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 inline 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 = 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) {
 479                         pr_err("lm_lock ret %d\n", ret);
 480                         GLOCK_BUG_ON(gl, 1);
 481                 }
 482         } else { /* lock_nolock */
 483                 finish_xmote(gl, target);
 484                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 485                         gfs2_glock_put(gl);
 486         }
 487
 488         spin_lock(&gl->gl_lockref.lock);
 489 }
 490
 491 /**
 492  * find_first_holder - find the first "holder" gh
 493  * @gl: the glock
 494  */
 495
 496 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
 497 {
 498         struct gfs2_holder *gh;
 499
 500         if (!list_empty(&gl->gl_holders)) {
 501                 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
 502                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 503                         return gh;
 504         }
 505         return NULL;
 506 }
 507
 508 /**
 509  * run_queue - do all outstanding tasks related to a glock
 510  * @gl: The glock in question
 511  * @nonblock: True if we must not block in run_queue
 512  *
 513  */
 514
 515 static void run_queue(struct gfs2_glock *gl, const int nonblock)
 516 __releases(&gl->gl_lockref.lock)
 517 __acquires(&gl->gl_lockref.lock)
 518 {
 519         struct gfs2_holder *gh = NULL;
 520         int ret;
 521
 522         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
 523                 return;
 524
 525         GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
 526
 527         if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
 528             gl->gl_demote_state != gl->gl_state) {
 529                 if (find_first_holder(gl))
 530                         goto out_unlock;
 531                 if (nonblock)
 532                         goto out_sched;
 533                 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
 534                 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
 535                 gl->gl_target = gl->gl_demote_state;
 536         } else {
 537                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
 538                         gfs2_demote_wake(gl);
 539                 ret = do_promote(gl);
 540                 if (ret == 0)
 541                         goto out_unlock;
 542                 if (ret == 2)
 543                         goto out;
 544                 gh = find_first_waiter(gl);
 545                 gl->gl_target = gh->gh_state;
 546                 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
 547                         do_error(gl, 0); /* Fail queued try locks */
 548         }
 549         do_xmote(gl, gh, gl->gl_target);
 550 out:
 551         return;
 552
 553 out_sched:
 554         clear_bit(GLF_LOCK, &gl->gl_flags);
 555         smp_mb__after_atomic();
 556         gl->gl_lockref.count++;
 557         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 558                 gl->gl_lockref.count--;
 559         return;
 560
 561 out_unlock:
 562         clear_bit(GLF_LOCK, &gl->gl_flags);
 563         smp_mb__after_atomic();
 564         return;
 565 }
 566
 567 static void delete_work_func(struct work_struct *work)
 568 {
 569         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
 570         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 571         struct gfs2_inode *ip;
 572         struct inode *inode;
 573         u64 no_addr = gl->gl_name.ln_number;
 574
 575         ip = gl->gl_object;
 576         /* Note: Unsafe to dereference ip as we don't hold right refs/locks */
 577
 578         if (ip)
 579                 inode = gfs2_ilookup(sdp->sd_vfs, no_addr, 1);
 580         else
 581                 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
 582         if (inode && !IS_ERR(inode)) {
 583                 d_prune_aliases(inode);
 584                 iput(inode);
 585         }
 586         gfs2_glock_put(gl);
 587 }
 588
 589 static void glock_work_func(struct work_struct *work)
 590 {
 591         unsigned long delay = 0;
 592         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
 593         int drop_ref = 0;
 594
 595         if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
 596                 finish_xmote(gl, gl->gl_reply);
 597                 drop_ref = 1;
 598         }
 599         spin_lock(&gl->gl_lockref.lock);
 600         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
 601             gl->gl_state != LM_ST_UNLOCKED &&
 602             gl->gl_demote_state != LM_ST_EXCLUSIVE) {
 603                 unsigned long holdtime, now = jiffies;
 604
 605                 holdtime = gl->gl_tchange + gl->gl_hold_time;
 606                 if (time_before(now, holdtime))
 607                         delay = holdtime - now;
 608
 609                 if (!delay) {
 610                         clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
 611                         set_bit(GLF_DEMOTE, &gl->gl_flags);
 612                 }
 613         }
 614         run_queue(gl, 0);
 615         spin_unlock(&gl->gl_lockref.lock);
 616         if (!delay)
 617                 gfs2_glock_put(gl);
 618         else {
 619                 if (gl->gl_name.ln_type != LM_TYPE_INODE)
 620                         delay = 0;
 621                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
 622                         gfs2_glock_put(gl);
 623         }
 624         if (drop_ref)
 625                 gfs2_glock_put(gl);
 626 }
 627
 628 /**
 629  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
 630  * @sdp: The GFS2 superblock
 631  * @number: the lock number
 632  * @glops: The glock_operations to use
 633  * @create: If 0, don't create the glock if it doesn't exist
 634  * @glp: the glock is returned here
 635  *
 636  * This does not lock a glock, just finds/creates structures for one.
