fs/gfs2/glock.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
   4  *
   5  * This copyrighted material is made available to anyone wishing to use,
   6  * modify, copy, or redistribute it subject to the terms and conditions
   7  * of the GNU General Public License version 2.
   8  */
   9
  10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  11
  12 #include <linux/sched.h>
  13 #include <linux/slab.h>
  14 #include <linux/spinlock.h>
  15 #include <linux/buffer_head.h>
  16 #include <linux/delay.h>
  17 #include <linux/sort.h>
  18 #include <linux/jhash.h>
  19 #include <linux/kallsyms.h>
  20 #include <linux/gfs2_ondisk.h>
  21 #include <linux/list.h>
  22 #include <linux/wait.h>
  23 #include <linux/module.h>
  24 #include <asm/uaccess.h>
  25 #include <linux/seq_file.h>
  26 #include <linux/debugfs.h>
  27 #include <linux/kthread.h>
  28 #include <linux/freezer.h>
  29 #include <linux/workqueue.h>
  30 #include <linux/jiffies.h>
  31 #include <linux/rcupdate.h>
  32 #include <linux/rculist_bl.h>
  33 #include <linux/bit_spinlock.h>
  34 #include <linux/percpu.h>
  35 #include <linux/list_sort.h>
  36 #include <linux/lockref.h>
  37 #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       (1 << 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 gfs2_inode *ip;
 579         struct inode *inode;
 580         u64 no_addr = gl->gl_name.ln_number;
 581
 582         /* If someone's using this glock to create a new dinode, the block must
 583            have been freed by another node, then re-used, in which case our
 584            iopen callback is too late after the fact. Ignore it. */
 585         if (test_bit(GLF_INODE_CREATING, &gl->gl_flags))
 586                 goto out;
 587
 588         ip = gl->gl_object;
 589         /* Note: Unsafe to dereference ip as we don't hold right refs/locks */
 590
 591         if (ip)
 592                 inode = gfs2_ilookup(sdp->sd_vfs, no_addr);
 593         else
 594                 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
 595         if (inode && !IS_ERR(inode)) {
 596                 d_prune_aliases(inode);
 597                 iput(inode);
 598         }
 599 out:
 600         gfs2_glock_put(gl);
 601 }
 602
 603 static void glock_work_func(struct work_struct *work)
 604 {
 605         unsigned long delay = 0;
 606         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
 607         int drop_ref = 0;
 608
 609         if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
 610                 finish_xmote(gl, gl->gl_reply);
 611                 drop_ref = 1;
 612         }
 613         spin_lock(&gl->gl_lockref.lock);
 614         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
 615             gl->gl_state != LM_ST_UNLOCKED &&
 616             gl->gl_demote_state != LM_ST_EXCLUSIVE) {
 617                 unsigned long holdtime, now = jiffies;
 618
 619                 holdtime = gl->gl_tchange + gl->gl_hold_time;
 620                 if (time_before(now, holdtime))
 621                         delay = holdtime - now;
 622
 623                 if (!delay) {
 624                         clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
 625                         set_bit(GLF_DEMOTE, &gl->gl_flags);
 626                 }
 627         }
 628         run_queue(gl, 0);
 629         spin_unlock(&gl->gl_lockref.lock);
 630         if (!delay)
 631                 gfs2_glock_put(gl);
 632         else {
 633                 if (gl->gl_name.ln_type != LM_TYPE_INODE)
 634                         delay = 0;
 635                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
 636                         gfs2_glock_put(gl);
 637         }
 638         if (drop_ref)
 639                 gfs2_glock_put(gl);
 640 }
 641
 642 /**
 643  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
 644  * @sdp: The GFS2 superblock
 645  * @number: the lock number
 646  * @glops: The glock_operations to use
 647  * @create: If 0, don't create the glock if it doesn't exist
 648  * @glp: the glock is returned here
 649  *
 650  * This does not lock a glock, just finds/creates structures for one.
 651  *
 652  * Returns: errno
 653  */
 654
 655 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
 656                    const struct gfs2_glock_operations *glops, int create,
 657                    struct gfs2_glock **glp)
 658 {
 659         struct super_block *s = sdp->sd_vfs;
 660         struct lm_lockname name = { .ln_number = number,
 661                                     .ln_type = glops->go_type,
 662                                     .ln_sbd = sdp };
 663         struct gfs2_glock *gl, *tmp = NULL;
 664         struct address_space *mapping;
 665         struct kmem_cache *cachep;
 666         int ret, tries = 0;
 667
 668         gl = rhashtable_lookup_fast(&gl_hash_table, &name, ht_parms);
 669         if (gl && !lockref_get_not_dead(&gl->gl_lockref))
 670                 gl = NULL;
 671
 672         *glp = gl;
 673         if (gl)
 674                 return 0;
 675         if (!create)
 676                 return -ENOENT;
 677
 678         if (glops->go_flags & GLOF_ASPACE)
 679                 cachep = gfs2_glock_aspace_cachep;
 680         else
 681                 cachep = gfs2_glock_cachep;
 682         gl = kmem_cache_alloc(cachep, GFP_NOFS);
 683         if (!gl)
 684                 return -ENOMEM;
 685
 686         memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
 687
 688         if (glops->go_flags & GLOF_LVB) {
 689                 gl->gl_lksb.sb_lvbptr = kzalloc(GFS2_MIN_LVB_SIZE, GFP_NOFS);
 690                 if (!gl->gl_lksb.sb_lvbptr) {
 691                         kmem_cache_free(cachep, gl);
 692                         return -ENOMEM;
 693                 }
 694         }
 695
 696         atomic_inc(&sdp->sd_glock_disposal);
 697         gl->gl_node.next = NULL;
 698         gl->gl_flags = 0;
 699         gl->gl_name = name;
 700         gl->gl_lockref.count = 1;
 701         gl->gl_state = LM_ST_UNLOCKED;
 702         gl->gl_target = LM_ST_UNLOCKED;
 703         gl->gl_demote_state = LM_ST_EXCLUSIVE;
 704         gl->gl_ops = glops;
 705         gl->gl_dstamp = ktime_set(0, 0);
 706         preempt_disable();
