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