 637  *
 638  * Returns: errno
 639  */
 640
 641 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
 642                    const struct gfs2_glock_operations *glops, int create,
 643                    struct gfs2_glock **glp)
 644 {
 645         struct super_block *s = sdp->sd_vfs;
 646         struct lm_lockname name = { .ln_number = number,
 647                                     .ln_type = glops->go_type,
 648                                     .ln_sbd = sdp };
 649         struct gfs2_glock *gl, *tmp = NULL;
 650         struct address_space *mapping;
 651         struct kmem_cache *cachep;
 652         int ret, tries = 0;
 653
 654         rcu_read_lock();
 655         gl = rhashtable_lookup_fast(&gl_hash_table, &name, ht_parms);
 656         if (gl && !lockref_get_not_dead(&gl->gl_lockref))
 657                 gl = NULL;
 658         rcu_read_unlock();
 659
 660         *glp = gl;
 661         if (gl)
 662                 return 0;
 663         if (!create)
 664                 return -ENOENT;
 665
 666         if (glops->go_flags & GLOF_ASPACE)
 667                 cachep = gfs2_glock_aspace_cachep;
 668         else
 669                 cachep = gfs2_glock_cachep;
 670         gl = kmem_cache_alloc(cachep, GFP_NOFS);
 671         if (!gl)
 672                 return -ENOMEM;
 673
 674         memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
 675
 676         if (glops->go_flags & GLOF_LVB) {
 677                 gl->gl_lksb.sb_lvbptr = kzalloc(GFS2_MIN_LVB_SIZE, GFP_NOFS);
 678                 if (!gl->gl_lksb.sb_lvbptr) {
 679                         kmem_cache_free(cachep, gl);
 680                         return -ENOMEM;
 681                 }
 682         }
 683
 684         atomic_inc(&sdp->sd_glock_disposal);
 685         gl->gl_node.next = NULL;
 686         gl->gl_flags = 0;
 687         gl->gl_name = name;
 688         gl->gl_lockref.count = 1;
 689         gl->gl_state = LM_ST_UNLOCKED;
 690         gl->gl_target = LM_ST_UNLOCKED;
 691         gl->gl_demote_state = LM_ST_EXCLUSIVE;
 692         gl->gl_ops = glops;
 693         gl->gl_dstamp = ktime_set(0, 0);
 694         preempt_disable();
 695         /* We use the global stats to estimate the initial per-glock stats */
 696         gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
 697         preempt_enable();
 698         gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
 699         gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
 700         gl->gl_tchange = jiffies;
 701         gl->gl_object = NULL;
 702         gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
 703         INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
 704         INIT_WORK(&gl->gl_delete, delete_work_func);
 705
 706         mapping = gfs2_glock2aspace(gl);
 707         if (mapping) {
 708                 mapping->a_ops = &gfs2_meta_aops;
 709                 mapping->host = s->s_bdev->bd_inode;
 710                 mapping->flags = 0;
 711                 mapping_set_gfp_mask(mapping, GFP_NOFS);
 712                 mapping->private_data = NULL;
 713                 mapping->writeback_index = 0;
 714         }
 715
 716 again:
 717         ret = rhashtable_lookup_insert_fast(&gl_hash_table, &gl->gl_node,
 718                                             ht_parms);
 719         if (ret == 0) {
 720                 *glp = gl;
 721                 return 0;
 722         }
 723
 724         if (ret == -EEXIST) {
 725                 ret = 0;
 726                 rcu_read_lock();
 727                 tmp = rhashtable_lookup_fast(&gl_hash_table, &name, ht_parms);
 728                 if (tmp == NULL || !lockref_get_not_dead(&tmp->gl_lockref)) {
 729                         if (++tries < 100) {
 730                                 rcu_read_unlock();
 731                                 cond_resched();
 732                                 goto again;
 733                         }
 734                         tmp = NULL;
 735                         ret = -ENOMEM;
 736                 }
 737                 rcu_read_unlock();
 738         } else {
 739                 WARN_ON_ONCE(ret);
 740         }
 741         kfree(gl->gl_lksb.sb_lvbptr);
 742         kmem_cache_free(cachep, gl);
 743         atomic_dec(&sdp->sd_glock_disposal);
 744         *glp = tmp;
 745
 746         return ret;
 747 }
 748
 749 /**
 750  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
 751  * @gl: the glock
 752  * @state: the state we're requesting
 753  * @flags: the modifier flags
 754  * @gh: the holder structure
 755  *
 756  */
 757
 758 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
 759                       struct gfs2_holder *gh)
 760 {
 761         INIT_LIST_HEAD(&gh->gh_list);
 762         gh->gh_gl = gl;
 763         gh->gh_ip = _RET_IP_;
 764         gh->gh_owner_pid = get_pid(task_pid(current));
 765         gh->gh_state = state;
 766         gh->gh_flags = flags;
 767         gh->gh_error = 0;
 768         gh->gh_iflags = 0;
 769         gfs2_glock_hold(gl);
 770 }
 771
 772 /**
 773  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
 774  * @state: the state we're requesting
 775  * @flags: the modifier flags
 776  * @gh: the holder structure
 777  *
 778  * Don't mess with the glock.
 779  *
 780  */
 781
 782 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
 783 {
 784         gh->gh_state = state;
 785         gh->gh_flags = flags;
 786         gh->gh_iflags = 0;
 787         gh->gh_ip = _RET_IP_;
 788         put_pid(gh->gh_owner_pid);
 789         gh->gh_owner_pid = get_pid(task_pid(current));
 790 }
 791
 792 /**
 793  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
 794  * @gh: the holder structure
 795  *
 796  */
 797
 798 void gfs2_holder_uninit(struct gfs2_holder *gh)
 799 {
 800         put_pid(gh->gh_owner_pid);
 801         gfs2_glock_put(gh->gh_gl);
 802         gh->gh_gl = NULL;
 803         gh->gh_ip = 0;
 804 }
 805
 806 /**
 807  * gfs2_glock_wait - wait on a glock acquisition
 808  * @gh: the glock holder
 809  *
 810  * Returns: 0 on success
 811  */
 812
 813 int gfs2_glock_wait(struct gfs2_holder *gh)
 814 {
 815         unsigned long time1 = jiffies;
 816
 817         might_sleep();
 818         wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
 819         if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
 820                 /* Lengthen the minimum hold time. */
 821                 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
 822                                               GL_GLOCK_HOLD_INCR,
 823                                               GL_GLOCK_MAX_HOLD);
 824         return gh->gh_error;
 825 }
 826
 827 /**
 828  * handle_callback - process a demote request
 829  * @gl: the glock
 830  * @state: the state the caller wants us to change to
 831  *
 832  * There are only two requests that we are going to see in actual
 833  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
 834  */
 835
 836 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
 837                             unsigned long delay, bool remote)
 838 {
 839         int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
 840
 841         set_bit(bit, &gl->gl_flags);
 842         if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
 843                 gl->gl_demote_state = state;
 844                 gl->gl_demote_time = jiffies;
 845         } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
 846                         gl->gl_demote_state != state) {
 847                 gl->gl_demote_state = LM_ST_UNLOCKED;
 848         }
 849         if (gl->gl_ops->go_callback)
 850                 gl->gl_ops->go_callback(gl, remote);
 851         trace_gfs2_demote_rq(gl, remote);
 852 }
 853
 854 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
 855 {
 856         struct va_format vaf;
 857         va_list args;
 858
 859         va_start(args, fmt);
 860
 861         if (seq) {
 862                 seq_vprintf(seq, fmt, args);
 863         } else {
 864                 vaf.fmt = fmt;
 865                 vaf.va = &args;
 866
 867                 pr_err("%pV", &vaf);
 868         }
 869
 870         va_end(args);
 871 }
 872
 873 /**
 874  * add_to_queue - Add a holder to the wait queue (but look for recursion)
 875  * @gh: the holder structure to add
 876  *
 877  * Eventually we should move the recursive locking trap to a
 878  * debugging option or something like that. This is the fast
 879  * path and needs to have the minimum number of distractions.