 707         /* We use the global stats to estimate the initial per-glock stats */
 708         gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
 709         preempt_enable();
 710         gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
 711         gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
 712         gl->gl_tchange = jiffies;
 713         gl->gl_object = NULL;
 714         gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
 715         INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
 716         INIT_WORK(&gl->gl_delete, delete_work_func);
 717
 718         mapping = gfs2_glock2aspace(gl);
 719         if (mapping) {
 720                 mapping->a_ops = &gfs2_meta_aops;
 721                 mapping->host = s->s_bdev->bd_inode;
 722                 mapping->flags = 0;
 723                 mapping_set_gfp_mask(mapping, GFP_NOFS);
 724                 mapping->private_data = NULL;
 725                 mapping->writeback_index = 0;
 726         }
 727
 728 again:
 729         ret = rhashtable_lookup_insert_fast(&gl_hash_table, &gl->gl_node,
 730                                             ht_parms);
 731         if (ret == 0) {
 732                 *glp = gl;
 733                 return 0;
 734         }
 735
 736         if (ret == -EEXIST) {
 737                 ret = 0;
 738                 tmp = rhashtable_lookup_fast(&gl_hash_table, &name, ht_parms);
 739                 if (tmp == NULL || !lockref_get_not_dead(&tmp->gl_lockref)) {
 740                         if (++tries < 100) {
 741                                 cond_resched();
 742                                 goto again;
 743                         }
 744                         tmp = NULL;
 745                         ret = -ENOMEM;
 746                 }
 747         } else {
 748                 WARN_ON_ONCE(ret);
 749         }
 750         kfree(gl->gl_lksb.sb_lvbptr);
 751         kmem_cache_free(cachep, gl);
 752         atomic_dec(&sdp->sd_glock_disposal);
 753         *glp = tmp;
 754
 755         return ret;
 756 }
 757
 758 /**
 759  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
 760  * @gl: the glock
 761  * @state: the state we're requesting
 762  * @flags: the modifier flags
 763  * @gh: the holder structure
 764  *
 765  */
 766
 767 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
 768                       struct gfs2_holder *gh)
 769 {
 770         INIT_LIST_HEAD(&gh->gh_list);
 771         gh->gh_gl = gl;
 772         gh->gh_ip = _RET_IP_;
 773         gh->gh_owner_pid = get_pid(task_pid(current));
 774         gh->gh_state = state;
 775         gh->gh_flags = flags;
 776         gh->gh_error = 0;
 777         gh->gh_iflags = 0;
 778         gfs2_glock_hold(gl);
 779 }
 780
 781 /**
 782  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
 783  * @state: the state we're requesting
 784  * @flags: the modifier flags
 785  * @gh: the holder structure
 786  *
 787  * Don't mess with the glock.
 788  *
 789  */
 790
 791 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
 792 {
 793         gh->gh_state = state;
 794         gh->gh_flags = flags;
 795         gh->gh_iflags = 0;
 796         gh->gh_ip = _RET_IP_;
 797         put_pid(gh->gh_owner_pid);
 798         gh->gh_owner_pid = get_pid(task_pid(current));
 799 }
 800
 801 /**
 802  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
 803  * @gh: the holder structure
 804  *
 805  */
 806
 807 void gfs2_holder_uninit(struct gfs2_holder *gh)
 808 {
 809         put_pid(gh->gh_owner_pid);
 810         gfs2_glock_put(gh->gh_gl);
 811         gh->gh_gl = NULL;
 812         gh->gh_ip = 0;
 813 }
 814
 815 /**
 816  * gfs2_glock_wait - wait on a glock acquisition
 817  * @gh: the glock holder
 818  *
 819  * Returns: 0 on success
 820  */
 821
 822 int gfs2_glock_wait(struct gfs2_holder *gh)
 823 {
 824         unsigned long time1 = jiffies;
 825
 826         might_sleep();
 827         wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
 828         if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
 829                 /* Lengthen the minimum hold time. */
 830                 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
 831                                               GL_GLOCK_HOLD_INCR,
 832                                               GL_GLOCK_MAX_HOLD);
 833         return gh->gh_error;
 834 }
 835
 836 /**
 837  * handle_callback - process a demote request
 838  * @gl: the glock
 839  * @state: the state the caller wants us to change to
 840  *
 841  * There are only two requests that we are going to see in actual
 842  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
 843  */
 844
 845 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
 846                             unsigned long delay, bool remote)
 847 {
 848         int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
 849
 850         set_bit(bit, &gl->gl_flags);
 851         if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
 852                 gl->gl_demote_state = state;
 853                 gl->gl_demote_time = jiffies;
 854         } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
 855                         gl->gl_demote_state != state) {
 856                 gl->gl_demote_state = LM_ST_UNLOCKED;
 857         }
 858         if (gl->gl_ops->go_callback)
 859                 gl->gl_ops->go_callback(gl, remote);
 860         trace_gfs2_demote_rq(gl, remote);
 861 }
 862
 863 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
 864 {
 865         struct va_format vaf;
 866         va_list args;
 867
 868         va_start(args, fmt);
 869
 870         if (seq) {
 871                 seq_vprintf(seq, fmt, args);
 872         } else {
 873                 vaf.fmt = fmt;
 874                 vaf.va = &args;
 875
 876                 pr_err("%pV", &vaf);
 877         }
 878
 879         va_end(args);
 880 }
 881
 882 /**
 883  * add_to_queue - Add a holder to the wait queue (but look for recursion)
 884  * @gh: the holder structure to add
 885  *
 886  * Eventually we should move the recursive locking trap to a
 887  * debugging option or something like that. This is the fast
 888  * path and needs to have the minimum number of distractions.
