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);
586 * do_xmote - Calls the DLM to change the state of a lock
587 * @gl: The lock state
588 * @gh: The holder (only for promotes)
589 * @target: The target lock state
593 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
594 __releases(&gl->gl_lockref.lock)
595 __acquires(&gl->gl_lockref.lock)
597 const struct gfs2_glock_operations *glops = gl->gl_ops;
598 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
599 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
602 if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) &&
603 gh && !(gh->gh_flags & LM_FLAG_NOEXP))
605 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
607 GLOCK_BUG_ON(gl, gl->gl_state == target);
608 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
609 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
612 * If another process is already doing the invalidate, let that
613 * finish first. The glock state machine will get back to this
614 * holder again later.
616 if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
619 do_error(gl, 0); /* Fail queued try locks */
622 set_bit(GLF_BLOCKING, &gl->gl_flags);
623 if ((gl->gl_req == LM_ST_UNLOCKED) ||
624 (gl->gl_state == LM_ST_EXCLUSIVE) ||
625 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
626 clear_bit(GLF_BLOCKING, &gl->gl_flags);
627 spin_unlock(&gl->gl_lockref.lock);
628 if (glops->go_sync) {
629 ret = glops->go_sync(gl);
630 /* If we had a problem syncing (due to io errors or whatever,
631 * we should not invalidate the metadata or tell dlm to
632 * release the glock to other nodes.
635 if (cmpxchg(&sdp->sd_log_error, 0, ret)) {
636 fs_err(sdp, "Error %d syncing glock \n", ret);
637 gfs2_dump_glock(NULL, gl, true);
642 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) {
644 * The call to go_sync should have cleared out the ail list.
645 * If there are still items, we have a problem. We ought to
646 * withdraw, but we can't because the withdraw code also uses
647 * glocks. Warn about the error, dump the glock, then fall
648 * through and wait for logd to do the withdraw for us.
650 if ((atomic_read(&gl->gl_ail_count) != 0) &&
651 (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) {
652 gfs2_glock_assert_warn(gl,
653 !atomic_read(&gl->gl_ail_count));
654 gfs2_dump_glock(NULL, gl, true);
656 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
657 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
663 * Check for an error encountered since we called go_sync and go_inval.
664 * If so, we can't withdraw from the glock code because the withdraw
665 * code itself uses glocks (see function signal_our_withdraw) to
666 * change the mount to read-only. Most importantly, we must not call
667 * dlm to unlock the glock until the journal is in a known good state
668 * (after journal replay) otherwise other nodes may use the object
669 * (rgrp or dinode) and then later, journal replay will corrupt the
670 * file system. The best we can do here is wait for the logd daemon
671 * to see sd_log_error and withdraw, and in the meantime, requeue the
674 * However, if we're just unlocking the lock (say, for unmount, when
675 * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
676 * then it's okay to tell dlm to unlock it.
678 if (unlikely(sdp->sd_log_error && !gfs2_withdrawn(sdp)))
679 gfs2_withdraw_delayed(sdp);
680 if (glock_blocked_by_withdraw(gl)) {
681 if (target != LM_ST_UNLOCKED ||
682 test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags)) {
683 gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD);
688 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
690 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
691 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
692 target == LM_ST_UNLOCKED &&
693 test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
694 finish_xmote(gl, target);
695 gfs2_glock_queue_work(gl, 0);
697 fs_err(sdp, "lm_lock ret %d\n", ret);
698 GLOCK_BUG_ON(gl, !gfs2_withdrawn(sdp));
700 } else { /* lock_nolock */
701 finish_xmote(gl, target);
702 gfs2_glock_queue_work(gl, 0);
705 spin_lock(&gl->gl_lockref.lock);
709 * find_first_holder - find the first "holder" gh
713 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
715 struct gfs2_holder *gh;
717 if (!list_empty(&gl->gl_holders)) {
718 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
719 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
726 * run_queue - do all outstanding tasks related to a glock
727 * @gl: The glock in question
728 * @nonblock: True if we must not block in run_queue
732 static void run_queue(struct gfs2_glock *gl, const int nonblock)
733 __releases(&gl->gl_lockref.lock)
734 __acquires(&gl->gl_lockref.lock)
736 struct gfs2_holder *gh = NULL;
739 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
742 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
744 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
745 gl->gl_demote_state != gl->gl_state) {
746 if (find_first_holder(gl))
750 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
751 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
752 gl->gl_target = gl->gl_demote_state;
754 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
755 gfs2_demote_wake(gl);
756 ret = do_promote(gl);
761 gh = find_first_waiter(gl);
762 gl->gl_target = gh->gh_state;
763 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
764 do_error(gl, 0); /* Fail queued try locks */
766 do_xmote(gl, gh, gl->gl_target);
771 clear_bit(GLF_LOCK, &gl->gl_flags);
772 smp_mb__after_atomic();
773 gl->gl_lockref.count++;
774 __gfs2_glock_queue_work(gl, 0);
778 clear_bit(GLF_LOCK, &gl->gl_flags);
779 smp_mb__after_atomic();
783 void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation)
785 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
787 if (ri->ri_magic == 0)
788 ri->ri_magic = cpu_to_be32(GFS2_MAGIC);
789 if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC))
790 ri->ri_generation_deleted = cpu_to_be64(generation);
793 bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation)
795 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
797 if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC))
799 return generation <= be64_to_cpu(ri->ri_generation_deleted);
802 static void gfs2_glock_poke(struct gfs2_glock *gl)
804 int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP;
805 struct gfs2_holder gh;
808 gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh);
809 error = gfs2_glock_nq(&gh);
812 gfs2_holder_uninit(&gh);
815 static bool gfs2_try_evict(struct gfs2_glock *gl)
817 struct gfs2_inode *ip;
818 bool evicted = false;
821 * If there is contention on the iopen glock and we have an inode, try
822 * to grab and release the inode so that it can be evicted. This will
823 * allow the remote node to go ahead and delete the inode without us
824 * having to do it, which will avoid rgrp glock thrashing.
