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_empty(mapping));
278 trace_gfs2_glock_put(gl);
279 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
283 * Cause the glock to be put in work queue context.
285 void gfs2_glock_queue_put(struct gfs2_glock *gl)
287 gfs2_glock_queue_work(gl, 0);
291 * gfs2_glock_put() - Decrement reference count on glock
292 * @gl: The glock to put
296 void gfs2_glock_put(struct gfs2_glock *gl)
298 if (lockref_put_or_lock(&gl->gl_lockref))
301 __gfs2_glock_put(gl);
305 * may_grant - check if its ok to grant a new lock
307 * @gh: The lock request which we wish to grant
309 * Returns: true if its ok to grant the lock
312 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
314 const struct gfs2_holder *gh_head = list_first_entry(&gl->gl_holders, const struct gfs2_holder, gh_list);
318 * Here we make a special exception to grant holders who agree
319 * to share the EX lock with other holders who also have the
320 * bit set. If the original holder has the LM_FLAG_NODE_SCOPE bit
321 * is set, we grant more holders with the bit set.
323 if (gh_head->gh_state == LM_ST_EXCLUSIVE &&
324 (gh_head->gh_flags & LM_FLAG_NODE_SCOPE) &&
325 gh->gh_state == LM_ST_EXCLUSIVE &&
326 (gh->gh_flags & LM_FLAG_NODE_SCOPE))
328 if ((gh->gh_state == LM_ST_EXCLUSIVE ||
329 gh_head->gh_state == LM_ST_EXCLUSIVE))
332 if (gl->gl_state == gh->gh_state)
334 if (gh->gh_flags & GL_EXACT)
336 if (gl->gl_state == LM_ST_EXCLUSIVE) {
337 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
339 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
342 if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
347 static void gfs2_holder_wake(struct gfs2_holder *gh)
349 clear_bit(HIF_WAIT, &gh->gh_iflags);
350 smp_mb__after_atomic();
351 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
352 if (gh->gh_flags & GL_ASYNC) {
353 struct gfs2_sbd *sdp = gh->gh_gl->gl_name.ln_sbd;
355 wake_up(&sdp->sd_async_glock_wait);
360 * do_error - Something unexpected has happened during a lock request
362 * @ret: The status from the DLM
365 static void do_error(struct gfs2_glock *gl, const int ret)
367 struct gfs2_holder *gh, *tmp;
369 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
370 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
372 if (ret & LM_OUT_ERROR)
374 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
375 gh->gh_error = GLR_TRYFAILED;
378 list_del_init(&gh->gh_list);
379 trace_gfs2_glock_queue(gh, 0);
380 gfs2_holder_wake(gh);
385 * do_promote - promote as many requests as possible on the current queue
388 * Returns: 1 if there is a blocked holder at the head of the list, or 2
389 * if a type specific operation is underway.
392 static int do_promote(struct gfs2_glock *gl)
393 __releases(&gl->gl_lockref.lock)
394 __acquires(&gl->gl_lockref.lock)
396 const struct gfs2_glock_operations *glops = gl->gl_ops;
397 struct gfs2_holder *gh, *tmp;
401 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
402 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
404 if (may_grant(gl, gh)) {
405 if (gh->gh_list.prev == &gl->gl_holders &&
407 spin_unlock(&gl->gl_lockref.lock);
408 /* FIXME: eliminate this eventually */
409 ret = glops->go_lock(gh);
410 spin_lock(&gl->gl_lockref.lock);
415 list_del_init(&gh->gh_list);
416 trace_gfs2_glock_queue(gh, 0);
417 gfs2_holder_wake(gh);
420 set_bit(HIF_HOLDER, &gh->gh_iflags);
421 trace_gfs2_promote(gh, 1);
422 gfs2_holder_wake(gh);
425 set_bit(HIF_HOLDER, &gh->gh_iflags);
426 trace_gfs2_promote(gh, 0);
427 gfs2_holder_wake(gh);
430 if (gh->gh_list.prev == &gl->gl_holders)
439 * find_first_waiter - find the first gh that's waiting for the glock
443 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
445 struct gfs2_holder *gh;
447 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
448 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
455 * state_change - record that the glock is now in a different state
457 * @new_state: the new state
460 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
464 held1 = (gl->gl_state != LM_ST_UNLOCKED);
465 held2 = (new_state != LM_ST_UNLOCKED);
467 if (held1 != held2) {
468 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
470 gl->gl_lockref.count++;
472 gl->gl_lockref.count--;
474 if (new_state != gl->gl_target)
475 /* shorten our minimum hold time */
476 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
478 gl->gl_state = new_state;
479 gl->gl_tchange = jiffies;
482 static void gfs2_set_demote(struct gfs2_glock *gl)
484 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
486 set_bit(GLF_DEMOTE, &gl->gl_flags);
488 wake_up(&sdp->sd_async_glock_wait);
491 static void gfs2_demote_wake(struct gfs2_glock *gl)
493 gl->gl_demote_state = LM_ST_EXCLUSIVE;
494 clear_bit(GLF_DEMOTE, &gl->gl_flags);
495 smp_mb__after_atomic();
496 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
500 * finish_xmote - The DLM has replied to one of our lock requests
502 * @ret: The status from the DLM
506 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
508 const struct gfs2_glock_operations *glops = gl->gl_ops;
509 struct gfs2_holder *gh;
510 unsigned state = ret & LM_OUT_ST_MASK;
513 spin_lock(&gl->gl_lockref.lock);
514 trace_gfs2_glock_state_change(gl, state);
515 state_change(gl, state);
516 gh = find_first_waiter(gl);
518 /* Demote to UN request arrived during demote to SH or DF */
519 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
520 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
521 gl->gl_target = LM_ST_UNLOCKED;
523 /* Check for state != intended state */
524 if (unlikely(state != gl->gl_target)) {
525 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
526 /* move to back of queue and try next entry */
527 if (ret & LM_OUT_CANCELED) {
528 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
529 list_move_tail(&gh->gh_list, &gl->gl_holders);
530 gh = find_first_waiter(gl);
531 gl->gl_target = gh->gh_state;
534 /* Some error or failed "try lock" - report it */
535 if ((ret & LM_OUT_ERROR) ||
536 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
537 gl->gl_target = gl->gl_state;
543 /* Unlocked due to conversion deadlock, try again */
546 do_xmote(gl, gh, gl->gl_target);
548 /* Conversion fails, unlock and try again */
551 do_xmote(gl, gh, LM_ST_UNLOCKED);
553 default: /* Everything else */
554 fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n",
555 gl->gl_target, state);
558 spin_unlock(&gl->gl_lockref.lock);
562 /* Fast path - we got what we asked for */
563 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
564 gfs2_demote_wake(gl);
565 if (state != LM_ST_UNLOCKED) {
566 if (glops->go_xmote_bh) {
567 spin_unlock(&gl->gl_lockref.lock);
568 rv = glops->go_xmote_bh(gl);
569 spin_lock(&gl->gl_lockref.lock);
580 clear_bit(GLF_LOCK, &gl->gl_flags);
582 spin_unlock(&gl->gl_lockref.lock);
585 static bool is_system_glock(struct gfs2_glock *gl)
587 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
588 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
590 if (gl == m_ip->i_gl)
596 * do_xmote - Calls the DLM to change the state of a lock
597 * @gl: The lock state
598 * @gh: The holder (only for promotes)
599 * @target: The target lock state
603 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
604 __releases(&gl->gl_lockref.lock)
605 __acquires(&gl->gl_lockref.lock)
607 const struct gfs2_glock_operations *glops = gl->gl_ops;
608 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
609 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
612 if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) &&
613 gh && !(gh->gh_flags & LM_FLAG_NOEXP))
615 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
617 GLOCK_BUG_ON(gl, gl->gl_state == target);
618 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
619 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
622 * If another process is already doing the invalidate, let that
623 * finish first. The glock state machine will get back to this
624 * holder again later.
