linux/fs/gfs2/glock.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   4 * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
   5 */
   6
   7#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   8
   9#include <linux/sched.h>
  10#include <linux/slab.h>
  11#include <linux/spinlock.h>
  12#include <linux/buffer_head.h>
  13#include <linux/delay.h>
  14#include <linux/sort.h>
  15#include <linux/hash.h>
  16#include <linux/jhash.h>
  17#include <linux/kallsyms.h>
  18#include <linux/gfs2_ondisk.h>
  19#include <linux/list.h>
  20#include <linux/wait.h>
  21#include <linux/module.h>
  22#include <linux/uaccess.h>
  23#include <linux/seq_file.h>
  24#include <linux/debugfs.h>
  25#include <linux/kthread.h>
  26#include <linux/freezer.h>
  27#include <linux/workqueue.h>
  28#include <linux/jiffies.h>
  29#include <linux/rcupdate.h>
  30#include <linux/rculist_bl.h>
  31#include <linux/bit_spinlock.h>
  32#include <linux/percpu.h>
  33#include <linux/list_sort.h>
  34#include <linux/lockref.h>
  35#include <linux/rhashtable.h>
  36
  37#include "gfs2.h"
  38#include "incore.h"
  39#include "glock.h"
  40#include "glops.h"
  41#include "inode.h"
  42#include "lops.h"
  43#include "meta_io.h"
  44#include "quota.h"
  45#include "super.h"
  46#include "util.h"
  47#include "bmap.h"
  48#define CREATE_TRACE_POINTS
  49#include "trace_gfs2.h"
  50
  51struct gfs2_glock_iter {
  52        struct gfs2_sbd *sdp;           /* incore superblock           */
  53        struct rhashtable_iter hti;     /* rhashtable iterator         */
  54        struct gfs2_glock *gl;          /* current glock struct        */
  55        loff_t last_pos;                /* last position               */
  56};
  57
  58typedef void (*glock_examiner) (struct gfs2_glock * gl);
  59
  60static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
  61
  62static struct dentry *gfs2_root;
  63static struct workqueue_struct *glock_workqueue;
  64struct workqueue_struct *gfs2_delete_workqueue;
  65static LIST_HEAD(lru_list);
  66static atomic_t lru_count = ATOMIC_INIT(0);
  67static DEFINE_SPINLOCK(lru_lock);
  68
  69#define GFS2_GL_HASH_SHIFT      15
  70#define GFS2_GL_HASH_SIZE       BIT(GFS2_GL_HASH_SHIFT)
  71
  72static const struct rhashtable_params ht_parms = {
  73        .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4,
  74        .key_len = offsetofend(struct lm_lockname, ln_type),
  75        .key_offset = offsetof(struct gfs2_glock, gl_name),
  76        .head_offset = offsetof(struct gfs2_glock, gl_node),
  77};
  78
  79static struct rhashtable gl_hash_table;
  80
  81#define GLOCK_WAIT_TABLE_BITS 12
  82#define GLOCK_WAIT_TABLE_SIZE (1 << GLOCK_WAIT_TABLE_BITS)
  83static wait_queue_head_t glock_wait_table[GLOCK_WAIT_TABLE_SIZE] __cacheline_aligned;
  84
  85struct wait_glock_queue {
  86        struct lm_lockname *name;
  87        wait_queue_entry_t wait;
  88};
  89
  90static int glock_wake_function(wait_queue_entry_t *wait, unsigned int mode,
  91                               int sync, void *key)
  92{
  93        struct wait_glock_queue *wait_glock =
  94                container_of(wait, struct wait_glock_queue, wait);
  95        struct lm_lockname *wait_name = wait_glock->name;
  96        struct lm_lockname *wake_name = key;
  97
  98        if (wake_name->ln_sbd != wait_name->ln_sbd ||
  99            wake_name->ln_number != wait_name->ln_number ||
 100            wake_name->ln_type != wait_name->ln_type)
 101                return 0;
 102        return autoremove_wake_function(wait, mode, sync, key);
 103}
 104
 105static wait_queue_head_t *glock_waitqueue(struct lm_lockname *name)
 106{
 107        u32 hash = jhash2((u32 *)name, ht_parms.key_len / 4, 0);
 108
 109        return glock_wait_table + hash_32(hash, GLOCK_WAIT_TABLE_BITS);
 110}
 111
 112/**
 113 * wake_up_glock  -  Wake up waiters on a glock
 114 * @gl: the glock
 115 */
 116static void wake_up_glock(struct gfs2_glock *gl)
 117{
 118        wait_queue_head_t *wq = glock_waitqueue(&gl->gl_name);
 119
 120        if (waitqueue_active(wq))
 121                __wake_up(wq, TASK_NORMAL, 1, &gl->gl_name);
 122}
 123
 124static void gfs2_glock_dealloc(struct rcu_head *rcu)
 125{
 126        struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
 127
 128        kfree(gl->gl_lksb.sb_lvbptr);
 129        if (gl->gl_ops->go_flags & GLOF_ASPACE)
 130                kmem_cache_free(gfs2_glock_aspace_cachep, gl);
 131        else
 132                kmem_cache_free(gfs2_glock_cachep, gl);
 133}
 134
 135/**
 136 * glock_blocked_by_withdraw - determine if we can still use a glock
 137 * @gl: the glock
 138 *
 139 * We need to allow some glocks to be enqueued, dequeued, promoted, and demoted
 140 * when we're withdrawn. For example, to maintain metadata integrity, we should
 141 * disallow the use of inode and rgrp glocks when withdrawn. Other glocks, like
 142 * iopen or the transaction glocks may be safely used because none of their
 143 * metadata goes through the journal. So in general, we should disallow all
 144 * glocks that are journaled, and allow all the others. One exception is:
 145 * we need to allow our active journal to be promoted and demoted so others
 146 * may recover it and we can reacquire it when they're done.
 147 */
 148static bool glock_blocked_by_withdraw(struct gfs2_glock *gl)
 149{
 150        struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 151
 152        if (likely(!gfs2_withdrawn(sdp)))
 153                return false;
 154        if (gl->gl_ops->go_flags & GLOF_NONDISK)
 155                return false;
 156        if (!sdp->sd_jdesc ||
 157            gl->gl_name.ln_number == sdp->sd_jdesc->jd_no_addr)
 158                return false;
 159        return true;
 160}
 161
 162void gfs2_glock_free(struct gfs2_glock *gl)
 163{
 164        struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 165
 166        gfs2_glock_assert_withdraw(gl, atomic_read(&gl->gl_revokes) == 0);
 167        rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
 168        smp_mb();
 169        wake_up_glock(gl);
 170        call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
 171        if (atomic_dec_and_test(&sdp->sd_glock_disposal))
 172                wake_up(&sdp->sd_glock_wait);
 173}
 174
 175/**
 176 * gfs2_glock_hold() - increment reference count on glock
 177 * @gl: The glock to hold
 178 *
 179 */
 180
 181void gfs2_glock_hold(struct gfs2_glock *gl)
 182{
 183        GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
 184        lockref_get(&gl->gl_lockref);
 185}
 186
 187/**
 188 * demote_ok - Check to see if it's ok to unlock a glock
 189 * @gl: the glock
 190 *
 191 * Returns: 1 if it's ok
 192 */
 193
 194static int demote_ok(const struct gfs2_glock *gl)
 195{
 196        const struct gfs2_glock_operations *glops = gl->gl_ops;
 197
 198        if (gl->gl_state == LM_ST_UNLOCKED)
 199                return 0;
 200        if (!list_empty(&gl->gl_holders))
 201                return 0;
 202        if (glops->go_demote_ok)
 203                return glops->go_demote_ok(gl);
 204        return 1;
 205}
 206
 207
 208void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
 209{
 210        if (!(gl->gl_ops->go_flags & GLOF_LRU))
 211                return;
 212
 213        spin_lock(&lru_lock);
 214
 215        list_move_tail(&gl->gl_lru, &lru_list);
 216
 217        if (!test_bit(GLF_LRU, &gl->gl_flags)) {
 218                set_bit(GLF_LRU, &gl->gl_flags);
 219                atomic_inc(&lru_count);
 220        }
 221
 222        spin_unlock(&lru_lock);
 223}
 224
 225static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
 226{
 227        if (!(gl->gl_ops->go_flags & GLOF_LRU))
 228                return;
 229
 230        spin_lock(&lru_lock);
 231        if (test_bit(GLF_LRU, &gl->gl_flags)) {
 232                list_del_init(&gl->gl_lru);
 233                atomic_dec(&lru_count);
 234                clear_bit(GLF_LRU, &gl->gl_flags);
 235        }
 236        spin_unlock(&lru_lock);
 237}
 238
 239/*
 240 * Enqueue the glock on the work queue.  Passes one glock reference on to the
 241 * work queue.
 242 */
 243static void __gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
 244        if (!queue_delayed_work(glock_workqueue, &gl->gl_work, delay)) {
 245                /*
 246                 * We are holding the lockref spinlock, and the work was still
 247                 * queued above.  The queued work (glock_work_func) takes that
 248                 * spinlock before dropping its glock reference(s), so it
 249                 * cannot have dropped them in the meantime.
 250                 */
 251                GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2);
 252                gl->gl_lockref.count--;
 253        }
 254}
 255
 256static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
 257        spin_lock(&gl->gl_lockref.lock);
 258        __gfs2_glock_queue_work(gl, delay);
 259        spin_unlock(&gl->gl_lockref.lock);
 260}
 261
 262static void __gfs2_glock_put(struct gfs2_glock *gl)
 263{
 264        struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 265        struct address_space *mapping = gfs2_glock2aspace(gl);
 266
 267        lockref_mark_dead(&gl->gl_lockref);
 268
 269        gfs2_glock_remove_from_lru(gl);
 270        spin_unlock(&gl->gl_lockref.lock);
 271        GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
 272        if (mapping) {
 273                truncate_inode_pages_final(mapping);
 274                if (!gfs2_withdrawn(sdp))
 275                        GLOCK_BUG_ON(gl, !mapping_empty(mapping));
 276        }
 277        trace_gfs2_glock_put(gl);
 278        sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
 279}
 280
 281/*
 282 * Cause the glock to be put in work queue context.