826 * The remote node is likely still holding the corresponding inode
827 * glock, so it will run before we get to verify that the delete has
830 spin_lock(&gl->gl_lockref.lock);
832 if (ip && !igrab(&ip->i_inode))
834 spin_unlock(&gl->gl_lockref.lock);
836 struct gfs2_glock *inode_gl = NULL;
838 gl->gl_no_formal_ino = ip->i_no_formal_ino;
839 set_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
840 d_prune_aliases(&ip->i_inode);
843 /* If the inode was evicted, gl->gl_object will now be NULL. */
844 spin_lock(&gl->gl_lockref.lock);
848 lockref_get(&inode_gl->gl_lockref);
849 clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
851 spin_unlock(&gl->gl_lockref.lock);
853 gfs2_glock_poke(inode_gl);
854 gfs2_glock_put(inode_gl);
861 static void delete_work_func(struct work_struct *work)
863 struct delayed_work *dwork = to_delayed_work(work);
864 struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete);
865 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
867 u64 no_addr = gl->gl_name.ln_number;
869 spin_lock(&gl->gl_lockref.lock);
870 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
871 spin_unlock(&gl->gl_lockref.lock);
873 if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
875 * If we can evict the inode, give the remote node trying to
876 * delete the inode some time before verifying that the delete
877 * has happened. Otherwise, if we cause contention on the inode glock
878 * immediately, the remote node will think that we still have
879 * the inode in use, and so it will give up waiting.
881 * If we can't evict the inode, signal to the remote node that
882 * the inode is still in use. We'll later try to delete the
883 * inode locally in gfs2_evict_inode.
885 * FIXME: We only need to verify that the remote node has
886 * deleted the inode because nodes before this remote delete
887 * rework won't cooperate. At a later time, when we no longer
888 * care about compatibility with such nodes, we can skip this
891 if (gfs2_try_evict(gl)) {
892 if (gfs2_queue_delete_work(gl, 5 * HZ))
898 inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino,
899 GFS2_BLKST_UNLINKED);
900 if (!IS_ERR_OR_NULL(inode)) {
901 d_prune_aliases(inode);
908 static void glock_work_func(struct work_struct *work)
910 unsigned long delay = 0;
911 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
912 unsigned int drop_refs = 1;
914 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
915 finish_xmote(gl, gl->gl_reply);
918 spin_lock(&gl->gl_lockref.lock);
919 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
920 gl->gl_state != LM_ST_UNLOCKED &&
921 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
922 unsigned long holdtime, now = jiffies;
924 holdtime = gl->gl_tchange + gl->gl_hold_time;
925 if (time_before(now, holdtime))
926 delay = holdtime - now;
929 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
935 /* Keep one glock reference for the work we requeue. */
937 if (gl->gl_name.ln_type != LM_TYPE_INODE)
939 __gfs2_glock_queue_work(gl, delay);
943 * Drop the remaining glock references manually here. (Mind that
944 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
947 gl->gl_lockref.count -= drop_refs;
948 if (!gl->gl_lockref.count) {
949 __gfs2_glock_put(gl);
952 spin_unlock(&gl->gl_lockref.lock);
955 static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
956 struct gfs2_glock *new)
958 struct wait_glock_queue wait;
959 wait_queue_head_t *wq = glock_waitqueue(name);
960 struct gfs2_glock *gl;
963 init_wait(&wait.wait);
964 wait.wait.func = glock_wake_function;
967 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
970 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
971 &new->gl_node, ht_parms);
975 gl = rhashtable_lookup_fast(&gl_hash_table,
978 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
985 finish_wait(wq, &wait.wait);
990 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
991 * @sdp: The GFS2 superblock
992 * @number: the lock number
993 * @glops: The glock_operations to use
994 * @create: If 0, don't create the glock if it doesn't exist
995 * @glp: the glock is returned here
997 * This does not lock a glock, just finds/creates structures for one.
