Glocks are not frozen if (a) the result of the dlm operation is
1240  * an error, (b) the locking operation was an unlock operation or
1241  * (c) if there is a "noexp" flagged request anywhere in the queue
1242  *
1243  * Returns: 1 if freezing should occur, 0 otherwise
1244  */
1245
1246 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1247 {
1248         const struct gfs2_holder *gh;
1249
1250         if (gl->gl_reply & ~LM_OUT_ST_MASK)
1251                 return 0;
1252         if (gl->gl_target == LM_ST_UNLOCKED)
1253                 return 0;
1254
1255         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1256                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1257                         continue;
1258                 if (LM_FLAG_NOEXP & gh->gh_flags)
1259                         return 0;
1260         }
1261
1262         return 1;
1263 }
1264
1265 /**
1266  * gfs2_glock_complete - Callback used by locking
1267  * @gl: Pointer to the glock
1268  * @ret: The return value from the dlm
1269  *
1270  * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1271  * to use a bitfield shared with other glock state fields.
1272  */
1273
1274 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1275 {
1276         struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1277
1278         spin_lock(&gl->gl_lockref.lock);
1279         gl->gl_reply = ret;
1280
1281         if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1282                 if (gfs2_should_freeze(gl)) {
1283                         set_bit(GLF_FROZEN, &gl->gl_flags);
1284                         spin_unlock(&gl->gl_lockref.lock);
1285                         return;
1286                 }
1287         }
1288
1289         gl->gl_lockref.count++;
1290         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1291         spin_unlock(&gl->gl_lockref.lock);
1292
1293         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1294                 gfs2_glock_put(gl);
1295 }
1296
1297 static int glock_cmp(void *priv, struct list_head *a, struct list_head *b)
1298 {
1299         struct gfs2_glock *gla, *glb;
1300
1301         gla = list_entry(a, struct gfs2_glock, gl_lru);
1302         glb = list_entry(b, struct gfs2_glock, gl_lru);
1303
1304         if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1305                 return 1;
1306         if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1307                 return -1;
1308
1309         return 0;
1310 }
1311
1312 /**
1313  * gfs2_dispose_glock_lru - Demote a list of glocks
1314  * @list: The list to dispose of
1315  *
1316  * Disposing of glocks may involve disk accesses, so that here we sort
1317  * the glocks by number (i.e. disk location of the inodes) so that if
1318  * there are any such accesses, they'll be sent in order (mostly).
1319  *
1320  * Must be called under the lru_lock, but may drop and retake this
1321  * lock. While the lru_lock is dropped, entries may vanish from the
1322  * list, but no new entries will appear on the list (since it is
1323  * private)
1324  */
1325
1326 static void gfs2_dispose_glock_lru(struct list_head *list)
1327 __releases(&lru_lock)
1328 __acquires(&lru_lock)
1329 {
1330         struct gfs2_glock *gl;
1331
1332         list_sort(NULL, list, glock_cmp);
1333
1334         while(!list_empty(list)) {
1335                 gl = list_entry(list->next, struct gfs2_glock, gl_lru);
1336                 list_del_init(&gl->gl_lru);
1337                 if (!spin_trylock(&gl->gl_lockref.lock)) {
1338 add_back_to_lru:
1339                         list_add(&gl->gl_lru, &lru_list);
1340                         atomic_inc(&lru_count);
1341                         continue;
1342                 }
1343                 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1344                         spin_unlock(&gl->gl_lockref.lock);
1345                         goto add_back_to_lru;
1346                 }
1347                 clear_bit(GLF_LRU, &gl->gl_flags);
1348                 gl->gl_lockref.count++;
1349                 if (demote_ok(gl))
1350                         handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1351                 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1352                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1353                         gl->gl_lockref.count--;
1354                 spin_unlock(&gl->gl_lockref.lock);
1355                 cond_resched_lock(&lru_lock);
1356         }
1357 }
1358
1359 /**
1360  * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1361  * @nr: The number of entries to scan
1362  *
1363  * This function selects the entries on the LRU which are able to
1364  * be demoted, and then kicks off the process by calling
1365  * gfs2_dispose_glock_lru() above.