 889  *
 890  */
 891
 892 static inline void add_to_queue(struct gfs2_holder *gh)
 893 __releases(&gl->gl_lockref.lock)
 894 __acquires(&gl->gl_lockref.lock)
 895 {
 896         struct gfs2_glock *gl = gh->gh_gl;
 897         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 898         struct list_head *insert_pt = NULL;
 899         struct gfs2_holder *gh2;
 900         int try_futile = 0;
 901
 902         BUG_ON(gh->gh_owner_pid == NULL);
 903         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
 904                 BUG();
 905
 906         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
 907                 if (test_bit(GLF_LOCK, &gl->gl_flags))
 908                         try_futile = !may_grant(gl, gh);
 909                 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 910                         goto fail;
 911         }
 912
 913         list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
 914                 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
 915                     (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
 916                         goto trap_recursive;
 917                 if (try_futile &&
 918                     !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
 919 fail:
 920                         gh->gh_error = GLR_TRYFAILED;
 921                         gfs2_holder_wake(gh);
 922                         return;
 923                 }
 924                 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
 925                         continue;
 926                 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
 927                         insert_pt = &gh2->gh_list;
 928         }
 929         set_bit(GLF_QUEUED, &gl->gl_flags);
 930         trace_gfs2_glock_queue(gh, 1);
 931         gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
 932         gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
 933         if (likely(insert_pt == NULL)) {
 934                 list_add_tail(&gh->gh_list, &gl->gl_holders);
 935                 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
 936                         goto do_cancel;
 937                 return;
 938         }
 939         list_add_tail(&gh->gh_list, insert_pt);
 940 do_cancel:
 941         gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
 942         if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
 943                 spin_unlock(&gl->gl_lockref.lock);
 944                 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
 945                         sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
 946                 spin_lock(&gl->gl_lockref.lock);
 947         }
 948         return;
 949
 950 trap_recursive:
 951         pr_err("original: %pSR\n", (void *)gh2->gh_ip);
 952         pr_err("pid: %d\n", pid_nr(gh2->gh_owner_pid));
 953         pr_err("lock type: %d req lock state : %d\n",
 954                gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
 955         pr_err("new: %pSR\n", (void *)gh->gh_ip);
 956         pr_err("pid: %d\n", pid_nr(gh->gh_owner_pid));
 957         pr_err("lock type: %d req lock state : %d\n",
 958                gh->gh_gl->gl_name.ln_type, gh->gh_state);
 959         gfs2_dump_glock(NULL, gl);
 960         BUG();
 961 }
 962
 963 /**
 964  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
 965  * @gh: the holder structure
 966  *
 967  * if (gh->gh_flags & GL_ASYNC), this never returns an error
 968  *
 969  * Returns: 0, GLR_TRYFAILED, or errno on failure
 970  */
 971
 972 int gfs2_glock_nq(struct gfs2_holder *gh)
 973 {
 974         struct gfs2_glock *gl = gh->gh_gl;
 975         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 976         int error = 0;
 977
 978         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 979                 return -EIO;
 980
 981         if (test_bit(GLF_LRU, &gl->gl_flags))
 982                 gfs2_glock_remove_from_lru(gl);
 983
 984         spin_lock(&gl->gl_lockref.lock);
 985         add_to_queue(gh);
 986         if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
 987                      test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
 988                 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
 989                 gl->gl_lockref.count++;
 990                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 991                         gl->gl_lockref.count--;
 992         }
 993         run_queue(gl, 1);
 994         spin_unlock(&gl->gl_lockref.lock);
 995
 996         if (!(gh->gh_flags & GL_ASYNC))
 997                 error = gfs2_glock_wait(gh);
 998
 999         return error;
1000 }
1001
1002 /**
1003  * gfs2_glock_poll - poll to see if an async request has been completed
1004  * @gh: the holder
1005  *
1006  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1007  */
1008
1009 int gfs2_glock_poll(struct gfs2_holder *gh)
1010 {
1011         return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1012 }
1013
1014 /**
1015  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1016  * @gh: the glock holder
1017  *
1018  */
1019
1020 void gfs2_glock_dq(struct gfs2_holder *gh)
1021 {
1022         struct gfs2_glock *gl = gh->gh_gl;
1023         const struct gfs2_glock_operations *glops = gl->gl_ops;
1024         unsigned delay = 0;
1025         int fast_path = 0;
1026
1027         spin_lock(&gl->gl_lockref.lock);
1028         if (gh->gh_flags & GL_NOCACHE)
1029                 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1030
1031         list_del_init(&gh->gh_list);
1032         clear_bit(HIF_HOLDER, &gh->gh_iflags);
1033         if (find_first_holder(gl) == NULL) {
1034                 if (glops->go_unlock) {
1035                         GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1036                         spin_unlock(&gl->gl_lockref.lock);
1037                         glops->go_unlock(gh);
1038                         spin_lock(&gl->gl_lockref.lock);
1039                         clear_bit(GLF_LOCK, &gl->gl_flags);
1040                 }
1041                 if (list_empty(&gl->gl_holders) &&
1042                     !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1043                     !test_bit(GLF_DEMOTE, &gl->gl_flags))
1044                         fast_path = 1;
1045         }
1046         if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl) &&
1047             (glops->go_flags & GLOF_LRU))
1048                 gfs2_glock_add_to_lru(gl);
1049
1050         trace_gfs2_glock_queue(gh, 0);
1051         spin_unlock(&gl->gl_lockref.lock);
1052         if (likely(fast_path))
1053                 return;
1054
1055         gfs2_glock_hold(gl);
1056         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1057             !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1058             gl->gl_name.ln_type == LM_TYPE_INODE)
1059                 delay = gl->gl_hold_time;
1060         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1061                 gfs2_glock_put(gl);
1062 }
1063
1064 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1065 {
1066         struct gfs2_glock *gl = gh->gh_gl;
1067         gfs2_glock_dq(gh);
1068         might_sleep();
1069         wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1070 }
1071
1072 /**
1073  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1074  * @gh: the holder structure
1075  *
1076  */
1077
1078 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1079 {
1080         gfs2_glock_dq(gh);
1081         gfs2_holder_uninit(gh);
1082 }
1083
1084 /**
1085  * gfs2_glock_nq_num - acquire a glock based on lock number
1086  * @sdp: the filesystem
1087  * @number: the lock number
1088  * @glops: the glock operations for the type of glock
1089  * @state: the state to acquire the glock in