 283 */
 284void gfs2_glock_queue_put(struct gfs2_glock *gl)
 285{
 286        gfs2_glock_queue_work(gl, 0);
 287}
 288
 289/**
 290 * gfs2_glock_put() - Decrement reference count on glock
 291 * @gl: The glock to put
 292 *
 293 */
 294
 295void gfs2_glock_put(struct gfs2_glock *gl)
 296{
 297        if (lockref_put_or_lock(&gl->gl_lockref))
 298                return;
 299
 300        __gfs2_glock_put(gl);
 301}
 302
 303/**
 304 * may_grant - check if its ok to grant a new lock
 305 * @gl: The glock
 306 * @gh: The lock request which we wish to grant
 307 *
 308 * Returns: true if its ok to grant the lock
 309 */
 310
 311static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
 312{
 313        const struct gfs2_holder *gh_head = list_first_entry(&gl->gl_holders, const struct gfs2_holder, gh_list);
 314
 315        if (gh != gh_head) {
 316                /**
 317                 * Here we make a special exception to grant holders who agree
 318                 * to share the EX lock with other holders who also have the
 319                 * bit set. If the original holder has the LM_FLAG_NODE_SCOPE bit
 320                 * is set, we grant more holders with the bit set.
 321                 */
 322                if (gh_head->gh_state == LM_ST_EXCLUSIVE &&
 323                    (gh_head->gh_flags & LM_FLAG_NODE_SCOPE) &&
 324                    gh->gh_state == LM_ST_EXCLUSIVE &&
 325                    (gh->gh_flags & LM_FLAG_NODE_SCOPE))
 326                        return 1;
 327                if ((gh->gh_state == LM_ST_EXCLUSIVE ||
 328                     gh_head->gh_state == LM_ST_EXCLUSIVE))
 329                        return 0;
 330        }
 331        if (gl->gl_state == gh->gh_state)
 332                return 1;
 333        if (gh->gh_flags & GL_EXACT)
 334                return 0;
 335        if (gl->gl_state == LM_ST_EXCLUSIVE) {
 336                if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
 337                        return 1;
 338                if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
 339                        return 1;
 340        }
 341        if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
 342                return 1;
 343        return 0;
 344}
 345
 346static void gfs2_holder_wake(struct gfs2_holder *gh)
 347{
 348        clear_bit(HIF_WAIT, &gh->gh_iflags);
 349        smp_mb__after_atomic();
 350        wake_up_bit(&gh->gh_iflags, HIF_WAIT);
 351        if (gh->gh_flags & GL_ASYNC) {
 352                struct gfs2_sbd *sdp = gh->gh_gl->gl_name.ln_sbd;
 353
 354                wake_up(&sdp->sd_async_glock_wait);
 355        }
 356}
 357
 358/**
 359 * do_error - Something unexpected has happened during a lock request
 360 * @gl: The glock
 361 * @ret: The status from the DLM
 362 */
 363
 364static void do_error(struct gfs2_glock *gl, const int ret)
 365{
 366        struct gfs2_holder *gh, *tmp;
 367
 368        list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
 369                if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 370                        continue;
 371                if (ret & LM_OUT_ERROR)
 372                        gh->gh_error = -EIO;
 373                else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
 374                        gh->gh_error = GLR_TRYFAILED;
 375                else
 376                        continue;
 377                list_del_init(&gh->gh_list);
 378                trace_gfs2_glock_queue(gh, 0);
 379                gfs2_holder_wake(gh);
 380        }
 381}
 382
 383/**
 384 * do_promote - promote as many requests as possible on the current queue
 385 * @gl: The glock
 386 * 
 387 * Returns: 1 if there is a blocked holder at the head of the list, or 2
 388 *          if a type specific operation is underway.
 389 */
 390
 391static int do_promote(struct gfs2_glock *gl)
 392__releases(&gl->gl_lockref.lock)
 393__acquires(&gl->gl_lockref.lock)
 394{
 395        const struct gfs2_glock_operations *glops = gl->gl_ops;
 396        struct gfs2_holder *gh, *tmp;
 397        int ret;
 398
 399restart:
 400        list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
 401                if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 402                        continue;
 403                if (may_grant(gl, gh)) {
 404                        if (gh->gh_list.prev == &gl->gl_holders &&
 405                            glops->go_lock) {
 406                                spin_unlock(&gl->gl_lockref.lock);
 407                                /* FIXME: eliminate this eventually */
 408                                ret = glops->go_lock(gh);
 409                                spin_lock(&gl->gl_lockref.lock);
 410                                if (ret) {
 411                                        if (ret == 1)
 412                                                return 2;
 413                                        gh->gh_error = ret;
 414                                        list_del_init(&gh->gh_list);
 415                                        trace_gfs2_glock_queue(gh, 0);
 416                                        gfs2_holder_wake(gh);
 417                                        goto restart;
 418                                }
 419                                set_bit(HIF_HOLDER, &gh->gh_iflags);
 420                                trace_gfs2_promote(gh, 1);
 421                                gfs2_holder_wake(gh);
 422                                goto restart;
 423                        }
 424                        set_bit(HIF_HOLDER, &gh->gh_iflags);
 425                        trace_gfs2_promote(gh, 0);
 426                        gfs2_holder_wake(gh);
 427                        continue;
 428                }
 429                if (gh->gh_list.prev == &gl->gl_holders)
 430                        return 1;
 431                do_error(gl, 0);
 432                break;
 433        }
 434        return 0;
 435}
 436
 437/**
 438 * find_first_waiter - find the first gh that's waiting for the glock
 439 * @gl: the glock
 440 */
 441
 442static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
 443{
 444        struct gfs2_holder *gh;
 445
 446        list_for_each_entry(gh, &gl->gl_holders, gh_list) {
 447                if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
 448                        return gh;
 449        }
 450        return NULL;
 451}
 452
 453/**
 454 * state_change - record that the glock is now in a different state
 455 * @gl: the glock
 456 * @new_state: the new state
 457 */
 458
 459static void state_change(struct gfs2_glock *gl, unsigned int new_state)
 460{
 461        int held1, held2;
 462
 463        held1 = (gl->gl_state != LM_ST_UNLOCKED);
 464        held2 = (new_state != LM_ST_UNLOCKED);
 465
 466        if (held1 != held2) {
 467                GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
 468                if (held2)
 469                        gl->gl_lockref.count++;
 470                else
 471                        gl->gl_lockref.count--;
 472        }
 473        if (new_state != gl->gl_target)
 474                /* shorten our minimum hold time */
 475                gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
 476                                       GL_GLOCK_MIN_HOLD);
 477        gl->gl_state = new_state;
 478        gl->gl_tchange = jiffies;
 479}
 480
 481static void gfs2_set_demote(struct gfs2_glock *gl)
 482{
 483        struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 484
 485        set_bit(GLF_DEMOTE, &gl->gl_flags);
 486        smp_mb();
 487        wake_up(&sdp->sd_async_glock_wait);
 488}
 489
 490static void gfs2_demote_wake(struct gfs2_glock *gl)
 491{
 492        gl->gl_demote_state = LM_ST_EXCLUSIVE;
 493        clear_bit(GLF_DEMOTE, &gl->gl_flags);
 494        smp_mb__after_atomic();
 495        wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
 496}
 497
 498/**
 499 * finish_xmote - The DLM has replied to one of our lock requests
 500 * @gl: The glock
 501 * @ret: The status from the DLM
 502 *
 503 */
 504
 505static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
 506{
 507        const struct gfs2_glock_operations *glops = gl->gl_ops;
 508        struct gfs2_holder *gh;
 509        unsigned state = ret & LM_OUT_ST_MASK;
 510        int rv;
 511
 512        spin_lock(&gl->gl_lockref.lock);
 513        trace_gfs2_glock_state_change(gl, state);
 514        state_change(gl, state);
 515        gh = find_first_waiter(gl);
 516
 517        /* Demote to UN request arrived during demote to SH or DF */
 518        if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
 519            state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
 520                gl->gl_target = LM_ST_UNLOCKED;
 521
 522        /* Check for state != intended state */
 523        if (unlikely(state != gl->gl_target)) {
 524                if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
 525                        /* move to back of queue and try next entry */
 526                        if (ret & LM_OUT_CANCELED) {
 527                                if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
 528                                        list_move_tail(&gh->gh_list, &gl->gl_holders);
 529                                gh = find_first_waiter(gl);
 530                                gl->gl_target = gh->gh_state;
 531                                goto retry;
 532                        }
 533                        /* Some error or failed "try lock" - report it */
 534                        if ((ret & LM_OUT_ERROR) ||
 535                            (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
 536                                gl->gl_target = gl->gl_state;
 537                                do_error(gl, ret);
 538                                goto out;
 539                        }
 540                }
 541                switch(state) {
 542                /* Unlocked due to conversion deadlock, try again */
 543                case LM_ST_UNLOCKED:
 544retry:
 545                        do_xmote(gl, gh, gl->gl_target);
 546                        break;
 547                /* Conversion fails, unlock and try again */
 548                case LM_ST_SHARED:
 549                case LM_ST_DEFERRED:
 550                        do_xmote(gl, gh, LM_ST_UNLOCKED);
 551                        break;
 552                default: /* Everything else */
 553                        fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n",
 554                               gl->gl_target, state);
 555                        GLOCK_BUG_ON(gl, 1);
 556                }
 557                spin_unlock(&gl->gl_lockref.lock);
 558                return;
 559        }
 560
 561        /* Fast path - we got what we asked for */
 562        if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
 563                gfs2_demote_wake(gl);
 564        if (state != LM_ST_UNLOCKED) {
 565                if (glops->go_xmote_bh) {
 566                        spin_unlock(&gl->gl_lockref.lock);
 567                        rv = glops->go_xmote_bh(gl);
 568                        spin_lock(&gl->gl_lockref.lock);
 569                        if (rv) {
 570                                do_error(gl, rv);
 571                                goto out;
 572                        }
 573                }
 574                rv = do_promote(gl);
 575                if (rv == 2)
 576                        goto out_locked;
 577        }
 578out:
 579        clear_bit(GLF_LOCK, &gl->gl_flags);
 580out_locked:
 581        spin_unlock(&gl->gl_lockref.lock);
 582}
 583
 584static bool is_system_glock(struct gfs2_glock *gl)
 585{
 586        struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 587        struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 588
 589        if (gl == m_ip->i_gl)
 590                return true;
 591        return false;
 592}
 593
 594/**
 595 * do_xmote - Calls the DLM to change the state of a lock
 596 * @gl: The lock state
 597 * @gh: The holder (only for promotes)
 598 * @target: The target lock state
 599 *
 600 */
 601
 602static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
 603__releases(&gl->gl_lockref.lock)
 604__acquires(&gl->gl_lockref.lock)
 605{
 606        const struct gfs2_glock_operations *glops = gl->gl_ops;
 607        struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 608        unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
 609        int ret;
 610
 611        if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) &&
 612            gh && !(gh->gh_flags & LM_FLAG_NOEXP))
 613                return;
 614        lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
 615                      LM_FLAG_PRIORITY);
 616        GLOCK_BUG_ON(gl, gl->gl_state == target);
 617        GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
 618        if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
 619            glops->go_inval) {
 620                /*
 621                 * If another process is already doing the invalidate, let that
 622                 * finish first.  The glock state machine will get back to this
 623                 * holder again later.
