1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
48 #define CREATE_TRACE_POINTS
49 #include "trace_gfs2.h"
51 struct 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 */
58 typedef void (*glock_examiner) (struct gfs2_glock * gl);
60 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
62 static struct dentry *gfs2_root;
63 static struct workqueue_struct *glock_workqueue;
64 struct workqueue_struct *gfs2_delete_workqueue;
65 static LIST_HEAD(lru_list);
66 static atomic_t lru_count = ATOMIC_INIT(0);
67 static DEFINE_SPINLOCK(lru_lock);
69 #define GFS2_GL_HASH_SHIFT 15
70 #define GFS2_GL_HASH_SIZE BIT(GFS2_GL_HASH_SHIFT)
72 static 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),
79 static struct rhashtable gl_hash_table;
81 #define GLOCK_WAIT_TABLE_BITS 12
82 #define GLOCK_WAIT_TABLE_SIZE (1 << GLOCK_WAIT_TABLE_BITS)
83 static wait_queue_head_t glock_wait_table[GLOCK_WAIT_TABLE_SIZE] __cacheline_aligned;
85 struct wait_glock_queue {
86 struct lm_lockname *name;
87 wait_queue_entry_t wait;
90 static int glock_wake_function(wait_queue_entry_t *wait, unsigned int mode,
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;
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)
102 return autoremove_wake_function(wait, mode, sync, key);
105 static wait_queue_head_t *glock_waitqueue(struct lm_lockname *name)
107 u32 hash = jhash2((u32 *)name, ht_parms.key_len / 4, 0);
109 return glock_wait_table + hash_32(hash, GLOCK_WAIT_TABLE_BITS);
113 * wake_up_glock - Wake up waiters on a glock
116 static void wake_up_glock(struct gfs2_glock *gl)
118 wait_queue_head_t *wq = glock_waitqueue(&gl->gl_name);
120 if (waitqueue_active(wq))
121 __wake_up(wq, TASK_NORMAL, 1, &gl->gl_name);
124 static void gfs2_glock_dealloc(struct rcu_head *rcu)
126 struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
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);
132 kmem_cache_free(gfs2_glock_cachep, gl);
136 * glock_blocked_by_withdraw - determine if we can still use a glock
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.
148 static bool glock_blocked_by_withdraw(struct gfs2_glock *gl)
150 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
152 if (likely(!gfs2_withdrawn(sdp)))
154 if (gl->gl_ops->go_flags & GLOF_NONDISK)
156 if (!sdp->sd_jdesc ||
157 gl->gl_name.ln_number == sdp->sd_jdesc->jd_no_addr)
162 void gfs2_glock_free(struct gfs2_glock *gl)
164 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
166 gfs2_glock_assert_withdraw(gl, atomic_read(&gl->gl_revokes) == 0);
167 rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
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);
176 * gfs2_glock_hold() - increment reference count on glock
177 * @gl: The glock to hold
181 void gfs2_glock_hold(struct gfs2_glock *gl)
183 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
184 lockref_get(&gl->gl_lockref);
188 * demote_ok - Check to see if it's ok to unlock a glock
191 * Returns: 1 if it's ok
194 static int demote_ok(const struct gfs2_glock *gl)
196 const struct gfs2_glock_operations *glops = gl->gl_ops;
198 if (gl->gl_state == LM_ST_UNLOCKED)
200 if (!list_empty(&gl->gl_holders))
202 if (glops->go_demote_ok)
203 return glops->go_demote_ok(gl);
208 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
210 if (!(gl->gl_ops->go_flags & GLOF_LRU))
213 spin_lock(&lru_lock);
215 list_del(&gl->gl_lru);
216 list_add_tail(&gl->gl_lru, &lru_list);
218 if (!test_bit(GLF_LRU, &gl->gl_flags)) {
219 set_bit(GLF_LRU, &gl->gl_flags);
220 atomic_inc(&lru_count);
223 spin_unlock(&lru_lock);
226 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
228 if (!(gl->gl_ops->go_flags & GLOF_LRU))
231 spin_lock(&lru_lock);
232 if (test_bit(GLF_LRU, &gl->gl_flags)) {
233 list_del_init(&gl->gl_lru);
234 atomic_dec(&lru_count);
235 clear_bit(GLF_LRU, &gl->gl_flags);
237 spin_unlock(&lru_lock);
241 * Enqueue the glock on the work queue. Passes one glock reference on to the
244 static void __gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
245 if (!queue_delayed_work(glock_workqueue, &gl->gl_work, delay)) {
247 * We are holding the lockref spinlock, and the work was still
248 * queued above. The queued work (glock_work_func) takes that
249 * spinlock before dropping its glock reference(s), so it
250 * cannot have dropped them in the meantime.
252 GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2);
253 gl->gl_lockref.count--;
257 static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
258 spin_lock(&gl->gl_lockref.lock);
259 __gfs2_glock_queue_work(gl, delay);
260 spin_unlock(&gl->gl_lockref.lock);
263 static void __gfs2_glock_put(struct gfs2_glock *gl)
265 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
266 struct address_space *mapping = gfs2_glock2aspace(gl);
268 lockref_mark_dead(&gl->gl_lockref);
270 gfs2_glock_remove_from_lru(gl);
271 spin_unlock(&gl->gl_lockref.lock);
272 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
274 truncate_inode_pages_final(mapping);
275 if (!gfs2_withdrawn(sdp))
276 GLOCK_BUG_ON(gl, mapping->nrpages ||
277 mapping->nrexceptional);
279 trace_gfs2_glock_put(gl);
280 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
284 * Cause the glock to be put in work queue context.
286 void gfs2_glock_queue_put(struct gfs2_glock *gl)
288 gfs2_glock_queue_work(gl, 0);
292 * gfs2_glock_put() - Decrement reference count on glock
293 * @gl: The glock to put
297 void gfs2_glock_put(struct gfs2_glock *gl)
299 if (lockref_put_or_lock(&gl->gl_lockref))
302 __gfs2_glock_put(gl);
306 * may_grant - check if its ok to grant a new lock
308 * @gh: The lock request which we wish to grant
310 * Returns: true if its ok to grant the lock
313 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
315 const struct gfs2_holder *gh_head = list_first_entry(&gl->gl_holders, const struct gfs2_holder, gh_list);
316 if ((gh->gh_state == LM_ST_EXCLUSIVE ||
317 gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
319 if (gl->gl_state == gh->gh_state)
321 if (gh->gh_flags & GL_EXACT)
323 if (gl->gl_state == LM_ST_EXCLUSIVE) {
324 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
326 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
329 if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
334 static void gfs2_holder_wake(struct gfs2_holder *gh)
336 clear_bit(HIF_WAIT, &gh->gh_iflags);
337 smp_mb__after_atomic();
338 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
339 if (gh->gh_flags & GL_ASYNC) {
340 struct gfs2_sbd *sdp = gh->gh_gl->gl_name.ln_sbd;
342 wake_up(&sdp->sd_async_glock_wait);
347 * do_error - Something unexpected has happened during a lock request
351 static void do_error(struct gfs2_glock *gl, const int ret)
353 struct gfs2_holder *gh, *tmp;
355 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
356 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
358 if (ret & LM_OUT_ERROR)
360 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
361 gh->gh_error = GLR_TRYFAILED;
364 list_del_init(&gh->gh_list);
365 trace_gfs2_glock_queue(gh, 0);
366 gfs2_holder_wake(gh);
371 * do_promote - promote as many requests as possible on the current queue
374 * Returns: 1 if there is a blocked holder at the head of the list, or 2
375 * if a type specific operation is underway.