1090  * @flags: modifier flags for the acquisition
1091  * @gh: the struct gfs2_holder
1092  *
1093  * Returns: errno
1094  */
1095
1096 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1097                       const struct gfs2_glock_operations *glops,
1098                       unsigned int state, u16 flags, struct gfs2_holder *gh)
1099 {
1100         struct gfs2_glock *gl;
1101         int error;
1102
1103         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1104         if (!error) {
1105                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1106                 gfs2_glock_put(gl);
1107         }
1108
1109         return error;
1110 }
1111
1112 /**
1113  * glock_compare - Compare two struct gfs2_glock structures for sorting
1114  * @arg_a: the first structure
1115  * @arg_b: the second structure
1116  *
1117  */
1118
1119 static int glock_compare(const void *arg_a, const void *arg_b)
1120 {
1121         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1122         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1123         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1124         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1125
1126         if (a->ln_number > b->ln_number)
1127                 return 1;
1128         if (a->ln_number < b->ln_number)
1129                 return -1;
1130         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1131         return 0;
1132 }
1133
1134 /**
1135  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1136  * @num_gh: the number of structures
1137  * @ghs: an array of struct gfs2_holder structures
1138  *
1139  * Returns: 0 on success (all glocks acquired),
1140  *          errno on failure (no glocks acquired)
1141  */
1142
1143 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1144                      struct gfs2_holder **p)
1145 {
1146         unsigned int x;
1147         int error = 0;
1148
1149         for (x = 0; x < num_gh; x++)
1150                 p[x] = &ghs[x];
1151
1152         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1153
1154         for (x = 0; x < num_gh; x++) {
1155                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1156
1157                 error = gfs2_glock_nq(p[x]);
1158                 if (error) {
1159                         while (x--)
1160                                 gfs2_glock_dq(p[x]);
1161                         break;
1162                 }
1163         }
1164
1165         return error;
1166 }
1167
1168 /**
1169  * gfs2_glock_nq_m - acquire multiple glocks
1170  * @num_gh: the number of structures
1171  * @ghs: an array of struct gfs2_holder structures
1172  *
1173  *
1174  * Returns: 0 on success (all glocks acquired),
1175  *          errno on failure (no glocks acquired)
1176  */
1177
1178 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1179 {
1180         struct gfs2_holder *tmp[4];
1181         struct gfs2_holder **pph = tmp;
1182         int error = 0;
1183
1184         switch(num_gh) {
1185         case 0:
1186                 return 0;
1187         case 1:
1188                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1189                 return gfs2_glock_nq(ghs);
1190         default:
1191                 if (num_gh <= 4)
1192                         break;
1193                 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1194                 if (!pph)
1195                         return -ENOMEM;
1196         }
1197
1198         error = nq_m_sync(num_gh, ghs, pph);
1199
1200         if (pph != tmp)
1201                 kfree(pph);
1202
1203         return error;
1204 }
1205
1206 /**
1207  * gfs2_glock_dq_m - release multiple glocks
1208  * @num_gh: the number of structures
1209  * @ghs: an array of struct gfs2_holder structures
1210  *
1211  */
1212
1213 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1214 {
1215         while (num_gh--)
1216                 gfs2_glock_dq(&ghs[num_gh]);
1217 }
1218
1219 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1220 {
1221         unsigned long delay = 0;
1222         unsigned long holdtime;
1223         unsigned long now = jiffies;
1224
1225         gfs2_glock_hold(gl);
1226         holdtime = gl->gl_tchange + gl->gl_hold_time;
1227         if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1228             gl->gl_name.ln_type == LM_TYPE_INODE) {
1229                 if (time_before(now, holdtime))
1230                         delay = holdtime - now;
1231                 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1232                         delay = gl->gl_hold_time;
1233         }
1234
1235         spin_lock(&gl->gl_lockref.lock);
1236         handle_callback(gl, state, delay, true);
1237         spin_unlock(&gl->gl_lockref.lock);
1238         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1239                 gfs2_glock_put(gl);
1240 }
1241
1242 /**
1243  * gfs2_should_freeze - Figure out if glock should be frozen
1244  * @gl: The glock in question
1245  *
1246  * Glocks are not frozen if (a) the result of the dlm operation is
1247  * an error, (b) the locking operation was an unlock operation or
1248  * (c) if there is a "noexp" flagged request anywhere in the queue
1249  *
1250  * Returns: 1 if freezing should occur, 0 otherwise
1251  */
1252
1253 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1254 {
1255         const struct gfs2_holder *gh;
1256
1257         if (gl->gl_reply & ~LM_OUT_ST_MASK)
1258                 return 0;
1259         if (gl->gl_target == LM_ST_UNLOCKED)
1260                 return 0;
1261
1262         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1263                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1264                         continue;
1265                 if (LM_FLAG_NOEXP & gh->gh_flags)
1266                         return 0;
1267         }
1268
1269         return 1;
1270 }
1271
1272 /**
1273  * gfs2_glock_complete - Callback used by locking
1274  * @gl: Pointer to the glock
1275  * @ret: The return value from the dlm
1276  *
1277  * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1278  * to use a bitfield shared with other glock state fields.
1279  */
1280
1281 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1282 {
1283         struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1284
1285         spin_lock(&gl->gl_lockref.lock);
1286         gl->gl_reply = ret;
1287
1288         if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1289                 if (gfs2_should_freeze(gl)) {
1290                         set_bit(GLF_FROZEN, &gl->gl_flags);
1291                         spin_unlock(&gl->gl_lockref.lock);
1292                         return;
1293                 }
1294         }
1295
1296         gl->gl_lockref.count++;
1297         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1298         spin_unlock(&gl->gl_lockref.lock);
1299
1300         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1301                 gfs2_glock_put(gl);
1302 }
1303
1304 static int glock_cmp(void *priv, struct list_head *a, struct list_head *b)
1305 {
1306         struct gfs2_glock *gla, *glb;
1307
1308         gla = list_entry(a, struct gfs2_glock, gl_lru);
1309         glb = list_entry(b, struct gfs2_glock, gl_lru);
1310
1311         if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1312                 return 1;
1313         if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1314                 return -1;
1315
1316         return 0;
1317 }
1318
1319 /**
1320  * gfs2_dispose_glock_lru - Demote a list of glocks
1321  * @list: The list to dispose of
1322  *
1323  * Disposing of glocks may involve disk accesses, so that here we sort
1324  * the glocks by number (i.e. disk location of the inodes) so that if
1325  * there are any such accesses, they'll be sent in order (mostly).