 624                 */
 625                if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
 626                                     &gl->gl_flags))
 627                        return;
 628                do_error(gl, 0); /* Fail queued try locks */
 629        }
 630        gl->gl_req = target;
 631        set_bit(GLF_BLOCKING, &gl->gl_flags);
 632        if ((gl->gl_req == LM_ST_UNLOCKED) ||
 633            (gl->gl_state == LM_ST_EXCLUSIVE) ||
 634            (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
 635                clear_bit(GLF_BLOCKING, &gl->gl_flags);
 636        spin_unlock(&gl->gl_lockref.lock);
 637        if (glops->go_sync) {
 638                ret = glops->go_sync(gl);
 639                /* If we had a problem syncing (due to io errors or whatever,
 640                 * we should not invalidate the metadata or tell dlm to
 641                 * release the glock to other nodes.
 642                 */
 643                if (ret) {
 644                        if (cmpxchg(&sdp->sd_log_error, 0, ret)) {
 645                                fs_err(sdp, "Error %d syncing glock \n", ret);
 646                                gfs2_dump_glock(NULL, gl, true);
 647                        }
 648                        goto skip_inval;
 649                }
 650        }
 651        if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) {
 652                /*
 653                 * The call to go_sync should have cleared out the ail list.
 654                 * If there are still items, we have a problem. We ought to
 655                 * withdraw, but we can't because the withdraw code also uses
 656                 * glocks. Warn about the error, dump the glock, then fall
 657                 * through and wait for logd to do the withdraw for us.
 658                 */
 659                if ((atomic_read(&gl->gl_ail_count) != 0) &&
 660                    (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) {
 661                        gfs2_glock_assert_warn(gl,
 662                                               !atomic_read(&gl->gl_ail_count));
 663                        gfs2_dump_glock(NULL, gl, true);
 664                }
 665                glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
 666                clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 667        }
 668
 669skip_inval:
 670        gfs2_glock_hold(gl);
 671        /*
 672         * Check for an error encountered since we called go_sync and go_inval.
 673         * If so, we can't withdraw from the glock code because the withdraw
 674         * code itself uses glocks (see function signal_our_withdraw) to
 675         * change the mount to read-only. Most importantly, we must not call
 676         * dlm to unlock the glock until the journal is in a known good state
 677         * (after journal replay) otherwise other nodes may use the object
 678         * (rgrp or dinode) and then later, journal replay will corrupt the
 679         * file system. The best we can do here is wait for the logd daemon
 680         * to see sd_log_error and withdraw, and in the meantime, requeue the
 681         * work for later.
 682         *
 683         * We make a special exception for some system glocks, such as the
 684         * system statfs inode glock, which needs to be granted before the
 685         * gfs2_quotad daemon can exit, and that exit needs to finish before
 686         * we can unmount the withdrawn file system.
 687         *
 688         * However, if we're just unlocking the lock (say, for unmount, when
 689         * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
 690         * then it's okay to tell dlm to unlock it.
 691         */
 692        if (unlikely(sdp->sd_log_error && !gfs2_withdrawn(sdp)))
 693                gfs2_withdraw_delayed(sdp);
 694        if (glock_blocked_by_withdraw(gl) &&
 695            (target != LM_ST_UNLOCKED ||
 696             test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags))) {
 697                if (!is_system_glock(gl)) {
 698                        gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD);
 699                        goto out;
 700                } else {
 701                        clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 702                }
 703        }
 704
 705        if (sdp->sd_lockstruct.ls_ops->lm_lock) {
 706                /* lock_dlm */
 707                ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
 708                if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
 709                    target == LM_ST_UNLOCKED &&
 710                    test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
 711                        finish_xmote(gl, target);
 712                        gfs2_glock_queue_work(gl, 0);
 713                } else if (ret) {
 714                        fs_err(sdp, "lm_lock ret %d\n", ret);
 715                        GLOCK_BUG_ON(gl, !gfs2_withdrawn(sdp));
 716                }
 717        } else { /* lock_nolock */
 718                finish_xmote(gl, target);
 719                gfs2_glock_queue_work(gl, 0);
 720        }
 721out:
 722        spin_lock(&gl->gl_lockref.lock);
 723}
 724
 725/**
 726 * find_first_holder - find the first "holder" gh
 727 * @gl: the glock
 728 */
 729
 730static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
 731{
 732        struct gfs2_holder *gh;
 733
 734        if (!list_empty(&gl->gl_holders)) {
 735                gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
 736                if (test_bit(HIF_HOLDER, &gh->gh_iflags))
 737                        return gh;
 738        }
 739        return NULL;
 740}
 741
 742/**
 743 * run_queue - do all outstanding tasks related to a glock
 744 * @gl: The glock in question
 745 * @nonblock: True if we must not block in run_queue
 746 *
 747 */
 748
 749static void run_queue(struct gfs2_glock *gl, const int nonblock)
 750__releases(&gl->gl_lockref.lock)
 751__acquires(&gl->gl_lockref.lock)
 752{
 753        struct gfs2_holder *gh = NULL;
 754        int ret;
 755
 756        if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
 757                return;
 758
 759        GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
 760
 761        if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
 762            gl->gl_demote_state != gl->gl_state) {
 763                if (find_first_holder(gl))
 764                        goto out_unlock;
 765                if (nonblock)
 766                        goto out_sched;
 767                set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
 768                GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
 769                gl->gl_target = gl->gl_demote_state;
 770        } else {
 771                if (test_bit(GLF_DEMOTE, &gl->gl_flags))
 772                        gfs2_demote_wake(gl);
 773                ret = do_promote(gl);
 774                if (ret == 0)
 775                        goto out_unlock;
 776                if (ret == 2)
 777                        goto out;
 778                gh = find_first_waiter(gl);
 779                gl->gl_target = gh->gh_state;
 780                if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
 781                        do_error(gl, 0); /* Fail queued try locks */
 782        }
 783        do_xmote(gl, gh, gl->gl_target);
 784out:
 785        return;
 786
 787out_sched:
 788        clear_bit(GLF_LOCK, &gl->gl_flags);
 789        smp_mb__after_atomic();
 790        gl->gl_lockref.count++;
 791        __gfs2_glock_queue_work(gl, 0);
 792        return;
 793
 794out_unlock:
 795        clear_bit(GLF_LOCK, &gl->gl_flags);
 796        smp_mb__after_atomic();
 797        return;
 798}
 799
 800void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation)
 801{
 802        struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
 803
 804        if (ri->ri_magic == 0)
 805                ri->ri_magic = cpu_to_be32(GFS2_MAGIC);
 806        if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC))
 807                ri->ri_generation_deleted = cpu_to_be64(generation);
 808}
 809
 810bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation)
 811{
 812        struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
 813
 814        if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC))
 815                return false;
 816        return generation <= be64_to_cpu(ri->ri_generation_deleted);
 817}
 818
 819static void gfs2_glock_poke(struct gfs2_glock *gl)
 820{
 821        int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP;
 822        struct gfs2_holder gh;
 823        int error;
 824
 825        gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh);
 826        error = gfs2_glock_nq(&gh);
 827        if (!error)
 828                gfs2_glock_dq(&gh);
 829        gfs2_holder_uninit(&gh);
 830}
 831
 832static bool gfs2_try_evict(struct gfs2_glock *gl)
 833{
 834        struct gfs2_inode *ip;
 835        bool evicted = false;
 836
 837        /*
 838         * If there is contention on the iopen glock and we have an inode, try
 839         * to grab and release the inode so that it can be evicted.  This will
 840         * allow the remote node to go ahead and delete the inode without us
 841         * having to do it, which will avoid rgrp glock thrashing.
 842         *
 843         * The remote node is likely still holding the corresponding inode
 844         * glock, so it will run before we get to verify that the delete has
 845         * happened below.
 846         */
 847        spin_lock(&gl->gl_lockref.lock);
 848        ip = gl->gl_object;
 849        if (ip && !igrab(&ip->i_inode))
 850                ip = NULL;
 851        spin_unlock(&gl->gl_lockref.lock);
 852        if (ip) {
 853                struct gfs2_glock *inode_gl = NULL;
 854
 855                gl->gl_no_formal_ino = ip->i_no_formal_ino;
 856                set_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
 857                d_prune_aliases(&ip->i_inode);
 858                iput(&ip->i_inode);
 859
 860                /* If the inode was evicted, gl->gl_object will now be NULL. */
 861                spin_lock(&gl->gl_lockref.lock);
 862                ip = gl->gl_object;
 863                if (ip) {
 864                        inode_gl = ip->i_gl;
 865                        lockref_get(&inode_gl->gl_lockref);
 866                        clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
 867                }
 868                spin_unlock(&gl->gl_lockref.lock);
 869                if (inode_gl) {
 870                        gfs2_glock_poke(inode_gl);
 871                        gfs2_glock_put(inode_gl);
 872                }
 873                evicted = !ip;
 874        }
 875        return evicted;
 876}
 877
 878static void delete_work_func(struct work_struct *work)
 879{
 880        struct delayed_work *dwork = to_delayed_work(work);
 881        struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete);
 882        struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 883        struct inode *inode;
 884        u64 no_addr = gl->gl_name.ln_number;
 885
 886        spin_lock(&gl->gl_lockref.lock);
 887        clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
 888        spin_unlock(&gl->gl_lockref.lock);
 889
 890        if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
 891                /*
 892                 * If we can evict the inode, give the remote node trying to
 893                 * delete the inode some time before verifying that the delete
 894                 * has happened.  Otherwise, if we cause contention on the inode glock
 895                 * immediately, the remote node will think that we still have
 896                 * the inode in use, and so it will give up waiting.