1002 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
1003 const struct gfs2_glock_operations *glops, int create,
1004 struct gfs2_glock **glp)
1006 struct super_block *s = sdp->sd_vfs;
1007 struct lm_lockname name = { .ln_number = number,
1008 .ln_type = glops->go_type,
1010 struct gfs2_glock *gl, *tmp;
1011 struct address_space *mapping;
1012 struct kmem_cache *cachep;
1015 gl = find_insert_glock(&name, NULL);
1023 if (glops->go_flags & GLOF_ASPACE)
1024 cachep = gfs2_glock_aspace_cachep;
1026 cachep = gfs2_glock_cachep;
1027 gl = kmem_cache_alloc(cachep, GFP_NOFS);
1031 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
1033 if (glops->go_flags & GLOF_LVB) {
1034 gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1035 if (!gl->gl_lksb.sb_lvbptr) {
1036 kmem_cache_free(cachep, gl);
1041 atomic_inc(&sdp->sd_glock_disposal);
1042 gl->gl_node.next = NULL;
1045 lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass);
1046 gl->gl_lockref.count = 1;
1047 gl->gl_state = LM_ST_UNLOCKED;
1048 gl->gl_target = LM_ST_UNLOCKED;
1049 gl->gl_demote_state = LM_ST_EXCLUSIVE;
1053 /* We use the global stats to estimate the initial per-glock stats */
1054 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
1056 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
1057 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
1058 gl->gl_tchange = jiffies;
1059 gl->gl_object = NULL;
1060 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
1061 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
1062 if (gl->gl_name.ln_type == LM_TYPE_IOPEN)
1063 INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func);
1065 mapping = gfs2_glock2aspace(gl);
1067 mapping->a_ops = &gfs2_meta_aops;
1068 mapping->host = s->s_bdev->bd_inode;
1070 mapping_set_gfp_mask(mapping, GFP_NOFS);
1071 mapping->private_data = NULL;
1072 mapping->writeback_index = 0;
1075 tmp = find_insert_glock(&name, gl);
1087 kfree(gl->gl_lksb.sb_lvbptr);
1088 kmem_cache_free(cachep, gl);
1089 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
1090 wake_up(&sdp->sd_glock_wait);
1097 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
1099 * @state: the state we're requesting
1100 * @flags: the modifier flags
1101 * @gh: the holder structure
1105 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
1106 struct gfs2_holder *gh)
1108 INIT_LIST_HEAD(&gh->gh_list);
1110 gh->gh_ip = _RET_IP_;
1111 gh->gh_owner_pid = get_pid(task_pid(current));
1112 gh->gh_state = state;
1113 gh->gh_flags = flags;
1116 gfs2_glock_hold(gl);
1120 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1121 * @state: the state we're requesting
1122 * @flags: the modifier flags
1123 * @gh: the holder structure
1125 * Don't mess with the glock.
1129 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
1131 gh->gh_state = state;
1132 gh->gh_flags = flags;
1134 gh->gh_ip = _RET_IP_;
1135 put_pid(gh->gh_owner_pid);
1136 gh->gh_owner_pid = get_pid(task_pid(current));
1140 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1141 * @gh: the holder structure
1145 void gfs2_holder_uninit(struct gfs2_holder *gh)
1147 put_pid(gh->gh_owner_pid);
1148 gfs2_glock_put(gh->gh_gl);
1149 gfs2_holder_mark_uninitialized(gh);
1153 static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
1154 unsigned long start_time)
1156 /* Have we waited longer that a second? */
1157 if (time_after(jiffies, start_time + HZ)) {
1158 /* Lengthen the minimum hold time. */
1159 gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
1165 * gfs2_glock_wait - wait on a glock acquisition
1166 * @gh: the glock holder
1168 * Returns: 0 on success
1171 int gfs2_glock_wait(struct gfs2_holder *gh)
1173 unsigned long start_time = jiffies;
1176 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1177 gfs2_glock_update_hold_time(gh->gh_gl, start_time);
1178 return gh->gh_error;
1181 static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
1185 for (i = 0; i < num_gh; i++)
1186 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
1192 * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1193 * @num_gh: the number of holders in the array
1194 * @ghs: the glock holder array
1196 * Returns: 0 on success, meaning all glocks have been granted and are held.
1197 * -ESTALE if the request timed out, meaning all glocks were released,
1198 * and the caller should retry the operation.
1201 int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
1203 struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
1204 int i, ret = 0, timeout = 0;
1205 unsigned long start_time = jiffies;
1210 * Total up the (minimum hold time * 2) of all glocks and use that to
1211 * determine the max amount of time we should wait.
1213 for (i = 0; i < num_gh; i++)
1214 timeout += ghs[i].gh_gl->gl_hold_time << 1;
1217 if (!wait_event_timeout(sdp->sd_async_glock_wait,
1218 !glocks_pending(num_gh, ghs), timeout))
1219 ret = -ESTALE; /* request timed out. */
1222 * If dlm granted all our requests, we need to adjust the glock
1223 * minimum hold time values according to how long we waited.
1225 * If our request timed out, we need to repeatedly release any held
1226 * glocks we acquired thus far to allow dlm to acquire the remaining
1227 * glocks without deadlocking. We cannot currently cancel outstanding
1228 * glock acquisitions.
1230 * The HIF_WAIT bit tells us which requests still need a response from
1233 * If dlm sent us any errors, we return the first error we find.
1235 keep_waiting = false;
1236 for (i = 0; i < num_gh; i++) {
1237 /* Skip holders we have already dequeued below. */
1238 if (!gfs2_holder_queued(&ghs[i]))
1240 /* Skip holders with a pending DLM response. */
1241 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags)) {
1242 keep_waiting = true;
1246 if (test_bit(HIF_HOLDER, &ghs[i].gh_iflags)) {
1248 gfs2_glock_dq(&ghs[i]);
1250 gfs2_glock_update_hold_time(ghs[i].gh_gl,
1254 ret = ghs[i].gh_error;
1261 * At this point, we've either acquired all locks or released them all.
1267 * handle_callback - process a demote request
1269 * @state: the state the caller wants us to change to
1270 * @delay: zero to demote immediately; otherwise pending demote
1271 * @remote: true if this came from a different cluster node
1273 * There are only two requests that we are going to see in actual
1274 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1277 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
1278 unsigned long delay, bool remote)
1281 set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1283 gfs2_set_demote(gl);
1284 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
1285 gl->gl_demote_state = state;
1286 gl->gl_demote_time = jiffies;
1287 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
1288 gl->gl_demote_state != state) {
1289 gl->gl_demote_state = LM_ST_UNLOCKED;
1291 if (gl->gl_ops->go_callback)
1292 gl->gl_ops->go_callback(gl, remote);
1293 trace_gfs2_demote_rq(gl, remote);
1296 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
1298 struct va_format vaf;
1301 va_start(args, fmt);
1304 seq_vprintf(seq, fmt, args);
1309 pr_err("%pV", &vaf);
1316 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1317 * @gh: the holder structure to add
1319 * Eventually we should move the recursive locking trap to a
1320 * debugging option or something like that. This is the fast
1321 * path and needs to have the minimum number of distractions.
