1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 #include <linux/sched.h>
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/buffer_head.h>
13 #include <linux/delay.h>
14 #include <linux/sort.h>
15 #include <linux/hash.h>
16 #include <linux/jhash.h>
17 #include <linux/kallsyms.h>
18 #include <linux/gfs2_ondisk.h>
19 #include <linux/list.h>
20 #include <linux/wait.h>
21 #include <linux/module.h>
22 #include <linux/uaccess.h>
23 #include <linux/seq_file.h>
24 #include <linux/debugfs.h>
25 #include <linux/kthread.h>
26 #include <linux/freezer.h>
27 #include <linux/workqueue.h>
28 #include <linux/jiffies.h>
29 #include <linux/rcupdate.h>
30 #include <linux/rculist_bl.h>
31 #include <linux/bit_spinlock.h>
32 #include <linux/percpu.h>
33 #include <linux/list_sort.h>
34 #include <linux/lockref.h>
35 #include <linux/rhashtable.h>
48 #define CREATE_TRACE_POINTS
49 #include "trace_gfs2.h"
51 struct gfs2_glock_iter {
52 struct gfs2_sbd *sdp; /* incore superblock */
53 struct rhashtable_iter hti; /* rhashtable iterator */
54 struct gfs2_glock *gl; /* current glock struct */
55 loff_t last_pos; /* last position */
58 typedef void (*glock_examiner) (struct gfs2_glock * gl);
60 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
62 static struct dentry *gfs2_root;
63 static struct workqueue_struct *glock_workqueue;
64 struct workqueue_struct *gfs2_delete_workqueue;
65 static LIST_HEAD(lru_list);
66 static atomic_t lru_count = ATOMIC_INIT(0);
67 static DEFINE_SPINLOCK(lru_lock);
69 #define GFS2_GL_HASH_SHIFT 15
70 #define GFS2_GL_HASH_SIZE BIT(GFS2_GL_HASH_SHIFT)
72 static const struct rhashtable_params ht_parms = {
73 .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4,
74 .key_len = offsetofend(struct lm_lockname, ln_type),
75 .key_offset = offsetof(struct gfs2_glock, gl_name),
76 .head_offset = offsetof(struct gfs2_glock, gl_node),
79 static struct rhashtable gl_hash_table;
81 #define GLOCK_WAIT_TABLE_BITS 12
82 #define GLOCK_WAIT_TABLE_SIZE (1 << GLOCK_WAIT_TABLE_BITS)
83 static wait_queue_head_t glock_wait_table[GLOCK_WAIT_TABLE_SIZE] __cacheline_aligned;
85 struct wait_glock_queue {
86 struct lm_lockname *name;
87 wait_queue_entry_t wait;
90 static int glock_wake_function(wait_queue_entry_t *wait, unsigned int mode,
93 struct wait_glock_queue *wait_glock =
94 container_of(wait, struct wait_glock_queue, wait);
95 struct lm_lockname *wait_name = wait_glock->name;
96 struct lm_lockname *wake_name = key;
98 if (wake_name->ln_sbd != wait_name->ln_sbd ||
99 wake_name->ln_number != wait_name->ln_number ||
100 wake_name->ln_type != wait_name->ln_type)
102 return autoremove_wake_function(wait, mode, sync, key);
105 static wait_queue_head_t *glock_waitqueue(struct lm_lockname *name)
107 u32 hash = jhash2((u32 *)name, ht_parms.key_len / 4, 0);
109 return glock_wait_table + hash_32(hash, GLOCK_WAIT_TABLE_BITS);
113 * wake_up_glock - Wake up waiters on a glock
116 static void wake_up_glock(struct gfs2_glock *gl)
118 wait_queue_head_t *wq = glock_waitqueue(&gl->gl_name);
120 if (waitqueue_active(wq))
121 __wake_up(wq, TASK_NORMAL, 1, &gl->gl_name);
124 static void gfs2_glock_dealloc(struct rcu_head *rcu)
126 struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
128 kfree(gl->gl_lksb.sb_lvbptr);
129 if (gl->gl_ops->go_flags & GLOF_ASPACE)
130 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
132 kmem_cache_free(gfs2_glock_cachep, gl);
136 * glock_blocked_by_withdraw - determine if we can still use a glock
139 * We need to allow some glocks to be enqueued, dequeued, promoted, and demoted
140 * when we're withdrawn. For example, to maintain metadata integrity, we should
141 * disallow the use of inode and rgrp glocks when withdrawn. Other glocks, like
142 * iopen or the transaction glocks may be safely used because none of their
143 * metadata goes through the journal. So in general, we should disallow all
144 * glocks that are journaled, and allow all the others. One exception is:
145 * we need to allow our active journal to be promoted and demoted so others
146 * may recover it and we can reacquire it when they're done.
148 static bool glock_blocked_by_withdraw(struct gfs2_glock *gl)
150 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
152 if (likely(!gfs2_withdrawn(sdp)))
154 if (gl->gl_ops->go_flags & GLOF_NONDISK)
156 if (!sdp->sd_jdesc ||
157 gl->gl_name.ln_number == sdp->sd_jdesc->jd_no_addr)
162 void gfs2_glock_free(struct gfs2_glock *gl)
164 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
166 gfs2_glock_assert_withdraw(gl, atomic_read(&gl->gl_revokes) == 0);
167 rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
170 call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
171 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
172 wake_up(&sdp->sd_glock_wait);
176 * gfs2_glock_hold() - increment reference count on glock
177 * @gl: The glock to hold
181 void gfs2_glock_hold(struct gfs2_glock *gl)
183 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
184 lockref_get(&gl->gl_lockref);
188 * demote_ok - Check to see if it's ok to unlock a glock
191 * Returns: 1 if it's ok
194 static int demote_ok(const struct gfs2_glock *gl)
196 const struct gfs2_glock_operations *glops = gl->gl_ops;
198 if (gl->gl_state == LM_ST_UNLOCKED)
200 if (!list_empty(&gl->gl_holders))
202 if (glops->go_demote_ok)
203 return glops->go_demote_ok(gl);
208 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
210 if (!(gl->gl_ops->go_flags & GLOF_LRU))
213 spin_lock(&lru_lock);
215 list_del(&gl->gl_lru);
216 list_add_tail(&gl->gl_lru, &lru_list);
218 if (!test_bit(GLF_LRU, &gl->gl_flags)) {
219 set_bit(GLF_LRU, &gl->gl_flags);
220 atomic_inc(&lru_count);
223 spin_unlock(&lru_lock);
226 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
228 if (!(gl->gl_ops->go_flags & GLOF_LRU))
231 spin_lock(&lru_lock);
232 if (test_bit(GLF_LRU, &gl->gl_flags)) {
233 list_del_init(&gl->gl_lru);
234 atomic_dec(&lru_count);
235 clear_bit(GLF_LRU, &gl->gl_flags);
237 spin_unlock(&lru_lock);
241 * Enqueue the glock on the work queue. Passes one glock reference on to the
244 static void __gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
245 if (!queue_delayed_work(glock_workqueue, &gl->gl_work, delay)) {
247 * We are holding the lockref spinlock, and the work was still
248 * queued above. The queued work (glock_work_func) takes that
249 * spinlock before dropping its glock reference(s), so it
250 * cannot have dropped them in the meantime.