626 if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
629 do_error(gl, 0); /* Fail queued try locks */
632 set_bit(GLF_BLOCKING, &gl->gl_flags);
633 if ((gl->gl_req == LM_ST_UNLOCKED) ||
634 (gl->gl_state == LM_ST_EXCLUSIVE) ||
635 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
636 clear_bit(GLF_BLOCKING, &gl->gl_flags);
637 spin_unlock(&gl->gl_lockref.lock);
638 if (glops->go_sync) {
639 ret = glops->go_sync(gl);
640 /* If we had a problem syncing (due to io errors or whatever,
641 * we should not invalidate the metadata or tell dlm to
642 * release the glock to other nodes.
645 if (cmpxchg(&sdp->sd_log_error, 0, ret)) {
646 fs_err(sdp, "Error %d syncing glock \n", ret);
647 gfs2_dump_glock(NULL, gl, true);
652 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) {
654 * The call to go_sync should have cleared out the ail list.
655 * If there are still items, we have a problem. We ought to
656 * withdraw, but we can't because the withdraw code also uses
657 * glocks. Warn about the error, dump the glock, then fall
658 * through and wait for logd to do the withdraw for us.
660 if ((atomic_read(&gl->gl_ail_count) != 0) &&
661 (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) {
662 gfs2_glock_assert_warn(gl,
663 !atomic_read(&gl->gl_ail_count));
664 gfs2_dump_glock(NULL, gl, true);
666 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
667 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
673 * Check for an error encountered since we called go_sync and go_inval.
674 * If so, we can't withdraw from the glock code because the withdraw
675 * code itself uses glocks (see function signal_our_withdraw) to
676 * change the mount to read-only. Most importantly, we must not call
677 * dlm to unlock the glock until the journal is in a known good state
678 * (after journal replay) otherwise other nodes may use the object
679 * (rgrp or dinode) and then later, journal replay will corrupt the
680 * file system. The best we can do here is wait for the logd daemon
681 * to see sd_log_error and withdraw, and in the meantime, requeue the
684 * We make a special exception for some system glocks, such as the
685 * system statfs inode glock, which needs to be granted before the
686 * gfs2_quotad daemon can exit, and that exit needs to finish before
687 * we can unmount the withdrawn file system.
689 * However, if we're just unlocking the lock (say, for unmount, when
690 * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
691 * then it's okay to tell dlm to unlock it.
693 if (unlikely(sdp->sd_log_error && !gfs2_withdrawn(sdp)))
694 gfs2_withdraw_delayed(sdp);
695 if (glock_blocked_by_withdraw(gl) &&
696 (target != LM_ST_UNLOCKED ||
697 test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags))) {
698 if (!is_system_glock(gl)) {
699 gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD);
702 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
706 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
708 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
709 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
710 target == LM_ST_UNLOCKED &&
711 test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
712 finish_xmote(gl, target);
713 gfs2_glock_queue_work(gl, 0);
715 fs_err(sdp, "lm_lock ret %d\n", ret);
716 GLOCK_BUG_ON(gl, !gfs2_withdrawn(sdp));
718 } else { /* lock_nolock */
719 finish_xmote(gl, target);
720 gfs2_glock_queue_work(gl, 0);
723 spin_lock(&gl->gl_lockref.lock);
727 * find_first_holder - find the first "holder" gh
731 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
733 struct gfs2_holder *gh;
735 if (!list_empty(&gl->gl_holders)) {
736 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
737 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
744 * run_queue - do all outstanding tasks related to a glock
745 * @gl: The glock in question
746 * @nonblock: True if we must not block in run_queue
750 static void run_queue(struct gfs2_glock *gl, const int nonblock)
751 __releases(&gl->gl_lockref.lock)
752 __acquires(&gl->gl_lockref.lock)
754 struct gfs2_holder *gh = NULL;
757 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
760 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
762 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
763 gl->gl_demote_state != gl->gl_state) {
764 if (find_first_holder(gl))
768 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
769 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
770 gl->gl_target = gl->gl_demote_state;
772 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
773 gfs2_demote_wake(gl);
774 ret = do_promote(gl);
779 gh = find_first_waiter(gl);
780 gl->gl_target = gh->gh_state;
781 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
782 do_error(gl, 0); /* Fail queued try locks */
784 do_xmote(gl, gh, gl->gl_target);
789 clear_bit(GLF_LOCK, &gl->gl_flags);
790 smp_mb__after_atomic();
791 gl->gl_lockref.count++;
792 __gfs2_glock_queue_work(gl, 0);
796 clear_bit(GLF_LOCK, &gl->gl_flags);
797 smp_mb__after_atomic();
801 void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation)
803 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
805 if (ri->ri_magic == 0)
806 ri->ri_magic = cpu_to_be32(GFS2_MAGIC);
807 if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC))
808 ri->ri_generation_deleted = cpu_to_be64(generation);
811 bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation)
813 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
815 if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC))
817 return generation <= be64_to_cpu(ri->ri_generation_deleted);
820 static void gfs2_glock_poke(struct gfs2_glock *gl)
822 int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP;
823 struct gfs2_holder gh;
826 gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh);
827 error = gfs2_glock_nq(&gh);
830 gfs2_holder_uninit(&gh);
833 static bool gfs2_try_evict(struct gfs2_glock *gl)
835 struct gfs2_inode *ip;
836 bool evicted = false;
839 * If there is contention on the iopen glock and we have an inode, try
840 * to grab and release the inode so that it can be evicted. This will
841 * allow the remote node to go ahead and delete the inode without us
842 * having to do it, which will avoid rgrp glock thrashing.