1325 static inline void add_to_queue(struct gfs2_holder *gh)
1326 __releases(&gl->gl_lockref.lock)
1327 __acquires(&gl->gl_lockref.lock)
1329 struct gfs2_glock *gl = gh->gh_gl;
1330 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1331 struct list_head *insert_pt = NULL;
1332 struct gfs2_holder *gh2;
1335 GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1336 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1337 GLOCK_BUG_ON(gl, true);
1339 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1340 if (test_bit(GLF_LOCK, &gl->gl_flags))
1341 try_futile = !may_grant(gl, gh);
1342 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1346 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1347 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
1348 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
1349 goto trap_recursive;
1351 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1353 gh->gh_error = GLR_TRYFAILED;
1354 gfs2_holder_wake(gh);
1357 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1359 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1360 insert_pt = &gh2->gh_list;
1362 trace_gfs2_glock_queue(gh, 1);
1363 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1364 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1365 if (likely(insert_pt == NULL)) {
1366 list_add_tail(&gh->gh_list, &gl->gl_holders);
1367 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1371 list_add_tail(&gh->gh_list, insert_pt);
1373 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
1374 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1375 spin_unlock(&gl->gl_lockref.lock);
1376 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1377 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1378 spin_lock(&gl->gl_lockref.lock);
1383 fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1384 fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1385 fs_err(sdp, "lock type: %d req lock state : %d\n",
1386 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1387 fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1388 fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1389 fs_err(sdp, "lock type: %d req lock state : %d\n",
1390 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1391 gfs2_dump_glock(NULL, gl, true);
1396 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1397 * @gh: the holder structure
1399 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1401 * Returns: 0, GLR_TRYFAILED, or errno on failure
1404 int gfs2_glock_nq(struct gfs2_holder *gh)
1406 struct gfs2_glock *gl = gh->gh_gl;
1409 if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP))
1412 if (test_bit(GLF_LRU, &gl->gl_flags))
1413 gfs2_glock_remove_from_lru(gl);
1415 spin_lock(&gl->gl_lockref.lock);
1417 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1418 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1419 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1420 gl->gl_lockref.count++;
1421 __gfs2_glock_queue_work(gl, 0);
1424 spin_unlock(&gl->gl_lockref.lock);
1426 if (!(gh->gh_flags & GL_ASYNC))
1427 error = gfs2_glock_wait(gh);
1433 * gfs2_glock_poll - poll to see if an async request has been completed
1436 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1439 int gfs2_glock_poll(struct gfs2_holder *gh)
1441 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1445 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1446 * @gh: the glock holder
1450 void gfs2_glock_dq(struct gfs2_holder *gh)
1452 struct gfs2_glock *gl = gh->gh_gl;
1453 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1457 spin_lock(&gl->gl_lockref.lock);
1459 * If we're in the process of file system withdraw, we cannot just
1460 * dequeue any glocks until our journal is recovered, lest we
1461 * introduce file system corruption. We need two exceptions to this
1462 * rule: We need to allow unlocking of nondisk glocks and the glock
1463 * for our own journal that needs recovery.
1465 if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) &&
1466 glock_blocked_by_withdraw(gl) &&
1467 gh->gh_gl != sdp->sd_jinode_gl) {
1468 sdp->sd_glock_dqs_held++;
1470 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
1471 TASK_UNINTERRUPTIBLE);
1473 if (gh->gh_flags & GL_NOCACHE)
1474 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1476 list_del_init(&gh->gh_list);
1477 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1478 if (find_first_holder(gl) == NULL) {
1479 if (list_empty(&gl->gl_holders) &&
1480 !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1481 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1484 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1485 gfs2_glock_add_to_lru(gl);
1487 trace_gfs2_glock_queue(gh, 0);
1488 if (unlikely(!fast_path)) {
1489 gl->gl_lockref.count++;
1490 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1491 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1492 gl->gl_name.ln_type == LM_TYPE_INODE)
1493 delay = gl->gl_hold_time;
1494 __gfs2_glock_queue_work(gl, delay);
1496 spin_unlock(&gl->gl_lockref.lock);
1499 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1501 struct gfs2_glock *gl = gh->gh_gl;
1504 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1508 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1509 * @gh: the holder structure
1513 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1516 gfs2_holder_uninit(gh);
1520 * gfs2_glock_nq_num - acquire a glock based on lock number
1521 * @sdp: the filesystem
1522 * @number: the lock number
1523 * @glops: the glock operations for the type of glock
1524 * @state: the state to acquire the glock in
1525 * @flags: modifier flags for the acquisition
1526 * @gh: the struct gfs2_holder
1531 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1532 const struct gfs2_glock_operations *glops,
1533 unsigned int state, u16 flags, struct gfs2_holder *gh)
1535 struct gfs2_glock *gl;
1538 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1540 error = gfs2_glock_nq_init(gl, state, flags, gh);
1548 * glock_compare - Compare two struct gfs2_glock structures for sorting