1366  */
1367
1368 static long gfs2_scan_glock_lru(int nr)
1369 {
1370         struct gfs2_glock *gl;
1371         LIST_HEAD(skipped);
1372         LIST_HEAD(dispose);
1373         long freed = 0;
1374
1375         spin_lock(&lru_lock);
1376         while ((nr-- >= 0) && !list_empty(&lru_list)) {
1377                 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1378
1379                 /* Test for being demotable */
1380                 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1381                         list_move(&gl->gl_lru, &dispose);
1382                         atomic_dec(&lru_count);
1383                         freed++;
1384                         continue;
1385                 }
1386
1387                 list_move(&gl->gl_lru, &skipped);
1388         }
1389         list_splice(&skipped, &lru_list);
1390         if (!list_empty(&dispose))
1391                 gfs2_dispose_glock_lru(&dispose);
1392         spin_unlock(&lru_lock);
1393
1394         return freed;
1395 }
1396
1397 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1398                                             struct shrink_control *sc)
1399 {
1400         if (!(sc->gfp_mask & __GFP_FS))
1401                 return SHRINK_STOP;
1402         return gfs2_scan_glock_lru(sc->nr_to_scan);
1403 }
1404
1405 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1406                                              struct shrink_control *sc)
1407 {
1408         return vfs_pressure_ratio(atomic_read(&lru_count));
1409 }
1410
1411 static struct shrinker glock_shrinker = {
1412         .seeks = DEFAULT_SEEKS,
1413         .count_objects = gfs2_glock_shrink_count,
1414         .scan_objects = gfs2_glock_shrink_scan,
1415 };
1416
1417 /**
1418  * examine_bucket - Call a function for glock in a hash bucket
1419  * @examiner: the function
1420  * @sdp: the filesystem
1421  * @bucket: the bucket
1422  *
1423  */
1424
1425 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1426 {
1427         struct gfs2_glock *gl;
1428         struct rhash_head *pos;
1429         const struct bucket_table *tbl;
1430         int i;
1431
1432         rcu_read_lock();
1433         tbl = rht_dereference_rcu(gl_hash_table.tbl, &gl_hash_table);
1434         for (i = 0; i < tbl->size; i++) {
1435                 rht_for_each_entry_rcu(gl, pos, tbl, i, gl_node) {
1436                         if ((gl->gl_name.ln_sbd == sdp) &&
1437                             lockref_get_not_dead(&gl->gl_lockref))
1438                                 examiner(gl);
1439                 }
1440         }
1441         rcu_read_unlock();
1442         cond_resched();
1443 }
1444
1445 /**
1446  * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1447  * @gl: The glock to thaw
1448  *
1449  */
1450
1451 static void thaw_glock(struct gfs2_glock *gl)
1452 {
1453         if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1454                 goto out;
1455         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1456         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0) {
1457 out:
1458                 gfs2_glock_put(gl);
1459         }
1460 }
1461
1462 /**
1463  * clear_glock - look at a glock and see if we can free it from glock cache
1464  * @gl: the glock to look at
1465  *
1466  */
1467
1468 static void clear_glock(struct gfs2_glock *gl)
1469 {
1470         gfs2_glock_remove_from_lru(gl);
1471
1472         spin_lock(&gl->gl_lockref.lock);
1473         if (gl->gl_state != LM_ST_UNLOCKED)
1474                 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1475         spin_unlock(&gl->gl_lockref.lock);
1476         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1477                 gfs2_glock_put(gl);
1478 }
1479
1480 /**
1481  * gfs2_glock_thaw - Thaw any frozen glocks
1482  * @sdp: The super block
1483  *
1484  */
1485
1486 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1487 {
1488         glock_hash_walk(thaw_glock, sdp);
1489 }
1490
1491 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1492 {
1493         spin_lock(&gl->gl_lockref.lock);
1494         gfs2_dump_glock(seq, gl);
1495         spin_unlock(&gl->gl_lockref.lock);
1496 }
1497
1498 static void dump_glock_func(struct gfs2_glock *gl)
1499 {
1500         dump_glock(NULL, gl);
1501 }
1502
1503 /**
1504  * gfs2_gl_hash_clear - Empty out the glock hash table
1505  * @sdp: the filesystem
1506  * @wait: wait until it's all gone
1507  *
1508  * Called when unmounting the filesystem.