 880  *
 881  */
 882
 883 static inline void add_to_queue(struct gfs2_holder *gh)
 884 __releases(&gl->gl_lockref.lock)
 885 __acquires(&gl->gl_lockref.lock)
 886 {
 887         struct gfs2_glock *gl = gh->gh_gl;
 888         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 889         struct list_head *insert_pt = NULL;
 890         struct gfs2_holder *gh2;
 891         int try_futile = 0;
 892
 893         BUG_ON(gh->gh_owner_pid == NULL);
 894         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
 895                 BUG();
 896
 897         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
 898                 if (test_bit(GLF_LOCK, &gl->gl_flags))
 899                         try_futile = !may_grant(gl, gh);
 900                 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
 901                         goto fail;
 902         }
 903
 904         list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
 905                 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
 906                     (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
 907                         goto trap_recursive;
 908                 if (try_futile &&
 909                     !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
 910 fail:
 911                         gh->gh_error = GLR_TRYFAILED;
 912                         gfs2_holder_wake(gh);
 913                         return;
 914                 }
 915                 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
 916                         continue;
 917                 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
 918                         insert_pt = &gh2->gh_list;
 919         }
 920         set_bit(GLF_QUEUED, &gl->gl_flags);
 921         trace_gfs2_glock_queue(gh, 1);
 922         gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
 923         gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
 924         if (likely(insert_pt == NULL)) {
 925                 list_add_tail(&gh->gh_list, &gl->gl_holders);
 926                 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
 927                         goto do_cancel;
 928                 return;
 929         }
 930         list_add_tail(&gh->gh_list, insert_pt);
 931 do_cancel:
 932         gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
 933         if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
 934                 spin_unlock(&gl->gl_lockref.lock);
 935                 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
 936                         sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
 937                 spin_lock(&gl->gl_lockref.lock);
 938         }
 939         return;
 940
 941 trap_recursive:
 942         pr_err("original: %pSR\n", (void *)gh2->gh_ip);
 943         pr_err("pid: %d\n", pid_nr(gh2->gh_owner_pid));
 944         pr_err("lock type: %d req lock state : %d\n",
 945                gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
 946         pr_err("new: %pSR\n", (void *)gh->gh_ip);
 947         pr_err("pid: %d\n", pid_nr(gh->gh_owner_pid));
 948         pr_err("lock type: %d req lock state : %d\n",
 949                gh->gh_gl->gl_name.ln_type, gh->gh_state);
 950         gfs2_dump_glock(NULL, gl);
 951         BUG();
 952 }
 953
 954 /**
 955  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
 956  * @gh: the holder structure
 957  *
 958  * if (gh->gh_flags & GL_ASYNC), this never returns an error
 959  *
 960  * Returns: 0, GLR_TRYFAILED, or errno on failure
 961  */
 962
 963 int gfs2_glock_nq(struct gfs2_holder *gh)
 964 {
 965         struct gfs2_glock *gl = gh->gh_gl;
 966         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 967         int error = 0;
 968
 969         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 970                 return -EIO;
 971
 972         if (test_bit(GLF_LRU, &gl->gl_flags))
 973                 gfs2_glock_remove_from_lru(gl);
 974
 975         spin_lock(&gl->gl_lockref.lock);
 976         add_to_queue(gh);
 977         if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
 978                      test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
 979                 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
 980                 gl->gl_lockref.count++;
 981                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
 982                         gl->gl_lockref.count--;
 983         }
 984         run_queue(gl, 1);
 985         spin_unlock(&gl->gl_lockref.lock);
 986
 987         if (!(gh->gh_flags & GL_ASYNC))
 988                 error = gfs2_glock_wait(gh);
 989
 990         return error;
 991 }
 992
 993 /**
 994  * gfs2_glock_poll - poll to see if an async request has been completed
 995  * @gh: the holder
 996  *
 997  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
 998  */
 999
1000 int gfs2_glock_poll(struct gfs2_holder *gh)
1001 {
1002         return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1003 }
1004
1005 /**
1006  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1007  * @gh: the glock holder
1008  *
1009  */
1010
1011 void gfs2_glock_dq(struct gfs2_holder *gh)
1012 {
1013         struct gfs2_glock *gl = gh->gh_gl;
1014         const struct gfs2_glock_operations *glops = gl->gl_ops;
1015         unsigned delay = 0;
1016         int fast_path = 0;
1017
1018         spin_lock(&gl->gl_lockref.lock);
1019         if (gh->gh_flags & GL_NOCACHE)
1020                 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1021
1022         list_del_init(&gh->gh_list);
1023         if (find_first_holder(gl) == NULL) {
1024                 if (glops->go_unlock) {
1025                         GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1026                         spin_unlock(&gl->gl_lockref.lock);
1027                         glops->go_unlock(gh);
1028                         spin_lock(&gl->gl_lockref.lock);
1029                         clear_bit(GLF_LOCK, &gl->gl_flags);
1030                 }
1031                 if (list_empty(&gl->gl_holders) &&
1032                     !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1033                     !test_bit(GLF_DEMOTE, &gl->gl_flags))
1034                         fast_path = 1;
1035         }
1036         if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl) &&
1037             (glops->go_flags & GLOF_LRU))
1038                 gfs2_glock_add_to_lru(gl);
1039
1040         trace_gfs2_glock_queue(gh, 0);
1041         spin_unlock(&gl->gl_lockref.lock);
1042         if (likely(fast_path))
1043                 return;
1044
1045         gfs2_glock_hold(gl);
1046         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1047             !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1048             gl->gl_name.ln_type == LM_TYPE_INODE)
1049                 delay = gl->gl_hold_time;
1050         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1051                 gfs2_glock_put(gl);
1052 }
1053
1054 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1055 {
1056         struct gfs2_glock *gl = gh->gh_gl;
1057         gfs2_glock_dq(gh);
1058         might_sleep();
1059         wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1060 }
1061
1062 /**
1063  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1064  * @gh: the holder structure
1065  *
1066  */
1067
1068 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1069 {
1070         gfs2_glock_dq(gh);
1071         gfs2_holder_uninit(gh);
1072 }
1073
1074 /**
1075  * gfs2_glock_nq_num - acquire a glock based on lock number
1076  * @sdp: the filesystem
1077  * @number: the lock number
1078  * @glops: the glock operations for the type of glock
1079  * @state: the state to acquire the glock in
1080  * @flags: modifier flags for the acquisition