1549 * @arg_a: the first structure
1550 * @arg_b: the second structure
1554 static int glock_compare(const void *arg_a, const void *arg_b)
1556 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1557 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1558 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1559 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1561 if (a->ln_number > b->ln_number)
1563 if (a->ln_number < b->ln_number)
1565 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1570 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1571 * @num_gh: the number of structures
1572 * @ghs: an array of struct gfs2_holder structures
1573 * @p: placeholder for the holder structure to pass back
1575 * Returns: 0 on success (all glocks acquired),
1576 * errno on failure (no glocks acquired)
1579 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1580 struct gfs2_holder **p)
1585 for (x = 0; x < num_gh; x++)
1588 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1590 for (x = 0; x < num_gh; x++) {
1591 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1593 error = gfs2_glock_nq(p[x]);
1596 gfs2_glock_dq(p[x]);
1605 * gfs2_glock_nq_m - acquire multiple glocks
1606 * @num_gh: the number of structures
1607 * @ghs: an array of struct gfs2_holder structures
1610 * Returns: 0 on success (all glocks acquired),
1611 * errno on failure (no glocks acquired)
1614 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1616 struct gfs2_holder *tmp[4];
1617 struct gfs2_holder **pph = tmp;
1624 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1625 return gfs2_glock_nq(ghs);
1629 pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *),
1635 error = nq_m_sync(num_gh, ghs, pph);
1644 * gfs2_glock_dq_m - release multiple glocks
1645 * @num_gh: the number of structures
1646 * @ghs: an array of struct gfs2_holder structures
1650 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1653 gfs2_glock_dq(&ghs[num_gh]);
1656 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1658 unsigned long delay = 0;
1659 unsigned long holdtime;
1660 unsigned long now = jiffies;
1662 gfs2_glock_hold(gl);
1663 spin_lock(&gl->gl_lockref.lock);
1664 holdtime = gl->gl_tchange + gl->gl_hold_time;
1665 if (!list_empty(&gl->gl_holders) &&
1666 gl->gl_name.ln_type == LM_TYPE_INODE) {
1667 if (time_before(now, holdtime))
1668 delay = holdtime - now;
1669 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1670 delay = gl->gl_hold_time;
1672 handle_callback(gl, state, delay, true);
1673 __gfs2_glock_queue_work(gl, delay);
1674 spin_unlock(&gl->gl_lockref.lock);
1678 * gfs2_should_freeze - Figure out if glock should be frozen
1679 * @gl: The glock in question
1681 * Glocks are not frozen if (a) the result of the dlm operation is
1682 * an error, (b) the locking operation was an unlock operation or
1683 * (c) if there is a "noexp" flagged request anywhere in the queue
1685 * Returns: 1 if freezing should occur, 0 otherwise
1688 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1690 const struct gfs2_holder *gh;
1692 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1694 if (gl->gl_target == LM_ST_UNLOCKED)
1697 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1698 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1700 if (LM_FLAG_NOEXP & gh->gh_flags)
1708 * gfs2_glock_complete - Callback used by locking
1709 * @gl: Pointer to the glock
1710 * @ret: The return value from the dlm
1712 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1713 * to use a bitfield shared with other glock state fields.
1716 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1718 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1720 spin_lock(&gl->gl_lockref.lock);
1723 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1724 if (gfs2_should_freeze(gl)) {
1725 set_bit(GLF_FROZEN, &gl->gl_flags);
1726 spin_unlock(&gl->gl_lockref.lock);
1731 gl->gl_lockref.count++;
1732 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1733 __gfs2_glock_queue_work(gl, 0);
1734 spin_unlock(&gl->gl_lockref.lock);
1737 static int glock_cmp(void *priv, const struct list_head *a,
1738 const struct list_head *b)
1740 struct gfs2_glock *gla, *glb;
1742 gla = list_entry(a, struct gfs2_glock, gl_lru);
1743 glb = list_entry(b, struct gfs2_glock, gl_lru);
1745 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1747 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1754 * gfs2_dispose_glock_lru - Demote a list of glocks
1755 * @list: The list to dispose of
1757 * Disposing of glocks may involve disk accesses, so that here we sort
1758 * the glocks by number (i.e. disk location of the inodes) so that if
1759 * there are any such accesses, they'll be sent in order (mostly).
1761 * Must be called under the lru_lock, but may drop and retake this
1762 * lock. While the lru_lock is dropped, entries may vanish from the
1763 * list, but no new entries will appear on the list (since it is
1767 static void gfs2_dispose_glock_lru(struct list_head *list)
1768 __releases(&lru_lock)
1769 __acquires(&lru_lock)
1771 struct gfs2_glock *gl;
1773 list_sort(NULL, list, glock_cmp);
1775 while(!list_empty(list)) {
1776 gl = list_first_entry(list, struct gfs2_glock, gl_lru);
1777 list_del_init(&gl->gl_lru);
1778 if (!spin_trylock(&gl->gl_lockref.lock)) {
1780 list_add(&gl->gl_lru, &lru_list);
1781 set_bit(GLF_LRU, &gl->gl_flags);
1782 atomic_inc(&lru_count);
1785 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1786 spin_unlock(&gl->gl_lockref.lock);
1787 goto add_back_to_lru;
1789 gl->gl_lockref.count++;
1791 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1792 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1793 __gfs2_glock_queue_work(gl, 0);
1794 spin_unlock(&gl->gl_lockref.lock);
1795 cond_resched_lock(&lru_lock);
1800 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1801 * @nr: The number of entries to scan
1803 * This function selects the entries on the LRU which are able to
1804 * be demoted, and then kicks off the process by calling
1805 * gfs2_dispose_glock_lru() above.