 897                 *
 898                 * If we can't evict the inode, signal to the remote node that
 899                 * the inode is still in use.  We'll later try to delete the
 900                 * inode locally in gfs2_evict_inode.
 901                 *
 902                 * FIXME: We only need to verify that the remote node has
 903                 * deleted the inode because nodes before this remote delete
 904                 * rework won't cooperate.  At a later time, when we no longer
 905                 * care about compatibility with such nodes, we can skip this
 906                 * step entirely.
 907                 */
 908                if (gfs2_try_evict(gl)) {
 909                        if (gfs2_queue_delete_work(gl, 5 * HZ))
 910                                return;
 911                }
 912                goto out;
 913        }
 914
 915        inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino,
 916                                    GFS2_BLKST_UNLINKED);
 917        if (!IS_ERR_OR_NULL(inode)) {
 918                d_prune_aliases(inode);
 919                iput(inode);
 920        }
 921out:
 922        gfs2_glock_put(gl);
 923}
 924
 925static void glock_work_func(struct work_struct *work)
 926{
 927        unsigned long delay = 0;
 928        struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
 929        unsigned int drop_refs = 1;
 930
 931        if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
 932                finish_xmote(gl, gl->gl_reply);
 933                drop_refs++;
 934        }
 935        spin_lock(&gl->gl_lockref.lock);
 936        if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
 937            gl->gl_state != LM_ST_UNLOCKED &&
 938            gl->gl_demote_state != LM_ST_EXCLUSIVE) {
 939                unsigned long holdtime, now = jiffies;
 940
 941                holdtime = gl->gl_tchange + gl->gl_hold_time;
 942                if (time_before(now, holdtime))
 943                        delay = holdtime - now;
 944
 945                if (!delay) {
 946                        clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
 947                        gfs2_set_demote(gl);
 948                }
 949        }
 950        run_queue(gl, 0);
 951        if (delay) {
 952                /* Keep one glock reference for the work we requeue. */
 953                drop_refs--;
 954                if (gl->gl_name.ln_type != LM_TYPE_INODE)
 955                        delay = 0;
 956                __gfs2_glock_queue_work(gl, delay);
 957        }
 958
 959        /*
 960         * Drop the remaining glock references manually here. (Mind that
 961         * __gfs2_glock_queue_work depends on the lockref spinlock begin held
 962         * here as well.)
 963         */
 964        gl->gl_lockref.count -= drop_refs;
 965        if (!gl->gl_lockref.count) {
 966                __gfs2_glock_put(gl);
 967                return;
 968        }
 969        spin_unlock(&gl->gl_lockref.lock);
 970}
 971
 972static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
 973                                            struct gfs2_glock *new)
 974{
 975        struct wait_glock_queue wait;
 976        wait_queue_head_t *wq = glock_waitqueue(name);
 977        struct gfs2_glock *gl;
 978
 979        wait.name = name;
 980        init_wait(&wait.wait);
 981        wait.wait.func = glock_wake_function;
 982
 983again:
 984        prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
 985        rcu_read_lock();
 986        if (new) {
 987                gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
 988                        &new->gl_node, ht_parms);
 989                if (IS_ERR(gl))
 990                        goto out;
 991        } else {
 992                gl = rhashtable_lookup_fast(&gl_hash_table,
 993                        name, ht_parms);
 994        }
 995        if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
 996                rcu_read_unlock();
 997                schedule();
 998                goto again;
 999        }
1000out:
1001        rcu_read_unlock();
1002        finish_wait(wq, &wait.wait);
1003        return gl;
1004}
1005
1006/**
1007 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
1008 * @sdp: The GFS2 superblock
1009 * @number: the lock number
1010 * @glops: The glock_operations to use
1011 * @create: If 0, don't create the glock if it doesn't exist
1012 * @glp: the glock is returned here
1013 *
1014 * This does not lock a glock, just finds/creates structures for one.
1015 *
1016 * Returns: errno
1017 */
1018
1019int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
1020                   const struct gfs2_glock_operations *glops, int create,
1021                   struct gfs2_glock **glp)
1022{
1023        struct super_block *s = sdp->sd_vfs;
1024        struct lm_lockname name = { .ln_number = number,
1025                                    .ln_type = glops->go_type,
1026                                    .ln_sbd = sdp };
1027        struct gfs2_glock *gl, *tmp;
1028        struct address_space *mapping;
1029        struct kmem_cache *cachep;
1030        int ret = 0;
1031
1032        gl = find_insert_glock(&name, NULL);
1033        if (gl) {
1034                *glp = gl;
1035                return 0;
1036        }
1037        if (!create)
1038                return -ENOENT;
1039
1040        if (glops->go_flags & GLOF_ASPACE)
1041                cachep = gfs2_glock_aspace_cachep;
1042        else
1043                cachep = gfs2_glock_cachep;
1044        gl = kmem_cache_alloc(cachep, GFP_NOFS);
1045        if (!gl)
1046                return -ENOMEM;
1047
1048        memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
1049
1050        if (glops->go_flags & GLOF_LVB) {
1051                gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1052                if (!gl->gl_lksb.sb_lvbptr) {
1053                        kmem_cache_free(cachep, gl);
1054                        return -ENOMEM;
1055                }
1056        }
1057
1058        atomic_inc(&sdp->sd_glock_disposal);
1059        gl->gl_node.next = NULL;
1060        gl->gl_flags = 0;
1061        gl->gl_name = name;
1062        lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass);
1063        gl->gl_lockref.count = 1;
1064        gl->gl_state = LM_ST_UNLOCKED;
1065        gl->gl_target = LM_ST_UNLOCKED;
1066        gl->gl_demote_state = LM_ST_EXCLUSIVE;
1067        gl->gl_ops = glops;
1068        gl->gl_dstamp = 0;
1069        preempt_disable();
1070        /* We use the global stats to estimate the initial per-glock stats */
1071        gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
1072        preempt_enable();
1073        gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
1074        gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
1075        gl->gl_tchange = jiffies;
1076        gl->gl_object = NULL;
1077        gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
1078        INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
1079        if (gl->gl_name.ln_type == LM_TYPE_IOPEN)
1080                INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func);
1081
1082        mapping = gfs2_glock2aspace(gl);
1083        if (mapping) {
1084                mapping->a_ops = &gfs2_meta_aops;
1085                mapping->host = s->s_bdev->bd_inode;
1086                mapping->flags = 0;
1087                mapping_set_gfp_mask(mapping, GFP_NOFS);
1088                mapping->private_data = NULL;
1089                mapping->writeback_index = 0;
1090        }
1091
1092        tmp = find_insert_glock(&name, gl);
1093        if (!tmp) {
1094                *glp = gl;
1095                goto out;
1096        }
1097        if (IS_ERR(tmp)) {
1098                ret = PTR_ERR(tmp);
1099                goto out_free;
1100        }
1101        *glp = tmp;
1102
1103out_free:
1104        kfree(gl->gl_lksb.sb_lvbptr);
1105        kmem_cache_free(cachep, gl);
1106        if (atomic_dec_and_test(&sdp->sd_glock_disposal))
1107                wake_up(&sdp->sd_glock_wait);
1108
1109out:
1110        return ret;
1111}
1112
1113/**
1114 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
1115 * @gl: the glock
1116 * @state: the state we're requesting
1117 * @flags: the modifier flags
1118 * @gh: the holder structure
1119 *
1120 */
1121
1122void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
1123                      struct gfs2_holder *gh)
1124{
1125        INIT_LIST_HEAD(&gh->gh_list);
1126        gh->gh_gl = gl;
1127        gh->gh_ip = _RET_IP_;
1128        gh->gh_owner_pid = get_pid(task_pid(current));
1129        gh->gh_state = state;
1130        gh->gh_flags = flags;
1131        gh->gh_error = 0;
1132        gh->gh_iflags = 0;
1133        gfs2_glock_hold(gl);
1134}
1135
1136/**
1137 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1138 * @state: the state we're requesting
1139 * @flags: the modifier flags
1140 * @gh: the holder structure
1141 *
1142 * Don't mess with the glock.
1143 *
1144 */
1145
1146void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
1147{
1148        gh->gh_state = state;
1149        gh->gh_flags = flags;
1150        gh->gh_iflags = 0;
1151        gh->gh_ip = _RET_IP_;
1152        put_pid(gh->gh_owner_pid);
1153        gh->gh_owner_pid = get_pid(task_pid(current));
1154}
1155
1156/**
1157 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1158 * @gh: the holder structure
1159 *
1160 */
1161
1162void gfs2_holder_uninit(struct gfs2_holder *gh)
1163{
1164        put_pid(gh->gh_owner_pid);
1165        gfs2_glock_put(gh->gh_gl);
1166        gfs2_holder_mark_uninitialized(gh);
1167        gh->gh_ip = 0;
1168}
1169
1170static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
1171                                        unsigned long start_time)
1172{
1173        /* Have we waited longer that a second? */
1174        if (time_after(jiffies, start_time + HZ)) {
1175                /* Lengthen the minimum hold time. */
1176                gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
1177                                       GL_GLOCK_MAX_HOLD);
1178        }
1179}
1180
1181/**
1182 * gfs2_glock_wait - wait on a glock acquisition
1183 * @gh: the glock holder
1184 *
1185 * Returns: 0 on success
1186 */
1187
1188int gfs2_glock_wait(struct gfs2_holder *gh)
1189{
1190        unsigned long start_time = jiffies;
1191
1192        might_sleep();
1193        wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1194        gfs2_glock_update_hold_time(gh->gh_gl, start_time);
1195        return gh->gh_error;
1196}
1197
1198static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
1199{
1200        int i;
1201
1202        for (i = 0; i < num_gh; i++)
1203                if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
1204                        return 1;
1205        return 0;
1206}
1207
1208/**
1209 * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1210 * @num_gh: the number of holders in the array
1211 * @ghs: the glock holder array
1212 *
1213 * Returns: 0 on success, meaning all glocks have been granted and are held.
