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 * @new_gh: the new holder to be promoted
410 static void demote_incompat_holders(struct gfs2_glock *gl,
411 struct gfs2_holder *new_gh)
413 struct gfs2_holder *gh, *tmp;
416 * Demote incompatible holders before we make ourselves eligible.
417 * (This holder may or may not allow auto-demoting, but we don't want
418 * to demote the new holder before it's even granted.)
420 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
422 * Since holders are at the front of the list, we stop when we
423 * find the first non-holder.
425 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
427 if (test_bit(HIF_MAY_DEMOTE, &gh->gh_iflags) &&
428 !may_grant(gl, new_gh, gh)) {
430 * We should not recurse into do_promote because
431 * __gfs2_glock_dq only calls handle_callback,
432 * gfs2_glock_add_to_lru and __gfs2_glock_queue_work.
440 * find_first_holder - find the first "holder" gh
444 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
446 struct gfs2_holder *gh;
448 if (!list_empty(&gl->gl_holders)) {
449 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder,
451 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
458 * find_first_strong_holder - find the first non-demoteable holder
461 * Find the first holder that doesn't have the HIF_MAY_DEMOTE flag set.
463 static inline struct gfs2_holder *
464 find_first_strong_holder(struct gfs2_glock *gl)
466 struct gfs2_holder *gh;
468 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
469 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
471 if (!test_bit(HIF_MAY_DEMOTE, &gh->gh_iflags))
478 * gfs2_instantiate - Call the glops instantiate function
479 * @gh: The glock holder
481 * Returns: 0 if instantiate was successful, or error.
483 int gfs2_instantiate(struct gfs2_holder *gh)
485 struct gfs2_glock *gl = gh->gh_gl;
486 const struct gfs2_glock_operations *glops = gl->gl_ops;
490 if (!test_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags))
494 * Since we unlock the lockref lock, we set a flag to indicate
495 * instantiate is in progress.
497 if (test_and_set_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags)) {
498 wait_on_bit(&gl->gl_flags, GLF_INSTANTIATE_IN_PROG,
499 TASK_UNINTERRUPTIBLE);
501 * Here we just waited for a different instantiate to finish.
502 * But that may not have been successful, as when a process
503 * locks an inode glock _before_ it has an actual inode to
504 * instantiate into. So we check again. This process might
505 * have an inode to instantiate, so might be successful.
510 ret = glops->go_instantiate(gl);
512 clear_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags);
513 clear_and_wake_up_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags);
519 return glops->go_held(gh);
524 * do_promote - promote as many requests as possible on the current queue
527 * Returns: 1 if there is a blocked holder at the head of the list
530 static int do_promote(struct gfs2_glock *gl)
532 struct gfs2_holder *gh, *tmp, *first_gh;
533 bool incompat_holders_demoted = false;
535 first_gh = find_first_strong_holder(gl);
536 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
537 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
539 if (!may_grant(gl, first_gh, gh)) {
541 * If we get here, it means we may not grant this holder for
542 * some reason. If this holder is the head of the list, it
543 * means we have a blocked holder at the head, so return 1.
545 if (list_is_first(&gh->gh_list, &gl->gl_holders))
550 if (!incompat_holders_demoted) {
551 demote_incompat_holders(gl, first_gh);
552 incompat_holders_demoted = true;
555 set_bit(HIF_HOLDER, &gh->gh_iflags);
556 trace_gfs2_promote(gh);
557 gfs2_holder_wake(gh);
563 * find_first_waiter - find the first gh that's waiting for the glock
567 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
569 struct gfs2_holder *gh;
571 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
572 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
579 * state_change - record that the glock is now in a different state
581 * @new_state: the new state
584 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
588 held1 = (gl->gl_state != LM_ST_UNLOCKED);
589 held2 = (new_state != LM_ST_UNLOCKED);
591 if (held1 != held2) {
592 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
594 gl->gl_lockref.count++;
596 gl->gl_lockref.count--;
598 if (new_state != gl->gl_target)
599 /* shorten our minimum hold time */
600 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
602 gl->gl_state = new_state;
603 gl->gl_tchange = jiffies;
606 static void gfs2_set_demote(struct gfs2_glock *gl)
608 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
610 set_bit(GLF_DEMOTE, &gl->gl_flags);
612 wake_up(&sdp->sd_async_glock_wait);
615 static void gfs2_demote_wake(struct gfs2_glock *gl)
617 gl->gl_demote_state = LM_ST_EXCLUSIVE;