1509  */
1510
1511 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1512 {
1513         set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1514         flush_workqueue(glock_workqueue);
1515         glock_hash_walk(clear_glock, sdp);
1516         flush_workqueue(glock_workqueue);
1517         wait_event_timeout(sdp->sd_glock_wait,
1518                            atomic_read(&sdp->sd_glock_disposal) == 0,
1519                            HZ * 600);
1520         glock_hash_walk(dump_glock_func, sdp);
1521 }
1522
1523 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1524 {
1525         struct gfs2_glock *gl = ip->i_gl;
1526         int ret;
1527
1528         ret = gfs2_truncatei_resume(ip);
1529         gfs2_assert_withdraw(gl->gl_name.ln_sbd, ret == 0);
1530
1531         spin_lock(&gl->gl_lockref.lock);
1532         clear_bit(GLF_LOCK, &gl->gl_flags);
1533         run_queue(gl, 1);
1534         spin_unlock(&gl->gl_lockref.lock);
1535 }
1536
1537 static const char *state2str(unsigned state)
1538 {
1539         switch(state) {
1540         case LM_ST_UNLOCKED:
1541                 return "UN";
1542         case LM_ST_SHARED:
1543                 return "SH";
1544         case LM_ST_DEFERRED:
1545                 return "DF";
1546         case LM_ST_EXCLUSIVE:
1547                 return "EX";
1548         }
1549         return "??";
1550 }
1551
1552 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
1553 {
1554         char *p = buf;
1555         if (flags & LM_FLAG_TRY)
1556                 *p++ = 't';
1557         if (flags & LM_FLAG_TRY_1CB)
1558                 *p++ = 'T';
1559         if (flags & LM_FLAG_NOEXP)
1560                 *p++ = 'e';
1561         if (flags & LM_FLAG_ANY)
1562                 *p++ = 'A';
1563         if (flags & LM_FLAG_PRIORITY)
1564                 *p++ = 'p';
1565         if (flags & GL_ASYNC)
1566                 *p++ = 'a';
1567         if (flags & GL_EXACT)
1568                 *p++ = 'E';
1569         if (flags & GL_NOCACHE)
1570                 *p++ = 'c';
1571         if (test_bit(HIF_HOLDER, &iflags))
1572                 *p++ = 'H';
1573         if (test_bit(HIF_WAIT, &iflags))
1574                 *p++ = 'W';
1575         if (test_bit(HIF_FIRST, &iflags))
1576                 *p++ = 'F';
1577         *p = 0;
1578         return buf;
1579 }
1580
1581 /**
1582  * dump_holder - print information about a glock holder
1583  * @seq: the seq_file struct
1584  * @gh: the glock holder
1585  *
1586  */
1587
1588 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1589 {
1590         struct task_struct *gh_owner = NULL;
1591         char flags_buf[32];
1592
1593         rcu_read_lock();
1594         if (gh->gh_owner_pid)
1595                 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1596         gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1597                        state2str(gh->gh_state),
1598                        hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1599                        gh->gh_error,
1600                        gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1601                        gh_owner ? gh_owner->comm : "(ended)",
1602                        (void *)gh->gh_ip);
1603         rcu_read_unlock();
1604 }
1605
1606 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1607 {
1608         const unsigned long *gflags = &gl->gl_flags;
1609         char *p = buf;
1610
1611         if (test_bit(GLF_LOCK, gflags))
1612                 *p++ = 'l';
1613         if (test_bit(GLF_DEMOTE, gflags))
1614                 *p++ = 'D';
1615         if (test_bit(GLF_PENDING_DEMOTE, gflags))
1616                 *p++ = 'd';
1617         if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1618                 *p++ = 'p';
1619         if (test_bit(GLF_DIRTY, gflags))
1620                 *p++ = 'y';
1621         if (test_bit(GLF_LFLUSH, gflags))
1622                 *p++ = 'f';
1623         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1624                 *p++ = 'i';