378 static int do_promote(struct gfs2_glock *gl)
379 __releases(&gl->gl_lockref.lock)
380 __acquires(&gl->gl_lockref.lock)
382 const struct gfs2_glock_operations *glops = gl->gl_ops;
383 struct gfs2_holder *gh, *tmp;
387 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
388 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
390 if (may_grant(gl, gh)) {
391 if (gh->gh_list.prev == &gl->gl_holders &&
393 spin_unlock(&gl->gl_lockref.lock);
394 /* FIXME: eliminate this eventually */
395 ret = glops->go_lock(gh);
396 spin_lock(&gl->gl_lockref.lock);
401 list_del_init(&gh->gh_list);
402 trace_gfs2_glock_queue(gh, 0);
403 gfs2_holder_wake(gh);
406 set_bit(HIF_HOLDER, &gh->gh_iflags);
407 trace_gfs2_promote(gh, 1);
408 gfs2_holder_wake(gh);
411 set_bit(HIF_HOLDER, &gh->gh_iflags);
412 trace_gfs2_promote(gh, 0);
413 gfs2_holder_wake(gh);
416 if (gh->gh_list.prev == &gl->gl_holders)
425 * find_first_waiter - find the first gh that's waiting for the glock
429 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
431 struct gfs2_holder *gh;
433 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
434 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
441 * state_change - record that the glock is now in a different state
443 * @new_state the new state
447 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
451 held1 = (gl->gl_state != LM_ST_UNLOCKED);
452 held2 = (new_state != LM_ST_UNLOCKED);
454 if (held1 != held2) {
455 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
457 gl->gl_lockref.count++;
459 gl->gl_lockref.count--;
461 if (new_state != gl->gl_target)
462 /* shorten our minimum hold time */
463 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
465 gl->gl_state = new_state;
466 gl->gl_tchange = jiffies;
469 static void gfs2_set_demote(struct gfs2_glock *gl)
471 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
473 set_bit(GLF_DEMOTE, &gl->gl_flags);
475 wake_up(&sdp->sd_async_glock_wait);
478 static void gfs2_demote_wake(struct gfs2_glock *gl)
480 gl->gl_demote_state = LM_ST_EXCLUSIVE;
481 clear_bit(GLF_DEMOTE, &gl->gl_flags);
482 smp_mb__after_atomic();
483 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
487 * finish_xmote - The DLM has replied to one of our lock requests
489 * @ret: The status from the DLM
493 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
495 const struct gfs2_glock_operations *glops = gl->gl_ops;
496 struct gfs2_holder *gh;
497 unsigned state = ret & LM_OUT_ST_MASK;
500 spin_lock(&gl->gl_lockref.lock);
501 trace_gfs2_glock_state_change(gl, state);
502 state_change(gl, state);
503 gh = find_first_waiter(gl);
505 /* Demote to UN request arrived during demote to SH or DF */
506 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
507 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
508 gl->gl_target = LM_ST_UNLOCKED;
510 /* Check for state != intended state */
511 if (unlikely(state != gl->gl_target)) {
512 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
513 /* move to back of queue and try next entry */
514 if (ret & LM_OUT_CANCELED) {
515 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
516 list_move_tail(&gh->gh_list, &gl->gl_holders);
517 gh = find_first_waiter(gl);
518 gl->gl_target = gh->gh_state;
521 /* Some error or failed "try lock" - report it */
522 if ((ret & LM_OUT_ERROR) ||
523 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
524 gl->gl_target = gl->gl_state;
530 /* Unlocked due to conversion deadlock, try again */
533 do_xmote(gl, gh, gl->gl_target);
535 /* Conversion fails, unlock and try again */
538 do_xmote(gl, gh, LM_ST_UNLOCKED);
540 default: /* Everything else */
541 fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n",
542 gl->gl_target, state);
545 spin_unlock(&gl->gl_lockref.lock);
549 /* Fast path - we got what we asked for */
550 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
551 gfs2_demote_wake(gl);
552 if (state != LM_ST_UNLOCKED) {
553 if (glops->go_xmote_bh) {
554 spin_unlock(&gl->gl_lockref.lock);
555 rv = glops->go_xmote_bh(gl, gh);
556 spin_lock(&gl->gl_lockref.lock);
567 clear_bit(GLF_LOCK, &gl->gl_flags);
569 spin_unlock(&gl->gl_lockref.lock);
573 * do_xmote - Calls the DLM to change the state of a lock
574 * @gl: The lock state
575 * @gh: The holder (only for promotes)
576 * @target: The target lock state
580 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
581 __releases(&gl->gl_lockref.lock)
582 __acquires(&gl->gl_lockref.lock)
584 const struct gfs2_glock_operations *glops = gl->gl_ops;
585 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
586 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
589 if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) &&
590 gh && !(gh->gh_flags & LM_FLAG_NOEXP))
592 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
594 GLOCK_BUG_ON(gl, gl->gl_state == target);
595 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
596 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
599 * If another process is already doing the invalidate, let that
600 * finish first. The glock state machine will get back to this
601 * holder again later.
603 if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
606 do_error(gl, 0); /* Fail queued try locks */
609 set_bit(GLF_BLOCKING, &gl->gl_flags);
610 if ((gl->gl_req == LM_ST_UNLOCKED) ||
611 (gl->gl_state == LM_ST_EXCLUSIVE) ||
612 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
613 clear_bit(GLF_BLOCKING, &gl->gl_flags);
614 spin_unlock(&gl->gl_lockref.lock);
615 if (glops->go_sync) {
616 ret = glops->go_sync(gl);
617 /* If we had a problem syncing (due to io errors or whatever,
618 * we should not invalidate the metadata or tell dlm to
619 * release the glock to other nodes.
622 if (cmpxchg(&sdp->sd_log_error, 0, ret)) {
623 fs_err(sdp, "Error %d syncing glock \n", ret);
624 gfs2_dump_glock(NULL, gl, true);
629 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) {
631 * The call to go_sync should have cleared out the ail list.
632 * If there are still items, we have a problem. We ought to
633 * withdraw, but we can't because the withdraw code also uses
634 * glocks. Warn about the error, dump the glock, then fall
635 * through and wait for logd to do the withdraw for us.
637 if ((atomic_read(&gl->gl_ail_count) != 0) &&
638 (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) {
639 gfs2_glock_assert_warn(gl,
640 !atomic_read(&gl->gl_ail_count));
641 gfs2_dump_glock(NULL, gl, true);
643 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
644 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
650 * Check for an error encountered since we called go_sync and go_inval.
