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);
61 static void __gfs2_glock_dq(struct gfs2_holder *gh);
63 static struct dentry *gfs2_root;
64 static struct workqueue_struct *glock_workqueue;
65 struct workqueue_struct *gfs2_delete_workqueue;
66 static LIST_HEAD(lru_list);
67 static atomic_t lru_count = ATOMIC_INIT(0);
68 static DEFINE_SPINLOCK(lru_lock);
70 #define GFS2_GL_HASH_SHIFT 15
71 #define GFS2_GL_HASH_SIZE BIT(GFS2_GL_HASH_SHIFT)
73 static const struct rhashtable_params ht_parms = {
74 .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4,
75 .key_len = offsetofend(struct lm_lockname, ln_type),
76 .key_offset = offsetof(struct gfs2_glock, gl_name),
77 .head_offset = offsetof(struct gfs2_glock, gl_node),
80 static struct rhashtable gl_hash_table;
82 #define GLOCK_WAIT_TABLE_BITS 12
83 #define GLOCK_WAIT_TABLE_SIZE (1 << GLOCK_WAIT_TABLE_BITS)
84 static wait_queue_head_t glock_wait_table[GLOCK_WAIT_TABLE_SIZE] __cacheline_aligned;
86 struct wait_glock_queue {
87 struct lm_lockname *name;
88 wait_queue_entry_t wait;
91 static int glock_wake_function(wait_queue_entry_t *wait, unsigned int mode,
94 struct wait_glock_queue *wait_glock =
95 container_of(wait, struct wait_glock_queue, wait);
96 struct lm_lockname *wait_name = wait_glock->name;
97 struct lm_lockname *wake_name = key;
99 if (wake_name->ln_sbd != wait_name->ln_sbd ||
100 wake_name->ln_number != wait_name->ln_number ||
101 wake_name->ln_type != wait_name->ln_type)
103 return autoremove_wake_function(wait, mode, sync, key);
106 static wait_queue_head_t *glock_waitqueue(struct lm_lockname *name)
108 u32 hash = jhash2((u32 *)name, ht_parms.key_len / 4, 0);
110 return glock_wait_table + hash_32(hash, GLOCK_WAIT_TABLE_BITS);
114 * wake_up_glock - Wake up waiters on a glock
117 static void wake_up_glock(struct gfs2_glock *gl)
119 wait_queue_head_t *wq = glock_waitqueue(&gl->gl_name);
121 if (waitqueue_active(wq))
122 __wake_up(wq, TASK_NORMAL, 1, &gl->gl_name);
125 static void gfs2_glock_dealloc(struct rcu_head *rcu)
127 struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
129 kfree(gl->gl_lksb.sb_lvbptr);
130 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
131 struct gfs2_glock_aspace *gla =
132 container_of(gl, struct gfs2_glock_aspace, glock);
133 kmem_cache_free(gfs2_glock_aspace_cachep, gla);
135 kmem_cache_free(gfs2_glock_cachep, gl);
139 * glock_blocked_by_withdraw - determine if we can still use a glock
142 * We need to allow some glocks to be enqueued, dequeued, promoted, and demoted
143 * when we're withdrawn. For example, to maintain metadata integrity, we should
144 * disallow the use of inode and rgrp glocks when withdrawn. Other glocks, like
145 * iopen or the transaction glocks may be safely used because none of their
146 * metadata goes through the journal. So in general, we should disallow all
147 * glocks that are journaled, and allow all the others. One exception is:
148 * we need to allow our active journal to be promoted and demoted so others
149 * may recover it and we can reacquire it when they're done.
151 static bool glock_blocked_by_withdraw(struct gfs2_glock *gl)
153 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
155 if (likely(!gfs2_withdrawn(sdp)))
157 if (gl->gl_ops->go_flags & GLOF_NONDISK)
159 if (!sdp->sd_jdesc ||
160 gl->gl_name.ln_number == sdp->sd_jdesc->jd_no_addr)
165 void gfs2_glock_free(struct gfs2_glock *gl)
167 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
169 gfs2_glock_assert_withdraw(gl, atomic_read(&gl->gl_revokes) == 0);
170 rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
173 call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
174 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
175 wake_up(&sdp->sd_glock_wait);
179 * gfs2_glock_hold() - increment reference count on glock
180 * @gl: The glock to hold
184 void gfs2_glock_hold(struct gfs2_glock *gl)
186 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
187 lockref_get(&gl->gl_lockref);
191 * demote_ok - Check to see if it's ok to unlock a glock
194 * Returns: 1 if it's ok
197 static int demote_ok(const struct gfs2_glock *gl)
199 const struct gfs2_glock_operations *glops = gl->gl_ops;
201 if (gl->gl_state == LM_ST_UNLOCKED)
204 * Note that demote_ok is used for the lru process of disposing of
205 * glocks. For this purpose, we don't care if the glock's holders
206 * have the HIF_MAY_DEMOTE flag set or not. If someone is using
207 * them, don't demote.
209 if (!list_empty(&gl->gl_holders))
211 if (glops->go_demote_ok)
212 return glops->go_demote_ok(gl);
217 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
219 if (!(gl->gl_ops->go_flags & GLOF_LRU))
222 spin_lock(&lru_lock);
224 list_move_tail(&gl->gl_lru, &lru_list);
226 if (!test_bit(GLF_LRU, &gl->gl_flags)) {
227 set_bit(GLF_LRU, &gl->gl_flags);
228 atomic_inc(&lru_count);
231 spin_unlock(&lru_lock);
234 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
236 if (!(gl->gl_ops->go_flags & GLOF_LRU))
239 spin_lock(&lru_lock);
240 if (test_bit(GLF_LRU, &gl->gl_flags)) {
241 list_del_init(&gl->gl_lru);
242 atomic_dec(&lru_count);
243 clear_bit(GLF_LRU, &gl->gl_flags);
245 spin_unlock(&lru_lock);
249 * Enqueue the glock on the work queue. Passes one glock reference on to the
252 static void __gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
253 if (!queue_delayed_work(glock_workqueue, &gl->gl_work, delay)) {
255 * We are holding the lockref spinlock, and the work was still
256 * queued above. The queued work (glock_work_func) takes that
257 * spinlock before dropping its glock reference(s), so it
258 * cannot have dropped them in the meantime.
260 GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2);
261 gl->gl_lockref.count--;
265 static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
266 spin_lock(&gl->gl_lockref.lock);
267 __gfs2_glock_queue_work(gl, delay);
268 spin_unlock(&gl->gl_lockref.lock);
271 static void __gfs2_glock_put(struct gfs2_glock *gl)
273 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
274 struct address_space *mapping = gfs2_glock2aspace(gl);
276 lockref_mark_dead(&gl->gl_lockref);
278 gfs2_glock_remove_from_lru(gl);
279 spin_unlock(&gl->gl_lockref.lock);
280 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
282 truncate_inode_pages_final(mapping);
283 if (!gfs2_withdrawn(sdp))
284 GLOCK_BUG_ON(gl, !mapping_empty(mapping));
286 trace_gfs2_glock_put(gl);
287 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
291 * Cause the glock to be put in work queue context.
293 void gfs2_glock_queue_put(struct gfs2_glock *gl)
295 gfs2_glock_queue_work(gl, 0);
299 * gfs2_glock_put() - Decrement reference count on glock
300 * @gl: The glock to put
304 void gfs2_glock_put(struct gfs2_glock *gl)
306 if (lockref_put_or_lock(&gl->gl_lockref))
309 __gfs2_glock_put(gl);
313 * may_grant - check if it's ok to grant a new lock
315 * @current_gh: One of the current holders of @gl
316 * @gh: The lock request which we wish to grant
318 * With our current compatibility rules, if a glock has one or more active
319 * holders (HIF_HOLDER flag set), any of those holders can be passed in as
320 * @current_gh; they are all the same as far as compatibility with the new @gh
323 * Returns true if it's ok to grant the lock.