844 * The remote node is likely still holding the corresponding inode
845 * glock, so it will run before we get to verify that the delete has
848 spin_lock(&gl->gl_lockref.lock);
850 if (ip && !igrab(&ip->i_inode))
852 spin_unlock(&gl->gl_lockref.lock);
854 struct gfs2_glock *inode_gl = NULL;
856 gl->gl_no_formal_ino = ip->i_no_formal_ino;
857 set_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
858 d_prune_aliases(&ip->i_inode);
861 /* If the inode was evicted, gl->gl_object will now be NULL. */
862 spin_lock(&gl->gl_lockref.lock);
866 lockref_get(&inode_gl->gl_lockref);
867 clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
869 spin_unlock(&gl->gl_lockref.lock);
871 gfs2_glock_poke(inode_gl);
872 gfs2_glock_put(inode_gl);
879 static void delete_work_func(struct work_struct *work)
881 struct delayed_work *dwork = to_delayed_work(work);
882 struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete);
883 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
885 u64 no_addr = gl->gl_name.ln_number;
887 spin_lock(&gl->gl_lockref.lock);
888 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
889 spin_unlock(&gl->gl_lockref.lock);
891 if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
893 * If we can evict the inode, give the remote node trying to
894 * delete the inode some time before verifying that the delete
895 * has happened. Otherwise, if we cause contention on the inode glock
896 * immediately, the remote node will think that we still have
897 * the inode in use, and so it will give up waiting.
899 * If we can't evict the inode, signal to the remote node that
900 * the inode is still in use. We'll later try to delete the
901 * inode locally in gfs2_evict_inode.
903 * FIXME: We only need to verify that the remote node has
904 * deleted the inode because nodes before this remote delete
905 * rework won't cooperate. At a later time, when we no longer
906 * care about compatibility with such nodes, we can skip this
909 if (gfs2_try_evict(gl)) {
910 if (gfs2_queue_delete_work(gl, 5 * HZ))
916 inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino,
917 GFS2_BLKST_UNLINKED);
918 if (!IS_ERR_OR_NULL(inode)) {
919 d_prune_aliases(inode);
926 static void glock_work_func(struct work_struct *work)
928 unsigned long delay = 0;
929 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
930 unsigned int drop_refs = 1;
932 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
933 finish_xmote(gl, gl->gl_reply);
936 spin_lock(&gl->gl_lockref.lock);
937 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
938 gl->gl_state != LM_ST_UNLOCKED &&
939 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
940 unsigned long holdtime, now = jiffies;
942 holdtime = gl->gl_tchange + gl->gl_hold_time;
943 if (time_before(now, holdtime))
944 delay = holdtime - now;
947 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
953 /* Keep one glock reference for the work we requeue. */
955 if (gl->gl_name.ln_type != LM_TYPE_INODE)
957 __gfs2_glock_queue_work(gl, delay);
961 * Drop the remaining glock references manually here. (Mind that
962 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
965 gl->gl_lockref.count -= drop_refs;
966 if (!gl->gl_lockref.count) {
967 __gfs2_glock_put(gl);
970 spin_unlock(&gl->gl_lockref.lock);
973 static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
974 struct gfs2_glock *new)
976 struct wait_glock_queue wait;
977 wait_queue_head_t *wq = glock_waitqueue(name);
978 struct gfs2_glock *gl;
981 init_wait(&wait.wait);
982 wait.wait.func = glock_wake_function;
985 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
988 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
989 &new->gl_node, ht_parms);
993 gl = rhashtable_lookup_fast(&gl_hash_table,
996 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
1003 finish_wait(wq, &wait.wait);
1008 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
1009 * @sdp: The GFS2 superblock
1010 * @number: the lock number
1011 * @glops: The glock_operations to use
1012 * @create: If 0, don't create the glock if it doesn't exist
1013 * @glp: the glock is returned here
1015 * This does not lock a glock, just finds/creates structures for one.
1020 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
1021 const struct gfs2_glock_operations *glops, int create,
1022 struct gfs2_glock **glp)
1024 struct super_block *s = sdp->sd_vfs;
1025 struct lm_lockname name = { .ln_number = number,
1026 .ln_type = glops->go_type,
1028 struct gfs2_glock *gl, *tmp;
1029 struct address_space *mapping;
1030 struct kmem_cache *cachep;
1033 gl = find_insert_glock(&name, NULL);
1041 if (glops->go_flags & GLOF_ASPACE)
1042 cachep = gfs2_glock_aspace_cachep;
1044 cachep = gfs2_glock_cachep;
1045 gl = kmem_cache_alloc(cachep, GFP_NOFS);
1049 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
1051 if (glops->go_flags & GLOF_LVB) {
1052 gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1053 if (!gl->gl_lksb.sb_lvbptr) {
1054 kmem_cache_free(cachep, gl);
1059 atomic_inc(&sdp->sd_glock_disposal);
1060 gl->gl_node.next = NULL;
1063 lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass);
1064 gl->gl_lockref.count = 1;
1065 gl->gl_state = LM_ST_UNLOCKED;
1066 gl->gl_target = LM_ST_UNLOCKED;
1067 gl->gl_demote_state = LM_ST_EXCLUSIVE;
1071 /* We use the global stats to estimate the initial per-glock stats */
1072 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
1074 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
1075 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
1076 gl->gl_tchange = jiffies;
1077 gl->gl_object = NULL;
1078 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
1079 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
1080 if (gl->gl_name.ln_type == LM_TYPE_IOPEN)
1081 INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func);
1083 mapping = gfs2_glock2aspace(gl);
1085 mapping->a_ops = &gfs2_meta_aops;
1086 mapping->host = s->s_bdev->bd_inode;
1088 mapping_set_gfp_mask(mapping, GFP_NOFS);
1089 mapping->private_data = NULL;
1090 mapping->writeback_index = 0;
1093 tmp = find_insert_glock(&name, gl);
1105 kfree(gl->gl_lksb.sb_lvbptr);
1106 kmem_cache_free(cachep, gl);
1107 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
1108 wake_up(&sdp->sd_glock_wait);
1115 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
1117 * @state: the state we're requesting
1118 * @flags: the modifier flags
1119 * @gh: the holder structure
1123 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
1124 struct gfs2_holder *gh)
1126 INIT_LIST_HEAD(&gh->gh_list);
1128 gh->gh_ip = _RET_IP_;
1129 gh->gh_owner_pid = get_pid(task_pid(current));
1130 gh->gh_state = state;
1131 gh->gh_flags = flags;
1134 gfs2_glock_hold(gl);
1138 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1139 * @state: the state we're requesting
1140 * @flags: the modifier flags
1141 * @gh: the holder structure
1143 * Don't mess with the glock.