651 * If so, we can't withdraw from the glock code because the withdraw
652 * code itself uses glocks (see function signal_our_withdraw) to
653 * change the mount to read-only. Most importantly, we must not call
654 * dlm to unlock the glock until the journal is in a known good state
655 * (after journal replay) otherwise other nodes may use the object
656 * (rgrp or dinode) and then later, journal replay will corrupt the
657 * file system. The best we can do here is wait for the logd daemon
658 * to see sd_log_error and withdraw, and in the meantime, requeue the
661 * However, if we're just unlocking the lock (say, for unmount, when
662 * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
663 * then it's okay to tell dlm to unlock it.
665 if (unlikely(sdp->sd_log_error && !gfs2_withdrawn(sdp)))
666 gfs2_withdraw_delayed(sdp);
667 if (glock_blocked_by_withdraw(gl)) {
668 if (target != LM_ST_UNLOCKED ||
669 test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags)) {
670 gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD);
675 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
677 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
678 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
679 target == LM_ST_UNLOCKED &&
680 test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
681 finish_xmote(gl, target);
682 gfs2_glock_queue_work(gl, 0);
684 fs_err(sdp, "lm_lock ret %d\n", ret);
685 GLOCK_BUG_ON(gl, !gfs2_withdrawn(sdp));
687 } else { /* lock_nolock */
688 finish_xmote(gl, target);
689 gfs2_glock_queue_work(gl, 0);
692 spin_lock(&gl->gl_lockref.lock);
696 * find_first_holder - find the first "holder" gh
700 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
702 struct gfs2_holder *gh;
704 if (!list_empty(&gl->gl_holders)) {
705 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
706 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
713 * run_queue - do all outstanding tasks related to a glock
714 * @gl: The glock in question
715 * @nonblock: True if we must not block in run_queue
719 static void run_queue(struct gfs2_glock *gl, const int nonblock)
720 __releases(&gl->gl_lockref.lock)
721 __acquires(&gl->gl_lockref.lock)
723 struct gfs2_holder *gh = NULL;
726 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
729 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
731 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
732 gl->gl_demote_state != gl->gl_state) {
733 if (find_first_holder(gl))
737 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
738 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
739 gl->gl_target = gl->gl_demote_state;
741 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
742 gfs2_demote_wake(gl);
743 ret = do_promote(gl);
748 gh = find_first_waiter(gl);
749 gl->gl_target = gh->gh_state;
750 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
751 do_error(gl, 0); /* Fail queued try locks */
753 do_xmote(gl, gh, gl->gl_target);
758 clear_bit(GLF_LOCK, &gl->gl_flags);
759 smp_mb__after_atomic();
760 gl->gl_lockref.count++;
761 __gfs2_glock_queue_work(gl, 0);
765 clear_bit(GLF_LOCK, &gl->gl_flags);
766 smp_mb__after_atomic();
770 void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation)
772 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
774 if (ri->ri_magic == 0)
775 ri->ri_magic = cpu_to_be32(GFS2_MAGIC);
776 if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC))
777 ri->ri_generation_deleted = cpu_to_be64(generation);
780 bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation)
782 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
784 if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC))
786 return generation <= be64_to_cpu(ri->ri_generation_deleted);
789 static void gfs2_glock_poke(struct gfs2_glock *gl)
791 int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP;
792 struct gfs2_holder gh;
795 gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh);
796 error = gfs2_glock_nq(&gh);
799 gfs2_holder_uninit(&gh);
802 static bool gfs2_try_evict(struct gfs2_glock *gl)
804 struct gfs2_inode *ip;
805 bool evicted = false;
808 * If there is contention on the iopen glock and we have an inode, try
809 * to grab and release the inode so that it can be evicted. This will
810 * allow the remote node to go ahead and delete the inode without us
811 * having to do it, which will avoid rgrp glock thrashing.
813 * The remote node is likely still holding the corresponding inode
814 * glock, so it will run before we get to verify that the delete has
817 spin_lock(&gl->gl_lockref.lock);
819 if (ip && !igrab(&ip->i_inode))
821 spin_unlock(&gl->gl_lockref.lock);
823 struct gfs2_glock *inode_gl = NULL;
825 gl->gl_no_formal_ino = ip->i_no_formal_ino;
826 set_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
827 d_prune_aliases(&ip->i_inode);
830 /* If the inode was evicted, gl->gl_object will now be NULL. */
831 spin_lock(&gl->gl_lockref.lock);
835 lockref_get(&inode_gl->gl_lockref);
836 clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
838 spin_unlock(&gl->gl_lockref.lock);
840 gfs2_glock_poke(inode_gl);
841 gfs2_glock_put(inode_gl);
848 static void delete_work_func(struct work_struct *work)
850 struct delayed_work *dwork = to_delayed_work(work);
851 struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete);
852 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
854 u64 no_addr = gl->gl_name.ln_number;
856 spin_lock(&gl->gl_lockref.lock);
857 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
858 spin_unlock(&gl->gl_lockref.lock);
860 /* If someone's using this glock to create a new dinode, the block must
861 have been freed by another node, then re-used, in which case our
862 iopen callback is too late after the fact. Ignore it. */
863 if (test_bit(GLF_INODE_CREATING, &gl->gl_flags))
866 if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
868 * If we can evict the inode, give the remote node trying to
869 * delete the inode some time before verifying that the delete
870 * has happened. Otherwise, if we cause contention on the inode glock
871 * immediately, the remote node will think that we still have
872 * the inode in use, and so it will give up waiting.
874 * If we can't evict the inode, signal to the remote node that
875 * the inode is still in use. We'll later try to delete the
876 * inode locally in gfs2_evict_inode.
878 * FIXME: We only need to verify that the remote node has
879 * deleted the inode because nodes before this remote delete
880 * rework won't cooperate. At a later time, when we no longer
881 * care about compatibility with such nodes, we can skip this
884 if (gfs2_try_evict(gl)) {
885 if (gfs2_queue_delete_work(gl, 5 * HZ))
891 inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino,
892 GFS2_BLKST_UNLINKED);
893 if (!IS_ERR_OR_NULL(inode)) {
894 d_prune_aliases(inode);
901 static void glock_work_func(struct work_struct *work)
903 unsigned long delay = 0;
904 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
905 unsigned int drop_refs = 1;
907 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
908 finish_xmote(gl, gl->gl_reply);
911 spin_lock(&gl->gl_lockref.lock);
912 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
913 gl->gl_state != LM_ST_UNLOCKED &&
914 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
915 unsigned long holdtime, now = jiffies;
917 holdtime = gl->gl_tchange + gl->gl_hold_time;
918 if (time_before(now, holdtime))
919 delay = holdtime - now;
922 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
928 /* Keep one glock reference for the work we requeue. */
930 if (gl->gl_name.ln_type != LM_TYPE_INODE)
932 __gfs2_glock_queue_work(gl, delay);
936 * Drop the remaining glock references manually here. (Mind that
937 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
940 gl->gl_lockref.count -= drop_refs;
941 if (!gl->gl_lockref.count) {
942 __gfs2_glock_put(gl);
945 spin_unlock(&gl->gl_lockref.lock);
948 static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
949 struct gfs2_glock *new)
951 struct wait_glock_queue wait;
952 wait_queue_head_t *wq = glock_waitqueue(name);
953 struct gfs2_glock *gl;
956 init_wait(&wait.wait);
957 wait.wait.func = glock_wake_function;
960 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
963 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
964 &new->gl_node, ht_parms);
968 gl = rhashtable_lookup_fast(&gl_hash_table,
971 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
978 finish_wait(wq, &wait.wait);
983 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
984 * @sdp: The GFS2 superblock
985 * @number: the lock number
986 * @glops: The glock_operations to use
987 * @create: If 0, don't create the glock if it doesn't exist
988 * @glp: the glock is returned here
990 * This does not lock a glock, just finds/creates structures for one.