1625         if (test_bit(GLF_REPLY_PENDING, gflags))
1626                 *p++ = 'r';
1627         if (test_bit(GLF_INITIAL, gflags))
1628                 *p++ = 'I';
1629         if (test_bit(GLF_FROZEN, gflags))
1630                 *p++ = 'F';
1631         if (test_bit(GLF_QUEUED, gflags))
1632                 *p++ = 'q';
1633         if (test_bit(GLF_LRU, gflags))
1634                 *p++ = 'L';
1635         if (gl->gl_object)
1636                 *p++ = 'o';
1637         if (test_bit(GLF_BLOCKING, gflags))
1638                 *p++ = 'b';
1639         *p = 0;
1640         return buf;
1641 }
1642
1643 /**
1644  * gfs2_dump_glock - print information about a glock
1645  * @seq: The seq_file struct
1646  * @gl: the glock
1647  *
1648  * The file format is as follows:
1649  * One line per object, capital letters are used to indicate objects
1650  * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1651  * other objects are indented by a single space and follow the glock to
1652  * which they are related. Fields are indicated by lower case letters
1653  * followed by a colon and the field value, except for strings which are in
1654  * [] so that its possible to see if they are composed of spaces for
1655  * example. The field's are n = number (id of the object), f = flags,
1656  * t = type, s = state, r = refcount, e = error, p = pid.
1657  *
1658  */
1659
1660 void gfs2_dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1661 {
1662         const struct gfs2_glock_operations *glops = gl->gl_ops;
1663         unsigned long long dtime;
1664         const struct gfs2_holder *gh;
1665         char gflags_buf[32];
1666
1667         dtime = jiffies - gl->gl_demote_time;
1668         dtime *= 1000000/HZ; /* demote time in uSec */
1669         if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1670                 dtime = 0;
1671         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",
1672                   state2str(gl->gl_state),
1673                   gl->gl_name.ln_type,
1674                   (unsigned long long)gl->gl_name.ln_number,
1675                   gflags2str(gflags_buf, gl),
1676                   state2str(gl->gl_target),
1677                   state2str(gl->gl_demote_state), dtime,
1678                   atomic_read(&gl->gl_ail_count),
1679                   atomic_read(&gl->gl_revokes),
1680                   (int)gl->gl_lockref.count, gl->gl_hold_time);
1681
1682         list_for_each_entry(gh, &gl->gl_holders, gh_list)
1683                 dump_holder(seq, gh);
1684
1685         if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1686                 glops->go_dump(seq, gl);
1687 }
1688
1689 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
1690 {
1691         struct gfs2_glock *gl = iter_ptr;
1692
1693         seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
1694                    gl->gl_name.ln_type,
1695                    (unsigned long long)gl->gl_name.ln_number,
1696                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
1697                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
1698                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
1699                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
1700                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
1701                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
1702                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
1703                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
1704         return 0;
1705 }
1706
1707 static const char *gfs2_gltype[] = {
1708         "type",
1709         "reserved",
1710         "nondisk",
1711         "inode",
1712         "rgrp",
1713         "meta",
1714         "iopen",
1715         "flock",
1716         "plock",
1717         "quota",
1718         "journal",
1719 };
1720
1721 static const char *gfs2_stype[] = {
1722         [GFS2_LKS_SRTT]         = "srtt",
1723         [GFS2_LKS_SRTTVAR]      = "srttvar",