1808 static long gfs2_scan_glock_lru(int nr)
1810 struct gfs2_glock *gl;
1815 spin_lock(&lru_lock);
1816 while ((nr-- >= 0) && !list_empty(&lru_list)) {
1817 gl = list_first_entry(&lru_list, struct gfs2_glock, gl_lru);
1819 /* Test for being demotable */
1820 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1821 list_move(&gl->gl_lru, &dispose);
1822 atomic_dec(&lru_count);
1823 clear_bit(GLF_LRU, &gl->gl_flags);
1828 list_move(&gl->gl_lru, &skipped);
1830 list_splice(&skipped, &lru_list);
1831 if (!list_empty(&dispose))
1832 gfs2_dispose_glock_lru(&dispose);
1833 spin_unlock(&lru_lock);
1838 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1839 struct shrink_control *sc)
1841 if (!(sc->gfp_mask & __GFP_FS))
1843 return gfs2_scan_glock_lru(sc->nr_to_scan);
1846 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1847 struct shrink_control *sc)
1849 return vfs_pressure_ratio(atomic_read(&lru_count));
1852 static struct shrinker glock_shrinker = {
1853 .seeks = DEFAULT_SEEKS,
1854 .count_objects = gfs2_glock_shrink_count,
1855 .scan_objects = gfs2_glock_shrink_scan,
1859 * glock_hash_walk - Call a function for glock in a hash bucket
1860 * @examiner: the function
1861 * @sdp: the filesystem
1863 * Note that the function can be called multiple times on the same
1864 * object. So the user must ensure that the function can cope with
1868 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1870 struct gfs2_glock *gl;
1871 struct rhashtable_iter iter;
1873 rhashtable_walk_enter(&gl_hash_table, &iter);
1876 rhashtable_walk_start(&iter);
1878 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl))
1879 if (gl->gl_name.ln_sbd == sdp &&
1880 lockref_get_not_dead(&gl->gl_lockref))
1883 rhashtable_walk_stop(&iter);
1884 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
1886 rhashtable_walk_exit(&iter);
1889 bool gfs2_queue_delete_work(struct gfs2_glock *gl, unsigned long delay)
1893 spin_lock(&gl->gl_lockref.lock);
1894 queued = queue_delayed_work(gfs2_delete_workqueue,
1895 &gl->gl_delete, delay);
1897 set_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1898 spin_unlock(&gl->gl_lockref.lock);
1902 void gfs2_cancel_delete_work(struct gfs2_glock *gl)
1904 if (cancel_delayed_work_sync(&gl->gl_delete)) {
1905 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1910 bool gfs2_delete_work_queued(const struct gfs2_glock *gl)
1912 return test_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1915 static void flush_delete_work(struct gfs2_glock *gl)
1917 if (gl->gl_name.ln_type == LM_TYPE_IOPEN) {
1918 if (cancel_delayed_work(&gl->gl_delete)) {
1919 queue_delayed_work(gfs2_delete_workqueue,
1923 gfs2_glock_queue_work(gl, 0);
1926 void gfs2_flush_delete_work(struct gfs2_sbd *sdp)
1928 glock_hash_walk(flush_delete_work, sdp);
1929 flush_workqueue(gfs2_delete_workqueue);
1933 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1934 * @gl: The glock to thaw
1938 static void thaw_glock(struct gfs2_glock *gl)
1940 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags)) {
1944 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1945 gfs2_glock_queue_work(gl, 0);
1949 * clear_glock - look at a glock and see if we can free it from glock cache
1950 * @gl: the glock to look at
1954 static void clear_glock(struct gfs2_glock *gl)
1956 gfs2_glock_remove_from_lru(gl);
1958 spin_lock(&gl->gl_lockref.lock);
1959 if (gl->gl_state != LM_ST_UNLOCKED)
1960 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1961 __gfs2_glock_queue_work(gl, 0);
1962 spin_unlock(&gl->gl_lockref.lock);
1966 * gfs2_glock_thaw - Thaw any frozen glocks
1967 * @sdp: The super block
1971 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1973 glock_hash_walk(thaw_glock, sdp);
1976 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
1978 spin_lock(&gl->gl_lockref.lock);
1979 gfs2_dump_glock(seq, gl, fsid);
1980 spin_unlock(&gl->gl_lockref.lock);
1983 static void dump_glock_func(struct gfs2_glock *gl)
1985 dump_glock(NULL, gl, true);
1989 * gfs2_gl_hash_clear - Empty out the glock hash table
1990 * @sdp: the filesystem
1992 * Called when unmounting the filesystem.