252 GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2);
253 gl->gl_lockref.count--;
257 static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
258 spin_lock(&gl->gl_lockref.lock);
259 __gfs2_glock_queue_work(gl, delay);
260 spin_unlock(&gl->gl_lockref.lock);
263 static void __gfs2_glock_put(struct gfs2_glock *gl)
265 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
266 struct address_space *mapping = gfs2_glock2aspace(gl);
268 lockref_mark_dead(&gl->gl_lockref);
270 gfs2_glock_remove_from_lru(gl);
271 spin_unlock(&gl->gl_lockref.lock);
272 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
274 truncate_inode_pages_final(mapping);
275 if (!gfs2_withdrawn(sdp))
276 GLOCK_BUG_ON(gl, mapping->nrpages ||
277 mapping->nrexceptional);
279 trace_gfs2_glock_put(gl);
280 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
284 * Cause the glock to be put in work queue context.
286 void gfs2_glock_queue_put(struct gfs2_glock *gl)
288 gfs2_glock_queue_work(gl, 0);
292 * gfs2_glock_put() - Decrement reference count on glock
293 * @gl: The glock to put
297 void gfs2_glock_put(struct gfs2_glock *gl)
299 if (lockref_put_or_lock(&gl->gl_lockref))
302 __gfs2_glock_put(gl);
306 * may_grant - check if its ok to grant a new lock
308 * @gh: The lock request which we wish to grant
310 * Returns: true if its ok to grant the lock
313 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
315 const struct gfs2_holder *gh_head = list_first_entry(&gl->gl_holders, const struct gfs2_holder, gh_list);
316 if ((gh->gh_state == LM_ST_EXCLUSIVE ||
317 gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
319 if (gl->gl_state == gh->gh_state)
321 if (gh->gh_flags & GL_EXACT)
323 if (gl->gl_state == LM_ST_EXCLUSIVE) {
324 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
326 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
329 if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
334 static void gfs2_holder_wake(struct gfs2_holder *gh)
336 clear_bit(HIF_WAIT, &gh->gh_iflags);
337 smp_mb__after_atomic();
338 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
339 if (gh->gh_flags & GL_ASYNC) {
340 struct gfs2_sbd *sdp = gh->gh_gl->gl_name.ln_sbd;
342 wake_up(&sdp->sd_async_glock_wait);
347 * do_error - Something unexpected has happened during a lock request
351 static void do_error(struct gfs2_glock *gl, const int ret)
353 struct gfs2_holder *gh, *tmp;
355 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
356 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
358 if (ret & LM_OUT_ERROR)
360 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
361 gh->gh_error = GLR_TRYFAILED;
364 list_del_init(&gh->gh_list);
365 trace_gfs2_glock_queue(gh, 0);
366 gfs2_holder_wake(gh);
371 * do_promote - promote as many requests as possible on the current queue
374 * Returns: 1 if there is a blocked holder at the head of the list, or 2
375 * if a type specific operation is underway.
378 static int do_promote(struct gfs2_glock *gl)
379 __releases(&gl->gl_lockref.lock)
380 __acquires(&gl->gl_lockref.lock)
382 const struct gfs2_glock_operations *glops = gl->gl_ops;
383 struct gfs2_holder *gh, *tmp;
387 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
388 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
390 if (may_grant(gl, gh)) {
391 if (gh->gh_list.prev == &gl->gl_holders &&
393 spin_unlock(&gl->gl_lockref.lock);
394 /* FIXME: eliminate this eventually */
395 ret = glops->go_lock(gh);
396 spin_lock(&gl->gl_lockref.lock);
401 list_del_init(&gh->gh_list);
402 trace_gfs2_glock_queue(gh, 0);
403 gfs2_holder_wake(gh);
406 set_bit(HIF_HOLDER, &gh->gh_iflags);
407 trace_gfs2_promote(gh, 1);
408 gfs2_holder_wake(gh);
411 set_bit(HIF_HOLDER, &gh->gh_iflags);
412 trace_gfs2_promote(gh, 0);
413 gfs2_holder_wake(gh);
416 if (gh->gh_list.prev == &gl->gl_holders)
425 * find_first_waiter - find the first gh that's waiting for the glock
429 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
431 struct gfs2_holder *gh;
433 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
434 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
441 * state_change - record that the glock is now in a different state
443 * @new_state the new state
447 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
451 held1 = (gl->gl_state != LM_ST_UNLOCKED);
452 held2 = (new_state != LM_ST_UNLOCKED);
454 if (held1 != held2) {
455 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
457 gl->gl_lockref.count++;
459 gl->gl_lockref.count--;
461 if (new_state != gl->gl_target)
462 /* shorten our minimum hold time */
463 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
465 gl->gl_state = new_state;
466 gl->gl_tchange = jiffies;
469 static void gfs2_set_demote(struct gfs2_glock *gl)
471 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
473 set_bit(GLF_DEMOTE, &gl->gl_flags);
475 wake_up(&sdp->sd_async_glock_wait);
478 static void gfs2_demote_wake(struct gfs2_glock *gl)
480 gl->gl_demote_state = LM_ST_EXCLUSIVE;
481 clear_bit(GLF_DEMOTE, &gl->gl_flags);
482 smp_mb__after_atomic();
483 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
487 * finish_xmote - The DLM has replied to one of our lock requests
489 * @ret: The status from the DLM
493 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
495 const struct gfs2_glock_operations *glops = gl->gl_ops;
496 struct gfs2_holder *gh;
497 unsigned state = ret & LM_OUT_ST_MASK;
500 spin_lock(&gl->gl_lockref.lock);
501 trace_gfs2_glock_state_change(gl, state);
502 state_change(gl, state);
503 gh = find_first_waiter(gl);
505 /* Demote to UN request arrived during demote to SH or DF */
506 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
507 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
508 gl->gl_target = LM_ST_UNLOCKED;
510 /* Check for state != intended state */
511 if (unlikely(state != gl->gl_target)) {
512 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
513 /* move to back of queue and try next entry */
514 if (ret & LM_OUT_CANCELED) {
515 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
516 list_move_tail(&gh->gh_list, &gl->gl_holders);
517 gh = find_first_waiter(gl);
518 gl->gl_target = gh->gh_state;
521 /* Some error or failed "try lock" - report it */
522 if ((ret & LM_OUT_ERROR) ||
523 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
524 gl->gl_target = gl->gl_state;
530 /* Unlocked due to conversion deadlock, try again */
533 do_xmote(gl, gh, gl->gl_target);
535 /* Conversion fails, unlock and try again */
538 do_xmote(gl, gh, LM_ST_UNLOCKED);
540 default: /* Everything else */
541 fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n",
542 gl->gl_target, state);
545 spin_unlock(&gl->gl_lockref.lock);
549 /* Fast path - we got what we asked for */
550 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
551 gfs2_demote_wake(gl);
552 if (state != LM_ST_UNLOCKED) {
553 if (glops->go_xmote_bh) {
554 spin_unlock(&gl->gl_lockref.lock);
555 rv = glops->go_xmote_bh(gl, gh);
556 spin_lock(&gl->gl_lockref.lock);
567 clear_bit(GLF_LOCK, &gl->gl_flags);
569 spin_unlock(&gl->gl_lockref.lock);
572 static bool is_system_glock(struct gfs2_glock *gl)
574 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
575 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
577 if (gl == m_ip->i_gl)
583 * do_xmote - Calls the DLM to change the state of a lock
584 * @gl: The lock state
585 * @gh: The holder (only for promotes)
586 * @target: The target lock state
590 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
591 __releases(&gl->gl_lockref.lock)
592 __acquires(&gl->gl_lockref.lock)
594 const struct gfs2_glock_operations *glops = gl->gl_ops;
595 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
596 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
599 if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) &&
600 gh && !(gh->gh_flags & LM_FLAG_NOEXP))
602 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
604 GLOCK_BUG_ON(gl, gl->gl_state == target);
605 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
606 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
609 * If another process is already doing the invalidate, let that
610 * finish first. The glock state machine will get back to this
611 * holder again later.