1147 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
1149 gh->gh_state = state;
1150 gh->gh_flags = flags;
1152 gh->gh_ip = _RET_IP_;
1153 put_pid(gh->gh_owner_pid);
1154 gh->gh_owner_pid = get_pid(task_pid(current));
1158 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1159 * @gh: the holder structure
1163 void gfs2_holder_uninit(struct gfs2_holder *gh)
1165 put_pid(gh->gh_owner_pid);
1166 gfs2_glock_put(gh->gh_gl);
1167 gfs2_holder_mark_uninitialized(gh);
1171 static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
1172 unsigned long start_time)
1174 /* Have we waited longer that a second? */
1175 if (time_after(jiffies, start_time + HZ)) {
1176 /* Lengthen the minimum hold time. */
1177 gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
1183 * gfs2_glock_wait - wait on a glock acquisition
1184 * @gh: the glock holder
1186 * Returns: 0 on success
1189 int gfs2_glock_wait(struct gfs2_holder *gh)
1191 unsigned long start_time = jiffies;
1194 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1195 gfs2_glock_update_hold_time(gh->gh_gl, start_time);
1196 return gh->gh_error;
1199 static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
1203 for (i = 0; i < num_gh; i++)
1204 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
1210 * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1211 * @num_gh: the number of holders in the array
1212 * @ghs: the glock holder array
1214 * Returns: 0 on success, meaning all glocks have been granted and are held.
1215 * -ESTALE if the request timed out, meaning all glocks were released,
1216 * and the caller should retry the operation.
1219 int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
1221 struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
1222 int i, ret = 0, timeout = 0;
1223 unsigned long start_time = jiffies;
1228 * Total up the (minimum hold time * 2) of all glocks and use that to
1229 * determine the max amount of time we should wait.
1231 for (i = 0; i < num_gh; i++)
1232 timeout += ghs[i].gh_gl->gl_hold_time << 1;
1235 if (!wait_event_timeout(sdp->sd_async_glock_wait,
1236 !glocks_pending(num_gh, ghs), timeout))
1237 ret = -ESTALE; /* request timed out. */
1240 * If dlm granted all our requests, we need to adjust the glock
1241 * minimum hold time values according to how long we waited.
1243 * If our request timed out, we need to repeatedly release any held
1244 * glocks we acquired thus far to allow dlm to acquire the remaining
1245 * glocks without deadlocking. We cannot currently cancel outstanding
1246 * glock acquisitions.
1248 * The HIF_WAIT bit tells us which requests still need a response from
1251 * If dlm sent us any errors, we return the first error we find.
1253 keep_waiting = false;
1254 for (i = 0; i < num_gh; i++) {
1255 /* Skip holders we have already dequeued below. */
1256 if (!gfs2_holder_queued(&ghs[i]))
1258 /* Skip holders with a pending DLM response. */
1259 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags)) {
1260 keep_waiting = true;
1264 if (test_bit(HIF_HOLDER, &ghs[i].gh_iflags)) {
1266 gfs2_glock_dq(&ghs[i]);
1268 gfs2_glock_update_hold_time(ghs[i].gh_gl,
1272 ret = ghs[i].gh_error;
1279 * At this point, we've either acquired all locks or released them all.
1285 * handle_callback - process a demote request
1287 * @state: the state the caller wants us to change to
1288 * @delay: zero to demote immediately; otherwise pending demote
1289 * @remote: true if this came from a different cluster node
1291 * There are only two requests that we are going to see in actual
1292 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1295 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
1296 unsigned long delay, bool remote)
1299 set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1301 gfs2_set_demote(gl);
1302 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
1303 gl->gl_demote_state = state;
1304 gl->gl_demote_time = jiffies;
1305 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
1306 gl->gl_demote_state != state) {
1307 gl->gl_demote_state = LM_ST_UNLOCKED;
1309 if (gl->gl_ops->go_callback)
1310 gl->gl_ops->go_callback(gl, remote);
1311 trace_gfs2_demote_rq(gl, remote);
1314 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
1316 struct va_format vaf;
1319 va_start(args, fmt);
1322 seq_vprintf(seq, fmt, args);
1327 pr_err("%pV", &vaf);
1334 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1335 * @gh: the holder structure to add
1337 * Eventually we should move the recursive locking trap to a
1338 * debugging option or something like that. This is the fast
1339 * path and needs to have the minimum number of distractions.