1081  * @gh: the struct gfs2_holder
1082  *
1083  * Returns: errno
1084  */
1085
1086 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1087                       const struct gfs2_glock_operations *glops,
1088                       unsigned int state, int flags, struct gfs2_holder *gh)
1089 {
1090         struct gfs2_glock *gl;
1091         int error;
1092
1093         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1094         if (!error) {
1095                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1096                 gfs2_glock_put(gl);
1097         }
1098
1099         return error;
1100 }
1101
1102 /**
1103  * glock_compare - Compare two struct gfs2_glock structures for sorting
1104  * @arg_a: the first structure
1105  * @arg_b: the second structure
1106  *
1107  */
1108
1109 static int glock_compare(const void *arg_a, const void *arg_b)
1110 {
1111         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1112         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1113         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1114         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1115
1116         if (a->ln_number > b->ln_number)
1117                 return 1;
1118         if (a->ln_number < b->ln_number)
1119                 return -1;
1120         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1121         return 0;
1122 }
1123
1124 /**
1125  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1126  * @num_gh: the number of structures
1127  * @ghs: an array of struct gfs2_holder structures
1128  *
1129  * Returns: 0 on success (all glocks acquired),
1130  *          errno on failure (no glocks acquired)
1131  */
1132
1133 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1134                      struct gfs2_holder **p)
1135 {
1136         unsigned int x;
1137         int error = 0;
1138
1139         for (x = 0; x < num_gh; x++)
1140                 p[x] = &ghs[x];
1141
1142         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1143
1144         for (x = 0; x < num_gh; x++) {
1145                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1146
1147                 error = gfs2_glock_nq(p[x]);
1148                 if (error) {
1149                         while (x--)
1150                                 gfs2_glock_dq(p[x]);
1151                         break;
1152                 }
1153         }
1154
1155         return error;
1156 }
1157
1158 /**
1159  * gfs2_glock_nq_m - acquire multiple glocks
1160  * @num_gh: the number of structures
1161  * @ghs: an array of struct gfs2_holder structures
1162  *
1163  *
1164  * Returns: 0 on success (all glocks acquired),
1165  *          errno on failure (no glocks acquired)
1166  */
1167
1168 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1169 {
1170         struct gfs2_holder *tmp[4];
1171         struct gfs2_holder **pph = tmp;
1172         int error = 0;
1173
1174         switch(num_gh) {
1175         case 0:
1176                 return 0;
1177         case 1:
1178                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1179                 return gfs2_glock_nq(ghs);
1180         default:
1181                 if (num_gh <= 4)
1182                         break;
1183                 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1184                 if (!pph)
1185                         return -ENOMEM;
1186         }
1187
1188         error = nq_m_sync(num_gh, ghs, pph);
1189
1190         if (pph != tmp)
1191                 kfree(pph);
1192
1193         return error;
1194 }
1195
1196 /**
1197  * gfs2_glock_dq_m - release multiple glocks
1198  * @num_gh: the number of structures
1199  * @ghs: an array of struct gfs2_holder structures
1200  *
1201  */
1202
1203 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1204 {
1205         while (num_gh--)
1206                 gfs2_glock_dq(&ghs[num_gh]);
1207 }
1208
1209 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1210 {
1211         unsigned long delay = 0;
1212         unsigned long holdtime;
1213         unsigned long now = jiffies;
1214
1215         gfs2_glock_hold(gl);
1216         holdtime = gl->gl_tchange + gl->gl_hold_time;
1217         if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1218             gl->gl_name.ln_type == LM_TYPE_INODE) {
1219                 if (time_before(now, holdtime))
1220                         delay = holdtime - now;
1221                 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1222                         delay = gl->gl_hold_time;
1223         }
1224
1225         spin_lock(&gl->gl_lockref.lock);
1226         handle_callback(gl, state, delay, true);
1227         spin_unlock(&gl->gl_lockref.lock);
1228         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1229                 gfs2_glock_put(gl);
1230 }
1231
1232 /**
1233  * gfs2_should_freeze - Figure out if glock should be frozen
1234  * @gl: The glock in question
1235  *
1236  * Glocks are not frozen if (a) the result of the dlm operation is
1237  * an error, (b) the locking operation was an unlock operation or
1238  * (c) if there is a "noexp" flagged request anywhere in the queue
1239  *
1240  * Returns: 1 if freezing should occur, 0 otherwise
1241  */
1242
1243 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1244 {
1245         const struct gfs2_holder *gh;
1246
1247         if (gl->gl_reply & ~LM_OUT_ST_MASK)
1248                 return 0;
1249         if (gl->gl_target == LM_ST_UNLOCKED)
1250                 return 0;
1251
1252         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1253                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1254                         continue;
1255                 if (LM_FLAG_NOEXP & gh->gh_flags)
1256                         return 0;
1257         }
1258
1259         return 1;
1260 }
1261
1262 /**
1263  * gfs2_glock_complete - Callback used by locking
1264  * @gl: Pointer to the glock
1265  * @ret: The return value from the dlm
1266  *
1267  * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1268  * to use a bitfield shared with other glock state fields.
1269  */
1270
1271 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1272 {
1273         struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1274
1275         spin_lock(&gl->gl_lockref.lock);
1276         gl->gl_reply = ret;
1277
1278         if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1279                 if (gfs2_should_freeze(gl)) {
1280                         set_bit(GLF_FROZEN, &gl->gl_flags);
1281                         spin_unlock(&gl->gl_lockref.lock);
1282                         return;
1283                 }
1284         }
1285
1286         gl->gl_lockref.count++;
1287         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1288         spin_unlock(&gl->gl_lockref.lock);
1289
1290         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1291                 gfs2_glock_put(gl);
1292 }
1293
1294 static int glock_cmp(void *priv, struct list_head *a, struct list_head *b)
1295 {
1296         struct gfs2_glock *gla, *glb;
1297
1298         gla = list_entry(a, struct gfs2_glock, gl_lru);
1299         glb = list_entry(b, struct gfs2_glock, gl_lru);
1300
1301         if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1302                 return 1;
1303         if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1304                 return -1;
1305
1306         return 0;
1307 }
1308
1309 /**
1310  * gfs2_dispose_glock_lru - Demote a list of glocks
1311  * @list: The list to dispose of
1312  *
1313  * Disposing of glocks may involve disk accesses, so that here we sort
1314  * the glocks by number (i.e. disk location of the inodes) so that if
1315  * there are any such accesses, they'll be sent in order (mostly).