326 static inline bool may_grant(struct gfs2_glock *gl,
327 struct gfs2_holder *current_gh,
328 struct gfs2_holder *gh)
331 GLOCK_BUG_ON(gl, !test_bit(HIF_HOLDER, ¤t_gh->gh_iflags));
333 switch(current_gh->gh_state) {
334 case LM_ST_EXCLUSIVE:
336 * Here we make a special exception to grant holders
337 * who agree to share the EX lock with other holders
338 * who also have the bit set. If the original holder
339 * has the LM_FLAG_NODE_SCOPE bit set, we grant more
340 * holders with the bit set.
342 return gh->gh_state == LM_ST_EXCLUSIVE &&
343 (current_gh->gh_flags & LM_FLAG_NODE_SCOPE) &&
344 (gh->gh_flags & LM_FLAG_NODE_SCOPE);
348 return gh->gh_state == current_gh->gh_state;
355 if (gl->gl_state == gh->gh_state)
357 if (gh->gh_flags & GL_EXACT)
359 if (gl->gl_state == LM_ST_EXCLUSIVE) {
360 return gh->gh_state == LM_ST_SHARED ||
361 gh->gh_state == LM_ST_DEFERRED;
363 if (gh->gh_flags & LM_FLAG_ANY)
364 return gl->gl_state != LM_ST_UNLOCKED;
368 static void gfs2_holder_wake(struct gfs2_holder *gh)
370 clear_bit(HIF_WAIT, &gh->gh_iflags);
371 smp_mb__after_atomic();
372 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
373 if (gh->gh_flags & GL_ASYNC) {
374 struct gfs2_sbd *sdp = gh->gh_gl->gl_name.ln_sbd;
376 wake_up(&sdp->sd_async_glock_wait);
381 * do_error - Something unexpected has happened during a lock request
383 * @ret: The status from the DLM
386 static void do_error(struct gfs2_glock *gl, const int ret)
388 struct gfs2_holder *gh, *tmp;
390 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
391 if (!test_bit(HIF_WAIT, &gh->gh_iflags))
393 if (ret & LM_OUT_ERROR)
395 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
396 gh->gh_error = GLR_TRYFAILED;
399 list_del_init(&gh->gh_list);
400 trace_gfs2_glock_queue(gh, 0);
401 gfs2_holder_wake(gh);
406 * demote_incompat_holders - demote incompatible demoteable holders
407 * @gl: the glock we want to promote
408 * @current_gh: the newly promoted holder
410 * We're passing the newly promoted holder in @current_gh, but actually, any of
411 * the strong holders would do.
413 static void demote_incompat_holders(struct gfs2_glock *gl,
414 struct gfs2_holder *current_gh)
416 struct gfs2_holder *gh, *tmp;
419 * Demote incompatible holders before we make ourselves eligible.
420 * (This holder may or may not allow auto-demoting, but we don't want
421 * to demote the new holder before it's even granted.)
423 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
425 * Since holders are at the front of the list, we stop when we
426 * find the first non-holder.
428 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
430 if (gh == current_gh)
432 if (test_bit(HIF_MAY_DEMOTE, &gh->gh_iflags) &&
433 !may_grant(gl, current_gh, gh)) {
435 * We should not recurse into do_promote because
436 * __gfs2_glock_dq only calls handle_callback,
437 * gfs2_glock_add_to_lru and __gfs2_glock_queue_work.
445 * find_first_holder - find the first "holder" gh
449 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
451 struct gfs2_holder *gh;
453 if (!list_empty(&gl->gl_holders)) {
454 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder,
456 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
463 * find_first_strong_holder - find the first non-demoteable holder
466 * Find the first holder that doesn't have the HIF_MAY_DEMOTE flag set.
468 static inline struct gfs2_holder *
469 find_first_strong_holder(struct gfs2_glock *gl)
471 struct gfs2_holder *gh;
473 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
474 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
476 if (!test_bit(HIF_MAY_DEMOTE, &gh->gh_iflags))
483 * gfs2_instantiate - Call the glops instantiate function
484 * @gh: The glock holder
486 * Returns: 0 if instantiate was successful, or error.
488 int gfs2_instantiate(struct gfs2_holder *gh)
490 struct gfs2_glock *gl = gh->gh_gl;
491 const struct gfs2_glock_operations *glops = gl->gl_ops;
495 if (!test_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags))
499 * Since we unlock the lockref lock, we set a flag to indicate
500 * instantiate is in progress.
502 if (test_and_set_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags)) {
503 wait_on_bit(&gl->gl_flags, GLF_INSTANTIATE_IN_PROG,
504 TASK_UNINTERRUPTIBLE);
506 * Here we just waited for a different instantiate to finish.
507 * But that may not have been successful, as when a process
508 * locks an inode glock _before_ it has an actual inode to
509 * instantiate into. So we check again. This process might
510 * have an inode to instantiate, so might be successful.
515 ret = glops->go_instantiate(gl);
517 clear_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags);
518 clear_and_wake_up_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags);
524 return glops->go_held(gh);
529 * do_promote - promote as many requests as possible on the current queue
532 * Returns: 1 if there is a blocked holder at the head of the list
535 static int do_promote(struct gfs2_glock *gl)
537 struct gfs2_holder *gh, *current_gh;
538 bool incompat_holders_demoted = false;
540 current_gh = find_first_strong_holder(gl);
541 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
542 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
544 if (!may_grant(gl, current_gh, gh)) {
546 * If we get here, it means we may not grant this
547 * holder for some reason. If this holder is at the
548 * head of the list, it means we have a blocked holder
549 * at the head, so return 1.
551 if (list_is_first(&gh->gh_list, &gl->gl_holders))
556 set_bit(HIF_HOLDER, &gh->gh_iflags);
557 trace_gfs2_promote(gh);
558 gfs2_holder_wake(gh);
559 if (!incompat_holders_demoted) {
561 demote_incompat_holders(gl, current_gh);
562 incompat_holders_demoted = true;
569 * find_first_waiter - find the first gh that's waiting for the glock
573 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
575 struct gfs2_holder *gh;
577 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
578 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
585 * state_change - record that the glock is now in a different state
587 * @new_state: the new state
590 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
594 held1 = (gl->gl_state != LM_ST_UNLOCKED);
595 held2 = (new_state != LM_ST_UNLOCKED);
597 if (held1 != held2) {
598 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
600 gl->gl_lockref.count++;
602 gl->gl_lockref.count--;
604 if (new_state != gl->gl_target)
605 /* shorten our minimum hold time */
606 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
608 gl->gl_state = new_state;
609 gl->gl_tchange = jiffies;
612 static void gfs2_set_demote(struct gfs2_glock *gl)
614 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
616 set_bit(GLF_DEMOTE, &gl->gl_flags);
618 wake_up(&sdp->sd_async_glock_wait);
621 static void gfs2_demote_wake(struct gfs2_glock *gl)
623 gl->gl_demote_state = LM_ST_EXCLUSIVE;
624 clear_bit(GLF_DEMOTE, &gl->gl_flags);
625 smp_mb__after_atomic();
626 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
630 * finish_xmote - The DLM has replied to one of our lock requests
632 * @ret: The status from the DLM
636 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
638 const struct gfs2_glock_operations *glops = gl->gl_ops;
639 struct gfs2_holder *gh;
640 unsigned state = ret & LM_OUT_ST_MASK;
642 spin_lock(&gl->gl_lockref.lock);
643 trace_gfs2_glock_state_change(gl, state);
644 state_change(gl, state);
645 gh = find_first_waiter(gl);
647 /* Demote to UN request arrived during demote to SH or DF */
648 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
649 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
650 gl->gl_target = LM_ST_UNLOCKED;
652 /* Check for state != intended state */
653 if (unlikely(state != gl->gl_target)) {
654 if (gh && (ret & LM_OUT_CANCELED))
655 gfs2_holder_wake(gh);
656 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
657 /* move to back of queue and try next entry */
658 if (ret & LM_OUT_CANCELED) {
659 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
660 list_move_tail(&gh->gh_list, &gl->gl_holders);
661 gh = find_first_waiter(gl);
662 gl->gl_target = gh->gh_state;
665 /* Some error or failed "try lock" - report it */
666 if ((ret & LM_OUT_ERROR) ||
667 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
668 gl->gl_target = gl->gl_state;
674 /* Unlocked due to conversion deadlock, try again */
677 do_xmote(gl, gh, gl->gl_target);
679 /* Conversion fails, unlock and try again */
682 do_xmote(gl, gh, LM_ST_UNLOCKED);
684 default: /* Everything else */
685 fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n",
686 gl->gl_target, state);
689 spin_unlock(&gl->gl_lockref.lock);
693 /* Fast path - we got what we asked for */
694 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
695 gfs2_demote_wake(gl);
696 if (state != LM_ST_UNLOCKED) {
697 if (glops->go_xmote_bh) {
700 spin_unlock(&gl->gl_lockref.lock);
701 rv = glops->go_xmote_bh(gl);
702 spin_lock(&gl->gl_lockref.lock);
711 clear_bit(GLF_LOCK, &gl->gl_flags);
712 spin_unlock(&gl->gl_lockref.lock);
715 static bool is_system_glock(struct gfs2_glock *gl)
717 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
718 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
720 if (gl == m_ip->i_gl)
726 * do_xmote - Calls the DLM to change the state of a lock
727 * @gl: The lock state
728 * @gh: The holder (only for promotes)
729 * @target: The target lock state
733 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
734 __releases(&gl->gl_lockref.lock)
735 __acquires(&gl->gl_lockref.lock)
737 const struct gfs2_glock_operations *glops = gl->gl_ops;
738 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
739 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
742 if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) &&
743 gh && !(gh->gh_flags & LM_FLAG_NOEXP))
745 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
747 GLOCK_BUG_ON(gl, gl->gl_state == target);
748 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
749 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
752 * If another process is already doing the invalidate, let that
753 * finish first. The glock state machine will get back to this
754 * holder again later.