995 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
996 const struct gfs2_glock_operations *glops, int create,
997 struct gfs2_glock **glp)
999 struct super_block *s = sdp->sd_vfs;
1000 struct lm_lockname name = { .ln_number = number,
1001 .ln_type = glops->go_type,
1003 struct gfs2_glock *gl, *tmp;
1004 struct address_space *mapping;
1005 struct kmem_cache *cachep;
1008 gl = find_insert_glock(&name, NULL);
1016 if (glops->go_flags & GLOF_ASPACE)
1017 cachep = gfs2_glock_aspace_cachep;
1019 cachep = gfs2_glock_cachep;
1020 gl = kmem_cache_alloc(cachep, GFP_NOFS);
1024 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
1026 if (glops->go_flags & GLOF_LVB) {
1027 gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1028 if (!gl->gl_lksb.sb_lvbptr) {
1029 kmem_cache_free(cachep, gl);
1034 atomic_inc(&sdp->sd_glock_disposal);
1035 gl->gl_node.next = NULL;
1038 lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass);
1039 gl->gl_lockref.count = 1;
1040 gl->gl_state = LM_ST_UNLOCKED;
1041 gl->gl_target = LM_ST_UNLOCKED;
1042 gl->gl_demote_state = LM_ST_EXCLUSIVE;
1046 /* We use the global stats to estimate the initial per-glock stats */
1047 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
1049 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
1050 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
1051 gl->gl_tchange = jiffies;
1052 gl->gl_object = NULL;
1053 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
1054 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
1055 if (gl->gl_name.ln_type == LM_TYPE_IOPEN)
1056 INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func);
1058 mapping = gfs2_glock2aspace(gl);
1060 mapping->a_ops = &gfs2_meta_aops;
1061 mapping->host = s->s_bdev->bd_inode;
1063 mapping_set_gfp_mask(mapping, GFP_NOFS);
1064 mapping->private_data = NULL;
1065 mapping->writeback_index = 0;
1068 tmp = find_insert_glock(&name, gl);
1080 kfree(gl->gl_lksb.sb_lvbptr);
1081 kmem_cache_free(cachep, gl);
1082 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
1083 wake_up(&sdp->sd_glock_wait);
1090 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
1092 * @state: the state we're requesting
1093 * @flags: the modifier flags
1094 * @gh: the holder structure
1098 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
1099 struct gfs2_holder *gh)
1101 INIT_LIST_HEAD(&gh->gh_list);
1103 gh->gh_ip = _RET_IP_;
1104 gh->gh_owner_pid = get_pid(task_pid(current));
1105 gh->gh_state = state;
1106 gh->gh_flags = flags;
1109 gfs2_glock_hold(gl);
1113 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1114 * @state: the state we're requesting
1115 * @flags: the modifier flags
1116 * @gh: the holder structure
1118 * Don't mess with the glock.
1122 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
1124 gh->gh_state = state;
1125 gh->gh_flags = flags;
1127 gh->gh_ip = _RET_IP_;
1128 put_pid(gh->gh_owner_pid);
1129 gh->gh_owner_pid = get_pid(task_pid(current));
1133 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1134 * @gh: the holder structure
1138 void gfs2_holder_uninit(struct gfs2_holder *gh)
1140 put_pid(gh->gh_owner_pid);
1141 gfs2_glock_put(gh->gh_gl);
1142 gfs2_holder_mark_uninitialized(gh);
1146 static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
1147 unsigned long start_time)
1149 /* Have we waited longer that a second? */
1150 if (time_after(jiffies, start_time + HZ)) {
1151 /* Lengthen the minimum hold time. */
1152 gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
1158 * gfs2_glock_wait - wait on a glock acquisition
1159 * @gh: the glock holder
1161 * Returns: 0 on success
1164 int gfs2_glock_wait(struct gfs2_holder *gh)
1166 unsigned long start_time = jiffies;
1169 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1170 gfs2_glock_update_hold_time(gh->gh_gl, start_time);
1171 return gh->gh_error;
1174 static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
1178 for (i = 0; i < num_gh; i++)
1179 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
1185 * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1186 * @num_gh: the number of holders in the array
1187 * @ghs: the glock holder array
1189 * Returns: 0 on success, meaning all glocks have been granted and are held.
1190 * -ESTALE if the request timed out, meaning all glocks were released,
1191 * and the caller should retry the operation.
1194 int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
1196 struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
1197 int i, ret = 0, timeout = 0;
1198 unsigned long start_time = jiffies;
1203 * Total up the (minimum hold time * 2) of all glocks and use that to
1204 * determine the max amount of time we should wait.
1206 for (i = 0; i < num_gh; i++)
1207 timeout += ghs[i].gh_gl->gl_hold_time << 1;
1210 if (!wait_event_timeout(sdp->sd_async_glock_wait,
1211 !glocks_pending(num_gh, ghs), timeout))
1212 ret = -ESTALE; /* request timed out. */
1215 * If dlm granted all our requests, we need to adjust the glock
1216 * minimum hold time values according to how long we waited.
1218 * If our request timed out, we need to repeatedly release any held
1219 * glocks we acquired thus far to allow dlm to acquire the remaining
1220 * glocks without deadlocking. We cannot currently cancel outstanding
1221 * glock acquisitions.
1223 * The HIF_WAIT bit tells us which requests still need a response from
1226 * If dlm sent us any errors, we return the first error we find.
1228 keep_waiting = false;
1229 for (i = 0; i < num_gh; i++) {
1230 /* Skip holders we have already dequeued below. */
1231 if (!gfs2_holder_queued(&ghs[i]))
1233 /* Skip holders with a pending DLM response. */
1234 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags)) {
1235 keep_waiting = true;
1239 if (test_bit(HIF_HOLDER, &ghs[i].gh_iflags)) {
1241 gfs2_glock_dq(&ghs[i]);
1243 gfs2_glock_update_hold_time(ghs[i].gh_gl,
1247 ret = ghs[i].gh_error;
1254 * At this point, we've either acquired all locks or released them all.