1995 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1997 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1998 flush_workqueue(glock_workqueue);
1999 glock_hash_walk(clear_glock, sdp);
2000 flush_workqueue(glock_workqueue);
2001 wait_event_timeout(sdp->sd_glock_wait,
2002 atomic_read(&sdp->sd_glock_disposal) == 0,
2004 glock_hash_walk(dump_glock_func, sdp);
2007 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
2009 struct gfs2_glock *gl = ip->i_gl;
2012 ret = gfs2_truncatei_resume(ip);
2013 gfs2_glock_assert_withdraw(gl, ret == 0);
2015 spin_lock(&gl->gl_lockref.lock);
2016 clear_bit(GLF_LOCK, &gl->gl_flags);
2018 spin_unlock(&gl->gl_lockref.lock);
2021 static const char *state2str(unsigned state)
2024 case LM_ST_UNLOCKED:
2028 case LM_ST_DEFERRED:
2030 case LM_ST_EXCLUSIVE:
2036 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
2039 if (flags & LM_FLAG_TRY)
2041 if (flags & LM_FLAG_TRY_1CB)
2043 if (flags & LM_FLAG_NOEXP)
2045 if (flags & LM_FLAG_ANY)
2047 if (flags & LM_FLAG_PRIORITY)
2049 if (flags & LM_FLAG_NODE_SCOPE)
2051 if (flags & GL_ASYNC)
2053 if (flags & GL_EXACT)
2055 if (flags & GL_NOCACHE)
2057 if (test_bit(HIF_HOLDER, &iflags))
2059 if (test_bit(HIF_WAIT, &iflags))
2061 if (test_bit(HIF_FIRST, &iflags))
2068 * dump_holder - print information about a glock holder
2069 * @seq: the seq_file struct
2070 * @gh: the glock holder
2071 * @fs_id_buf: pointer to file system id (if requested)
2075 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
2076 const char *fs_id_buf)
2078 struct task_struct *gh_owner = NULL;
2082 if (gh->gh_owner_pid)
2083 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
2084 gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2085 fs_id_buf, state2str(gh->gh_state),
2086 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
2088 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
2089 gh_owner ? gh_owner->comm : "(ended)",
2094 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
2096 const unsigned long *gflags = &gl->gl_flags;
2099 if (test_bit(GLF_LOCK, gflags))
2101 if (test_bit(GLF_DEMOTE, gflags))
2103 if (test_bit(GLF_PENDING_DEMOTE, gflags))
2105 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
2107 if (test_bit(GLF_DIRTY, gflags))
2109 if (test_bit(GLF_LFLUSH, gflags))
2111 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
2113 if (test_bit(GLF_REPLY_PENDING, gflags))
2115 if (test_bit(GLF_INITIAL, gflags))
2117 if (test_bit(GLF_FROZEN, gflags))
2119 if (!list_empty(&gl->gl_holders))
2121 if (test_bit(GLF_LRU, gflags))
2125 if (test_bit(GLF_BLOCKING, gflags))
2127 if (test_bit(GLF_PENDING_DELETE, gflags))
2129 if (test_bit(GLF_FREEING, gflags))
2136 * gfs2_dump_glock - print information about a glock
2137 * @seq: The seq_file struct
2139 * @fsid: If true, also dump the file system id
2141 * The file format is as follows:
2142 * One line per object, capital letters are used to indicate objects
2143 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2144 * other objects are indented by a single space and follow the glock to
2145 * which they are related. Fields are indicated by lower case letters
2146 * followed by a colon and the field value, except for strings which are in
2147 * [] so that its possible to see if they are composed of spaces for
2148 * example. The field's are n = number (id of the object), f = flags,
2149 * t = type, s = state, r = refcount, e = error, p = pid.
2153 void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2155 const struct gfs2_glock_operations *glops = gl->gl_ops;
2156 unsigned long long dtime;
2157 const struct gfs2_holder *gh;
2158 char gflags_buf[32];
2159 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2160 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2161 unsigned long nrpages = 0;
2163 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
2164 struct address_space *mapping = gfs2_glock2aspace(gl);
2166 nrpages = mapping->nrpages;
2168 memset(fs_id_buf, 0, sizeof(fs_id_buf));
2169 if (fsid && sdp) /* safety precaution */
2170 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2171 dtime = jiffies - gl->gl_demote_time;
2172 dtime *= 1000000/HZ; /* demote time in uSec */
2173 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
2175 gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2176 "v:%d r:%d m:%ld p:%lu\n",
2177 fs_id_buf, state2str(gl->gl_state),
2178 gl->gl_name.ln_type,
2179 (unsigned long long)gl->gl_name.ln_number,
2180 gflags2str(gflags_buf, gl),
2181 state2str(gl->gl_target),
2182 state2str(gl->gl_demote_state), dtime,
2183 atomic_read(&gl->gl_ail_count),
2184 atomic_read(&gl->gl_revokes),
2185 (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages);
2187 list_for_each_entry(gh, &gl->gl_holders, gh_list)
2188 dump_holder(seq, gh, fs_id_buf);
2190 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
2191 glops->go_dump(seq, gl, fs_id_buf);
2194 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
2196 struct gfs2_glock *gl = iter_ptr;
2198 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2199 gl->gl_name.ln_type,
2200 (unsigned long long)gl->gl_name.ln_number,
2201 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
2202 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
2203 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
2204 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
2205 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
2206 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
2207 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
2208 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
2212 static const char *gfs2_gltype[] = {
2226 static const char *gfs2_stype[] = {
2227 [GFS2_LKS_SRTT] = "srtt",
2228 [GFS2_LKS_SRTTVAR] = "srttvar",
2229 [GFS2_LKS_SRTTB] = "srttb",
2230 [GFS2_LKS_SRTTVARB] = "srttvarb",
2231 [GFS2_LKS_SIRT] = "sirt",
2232 [GFS2_LKS_SIRTVAR] = "sirtvar",
2233 [GFS2_LKS_DCOUNT] = "dlm",
2234 [GFS2_LKS_QCOUNT] = "queue",