618 clear_bit(GLF_DEMOTE, &gl->gl_flags);
619 smp_mb__after_atomic();
620 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
624 * finish_xmote - The DLM has replied to one of our lock requests
626 * @ret: The status from the DLM
630 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
632 const struct gfs2_glock_operations *glops = gl->gl_ops;
633 struct gfs2_holder *gh;
634 unsigned state = ret & LM_OUT_ST_MASK;
636 spin_lock(&gl->gl_lockref.lock);
637 trace_gfs2_glock_state_change(gl, state);
638 state_change(gl, state);
639 gh = find_first_waiter(gl);
641 /* Demote to UN request arrived during demote to SH or DF */
642 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
643 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
644 gl->gl_target = LM_ST_UNLOCKED;
646 /* Check for state != intended state */
647 if (unlikely(state != gl->gl_target)) {
648 if (gh && (ret & LM_OUT_CANCELED))
649 gfs2_holder_wake(gh);
650 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
651 /* move to back of queue and try next entry */
652 if (ret & LM_OUT_CANCELED) {
653 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
654 list_move_tail(&gh->gh_list, &gl->gl_holders);
655 gh = find_first_waiter(gl);
656 gl->gl_target = gh->gh_state;
659 /* Some error or failed "try lock" - report it */
660 if ((ret & LM_OUT_ERROR) ||
661 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
662 gl->gl_target = gl->gl_state;
668 /* Unlocked due to conversion deadlock, try again */
671 do_xmote(gl, gh, gl->gl_target);
673 /* Conversion fails, unlock and try again */
676 do_xmote(gl, gh, LM_ST_UNLOCKED);
678 default: /* Everything else */
679 fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n",
680 gl->gl_target, state);
683 spin_unlock(&gl->gl_lockref.lock);
687 /* Fast path - we got what we asked for */
688 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
689 gfs2_demote_wake(gl);
690 if (state != LM_ST_UNLOCKED) {
691 if (glops->go_xmote_bh) {
694 spin_unlock(&gl->gl_lockref.lock);
695 rv = glops->go_xmote_bh(gl);
696 spin_lock(&gl->gl_lockref.lock);
705 clear_bit(GLF_LOCK, &gl->gl_flags);
706 spin_unlock(&gl->gl_lockref.lock);
709 static bool is_system_glock(struct gfs2_glock *gl)
711 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
712 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
714 if (gl == m_ip->i_gl)
720 * do_xmote - Calls the DLM to change the state of a lock
721 * @gl: The lock state
722 * @gh: The holder (only for promotes)
723 * @target: The target lock state
727 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
728 __releases(&gl->gl_lockref.lock)
729 __acquires(&gl->gl_lockref.lock)
731 const struct gfs2_glock_operations *glops = gl->gl_ops;
732 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
733 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
736 if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) &&
737 gh && !(gh->gh_flags & LM_FLAG_NOEXP))
739 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
741 GLOCK_BUG_ON(gl, gl->gl_state == target);
742 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
743 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
746 * If another process is already doing the invalidate, let that
747 * finish first. The glock state machine will get back to this
748 * holder again later.
750 if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
753 do_error(gl, 0); /* Fail queued try locks */
756 set_bit(GLF_BLOCKING, &gl->gl_flags);
757 if ((gl->gl_req == LM_ST_UNLOCKED) ||
758 (gl->gl_state == LM_ST_EXCLUSIVE) ||
759 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
760 clear_bit(GLF_BLOCKING, &gl->gl_flags);
761 spin_unlock(&gl->gl_lockref.lock);
762 if (glops->go_sync) {
763 ret = glops->go_sync(gl);
764 /* If we had a problem syncing (due to io errors or whatever,
765 * we should not invalidate the metadata or tell dlm to
766 * release the glock to other nodes.
769 if (cmpxchg(&sdp->sd_log_error, 0, ret)) {
770 fs_err(sdp, "Error %d syncing glock \n", ret);
771 gfs2_dump_glock(NULL, gl, true);
776 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) {
778 * The call to go_sync should have cleared out the ail list.
779 * If there are still items, we have a problem. We ought to
780 * withdraw, but we can't because the withdraw code also uses
781 * glocks. Warn about the error, dump the glock, then fall
782 * through and wait for logd to do the withdraw for us.
784 if ((atomic_read(&gl->gl_ail_count) != 0) &&
785 (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) {
786 gfs2_glock_assert_warn(gl,
787 !atomic_read(&gl->gl_ail_count));
788 gfs2_dump_glock(NULL, gl, true);
790 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
791 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
797 * Check for an error encountered since we called go_sync and go_inval.