1260 * handle_callback - process a demote request
1262 * @state: the state the caller wants us to change to
1264 * There are only two requests that we are going to see in actual
1265 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1268 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
1269 unsigned long delay, bool remote)
1272 set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1274 gfs2_set_demote(gl);
1275 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
1276 gl->gl_demote_state = state;
1277 gl->gl_demote_time = jiffies;
1278 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
1279 gl->gl_demote_state != state) {
1280 gl->gl_demote_state = LM_ST_UNLOCKED;
1282 if (gl->gl_ops->go_callback)
1283 gl->gl_ops->go_callback(gl, remote);
1284 trace_gfs2_demote_rq(gl, remote);
1287 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
1289 struct va_format vaf;
1292 va_start(args, fmt);
1295 seq_vprintf(seq, fmt, args);
1300 pr_err("%pV", &vaf);
1307 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1308 * @gh: the holder structure to add
1310 * Eventually we should move the recursive locking trap to a
1311 * debugging option or something like that. This is the fast
1312 * path and needs to have the minimum number of distractions.
1316 static inline void add_to_queue(struct gfs2_holder *gh)
1317 __releases(&gl->gl_lockref.lock)
1318 __acquires(&gl->gl_lockref.lock)
1320 struct gfs2_glock *gl = gh->gh_gl;
1321 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1322 struct list_head *insert_pt = NULL;
1323 struct gfs2_holder *gh2;
1326 GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1327 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1328 GLOCK_BUG_ON(gl, true);
1330 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1331 if (test_bit(GLF_LOCK, &gl->gl_flags))
1332 try_futile = !may_grant(gl, gh);
1333 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1337 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1338 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
1339 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
1340 goto trap_recursive;
1342 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1344 gh->gh_error = GLR_TRYFAILED;
1345 gfs2_holder_wake(gh);
1348 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1350 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1351 insert_pt = &gh2->gh_list;
1353 trace_gfs2_glock_queue(gh, 1);
1354 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1355 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1356 if (likely(insert_pt == NULL)) {
1357 list_add_tail(&gh->gh_list, &gl->gl_holders);
1358 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1362 list_add_tail(&gh->gh_list, insert_pt);
1364 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
1365 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1366 spin_unlock(&gl->gl_lockref.lock);
1367 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1368 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1369 spin_lock(&gl->gl_lockref.lock);
1374 fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1375 fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1376 fs_err(sdp, "lock type: %d req lock state : %d\n",
1377 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1378 fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1379 fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1380 fs_err(sdp, "lock type: %d req lock state : %d\n",
1381 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1382 gfs2_dump_glock(NULL, gl, true);
1387 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1388 * @gh: the holder structure
1390 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1392 * Returns: 0, GLR_TRYFAILED, or errno on failure
1395 int gfs2_glock_nq(struct gfs2_holder *gh)
1397 struct gfs2_glock *gl = gh->gh_gl;
1400 if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP))
1403 if (test_bit(GLF_LRU, &gl->gl_flags))
1404 gfs2_glock_remove_from_lru(gl);
1406 spin_lock(&gl->gl_lockref.lock);
1408 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1409 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1410 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1411 gl->gl_lockref.count++;
1412 __gfs2_glock_queue_work(gl, 0);
1415 spin_unlock(&gl->gl_lockref.lock);
1417 if (!(gh->gh_flags & GL_ASYNC))
1418 error = gfs2_glock_wait(gh);
1424 * gfs2_glock_poll - poll to see if an async request has been completed
1427 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1430 int gfs2_glock_poll(struct gfs2_holder *gh)
1432 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1436 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1437 * @gh: the glock holder
1441 void gfs2_glock_dq(struct gfs2_holder *gh)
1443 struct gfs2_glock *gl = gh->gh_gl;
1444 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1448 spin_lock(&gl->gl_lockref.lock);
1450 * If we're in the process of file system withdraw, we cannot just
1451 * dequeue any glocks until our journal is recovered, lest we
1452 * introduce file system corruption. We need two exceptions to this
1453 * rule: We need to allow unlocking of nondisk glocks and the glock
1454 * for our own journal that needs recovery.
1456 if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) &&
1457 glock_blocked_by_withdraw(gl) &&
1458 gh->gh_gl != sdp->sd_jinode_gl) {
1459 sdp->sd_glock_dqs_held++;
1461 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
1462 TASK_UNINTERRUPTIBLE);
1464 if (gh->gh_flags & GL_NOCACHE)
1465 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1467 list_del_init(&gh->gh_list);
1468 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1469 if (find_first_holder(gl) == NULL) {
1470 if (list_empty(&gl->gl_holders) &&
1471 !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1472 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1475 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1476 gfs2_glock_add_to_lru(gl);
1478 trace_gfs2_glock_queue(gh, 0);
1479 if (unlikely(!fast_path)) {
1480 gl->gl_lockref.count++;
1481 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1482 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1483 gl->gl_name.ln_type == LM_TYPE_INODE)
1484 delay = gl->gl_hold_time;
1485 __gfs2_glock_queue_work(gl, delay);
1487 spin_unlock(&gl->gl_lockref.lock);
1490 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1492 struct gfs2_glock *gl = gh->gh_gl;
1495 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1499 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1500 * @gh: the holder structure
1504 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1507 gfs2_holder_uninit(gh);
1511 * gfs2_glock_nq_num - acquire a glock based on lock number
1512 * @sdp: the filesystem
1513 * @number: the lock number
1514 * @glops: the glock operations for the type of glock
1515 * @state: the state to acquire the glock in
1516 * @flags: modifier flags for the acquisition
1517 * @gh: the struct gfs2_holder
1522 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1523 const struct gfs2_glock_operations *glops,
1524 unsigned int state, u16 flags, struct gfs2_holder *gh)
1526 struct gfs2_glock *gl;
1529 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1531 error = gfs2_glock_nq_init(gl, state, flags, gh);
1539 * glock_compare - Compare two struct gfs2_glock structures for sorting
1540 * @arg_a: the first structure
1541 * @arg_b: the second structure
1545 static int glock_compare(const void *arg_a, const void *arg_b)
1547 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1548 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1549 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1550 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1552 if (a->ln_number > b->ln_number)
1554 if (a->ln_number < b->ln_number)
1556 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1561 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1562 * @num_gh: the number of structures
1563 * @ghs: an array of struct gfs2_holder structures
1565 * Returns: 0 on success (all glocks acquired),
1566 * errno on failure (no glocks acquired)
1569 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1570 struct gfs2_holder **p)
1575 for (x = 0; x < num_gh; x++)
1578 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1580 for (x = 0; x < num_gh; x++) {
1581 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1583 error = gfs2_glock_nq(p[x]);
1586 gfs2_glock_dq(p[x]);
1595 * gfs2_glock_nq_m - acquire multiple glocks
1596 * @num_gh: the number of structures
1597 * @ghs: an array of struct gfs2_holder structures
1600 * Returns: 0 on success (all glocks acquired),
1601 * errno on failure (no glocks acquired)
1604 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1606 struct gfs2_holder *tmp[4];
1607 struct gfs2_holder **pph = tmp;
1614 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1615 return gfs2_glock_nq(ghs);
1619 pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *),
1625 error = nq_m_sync(num_gh, ghs, pph);
1634 * gfs2_glock_dq_m - release multiple glocks
1635 * @num_gh: the number of structures
1636 * @ghs: an array of struct gfs2_holder structures
1640 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1643 gfs2_glock_dq(&ghs[num_gh]);
1646 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1648 unsigned long delay = 0;
1649 unsigned long holdtime;
1650 unsigned long now = jiffies;
1652 gfs2_glock_hold(gl);
1653 spin_lock(&gl->gl_lockref.lock);
1654 holdtime = gl->gl_tchange + gl->gl_hold_time;
1655 if (!list_empty(&gl->gl_holders) &&
1656 gl->gl_name.ln_type == LM_TYPE_INODE) {
1657 if (time_before(now, holdtime))
1658 delay = holdtime - now;
1659 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1660 delay = gl->gl_hold_time;
1662 handle_callback(gl, state, delay, true);
1663 __gfs2_glock_queue_work(gl, delay);
1664 spin_unlock(&gl->gl_lockref.lock);
1668 * gfs2_should_freeze - Figure out if glock should be frozen
1669 * @gl: The glock in question
1671 * Glocks are not frozen if (a) the result of the dlm operation is
1672 * an error, (b) the locking operation was an unlock operation or
1673 * (c) if there is a "noexp" flagged request anywhere in the queue
1675 * Returns: 1 if freezing should occur, 0 otherwise
1678 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1680 const struct gfs2_holder *gh;
1682 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1684 if (gl->gl_target == LM_ST_UNLOCKED)
1687 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1688 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1690 if (LM_FLAG_NOEXP & gh->gh_flags)
1698 * gfs2_glock_complete - Callback used by locking
1699 * @gl: Pointer to the glock
1700 * @ret: The return value from the dlm
1702 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1703 * to use a bitfield shared with other glock state fields.