756 if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
759 do_error(gl, 0); /* Fail queued try locks */
762 set_bit(GLF_BLOCKING, &gl->gl_flags);
763 if ((gl->gl_req == LM_ST_UNLOCKED) ||
764 (gl->gl_state == LM_ST_EXCLUSIVE) ||
765 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
766 clear_bit(GLF_BLOCKING, &gl->gl_flags);
767 spin_unlock(&gl->gl_lockref.lock);
768 if (glops->go_sync) {
769 ret = glops->go_sync(gl);
770 /* If we had a problem syncing (due to io errors or whatever,
771 * we should not invalidate the metadata or tell dlm to
772 * release the glock to other nodes.
775 if (cmpxchg(&sdp->sd_log_error, 0, ret)) {
776 fs_err(sdp, "Error %d syncing glock \n", ret);
777 gfs2_dump_glock(NULL, gl, true);
782 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) {
784 * The call to go_sync should have cleared out the ail list.
785 * If there are still items, we have a problem. We ought to
786 * withdraw, but we can't because the withdraw code also uses
787 * glocks. Warn about the error, dump the glock, then fall
788 * through and wait for logd to do the withdraw for us.
790 if ((atomic_read(&gl->gl_ail_count) != 0) &&
791 (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) {
792 gfs2_glock_assert_warn(gl,
793 !atomic_read(&gl->gl_ail_count));
794 gfs2_dump_glock(NULL, gl, true);
796 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
797 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
803 * Check for an error encountered since we called go_sync and go_inval.
804 * If so, we can't withdraw from the glock code because the withdraw
805 * code itself uses glocks (see function signal_our_withdraw) to
806 * change the mount to read-only. Most importantly, we must not call
807 * dlm to unlock the glock until the journal is in a known good state
808 * (after journal replay) otherwise other nodes may use the object
809 * (rgrp or dinode) and then later, journal replay will corrupt the
810 * file system. The best we can do here is wait for the logd daemon
811 * to see sd_log_error and withdraw, and in the meantime, requeue the
814 * We make a special exception for some system glocks, such as the
815 * system statfs inode glock, which needs to be granted before the
816 * gfs2_quotad daemon can exit, and that exit needs to finish before
817 * we can unmount the withdrawn file system.
819 * However, if we're just unlocking the lock (say, for unmount, when
820 * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
821 * then it's okay to tell dlm to unlock it.
823 if (unlikely(sdp->sd_log_error && !gfs2_withdrawn(sdp)))
824 gfs2_withdraw_delayed(sdp);
825 if (glock_blocked_by_withdraw(gl) &&
826 (target != LM_ST_UNLOCKED ||
827 test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags))) {
828 if (!is_system_glock(gl)) {
829 gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD);
832 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
836 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
838 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
839 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
840 target == LM_ST_UNLOCKED &&
841 test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
842 finish_xmote(gl, target);
843 gfs2_glock_queue_work(gl, 0);
845 fs_err(sdp, "lm_lock ret %d\n", ret);
846 GLOCK_BUG_ON(gl, !gfs2_withdrawn(sdp));
848 } else { /* lock_nolock */
849 finish_xmote(gl, target);
850 gfs2_glock_queue_work(gl, 0);
853 spin_lock(&gl->gl_lockref.lock);
857 * run_queue - do all outstanding tasks related to a glock
858 * @gl: The glock in question
859 * @nonblock: True if we must not block in run_queue
863 static void run_queue(struct gfs2_glock *gl, const int nonblock)
864 __releases(&gl->gl_lockref.lock)
865 __acquires(&gl->gl_lockref.lock)
867 struct gfs2_holder *gh = NULL;
869 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
872 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
874 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
875 gl->gl_demote_state != gl->gl_state) {
876 if (find_first_holder(gl))
880 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
881 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
882 gl->gl_target = gl->gl_demote_state;
884 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
885 gfs2_demote_wake(gl);
886 if (do_promote(gl) == 0)
888 gh = find_first_waiter(gl);
889 gl->gl_target = gh->gh_state;
890 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
891 do_error(gl, 0); /* Fail queued try locks */
893 do_xmote(gl, gh, gl->gl_target);
897 clear_bit(GLF_LOCK, &gl->gl_flags);
898 smp_mb__after_atomic();
899 gl->gl_lockref.count++;
900 __gfs2_glock_queue_work(gl, 0);
904 clear_bit(GLF_LOCK, &gl->gl_flags);
905 smp_mb__after_atomic();
909 void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation)
911 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
913 if (ri->ri_magic == 0)
914 ri->ri_magic = cpu_to_be32(GFS2_MAGIC);
915 if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC))
916 ri->ri_generation_deleted = cpu_to_be64(generation);
919 bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation)
921 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
923 if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC))
925 return generation <= be64_to_cpu(ri->ri_generation_deleted);
928 static void gfs2_glock_poke(struct gfs2_glock *gl)
930 int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP;
931 struct gfs2_holder gh;
934 __gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh, _RET_IP_);
935 error = gfs2_glock_nq(&gh);
938 gfs2_holder_uninit(&gh);
941 static bool gfs2_try_evict(struct gfs2_glock *gl)
943 struct gfs2_inode *ip;
944 bool evicted = false;
947 * If there is contention on the iopen glock and we have an inode, try
948 * to grab and release the inode so that it can be evicted. This will
949 * allow the remote node to go ahead and delete the inode without us
950 * having to do it, which will avoid rgrp glock thrashing.
952 * The remote node is likely still holding the corresponding inode
953 * glock, so it will run before we get to verify that the delete has
956 spin_lock(&gl->gl_lockref.lock);
958 if (ip && !igrab(&ip->i_inode))
960 spin_unlock(&gl->gl_lockref.lock);
962 struct gfs2_glock *inode_gl = NULL;
964 gl->gl_no_formal_ino = ip->i_no_formal_ino;
965 set_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
966 d_prune_aliases(&ip->i_inode);
969 /* If the inode was evicted, gl->gl_object will now be NULL. */
970 spin_lock(&gl->gl_lockref.lock);
974 lockref_get(&inode_gl->gl_lockref);
975 clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
977 spin_unlock(&gl->gl_lockref.lock);
979 gfs2_glock_poke(inode_gl);
980 gfs2_glock_put(inode_gl);
987 static void delete_work_func(struct work_struct *work)
989 struct delayed_work *dwork = to_delayed_work(work);
990 struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete);
991 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
993 u64 no_addr = gl->gl_name.ln_number;
995 spin_lock(&gl->gl_lockref.lock);
996 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
997 spin_unlock(&gl->gl_lockref.lock);
999 if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
1001 * If we can evict the inode, give the remote node trying to
1002 * delete the inode some time before verifying that the delete
1003 * has happened. Otherwise, if we cause contention on the inode glock
1004 * immediately, the remote node will think that we still have
1005 * the inode in use, and so it will give up waiting.