1214 *          -ESTALE if the request timed out, meaning all glocks were released,
1215 *          and the caller should retry the operation.
1216 */
1217
1218int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
1219{
1220        struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
1221        int i, ret = 0, timeout = 0;
1222        unsigned long start_time = jiffies;
1223        bool keep_waiting;
1224
1225        might_sleep();
1226        /*
1227         * Total up the (minimum hold time * 2) of all glocks and use that to
1228         * determine the max amount of time we should wait.
1229         */
1230        for (i = 0; i < num_gh; i++)
1231                timeout += ghs[i].gh_gl->gl_hold_time << 1;
1232
1233wait_for_dlm:
1234        if (!wait_event_timeout(sdp->sd_async_glock_wait,
1235                                !glocks_pending(num_gh, ghs), timeout))
1236                ret = -ESTALE; /* request timed out. */
1237
1238        /*
1239         * If dlm granted all our requests, we need to adjust the glock
1240         * minimum hold time values according to how long we waited.
1241         *
1242         * If our request timed out, we need to repeatedly release any held
1243         * glocks we acquired thus far to allow dlm to acquire the remaining
1244         * glocks without deadlocking.  We cannot currently cancel outstanding
1245         * glock acquisitions.
1246         *
1247         * The HIF_WAIT bit tells us which requests still need a response from
1248         * dlm.
1249         *
1250         * If dlm sent us any errors, we return the first error we find.
1251         */
1252        keep_waiting = false;
1253        for (i = 0; i < num_gh; i++) {
1254                /* Skip holders we have already dequeued below. */
1255                if (!gfs2_holder_queued(&ghs[i]))
1256                        continue;
1257                /* Skip holders with a pending DLM response. */
1258                if (test_bit(HIF_WAIT, &ghs[i].gh_iflags)) {
1259                        keep_waiting = true;
1260                        continue;
1261                }
1262
1263                if (test_bit(HIF_HOLDER, &ghs[i].gh_iflags)) {
1264                        if (ret == -ESTALE)
1265                                gfs2_glock_dq(&ghs[i]);
1266                        else
1267                                gfs2_glock_update_hold_time(ghs[i].gh_gl,
1268                                                            start_time);
1269                }
1270                if (!ret)
1271                        ret = ghs[i].gh_error;
1272        }
1273
1274        if (keep_waiting)
1275                goto wait_for_dlm;
1276
1277        /*
1278         * At this point, we've either acquired all locks or released them all.
1279         */
1280        return ret;
1281}
1282
1283/**
1284 * handle_callback - process a demote request
1285 * @gl: the glock
1286 * @state: the state the caller wants us to change to
1287 * @delay: zero to demote immediately; otherwise pending demote
1288 * @remote: true if this came from a different cluster node
1289 *
1290 * There are only two requests that we are going to see in actual
1291 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1292 */
1293
1294static void handle_callback(struct gfs2_glock *gl, unsigned int state,
1295                            unsigned long delay, bool remote)
1296{
1297        if (delay)
1298                set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1299        else
1300                gfs2_set_demote(gl);
1301        if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
1302                gl->gl_demote_state = state;
1303                gl->gl_demote_time = jiffies;
1304        } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
1305                        gl->gl_demote_state != state) {
1306                gl->gl_demote_state = LM_ST_UNLOCKED;
1307        }
1308        if (gl->gl_ops->go_callback)
1309                gl->gl_ops->go_callback(gl, remote);
1310        trace_gfs2_demote_rq(gl, remote);
1311}
1312
1313void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
1314{
1315        struct va_format vaf;
1316        va_list args;
1317
1318        va_start(args, fmt);
1319
1320        if (seq) {
1321                seq_vprintf(seq, fmt, args);
1322        } else {
1323                vaf.fmt = fmt;
1324                vaf.va = &args;
1325
1326                pr_err("%pV", &vaf);
1327        }
1328
1329        va_end(args);
1330}
1331
1332/**
1333 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1334 * @gh: the holder structure to add
1335 *
1336 * Eventually we should move the recursive locking trap to a
1337 * debugging option or something like that. This is the fast
1338 * path and needs to have the minimum number of distractions.
1339 * 
1340 */
1341
1342static inline void add_to_queue(struct gfs2_holder *gh)
1343__releases(&gl->gl_lockref.lock)
1344__acquires(&gl->gl_lockref.lock)
1345{
1346        struct gfs2_glock *gl = gh->gh_gl;
1347        struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1348        struct list_head *insert_pt = NULL;
1349        struct gfs2_holder *gh2;
1350        int try_futile = 0;
1351
1352        GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1353        if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1354                GLOCK_BUG_ON(gl, true);
1355
1356        if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1357                if (test_bit(GLF_LOCK, &gl->gl_flags))
1358                        try_futile = !may_grant(gl, gh);
1359                if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1360                        goto fail;
1361        }
1362
1363        list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1364                if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
1365                    (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
1366                        goto trap_recursive;
1367                if (try_futile &&
1368                    !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1369fail:
1370                        gh->gh_error = GLR_TRYFAILED;
1371                        gfs2_holder_wake(gh);
1372                        return;
1373                }
1374                if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1375                        continue;
1376                if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1377                        insert_pt = &gh2->gh_list;
1378        }
1379        trace_gfs2_glock_queue(gh, 1);
1380        gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1381        gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1382        if (likely(insert_pt == NULL)) {
1383                list_add_tail(&gh->gh_list, &gl->gl_holders);
1384                if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1385                        goto do_cancel;
1386                return;
1387        }
1388        list_add_tail(&gh->gh_list, insert_pt);
1389do_cancel:
1390        gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
1391        if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1392                spin_unlock(&gl->gl_lockref.lock);
1393                if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1394                        sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1395                spin_lock(&gl->gl_lockref.lock);
1396        }
1397        return;
1398
1399trap_recursive:
1400        fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1401        fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1402        fs_err(sdp, "lock type: %d req lock state : %d\n",
1403               gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1404        fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1405        fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1406        fs_err(sdp, "lock type: %d req lock state : %d\n",
1407               gh->gh_gl->gl_name.ln_type, gh->gh_state);
1408        gfs2_dump_glock(NULL, gl, true);
1409        BUG();
1410}
1411
1412/**
1413 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1414 * @gh: the holder structure
1415 *
1416 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1417 *
1418 * Returns: 0, GLR_TRYFAILED, or errno on failure
1419 */
1420
1421int gfs2_glock_nq(struct gfs2_holder *gh)
1422{
1423        struct gfs2_glock *gl = gh->gh_gl;
1424        int error = 0;
1425
1426        if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP))
1427                return -EIO;
1428
1429        if (test_bit(GLF_LRU, &gl->gl_flags))
1430                gfs2_glock_remove_from_lru(gl);
1431
1432        spin_lock(&gl->gl_lockref.lock);
1433        add_to_queue(gh);
1434        if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1435                     test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1436                set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1437                gl->gl_lockref.count++;
1438                __gfs2_glock_queue_work(gl, 0);
1439        }
1440        run_queue(gl, 1);
1441        spin_unlock(&gl->gl_lockref.lock);
1442
1443        if (!(gh->gh_flags & GL_ASYNC))
1444                error = gfs2_glock_wait(gh);
1445
1446        return error;
1447}
1448
1449/**
1450 * gfs2_glock_poll - poll to see if an async request has been completed
1451 * @gh: the holder
1452 *
1453 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1454 */
1455
1456int gfs2_glock_poll(struct gfs2_holder *gh)
1457{
1458        return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1459}
1460
1461/**
1462 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1463 * @gh: the glock holder
1464 *
1465 */
1466
1467void gfs2_glock_dq(struct gfs2_holder *gh)
1468{
1469        struct gfs2_glock *gl = gh->gh_gl;
1470        struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1471        unsigned delay = 0;
1472        int fast_path = 0;
1473
1474        spin_lock(&gl->gl_lockref.lock);
1475        /*
1476         * If we're in the process of file system withdraw, we cannot just
1477         * dequeue any glocks until our journal is recovered, lest we
1478         * introduce file system corruption. We need two exceptions to this
1479         * rule: We need to allow unlocking of nondisk glocks and the glock
1480         * for our own journal that needs recovery.