1316  *
1317  * Must be called under the lru_lock, but may drop and retake this
1318  * lock. While the lru_lock is dropped, entries may vanish from the
1319  * list, but no new entries will appear on the list (since it is
1320  * private)
1321  */
1322
1323 static void gfs2_dispose_glock_lru(struct list_head *list)
1324 __releases(&lru_lock)
1325 __acquires(&lru_lock)
1326 {
1327         struct gfs2_glock *gl;
1328
1329         list_sort(NULL, list, glock_cmp);
1330
1331         while(!list_empty(list)) {
1332                 gl = list_entry(list->next, struct gfs2_glock, gl_lru);
1333                 list_del_init(&gl->gl_lru);
1334                 if (!spin_trylock(&gl->gl_lockref.lock)) {
1335 add_back_to_lru:
1336                         list_add(&gl->gl_lru, &lru_list);
1337                         atomic_inc(&lru_count);
1338                         continue;
1339                 }
1340                 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1341                         spin_unlock(&gl->gl_lockref.lock);
1342                         goto add_back_to_lru;
1343                 }
1344                 clear_bit(GLF_LRU, &gl->gl_flags);
1345                 gl->gl_lockref.count++;
1346                 if (demote_ok(gl))
1347                         handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1348                 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1349                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1350                         gl->gl_lockref.count--;
1351                 spin_unlock(&gl->gl_lockref.lock);
1352                 cond_resched_lock(&lru_lock);
1353         }
1354 }
1355
1356 /**
1357  * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1358  * @nr: The number of entries to scan
1359  *
1360  * This function selects the entries on the LRU which are able to
1361  * be demoted, and then kicks off the process by calling
1362  * gfs2_dispose_glock_lru() above.
1363  */
1364
1365 static long gfs2_scan_glock_lru(int nr)
1366 {
1367         struct gfs2_glock *gl;
1368         LIST_HEAD(skipped);
1369         LIST_HEAD(dispose);
1370         long freed = 0;
1371
1372         spin_lock(&lru_lock);
1373         while ((nr-- >= 0) && !list_empty(&lru_list)) {
1374                 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1375
1376                 /* Test for being demotable */
1377                 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1378                         list_move(&gl->gl_lru, &dispose);
1379                         atomic_dec(&lru_count);
1380                         freed++;
1381                         continue;
1382                 }
1383
1384                 list_move(&gl->gl_lru, &skipped);
1385         }
1386         list_splice(&skipped, &lru_list);
1387         if (!list_empty(&dispose))
1388                 gfs2_dispose_glock_lru(&dispose);
1389         spin_unlock(&lru_lock);
1390
1391         return freed;
1392 }
1393
1394 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1395                                             struct shrink_control *sc)
1396 {
1397         if (!(sc->gfp_mask & __GFP_FS))
1398                 return SHRINK_STOP;
1399         return gfs2_scan_glock_lru(sc->nr_to_scan);
1400 }
1401
1402 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1403                                              struct shrink_control *sc)
1404 {
1405         return vfs_pressure_ratio(atomic_read(&lru_count));
1406 }
1407
1408 static struct shrinker glock_shrinker = {
1409         .seeks = DEFAULT_SEEKS,
1410         .count_objects = gfs2_glock_shrink_count,
1411         .scan_objects = gfs2_glock_shrink_scan,
1412 };
1413
1414 /**
1415  * examine_bucket - Call a function for glock in a hash bucket
1416  * @examiner: the function
1417  * @sdp: the filesystem
1418  * @bucket: the bucket
1419  *
1420  */
1421
1422 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1423 {
1424         struct gfs2_glock *gl;
1425         struct rhash_head *pos, *next;
1426         const struct bucket_table *tbl;
1427         int i;
1428
1429         rcu_read_lock();
1430         tbl = rht_dereference_rcu(gl_hash_table.tbl, &gl_hash_table);
1431         for (i = 0; i < tbl->size; i++) {
1432                 rht_for_each_entry_safe(gl, pos, next, tbl, i, gl_node) {
1433                         if ((gl->gl_name.ln_sbd == sdp) &&
1434                             lockref_get_not_dead(&gl->gl_lockref))
1435                                 examiner(gl);
1436                 }
1437         }
1438         rcu_read_unlock();
1439         cond_resched();
1440 }
1441
1442 /**
1443  * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1444  * @gl: The glock to thaw
1445  *
1446  */
1447
1448 static void thaw_glock(struct gfs2_glock *gl)
1449 {
1450         if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1451                 goto out;
1452         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1453         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0) {
1454 out:
1455                 gfs2_glock_put(gl);
1456         }
1457 }
1458
1459 /**
1460  * clear_glock - look at a glock and see if we can free it from glock cache
1461  * @gl: the glock to look at
1462  *
1463  */
1464
1465 static void clear_glock(struct gfs2_glock *gl)
1466 {
1467         gfs2_glock_remove_from_lru(gl);
1468
1469         spin_lock(&gl->gl_lockref.lock);
1470         if (gl->gl_state != LM_ST_UNLOCKED)
1471                 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1472         spin_unlock(&gl->gl_lockref.lock);
1473         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1474                 gfs2_glock_put(gl);
1475 }
1476
1477 /**
1478  * gfs2_glock_thaw - Thaw any frozen glocks
1479  * @sdp: The super block
1480  *
1481  */
1482
1483 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1484 {
1485         glock_hash_walk(thaw_glock, sdp);
1486 }
1487
1488 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1489 {
1490         spin_lock(&gl->gl_lockref.lock);
1491         gfs2_dump_glock(seq, gl);
1492         spin_unlock(&gl->gl_lockref.lock);
1493 }
1494
1495 static void dump_glock_func(struct gfs2_glock *gl)
1496 {
1497         dump_glock(NULL, gl);
1498 }
1499
1500 /**
1501  * gfs2_gl_hash_clear - Empty out the glock hash table
1502  * @sdp: the filesystem
1503  * @wait: wait until it's all gone
1504  *
1505  * Called when unmounting the filesystem.