1007 * If we can't evict the inode, signal to the remote node that
1008 * the inode is still in use. We'll later try to delete the
1009 * inode locally in gfs2_evict_inode.
1011 * FIXME: We only need to verify that the remote node has
1012 * deleted the inode because nodes before this remote delete
1013 * rework won't cooperate. At a later time, when we no longer
1014 * care about compatibility with such nodes, we can skip this
1017 if (gfs2_try_evict(gl)) {
1018 if (gfs2_queue_delete_work(gl, 5 * HZ))
1024 inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino,
1025 GFS2_BLKST_UNLINKED);
1026 if (!IS_ERR_OR_NULL(inode)) {
1027 d_prune_aliases(inode);
1034 static void glock_work_func(struct work_struct *work)
1036 unsigned long delay = 0;
1037 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
1038 unsigned int drop_refs = 1;
1040 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
1041 finish_xmote(gl, gl->gl_reply);
1044 spin_lock(&gl->gl_lockref.lock);
1045 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1046 gl->gl_state != LM_ST_UNLOCKED &&
1047 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
1048 unsigned long holdtime, now = jiffies;
1050 holdtime = gl->gl_tchange + gl->gl_hold_time;
1051 if (time_before(now, holdtime))
1052 delay = holdtime - now;
1055 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1056 gfs2_set_demote(gl);
1061 /* Keep one glock reference for the work we requeue. */
1063 if (gl->gl_name.ln_type != LM_TYPE_INODE)
1065 __gfs2_glock_queue_work(gl, delay);
1069 * Drop the remaining glock references manually here. (Mind that
1070 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
1073 gl->gl_lockref.count -= drop_refs;
1074 if (!gl->gl_lockref.count) {
1075 __gfs2_glock_put(gl);
1078 spin_unlock(&gl->gl_lockref.lock);
1081 static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
1082 struct gfs2_glock *new)
1084 struct wait_glock_queue wait;
1085 wait_queue_head_t *wq = glock_waitqueue(name);
1086 struct gfs2_glock *gl;
1089 init_wait(&wait.wait);
1090 wait.wait.func = glock_wake_function;
1093 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1096 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
1097 &new->gl_node, ht_parms);
1101 gl = rhashtable_lookup_fast(&gl_hash_table,
1104 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
1111 finish_wait(wq, &wait.wait);
1116 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
1117 * @sdp: The GFS2 superblock
1118 * @number: the lock number
1119 * @glops: The glock_operations to use
1120 * @create: If 0, don't create the glock if it doesn't exist
1121 * @glp: the glock is returned here
1123 * This does not lock a glock, just finds/creates structures for one.
1128 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
1129 const struct gfs2_glock_operations *glops, int create,
1130 struct gfs2_glock **glp)
1132 struct super_block *s = sdp->sd_vfs;
1133 struct lm_lockname name = { .ln_number = number,
1134 .ln_type = glops->go_type,
1136 struct gfs2_glock *gl, *tmp;
1137 struct address_space *mapping;
1140 gl = find_insert_glock(&name, NULL);
1148 if (glops->go_flags & GLOF_ASPACE) {
1149 struct gfs2_glock_aspace *gla =
1150 kmem_cache_alloc(gfs2_glock_aspace_cachep, GFP_NOFS);
1155 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_NOFS);
1159 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
1162 if (glops->go_flags & GLOF_LVB) {
1163 gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1164 if (!gl->gl_lksb.sb_lvbptr) {
1165 gfs2_glock_dealloc(&gl->gl_rcu);
1170 atomic_inc(&sdp->sd_glock_disposal);
1171 gl->gl_node.next = NULL;
1172 gl->gl_flags = glops->go_instantiate ? BIT(GLF_INSTANTIATE_NEEDED) : 0;
1174 lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass);
1175 gl->gl_lockref.count = 1;
1176 gl->gl_state = LM_ST_UNLOCKED;
1177 gl->gl_target = LM_ST_UNLOCKED;
1178 gl->gl_demote_state = LM_ST_EXCLUSIVE;
1181 /* We use the global stats to estimate the initial per-glock stats */
1182 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
1184 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
1185 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
1186 gl->gl_tchange = jiffies;
1187 gl->gl_object = NULL;
1188 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
1189 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
1190 if (gl->gl_name.ln_type == LM_TYPE_IOPEN)
1191 INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func);
1193 mapping = gfs2_glock2aspace(gl);
1195 mapping->a_ops = &gfs2_meta_aops;
1196 mapping->host = s->s_bdev->bd_inode;
1198 mapping_set_gfp_mask(mapping, GFP_NOFS);
1199 mapping->private_data = NULL;
1200 mapping->writeback_index = 0;
1203 tmp = find_insert_glock(&name, gl);
1215 gfs2_glock_dealloc(&gl->gl_rcu);
1216 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
1217 wake_up(&sdp->sd_glock_wait);
1224 * __gfs2_holder_init - initialize a struct gfs2_holder in the default way
1226 * @state: the state we're requesting
1227 * @flags: the modifier flags
1228 * @gh: the holder structure
1232 void __gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
1233 struct gfs2_holder *gh, unsigned long ip)
1235 INIT_LIST_HEAD(&gh->gh_list);
1238 gh->gh_owner_pid = get_pid(task_pid(current));
1239 gh->gh_state = state;
1240 gh->gh_flags = flags;
1242 gfs2_glock_hold(gl);
1246 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1247 * @state: the state we're requesting
1248 * @flags: the modifier flags
1249 * @gh: the holder structure
1251 * Don't mess with the glock.
1255 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
1257 gh->gh_state = state;
1258 gh->gh_flags = flags;
1260 gh->gh_ip = _RET_IP_;
1261 put_pid(gh->gh_owner_pid);
1262 gh->gh_owner_pid = get_pid(task_pid(current));
1266 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1267 * @gh: the holder structure
1271 void gfs2_holder_uninit(struct gfs2_holder *gh)
1273 put_pid(gh->gh_owner_pid);
1274 gfs2_glock_put(gh->gh_gl);
1275 gfs2_holder_mark_uninitialized(gh);
1279 static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
1280 unsigned long start_time)
1282 /* Have we waited longer that a second? */
1283 if (time_after(jiffies, start_time + HZ)) {
1284 /* Lengthen the minimum hold time. */
1285 gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
1291 * gfs2_glock_holder_ready - holder is ready and its error code can be collected
1292 * @gh: the glock holder
1294 * Called when a glock holder no longer needs to be waited for because it is
1295 * now either held (HIF_HOLDER set; gh_error == 0), or acquiring the lock has
1296 * failed (gh_error != 0).
1299 int gfs2_glock_holder_ready(struct gfs2_holder *gh)
1301 if (gh->gh_error || (gh->gh_flags & GL_SKIP))
1302 return gh->gh_error;
1303 gh->gh_error = gfs2_instantiate(gh);
1306 return gh->gh_error;
1310 * gfs2_glock_wait - wait on a glock acquisition
1311 * @gh: the glock holder
1313 * Returns: 0 on success
1316 int gfs2_glock_wait(struct gfs2_holder *gh)
1318 unsigned long start_time = jiffies;
1321 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1322 gfs2_glock_update_hold_time(gh->gh_gl, start_time);
1323 return gfs2_glock_holder_ready(gh);
1326 static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
1330 for (i = 0; i < num_gh; i++)
1331 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
1337 * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1338 * @num_gh: the number of holders in the array
1339 * @ghs: the glock holder array
1341 * Returns: 0 on success, meaning all glocks have been granted and are held.