613 if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
616 do_error(gl, 0); /* Fail queued try locks */
619 set_bit(GLF_BLOCKING, &gl->gl_flags);
620 if ((gl->gl_req == LM_ST_UNLOCKED) ||
621 (gl->gl_state == LM_ST_EXCLUSIVE) ||
622 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
623 clear_bit(GLF_BLOCKING, &gl->gl_flags);
624 spin_unlock(&gl->gl_lockref.lock);
625 if (glops->go_sync) {
626 ret = glops->go_sync(gl);
627 /* If we had a problem syncing (due to io errors or whatever,
628 * we should not invalidate the metadata or tell dlm to
629 * release the glock to other nodes.
632 if (cmpxchg(&sdp->sd_log_error, 0, ret)) {
633 fs_err(sdp, "Error %d syncing glock \n", ret);
634 gfs2_dump_glock(NULL, gl, true);
639 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) {
641 * The call to go_sync should have cleared out the ail list.
642 * If there are still items, we have a problem. We ought to
643 * withdraw, but we can't because the withdraw code also uses
644 * glocks. Warn about the error, dump the glock, then fall
645 * through and wait for logd to do the withdraw for us.
647 if ((atomic_read(&gl->gl_ail_count) != 0) &&
648 (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) {
649 gfs2_glock_assert_warn(gl,
650 !atomic_read(&gl->gl_ail_count));
651 gfs2_dump_glock(NULL, gl, true);
653 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
654 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
660 * Check for an error encountered since we called go_sync and go_inval.
661 * If so, we can't withdraw from the glock code because the withdraw
662 * code itself uses glocks (see function signal_our_withdraw) to
663 * change the mount to read-only. Most importantly, we must not call
664 * dlm to unlock the glock until the journal is in a known good state
665 * (after journal replay) otherwise other nodes may use the object
666 * (rgrp or dinode) and then later, journal replay will corrupt the
667 * file system. The best we can do here is wait for the logd daemon
668 * to see sd_log_error and withdraw, and in the meantime, requeue the
671 * We make a special exception for some system glocks, such as the
672 * system statfs inode glock, which needs to be granted before the
673 * gfs2_quotad daemon can exit, and that exit needs to finish before
674 * we can unmount the withdrawn file system.
676 * However, if we're just unlocking the lock (say, for unmount, when
677 * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
678 * then it's okay to tell dlm to unlock it.
680 if (unlikely(sdp->sd_log_error && !gfs2_withdrawn(sdp)))
681 gfs2_withdraw_delayed(sdp);
682 if (glock_blocked_by_withdraw(gl) &&
683 (target != LM_ST_UNLOCKED ||
684 test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags))) {
685 if (!is_system_glock(gl)) {
686 gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD);
689 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
693 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
694 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
697 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
698 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
699 target == LM_ST_UNLOCKED &&
700 test_bit(DFL_UNMOUNT, &ls->ls_recover_flags)) {
701 finish_xmote(gl, target);
702 gfs2_glock_queue_work(gl, 0);
704 fs_err(sdp, "lm_lock ret %d\n", ret);
705 GLOCK_BUG_ON(gl, !gfs2_withdrawn(sdp));
707 } else { /* lock_nolock */
708 finish_xmote(gl, target);
709 gfs2_glock_queue_work(gl, 0);
712 spin_lock(&gl->gl_lockref.lock);
716 * find_first_holder - find the first "holder" gh
720 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
722 struct gfs2_holder *gh;
724 if (!list_empty(&gl->gl_holders)) {
725 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
726 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
733 * run_queue - do all outstanding tasks related to a glock
734 * @gl: The glock in question
735 * @nonblock: True if we must not block in run_queue
739 static void run_queue(struct gfs2_glock *gl, const int nonblock)
740 __releases(&gl->gl_lockref.lock)
741 __acquires(&gl->gl_lockref.lock)
743 struct gfs2_holder *gh = NULL;
746 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
749 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
751 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
752 gl->gl_demote_state != gl->gl_state) {
753 if (find_first_holder(gl))
757 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
758 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
759 gl->gl_target = gl->gl_demote_state;
761 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
762 gfs2_demote_wake(gl);
763 ret = do_promote(gl);
768 gh = find_first_waiter(gl);
769 gl->gl_target = gh->gh_state;
770 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
771 do_error(gl, 0); /* Fail queued try locks */
773 do_xmote(gl, gh, gl->gl_target);
778 clear_bit(GLF_LOCK, &gl->gl_flags);
779 smp_mb__after_atomic();
780 gl->gl_lockref.count++;
781 __gfs2_glock_queue_work(gl, 0);
785 clear_bit(GLF_LOCK, &gl->gl_flags);
786 smp_mb__after_atomic();
790 void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation)
792 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
794 if (ri->ri_magic == 0)
795 ri->ri_magic = cpu_to_be32(GFS2_MAGIC);
796 if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC))
797 ri->ri_generation_deleted = cpu_to_be64(generation);
800 bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation)
802 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
804 if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC))
806 return generation <= be64_to_cpu(ri->ri_generation_deleted);
809 static void gfs2_glock_poke(struct gfs2_glock *gl)
811 int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP;
812 struct gfs2_holder gh;
815 gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh);
816 error = gfs2_glock_nq(&gh);
819 gfs2_holder_uninit(&gh);
822 static bool gfs2_try_evict(struct gfs2_glock *gl)
824 struct gfs2_inode *ip;
825 bool evicted = false;
828 * If there is contention on the iopen glock and we have an inode, try
829 * to grab and release the inode so that it can be evicted. This will
830 * allow the remote node to go ahead and delete the inode without us
831 * having to do it, which will avoid rgrp glock thrashing.
833 * The remote node is likely still holding the corresponding inode
834 * glock, so it will run before we get to verify that the delete has
837 spin_lock(&gl->gl_lockref.lock);
839 if (ip && !igrab(&ip->i_inode))
841 spin_unlock(&gl->gl_lockref.lock);
843 struct gfs2_glock *inode_gl = NULL;
845 gl->gl_no_formal_ino = ip->i_no_formal_ino;
846 set_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
847 d_prune_aliases(&ip->i_inode);
850 /* If the inode was evicted, gl->gl_object will now be NULL. */
851 spin_lock(&gl->gl_lockref.lock);
855 lockref_get(&inode_gl->gl_lockref);
856 clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
858 spin_unlock(&gl->gl_lockref.lock);
860 gfs2_glock_poke(inode_gl);
861 gfs2_glock_put(inode_gl);
868 static void delete_work_func(struct work_struct *work)
870 struct delayed_work *dwork = to_delayed_work(work);
871 struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete);
872 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
874 u64 no_addr = gl->gl_name.ln_number;
876 spin_lock(&gl->gl_lockref.lock);
877 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
878 spin_unlock(&gl->gl_lockref.lock);
880 /* If someone's using this glock to create a new dinode, the block must
881 have been freed by another node, then re-used, in which case our
882 iopen callback is too late after the fact. Ignore it. */
883 if (test_bit(GLF_INODE_CREATING, &gl->gl_flags))
886 if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
888 * If we can evict the inode, give the remote node trying to
889 * delete the inode some time before verifying that the delete
890 * has happened. Otherwise, if we cause contention on the inode glock
891 * immediately, the remote node will think that we still have
892 * the inode in use, and so it will give up waiting.