1724         [GFS2_LKS_SRTTB]        = "srttb",
1725         [GFS2_LKS_SRTTVARB]     = "srttvarb",
1726         [GFS2_LKS_SIRT]         = "sirt",
1727         [GFS2_LKS_SIRTVAR]      = "sirtvar",
1728         [GFS2_LKS_DCOUNT]       = "dlm",
1729         [GFS2_LKS_QCOUNT]       = "queue",
1730 };
1731
1732 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1733
1734 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
1735 {
1736         struct gfs2_sbd *sdp = seq->private;
1737         loff_t pos = *(loff_t *)iter_ptr;
1738         unsigned index = pos >> 3;
1739         unsigned subindex = pos & 0x07;
1740         int i;
1741
1742         if (index == 0 && subindex != 0)
1743                 return 0;
1744
1745         seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
1746                    (index == 0) ? "cpu": gfs2_stype[subindex]);
1747
1748         for_each_possible_cpu(i) {
1749                 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
1750
1751                 if (index == 0)
1752                         seq_printf(seq, " %15u", i);
1753                 else
1754                         seq_printf(seq, " %15llu", (unsigned long long)lkstats->
1755                                    lkstats[index - 1].stats[subindex]);
1756         }
1757         seq_putc(seq, '\n');
1758         return 0;
1759 }
1760
1761 int __init gfs2_glock_init(void)
1762 {
1763         int ret;
1764
1765         ret = rhashtable_init(&gl_hash_table, &ht_parms);
1766         if (ret < 0)
1767                 return ret;
1768
1769         glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1770                                           WQ_HIGHPRI | WQ_FREEZABLE, 0);
1771         if (!glock_workqueue) {
1772                 rhashtable_destroy(&gl_hash_table);
1773                 return -ENOMEM;
1774         }
1775         gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1776                                                 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1777                                                 0);
1778         if (!gfs2_delete_workqueue) {
1779                 destroy_workqueue(glock_workqueue);
1780                 rhashtable_destroy(&gl_hash_table);
1781                 return -ENOMEM;
1782         }
1783
1784         ret = register_shrinker(&glock_shrinker);
1785         if (ret) {
1786                 destroy_workqueue(gfs2_delete_workqueue);
1787                 destroy_workqueue(glock_workqueue);
1788                 rhashtable_destroy(&gl_hash_table);
1789                 return ret;
1790         }
1791
1792         return 0;
1793 }
1794
1795 void gfs2_glock_exit(void)
1796 {
1797         unregister_shrinker(&glock_shrinker);
1798         rhashtable_destroy(&gl_hash_table);
1799         destroy_workqueue(glock_workqueue);
1800         destroy_workqueue(gfs2_delete_workqueue);
1801 }
1802
1803 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1804 {
1805         while ((gi->gl = rhashtable_walk_next(&gi->hti))) {
1806                 if (IS_ERR(gi->gl)) {
1807                         if (PTR_ERR(gi->gl) == -EAGAIN)
1808                                 continue;
1809                         gi->gl = NULL;
1810                         return;
1811                 }
1812                 /* Skip entries for other sb and dead entries */
1813                 if (gi->sdp == gi->gl->gl_name.ln_sbd &&
1814                     !__lockref_is_dead(&gi->gl->gl_lockref))
1815                         return;
1816         }
1817 }
1818
1819 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1820 {
1821         struct gfs2_glock_iter *gi = seq->private;
1822         loff_t n = *pos;
1823         int ret;
1824
1825         if (gi->last_pos <= *pos)
1826                 n = (*pos - gi->last_pos);
1827
1828         ret = rhashtable_walk_start(&gi->hti);
1829         if (ret)
1830                 return NULL;
1831
1832         do {
1833                 gfs2_glock_iter_next(gi);