1342 * -ESTALE if the request timed out, meaning all glocks were released,
1343 * and the caller should retry the operation.
1346 int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
1348 struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
1349 int i, ret = 0, timeout = 0;
1350 unsigned long start_time = jiffies;
1354 * Total up the (minimum hold time * 2) of all glocks and use that to
1355 * determine the max amount of time we should wait.
1357 for (i = 0; i < num_gh; i++)
1358 timeout += ghs[i].gh_gl->gl_hold_time << 1;
1360 if (!wait_event_timeout(sdp->sd_async_glock_wait,
1361 !glocks_pending(num_gh, ghs), timeout)) {
1362 ret = -ESTALE; /* request timed out. */
1366 for (i = 0; i < num_gh; i++) {
1367 struct gfs2_holder *gh = &ghs[i];
1370 if (test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1371 gfs2_glock_update_hold_time(gh->gh_gl,
1374 ret2 = gfs2_glock_holder_ready(gh);
1381 for (i = 0; i < num_gh; i++) {
1382 struct gfs2_holder *gh = &ghs[i];
1391 * handle_callback - process a demote request
1393 * @state: the state the caller wants us to change to
1394 * @delay: zero to demote immediately; otherwise pending demote
1395 * @remote: true if this came from a different cluster node
1397 * There are only two requests that we are going to see in actual
1398 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1401 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
1402 unsigned long delay, bool remote)
1405 set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1407 gfs2_set_demote(gl);
1408 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
1409 gl->gl_demote_state = state;
1410 gl->gl_demote_time = jiffies;
1411 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
1412 gl->gl_demote_state != state) {
1413 gl->gl_demote_state = LM_ST_UNLOCKED;
1415 if (gl->gl_ops->go_callback)
1416 gl->gl_ops->go_callback(gl, remote);
1417 trace_gfs2_demote_rq(gl, remote);
1420 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
1422 struct va_format vaf;
1425 va_start(args, fmt);
1428 seq_vprintf(seq, fmt, args);
1433 pr_err("%pV", &vaf);
1440 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1441 * @gh: the holder structure to add
1443 * Eventually we should move the recursive locking trap to a
1444 * debugging option or something like that. This is the fast
1445 * path and needs to have the minimum number of distractions.
1449 static inline void add_to_queue(struct gfs2_holder *gh)
1450 __releases(&gl->gl_lockref.lock)
1451 __acquires(&gl->gl_lockref.lock)
1453 struct gfs2_glock *gl = gh->gh_gl;
1454 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1455 struct list_head *insert_pt = NULL;
1456 struct gfs2_holder *gh2;
1459 GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1460 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1461 GLOCK_BUG_ON(gl, true);
1463 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1464 if (test_bit(GLF_LOCK, &gl->gl_flags)) {
1465 struct gfs2_holder *current_gh;
1467 current_gh = find_first_strong_holder(gl);
1468 try_futile = !may_grant(gl, current_gh, gh);
1470 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1474 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1475 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
1476 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK) &&
1477 !test_bit(HIF_MAY_DEMOTE, &gh2->gh_iflags)))
1478 goto trap_recursive;
1480 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1482 gh->gh_error = GLR_TRYFAILED;
1483 gfs2_holder_wake(gh);
1486 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1488 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1489 insert_pt = &gh2->gh_list;
1491 trace_gfs2_glock_queue(gh, 1);
1492 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1493 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1494 if (likely(insert_pt == NULL)) {
1495 list_add_tail(&gh->gh_list, &gl->gl_holders);
1496 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1500 list_add_tail(&gh->gh_list, insert_pt);
1502 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
1503 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1504 spin_unlock(&gl->gl_lockref.lock);
1505 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1506 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1507 spin_lock(&gl->gl_lockref.lock);
1512 fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1513 fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1514 fs_err(sdp, "lock type: %d req lock state : %d\n",
1515 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1516 fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1517 fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1518 fs_err(sdp, "lock type: %d req lock state : %d\n",
1519 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1520 gfs2_dump_glock(NULL, gl, true);
1525 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1526 * @gh: the holder structure
1528 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1530 * Returns: 0, GLR_TRYFAILED, or errno on failure
1533 int gfs2_glock_nq(struct gfs2_holder *gh)
1535 struct gfs2_glock *gl = gh->gh_gl;
1538 if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP))
1541 if (test_bit(GLF_LRU, &gl->gl_flags))
1542 gfs2_glock_remove_from_lru(gl);
1545 spin_lock(&gl->gl_lockref.lock);
1547 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1548 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1549 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1550 gl->gl_lockref.count++;
1551 __gfs2_glock_queue_work(gl, 0);
1554 spin_unlock(&gl->gl_lockref.lock);
1556 if (!(gh->gh_flags & GL_ASYNC))
1557 error = gfs2_glock_wait(gh);
1563 * gfs2_glock_poll - poll to see if an async request has been completed
1566 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1569 int gfs2_glock_poll(struct gfs2_holder *gh)
1571 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1574 static inline bool needs_demote(struct gfs2_glock *gl)
1576 return (test_bit(GLF_DEMOTE, &gl->gl_flags) ||
1577 test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags));
1580 static void __gfs2_glock_dq(struct gfs2_holder *gh)
1582 struct gfs2_glock *gl = gh->gh_gl;
1583 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1588 * This while loop is similar to function demote_incompat_holders:
1589 * If the glock is due to be demoted (which may be from another node
1590 * or even if this holder is GL_NOCACHE), the weak holders are
1591 * demoted as well, allowing the glock to be demoted.
1595 * If we're in the process of file system withdraw, we cannot
1596 * just dequeue any glocks until our journal is recovered, lest
1597 * we introduce file system corruption. We need two exceptions
1598 * to this rule: We need to allow unlocking of nondisk glocks
1599 * and the glock for our own journal that needs recovery.
1601 if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) &&
1602 glock_blocked_by_withdraw(gl) &&
1603 gh->gh_gl != sdp->sd_jinode_gl) {
1604 sdp->sd_glock_dqs_held++;
1605 spin_unlock(&gl->gl_lockref.lock);
1607 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
1608 TASK_UNINTERRUPTIBLE);
1609 spin_lock(&gl->gl_lockref.lock);
1613 * This holder should not be cached, so mark it for demote.
1614 * Note: this should be done before the check for needs_demote
1617 if (gh->gh_flags & GL_NOCACHE)
1618 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1620 list_del_init(&gh->gh_list);
1621 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1622 trace_gfs2_glock_queue(gh, 0);
1625 * If there hasn't been a demote request we are done.
1626 * (Let the remaining holders, if any, keep holding it.)
1628 if (!needs_demote(gl)) {
1629 if (list_empty(&gl->gl_holders))
1634 * If we have another strong holder (we cannot auto-demote)
1635 * we are done. It keeps holding it until it is done.
1637 if (find_first_strong_holder(gl))
1641 * If we have a weak holder at the head of the list, it
1642 * (and all others like it) must be auto-demoted. If there
1643 * are no more weak holders, we exit the while loop.