1343 static inline void add_to_queue(struct gfs2_holder *gh)
1344 __releases(&gl->gl_lockref.lock)
1345 __acquires(&gl->gl_lockref.lock)
1347 struct gfs2_glock *gl = gh->gh_gl;
1348 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1349 struct list_head *insert_pt = NULL;
1350 struct gfs2_holder *gh2;
1353 GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1354 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1355 GLOCK_BUG_ON(gl, true);
1357 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1358 if (test_bit(GLF_LOCK, &gl->gl_flags))
1359 try_futile = !may_grant(gl, gh);
1360 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1364 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1365 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
1366 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
1367 goto trap_recursive;
1369 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1371 gh->gh_error = GLR_TRYFAILED;
1372 gfs2_holder_wake(gh);
1375 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1377 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1378 insert_pt = &gh2->gh_list;
1380 trace_gfs2_glock_queue(gh, 1);
1381 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1382 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1383 if (likely(insert_pt == NULL)) {
1384 list_add_tail(&gh->gh_list, &gl->gl_holders);
1385 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1389 list_add_tail(&gh->gh_list, insert_pt);
1391 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
1392 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1393 spin_unlock(&gl->gl_lockref.lock);
1394 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1395 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1396 spin_lock(&gl->gl_lockref.lock);
1401 fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1402 fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1403 fs_err(sdp, "lock type: %d req lock state : %d\n",
1404 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1405 fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1406 fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1407 fs_err(sdp, "lock type: %d req lock state : %d\n",
1408 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1409 gfs2_dump_glock(NULL, gl, true);
1414 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1415 * @gh: the holder structure
1417 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1419 * Returns: 0, GLR_TRYFAILED, or errno on failure
1422 int gfs2_glock_nq(struct gfs2_holder *gh)
1424 struct gfs2_glock *gl = gh->gh_gl;
1427 if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP))
1430 if (test_bit(GLF_LRU, &gl->gl_flags))
1431 gfs2_glock_remove_from_lru(gl);
1433 spin_lock(&gl->gl_lockref.lock);
1435 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1436 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1437 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1438 gl->gl_lockref.count++;
1439 __gfs2_glock_queue_work(gl, 0);
1442 spin_unlock(&gl->gl_lockref.lock);
1444 if (!(gh->gh_flags & GL_ASYNC))
1445 error = gfs2_glock_wait(gh);
1451 * gfs2_glock_poll - poll to see if an async request has been completed
1454 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1457 int gfs2_glock_poll(struct gfs2_holder *gh)
1459 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1463 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1464 * @gh: the glock holder
1468 void gfs2_glock_dq(struct gfs2_holder *gh)
1470 struct gfs2_glock *gl = gh->gh_gl;
1471 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1475 spin_lock(&gl->gl_lockref.lock);
1477 * If we're in the process of file system withdraw, we cannot just
1478 * dequeue any glocks until our journal is recovered, lest we
1479 * introduce file system corruption. We need two exceptions to this
1480 * rule: We need to allow unlocking of nondisk glocks and the glock
1481 * for our own journal that needs recovery.
1483 if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) &&
1484 glock_blocked_by_withdraw(gl) &&
1485 gh->gh_gl != sdp->sd_jinode_gl) {
1486 sdp->sd_glock_dqs_held++;
1487 spin_unlock(&gl->gl_lockref.lock);
1489 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
1490 TASK_UNINTERRUPTIBLE);
1491 spin_lock(&gl->gl_lockref.lock);
1493 if (gh->gh_flags & GL_NOCACHE)
1494 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1496 list_del_init(&gh->gh_list);
1497 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1498 if (find_first_holder(gl) == NULL) {
1499 if (list_empty(&gl->gl_holders) &&
1500 !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1501 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1504 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1505 gfs2_glock_add_to_lru(gl);
1507 trace_gfs2_glock_queue(gh, 0);
1508 if (unlikely(!fast_path)) {
1509 gl->gl_lockref.count++;
1510 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1511 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1512 gl->gl_name.ln_type == LM_TYPE_INODE)
1513 delay = gl->gl_hold_time;
1514 __gfs2_glock_queue_work(gl, delay);
1516 spin_unlock(&gl->gl_lockref.lock);
1519 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1521 struct gfs2_glock *gl = gh->gh_gl;
1524 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1528 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1529 * @gh: the holder structure
1533 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1536 gfs2_holder_uninit(gh);
1540 * gfs2_glock_nq_num - acquire a glock based on lock number
1541 * @sdp: the filesystem
1542 * @number: the lock number
1543 * @glops: the glock operations for the type of glock
1544 * @state: the state to acquire the glock in
1545 * @flags: modifier flags for the acquisition
1546 * @gh: the struct gfs2_holder
1551 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1552 const struct gfs2_glock_operations *glops,
1553 unsigned int state, u16 flags, struct gfs2_holder *gh)
1555 struct gfs2_glock *gl;
1558 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1560 error = gfs2_glock_nq_init(gl, state, flags, gh);
1568 * glock_compare - Compare two struct gfs2_glock structures for sorting
1569 * @arg_a: the first structure
1570 * @arg_b: the second structure
1574 static int glock_compare(const void *arg_a, const void *arg_b)
1576 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1577 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1578 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1579 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1581 if (a->ln_number > b->ln_number)
1583 if (a->ln_number < b->ln_number)
1585 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1590 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1591 * @num_gh: the number of structures
1592 * @ghs: an array of struct gfs2_holder structures
1593 * @p: placeholder for the holder structure to pass back
1595 * Returns: 0 on success (all glocks acquired),
1596 * errno on failure (no glocks acquired)
1599 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1600 struct gfs2_holder **p)
1605 for (x = 0; x < num_gh; x++)
1608 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1610 for (x = 0; x < num_gh; x++) {
1611 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1613 error = gfs2_glock_nq(p[x]);
1616 gfs2_glock_dq(p[x]);
1625 * gfs2_glock_nq_m - acquire multiple glocks
1626 * @num_gh: the number of structures
1627 * @ghs: an array of struct gfs2_holder structures
1630 * Returns: 0 on success (all glocks acquired),
1631 * errno on failure (no glocks acquired)
1634 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1636 struct gfs2_holder *tmp[4];
1637 struct gfs2_holder **pph = tmp;
1644 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1645 return gfs2_glock_nq(ghs);
1649 pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *),
1655 error = nq_m_sync(num_gh, ghs, pph);
1664 * gfs2_glock_dq_m - release multiple glocks
1665 * @num_gh: the number of structures
1666 * @ghs: an array of struct gfs2_holder structures
1670 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1673 gfs2_glock_dq(&ghs[num_gh]);
1676 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1678 unsigned long delay = 0;
1679 unsigned long holdtime;
1680 unsigned long now = jiffies;
1682 gfs2_glock_hold(gl);
1683 spin_lock(&gl->gl_lockref.lock);
1684 holdtime = gl->gl_tchange + gl->gl_hold_time;
1685 if (!list_empty(&gl->gl_holders) &&
1686 gl->gl_name.ln_type == LM_TYPE_INODE) {
1687 if (time_before(now, holdtime))
1688 delay = holdtime - now;
1689 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1690 delay = gl->gl_hold_time;
1692 handle_callback(gl, state, delay, true);
1693 __gfs2_glock_queue_work(gl, delay);
1694 spin_unlock(&gl->gl_lockref.lock);
1698 * gfs2_should_freeze - Figure out if glock should be frozen
1699 * @gl: The glock in question
1701 * Glocks are not frozen if (a) the result of the dlm operation is
1702 * an error, (b) the locking operation was an unlock operation or
1703 * (c) if there is a "noexp" flagged request anywhere in the queue
1705 * Returns: 1 if freezing should occur, 0 otherwise
1708 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1710 const struct gfs2_holder *gh;
1712 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1714 if (gl->gl_target == LM_ST_UNLOCKED)
1717 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1718 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1720 if (LM_FLAG_NOEXP & gh->gh_flags)
1728 * gfs2_glock_complete - Callback used by locking
1729 * @gl: Pointer to the glock
1730 * @ret: The return value from the dlm
1732 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1733 * to use a bitfield shared with other glock state fields.