1834         } while (gi->gl && n--);
1835
1836         gi->last_pos = *pos;
1837         return gi->gl;
1838 }
1839
1840 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1841                                  loff_t *pos)
1842 {
1843         struct gfs2_glock_iter *gi = seq->private;
1844
1845         (*pos)++;
1846         gi->last_pos = *pos;
1847         gfs2_glock_iter_next(gi);
1848         return gi->gl;
1849 }
1850
1851 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1852 {
1853         struct gfs2_glock_iter *gi = seq->private;
1854
1855         gi->gl = NULL;
1856         rhashtable_walk_stop(&gi->hti);
1857 }
1858
1859 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1860 {
1861         dump_glock(seq, iter_ptr);
1862         return 0;
1863 }
1864
1865 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
1866 {
1867         preempt_disable();
1868         if (*pos >= GFS2_NR_SBSTATS)
1869                 return NULL;
1870         return pos;
1871 }
1872
1873 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
1874                                    loff_t *pos)
1875 {
1876         (*pos)++;
1877         if (*pos >= GFS2_NR_SBSTATS)
1878                 return NULL;
1879         return pos;
1880 }
1881
1882 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
1883 {
1884         preempt_enable();
1885 }
1886
1887 static const struct seq_operations gfs2_glock_seq_ops = {
1888         .start = gfs2_glock_seq_start,
1889         .next  = gfs2_glock_seq_next,
1890         .stop  = gfs2_glock_seq_stop,
1891         .show  = gfs2_glock_seq_show,
1892 };
1893
1894 static const struct seq_operations gfs2_glstats_seq_ops = {
1895         .start = gfs2_glock_seq_start,
1896         .next  = gfs2_glock_seq_next,
1897         .stop  = gfs2_glock_seq_stop,
1898         .show  = gfs2_glstats_seq_show,
1899 };
1900
1901 static const struct seq_operations gfs2_sbstats_seq_ops = {
1902         .start = gfs2_sbstats_seq_start,
1903         .next  = gfs2_sbstats_seq_next,
1904         .stop  = gfs2_sbstats_seq_stop,
1905         .show  = gfs2_sbstats_seq_show,
1906 };
1907
1908 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
1909
1910 static int gfs2_glocks_open(struct inode *inode, struct file *file)
1911 {
1912         int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1913                                    sizeof(struct gfs2_glock_iter));
1914         if (ret == 0) {
1915                 struct seq_file *seq = file->private_data;
1916                 struct gfs2_glock_iter *gi = seq->private;
1917
1918                 gi->sdp = inode->i_private;
1919                 gi->last_pos = 0;
1920                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
1921                 if (seq->buf)
1922                         seq->size = GFS2_SEQ_GOODSIZE;
1923                 gi->gl = NULL;
1924                 ret = rhashtable_walk_init(&gl_hash_table, &gi->hti, GFP_KERNEL);
1925         }
1926         return ret;
1927 }
1928
1929 static int gfs2_glocks_release(struct inode *inode, struct file *file)
1930 {
1931         struct seq_file *seq = file->private_data;
1932         struct gfs2_glock_iter *gi = seq->private;
1933
1934         gi->gl = NULL;
1935         rhashtable_walk_exit(&gi->hti);
1936         return seq_release_private(inode, file);
1937 }
1938
1939 static int gfs2_glstats_open(struct inode *inode, struct file *file)
1940 {
1941         int ret = seq_open_private(file, &gfs2_glstats_seq_ops,
1942                                    sizeof(struct gfs2_glock_iter));
1943         if (ret == 0) {
1944                 struct seq_file *seq = file->private_data;
1945                 struct gfs2_glock_iter *gi = seq->private;
1946                 gi->sdp = inode->i_private;
1947                 gi->last_pos = 0;
1948                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
1949                 if (seq->buf)
1950                         seq->size = GFS2_SEQ_GOODSIZE;
1951                 gi->gl = NULL;