1645 gh = find_first_holder(gl);
1648 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1649 gfs2_glock_add_to_lru(gl);
1651 if (unlikely(!fast_path)) {
1652 gl->gl_lockref.count++;
1653 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1654 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1655 gl->gl_name.ln_type == LM_TYPE_INODE)
1656 delay = gl->gl_hold_time;
1657 __gfs2_glock_queue_work(gl, delay);
1662 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1663 * @gh: the glock holder
1666 void gfs2_glock_dq(struct gfs2_holder *gh)
1668 struct gfs2_glock *gl = gh->gh_gl;
1670 spin_lock(&gl->gl_lockref.lock);
1671 if (list_is_first(&gh->gh_list, &gl->gl_holders) &&
1672 !test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1673 spin_unlock(&gl->gl_lockref.lock);
1674 gl->gl_name.ln_sbd->sd_lockstruct.ls_ops->lm_cancel(gl);
1675 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1676 spin_lock(&gl->gl_lockref.lock);
1679 __gfs2_glock_dq(gh);
1680 spin_unlock(&gl->gl_lockref.lock);
1683 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1685 struct gfs2_glock *gl = gh->gh_gl;
1688 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1692 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1693 * @gh: the holder structure
1697 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1700 gfs2_holder_uninit(gh);
1704 * gfs2_glock_nq_num - acquire a glock based on lock number
1705 * @sdp: the filesystem
1706 * @number: the lock number
1707 * @glops: the glock operations for the type of glock
1708 * @state: the state to acquire the glock in
1709 * @flags: modifier flags for the acquisition
1710 * @gh: the struct gfs2_holder
1715 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1716 const struct gfs2_glock_operations *glops,
1717 unsigned int state, u16 flags, struct gfs2_holder *gh)
1719 struct gfs2_glock *gl;
1722 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1724 error = gfs2_glock_nq_init(gl, state, flags, gh);
1732 * glock_compare - Compare two struct gfs2_glock structures for sorting
1733 * @arg_a: the first structure
1734 * @arg_b: the second structure
1738 static int glock_compare(const void *arg_a, const void *arg_b)
1740 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1741 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1742 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1743 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1745 if (a->ln_number > b->ln_number)
1747 if (a->ln_number < b->ln_number)
1749 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1754 * nq_m_sync - synchronously acquire more than one glock in deadlock free order
1755 * @num_gh: the number of structures
1756 * @ghs: an array of struct gfs2_holder structures
1757 * @p: placeholder for the holder structure to pass back
1759 * Returns: 0 on success (all glocks acquired),
1760 * errno on failure (no glocks acquired)
1763 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1764 struct gfs2_holder **p)
1769 for (x = 0; x < num_gh; x++)
1772 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1774 for (x = 0; x < num_gh; x++) {
1775 error = gfs2_glock_nq(p[x]);
1778 gfs2_glock_dq(p[x]);
1787 * gfs2_glock_nq_m - acquire multiple glocks
1788 * @num_gh: the number of structures
1789 * @ghs: an array of struct gfs2_holder structures
1791 * Returns: 0 on success (all glocks acquired),
1792 * errno on failure (no glocks acquired)
1795 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1797 struct gfs2_holder *tmp[4];
1798 struct gfs2_holder **pph = tmp;
1805 return gfs2_glock_nq(ghs);
1809 pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *),
1815 error = nq_m_sync(num_gh, ghs, pph);
1824 * gfs2_glock_dq_m - release multiple glocks
1825 * @num_gh: the number of structures
1826 * @ghs: an array of struct gfs2_holder structures
1830 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1833 gfs2_glock_dq(&ghs[num_gh]);
1836 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1838 unsigned long delay = 0;
1839 unsigned long holdtime;
1840 unsigned long now = jiffies;
1842 gfs2_glock_hold(gl);
1843 spin_lock(&gl->gl_lockref.lock);
1844 holdtime = gl->gl_tchange + gl->gl_hold_time;
1845 if (!list_empty(&gl->gl_holders) &&
1846 gl->gl_name.ln_type == LM_TYPE_INODE) {
1847 if (time_before(now, holdtime))
1848 delay = holdtime - now;
1849 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1850 delay = gl->gl_hold_time;
1853 * Note 1: We cannot call demote_incompat_holders from handle_callback
1854 * or gfs2_set_demote due to recursion problems like: gfs2_glock_dq ->
1855 * handle_callback -> demote_incompat_holders -> gfs2_glock_dq
1856 * Plus, we only want to demote the holders if the request comes from
1857 * a remote cluster node because local holder conflicts are resolved
1860 * Note 2: if a remote node wants this glock in EX mode, lock_dlm will
1861 * request that we set our state to UNLOCKED. Here we mock up a holder
1862 * to make it look like someone wants the lock EX locally. Any SH
1863 * and DF requests should be able to share the lock without demoting.
1865 * Note 3: We only want to demote the demoteable holders when there
1866 * are no more strong holders. The demoteable holders might as well
1867 * keep the glock until the last strong holder is done with it.
1869 if (!find_first_strong_holder(gl)) {
1870 struct gfs2_holder mock_gh = {
1872 .gh_state = (state == LM_ST_UNLOCKED) ?
1873 LM_ST_EXCLUSIVE : state,
1874 .gh_iflags = BIT(HIF_HOLDER)
1877 demote_incompat_holders(gl, &mock_gh);
1879 handle_callback(gl, state, delay, true);
1880 __gfs2_glock_queue_work(gl, delay);
1881 spin_unlock(&gl->gl_lockref.lock);
1885 * gfs2_should_freeze - Figure out if glock should be frozen
1886 * @gl: The glock in question
1888 * Glocks are not frozen if (a) the result of the dlm operation is
1889 * an error, (b) the locking operation was an unlock operation or
1890 * (c) if there is a "noexp" flagged request anywhere in the queue
1892 * Returns: 1 if freezing should occur, 0 otherwise
1895 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1897 const struct gfs2_holder *gh;
1899 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1901 if (gl->gl_target == LM_ST_UNLOCKED)
1904 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1905 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1907 if (LM_FLAG_NOEXP & gh->gh_flags)
1915 * gfs2_glock_complete - Callback used by locking
1916 * @gl: Pointer to the glock
1917 * @ret: The return value from the dlm
1919 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1920 * to use a bitfield shared with other glock state fields.
1923 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1925 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1927 spin_lock(&gl->gl_lockref.lock);
1930 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1931 if (gfs2_should_freeze(gl)) {
1932 set_bit(GLF_FROZEN, &gl->gl_flags);
1933 spin_unlock(&gl->gl_lockref.lock);
1938 gl->gl_lockref.count++;
1939 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1940 __gfs2_glock_queue_work(gl, 0);
1941 spin_unlock(&gl->gl_lockref.lock);
1944 static int glock_cmp(void *priv, const struct list_head *a,
1945 const struct list_head *b)
1947 struct gfs2_glock *gla, *glb;
1949 gla = list_entry(a, struct gfs2_glock, gl_lru);
1950 glb = list_entry(b, struct gfs2_glock, gl_lru);
1952 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1954 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1961 * gfs2_dispose_glock_lru - Demote a list of glocks
1962 * @list: The list to dispose of
1964 * Disposing of glocks may involve disk accesses, so that here we sort
1965 * the glocks by number (i.e. disk location of the inodes) so that if
1966 * there are any such accesses, they'll be sent in order (mostly).
1968 * Must be called under the lru_lock, but may drop and retake this
1969 * lock. While the lru_lock is dropped, entries may vanish from the
1970 * list, but no new entries will appear on the list (since it is
1974 static void gfs2_dispose_glock_lru(struct list_head *list)
1975 __releases(&lru_lock)
1976 __acquires(&lru_lock)
1978 struct gfs2_glock *gl;
1980 list_sort(NULL, list, glock_cmp);
1982 while(!list_empty(list)) {
1983 gl = list_first_entry(list, struct gfs2_glock, gl_lru);
1984 list_del_init(&gl->gl_lru);
1985 clear_bit(GLF_LRU, &gl->gl_flags);
1986 if (!spin_trylock(&gl->gl_lockref.lock)) {
1988 list_add(&gl->gl_lru, &lru_list);
1989 set_bit(GLF_LRU, &gl->gl_flags);
1990 atomic_inc(&lru_count);
1993 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1994 spin_unlock(&gl->gl_lockref.lock);
1995 goto add_back_to_lru;
1997 gl->gl_lockref.count++;
1999 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2000 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
2001 __gfs2_glock_queue_work(gl, 0);
2002 spin_unlock(&gl->gl_lockref.lock);
2003 cond_resched_lock(&lru_lock);
2008 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
2009 * @nr: The number of entries to scan
2011 * This function selects the entries on the LRU which are able to
2012 * be demoted, and then kicks off the process by calling
2013 * gfs2_dispose_glock_lru() above.