1326  *
1327  * Must be called under the lru_lock, but may drop and retake this
1328  * lock. While the lru_lock is dropped, entries may vanish from the
1329  * list, but no new entries will appear on the list (since it is
1330  * private)
1331  */
1332
1333 static void gfs2_dispose_glock_lru(struct list_head *list)
1334 __releases(&lru_lock)
1335 __acquires(&lru_lock)
1336 {
1337         struct gfs2_glock *gl;
1338
1339         list_sort(NULL, list, glock_cmp);
1340
1341         while(!list_empty(list)) {
1342                 gl = list_entry(list->next, struct gfs2_glock, gl_lru);
1343                 list_del_init(&gl->gl_lru);
1344                 if (!spin_trylock(&gl->gl_lockref.lock)) {
1345 add_back_to_lru:
1346                         list_add(&gl->gl_lru, &lru_list);
1347                         atomic_inc(&lru_count);
1348                         continue;
1349                 }
1350                 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1351                         spin_unlock(&gl->gl_lockref.lock);
1352                         goto add_back_to_lru;
1353                 }
1354                 clear_bit(GLF_LRU, &gl->gl_flags);
1355                 gl->gl_lockref.count++;
1356                 if (demote_ok(gl))
1357                         handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1358                 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1359                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1360                         gl->gl_lockref.count--;
1361                 spin_unlock(&gl->gl_lockref.lock);
1362                 cond_resched_lock(&lru_lock);
1363         }
1364 }
1365
1366 /**
1367  * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1368  * @nr: The number of entries to scan
1369  *
1370  * This function selects the entries on the LRU which are able to
1371  * be demoted, and then kicks off the process by calling
1372  * gfs2_dispose_glock_lru() above.
1373  */
1374
1375 static long gfs2_scan_glock_lru(int nr)
1376 {
1377         struct gfs2_glock *gl;
1378         LIST_HEAD(skipped);
1379         LIST_HEAD(dispose);
1380         long freed = 0;
1381
1382         spin_lock(&lru_lock);
1383         while ((nr-- >= 0) && !list_empty(&lru_list)) {
1384                 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1385
1386                 /* Test for being demotable */
1387                 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1388                         list_move(&gl->gl_lru, &dispose);
1389                         atomic_dec(&lru_count);
1390                         freed++;
1391                         continue;
1392                 }
1393
1394                 list_move(&gl->gl_lru, &skipped);
1395         }
1396         list_splice(&skipped, &lru_list);
1397         if (!list_empty(&dispose))
1398                 gfs2_dispose_glock_lru(&dispose);
1399         spin_unlock(&lru_lock);
1400
1401         return freed;
1402 }
1403
1404 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1405                                             struct shrink_control *sc)
1406 {
1407         if (!(sc->gfp_mask & __GFP_FS))
1408                 return SHRINK_STOP;
1409         return gfs2_scan_glock_lru(sc->nr_to_scan);
1410 }
1411
1412 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1413                                              struct shrink_control *sc)
1414 {
1415         return vfs_pressure_ratio(atomic_read(&lru_count));
1416 }
1417
1418 static struct shrinker glock_shrinker = {
1419         .seeks = DEFAULT_SEEKS,
1420         .count_objects = gfs2_glock_shrink_count,
1421         .scan_objects = gfs2_glock_shrink_scan,
1422 };
1423
1424 /**
1425  * examine_bucket - Call a function for glock in a hash bucket
1426  * @examiner: the function
1427  * @sdp: the filesystem
1428  * @bucket: the bucket
1429  *
1430  */
1431
1432 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1433 {
1434         struct gfs2_glock *gl;
1435         struct rhash_head *pos;
1436         const struct bucket_table *tbl;
1437         int i;
1438
1439         rcu_read_lock();
1440         tbl = rht_dereference_rcu(gl_hash_table.tbl, &gl_hash_table);
1441         for (i = 0; i < tbl->size; i++) {
1442                 rht_for_each_entry_rcu(gl, pos, tbl, i, gl_node) {
1443                         if ((gl->gl_name.ln_sbd == sdp) &&
1444                             lockref_get_not_dead(&gl->gl_lockref))
1445                                 examiner(gl);
1446                 }
1447         }
1448         rcu_read_unlock();
1449         cond_resched();
1450 }
1451
1452 /**
1453  * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1454  * @gl: The glock to thaw
1455  *
1456  */
1457
1458 static void thaw_glock(struct gfs2_glock *gl)
1459 {
1460         if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1461                 goto out;
1462         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1463         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0) {
1464 out:
1465                 gfs2_glock_put(gl);
1466         }
1467 }
1468
1469 /**
1470  * clear_glock - look at a glock and see if we can free it from glock cache
1471  * @gl: the glock to look at
1472  *
1473  */
1474
1475 static void clear_glock(struct gfs2_glock *gl)
1476 {
1477         gfs2_glock_remove_from_lru(gl);
1478
1479         spin_lock(&gl->gl_lockref.lock);
1480         if (gl->gl_state != LM_ST_UNLOCKED)
1481                 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1482         spin_unlock(&gl->gl_lockref.lock);
1483         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1484                 gfs2_glock_put(gl);
1485 }
1486
1487 /**
1488  * gfs2_glock_thaw - Thaw any frozen glocks
1489  * @sdp: The super block
1490  *
1491  */
1492
1493 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1494 {
1495         glock_hash_walk(thaw_glock, sdp);
1496 }
1497
1498 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1499 {
1500         spin_lock(&gl->gl_lockref.lock);
1501         gfs2_dump_glock(seq, gl);
1502         spin_unlock(&gl->gl_lockref.lock);
1503 }
1504
1505 static void dump_glock_func(struct gfs2_glock *gl)
1506 {
1507         dump_glock(NULL, gl);
1508 }
1509
1510 /**
1511  * gfs2_gl_hash_clear - Empty out the glock hash table
1512  * @sdp: the filesystem
1513  * @wait: wait until it's all gone
1514  *
1515  * Called when unmounting the filesystem.