2237 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2239 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
2241 struct gfs2_sbd *sdp = seq->private;
2242 loff_t pos = *(loff_t *)iter_ptr;
2243 unsigned index = pos >> 3;
2244 unsigned subindex = pos & 0x07;
2247 if (index == 0 && subindex != 0)
2250 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
2251 (index == 0) ? "cpu": gfs2_stype[subindex]);
2253 for_each_possible_cpu(i) {
2254 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
2257 seq_printf(seq, " %15u", i);
2259 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
2260 lkstats[index - 1].stats[subindex]);
2262 seq_putc(seq, '\n');
2266 int __init gfs2_glock_init(void)
2270 ret = rhashtable_init(&gl_hash_table, &ht_parms);
2274 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
2275 WQ_HIGHPRI | WQ_FREEZABLE, 0);
2276 if (!glock_workqueue) {
2277 rhashtable_destroy(&gl_hash_table);
2280 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
2281 WQ_MEM_RECLAIM | WQ_FREEZABLE,
2283 if (!gfs2_delete_workqueue) {
2284 destroy_workqueue(glock_workqueue);
2285 rhashtable_destroy(&gl_hash_table);
2289 ret = register_shrinker(&glock_shrinker);
2291 destroy_workqueue(gfs2_delete_workqueue);
2292 destroy_workqueue(glock_workqueue);
2293 rhashtable_destroy(&gl_hash_table);
2297 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2298 init_waitqueue_head(glock_wait_table + i);
2303 void gfs2_glock_exit(void)
2305 unregister_shrinker(&glock_shrinker);
2306 rhashtable_destroy(&gl_hash_table);
2307 destroy_workqueue(glock_workqueue);
2308 destroy_workqueue(gfs2_delete_workqueue);
2311 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2313 struct gfs2_glock *gl = gi->gl;
2318 if (!lockref_put_not_zero(&gl->gl_lockref))
2319 gfs2_glock_queue_put(gl);
2322 gl = rhashtable_walk_next(&gi->hti);
2323 if (IS_ERR_OR_NULL(gl)) {
2324 if (gl == ERR_PTR(-EAGAIN)) {
2331 if (gl->gl_name.ln_sbd != gi->sdp)
2334 if (!lockref_get_not_dead(&gl->gl_lockref))
2338 if (__lockref_is_dead(&gl->gl_lockref))
2346 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2349 struct gfs2_glock_iter *gi = seq->private;
2353 * We can either stay where we are, skip to the next hash table
2354 * entry, or start from the beginning.
2356 if (*pos < gi->last_pos) {
2357 rhashtable_walk_exit(&gi->hti);
2358 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2361 n = *pos - gi->last_pos;
2364 rhashtable_walk_start(&gi->hti);
2366 gfs2_glock_iter_next(gi, n);
2367 gi->last_pos = *pos;
2371 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2374 struct gfs2_glock_iter *gi = seq->private;
2377 gi->last_pos = *pos;
2378 gfs2_glock_iter_next(gi, 1);
2382 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2385 struct gfs2_glock_iter *gi = seq->private;
2387 rhashtable_walk_stop(&gi->hti);
2390 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2392 dump_glock(seq, iter_ptr, false);
2396 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2399 if (*pos >= GFS2_NR_SBSTATS)
2404 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2408 if (*pos >= GFS2_NR_SBSTATS)
2413 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2418 static const struct seq_operations gfs2_glock_seq_ops = {
2419 .start = gfs2_glock_seq_start,
2420 .next = gfs2_glock_seq_next,
2421 .stop = gfs2_glock_seq_stop,
2422 .show = gfs2_glock_seq_show,
2425 static const struct seq_operations gfs2_glstats_seq_ops = {
2426 .start = gfs2_glock_seq_start,
2427 .next = gfs2_glock_seq_next,
2428 .stop = gfs2_glock_seq_stop,
2429 .show = gfs2_glstats_seq_show,
2432 static const struct seq_operations gfs2_sbstats_sops = {
2433 .start = gfs2_sbstats_seq_start,
2434 .next = gfs2_sbstats_seq_next,
2435 .stop = gfs2_sbstats_seq_stop,
2436 .show = gfs2_sbstats_seq_show,
2439 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2441 static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2442 const struct seq_operations *ops)
2444 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2446 struct seq_file *seq = file->private_data;
2447 struct gfs2_glock_iter *gi = seq->private;
2449 gi->sdp = inode->i_private;
2450 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2452 seq->size = GFS2_SEQ_GOODSIZE;
2454 * Initially, we are "before" the first hash table entry; the
2455 * first call to rhashtable_walk_next gets us the first entry.
2459 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2464 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2466 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2469 static int gfs2_glocks_release(struct inode *inode, struct file *file)
2471 struct seq_file *seq = file->private_data;
2472 struct gfs2_glock_iter *gi = seq->private;
2475 gfs2_glock_put(gi->gl);
2476 rhashtable_walk_exit(&gi->hti);
2477 return seq_release_private(inode, file);
2480 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2482 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2485 static const struct file_operations gfs2_glocks_fops = {
2486 .owner = THIS_MODULE,
2487 .open = gfs2_glocks_open,
2489 .llseek = seq_lseek,
2490 .release = gfs2_glocks_release,
2493 static const struct file_operations gfs2_glstats_fops = {
2494 .owner = THIS_MODULE,
2495 .open = gfs2_glstats_open,
2497 .llseek = seq_lseek,
2498 .release = gfs2_glocks_release,
2501 DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats);
2503 void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2505 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2507 debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2510 debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2511 &gfs2_glstats_fops);
2513 debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2514 &gfs2_sbstats_fops);
2517 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2519 debugfs_remove_recursive(sdp->debugfs_dir);
2520 sdp->debugfs_dir = NULL;
2523 void gfs2_register_debugfs(void)
2525 gfs2_root = debugfs_create_dir("gfs2", NULL);
2528 void gfs2_unregister_debugfs(void)
2530 debugfs_remove(gfs2_root);