1706 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1708 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1710 spin_lock(&gl->gl_lockref.lock);
1713 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1714 if (gfs2_should_freeze(gl)) {
1715 set_bit(GLF_FROZEN, &gl->gl_flags);
1716 spin_unlock(&gl->gl_lockref.lock);
1721 gl->gl_lockref.count++;
1722 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1723 __gfs2_glock_queue_work(gl, 0);
1724 spin_unlock(&gl->gl_lockref.lock);
1727 static int glock_cmp(void *priv, struct list_head *a, struct list_head *b)
1729 struct gfs2_glock *gla, *glb;
1731 gla = list_entry(a, struct gfs2_glock, gl_lru);
1732 glb = list_entry(b, struct gfs2_glock, gl_lru);
1734 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1736 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1743 * gfs2_dispose_glock_lru - Demote a list of glocks
1744 * @list: The list to dispose of
1746 * Disposing of glocks may involve disk accesses, so that here we sort
1747 * the glocks by number (i.e. disk location of the inodes) so that if
1748 * there are any such accesses, they'll be sent in order (mostly).
1750 * Must be called under the lru_lock, but may drop and retake this
1751 * lock. While the lru_lock is dropped, entries may vanish from the
1752 * list, but no new entries will appear on the list (since it is
1756 static void gfs2_dispose_glock_lru(struct list_head *list)
1757 __releases(&lru_lock)
1758 __acquires(&lru_lock)
1760 struct gfs2_glock *gl;
1762 list_sort(NULL, list, glock_cmp);
1764 while(!list_empty(list)) {
1765 gl = list_first_entry(list, struct gfs2_glock, gl_lru);
1766 list_del_init(&gl->gl_lru);
1767 if (!spin_trylock(&gl->gl_lockref.lock)) {
1769 list_add(&gl->gl_lru, &lru_list);
1770 set_bit(GLF_LRU, &gl->gl_flags);
1771 atomic_inc(&lru_count);
1774 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1775 spin_unlock(&gl->gl_lockref.lock);
1776 goto add_back_to_lru;
1778 gl->gl_lockref.count++;
1780 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1781 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1782 __gfs2_glock_queue_work(gl, 0);
1783 spin_unlock(&gl->gl_lockref.lock);
1784 cond_resched_lock(&lru_lock);
1789 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1790 * @nr: The number of entries to scan
1792 * This function selects the entries on the LRU which are able to
1793 * be demoted, and then kicks off the process by calling
1794 * gfs2_dispose_glock_lru() above.
1797 static long gfs2_scan_glock_lru(int nr)
1799 struct gfs2_glock *gl;
1804 spin_lock(&lru_lock);
1805 while ((nr-- >= 0) && !list_empty(&lru_list)) {
1806 gl = list_first_entry(&lru_list, struct gfs2_glock, gl_lru);
1808 /* Test for being demotable */
1809 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1810 list_move(&gl->gl_lru, &dispose);
1811 atomic_dec(&lru_count);
1812 clear_bit(GLF_LRU, &gl->gl_flags);
1817 list_move(&gl->gl_lru, &skipped);
1819 list_splice(&skipped, &lru_list);
1820 if (!list_empty(&dispose))
1821 gfs2_dispose_glock_lru(&dispose);
1822 spin_unlock(&lru_lock);
1827 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1828 struct shrink_control *sc)
1830 if (!(sc->gfp_mask & __GFP_FS))
1832 return gfs2_scan_glock_lru(sc->nr_to_scan);
1835 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1836 struct shrink_control *sc)
1838 return vfs_pressure_ratio(atomic_read(&lru_count));
1841 static struct shrinker glock_shrinker = {
1842 .seeks = DEFAULT_SEEKS,
1843 .count_objects = gfs2_glock_shrink_count,
1844 .scan_objects = gfs2_glock_shrink_scan,
1848 * glock_hash_walk - Call a function for glock in a hash bucket
1849 * @examiner: the function
1850 * @sdp: the filesystem
1852 * Note that the function can be called multiple times on the same
1853 * object. So the user must ensure that the function can cope with
1857 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1859 struct gfs2_glock *gl;
1860 struct rhashtable_iter iter;
1862 rhashtable_walk_enter(&gl_hash_table, &iter);
1865 rhashtable_walk_start(&iter);
1867 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl))
1868 if (gl->gl_name.ln_sbd == sdp &&
1869 lockref_get_not_dead(&gl->gl_lockref))
1872 rhashtable_walk_stop(&iter);
1873 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
1875 rhashtable_walk_exit(&iter);
1878 bool gfs2_queue_delete_work(struct gfs2_glock *gl, unsigned long delay)
1882 spin_lock(&gl->gl_lockref.lock);
1883 queued = queue_delayed_work(gfs2_delete_workqueue,
1884 &gl->gl_delete, delay);
1886 set_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1887 spin_unlock(&gl->gl_lockref.lock);
1891 void gfs2_cancel_delete_work(struct gfs2_glock *gl)
1893 if (cancel_delayed_work_sync(&gl->gl_delete)) {
1894 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1899 bool gfs2_delete_work_queued(const struct gfs2_glock *gl)
1901 return test_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1904 static void flush_delete_work(struct gfs2_glock *gl)
1906 if (gl->gl_name.ln_type == LM_TYPE_IOPEN) {
1907 if (cancel_delayed_work(&gl->gl_delete)) {
1908 queue_delayed_work(gfs2_delete_workqueue,
1912 gfs2_glock_queue_work(gl, 0);
1915 void gfs2_flush_delete_work(struct gfs2_sbd *sdp)
1917 glock_hash_walk(flush_delete_work, sdp);
1918 flush_workqueue(gfs2_delete_workqueue);
1922 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1923 * @gl: The glock to thaw
1927 static void thaw_glock(struct gfs2_glock *gl)
1929 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags)) {
1933 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1934 gfs2_glock_queue_work(gl, 0);
1938 * clear_glock - look at a glock and see if we can free it from glock cache
1939 * @gl: the glock to look at
1943 static void clear_glock(struct gfs2_glock *gl)
1945 gfs2_glock_remove_from_lru(gl);
1947 spin_lock(&gl->gl_lockref.lock);
1948 if (gl->gl_state != LM_ST_UNLOCKED)
1949 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1950 __gfs2_glock_queue_work(gl, 0);
1951 spin_unlock(&gl->gl_lockref.lock);
1955 * gfs2_glock_thaw - Thaw any frozen glocks
1956 * @sdp: The super block
1960 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1962 glock_hash_walk(thaw_glock, sdp);
1965 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
1967 spin_lock(&gl->gl_lockref.lock);
1968 gfs2_dump_glock(seq, gl, fsid);
1969 spin_unlock(&gl->gl_lockref.lock);
1972 static void dump_glock_func(struct gfs2_glock *gl)
1974 dump_glock(NULL, gl, true);
1978 * gfs2_gl_hash_clear - Empty out the glock hash table
1979 * @sdp: the filesystem
1980 * @wait: wait until it's all gone
1982 * Called when unmounting the filesystem.