1481         */
1482        if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) &&
1483            glock_blocked_by_withdraw(gl) &&
1484            gh->gh_gl != sdp->sd_jinode_gl) {
1485                sdp->sd_glock_dqs_held++;
1486                spin_unlock(&gl->gl_lockref.lock);
1487                might_sleep();
1488                wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
1489                            TASK_UNINTERRUPTIBLE);
1490                spin_lock(&gl->gl_lockref.lock);
1491        }
1492        if (gh->gh_flags & GL_NOCACHE)
1493                handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1494
1495        list_del_init(&gh->gh_list);
1496        clear_bit(HIF_HOLDER, &gh->gh_iflags);
1497        if (list_empty(&gl->gl_holders) &&
1498            !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1499            !test_bit(GLF_DEMOTE, &gl->gl_flags))
1500                fast_path = 1;
1501
1502        if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1503                gfs2_glock_add_to_lru(gl);
1504
1505        trace_gfs2_glock_queue(gh, 0);
1506        if (unlikely(!fast_path)) {
1507                gl->gl_lockref.count++;
1508                if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1509                    !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1510                    gl->gl_name.ln_type == LM_TYPE_INODE)
1511                        delay = gl->gl_hold_time;
1512                __gfs2_glock_queue_work(gl, delay);
1513        }
1514        spin_unlock(&gl->gl_lockref.lock);
1515}
1516
1517void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1518{
1519        struct gfs2_glock *gl = gh->gh_gl;
1520        gfs2_glock_dq(gh);
1521        might_sleep();
1522        wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1523}
1524
1525/**
1526 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1527 * @gh: the holder structure
1528 *
1529 */
1530
1531void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1532{
1533        gfs2_glock_dq(gh);
1534        gfs2_holder_uninit(gh);
1535}
1536
1537/**
1538 * gfs2_glock_nq_num - acquire a glock based on lock number
1539 * @sdp: the filesystem
1540 * @number: the lock number
1541 * @glops: the glock operations for the type of glock
1542 * @state: the state to acquire the glock in
1543 * @flags: modifier flags for the acquisition
1544 * @gh: the struct gfs2_holder
1545 *
1546 * Returns: errno
1547 */
1548
1549int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1550                      const struct gfs2_glock_operations *glops,
1551                      unsigned int state, u16 flags, struct gfs2_holder *gh)
1552{
1553        struct gfs2_glock *gl;
1554        int error;
1555
1556        error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1557        if (!error) {
1558                error = gfs2_glock_nq_init(gl, state, flags, gh);
1559                gfs2_glock_put(gl);
1560        }
1561
1562        return error;
1563}
1564
1565/**
1566 * glock_compare - Compare two struct gfs2_glock structures for sorting
1567 * @arg_a: the first structure
1568 * @arg_b: the second structure
1569 *
1570 */
1571
1572static int glock_compare(const void *arg_a, const void *arg_b)
1573{
1574        const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1575        const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1576        const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1577        const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1578
1579        if (a->ln_number > b->ln_number)
1580                return 1;
1581        if (a->ln_number < b->ln_number)
1582                return -1;
1583        BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1584        return 0;
1585}
1586
1587/**
1588 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1589 * @num_gh: the number of structures
1590 * @ghs: an array of struct gfs2_holder structures
1591 * @p: placeholder for the holder structure to pass back
1592 *
1593 * Returns: 0 on success (all glocks acquired),
1594 *          errno on failure (no glocks acquired)
1595 */
1596
1597static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1598                     struct gfs2_holder **p)
1599{
1600        unsigned int x;
1601        int error = 0;
1602
1603        for (x = 0; x < num_gh; x++)
1604                p[x] = &ghs[x];
1605
1606        sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1607
1608        for (x = 0; x < num_gh; x++) {
1609                p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1610
1611                error = gfs2_glock_nq(p[x]);
1612                if (error) {
1613                        while (x--)
1614                                gfs2_glock_dq(p[x]);
1615                        break;
1616                }
1617        }
1618
1619        return error;
1620}
1621
1622/**
1623 * gfs2_glock_nq_m - acquire multiple glocks
1624 * @num_gh: the number of structures
1625 * @ghs: an array of struct gfs2_holder structures
1626 *
1627 *
1628 * Returns: 0 on success (all glocks acquired),
1629 *          errno on failure (no glocks acquired)
1630 */
1631
1632int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1633{
1634        struct gfs2_holder *tmp[4];
1635        struct gfs2_holder **pph = tmp;
1636        int error = 0;
1637
1638        switch(num_gh) {
1639        case 0:
1640                return 0;
1641        case 1:
1642                ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1643                return gfs2_glock_nq(ghs);
1644        default:
1645                if (num_gh <= 4)
1646                        break;
1647                pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *),
1648                                    GFP_NOFS);
1649                if (!pph)
1650                        return -ENOMEM;
1651        }
1652
1653        error = nq_m_sync(num_gh, ghs, pph);
1654
1655        if (pph != tmp)
1656                kfree(pph);
1657
1658        return error;
1659}
1660
1661/**
1662 * gfs2_glock_dq_m - release multiple glocks
1663 * @num_gh: the number of structures
1664 * @ghs: an array of struct gfs2_holder structures
1665 *
1666 */
1667
1668void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1669{
1670        while (num_gh--)
1671                gfs2_glock_dq(&ghs[num_gh]);
1672}
1673
1674void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1675{
1676        unsigned long delay = 0;
1677        unsigned long holdtime;
1678        unsigned long now = jiffies;
1679
1680        gfs2_glock_hold(gl);
1681        spin_lock(&gl->gl_lockref.lock);
1682        holdtime = gl->gl_tchange + gl->gl_hold_time;
1683        if (!list_empty(&gl->gl_holders) &&
1684            gl->gl_name.ln_type == LM_TYPE_INODE) {
1685                if (time_before(now, holdtime))
1686                        delay = holdtime - now;
1687                if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1688                        delay = gl->gl_hold_time;
1689        }
1690        handle_callback(gl, state, delay, true);
1691        __gfs2_glock_queue_work(gl, delay);
1692        spin_unlock(&gl->gl_lockref.lock);
1693}
1694
1695/**
1696 * gfs2_should_freeze - Figure out if glock should be frozen
1697 * @gl: The glock in question
1698 *
1699 * Glocks are not frozen if (a) the result of the dlm operation is
1700 * an error, (b) the locking operation was an unlock operation or
1701 * (c) if there is a "noexp" flagged request anywhere in the queue
1702 *
1703 * Returns: 1 if freezing should occur, 0 otherwise
1704 */
1705
1706static int gfs2_should_freeze(const struct gfs2_glock *gl)
1707{
1708        const struct gfs2_holder *gh;
1709
1710        if (gl->gl_reply & ~LM_OUT_ST_MASK)
1711                return 0;
1712        if (gl->gl_target == LM_ST_UNLOCKED)
1713                return 0;
1714
1715        list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1716                if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1717                        continue;
1718                if (LM_FLAG_NOEXP & gh->gh_flags)
1719                        return 0;
1720        }
1721
1722        return 1;
1723}
1724
1725/**
1726 * gfs2_glock_complete - Callback used by locking
1727 * @gl: Pointer to the glock
1728 * @ret: The return value from the dlm
1729 *
1730 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1731 * to use a bitfield shared with other glock state fields.
1732 */
1733
1734void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1735{
1736        struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1737
1738        spin_lock(&gl->gl_lockref.lock);
1739        gl->gl_reply = ret;
1740
1741        if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1742                if (gfs2_should_freeze(gl)) {
1743                        set_bit(GLF_FROZEN, &gl->gl_flags);
1744                        spin_unlock(&gl->gl_lockref.lock);
1745                        return;
1746                }
1747        }
1748
1749        gl->gl_lockref.count++;
1750        set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1751        __gfs2_glock_queue_work(gl, 0);
1752        spin_unlock(&gl->gl_lockref.lock);
1753}
1754
1755static int glock_cmp(void *priv, const struct list_head *a,
1756                     const struct list_head *b)
1757{
1758        struct gfs2_glock *gla, *glb;
1759
1760        gla = list_entry(a, struct gfs2_glock, gl_lru);
1761        glb = list_entry(b, struct gfs2_glock, gl_lru);
1762
1763        if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1764                return 1;
1765        if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1766                return -1;
1767
1768        return 0;
1769}
1770
1771/**
1772 * gfs2_dispose_glock_lru - Demote a list of glocks
1773 * @list: The list to dispose of
1774 *
1775 * Disposing of glocks may involve disk accesses, so that here we sort
1776 * the glocks by number (i.e. disk location of the inodes) so that if
1777 * there are any such accesses, they'll be sent in order (mostly).
1778 *
1779 * Must be called under the lru_lock, but may drop and retake this
1780 * lock. While the lru_lock is dropped, entries may vanish from the
1781 * list, but no new entries will appear on the list (since it is
1782 * private)
1783 */
1784
1785static void gfs2_dispose_glock_lru(struct list_head *list)
1786__releases(&lru_lock)
1787__acquires(&lru_lock)
1788{
1789        struct gfs2_glock *gl;
1790
1791        list_sort(NULL, list, glock_cmp);
1792
1793        while(!list_empty(list)) {
1794                gl = list_first_entry(list, struct gfs2_glock, gl_lru);
1795                list_del_init(&gl->gl_lru);
1796                clear_bit(GLF_LRU, &gl->gl_flags);
1797                if (!spin_trylock(&gl->gl_lockref.lock)) {
1798add_back_to_lru:
1799                        list_add(&gl->gl_lru, &lru_list);
1800                        set_bit(GLF_LRU, &gl->gl_flags);
1801                        atomic_inc(&lru_count);
1802                        continue;
1803                }
1804                if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1805                        spin_unlock(&gl->gl_lockref.lock);
1806                        goto add_back_to_lru;
1807                }
1808                gl->gl_lockref.count++;
1809                if (demote_ok(gl))
1810                        handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1811                WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1812                __gfs2_glock_queue_work(gl, 0);
1813                spin_unlock(&gl->gl_lockref.lock);
1814                cond_resched_lock(&lru_lock);
1815        }
1816}
1817
1818/**
1819 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1820 * @nr: The number of entries to scan
1821 *
1822 * This function selects the entries on the LRU which are able to
1823 * be demoted, and then kicks off the process by calling
1824 * gfs2_dispose_glock_lru() above.
1825 */
1826
1827static long gfs2_scan_glock_lru(int nr)
1828{
1829        struct gfs2_glock *gl;
1830        LIST_HEAD(skipped);
1831        LIST_HEAD(dispose);
1832        long freed = 0;
1833
1834        spin_lock(&lru_lock);
1835        while ((nr-- >= 0) && !list_empty(&lru_list)) {
1836                gl = list_first_entry(&lru_list, struct gfs2_glock, gl_lru);
1837
1838                /* Test for being demotable */
1839                if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1840                        list_move(&gl->gl_lru, &dispose);
1841                        atomic_dec(&lru_count);
1842                        freed++;
1843                        continue;
1844                }
1845
1846                list_move(&gl->gl_lru, &skipped);
1847        }
1848        list_splice(&skipped, &lru_list);
1849        if (!list_empty(&dispose))
1850                gfs2_dispose_glock_lru(&dispose);
1851        spin_unlock(&lru_lock);
1852
1853        return freed;
1854}
1855
1856static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1857                                            struct shrink_control *sc)
1858{
1859        if (!(sc->gfp_mask & __GFP_FS))
1860                return SHRINK_STOP;
1861        return gfs2_scan_glock_lru(sc->nr_to_scan);
1862}
1863
1864static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1865                                             struct shrink_control *sc)
1866{
1867        return vfs_pressure_ratio(atomic_read(&lru_count));
1868}
1869
1870static struct shrinker glock_shrinker = {
1871        .seeks = DEFAULT_SEEKS,
1872        .count_objects = gfs2_glock_shrink_count,
1873        .scan_objects = gfs2_glock_shrink_scan,
1874};
1875
1876/**
1877 * glock_hash_walk - Call a function for glock in a hash bucket
1878 * @examiner: the function
1879 * @sdp: the filesystem
1880 *
1881 * Note that the function can be called multiple times on the same
1882 * object.  So the user must ensure that the function can cope with
1883 * that.