1736 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1738 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1740 spin_lock(&gl->gl_lockref.lock);
1743 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1744 if (gfs2_should_freeze(gl)) {
1745 set_bit(GLF_FROZEN, &gl->gl_flags);
1746 spin_unlock(&gl->gl_lockref.lock);
1751 gl->gl_lockref.count++;
1752 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1753 __gfs2_glock_queue_work(gl, 0);
1754 spin_unlock(&gl->gl_lockref.lock);
1757 static int glock_cmp(void *priv, const struct list_head *a,
1758 const struct list_head *b)
1760 struct gfs2_glock *gla, *glb;
1762 gla = list_entry(a, struct gfs2_glock, gl_lru);
1763 glb = list_entry(b, struct gfs2_glock, gl_lru);
1765 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1767 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1774 * gfs2_dispose_glock_lru - Demote a list of glocks
1775 * @list: The list to dispose of
1777 * Disposing of glocks may involve disk accesses, so that here we sort
1778 * the glocks by number (i.e. disk location of the inodes) so that if
1779 * there are any such accesses, they'll be sent in order (mostly).
1781 * Must be called under the lru_lock, but may drop and retake this
1782 * lock. While the lru_lock is dropped, entries may vanish from the
1783 * list, but no new entries will appear on the list (since it is
1787 static void gfs2_dispose_glock_lru(struct list_head *list)
1788 __releases(&lru_lock)
1789 __acquires(&lru_lock)
1791 struct gfs2_glock *gl;
1793 list_sort(NULL, list, glock_cmp);
1795 while(!list_empty(list)) {
1796 gl = list_first_entry(list, struct gfs2_glock, gl_lru);
1797 list_del_init(&gl->gl_lru);
1798 clear_bit(GLF_LRU, &gl->gl_flags);
1799 if (!spin_trylock(&gl->gl_lockref.lock)) {
1801 list_add(&gl->gl_lru, &lru_list);
1802 set_bit(GLF_LRU, &gl->gl_flags);
1803 atomic_inc(&lru_count);
1806 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1807 spin_unlock(&gl->gl_lockref.lock);
1808 goto add_back_to_lru;
1810 gl->gl_lockref.count++;
1812 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1813 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1814 __gfs2_glock_queue_work(gl, 0);
1815 spin_unlock(&gl->gl_lockref.lock);
1816 cond_resched_lock(&lru_lock);
1821 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1822 * @nr: The number of entries to scan
1824 * This function selects the entries on the LRU which are able to
1825 * be demoted, and then kicks off the process by calling
1826 * gfs2_dispose_glock_lru() above.
1829 static long gfs2_scan_glock_lru(int nr)
1831 struct gfs2_glock *gl;
1836 spin_lock(&lru_lock);
1837 while ((nr-- >= 0) && !list_empty(&lru_list)) {
1838 gl = list_first_entry(&lru_list, struct gfs2_glock, gl_lru);
1840 /* Test for being demotable */
1841 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1842 list_move(&gl->gl_lru, &dispose);
1843 atomic_dec(&lru_count);
1848 list_move(&gl->gl_lru, &skipped);
1850 list_splice(&skipped, &lru_list);
1851 if (!list_empty(&dispose))
1852 gfs2_dispose_glock_lru(&dispose);
1853 spin_unlock(&lru_lock);
1858 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1859 struct shrink_control *sc)
1861 if (!(sc->gfp_mask & __GFP_FS))
1863 return gfs2_scan_glock_lru(sc->nr_to_scan);
1866 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1867 struct shrink_control *sc)
1869 return vfs_pressure_ratio(atomic_read(&lru_count));
1872 static struct shrinker glock_shrinker = {
1873 .seeks = DEFAULT_SEEKS,
1874 .count_objects = gfs2_glock_shrink_count,
1875 .scan_objects = gfs2_glock_shrink_scan,
1879 * glock_hash_walk - Call a function for glock in a hash bucket
1880 * @examiner: the function
1881 * @sdp: the filesystem
1883 * Note that the function can be called multiple times on the same
1884 * object. So the user must ensure that the function can cope with
1888 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1890 struct gfs2_glock *gl;
1891 struct rhashtable_iter iter;
1893 rhashtable_walk_enter(&gl_hash_table, &iter);
1896 rhashtable_walk_start(&iter);
1898 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl))
1899 if (gl->gl_name.ln_sbd == sdp &&
1900 lockref_get_not_dead(&gl->gl_lockref))
1903 rhashtable_walk_stop(&iter);
1904 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
1906 rhashtable_walk_exit(&iter);
1909 bool gfs2_queue_delete_work(struct gfs2_glock *gl, unsigned long delay)
1913 spin_lock(&gl->gl_lockref.lock);
1914 queued = queue_delayed_work(gfs2_delete_workqueue,
1915 &gl->gl_delete, delay);
1917 set_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1918 spin_unlock(&gl->gl_lockref.lock);
1922 void gfs2_cancel_delete_work(struct gfs2_glock *gl)
1924 if (cancel_delayed_work_sync(&gl->gl_delete)) {
1925 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1930 bool gfs2_delete_work_queued(const struct gfs2_glock *gl)
1932 return test_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1935 static void flush_delete_work(struct gfs2_glock *gl)
1937 if (gl->gl_name.ln_type == LM_TYPE_IOPEN) {
1938 if (cancel_delayed_work(&gl->gl_delete)) {
1939 queue_delayed_work(gfs2_delete_workqueue,
1943 gfs2_glock_queue_work(gl, 0);
1946 void gfs2_flush_delete_work(struct gfs2_sbd *sdp)
1948 glock_hash_walk(flush_delete_work, sdp);
1949 flush_workqueue(gfs2_delete_workqueue);
1953 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1954 * @gl: The glock to thaw
1958 static void thaw_glock(struct gfs2_glock *gl)
1960 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags)) {
1964 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1965 gfs2_glock_queue_work(gl, 0);
1969 * clear_glock - look at a glock and see if we can free it from glock cache
1970 * @gl: the glock to look at
1974 static void clear_glock(struct gfs2_glock *gl)
1976 gfs2_glock_remove_from_lru(gl);
1978 spin_lock(&gl->gl_lockref.lock);
1979 if (gl->gl_state != LM_ST_UNLOCKED)
1980 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1981 __gfs2_glock_queue_work(gl, 0);
1982 spin_unlock(&gl->gl_lockref.lock);
1986 * gfs2_glock_thaw - Thaw any frozen glocks
1987 * @sdp: The super block
1991 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1993 glock_hash_walk(thaw_glock, sdp);
1996 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
1998 spin_lock(&gl->gl_lockref.lock);
1999 gfs2_dump_glock(seq, gl, fsid);
2000 spin_unlock(&gl->gl_lockref.lock);
2003 static void dump_glock_func(struct gfs2_glock *gl)
2005 dump_glock(NULL, gl, true);
2009 * gfs2_gl_hash_clear - Empty out the glock hash table
2010 * @sdp: the filesystem
2012 * Called when unmounting the filesystem.