894 * If we can't evict the inode, signal to the remote node that
895 * the inode is still in use. We'll later try to delete the
896 * inode locally in gfs2_evict_inode.
898 * FIXME: We only need to verify that the remote node has
899 * deleted the inode because nodes before this remote delete
900 * rework won't cooperate. At a later time, when we no longer
901 * care about compatibility with such nodes, we can skip this
904 if (gfs2_try_evict(gl)) {
905 if (gfs2_queue_delete_work(gl, 5 * HZ))
911 inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino,
912 GFS2_BLKST_UNLINKED);
913 if (!IS_ERR_OR_NULL(inode)) {
914 d_prune_aliases(inode);
921 static void glock_work_func(struct work_struct *work)
923 unsigned long delay = 0;
924 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
925 unsigned int drop_refs = 1;
927 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
928 finish_xmote(gl, gl->gl_reply);
931 spin_lock(&gl->gl_lockref.lock);
932 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
933 gl->gl_state != LM_ST_UNLOCKED &&
934 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
935 unsigned long holdtime, now = jiffies;
937 holdtime = gl->gl_tchange + gl->gl_hold_time;
938 if (time_before(now, holdtime))
939 delay = holdtime - now;
942 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
948 /* Keep one glock reference for the work we requeue. */
950 if (gl->gl_name.ln_type != LM_TYPE_INODE)
952 __gfs2_glock_queue_work(gl, delay);
956 * Drop the remaining glock references manually here. (Mind that
957 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
960 gl->gl_lockref.count -= drop_refs;
961 if (!gl->gl_lockref.count) {
962 __gfs2_glock_put(gl);
965 spin_unlock(&gl->gl_lockref.lock);
968 static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
969 struct gfs2_glock *new)
971 struct wait_glock_queue wait;
972 wait_queue_head_t *wq = glock_waitqueue(name);
973 struct gfs2_glock *gl;
976 init_wait(&wait.wait);
977 wait.wait.func = glock_wake_function;
980 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
983 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
984 &new->gl_node, ht_parms);
988 gl = rhashtable_lookup_fast(&gl_hash_table,
991 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
998 finish_wait(wq, &wait.wait);
1003 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
1004 * @sdp: The GFS2 superblock
1005 * @number: the lock number
1006 * @glops: The glock_operations to use
1007 * @create: If 0, don't create the glock if it doesn't exist
1008 * @glp: the glock is returned here
1010 * This does not lock a glock, just finds/creates structures for one.
1015 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
1016 const struct gfs2_glock_operations *glops, int create,
1017 struct gfs2_glock **glp)
1019 struct super_block *s = sdp->sd_vfs;
1020 struct lm_lockname name = { .ln_number = number,
1021 .ln_type = glops->go_type,
1023 struct gfs2_glock *gl, *tmp;
1024 struct address_space *mapping;
1025 struct kmem_cache *cachep;
1028 gl = find_insert_glock(&name, NULL);
1036 if (glops->go_flags & GLOF_ASPACE)
1037 cachep = gfs2_glock_aspace_cachep;
1039 cachep = gfs2_glock_cachep;
1040 gl = kmem_cache_alloc(cachep, GFP_NOFS);
1044 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
1046 if (glops->go_flags & GLOF_LVB) {
1047 gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1048 if (!gl->gl_lksb.sb_lvbptr) {
1049 kmem_cache_free(cachep, gl);
1054 atomic_inc(&sdp->sd_glock_disposal);
1055 gl->gl_node.next = NULL;
1058 lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass);
1059 gl->gl_lockref.count = 1;
1060 gl->gl_state = LM_ST_UNLOCKED;
1061 gl->gl_target = LM_ST_UNLOCKED;
1062 gl->gl_demote_state = LM_ST_EXCLUSIVE;
1066 /* We use the global stats to estimate the initial per-glock stats */
1067 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
1069 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
1070 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
1071 gl->gl_tchange = jiffies;
1072 gl->gl_object = NULL;
1073 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
1074 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
1075 if (gl->gl_name.ln_type == LM_TYPE_IOPEN)
1076 INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func);
1078 mapping = gfs2_glock2aspace(gl);
1080 mapping->a_ops = &gfs2_meta_aops;
1081 mapping->host = s->s_bdev->bd_inode;
1083 mapping_set_gfp_mask(mapping, GFP_NOFS);
1084 mapping->private_data = NULL;
1085 mapping->writeback_index = 0;
1088 tmp = find_insert_glock(&name, gl);
1100 kfree(gl->gl_lksb.sb_lvbptr);
1101 kmem_cache_free(cachep, gl);
1102 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
1103 wake_up(&sdp->sd_glock_wait);
1110 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
1112 * @state: the state we're requesting
1113 * @flags: the modifier flags
1114 * @gh: the holder structure
1118 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
1119 struct gfs2_holder *gh)
1121 INIT_LIST_HEAD(&gh->gh_list);
1123 gh->gh_ip = _RET_IP_;
1124 gh->gh_owner_pid = get_pid(task_pid(current));
1125 gh->gh_state = state;
1126 gh->gh_flags = flags;
1129 gfs2_glock_hold(gl);
1133 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1134 * @state: the state we're requesting
1135 * @flags: the modifier flags
1136 * @gh: the holder structure
1138 * Don't mess with the glock.
1142 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
1144 gh->gh_state = state;
1145 gh->gh_flags = flags;
1147 gh->gh_ip = _RET_IP_;
1148 put_pid(gh->gh_owner_pid);
1149 gh->gh_owner_pid = get_pid(task_pid(current));
1153 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1154 * @gh: the holder structure
1158 void gfs2_holder_uninit(struct gfs2_holder *gh)
1160 put_pid(gh->gh_owner_pid);
1161 gfs2_glock_put(gh->gh_gl);
1162 gfs2_holder_mark_uninitialized(gh);
1166 static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
1167 unsigned long start_time)
1169 /* Have we waited longer that a second? */
1170 if (time_after(jiffies, start_time + HZ)) {
1171 /* Lengthen the minimum hold time. */
1172 gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
1178 * gfs2_glock_wait - wait on a glock acquisition
1179 * @gh: the glock holder
1181 * Returns: 0 on success
1184 int gfs2_glock_wait(struct gfs2_holder *gh)
1186 unsigned long start_time = jiffies;
1189 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1190 gfs2_glock_update_hold_time(gh->gh_gl, start_time);
1191 return gh->gh_error;
1194 static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
1198 for (i = 0; i < num_gh; i++)
1199 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
1205 * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1206 * @num_gh: the number of holders in the array
1207 * @ghs: the glock holder array
1209 * Returns: 0 on success, meaning all glocks have been granted and are held.
1210 * -ESTALE if the request timed out, meaning all glocks were released,
1211 * and the caller should retry the operation.
1214 int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
1216 struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
1217 int i, ret = 0, timeout = 0;
1218 unsigned long start_time = jiffies;
1223 * Total up the (minimum hold time * 2) of all glocks and use that to
1224 * determine the max amount of time we should wait.