1985 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1987 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1988 flush_workqueue(glock_workqueue);
1989 glock_hash_walk(clear_glock, sdp);
1990 flush_workqueue(glock_workqueue);
1991 wait_event_timeout(sdp->sd_glock_wait,
1992 atomic_read(&sdp->sd_glock_disposal) == 0,
1994 glock_hash_walk(dump_glock_func, sdp);
1997 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1999 struct gfs2_glock *gl = ip->i_gl;
2002 ret = gfs2_truncatei_resume(ip);
2003 gfs2_glock_assert_withdraw(gl, ret == 0);
2005 spin_lock(&gl->gl_lockref.lock);
2006 clear_bit(GLF_LOCK, &gl->gl_flags);
2008 spin_unlock(&gl->gl_lockref.lock);
2011 static const char *state2str(unsigned state)
2014 case LM_ST_UNLOCKED:
2018 case LM_ST_DEFERRED:
2020 case LM_ST_EXCLUSIVE:
2026 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
2029 if (flags & LM_FLAG_TRY)
2031 if (flags & LM_FLAG_TRY_1CB)
2033 if (flags & LM_FLAG_NOEXP)
2035 if (flags & LM_FLAG_ANY)
2037 if (flags & LM_FLAG_PRIORITY)
2039 if (flags & GL_ASYNC)
2041 if (flags & GL_EXACT)
2043 if (flags & GL_NOCACHE)
2045 if (test_bit(HIF_HOLDER, &iflags))
2047 if (test_bit(HIF_WAIT, &iflags))
2049 if (test_bit(HIF_FIRST, &iflags))
2056 * dump_holder - print information about a glock holder
2057 * @seq: the seq_file struct
2058 * @gh: the glock holder
2059 * @fs_id_buf: pointer to file system id (if requested)
2063 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
2064 const char *fs_id_buf)
2066 struct task_struct *gh_owner = NULL;
2070 if (gh->gh_owner_pid)
2071 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
2072 gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2073 fs_id_buf, state2str(gh->gh_state),
2074 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
2076 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
2077 gh_owner ? gh_owner->comm : "(ended)",
2082 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
2084 const unsigned long *gflags = &gl->gl_flags;
2087 if (test_bit(GLF_LOCK, gflags))
2089 if (test_bit(GLF_DEMOTE, gflags))
2091 if (test_bit(GLF_PENDING_DEMOTE, gflags))
2093 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
2095 if (test_bit(GLF_DIRTY, gflags))
2097 if (test_bit(GLF_LFLUSH, gflags))
2099 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
2101 if (test_bit(GLF_REPLY_PENDING, gflags))
2103 if (test_bit(GLF_INITIAL, gflags))
2105 if (test_bit(GLF_FROZEN, gflags))
2107 if (!list_empty(&gl->gl_holders))
2109 if (test_bit(GLF_LRU, gflags))
2113 if (test_bit(GLF_BLOCKING, gflags))
2115 if (test_bit(GLF_INODE_CREATING, gflags))
2117 if (test_bit(GLF_PENDING_DELETE, gflags))
2119 if (test_bit(GLF_FREEING, gflags))
2126 * gfs2_dump_glock - print information about a glock
2127 * @seq: The seq_file struct
2129 * @fsid: If true, also dump the file system id
2131 * The file format is as follows:
2132 * One line per object, capital letters are used to indicate objects
2133 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2134 * other objects are indented by a single space and follow the glock to
2135 * which they are related. Fields are indicated by lower case letters
2136 * followed by a colon and the field value, except for strings which are in
2137 * [] so that its possible to see if they are composed of spaces for
2138 * example. The field's are n = number (id of the object), f = flags,
2139 * t = type, s = state, r = refcount, e = error, p = pid.