2015 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
2017 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
2018 flush_workqueue(glock_workqueue);
2019 glock_hash_walk(clear_glock, sdp);
2020 flush_workqueue(glock_workqueue);
2021 wait_event_timeout(sdp->sd_glock_wait,
2022 atomic_read(&sdp->sd_glock_disposal) == 0,
2024 glock_hash_walk(dump_glock_func, sdp);
2027 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
2029 struct gfs2_glock *gl = ip->i_gl;
2032 ret = gfs2_truncatei_resume(ip);
2033 gfs2_glock_assert_withdraw(gl, ret == 0);
2035 spin_lock(&gl->gl_lockref.lock);
2036 clear_bit(GLF_LOCK, &gl->gl_flags);
2038 spin_unlock(&gl->gl_lockref.lock);
2041 static const char *state2str(unsigned state)
2044 case LM_ST_UNLOCKED:
2048 case LM_ST_DEFERRED:
2050 case LM_ST_EXCLUSIVE:
2056 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
2059 if (flags & LM_FLAG_TRY)
2061 if (flags & LM_FLAG_TRY_1CB)
2063 if (flags & LM_FLAG_NOEXP)
2065 if (flags & LM_FLAG_ANY)
2067 if (flags & LM_FLAG_PRIORITY)
2069 if (flags & LM_FLAG_NODE_SCOPE)
2071 if (flags & GL_ASYNC)
2073 if (flags & GL_EXACT)
2075 if (flags & GL_NOCACHE)
2077 if (test_bit(HIF_HOLDER, &iflags))
2079 if (test_bit(HIF_WAIT, &iflags))
2081 if (test_bit(HIF_FIRST, &iflags))
2088 * dump_holder - print information about a glock holder
2089 * @seq: the seq_file struct
2090 * @gh: the glock holder
2091 * @fs_id_buf: pointer to file system id (if requested)
2095 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
2096 const char *fs_id_buf)
2098 struct task_struct *gh_owner = NULL;
2102 if (gh->gh_owner_pid)
2103 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
2104 gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2105 fs_id_buf, state2str(gh->gh_state),
2106 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
2108 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
2109 gh_owner ? gh_owner->comm : "(ended)",
2114 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
2116 const unsigned long *gflags = &gl->gl_flags;
2119 if (test_bit(GLF_LOCK, gflags))
2121 if (test_bit(GLF_DEMOTE, gflags))
2123 if (test_bit(GLF_PENDING_DEMOTE, gflags))
2125 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
2127 if (test_bit(GLF_DIRTY, gflags))
2129 if (test_bit(GLF_LFLUSH, gflags))
2131 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
2133 if (test_bit(GLF_REPLY_PENDING, gflags))
2135 if (test_bit(GLF_INITIAL, gflags))
2137 if (test_bit(GLF_FROZEN, gflags))
2139 if (!list_empty(&gl->gl_holders))
2141 if (test_bit(GLF_LRU, gflags))
2145 if (test_bit(GLF_BLOCKING, gflags))
2147 if (test_bit(GLF_PENDING_DELETE, gflags))
2149 if (test_bit(GLF_FREEING, gflags))
2156 * gfs2_dump_glock - print information about a glock
2157 * @seq: The seq_file struct
2159 * @fsid: If true, also dump the file system id
2161 * The file format is as follows:
2162 * One line per object, capital letters are used to indicate objects
2163 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2164 * other objects are indented by a single space and follow the glock to
2165 * which they are related. Fields are indicated by lower case letters
2166 * followed by a colon and the field value, except for strings which are in
2167 * [] so that its possible to see if they are composed of spaces for
2168 * example. The field's are n = number (id of the object), f = flags,
2169 * t = type, s = state, r = refcount, e = error, p = pid.