1506  */
1507
1508 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1509 {
1510         set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1511         flush_workqueue(glock_workqueue);
1512         glock_hash_walk(clear_glock, sdp);
1513         flush_workqueue(glock_workqueue);
1514         wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
1515         glock_hash_walk(dump_glock_func, sdp);
1516 }
1517
1518 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1519 {
1520         struct gfs2_glock *gl = ip->i_gl;
1521         int ret;
1522
1523         ret = gfs2_truncatei_resume(ip);
1524         gfs2_assert_withdraw(gl->gl_name.ln_sbd, ret == 0);
1525
1526         spin_lock(&gl->gl_lockref.lock);
1527         clear_bit(GLF_LOCK, &gl->gl_flags);
1528         run_queue(gl, 1);
1529         spin_unlock(&gl->gl_lockref.lock);
1530 }
1531
1532 static const char *state2str(unsigned state)
1533 {
1534         switch(state) {
1535         case LM_ST_UNLOCKED:
1536                 return "UN";
1537         case LM_ST_SHARED:
1538                 return "SH";
1539         case LM_ST_DEFERRED:
1540                 return "DF";
1541         case LM_ST_EXCLUSIVE:
1542                 return "EX";
1543         }
1544         return "??";
1545 }
1546
1547 static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1548 {
1549         char *p = buf;
1550         if (flags & LM_FLAG_TRY)
1551                 *p++ = 't';
1552         if (flags & LM_FLAG_TRY_1CB)
1553                 *p++ = 'T';
1554         if (flags & LM_FLAG_NOEXP)
1555                 *p++ = 'e';
1556         if (flags & LM_FLAG_ANY)
1557                 *p++ = 'A';
1558         if (flags & LM_FLAG_PRIORITY)
1559                 *p++ = 'p';
1560         if (flags & GL_ASYNC)
1561                 *p++ = 'a';
1562         if (flags & GL_EXACT)
1563                 *p++ = 'E';
1564         if (flags & GL_NOCACHE)
1565                 *p++ = 'c';
1566         if (test_bit(HIF_HOLDER, &iflags))
1567                 *p++ = 'H';
1568         if (test_bit(HIF_WAIT, &iflags))
1569                 *p++ = 'W';
1570         if (test_bit(HIF_FIRST, &iflags))
1571                 *p++ = 'F';
1572         *p = 0;
1573         return buf;
1574 }
1575
1576 /**
1577  * dump_holder - print information about a glock holder
1578  * @seq: the seq_file struct
1579  * @gh: the glock holder
1580  *
1581  */
1582
1583 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1584 {
1585         struct task_struct *gh_owner = NULL;
1586         char flags_buf[32];
1587
1588         rcu_read_lock();
1589         if (gh->gh_owner_pid)
1590                 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1591         gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1592                        state2str(gh->gh_state),
1593                        hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1594                        gh->gh_error,
1595                        gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1596                        gh_owner ? gh_owner->comm : "(ended)",
1597                        (void *)gh->gh_ip);
1598         rcu_read_unlock();
1599 }
1600
1601 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1602 {
1603         const unsigned long *gflags = &gl->gl_flags;
1604         char *p = buf;
1605
1606         if (test_bit(GLF_LOCK, gflags))
1607                 *p++ = 'l';
1608         if (test_bit(GLF_DEMOTE, gflags))
1609                 *p++ = 'D';
1610         if (test_bit(GLF_PENDING_DEMOTE, gflags))
1611                 *p++ = 'd';
1612         if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1613                 *p++ = 'p';
1614         if (test_bit(GLF_DIRTY, gflags))
1615                 *p++ = 'y';
1616         if (test_bit(GLF_LFLUSH, gflags))
1617                 *p++ = 'f';
1618         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1619                 *p++ = 'i';
1620         if (test_bit(GLF_REPLY_PENDING, gflags))
1621                 *p++ = 'r';
1622         if (test_bit(GLF_INITIAL, gflags))
1623                 *p++ = 'I';
1624         if (test_bit(GLF_FROZEN, gflags))
1625                 *p++ = 'F';
1626         if (test_bit(GLF_QUEUED, gflags))
1627                 *p++ = 'q';
1628         if (test_bit(GLF_LRU, gflags))
1629                 *p++ = 'L';
1630         if (gl->gl_object)
1631                 *p++ = 'o';
1632         if (test_bit(GLF_BLOCKING, gflags))
1633                 *p++ = 'b';
1634         *p = 0;
1635         return buf;
1636 }
1637
1638 /**
1639  * gfs2_dump_glock - print information about a glock
1640  * @seq: The seq_file struct
1641  * @gl: the glock
1642  *
1643  * The file format is as follows:
1644  * One line per object, capital letters are used to indicate objects
1645  * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1646  * other objects are indented by a single space and follow the glock to
1647  * which they are related. Fields are indicated by lower case letters
1648  * followed by a colon and the field value, except for strings which are in
1649  * [] so that its possible to see if they are composed of spaces for
1650  * example. The field's are n = number (id of the object), f = flags,
1651  * t = type, s = state, r = refcount, e = error, p = pid.