1226 for (i = 0; i < num_gh; i++)
1227 timeout += ghs[i].gh_gl->gl_hold_time << 1;
1230 if (!wait_event_timeout(sdp->sd_async_glock_wait,
1231 !glocks_pending(num_gh, ghs), timeout))
1232 ret = -ESTALE; /* request timed out. */
1235 * If dlm granted all our requests, we need to adjust the glock
1236 * minimum hold time values according to how long we waited.
1238 * If our request timed out, we need to repeatedly release any held
1239 * glocks we acquired thus far to allow dlm to acquire the remaining
1240 * glocks without deadlocking. We cannot currently cancel outstanding
1241 * glock acquisitions.
1243 * The HIF_WAIT bit tells us which requests still need a response from
1246 * If dlm sent us any errors, we return the first error we find.
1248 keep_waiting = false;
1249 for (i = 0; i < num_gh; i++) {
1250 /* Skip holders we have already dequeued below. */
1251 if (!gfs2_holder_queued(&ghs[i]))
1253 /* Skip holders with a pending DLM response. */
1254 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags)) {
1255 keep_waiting = true;
1259 if (test_bit(HIF_HOLDER, &ghs[i].gh_iflags)) {
1261 gfs2_glock_dq(&ghs[i]);
1263 gfs2_glock_update_hold_time(ghs[i].gh_gl,
1267 ret = ghs[i].gh_error;
1274 * At this point, we've either acquired all locks or released them all.
1280 * handle_callback - process a demote request
1282 * @state: the state the caller wants us to change to
1284 * There are only two requests that we are going to see in actual
1285 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1288 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
1289 unsigned long delay, bool remote)
1292 set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1294 gfs2_set_demote(gl);
1295 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
1296 gl->gl_demote_state = state;
1297 gl->gl_demote_time = jiffies;
1298 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
1299 gl->gl_demote_state != state) {
1300 gl->gl_demote_state = LM_ST_UNLOCKED;
1302 if (gl->gl_ops->go_callback)
1303 gl->gl_ops->go_callback(gl, remote);
1304 trace_gfs2_demote_rq(gl, remote);
1307 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
1309 struct va_format vaf;
1312 va_start(args, fmt);
1315 seq_vprintf(seq, fmt, args);
1320 pr_err("%pV", &vaf);
1327 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1328 * @gh: the holder structure to add
1330 * Eventually we should move the recursive locking trap to a
1331 * debugging option or something like that. This is the fast
1332 * path and needs to have the minimum number of distractions.
1336 static inline void add_to_queue(struct gfs2_holder *gh)
1337 __releases(&gl->gl_lockref.lock)
1338 __acquires(&gl->gl_lockref.lock)
1340 struct gfs2_glock *gl = gh->gh_gl;
1341 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1342 struct list_head *insert_pt = NULL;
1343 struct gfs2_holder *gh2;
1346 GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1347 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1348 GLOCK_BUG_ON(gl, true);
1350 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1351 if (test_bit(GLF_LOCK, &gl->gl_flags))
1352 try_futile = !may_grant(gl, gh);
1353 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1357 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1358 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
1359 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
1360 goto trap_recursive;
1362 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1364 gh->gh_error = GLR_TRYFAILED;
1365 gfs2_holder_wake(gh);
1368 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1370 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1371 insert_pt = &gh2->gh_list;
1373 trace_gfs2_glock_queue(gh, 1);
1374 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1375 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1376 if (likely(insert_pt == NULL)) {
1377 list_add_tail(&gh->gh_list, &gl->gl_holders);
1378 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1382 list_add_tail(&gh->gh_list, insert_pt);
1384 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
1385 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1386 spin_unlock(&gl->gl_lockref.lock);
1387 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1388 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1389 spin_lock(&gl->gl_lockref.lock);
1394 fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1395 fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1396 fs_err(sdp, "lock type: %d req lock state : %d\n",
1397 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1398 fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1399 fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1400 fs_err(sdp, "lock type: %d req lock state : %d\n",
1401 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1402 gfs2_dump_glock(NULL, gl, true);
1407 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1408 * @gh: the holder structure
1410 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1412 * Returns: 0, GLR_TRYFAILED, or errno on failure
1415 int gfs2_glock_nq(struct gfs2_holder *gh)
1417 struct gfs2_glock *gl = gh->gh_gl;
1420 if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP))
1423 if (test_bit(GLF_LRU, &gl->gl_flags))
1424 gfs2_glock_remove_from_lru(gl);
1426 spin_lock(&gl->gl_lockref.lock);
1428 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1429 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1430 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1431 gl->gl_lockref.count++;
1432 __gfs2_glock_queue_work(gl, 0);
1435 spin_unlock(&gl->gl_lockref.lock);
1437 if (!(gh->gh_flags & GL_ASYNC))
1438 error = gfs2_glock_wait(gh);
1444 * gfs2_glock_poll - poll to see if an async request has been completed
1447 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1450 int gfs2_glock_poll(struct gfs2_holder *gh)
1452 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1456 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1457 * @gh: the glock holder
1461 void gfs2_glock_dq(struct gfs2_holder *gh)
1463 struct gfs2_glock *gl = gh->gh_gl;
1464 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1468 spin_lock(&gl->gl_lockref.lock);
1470 * If we're in the process of file system withdraw, we cannot just
1471 * dequeue any glocks until our journal is recovered, lest we
1472 * introduce file system corruption. We need two exceptions to this
1473 * rule: We need to allow unlocking of nondisk glocks and the glock
1474 * for our own journal that needs recovery.