2173 void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2175 const struct gfs2_glock_operations *glops = gl->gl_ops;
2176 unsigned long long dtime;
2177 const struct gfs2_holder *gh;
2178 char gflags_buf[32];
2179 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2180 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2181 unsigned long nrpages = 0;
2183 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
2184 struct address_space *mapping = gfs2_glock2aspace(gl);
2186 nrpages = mapping->nrpages;
2188 memset(fs_id_buf, 0, sizeof(fs_id_buf));
2189 if (fsid && sdp) /* safety precaution */
2190 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2191 dtime = jiffies - gl->gl_demote_time;
2192 dtime *= 1000000/HZ; /* demote time in uSec */
2193 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
2195 gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2196 "v:%d r:%d m:%ld p:%lu\n",
2197 fs_id_buf, state2str(gl->gl_state),
2198 gl->gl_name.ln_type,
2199 (unsigned long long)gl->gl_name.ln_number,
2200 gflags2str(gflags_buf, gl),
2201 state2str(gl->gl_target),
2202 state2str(gl->gl_demote_state), dtime,
2203 atomic_read(&gl->gl_ail_count),
2204 atomic_read(&gl->gl_revokes),
2205 (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages);
2207 list_for_each_entry(gh, &gl->gl_holders, gh_list)
2208 dump_holder(seq, gh, fs_id_buf);
2210 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
2211 glops->go_dump(seq, gl, fs_id_buf);
2214 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
2216 struct gfs2_glock *gl = iter_ptr;
2218 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2219 gl->gl_name.ln_type,
2220 (unsigned long long)gl->gl_name.ln_number,
2221 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
2222 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
2223 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
2224 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
2225 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
2226 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
2227 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
2228 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
2232 static const char *gfs2_gltype[] = {
2246 static const char *gfs2_stype[] = {
2247 [GFS2_LKS_SRTT] = "srtt",
2248 [GFS2_LKS_SRTTVAR] = "srttvar",
2249 [GFS2_LKS_SRTTB] = "srttb",
2250 [GFS2_LKS_SRTTVARB] = "srttvarb",
2251 [GFS2_LKS_SIRT] = "sirt",
2252 [GFS2_LKS_SIRTVAR] = "sirtvar",
2253 [GFS2_LKS_DCOUNT] = "dlm",
2254 [GFS2_LKS_QCOUNT] = "queue",
2257 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2259 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
2261 struct gfs2_sbd *sdp = seq->private;
2262 loff_t pos = *(loff_t *)iter_ptr;
2263 unsigned index = pos >> 3;
2264 unsigned subindex = pos & 0x07;
2267 if (index == 0 && subindex != 0)
2270 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
2271 (index == 0) ? "cpu": gfs2_stype[subindex]);
2273 for_each_possible_cpu(i) {
2274 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
2277 seq_printf(seq, " %15u", i);
2279 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
2280 lkstats[index - 1].stats[subindex]);
2282 seq_putc(seq, '\n');
2286 int __init gfs2_glock_init(void)
2290 ret = rhashtable_init(&gl_hash_table, &ht_parms);
2294 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
2295 WQ_HIGHPRI | WQ_FREEZABLE, 0);
2296 if (!glock_workqueue) {
2297 rhashtable_destroy(&gl_hash_table);
2300 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
2301 WQ_MEM_RECLAIM | WQ_FREEZABLE,
2303 if (!gfs2_delete_workqueue) {
2304 destroy_workqueue(glock_workqueue);
2305 rhashtable_destroy(&gl_hash_table);
2309 ret = register_shrinker(&glock_shrinker);
2311 destroy_workqueue(gfs2_delete_workqueue);
2312 destroy_workqueue(glock_workqueue);
2313 rhashtable_destroy(&gl_hash_table);
2317 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2318 init_waitqueue_head(glock_wait_table + i);
2323 void gfs2_glock_exit(void)
2325 unregister_shrinker(&glock_shrinker);
2326 rhashtable_destroy(&gl_hash_table);
2327 destroy_workqueue(glock_workqueue);
2328 destroy_workqueue(gfs2_delete_workqueue);
2331 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2333 struct gfs2_glock *gl = gi->gl;
2338 if (!lockref_put_not_zero(&gl->gl_lockref))
2339 gfs2_glock_queue_put(gl);
2342 gl = rhashtable_walk_next(&gi->hti);
2343 if (IS_ERR_OR_NULL(gl)) {
2344 if (gl == ERR_PTR(-EAGAIN)) {
2351 if (gl->gl_name.ln_sbd != gi->sdp)
2354 if (!lockref_get_not_dead(&gl->gl_lockref))
2358 if (__lockref_is_dead(&gl->gl_lockref))
2366 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2369 struct gfs2_glock_iter *gi = seq->private;
2373 * We can either stay where we are, skip to the next hash table
2374 * entry, or start from the beginning.
2376 if (*pos < gi->last_pos) {
2377 rhashtable_walk_exit(&gi->hti);
2378 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2381 n = *pos - gi->last_pos;
2384 rhashtable_walk_start(&gi->hti);
2386 gfs2_glock_iter_next(gi, n);
2387 gi->last_pos = *pos;
2391 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2394 struct gfs2_glock_iter *gi = seq->private;
2397 gi->last_pos = *pos;
2398 gfs2_glock_iter_next(gi, 1);
2402 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2405 struct gfs2_glock_iter *gi = seq->private;
2407 rhashtable_walk_stop(&gi->hti);
2410 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2412 dump_glock(seq, iter_ptr, false);
2416 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2419 if (*pos >= GFS2_NR_SBSTATS)
2424 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2428 if (*pos >= GFS2_NR_SBSTATS)
2433 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2438 static const struct seq_operations gfs2_glock_seq_ops = {
2439 .start = gfs2_glock_seq_start,
2440 .next = gfs2_glock_seq_next,
2441 .stop = gfs2_glock_seq_stop,
2442 .show = gfs2_glock_seq_show,
2445 static const struct seq_operations gfs2_glstats_seq_ops = {
2446 .start = gfs2_glock_seq_start,
2447 .next = gfs2_glock_seq_next,
2448 .stop = gfs2_glock_seq_stop,
2449 .show = gfs2_glstats_seq_show,
2452 static const struct seq_operations gfs2_sbstats_sops = {
2453 .start = gfs2_sbstats_seq_start,
2454 .next = gfs2_sbstats_seq_next,
2455 .stop = gfs2_sbstats_seq_stop,
2456 .show = gfs2_sbstats_seq_show,
2459 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2461 static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2462 const struct seq_operations *ops)
2464 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2466 struct seq_file *seq = file->private_data;
2467 struct gfs2_glock_iter *gi = seq->private;
2469 gi->sdp = inode->i_private;
2470 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2472 seq->size = GFS2_SEQ_GOODSIZE;
2474 * Initially, we are "before" the first hash table entry; the
2475 * first call to rhashtable_walk_next gets us the first entry.
2479 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2484 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2486 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2489 static int gfs2_glocks_release(struct inode *inode, struct file *file)
2491 struct seq_file *seq = file->private_data;
2492 struct gfs2_glock_iter *gi = seq->private;
2495 gfs2_glock_put(gi->gl);
2496 rhashtable_walk_exit(&gi->hti);
2497 return seq_release_private(inode, file);
2500 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2502 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2505 static const struct file_operations gfs2_glocks_fops = {
2506 .owner = THIS_MODULE,
2507 .open = gfs2_glocks_open,
2509 .llseek = seq_lseek,
2510 .release = gfs2_glocks_release,
2513 static const struct file_operations gfs2_glstats_fops = {
2514 .owner = THIS_MODULE,
2515 .open = gfs2_glstats_open,
2517 .llseek = seq_lseek,
2518 .release = gfs2_glocks_release,
2521 DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats);
2523 void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2525 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2527 debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2530 debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2531 &gfs2_glstats_fops);
2533 debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2534 &gfs2_sbstats_fops);
2537 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2539 debugfs_remove_recursive(sdp->debugfs_dir);
2540 sdp->debugfs_dir = NULL;
2543 void gfs2_register_debugfs(void)
2545 gfs2_root = debugfs_create_dir("gfs2", NULL);
2548 void gfs2_unregister_debugfs(void)
2550 debugfs_remove(gfs2_root);