2016 static long gfs2_scan_glock_lru(int nr)
2018 struct gfs2_glock *gl;
2023 spin_lock(&lru_lock);
2024 while ((nr-- >= 0) && !list_empty(&lru_list)) {
2025 gl = list_first_entry(&lru_list, struct gfs2_glock, gl_lru);
2027 /* Test for being demotable */
2028 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
2029 list_move(&gl->gl_lru, &dispose);
2030 atomic_dec(&lru_count);
2035 list_move(&gl->gl_lru, &skipped);
2037 list_splice(&skipped, &lru_list);
2038 if (!list_empty(&dispose))
2039 gfs2_dispose_glock_lru(&dispose);
2040 spin_unlock(&lru_lock);
2045 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
2046 struct shrink_control *sc)
2048 if (!(sc->gfp_mask & __GFP_FS))
2050 return gfs2_scan_glock_lru(sc->nr_to_scan);
2053 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
2054 struct shrink_control *sc)
2056 return vfs_pressure_ratio(atomic_read(&lru_count));
2059 static struct shrinker glock_shrinker = {
2060 .seeks = DEFAULT_SEEKS,
2061 .count_objects = gfs2_glock_shrink_count,
2062 .scan_objects = gfs2_glock_shrink_scan,
2066 * glock_hash_walk - Call a function for glock in a hash bucket
2067 * @examiner: the function
2068 * @sdp: the filesystem
2070 * Note that the function can be called multiple times on the same
2071 * object. So the user must ensure that the function can cope with
2075 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
2077 struct gfs2_glock *gl;
2078 struct rhashtable_iter iter;
2080 rhashtable_walk_enter(&gl_hash_table, &iter);
2083 rhashtable_walk_start(&iter);
2085 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl)) {
2086 if (gl->gl_name.ln_sbd == sdp)
2090 rhashtable_walk_stop(&iter);
2091 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
2093 rhashtable_walk_exit(&iter);
2096 bool gfs2_queue_delete_work(struct gfs2_glock *gl, unsigned long delay)
2100 spin_lock(&gl->gl_lockref.lock);
2101 queued = queue_delayed_work(gfs2_delete_workqueue,
2102 &gl->gl_delete, delay);
2104 set_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2105 spin_unlock(&gl->gl_lockref.lock);
2109 void gfs2_cancel_delete_work(struct gfs2_glock *gl)
2111 if (cancel_delayed_work(&gl->gl_delete)) {
2112 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2117 bool gfs2_delete_work_queued(const struct gfs2_glock *gl)
2119 return test_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2122 static void flush_delete_work(struct gfs2_glock *gl)
2124 if (gl->gl_name.ln_type == LM_TYPE_IOPEN) {
2125 if (cancel_delayed_work(&gl->gl_delete)) {
2126 queue_delayed_work(gfs2_delete_workqueue,
2132 void gfs2_flush_delete_work(struct gfs2_sbd *sdp)
2134 glock_hash_walk(flush_delete_work, sdp);
2135 flush_workqueue(gfs2_delete_workqueue);
2139 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
2140 * @gl: The glock to thaw
2144 static void thaw_glock(struct gfs2_glock *gl)
2146 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
2148 if (!lockref_get_not_dead(&gl->gl_lockref))
2150 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
2151 gfs2_glock_queue_work(gl, 0);
2155 * clear_glock - look at a glock and see if we can free it from glock cache
2156 * @gl: the glock to look at
2160 static void clear_glock(struct gfs2_glock *gl)
2162 gfs2_glock_remove_from_lru(gl);
2164 spin_lock(&gl->gl_lockref.lock);
2165 if (!__lockref_is_dead(&gl->gl_lockref)) {
2166 gl->gl_lockref.count++;
2167 if (gl->gl_state != LM_ST_UNLOCKED)
2168 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2169 __gfs2_glock_queue_work(gl, 0);
2171 spin_unlock(&gl->gl_lockref.lock);
2175 * gfs2_glock_thaw - Thaw any frozen glocks
2176 * @sdp: The super block
2180 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
2182 glock_hash_walk(thaw_glock, sdp);
2185 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2187 spin_lock(&gl->gl_lockref.lock);
2188 gfs2_dump_glock(seq, gl, fsid);
2189 spin_unlock(&gl->gl_lockref.lock);
2192 static void dump_glock_func(struct gfs2_glock *gl)
2194 dump_glock(NULL, gl, true);
2198 * gfs2_gl_hash_clear - Empty out the glock hash table
2199 * @sdp: the filesystem
2201 * Called when unmounting the filesystem.
2204 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
2206 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
2207 flush_workqueue(glock_workqueue);
2208 glock_hash_walk(clear_glock, sdp);
2209 flush_workqueue(glock_workqueue);
2210 wait_event_timeout(sdp->sd_glock_wait,
2211 atomic_read(&sdp->sd_glock_disposal) == 0,
2213 glock_hash_walk(dump_glock_func, sdp);
2216 static const char *state2str(unsigned state)
2219 case LM_ST_UNLOCKED:
2223 case LM_ST_DEFERRED:
2225 case LM_ST_EXCLUSIVE:
2231 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
2234 if (flags & LM_FLAG_TRY)
2236 if (flags & LM_FLAG_TRY_1CB)
2238 if (flags & LM_FLAG_NOEXP)
2240 if (flags & LM_FLAG_ANY)
2242 if (flags & LM_FLAG_PRIORITY)
2244 if (flags & LM_FLAG_NODE_SCOPE)
2246 if (flags & GL_ASYNC)
2248 if (flags & GL_EXACT)
2250 if (flags & GL_NOCACHE)
2252 if (test_bit(HIF_HOLDER, &iflags))
2254 if (test_bit(HIF_WAIT, &iflags))
2256 if (test_bit(HIF_MAY_DEMOTE, &iflags))
2258 if (flags & GL_SKIP)
2265 * dump_holder - print information about a glock holder
2266 * @seq: the seq_file struct
2267 * @gh: the glock holder
2268 * @fs_id_buf: pointer to file system id (if requested)
2272 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
2273 const char *fs_id_buf)
2275 struct task_struct *gh_owner = NULL;
2279 if (gh->gh_owner_pid)
2280 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
2281 gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2282 fs_id_buf, state2str(gh->gh_state),
2283 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
2285 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
2286 gh_owner ? gh_owner->comm : "(ended)",
2291 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
2293 const unsigned long *gflags = &gl->gl_flags;
2296 if (test_bit(GLF_LOCK, gflags))
2298 if (test_bit(GLF_DEMOTE, gflags))
2300 if (test_bit(GLF_PENDING_DEMOTE, gflags))
2302 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
2304 if (test_bit(GLF_DIRTY, gflags))
2306 if (test_bit(GLF_LFLUSH, gflags))
2308 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
2310 if (test_bit(GLF_REPLY_PENDING, gflags))
2312 if (test_bit(GLF_INITIAL, gflags))
2314 if (test_bit(GLF_FROZEN, gflags))
2316 if (!list_empty(&gl->gl_holders))
2318 if (test_bit(GLF_LRU, gflags))
2322 if (test_bit(GLF_BLOCKING, gflags))
2324 if (test_bit(GLF_PENDING_DELETE, gflags))
2326 if (test_bit(GLF_FREEING, gflags))
2328 if (test_bit(GLF_INSTANTIATE_NEEDED, gflags))
2330 if (test_bit(GLF_INSTANTIATE_IN_PROG, gflags))
2337 * gfs2_dump_glock - print information about a glock
2338 * @seq: The seq_file struct
2340 * @fsid: If true, also dump the file system id
2342 * The file format is as follows:
2343 * One line per object, capital letters are used to indicate objects
2344 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2345 * other objects are indented by a single space and follow the glock to
2346 * which they are related. Fields are indicated by lower case letters
2347 * followed by a colon and the field value, except for strings which are in
2348 * [] so that its possible to see if they are composed of spaces for
2349 * example. The field's are n = number (id of the object), f = flags,
2350 * t = type, s = state, r = refcount, e = error, p = pid.