1952                 ret = rhashtable_walk_init(&gl_hash_table, &gi->hti, GFP_KERNEL);
1953         }
1954         return ret;
1955 }
1956
1957 static int gfs2_sbstats_open(struct inode *inode, struct file *file)
1958 {
1959         int ret = seq_open(file, &gfs2_sbstats_seq_ops);
1960         if (ret == 0) {
1961                 struct seq_file *seq = file->private_data;
1962                 seq->private = inode->i_private;  /* sdp */
1963         }
1964         return ret;
1965 }
1966
1967 static const struct file_operations gfs2_glocks_fops = {
1968         .owner   = THIS_MODULE,
1969         .open    = gfs2_glocks_open,
1970         .read    = seq_read,
1971         .llseek  = seq_lseek,
1972         .release = gfs2_glocks_release,
1973 };
1974
1975 static const struct file_operations gfs2_glstats_fops = {
1976         .owner   = THIS_MODULE,
1977         .open    = gfs2_glstats_open,
1978         .read    = seq_read,
1979         .llseek  = seq_lseek,
1980         .release = gfs2_glocks_release,
1981 };
1982
1983 static const struct file_operations gfs2_sbstats_fops = {
1984         .owner   = THIS_MODULE,
1985         .open    = gfs2_sbstats_open,
1986         .read    = seq_read,
1987         .llseek  = seq_lseek,
1988         .release = seq_release,
1989 };
1990
1991 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
1992 {
1993         struct dentry *dent;
1994
1995         dent = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
1996         if (IS_ERR_OR_NULL(dent))
1997                 goto fail;
1998         sdp->debugfs_dir = dent;
1999
2000         dent = debugfs_create_file("glocks",
2001                                    S_IFREG | S_IRUGO,
2002                                    sdp->debugfs_dir, sdp,
2003                                    &gfs2_glocks_fops);
2004         if (IS_ERR_OR_NULL(dent))
2005                 goto fail;
2006         sdp->debugfs_dentry_glocks = dent;
2007
2008         dent = debugfs_create_file("glstats",
2009                                    S_IFREG | S_IRUGO,
2010                                    sdp->debugfs_dir, sdp,
2011                                    &gfs2_glstats_fops);
2012         if (IS_ERR_OR_NULL(dent))
2013                 goto fail;
2014         sdp->debugfs_dentry_glstats = dent;
2015
2016         dent = debugfs_create_file("sbstats",
2017                                    S_IFREG | S_IRUGO,
2018                                    sdp->debugfs_dir, sdp,
2019                                    &gfs2_sbstats_fops);
2020         if (IS_ERR_OR_NULL(dent))
2021                 goto fail;
2022         sdp->debugfs_dentry_sbstats = dent;
2023
2024         return 0;
2025 fail:
2026         gfs2_delete_debugfs_file(sdp);
2027         return dent ? PTR_ERR(dent) : -ENOMEM;
2028 }
2029
2030 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2031 {
2032         if (sdp->debugfs_dir) {
2033                 if (sdp->debugfs_dentry_glocks) {
2034                         debugfs_remove(sdp->debugfs_dentry_glocks);
2035                         sdp->debugfs_dentry_glocks = NULL;
2036                 }
2037                 if (sdp->debugfs_dentry_glstats) {
2038                         debugfs_remove(sdp->debugfs_dentry_glstats);
2039                         sdp->debugfs_dentry_glstats = NULL;
2040                 }
2041                 if (sdp->debugfs_dentry_sbstats) {
2042                         debugfs_remove(sdp->debugfs_dentry_sbstats);
2043                         sdp->debugfs_dentry_sbstats = NULL;
2044                 }
2045                 debugfs_remove(sdp->debugfs_dir);
2046                 sdp->debugfs_dir = NULL;
2047         }
2048 }
2049
2050 int gfs2_register_debugfs(void)
2051 {
2052         gfs2_root = debugfs_create_dir("gfs2", NULL);
2053         if (IS_ERR(gfs2_root))
2054                 return PTR_ERR(gfs2_root);
2055         return gfs2_root ? 0 : -ENOMEM;
2056 }
2057
2058 void gfs2_unregister_debugfs(void)
2059 {
2060         debugfs_remove(gfs2_root);
2061         gfs2_root = NULL;
2062 }