1516  */
1517
1518 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1519 {
1520         set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1521         flush_workqueue(glock_workqueue);
1522         glock_hash_walk(clear_glock, sdp);
1523         flush_workqueue(glock_workqueue);
1524         wait_event_timeout(sdp->sd_glock_wait,
1525                            atomic_read(&sdp->sd_glock_disposal) == 0,
1526                            HZ * 600);
1527         glock_hash_walk(dump_glock_func, sdp);
1528 }
1529
1530 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1531 {
1532         struct gfs2_glock *gl = ip->i_gl;
1533         int ret;
1534
1535         ret = gfs2_truncatei_resume(ip);
1536         gfs2_assert_withdraw(gl->gl_name.ln_sbd, ret == 0);
1537
1538         spin_lock(&gl->gl_lockref.lock);
1539         clear_bit(GLF_LOCK, &gl->gl_flags);
1540         run_queue(gl, 1);
1541         spin_unlock(&gl->gl_lockref.lock);
1542 }
1543
1544 static const char *state2str(unsigned state)
1545 {
1546         switch(state) {
1547         case LM_ST_UNLOCKED:
1548                 return "UN";
1549         case LM_ST_SHARED:
1550                 return "SH";
1551         case LM_ST_DEFERRED:
1552                 return "DF";
1553         case LM_ST_EXCLUSIVE:
1554                 return "EX";
1555         }
1556         return "??";
1557 }
1558
1559 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
1560 {
1561         char *p = buf;
1562         if (flags & LM_FLAG_TRY)
1563                 *p++ = 't';
1564         if (flags & LM_FLAG_TRY_1CB)
1565                 *p++ = 'T';
1566         if (flags & LM_FLAG_NOEXP)
1567                 *p++ = 'e';
1568         if (flags & LM_FLAG_ANY)
1569                 *p++ = 'A';
1570         if (flags & LM_FLAG_PRIORITY)
1571                 *p++ = 'p';
1572         if (flags & GL_ASYNC)
1573                 *p++ = 'a';
1574         if (flags & GL_EXACT)
1575                 *p++ = 'E';
1576         if (flags & GL_NOCACHE)
1577                 *p++ = 'c';
1578         if (test_bit(HIF_HOLDER, &iflags))
1579                 *p++ = 'H';
1580         if (test_bit(HIF_WAIT, &iflags))
1581                 *p++ = 'W';
1582         if (test_bit(HIF_FIRST, &iflags))
1583                 *p++ = 'F';
1584         *p = 0;
1585         return buf;
1586 }
1587
1588 /**
1589  * dump_holder - print information about a glock holder
1590  * @seq: the seq_file struct
1591  * @gh: the glock holder
1592  *
1593  */
1594
1595 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1596 {
1597         struct task_struct *gh_owner = NULL;
1598         char flags_buf[32];
1599
1600         rcu_read_lock();
1601         if (gh->gh_owner_pid)
1602                 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1603         gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1604                        state2str(gh->gh_state),
1605                        hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1606                        gh->gh_error,
1607                        gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1608                        gh_owner ? gh_owner->comm : "(ended)",
1609                        (void *)gh->gh_ip);
1610         rcu_read_unlock();
1611 }
1612
1613 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1614 {
1615         const unsigned long *gflags = &gl->gl_flags;
1616         char *p = buf;
1617
1618         if (test_bit(GLF_LOCK, gflags))
1619                 *p++ = 'l';
1620         if (test_bit(GLF_DEMOTE, gflags))
1621                 *p++ = 'D';
1622         if (test_bit(GLF_PENDING_DEMOTE, gflags))
1623                 *p++ = 'd';
1624         if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1625                 *p++ = 'p';
1626         if (test_bit(GLF_DIRTY, gflags))
1627                 *p++ = 'y';
1628         if (test_bit(GLF_LFLUSH, gflags))
1629                 *p++ = 'f';
1630         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1631                 *p++ = 'i';
1632         if (test_bit(GLF_REPLY_PENDING, gflags))
1633                 *p++ = 'r';
1634         if (test_bit(GLF_INITIAL, gflags))
1635                 *p++ = 'I';
1636         if (test_bit(GLF_FROZEN, gflags))
1637                 *p++ = 'F';
1638         if (test_bit(GLF_QUEUED, gflags))
1639                 *p++ = 'q';
1640         if (test_bit(GLF_LRU, gflags))
1641                 *p++ = 'L';
1642         if (gl->gl_object)
1643                 *p++ = 'o';
1644         if (test_bit(GLF_BLOCKING, gflags))
1645                 *p++ = 'b';
1646         *p = 0;
1647         return buf;
1648 }
1649
1650 /**
1651  * gfs2_dump_glock - print information about a glock
1652  * @seq: The seq_file struct
1653  * @gl: the glock
1654  *
1655  * The file format is as follows:
1656  * One line per object, capital letters are used to indicate objects
1657  * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1658  * other objects are indented by a single space and follow the glock to
1659  * which they are related. Fields are indicated by lower case letters
1660  * followed by a colon and the field value, except for strings which are in
1661  * [] so that its possible to see if they are composed of spaces for
1662  * example. The field's are n = number (id of the object), f = flags,
1663  * t = type, s = state, r = refcount, e = error, p = pid.