2354 void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2356 const struct gfs2_glock_operations *glops = gl->gl_ops;
2357 unsigned long long dtime;
2358 const struct gfs2_holder *gh;
2359 char gflags_buf[32];
2360 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2361 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2362 unsigned long nrpages = 0;
2364 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
2365 struct address_space *mapping = gfs2_glock2aspace(gl);
2367 nrpages = mapping->nrpages;
2369 memset(fs_id_buf, 0, sizeof(fs_id_buf));
2370 if (fsid && sdp) /* safety precaution */
2371 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2372 dtime = jiffies - gl->gl_demote_time;
2373 dtime *= 1000000/HZ; /* demote time in uSec */
2374 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
2376 gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2377 "v:%d r:%d m:%ld p:%lu\n",
2378 fs_id_buf, state2str(gl->gl_state),
2379 gl->gl_name.ln_type,
2380 (unsigned long long)gl->gl_name.ln_number,
2381 gflags2str(gflags_buf, gl),
2382 state2str(gl->gl_target),
2383 state2str(gl->gl_demote_state), dtime,
2384 atomic_read(&gl->gl_ail_count),
2385 atomic_read(&gl->gl_revokes),
2386 (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages);
2388 list_for_each_entry(gh, &gl->gl_holders, gh_list)
2389 dump_holder(seq, gh, fs_id_buf);
2391 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
2392 glops->go_dump(seq, gl, fs_id_buf);
2395 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
2397 struct gfs2_glock *gl = iter_ptr;
2399 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2400 gl->gl_name.ln_type,
2401 (unsigned long long)gl->gl_name.ln_number,
2402 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
2403 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
2404 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
2405 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
2406 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
2407 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
2408 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
2409 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
2413 static const char *gfs2_gltype[] = {
2427 static const char *gfs2_stype[] = {
2428 [GFS2_LKS_SRTT] = "srtt",
2429 [GFS2_LKS_SRTTVAR] = "srttvar",
2430 [GFS2_LKS_SRTTB] = "srttb",
2431 [GFS2_LKS_SRTTVARB] = "srttvarb",
2432 [GFS2_LKS_SIRT] = "sirt",
2433 [GFS2_LKS_SIRTVAR] = "sirtvar",
2434 [GFS2_LKS_DCOUNT] = "dlm",
2435 [GFS2_LKS_QCOUNT] = "queue",
2438 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2440 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
2442 struct gfs2_sbd *sdp = seq->private;
2443 loff_t pos = *(loff_t *)iter_ptr;
2444 unsigned index = pos >> 3;
2445 unsigned subindex = pos & 0x07;
2448 if (index == 0 && subindex != 0)
2451 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
2452 (index == 0) ? "cpu": gfs2_stype[subindex]);
2454 for_each_possible_cpu(i) {
2455 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
2458 seq_printf(seq, " %15u", i);
2460 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
2461 lkstats[index - 1].stats[subindex]);
2463 seq_putc(seq, '\n');
2467 int __init gfs2_glock_init(void)
2471 ret = rhashtable_init(&gl_hash_table, &ht_parms);
2475 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
2476 WQ_HIGHPRI | WQ_FREEZABLE, 0);
2477 if (!glock_workqueue) {
2478 rhashtable_destroy(&gl_hash_table);
2481 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
2482 WQ_MEM_RECLAIM | WQ_FREEZABLE,
2484 if (!gfs2_delete_workqueue) {
2485 destroy_workqueue(glock_workqueue);
2486 rhashtable_destroy(&gl_hash_table);
2490 ret = register_shrinker(&glock_shrinker, "gfs2-glock");
2492 destroy_workqueue(gfs2_delete_workqueue);
2493 destroy_workqueue(glock_workqueue);
2494 rhashtable_destroy(&gl_hash_table);
2498 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2499 init_waitqueue_head(glock_wait_table + i);
2504 void gfs2_glock_exit(void)
2506 unregister_shrinker(&glock_shrinker);
2507 rhashtable_destroy(&gl_hash_table);
2508 destroy_workqueue(glock_workqueue);
2509 destroy_workqueue(gfs2_delete_workqueue);
2512 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2514 struct gfs2_glock *gl = gi->gl;
2519 if (!lockref_put_not_zero(&gl->gl_lockref))
2520 gfs2_glock_queue_put(gl);
2523 gl = rhashtable_walk_next(&gi->hti);
2524 if (IS_ERR_OR_NULL(gl)) {
2525 if (gl == ERR_PTR(-EAGAIN)) {
2532 if (gl->gl_name.ln_sbd != gi->sdp)
2535 if (!lockref_get_not_dead(&gl->gl_lockref))
2539 if (__lockref_is_dead(&gl->gl_lockref))
2547 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2550 struct gfs2_glock_iter *gi = seq->private;
2554 * We can either stay where we are, skip to the next hash table
2555 * entry, or start from the beginning.
2557 if (*pos < gi->last_pos) {
2558 rhashtable_walk_exit(&gi->hti);
2559 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2562 n = *pos - gi->last_pos;
2565 rhashtable_walk_start(&gi->hti);
2567 gfs2_glock_iter_next(gi, n);
2568 gi->last_pos = *pos;
2572 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2575 struct gfs2_glock_iter *gi = seq->private;
2578 gi->last_pos = *pos;
2579 gfs2_glock_iter_next(gi, 1);
2583 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2586 struct gfs2_glock_iter *gi = seq->private;
2588 rhashtable_walk_stop(&gi->hti);
2591 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2593 dump_glock(seq, iter_ptr, false);
2597 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2600 if (*pos >= GFS2_NR_SBSTATS)
2605 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2609 if (*pos >= GFS2_NR_SBSTATS)
2614 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2619 static const struct seq_operations gfs2_glock_seq_ops = {
2620 .start = gfs2_glock_seq_start,
2621 .next = gfs2_glock_seq_next,
2622 .stop = gfs2_glock_seq_stop,
2623 .show = gfs2_glock_seq_show,
2626 static const struct seq_operations gfs2_glstats_seq_ops = {
2627 .start = gfs2_glock_seq_start,
2628 .next = gfs2_glock_seq_next,
2629 .stop = gfs2_glock_seq_stop,
2630 .show = gfs2_glstats_seq_show,
2633 static const struct seq_operations gfs2_sbstats_sops = {
2634 .start = gfs2_sbstats_seq_start,
2635 .next = gfs2_sbstats_seq_next,
2636 .stop = gfs2_sbstats_seq_stop,
2637 .show = gfs2_sbstats_seq_show,
2640 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2642 static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2643 const struct seq_operations *ops)
2645 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2647 struct seq_file *seq = file->private_data;
2648 struct gfs2_glock_iter *gi = seq->private;
2650 gi->sdp = inode->i_private;
2651 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2653 seq->size = GFS2_SEQ_GOODSIZE;
2655 * Initially, we are "before" the first hash table entry; the
2656 * first call to rhashtable_walk_next gets us the first entry.
2660 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2665 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2667 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2670 static int gfs2_glocks_release(struct inode *inode, struct file *file)
2672 struct seq_file *seq = file->private_data;
2673 struct gfs2_glock_iter *gi = seq->private;
2676 gfs2_glock_put(gi->gl);
2677 rhashtable_walk_exit(&gi->hti);
2678 return seq_release_private(inode, file);
2681 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2683 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2686 static const struct file_operations gfs2_glocks_fops = {
2687 .owner = THIS_MODULE,
2688 .open = gfs2_glocks_open,
2690 .llseek = seq_lseek,
2691 .release = gfs2_glocks_release,
2694 static const struct file_operations gfs2_glstats_fops = {
2695 .owner = THIS_MODULE,
2696 .open = gfs2_glstats_open,
2698 .llseek = seq_lseek,
2699 .release = gfs2_glocks_release,
2702 DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats);
2704 void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2706 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2708 debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2711 debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2712 &gfs2_glstats_fops);
2714 debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2715 &gfs2_sbstats_fops);
2718 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2720 debugfs_remove_recursive(sdp->debugfs_dir);
2721 sdp->debugfs_dir = NULL;
2724 void gfs2_register_debugfs(void)
2726 gfs2_root = debugfs_create_dir("gfs2", NULL);
2729 void gfs2_unregister_debugfs(void)
2731 debugfs_remove(gfs2_root);