1652  *
1653  */
1654
1655 void gfs2_dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1656 {
1657         const struct gfs2_glock_operations *glops = gl->gl_ops;
1658         unsigned long long dtime;
1659         const struct gfs2_holder *gh;
1660         char gflags_buf[32];
1661
1662         dtime = jiffies - gl->gl_demote_time;
1663         dtime *= 1000000/HZ; /* demote time in uSec */
1664         if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1665                 dtime = 0;
1666         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",
1667                   state2str(gl->gl_state),
1668                   gl->gl_name.ln_type,
1669                   (unsigned long long)gl->gl_name.ln_number,
1670                   gflags2str(gflags_buf, gl),
1671                   state2str(gl->gl_target),
1672                   state2str(gl->gl_demote_state), dtime,
1673                   atomic_read(&gl->gl_ail_count),
1674                   atomic_read(&gl->gl_revokes),
1675                   (int)gl->gl_lockref.count, gl->gl_hold_time);
1676
1677         list_for_each_entry(gh, &gl->gl_holders, gh_list)
1678                 dump_holder(seq, gh);
1679
1680         if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1681                 glops->go_dump(seq, gl);
1682 }
1683
1684 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
1685 {
1686         struct gfs2_glock *gl = iter_ptr;
1687
1688         seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
1689                    gl->gl_name.ln_type,
1690                    (unsigned long long)gl->gl_name.ln_number,
1691                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
1692                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
1693                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
1694                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
1695                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
1696                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
1697                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
1698                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
1699         return 0;
1700 }
1701
1702 static const char *gfs2_gltype[] = {
1703         "type",
1704         "reserved",
1705         "nondisk",
1706         "inode",
1707         "rgrp",
1708         "meta",
1709         "iopen",
1710         "flock",
1711         "plock",
1712         "quota",
1713         "journal",
1714 };
1715
1716 static const char *gfs2_stype[] = {
1717         [GFS2_LKS_SRTT]         = "srtt",
1718         [GFS2_LKS_SRTTVAR]      = "srttvar",
1719         [GFS2_LKS_SRTTB]        = "srttb",
1720         [GFS2_LKS_SRTTVARB]     = "srttvarb",
1721         [GFS2_LKS_SIRT]         = "sirt",
1722         [GFS2_LKS_SIRTVAR]      = "sirtvar",
1723         [GFS2_LKS_DCOUNT]       = "dlm",
1724         [GFS2_LKS_QCOUNT]       = "queue",
1725 };
1726
1727 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1728
1729 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
1730 {
1731         struct gfs2_sbd *sdp = seq->private;
1732         loff_t pos = *(loff_t *)iter_ptr;
1733         unsigned index = pos >> 3;
1734         unsigned subindex = pos & 0x07;
1735         int i;
1736
1737         if (index == 0 && subindex != 0)
1738                 return 0;
1739
1740         seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
1741                    (index == 0) ? "cpu": gfs2_stype[subindex]);
1742
1743         for_each_possible_cpu(i) {
1744                 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
1745
1746                 if (index == 0)
1747                         seq_printf(seq, " %15u", i);
1748                 else
1749                         seq_printf(seq, " %15llu", (unsigned long long)lkstats->
1750                                    lkstats[index - 1].stats[subindex]);
1751         }
1752         seq_putc(seq, '\n');
1753         return 0;
1754 }
1755
1756 int __init gfs2_glock_init(void)
1757 {
1758         int ret;
1759
1760         ret = rhashtable_init(&gl_hash_table, &ht_parms);
1761         if (ret < 0)
1762                 return ret;
1763
1764         glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1765                                           WQ_HIGHPRI | WQ_FREEZABLE, 0);
1766         if (!glock_workqueue) {
1767                 rhashtable_destroy(&gl_hash_table);
1768                 return -ENOMEM;
1769         }
1770         gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1771                                                 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1772                                                 0);
1773         if (!gfs2_delete_workqueue) {
1774                 destroy_workqueue(glock_workqueue);
1775                 rhashtable_destroy(&gl_hash_table);
1776                 return -ENOMEM;
1777         }
1778
1779         register_shrinker(&glock_shrinker);
1780
1781         return 0;
1782 }
1783
1784 void gfs2_glock_exit(void)
1785 {
1786         unregister_shrinker(&glock_shrinker);
1787         rhashtable_destroy(&gl_hash_table);
1788         destroy_workqueue(glock_workqueue);
1789         destroy_workqueue(gfs2_delete_workqueue);
1790 }
1791
1792 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1793 {
1794         do {
1795                 gi->gl = rhashtable_walk_next(&gi->hti);
1796                 if (IS_ERR(gi->gl)) {
1797                         if (PTR_ERR(gi->gl) == -EAGAIN)
1798                                 continue;
1799                         gi->gl = NULL;
1800                 }
1801         /* Skip entries for other sb and dead entries */
1802         } while ((gi->gl) && ((gi->sdp != gi->gl->gl_name.ln_sbd) ||
1803                               __lockref_is_dead(&gi->gl->gl_lockref)));
1804 }
1805
1806 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1807 {
1808         struct gfs2_glock_iter *gi = seq->private;
1809         loff_t n = *pos;
1810         int ret;
1811
1812         if (gi->last_pos <= *pos)
1813                 n = (*pos - gi->last_pos);
1814
1815         ret = rhashtable_walk_start(&gi->hti);
1816         if (ret)
1817                 return NULL;
1818
1819         do {
1820                 gfs2_glock_iter_next(gi);
1821         } while (gi->gl && n--);
1822
1823         gi->last_pos = *pos;
1824         return gi->gl;
1825 }
1826
1827 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1828                                  loff_t *pos)
1829 {
1830         struct gfs2_glock_iter *gi = seq->private;
1831
1832         (*pos)++;
1833         gi->last_pos = *pos;
1834         gfs2_glock_iter_next(gi);
1835         return gi->gl;
1836 }
1837
1838 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1839 {
1840         struct gfs2_glock_iter *gi = seq->private;
1841
1842         gi->gl = NULL;
1843         rhashtable_walk_stop(&gi->hti);
1844 }
1845
1846 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1847 {
1848         dump_glock(seq, iter_ptr);
1849         return 0;
1850 }
1851
1852 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
1853 {