1476 if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) &&
1477 glock_blocked_by_withdraw(gl) &&
1478 gh->gh_gl != sdp->sd_jinode_gl) {
1479 sdp->sd_glock_dqs_held++;
1480 spin_unlock(&gl->gl_lockref.lock);
1482 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
1483 TASK_UNINTERRUPTIBLE);
1484 spin_lock(&gl->gl_lockref.lock);
1486 if (gh->gh_flags & GL_NOCACHE)
1487 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1489 list_del_init(&gh->gh_list);
1490 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1491 if (find_first_holder(gl) == NULL) {
1492 if (list_empty(&gl->gl_holders) &&
1493 !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1494 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1497 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1498 gfs2_glock_add_to_lru(gl);
1500 trace_gfs2_glock_queue(gh, 0);
1501 if (unlikely(!fast_path)) {
1502 gl->gl_lockref.count++;
1503 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1504 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1505 gl->gl_name.ln_type == LM_TYPE_INODE)
1506 delay = gl->gl_hold_time;
1507 __gfs2_glock_queue_work(gl, delay);
1509 spin_unlock(&gl->gl_lockref.lock);
1512 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1514 struct gfs2_glock *gl = gh->gh_gl;
1517 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1521 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1522 * @gh: the holder structure
1526 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1529 gfs2_holder_uninit(gh);
1533 * gfs2_glock_nq_num - acquire a glock based on lock number
1534 * @sdp: the filesystem
1535 * @number: the lock number
1536 * @glops: the glock operations for the type of glock
1537 * @state: the state to acquire the glock in
1538 * @flags: modifier flags for the acquisition
1539 * @gh: the struct gfs2_holder
1544 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1545 const struct gfs2_glock_operations *glops,
1546 unsigned int state, u16 flags, struct gfs2_holder *gh)
1548 struct gfs2_glock *gl;
1551 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1553 error = gfs2_glock_nq_init(gl, state, flags, gh);
1561 * glock_compare - Compare two struct gfs2_glock structures for sorting
1562 * @arg_a: the first structure
1563 * @arg_b: the second structure
1567 static int glock_compare(const void *arg_a, const void *arg_b)
1569 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1570 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1571 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1572 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1574 if (a->ln_number > b->ln_number)
1576 if (a->ln_number < b->ln_number)
1578 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1583 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1584 * @num_gh: the number of structures
1585 * @ghs: an array of struct gfs2_holder structures
1587 * Returns: 0 on success (all glocks acquired),
1588 * errno on failure (no glocks acquired)
1591 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1592 struct gfs2_holder **p)
1597 for (x = 0; x < num_gh; x++)
1600 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1602 for (x = 0; x < num_gh; x++) {
1603 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1605 error = gfs2_glock_nq(p[x]);
1608 gfs2_glock_dq(p[x]);
1617 * gfs2_glock_nq_m - acquire multiple glocks
1618 * @num_gh: the number of structures
1619 * @ghs: an array of struct gfs2_holder structures
1622 * Returns: 0 on success (all glocks acquired),
1623 * errno on failure (no glocks acquired)
1626 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1628 struct gfs2_holder *tmp[4];
1629 struct gfs2_holder **pph = tmp;
1636 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1637 return gfs2_glock_nq(ghs);
1641 pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *),
1647 error = nq_m_sync(num_gh, ghs, pph);
1656 * gfs2_glock_dq_m - release multiple glocks
1657 * @num_gh: the number of structures
1658 * @ghs: an array of struct gfs2_holder structures
1662 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1665 gfs2_glock_dq(&ghs[num_gh]);
1668 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1670 unsigned long delay = 0;
1671 unsigned long holdtime;
1672 unsigned long now = jiffies;
1674 gfs2_glock_hold(gl);
1675 spin_lock(&gl->gl_lockref.lock);
1676 holdtime = gl->gl_tchange + gl->gl_hold_time;
1677 if (!list_empty(&gl->gl_holders) &&
1678 gl->gl_name.ln_type == LM_TYPE_INODE) {
1679 if (time_before(now, holdtime))
1680 delay = holdtime - now;
1681 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1682 delay = gl->gl_hold_time;
1684 handle_callback(gl, state, delay, true);
1685 __gfs2_glock_queue_work(gl, delay);
1686 spin_unlock(&gl->gl_lockref.lock);
1690 * gfs2_should_freeze - Figure out if glock should be frozen
1691 * @gl: The glock in question
1693 * Glocks are not frozen if (a) the result of the dlm operation is
1694 * an error, (b) the locking operation was an unlock operation or
1695 * (c) if there is a "noexp" flagged request anywhere in the queue
1697 * Returns: 1 if freezing should occur, 0 otherwise
1700 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1702 const struct gfs2_holder *gh;
1704 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1706 if (gl->gl_target == LM_ST_UNLOCKED)
1709 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1710 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1712 if (LM_FLAG_NOEXP & gh->gh_flags)
1720 * gfs2_glock_complete - Callback used by locking
1721 * @gl: Pointer to the glock
1722 * @ret: The return value from the dlm
1724 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1725 * to use a bitfield shared with other glock state fields.
1728 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1730 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1732 spin_lock(&gl->gl_lockref.lock);
1735 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1736 if (gfs2_should_freeze(gl)) {
1737 set_bit(GLF_FROZEN, &gl->gl_flags);
1738 spin_unlock(&gl->gl_lockref.lock);
1743 gl->gl_lockref.count++;
1744 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1745 __gfs2_glock_queue_work(gl, 0);
1746 spin_unlock(&gl->gl_lockref.lock);
1749 static int glock_cmp(void *priv, const struct list_head *a,
1750 const struct list_head *b)
1752 struct gfs2_glock *gla, *glb;
1754 gla = list_entry(a, struct gfs2_glock, gl_lru);
1755 glb = list_entry(b, struct gfs2_glock, gl_lru);
1757 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1759 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1766 * gfs2_dispose_glock_lru - Demote a list of glocks
1767 * @list: The list to dispose of
1769 * Disposing of glocks may involve disk accesses, so that here we sort
1770 * the glocks by number (i.e. disk location of the inodes) so that if
1771 * there are any such accesses, they'll be sent in order (mostly).
1773 * Must be called under the lru_lock, but may drop and retake this
1774 * lock. While the lru_lock is dropped, entries may vanish from the
1775 * list, but no new entries will appear on the list (since it is
1779 static void gfs2_dispose_glock_lru(struct list_head *list)
1780 __releases(&lru_lock)
1781 __acquires(&lru_lock)
1783 struct gfs2_glock *gl;
1785 list_sort(NULL, list, glock_cmp);
1787 while(!list_empty(list)) {
1788 gl = list_first_entry(list, struct gfs2_glock, gl_lru);
1789 list_del_init(&gl->gl_lru);
1790 clear_bit(GLF_LRU, &gl->gl_flags);
1791 if (!spin_trylock(&gl->gl_lockref.lock)) {
1793 list_add(&gl->gl_lru, &lru_list);
1794 set_bit(GLF_LRU, &gl->gl_flags);
1795 atomic_inc(&lru_count);
1798 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1799 spin_unlock(&gl->gl_lockref.lock);
1800 goto add_back_to_lru;
1802 gl->gl_lockref.count++;
1804 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1805 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1806 __gfs2_glock_queue_work(gl, 0);
1807 spin_unlock(&gl->gl_lockref.lock);
1808 cond_resched_lock(&lru_lock);
1813 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1814 * @nr: The number of entries to scan
1816 * This function selects the entries on the LRU which are able to
1817 * be demoted, and then kicks off the process by calling
1818 * gfs2_dispose_glock_lru() above.