1884 */
1885
1886static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1887{
1888        struct gfs2_glock *gl;
1889        struct rhashtable_iter iter;
1890
1891        rhashtable_walk_enter(&gl_hash_table, &iter);
1892
1893        do {
1894                rhashtable_walk_start(&iter);
1895
1896                while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl))
1897                        if (gl->gl_name.ln_sbd == sdp &&
1898                            lockref_get_not_dead(&gl->gl_lockref))
1899                                examiner(gl);
1900
1901                rhashtable_walk_stop(&iter);
1902        } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
1903
1904        rhashtable_walk_exit(&iter);
1905}
1906
1907bool gfs2_queue_delete_work(struct gfs2_glock *gl, unsigned long delay)
1908{
1909        bool queued;
1910
1911        spin_lock(&gl->gl_lockref.lock);
1912        queued = queue_delayed_work(gfs2_delete_workqueue,
1913                                    &gl->gl_delete, delay);
1914        if (queued)
1915                set_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1916        spin_unlock(&gl->gl_lockref.lock);
1917        return queued;
1918}
1919
1920void gfs2_cancel_delete_work(struct gfs2_glock *gl)
1921{
1922        if (cancel_delayed_work_sync(&gl->gl_delete)) {
1923                clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1924                gfs2_glock_put(gl);
1925        }
1926}
1927
1928bool gfs2_delete_work_queued(const struct gfs2_glock *gl)
1929{
1930        return test_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1931}
1932
1933static void flush_delete_work(struct gfs2_glock *gl)
1934{
1935        if (gl->gl_name.ln_type == LM_TYPE_IOPEN) {
1936                if (cancel_delayed_work(&gl->gl_delete)) {
1937                        queue_delayed_work(gfs2_delete_workqueue,
1938                                           &gl->gl_delete, 0);
1939                }
1940        }
1941        gfs2_glock_queue_work(gl, 0);
1942}
1943
1944void gfs2_flush_delete_work(struct gfs2_sbd *sdp)
1945{
1946        glock_hash_walk(flush_delete_work, sdp);
1947        flush_workqueue(gfs2_delete_workqueue);
1948}
1949
1950/**
1951 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1952 * @gl: The glock to thaw
1953 *
1954 */
1955
1956static void thaw_glock(struct gfs2_glock *gl)
1957{
1958        if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags)) {
1959                gfs2_glock_put(gl);
1960                return;
1961        }
1962        set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1963        gfs2_glock_queue_work(gl, 0);
1964}
1965
1966/**
1967 * clear_glock - look at a glock and see if we can free it from glock cache
1968 * @gl: the glock to look at
1969 *
1970 */
1971
1972static void clear_glock(struct gfs2_glock *gl)
1973{
1974        gfs2_glock_remove_from_lru(gl);
1975
1976        spin_lock(&gl->gl_lockref.lock);
1977        if (gl->gl_state != LM_ST_UNLOCKED)
1978                handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1979        __gfs2_glock_queue_work(gl, 0);
1980        spin_unlock(&gl->gl_lockref.lock);
1981}
1982
1983/**
1984 * gfs2_glock_thaw - Thaw any frozen glocks
1985 * @sdp: The super block
1986 *
1987 */
1988
1989void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1990{
1991        glock_hash_walk(thaw_glock, sdp);
1992}
1993
1994static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
1995{
1996        spin_lock(&gl->gl_lockref.lock);
1997        gfs2_dump_glock(seq, gl, fsid);
1998        spin_unlock(&gl->gl_lockref.lock);
1999}
2000
2001static void dump_glock_func(struct gfs2_glock *gl)
2002{
2003        dump_glock(NULL, gl, true);
2004}
2005
2006/**
2007 * gfs2_gl_hash_clear - Empty out the glock hash table
2008 * @sdp: the filesystem
2009 *
2010 * Called when unmounting the filesystem.
2011 */
2012
2013void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
2014{
2015        set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
2016        flush_workqueue(glock_workqueue);
2017        glock_hash_walk(clear_glock, sdp);
2018        flush_workqueue(glock_workqueue);
2019        wait_event_timeout(sdp->sd_glock_wait,
2020                           atomic_read(&sdp->sd_glock_disposal) == 0,
2021                           HZ * 600);
2022        glock_hash_walk(dump_glock_func, sdp);
2023}
2024
2025void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
2026{
2027        struct gfs2_glock *gl = ip->i_gl;
2028        int ret;
2029
2030        ret = gfs2_truncatei_resume(ip);
2031        gfs2_glock_assert_withdraw(gl, ret == 0);
2032
2033        spin_lock(&gl->gl_lockref.lock);
2034        clear_bit(GLF_LOCK, &gl->gl_flags);
2035        run_queue(gl, 1);
2036        spin_unlock(&gl->gl_lockref.lock);
2037}
2038
2039static const char *state2str(unsigned state)
2040{
2041        switch(state) {
2042        case LM_ST_UNLOCKED:
2043                return "UN";
2044        case LM_ST_SHARED:
2045                return "SH";
2046        case LM_ST_DEFERRED:
2047                return "DF";
2048        case LM_ST_EXCLUSIVE:
2049                return "EX";
2050        }
2051        return "??";
2052}
2053
2054static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
2055{
2056        char *p = buf;
2057        if (flags & LM_FLAG_TRY)
2058                *p++ = 't';
2059        if (flags & LM_FLAG_TRY_1CB)
2060                *p++ = 'T';
2061        if (flags & LM_FLAG_NOEXP)
2062                *p++ = 'e';
2063        if (flags & LM_FLAG_ANY)
2064                *p++ = 'A';
2065        if (flags & LM_FLAG_PRIORITY)
2066                *p++ = 'p';
2067        if (flags & LM_FLAG_NODE_SCOPE)
2068                *p++ = 'n';
2069        if (flags & GL_ASYNC)
2070                *p++ = 'a';
2071        if (flags & GL_EXACT)
2072                *p++ = 'E';
2073        if (flags & GL_NOCACHE)
2074                *p++ = 'c';
2075        if (test_bit(HIF_HOLDER, &iflags))
2076                *p++ = 'H';
2077        if (test_bit(HIF_WAIT, &iflags))
2078                *p++ = 'W';
2079        *p = 0;
2080        return buf;
2081}
2082
2083/**
2084 * dump_holder - print information about a glock holder
2085 * @seq: the seq_file struct
2086 * @gh: the glock holder
2087 * @fs_id_buf: pointer to file system id (if requested)
2088 *
2089 */
2090
2091static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
2092                        const char *fs_id_buf)
2093{
2094        struct task_struct *gh_owner = NULL;
2095        char flags_buf[32];
2096
2097        rcu_read_lock();
2098        if (gh->gh_owner_pid)
2099                gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
2100        gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2101                       fs_id_buf, state2str(gh->gh_state),
2102                       hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
2103                       gh->gh_error,
2104                       gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
2105                       gh_owner ? gh_owner->comm : "(ended)",
2106                       (void *)gh->gh_ip);
2107        rcu_read_unlock();
2108}
2109
2110static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
2111{
2112        const unsigned long *gflags = &gl->gl_flags;
2113        char *p = buf;
2114
2115        if (test_bit(GLF_LOCK, gflags))
2116                *p++ = 'l';
2117        if (test_bit(GLF_DEMOTE, gflags))
2118                *p++ = 'D';
2119        if (test_bit(GLF_PENDING_DEMOTE, gflags))
2120                *p++ = 'd';
2121        if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
2122                *p++ = 'p';
2123        if (test_bit(GLF_DIRTY, gflags))
2124                *p++ = 'y';
2125        if (test_bit(GLF_LFLUSH, gflags))
2126                *p++ = 'f';
2127        if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
2128                *p++ = 'i';
2129        if (test_bit(GLF_REPLY_PENDING, gflags))
2130                *p++ = 'r';
2131        if (test_bit(GLF_INITIAL, gflags))
2132                *p++ = 'I';
2133        if (test_bit(GLF_FROZEN, gflags))
2134                *p++ = 'F';
2135        if (!list_empty(&gl->gl_holders))
2136                *p++ = 'q';
2137        if (test_bit(GLF_LRU, gflags))
2138                *p++ = 'L';
2139        if (gl->gl_object)
2140                *p++ = 'o';
2141        if (test_bit(GLF_BLOCKING, gflags))
2142                *p++ = 'b';
2143        if (test_bit(GLF_PENDING_DELETE, gflags))
2144                *p++ = 'P';
2145        if (test_bit(GLF_FREEING, gflags))
2146                *p++ = 'x';
2147        *p = 0;
2148        return buf;
2149}
2150
2151/**
2152 * gfs2_dump_glock - print information about a glock
2153 * @seq: The seq_file struct
2154 * @gl: the glock
2155 * @fsid: If true, also dump the file system id
2156 *
2157 * The file format is as follows:
2158 * One line per object, capital letters are used to indicate objects
2159 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2160 * other objects are indented by a single space and follow the glock to
2161 * which they are related. Fields are indicated by lower case letters
2162 * followed by a colon and the field value, except for strings which are in
2163 * [] so that its possible to see if they are composed of spaces for
2164 * example. The field's are n = number (id of the object), f = flags,
2165 * t = type, s = state, r = refcount, e = error, p = pid.