1854         preempt_disable();
1855         if (*pos >= GFS2_NR_SBSTATS)
1856                 return NULL;
1857         return pos;
1858 }
1859
1860 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
1861                                    loff_t *pos)
1862 {
1863         (*pos)++;
1864         if (*pos >= GFS2_NR_SBSTATS)
1865                 return NULL;
1866         return pos;
1867 }
1868
1869 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
1870 {
1871         preempt_enable();
1872 }
1873
1874 static const struct seq_operations gfs2_glock_seq_ops = {
1875         .start = gfs2_glock_seq_start,
1876         .next  = gfs2_glock_seq_next,
1877         .stop  = gfs2_glock_seq_stop,
1878         .show  = gfs2_glock_seq_show,
1879 };
1880
1881 static const struct seq_operations gfs2_glstats_seq_ops = {
1882         .start = gfs2_glock_seq_start,
1883         .next  = gfs2_glock_seq_next,
1884         .stop  = gfs2_glock_seq_stop,
1885         .show  = gfs2_glstats_seq_show,
1886 };
1887
1888 static const struct seq_operations gfs2_sbstats_seq_ops = {
1889         .start = gfs2_sbstats_seq_start,
1890         .next  = gfs2_sbstats_seq_next,
1891         .stop  = gfs2_sbstats_seq_stop,
1892         .show  = gfs2_sbstats_seq_show,
1893 };
1894
1895 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
1896
1897 static int gfs2_glocks_open(struct inode *inode, struct file *file)
1898 {
1899         int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1900                                    sizeof(struct gfs2_glock_iter));
1901         if (ret == 0) {
1902                 struct seq_file *seq = file->private_data;
1903                 struct gfs2_glock_iter *gi = seq->private;
1904
1905                 gi->sdp = inode->i_private;
1906                 gi->last_pos = 0;
1907                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
1908                 if (seq->buf)
1909                         seq->size = GFS2_SEQ_GOODSIZE;
1910                 gi->gl = NULL;
1911                 ret = rhashtable_walk_init(&gl_hash_table, &gi->hti);
1912         }
1913         return ret;
1914 }
1915
1916 static int gfs2_glocks_release(struct inode *inode, struct file *file)
1917 {
1918         struct seq_file *seq = file->private_data;
1919         struct gfs2_glock_iter *gi = seq->private;
1920
1921         gi->gl = NULL;
1922         rhashtable_walk_exit(&gi->hti);
1923         return seq_release_private(inode, file);
1924 }
1925
1926 static int gfs2_glstats_open(struct inode *inode, struct file *file)
1927 {
1928         int ret = seq_open_private(file, &gfs2_glstats_seq_ops,
1929                                    sizeof(struct gfs2_glock_iter));
1930         if (ret == 0) {
1931                 struct seq_file *seq = file->private_data;
1932                 struct gfs2_glock_iter *gi = seq->private;
1933                 gi->sdp = inode->i_private;
1934                 gi->last_pos = 0;
1935                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
1936                 if (seq->buf)
1937                         seq->size = GFS2_SEQ_GOODSIZE;
1938                 gi->gl = NULL;
1939                 ret = rhashtable_walk_init(&gl_hash_table, &gi->hti);
1940         }
1941         return ret;
1942 }
1943
1944 static int gfs2_sbstats_open(struct inode *inode, struct file *file)
1945 {
1946         int ret = seq_open(file, &gfs2_sbstats_seq_ops);
1947         if (ret == 0) {
1948                 struct seq_file *seq = file->private_data;
1949                 seq->private = inode->i_private;  /* sdp */
1950         }
1951         return ret;
1952 }
1953
1954 static const struct file_operations gfs2_glocks_fops = {
1955         .owner   = THIS_MODULE,
1956         .open    = gfs2_glocks_open,
1957         .read    = seq_read,
1958         .llseek  = seq_lseek,
1959         .release = gfs2_glocks_release,
1960 };
1961
1962 static const struct file_operations gfs2_glstats_fops = {
1963         .owner   = THIS_MODULE,
1964         .open    = gfs2_glstats_open,
1965         .read    = seq_read,
1966         .llseek  = seq_lseek,
1967         .release = gfs2_glocks_release,
1968 };
1969
1970 static const struct file_operations gfs2_sbstats_fops = {
1971         .owner   = THIS_MODULE,
1972         .open    = gfs2_sbstats_open,
1973         .read    = seq_read,
1974         .llseek  = seq_lseek,
1975         .release = seq_release,
1976 };
1977
1978 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
1979 {
1980         struct dentry *dent;
1981
1982         dent = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
1983         if (IS_ERR_OR_NULL(dent))
1984                 goto fail;
1985         sdp->debugfs_dir = dent;
1986
1987         dent = debugfs_create_file("glocks",
1988                                    S_IFREG | S_IRUGO,
1989                                    sdp->debugfs_dir, sdp,
1990                                    &gfs2_glocks_fops);
1991         if (IS_ERR_OR_NULL(dent))
1992                 goto fail;
1993         sdp->debugfs_dentry_glocks = dent;
1994
1995         dent = debugfs_create_file("glstats",
1996                                    S_IFREG | S_IRUGO,
1997                                    sdp->debugfs_dir, sdp,
1998                                    &gfs2_glstats_fops);
1999         if (IS_ERR_OR_NULL(dent))
2000                 goto fail;
2001         sdp->debugfs_dentry_glstats = dent;
2002
2003         dent = debugfs_create_file("sbstats",
2004                                    S_IFREG | S_IRUGO,
2005                                    sdp->debugfs_dir, sdp,
2006                                    &gfs2_sbstats_fops);
2007         if (IS_ERR_OR_NULL(dent))
2008                 goto fail;
2009         sdp->debugfs_dentry_sbstats = dent;
2010
2011         return 0;
2012 fail:
2013         gfs2_delete_debugfs_file(sdp);
2014         return dent ? PTR_ERR(dent) : -ENOMEM;
2015 }
2016
2017 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2018 {
2019         if (sdp->debugfs_dir) {
2020                 if (sdp->debugfs_dentry_glocks) {
2021                         debugfs_remove(sdp->debugfs_dentry_glocks);
2022                         sdp->debugfs_dentry_glocks = NULL;
2023                 }
2024                 if (sdp->debugfs_dentry_glstats) {
2025                         debugfs_remove(sdp->debugfs_dentry_glstats);
2026                         sdp->debugfs_dentry_glstats = NULL;
2027                 }
2028                 if (sdp->debugfs_dentry_sbstats) {
2029                         debugfs_remove(sdp->debugfs_dentry_sbstats);
2030                         sdp->debugfs_dentry_sbstats = NULL;
2031                 }
2032                 debugfs_remove(sdp->debugfs_dir);
2033                 sdp->debugfs_dir = NULL;
2034         }
2035 }
2036
2037 int gfs2_register_debugfs(void)
2038 {
2039         gfs2_root = debugfs_create_dir("gfs2", NULL);
2040         if (IS_ERR(gfs2_root))
2041                 return PTR_ERR(gfs2_root);
2042         return gfs2_root ? 0 : -ENOMEM;
2043 }
2044
2045 void gfs2_unregister_debugfs(void)
2046 {
2047         debugfs_remove(gfs2_root);
2048         gfs2_root = NULL;
2049 }