1821 static long gfs2_scan_glock_lru(int nr)
1823 struct gfs2_glock *gl;
1828 spin_lock(&lru_lock);
1829 while ((nr-- >= 0) && !list_empty(&lru_list)) {
1830 gl = list_first_entry(&lru_list, struct gfs2_glock, gl_lru);
1832 /* Test for being demotable */
1833 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1834 list_move(&gl->gl_lru, &dispose);
1835 atomic_dec(&lru_count);
1840 list_move(&gl->gl_lru, &skipped);
1842 list_splice(&skipped, &lru_list);
1843 if (!list_empty(&dispose))
1844 gfs2_dispose_glock_lru(&dispose);
1845 spin_unlock(&lru_lock);
1850 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1851 struct shrink_control *sc)
1853 if (!(sc->gfp_mask & __GFP_FS))
1855 return gfs2_scan_glock_lru(sc->nr_to_scan);
1858 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1859 struct shrink_control *sc)
1861 return vfs_pressure_ratio(atomic_read(&lru_count));
1864 static struct shrinker glock_shrinker = {
1865 .seeks = DEFAULT_SEEKS,
1866 .count_objects = gfs2_glock_shrink_count,
1867 .scan_objects = gfs2_glock_shrink_scan,
1871 * glock_hash_walk - Call a function for glock in a hash bucket
1872 * @examiner: the function
1873 * @sdp: the filesystem
1875 * Note that the function can be called multiple times on the same
1876 * object. So the user must ensure that the function can cope with
1880 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1882 struct gfs2_glock *gl;
1883 struct rhashtable_iter iter;
1885 rhashtable_walk_enter(&gl_hash_table, &iter);
1888 rhashtable_walk_start(&iter);
1890 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl)) {
1891 if (gl->gl_name.ln_sbd == sdp)
1895 rhashtable_walk_stop(&iter);
1896 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
1898 rhashtable_walk_exit(&iter);
1901 bool gfs2_queue_delete_work(struct gfs2_glock *gl, unsigned long delay)
1905 spin_lock(&gl->gl_lockref.lock);
1906 queued = queue_delayed_work(gfs2_delete_workqueue,
1907 &gl->gl_delete, delay);
1909 set_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1910 spin_unlock(&gl->gl_lockref.lock);
1914 void gfs2_cancel_delete_work(struct gfs2_glock *gl)
1916 if (cancel_delayed_work(&gl->gl_delete)) {
1917 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1922 bool gfs2_delete_work_queued(const struct gfs2_glock *gl)
1924 return test_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1927 static void flush_delete_work(struct gfs2_glock *gl)
1929 if (gl->gl_name.ln_type == LM_TYPE_IOPEN) {
1930 if (cancel_delayed_work(&gl->gl_delete)) {
1931 queue_delayed_work(gfs2_delete_workqueue,
1937 void gfs2_flush_delete_work(struct gfs2_sbd *sdp)
1939 glock_hash_walk(flush_delete_work, sdp);
1940 flush_workqueue(gfs2_delete_workqueue);
1944 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1945 * @gl: The glock to thaw
1949 static void thaw_glock(struct gfs2_glock *gl)
1951 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1953 if (!lockref_get_not_dead(&gl->gl_lockref))
1955 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1956 gfs2_glock_queue_work(gl, 0);
1960 * clear_glock - look at a glock and see if we can free it from glock cache
1961 * @gl: the glock to look at
1965 static void clear_glock(struct gfs2_glock *gl)
1967 gfs2_glock_remove_from_lru(gl);
1969 spin_lock(&gl->gl_lockref.lock);
1970 if (!__lockref_is_dead(&gl->gl_lockref)) {
1971 gl->gl_lockref.count++;
1972 if (gl->gl_state != LM_ST_UNLOCKED)
1973 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1974 __gfs2_glock_queue_work(gl, 0);
1976 spin_unlock(&gl->gl_lockref.lock);
1980 * gfs2_glock_thaw - Thaw any frozen glocks
1981 * @sdp: The super block
1985 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1987 glock_hash_walk(thaw_glock, sdp);
1990 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
1992 spin_lock(&gl->gl_lockref.lock);
1993 gfs2_dump_glock(seq, gl, fsid);
1994 spin_unlock(&gl->gl_lockref.lock);
1997 static void dump_glock_func(struct gfs2_glock *gl)
1999 dump_glock(NULL, gl, true);
2003 * gfs2_gl_hash_clear - Empty out the glock hash table
2004 * @sdp: the filesystem
2005 * @wait: wait until it's all gone
2007 * Called when unmounting the filesystem.
2010 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
2012 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
2013 flush_workqueue(glock_workqueue);
2014 glock_hash_walk(clear_glock, sdp);
2015 flush_workqueue(glock_workqueue);
2016 wait_event_timeout(sdp->sd_glock_wait,
2017 atomic_read(&sdp->sd_glock_disposal) == 0,
2019 glock_hash_walk(dump_glock_func, sdp);
2022 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
2024 struct gfs2_glock *gl = ip->i_gl;
2027 ret = gfs2_truncatei_resume(ip);
2028 gfs2_glock_assert_withdraw(gl, ret == 0);
2030 spin_lock(&gl->gl_lockref.lock);
2031 clear_bit(GLF_LOCK, &gl->gl_flags);
2033 spin_unlock(&gl->gl_lockref.lock);
2036 static const char *state2str(unsigned state)
2039 case LM_ST_UNLOCKED:
2043 case LM_ST_DEFERRED:
2045 case LM_ST_EXCLUSIVE:
2051 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
2054 if (flags & LM_FLAG_TRY)
2056 if (flags & LM_FLAG_TRY_1CB)
2058 if (flags & LM_FLAG_NOEXP)
2060 if (flags & LM_FLAG_ANY)
2062 if (flags & LM_FLAG_PRIORITY)
2064 if (flags & GL_ASYNC)
2066 if (flags & GL_EXACT)
2068 if (flags & GL_NOCACHE)
2070 if (test_bit(HIF_HOLDER, &iflags))
2072 if (test_bit(HIF_WAIT, &iflags))
2074 if (test_bit(HIF_FIRST, &iflags))
2081 * dump_holder - print information about a glock holder
2082 * @seq: the seq_file struct
2083 * @gh: the glock holder
2084 * @fs_id_buf: pointer to file system id (if requested)
2088 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
2089 const char *fs_id_buf)
2091 struct task_struct *gh_owner = NULL;
2095 if (gh->gh_owner_pid)
2096 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
2097 gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2098 fs_id_buf, state2str(gh->gh_state),
2099 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
2101 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
2102 gh_owner ? gh_owner->comm : "(ended)",
2107 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
2109 const unsigned long *gflags = &gl->gl_flags;
2112 if (test_bit(GLF_LOCK, gflags))
2114 if (test_bit(GLF_DEMOTE, gflags))
2116 if (test_bit(GLF_PENDING_DEMOTE, gflags))
2118 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
2120 if (test_bit(GLF_DIRTY, gflags))
2122 if (test_bit(GLF_LFLUSH, gflags))
2124 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
2126 if (test_bit(GLF_REPLY_PENDING, gflags))
2128 if (test_bit(GLF_INITIAL, gflags))
2130 if (test_bit(GLF_FROZEN, gflags))
2132 if (!list_empty(&gl->gl_holders))
2134 if (test_bit(GLF_LRU, gflags))
2138 if (test_bit(GLF_BLOCKING, gflags))
2140 if (test_bit(GLF_INODE_CREATING, gflags))
2142 if (test_bit(GLF_PENDING_DELETE, gflags))
2144 if (test_bit(GLF_FREEING, gflags))
2151 * gfs2_dump_glock - print information about a glock
2152 * @seq: The seq_file struct
2154 * @fsid: If true, also dump the file system id
2156 * The file format is as follows:
2157 * One line per object, capital letters are used to indicate objects
2158 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2159 * other objects are indented by a single space and follow the glock to
2160 * which they are related. Fields are indicated by lower case letters
2161 * followed by a colon and the field value, except for strings which are in
2162 * [] so that its possible to see if they are composed of spaces for
2163 * example. The field's are n = number (id of the object), f = flags,
2164 * t = type, s = state, r = refcount, e = error, p = pid.