1664  *
1665  */
1666
1667 void gfs2_dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1668 {
1669         const struct gfs2_glock_operations *glops = gl->gl_ops;
1670         unsigned long long dtime;
1671         const struct gfs2_holder *gh;
1672         char gflags_buf[32];
1673
1674         dtime = jiffies - gl->gl_demote_time;
1675         dtime *= 1000000/HZ; /* demote time in uSec */
1676         if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1677                 dtime = 0;
1678         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",
1679                   state2str(gl->gl_state),
1680                   gl->gl_name.ln_type,
1681                   (unsigned long long)gl->gl_name.ln_number,
1682                   gflags2str(gflags_buf, gl),
1683                   state2str(gl->gl_target),
1684                   state2str(gl->gl_demote_state), dtime,
1685                   atomic_read(&gl->gl_ail_count),
1686                   atomic_read(&gl->gl_revokes),
1687                   (int)gl->gl_lockref.count, gl->gl_hold_time);
1688
1689         list_for_each_entry(gh, &gl->gl_holders, gh_list)
1690                 dump_holder(seq, gh);
1691
1692         if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1693                 glops->go_dump(seq, gl);
1694 }
1695
1696 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
1697 {
1698         struct gfs2_glock *gl = iter_ptr;
1699
1700         seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
1701                    gl->gl_name.ln_type,
1702                    (unsigned long long)gl->gl_name.ln_number,
1703                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
1704                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
1705                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
1706                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
1707                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
1708                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
1709                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
1710                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
1711         return 0;
1712 }
1713
1714 static const char *gfs2_gltype[] = {
1715         "type",
1716         "reserved",
1717         "nondisk",
1718         "inode",
1719         "rgrp",
1720         "meta",
1721         "iopen",
1722         "flock",
1723         "plock",
1724         "quota",
1725         "journal",
1726 };
1727
1728 static const char *gfs2_stype[] = {
1729         [GFS2_LKS_SRTT]         = "srtt",
1730         [GFS2_LKS_SRTTVAR]      = "srttvar",
1731         [GFS2_LKS_SRTTB]        = "srttb",
1732         [GFS2_LKS_SRTTVARB]     = "srttvarb",
1733         [GFS2_LKS_SIRT]         = "sirt",
1734         [GFS2_LKS_SIRTVAR]      = "sirtvar",
1735         [GFS2_LKS_DCOUNT]       = "dlm",
1736         [GFS2_LKS_QCOUNT]       = "queue",
1737 };
1738
1739 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1740
1741 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
1742 {
1743         struct gfs2_sbd *sdp = seq->private;
1744         loff_t pos = *(loff_t *)iter_ptr;
1745         unsigned index = pos >> 3;
1746         unsigned subindex = pos & 0x07;
1747         int i;
1748
1749         if (index == 0 && subindex != 0)
1750                 return 0;
1751
1752         seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
1753                    (index == 0) ? "cpu": gfs2_stype[subindex]);
1754
1755         for_each_possible_cpu(i) {
1756                 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
1757
1758                 if (index == 0)
1759                         seq_printf(seq, " %15u", i);
1760                 else
1761                         seq_printf(seq, " %15llu", (unsigned long long)lkstats->
1762                                    lkstats[index - 1].stats[subindex]);
1763         }
1764         seq_putc(seq, '\n');
1765         return 0;
1766 }
1767
1768 int __init gfs2_glock_init(void)
1769 {
1770         int ret;
1771
1772         ret = rhashtable_init(&gl_hash_table, &ht_parms);
1773         if (ret < 0)
1774                 return ret;
1775
1776         glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1777                                           WQ_HIGHPRI | WQ_FREEZABLE, 0);
1778         if (!glock_workqueue) {
1779                 rhashtable_destroy(&gl_hash_table);
1780                 return -ENOMEM;
1781         }
1782         gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1783                                                 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1784                                                 0);
1785         if (!gfs2_delete_workqueue) {
1786                 destroy_workqueue(glock_workqueue);
1787                 rhashtable_destroy(&gl_hash_table);
1788                 return -ENOMEM;
1789         }
1790
1791         register_shrinker(&glock_shrinker);
1792
1793         return 0;
1794 }
1795
1796 void gfs2_glock_exit(void)
1797 {
1798         unregister_shrinker(&glock_shrinker);
1799         rhashtable_destroy(&gl_hash_table);
1800         destroy_workqueue(glock_workqueue);
1801         destroy_workqueue(gfs2_delete_workqueue);
1802 }
1803
1804 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1805 {
1806         do {
1807                 gi->gl = rhashtable_walk_next(&gi->hti);
1808                 if (IS_ERR(gi->gl)) {
1809                         if (PTR_ERR(gi->gl) == -EAGAIN)
1810                                 continue;
1811                         gi->gl = NULL;
1812                 }
1813         /* Skip entries for other sb and dead entries */
1814         } while ((gi->gl) && ((gi->sdp != gi->gl->gl_name.ln_sbd) ||
1815                               __lockref_is_dead(&gi->gl->gl_lockref)));
1816 }
1817
1818 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1819 {
1820         struct gfs2_glock_iter *gi = seq->private;
1821         loff_t n = *pos;
1822         int ret;
1823
1824         if (gi->last_pos <= *pos)
1825                 n = (*pos - gi->last_pos);
1826
1827         ret = rhashtable_walk_start(&gi->hti);
1828         if (ret)
1829                 return NULL;
1830
1831         do {
1832                 gfs2_glock_iter_next(gi);
1833         } while (gi->gl && n--);
1834
1835         gi->last_pos = *pos;
1836         return gi->gl;
1837 }
1838
1839 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1840                                  loff_t *pos)
1841 {
1842         struct gfs2_glock_iter *gi = seq->private;
1843
1844         (*pos)++;
1845         gi->last_pos = *pos;
1846         gfs2_glock_iter_next(gi);
1847         return gi->gl;
1848 }
1849
1850 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1851 {
1852         struct gfs2_glock_iter *gi = seq->private;
1853
1854         gi->gl = NULL;
1855         rhashtable_walk_stop(&gi->hti);
1856 }
1857
1858 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1859 {
1860         dump_glock(seq, iter_ptr);
1861         return 0;
1862 }
1863
1864 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
1865 {