2166 *
2167 */
2168
2169void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2170{
2171        const struct gfs2_glock_operations *glops = gl->gl_ops;
2172        unsigned long long dtime;
2173        const struct gfs2_holder *gh;
2174        char gflags_buf[32];
2175        struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2176        char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2177        unsigned long nrpages = 0;
2178
2179        if (gl->gl_ops->go_flags & GLOF_ASPACE) {
2180                struct address_space *mapping = gfs2_glock2aspace(gl);
2181
2182                nrpages = mapping->nrpages;
2183        }
2184        memset(fs_id_buf, 0, sizeof(fs_id_buf));
2185        if (fsid && sdp) /* safety precaution */
2186                sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2187        dtime = jiffies - gl->gl_demote_time;
2188        dtime *= 1000000/HZ; /* demote time in uSec */
2189        if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
2190                dtime = 0;
2191        gfs2_print_dbg(seq, "%sG:  s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2192                       "v:%d r:%d m:%ld p:%lu\n",
2193                       fs_id_buf, state2str(gl->gl_state),
2194                       gl->gl_name.ln_type,
2195                       (unsigned long long)gl->gl_name.ln_number,
2196                       gflags2str(gflags_buf, gl),
2197                       state2str(gl->gl_target),
2198                       state2str(gl->gl_demote_state), dtime,
2199                       atomic_read(&gl->gl_ail_count),
2200                       atomic_read(&gl->gl_revokes),
2201                       (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages);
2202
2203        list_for_each_entry(gh, &gl->gl_holders, gh_list)
2204                dump_holder(seq, gh, fs_id_buf);
2205
2206        if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
2207                glops->go_dump(seq, gl, fs_id_buf);
2208}
2209
2210static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
2211{
2212        struct gfs2_glock *gl = iter_ptr;
2213
2214        seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2215                   gl->gl_name.ln_type,
2216                   (unsigned long long)gl->gl_name.ln_number,
2217                   (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
2218                   (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
2219                   (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
2220                   (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
2221                   (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
2222                   (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
2223                   (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
2224                   (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
2225        return 0;
2226}
2227
2228static const char *gfs2_gltype[] = {
2229        "type",
2230        "reserved",
2231        "nondisk",
2232        "inode",
2233        "rgrp",
2234        "meta",
2235        "iopen",
2236        "flock",
2237        "plock",
2238        "quota",
2239        "journal",
2240};
2241
2242static const char *gfs2_stype[] = {
2243        [GFS2_LKS_SRTT]         = "srtt",
2244        [GFS2_LKS_SRTTVAR]      = "srttvar",
2245        [GFS2_LKS_SRTTB]        = "srttb",
2246        [GFS2_LKS_SRTTVARB]     = "srttvarb",
2247        [GFS2_LKS_SIRT]         = "sirt",
2248        [GFS2_LKS_SIRTVAR]      = "sirtvar",
2249        [GFS2_LKS_DCOUNT]       = "dlm",
2250        [GFS2_LKS_QCOUNT]       = "queue",
2251};
2252
2253#define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2254
2255static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
2256{
2257        struct gfs2_sbd *sdp = seq->private;
2258        loff_t pos = *(loff_t *)iter_ptr;
2259        unsigned index = pos >> 3;
2260        unsigned subindex = pos & 0x07;
2261        int i;
2262
2263        if (index == 0 && subindex != 0)
2264                return 0;
2265
2266        seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
2267                   (index == 0) ? "cpu": gfs2_stype[subindex]);
2268
2269        for_each_possible_cpu(i) {
2270                const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
2271
2272                if (index == 0)
2273                        seq_printf(seq, " %15u", i);
2274                else
2275                        seq_printf(seq, " %15llu", (unsigned long long)lkstats->
2276                                   lkstats[index - 1].stats[subindex]);
2277        }
2278        seq_putc(seq, '\n');
2279        return 0;
2280}
2281
2282int __init gfs2_glock_init(void)
2283{
2284        int i, ret;
2285
2286        ret = rhashtable_init(&gl_hash_table, &ht_parms);
2287        if (ret < 0)
2288                return ret;
2289
2290        glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
2291                                          WQ_HIGHPRI | WQ_FREEZABLE, 0);
2292        if (!glock_workqueue) {
2293                rhashtable_destroy(&gl_hash_table);
2294                return -ENOMEM;
2295        }
2296        gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
2297                                                WQ_MEM_RECLAIM | WQ_FREEZABLE,
2298                                                0);
2299        if (!gfs2_delete_workqueue) {
2300                destroy_workqueue(glock_workqueue);
2301                rhashtable_destroy(&gl_hash_table);
2302                return -ENOMEM;
2303        }
2304
2305        ret = register_shrinker(&glock_shrinker);
2306        if (ret) {
2307                destroy_workqueue(gfs2_delete_workqueue);
2308                destroy_workqueue(glock_workqueue);
2309                rhashtable_destroy(&gl_hash_table);
2310                return ret;
2311        }
2312
2313        for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2314                init_waitqueue_head(glock_wait_table + i);
2315
2316        return 0;
2317}
2318
2319void gfs2_glock_exit(void)
2320{
2321        unregister_shrinker(&glock_shrinker);
2322        rhashtable_destroy(&gl_hash_table);
2323        destroy_workqueue(glock_workqueue);
2324        destroy_workqueue(gfs2_delete_workqueue);
2325}
2326
2327static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2328{
2329        struct gfs2_glock *gl = gi->gl;
2330
2331        if (gl) {
2332                if (n == 0)
2333                        return;
2334                if (!lockref_put_not_zero(&gl->gl_lockref))
2335                        gfs2_glock_queue_put(gl);
2336        }
2337        for (;;) {
2338                gl = rhashtable_walk_next(&gi->hti);
2339                if (IS_ERR_OR_NULL(gl)) {
2340                        if (gl == ERR_PTR(-EAGAIN)) {
2341                                n = 1;
2342                                continue;
2343                        }
2344                        gl = NULL;
2345                        break;
2346                }
2347                if (gl->gl_name.ln_sbd != gi->sdp)
2348                        continue;
2349                if (n <= 1) {
2350                        if (!lockref_get_not_dead(&gl->gl_lockref))
2351                                continue;
2352                        break;
2353                } else {
2354                        if (__lockref_is_dead(&gl->gl_lockref))
2355                                continue;
2356                        n--;
2357                }
2358        }
2359        gi->gl = gl;
2360}
2361
2362static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2363        __acquires(RCU)
2364{
2365        struct gfs2_glock_iter *gi = seq->private;
2366        loff_t n;
2367
2368        /*
2369         * We can either stay where we are, skip to the next hash table
2370         * entry, or start from the beginning.
2371         */
2372        if (*pos < gi->last_pos) {
2373                rhashtable_walk_exit(&gi->hti);
2374                rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2375                n = *pos + 1;
2376        } else {
2377                n = *pos - gi->last_pos;
2378        }
2379
2380        rhashtable_walk_start(&gi->hti);
2381
2382        gfs2_glock_iter_next(gi, n);
2383        gi->last_pos = *pos;
2384        return gi->gl;
2385}
2386
2387static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2388                                 loff_t *pos)
2389{
2390        struct gfs2_glock_iter *gi = seq->private;
2391
2392        (*pos)++;
2393        gi->last_pos = *pos;
2394        gfs2_glock_iter_next(gi, 1);
2395        return gi->gl;
2396}
2397
2398static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2399        __releases(RCU)
2400{
2401        struct gfs2_glock_iter *gi = seq->private;
2402
2403        rhashtable_walk_stop(&gi->hti);
2404}
2405
2406static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2407{
2408        dump_glock(seq, iter_ptr, false);
2409        return 0;
2410}
2411
2412static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2413{
2414        preempt_disable();
2415        if (*pos >= GFS2_NR_SBSTATS)
2416                return NULL;
2417        return pos;
2418}
2419
2420static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2421                                   loff_t *pos)
2422{
2423        (*pos)++;
2424        if (*pos >= GFS2_NR_SBSTATS)
2425                return NULL;
2426        return pos;
2427}
2428
2429static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2430{
2431        preempt_enable();
2432}
2433
2434static const struct seq_operations gfs2_glock_seq_ops = {
2435        .start = gfs2_glock_seq_start,
2436        .next  = gfs2_glock_seq_next,
2437        .stop  = gfs2_glock_seq_stop,
2438        .show  = gfs2_glock_seq_show,
2439};
2440
2441static const struct seq_operations gfs2_glstats_seq_ops = {
2442        .start = gfs2_glock_seq_start,
2443        .next  = gfs2_glock_seq_next,
2444        .stop  = gfs2_glock_seq_stop,
2445        .show  = gfs2_glstats_seq_show,
2446};
2447
2448static const struct seq_operations gfs2_sbstats_sops = {
2449        .start = gfs2_sbstats_seq_start,
2450        .next  = gfs2_sbstats_seq_next,
2451        .stop  = gfs2_sbstats_seq_stop,
2452        .show  = gfs2_sbstats_seq_show,
2453};
2454
2455#define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2456
2457static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2458                              const struct seq_operations *ops)
2459{
2460        int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2461        if (ret == 0) {
2462                struct seq_file *seq = file->private_data;
2463                struct gfs2_glock_iter *gi = seq->private;
2464
2465                gi->sdp = inode->i_private;
2466                seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2467                if (seq->buf)
2468                        seq->size = GFS2_SEQ_GOODSIZE;
2469                /*
2470                 * Initially, we are "before" the first hash table entry; the
2471                 * first call to rhashtable_walk_next gets us the first entry.
2472                 */
2473                gi->last_pos = -1;
2474                gi->gl = NULL;
2475                rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2476        }
2477        return ret;
2478}
2479
2480static int gfs2_glocks_open(struct inode *inode, struct file *file)
2481{
2482        return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2483}
2484
2485static int gfs2_glocks_release(struct inode *inode, struct file *file)
2486{
2487        struct seq_file *seq = file->private_data;
2488        struct gfs2_glock_iter *gi = seq->private;
2489
2490        if (gi->gl)
2491                gfs2_glock_put(gi->gl);
2492        rhashtable_walk_exit(&gi->hti);
2493        return seq_release_private(inode, file);
2494}
2495
2496static int gfs2_glstats_open(struct inode *inode, struct file *file)
2497{
2498        return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2499}
2500
2501static const struct file_operations gfs2_glocks_fops = {
2502        .owner   = THIS_MODULE,
2503        .open    = gfs2_glocks_open,
2504        .read    = seq_read,
2505        .llseek  = seq_lseek,
2506        .release = gfs2_glocks_release,
2507};
2508
2509static const struct file_operations gfs2_glstats_fops = {
2510        .owner   = THIS_MODULE,
2511        .open    = gfs2_glstats_open,
2512        .read    = seq_read,
2513        .llseek  = seq_lseek,
2514        .release = gfs2_glocks_release,
2515};
2516
2517DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats);
2518
2519void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2520{
2521        sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2522
2523        debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2524                            &gfs2_glocks_fops);
2525
2526        debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2527                            &gfs2_glstats_fops);
2528
2529        debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2530                            &gfs2_sbstats_fops);
2531}
2532
2533void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2534{
2535        debugfs_remove_recursive(sdp->debugfs_dir);
2536        sdp->debugfs_dir = NULL;
2537}
2538
2539void gfs2_register_debugfs(void)
2540{
2541        gfs2_root = debugfs_create_dir("gfs2", NULL);
2542}
2543
2544void gfs2_unregister_debugfs(void)
2545{
2546        debugfs_remove(gfs2_root);
2547        gfs2_root = NULL;
2548}
2549