2168 void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2170 const struct gfs2_glock_operations *glops = gl->gl_ops;
2171 unsigned long long dtime;
2172 const struct gfs2_holder *gh;
2173 char gflags_buf[32];
2174 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2175 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2176 unsigned long nrpages = 0;
2178 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
2179 struct address_space *mapping = gfs2_glock2aspace(gl);
2181 nrpages = mapping->nrpages;
2183 memset(fs_id_buf, 0, sizeof(fs_id_buf));
2184 if (fsid && sdp) /* safety precaution */
2185 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2186 dtime = jiffies - gl->gl_demote_time;
2187 dtime *= 1000000/HZ; /* demote time in uSec */
2188 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
2190 gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2191 "v:%d r:%d m:%ld p:%lu\n",
2192 fs_id_buf, state2str(gl->gl_state),
2193 gl->gl_name.ln_type,
2194 (unsigned long long)gl->gl_name.ln_number,
2195 gflags2str(gflags_buf, gl),
2196 state2str(gl->gl_target),
2197 state2str(gl->gl_demote_state), dtime,
2198 atomic_read(&gl->gl_ail_count),
2199 atomic_read(&gl->gl_revokes),
2200 (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages);
2202 list_for_each_entry(gh, &gl->gl_holders, gh_list)
2203 dump_holder(seq, gh, fs_id_buf);
2205 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
2206 glops->go_dump(seq, gl, fs_id_buf);
2209 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
2211 struct gfs2_glock *gl = iter_ptr;
2213 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2214 gl->gl_name.ln_type,
2215 (unsigned long long)gl->gl_name.ln_number,
2216 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
2217 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
2218 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
2219 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
2220 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
2221 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
2222 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
2223 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
2227 static const char *gfs2_gltype[] = {
2241 static const char *gfs2_stype[] = {
2242 [GFS2_LKS_SRTT] = "srtt",
2243 [GFS2_LKS_SRTTVAR] = "srttvar",
2244 [GFS2_LKS_SRTTB] = "srttb",
2245 [GFS2_LKS_SRTTVARB] = "srttvarb",
2246 [GFS2_LKS_SIRT] = "sirt",
2247 [GFS2_LKS_SIRTVAR] = "sirtvar",
2248 [GFS2_LKS_DCOUNT] = "dlm",
2249 [GFS2_LKS_QCOUNT] = "queue",
2252 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2254 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
2256 struct gfs2_sbd *sdp = seq->private;
2257 loff_t pos = *(loff_t *)iter_ptr;
2258 unsigned index = pos >> 3;
2259 unsigned subindex = pos & 0x07;
2262 if (index == 0 && subindex != 0)
2265 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
2266 (index == 0) ? "cpu": gfs2_stype[subindex]);
2268 for_each_possible_cpu(i) {
2269 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
2272 seq_printf(seq, " %15u", i);
2274 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
2275 lkstats[index - 1].stats[subindex]);
2277 seq_putc(seq, '\n');
2281 int __init gfs2_glock_init(void)
2285 ret = rhashtable_init(&gl_hash_table, &ht_parms);
2289 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
2290 WQ_HIGHPRI | WQ_FREEZABLE, 0);
2291 if (!glock_workqueue) {
2292 rhashtable_destroy(&gl_hash_table);
2295 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
2296 WQ_MEM_RECLAIM | WQ_FREEZABLE,
2298 if (!gfs2_delete_workqueue) {
2299 destroy_workqueue(glock_workqueue);
2300 rhashtable_destroy(&gl_hash_table);
2304 ret = register_shrinker(&glock_shrinker);
2306 destroy_workqueue(gfs2_delete_workqueue);
2307 destroy_workqueue(glock_workqueue);
2308 rhashtable_destroy(&gl_hash_table);
2312 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2313 init_waitqueue_head(glock_wait_table + i);
2318 void gfs2_glock_exit(void)
2320 unregister_shrinker(&glock_shrinker);
2321 rhashtable_destroy(&gl_hash_table);
2322 destroy_workqueue(glock_workqueue);
2323 destroy_workqueue(gfs2_delete_workqueue);
2326 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2328 struct gfs2_glock *gl = gi->gl;
2333 if (!lockref_put_not_zero(&gl->gl_lockref))
2334 gfs2_glock_queue_put(gl);
2337 gl = rhashtable_walk_next(&gi->hti);
2338 if (IS_ERR_OR_NULL(gl)) {
2339 if (gl == ERR_PTR(-EAGAIN)) {
2346 if (gl->gl_name.ln_sbd != gi->sdp)
2349 if (!lockref_get_not_dead(&gl->gl_lockref))
2353 if (__lockref_is_dead(&gl->gl_lockref))
2361 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2364 struct gfs2_glock_iter *gi = seq->private;
2368 * We can either stay where we are, skip to the next hash table
2369 * entry, or start from the beginning.
2371 if (*pos < gi->last_pos) {
2372 rhashtable_walk_exit(&gi->hti);
2373 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2376 n = *pos - gi->last_pos;
2379 rhashtable_walk_start(&gi->hti);
2381 gfs2_glock_iter_next(gi, n);
2382 gi->last_pos = *pos;
2386 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2389 struct gfs2_glock_iter *gi = seq->private;
2392 gi->last_pos = *pos;
2393 gfs2_glock_iter_next(gi, 1);
2397 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2400 struct gfs2_glock_iter *gi = seq->private;
2402 rhashtable_walk_stop(&gi->hti);
2405 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2407 dump_glock(seq, iter_ptr, false);
2411 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2414 if (*pos >= GFS2_NR_SBSTATS)
2419 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2423 if (*pos >= GFS2_NR_SBSTATS)
2428 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2433 static const struct seq_operations gfs2_glock_seq_ops = {
2434 .start = gfs2_glock_seq_start,
2435 .next = gfs2_glock_seq_next,
2436 .stop = gfs2_glock_seq_stop,
2437 .show = gfs2_glock_seq_show,
2440 static const struct seq_operations gfs2_glstats_seq_ops = {
2441 .start = gfs2_glock_seq_start,
2442 .next = gfs2_glock_seq_next,
2443 .stop = gfs2_glock_seq_stop,
2444 .show = gfs2_glstats_seq_show,
2447 static const struct seq_operations gfs2_sbstats_sops = {
2448 .start = gfs2_sbstats_seq_start,
2449 .next = gfs2_sbstats_seq_next,
2450 .stop = gfs2_sbstats_seq_stop,
2451 .show = gfs2_sbstats_seq_show,
2454 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2456 static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2457 const struct seq_operations *ops)
2459 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2461 struct seq_file *seq = file->private_data;
2462 struct gfs2_glock_iter *gi = seq->private;
2464 gi->sdp = inode->i_private;
2465 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2467 seq->size = GFS2_SEQ_GOODSIZE;
2469 * Initially, we are "before" the first hash table entry; the
2470 * first call to rhashtable_walk_next gets us the first entry.
2474 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2479 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2481 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2484 static int gfs2_glocks_release(struct inode *inode, struct file *file)
2486 struct seq_file *seq = file->private_data;
2487 struct gfs2_glock_iter *gi = seq->private;
2490 gfs2_glock_put(gi->gl);
2491 rhashtable_walk_exit(&gi->hti);
2492 return seq_release_private(inode, file);
2495 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2497 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2500 static const struct file_operations gfs2_glocks_fops = {
2501 .owner = THIS_MODULE,
2502 .open = gfs2_glocks_open,
2504 .llseek = seq_lseek,
2505 .release = gfs2_glocks_release,
2508 static const struct file_operations gfs2_glstats_fops = {
2509 .owner = THIS_MODULE,
2510 .open = gfs2_glstats_open,
2512 .llseek = seq_lseek,
2513 .release = gfs2_glocks_release,
2516 DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats);
2518 void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2520 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2522 debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2525 debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2526 &gfs2_glstats_fops);
2528 debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2529 &gfs2_sbstats_fops);
2532 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2534 debugfs_remove_recursive(sdp->debugfs_dir);
2535 sdp->debugfs_dir = NULL;
2538 void gfs2_register_debugfs(void)
2540 gfs2_root = debugfs_create_dir("gfs2", NULL);
2543 void gfs2_unregister_debugfs(void)
2545 debugfs_remove(gfs2_root);