1866         preempt_disable();
1867         if (*pos >= GFS2_NR_SBSTATS)
1868                 return NULL;
1869         return pos;
1870 }
1871
1872 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
1873                                    loff_t *pos)
1874 {
1875         (*pos)++;
1876         if (*pos >= GFS2_NR_SBSTATS)
1877                 return NULL;
1878         return pos;
1879 }
1880
1881 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
1882 {
1883         preempt_enable();
1884 }
1885
1886 static const struct seq_operations gfs2_glock_seq_ops = {
1887         .start = gfs2_glock_seq_start,
1888         .next  = gfs2_glock_seq_next,
1889         .stop  = gfs2_glock_seq_stop,
1890         .show  = gfs2_glock_seq_show,
1891 };
1892
1893 static const struct seq_operations gfs2_glstats_seq_ops = {
1894         .start = gfs2_glock_seq_start,
1895         .next  = gfs2_glock_seq_next,
1896         .stop  = gfs2_glock_seq_stop,
1897         .show  = gfs2_glstats_seq_show,
1898 };
1899
1900 static const struct seq_operations gfs2_sbstats_seq_ops = {
1901         .start = gfs2_sbstats_seq_start,
1902         .next  = gfs2_sbstats_seq_next,
1903         .stop  = gfs2_sbstats_seq_stop,
1904         .show  = gfs2_sbstats_seq_show,
1905 };
1906
1907 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
1908
1909 static int gfs2_glocks_open(struct inode *inode, struct file *file)
1910 {
1911         int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1912                                    sizeof(struct gfs2_glock_iter));
1913         if (ret == 0) {
1914                 struct seq_file *seq = file->private_data;
1915                 struct gfs2_glock_iter *gi = seq->private;
1916
1917                 gi->sdp = inode->i_private;
1918                 gi->last_pos = 0;
1919                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
1920                 if (seq->buf)
1921                         seq->size = GFS2_SEQ_GOODSIZE;
1922                 gi->gl = NULL;
1923                 ret = rhashtable_walk_init(&gl_hash_table, &gi->hti, GFP_KERNEL);
1924         }
1925         return ret;
1926 }
1927
1928 static int gfs2_glocks_release(struct inode *inode, struct file *file)
1929 {
1930         struct seq_file *seq = file->private_data;
1931         struct gfs2_glock_iter *gi = seq->private;
1932
1933         gi->gl = NULL;
1934         rhashtable_walk_exit(&gi->hti);
1935         return seq_release_private(inode, file);
1936 }
1937
1938 static int gfs2_glstats_open(struct inode *inode, struct file *file)
1939 {
1940         int ret = seq_open_private(file, &gfs2_glstats_seq_ops,
1941                                    sizeof(struct gfs2_glock_iter));
1942         if (ret == 0) {
1943                 struct seq_file *seq = file->private_data;
1944                 struct gfs2_glock_iter *gi = seq->private;
1945                 gi->sdp = inode->i_private;
1946                 gi->last_pos = 0;
1947                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
1948                 if (seq->buf)
1949                         seq->size = GFS2_SEQ_GOODSIZE;
1950                 gi->gl = NULL;
1951                 ret = rhashtable_walk_init(&gl_hash_table, &gi->hti, GFP_KERNEL);
1952         }
1953         return ret;
1954 }
1955
1956 static int gfs2_sbstats_open(struct inode *inode, struct file *file)
1957 {
1958         int ret = seq_open(file, &gfs2_sbstats_seq_ops);
1959         if (ret == 0) {
1960                 struct seq_file *seq = file->private_data;
1961                 seq->private = inode->i_private;  /* sdp */
1962         }
1963         return ret;
1964 }
1965
1966 static const struct file_operations gfs2_glocks_fops = {
1967         .owner   = THIS_MODULE,
1968         .open    = gfs2_glocks_open,
1969         .read    = seq_read,
1970         .llseek  = seq_lseek,
1971         .release = gfs2_glocks_release,
1972 };
1973
1974 static const struct file_operations gfs2_glstats_fops = {
1975         .owner   = THIS_MODULE,
1976         .open    = gfs2_glstats_open,
1977         .read    = seq_read,
1978         .llseek  = seq_lseek,
1979         .release = gfs2_glocks_release,
1980 };
1981
1982 static const struct file_operations gfs2_sbstats_fops = {
1983         .owner   = THIS_MODULE,
1984         .open    = gfs2_sbstats_open,
1985         .read    = seq_read,
1986         .llseek  = seq_lseek,
1987         .release = seq_release,
1988 };
1989
1990 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
1991 {
1992         struct dentry *dent;
1993
1994         dent = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
1995         if (IS_ERR_OR_NULL(dent))
1996                 goto fail;
1997         sdp->debugfs_dir = dent;
1998
1999         dent = debugfs_create_file("glocks",
2000                                    S_IFREG | S_IRUGO,
2001                                    sdp->debugfs_dir, sdp,
2002                                    &gfs2_glocks_fops);
2003         if (IS_ERR_OR_NULL(dent))
2004                 goto fail;
2005         sdp->debugfs_dentry_glocks = dent;
2006
2007         dent = debugfs_create_file("glstats",
2008                                    S_IFREG | S_IRUGO,
2009                                    sdp->debugfs_dir, sdp,
2010                                    &gfs2_glstats_fops);
2011         if (IS_ERR_OR_NULL(dent))
2012                 goto fail;
2013         sdp->debugfs_dentry_glstats = dent;
2014
2015         dent = debugfs_create_file("sbstats",
2016                                    S_IFREG | S_IRUGO,
2017                                    sdp->debugfs_dir, sdp,
2018                                    &gfs2_sbstats_fops);
2019         if (IS_ERR_OR_NULL(dent))
2020                 goto fail;
2021         sdp->debugfs_dentry_sbstats = dent;
2022
2023         return 0;
2024 fail:
2025         gfs2_delete_debugfs_file(sdp);
2026         return dent ? PTR_ERR(dent) : -ENOMEM;
2027 }
2028
2029 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2030 {
2031         if (sdp->debugfs_dir) {
2032                 if (sdp->debugfs_dentry_glocks) {
2033                         debugfs_remove(sdp->debugfs_dentry_glocks);
2034                         sdp->debugfs_dentry_glocks = NULL;
2035                 }
2036                 if (sdp->debugfs_dentry_glstats) {
2037                         debugfs_remove(sdp->debugfs_dentry_glstats);
2038                         sdp->debugfs_dentry_glstats = NULL;
2039                 }
2040                 if (sdp->debugfs_dentry_sbstats) {
2041                         debugfs_remove(sdp->debugfs_dentry_sbstats);
2042                         sdp->debugfs_dentry_sbstats = NULL;
2043                 }
2044                 debugfs_remove(sdp->debugfs_dir);
2045                 sdp->debugfs_dir = NULL;
2046         }
2047 }
2048
2049 int gfs2_register_debugfs(void)
2050 {
2051         gfs2_root = debugfs_create_dir("gfs2", NULL);
2052         if (IS_ERR(gfs2_root))
2053                 return PTR_ERR(gfs2_root);
2054         return gfs2_root ? 0 : -ENOMEM;
2055 }
2056
2057 void gfs2_unregister_debugfs(void)
2058 {
2059         debugfs_remove(gfs2_root);
2060         gfs2_root = NULL;
2061 }