2143 void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2145 const struct gfs2_glock_operations *glops = gl->gl_ops;
2146 unsigned long long dtime;
2147 const struct gfs2_holder *gh;
2148 char gflags_buf[32];
2149 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2150 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2151 unsigned long nrpages = 0;
2153 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
2154 struct address_space *mapping = gfs2_glock2aspace(gl);
2156 nrpages = mapping->nrpages;
2158 memset(fs_id_buf, 0, sizeof(fs_id_buf));
2159 if (fsid && sdp) /* safety precaution */
2160 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2161 dtime = jiffies - gl->gl_demote_time;
2162 dtime *= 1000000/HZ; /* demote time in uSec */
2163 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
2165 gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2166 "v:%d r:%d m:%ld p:%lu\n",
2167 fs_id_buf, state2str(gl->gl_state),
2168 gl->gl_name.ln_type,
2169 (unsigned long long)gl->gl_name.ln_number,
2170 gflags2str(gflags_buf, gl),
2171 state2str(gl->gl_target),
2172 state2str(gl->gl_demote_state), dtime,
2173 atomic_read(&gl->gl_ail_count),
2174 atomic_read(&gl->gl_revokes),
2175 (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages);
2177 list_for_each_entry(gh, &gl->gl_holders, gh_list)
2178 dump_holder(seq, gh, fs_id_buf);
2180 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
2181 glops->go_dump(seq, gl, fs_id_buf);
2184 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
2186 struct gfs2_glock *gl = iter_ptr;
2188 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2189 gl->gl_name.ln_type,
2190 (unsigned long long)gl->gl_name.ln_number,
2191 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
2192 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
2193 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
2194 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
2195 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
2196 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
2197 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
2198 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
2202 static const char *gfs2_gltype[] = {
2216 static const char *gfs2_stype[] = {
2217 [GFS2_LKS_SRTT] = "srtt",
2218 [GFS2_LKS_SRTTVAR] = "srttvar",
2219 [GFS2_LKS_SRTTB] = "srttb",
2220 [GFS2_LKS_SRTTVARB] = "srttvarb",
2221 [GFS2_LKS_SIRT] = "sirt",
2222 [GFS2_LKS_SIRTVAR] = "sirtvar",
2223 [GFS2_LKS_DCOUNT] = "dlm",
2224 [GFS2_LKS_QCOUNT] = "queue",
2227 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2229 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
2231 struct gfs2_sbd *sdp = seq->private;
2232 loff_t pos = *(loff_t *)iter_ptr;
2233 unsigned index = pos >> 3;
2234 unsigned subindex = pos & 0x07;
2237 if (index == 0 && subindex != 0)
2240 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
2241 (index == 0) ? "cpu": gfs2_stype[subindex]);
2243 for_each_possible_cpu(i) {
2244 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
2247 seq_printf(seq, " %15u", i);
2249 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
2250 lkstats[index - 1].stats[subindex]);
2252 seq_putc(seq, '\n');
2256 int __init gfs2_glock_init(void)
2260 ret = rhashtable_init(&gl_hash_table, &ht_parms);
2264 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
2265 WQ_HIGHPRI | WQ_FREEZABLE, 0);
2266 if (!glock_workqueue) {
2267 rhashtable_destroy(&gl_hash_table);
2270 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
2271 WQ_MEM_RECLAIM | WQ_FREEZABLE,
2273 if (!gfs2_delete_workqueue) {
2274 destroy_workqueue(glock_workqueue);
2275 rhashtable_destroy(&gl_hash_table);
2279 ret = register_shrinker(&glock_shrinker);
2281 destroy_workqueue(gfs2_delete_workqueue);
2282 destroy_workqueue(glock_workqueue);
2283 rhashtable_destroy(&gl_hash_table);
2287 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2288 init_waitqueue_head(glock_wait_table + i);
2293 void gfs2_glock_exit(void)
2295 unregister_shrinker(&glock_shrinker);
2296 rhashtable_destroy(&gl_hash_table);
2297 destroy_workqueue(glock_workqueue);
2298 destroy_workqueue(gfs2_delete_workqueue);
2301 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2303 struct gfs2_glock *gl = gi->gl;
2308 if (!lockref_put_not_zero(&gl->gl_lockref))
2309 gfs2_glock_queue_put(gl);
2312 gl = rhashtable_walk_next(&gi->hti);
2313 if (IS_ERR_OR_NULL(gl)) {
2314 if (gl == ERR_PTR(-EAGAIN)) {
2321 if (gl->gl_name.ln_sbd != gi->sdp)
2324 if (!lockref_get_not_dead(&gl->gl_lockref))
2328 if (__lockref_is_dead(&gl->gl_lockref))
2336 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2339 struct gfs2_glock_iter *gi = seq->private;
2343 * We can either stay where we are, skip to the next hash table
2344 * entry, or start from the beginning.
2346 if (*pos < gi->last_pos) {
2347 rhashtable_walk_exit(&gi->hti);
2348 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2351 n = *pos - gi->last_pos;
2354 rhashtable_walk_start(&gi->hti);
2356 gfs2_glock_iter_next(gi, n);
2357 gi->last_pos = *pos;
2361 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2364 struct gfs2_glock_iter *gi = seq->private;
2367 gi->last_pos = *pos;
2368 gfs2_glock_iter_next(gi, 1);
2372 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2375 struct gfs2_glock_iter *gi = seq->private;
2377 rhashtable_walk_stop(&gi->hti);
2380 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2382 dump_glock(seq, iter_ptr, false);
2386 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2389 if (*pos >= GFS2_NR_SBSTATS)
2394 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2398 if (*pos >= GFS2_NR_SBSTATS)
2403 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2408 static const struct seq_operations gfs2_glock_seq_ops = {
2409 .start = gfs2_glock_seq_start,
2410 .next = gfs2_glock_seq_next,
2411 .stop = gfs2_glock_seq_stop,
2412 .show = gfs2_glock_seq_show,
2415 static const struct seq_operations gfs2_glstats_seq_ops = {
2416 .start = gfs2_glock_seq_start,
2417 .next = gfs2_glock_seq_next,
2418 .stop = gfs2_glock_seq_stop,
2419 .show = gfs2_glstats_seq_show,
2422 static const struct seq_operations gfs2_sbstats_sops = {
2423 .start = gfs2_sbstats_seq_start,
2424 .next = gfs2_sbstats_seq_next,
2425 .stop = gfs2_sbstats_seq_stop,
2426 .show = gfs2_sbstats_seq_show,
2429 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2431 static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2432 const struct seq_operations *ops)
2434 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2436 struct seq_file *seq = file->private_data;
2437 struct gfs2_glock_iter *gi = seq->private;
2439 gi->sdp = inode->i_private;
2440 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2442 seq->size = GFS2_SEQ_GOODSIZE;
2444 * Initially, we are "before" the first hash table entry; the
2445 * first call to rhashtable_walk_next gets us the first entry.
2449 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2454 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2456 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2459 static int gfs2_glocks_release(struct inode *inode, struct file *file)
2461 struct seq_file *seq = file->private_data;
2462 struct gfs2_glock_iter *gi = seq->private;
2465 gfs2_glock_put(gi->gl);
2466 rhashtable_walk_exit(&gi->hti);
2467 return seq_release_private(inode, file);
2470 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2472 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2475 static const struct file_operations gfs2_glocks_fops = {
2476 .owner = THIS_MODULE,
2477 .open = gfs2_glocks_open,
2479 .llseek = seq_lseek,
2480 .release = gfs2_glocks_release,
2483 static const struct file_operations gfs2_glstats_fops = {
2484 .owner = THIS_MODULE,
2485 .open = gfs2_glstats_open,
2487 .llseek = seq_lseek,
2488 .release = gfs2_glocks_release,
2491 DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats);
2493 void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2495 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2497 debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2500 debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2501 &gfs2_glstats_fops);
2503 debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2504 &gfs2_sbstats_fops);
2507 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2509 debugfs_remove_recursive(sdp->debugfs_dir);
2510 sdp->debugfs_dir = NULL;
2513 void gfs2_register_debugfs(void)
2515 gfs2_root = debugfs_create_dir("gfs2", NULL);
2518 void gfs2_unregister_debugfs(void)
2520 debugfs_remove(gfs2_root);