798 * If so, we can't withdraw from the glock code because the withdraw
799 * code itself uses glocks (see function signal_our_withdraw) to
800 * change the mount to read-only. Most importantly, we must not call
801 * dlm to unlock the glock until the journal is in a known good state
802 * (after journal replay) otherwise other nodes may use the object
803 * (rgrp or dinode) and then later, journal replay will corrupt the
804 * file system. The best we can do here is wait for the logd daemon
805 * to see sd_log_error and withdraw, and in the meantime, requeue the
808 * We make a special exception for some system glocks, such as the
809 * system statfs inode glock, which needs to be granted before the
810 * gfs2_quotad daemon can exit, and that exit needs to finish before
811 * we can unmount the withdrawn file system.
813 * However, if we're just unlocking the lock (say, for unmount, when
814 * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
815 * then it's okay to tell dlm to unlock it.
817 if (unlikely(sdp->sd_log_error && !gfs2_withdrawn(sdp)))
818 gfs2_withdraw_delayed(sdp);
819 if (glock_blocked_by_withdraw(gl) &&
820 (target != LM_ST_UNLOCKED ||
821 test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags))) {
822 if (!is_system_glock(gl)) {
823 gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD);
826 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
830 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
832 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
833 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
834 target == LM_ST_UNLOCKED &&
835 test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
836 finish_xmote(gl, target);
837 gfs2_glock_queue_work(gl, 0);
839 fs_err(sdp, "lm_lock ret %d\n", ret);
840 GLOCK_BUG_ON(gl, !gfs2_withdrawn(sdp));
842 } else { /* lock_nolock */
843 finish_xmote(gl, target);
844 gfs2_glock_queue_work(gl, 0);
847 spin_lock(&gl->gl_lockref.lock);
851 * run_queue - do all outstanding tasks related to a glock
852 * @gl: The glock in question
853 * @nonblock: True if we must not block in run_queue
857 static void run_queue(struct gfs2_glock *gl, const int nonblock)
858 __releases(&gl->gl_lockref.lock)
859 __acquires(&gl->gl_lockref.lock)
861 struct gfs2_holder *gh = NULL;
863 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
866 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
868 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
869 gl->gl_demote_state != gl->gl_state) {
870 if (find_first_holder(gl))
874 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
875 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
876 gl->gl_target = gl->gl_demote_state;
878 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
879 gfs2_demote_wake(gl);
880 if (do_promote(gl) == 0)
882 gh = find_first_waiter(gl);
883 gl->gl_target = gh->gh_state;
884 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
885 do_error(gl, 0); /* Fail queued try locks */
887 do_xmote(gl, gh, gl->gl_target);
891 clear_bit(GLF_LOCK, &gl->gl_flags);
892 smp_mb__after_atomic();
893 gl->gl_lockref.count++;
894 __gfs2_glock_queue_work(gl, 0);
898 clear_bit(GLF_LOCK, &gl->gl_flags);
899 smp_mb__after_atomic();
903 void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation)
905 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
907 if (ri->ri_magic == 0)
908 ri->ri_magic = cpu_to_be32(GFS2_MAGIC);
909 if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC))
910 ri->ri_generation_deleted = cpu_to_be64(generation);
913 bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation)
915 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
917 if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC))
919 return generation <= be64_to_cpu(ri->ri_generation_deleted);
922 static void gfs2_glock_poke(struct gfs2_glock *gl)
924 int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP;
925 struct gfs2_holder gh;
928 __gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh, _RET_IP_);
929 error = gfs2_glock_nq(&gh);
932 gfs2_holder_uninit(&gh);
935 static bool gfs2_try_evict(struct gfs2_glock *gl)
937 struct gfs2_inode *ip;
938 bool evicted = false;
941 * If there is contention on the iopen glock and we have an inode, try
942 * to grab and release the inode so that it can be evicted. This will
943 * allow the remote node to go ahead and delete the inode without us
944 * having to do it, which will avoid rgrp glock thrashing.
946 * The remote node is likely still holding the corresponding inode
947 * glock, so it will run before we get to verify that the delete has
950 spin_lock(&gl->gl_lockref.lock);
952 if (ip && !igrab(&ip->i_inode))
954 spin_unlock(&gl->gl_lockref.lock);
956 struct gfs2_glock *inode_gl = NULL;
958 gl->gl_no_formal_ino = ip->i_no_formal_ino;
959 set_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
960 d_prune_aliases(&ip->i_inode);
963 /* If the inode was evicted, gl->gl_object will now be NULL. */
964 spin_lock(&gl->gl_lockref.lock);
968 lockref_get(&inode_gl->gl_lockref);
969 clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
971 spin_unlock(&gl->gl_lockref.lock);
973 gfs2_glock_poke(inode_gl);
974 gfs2_glock_put(inode_gl);
981 static void delete_work_func(struct work_struct *work)
983 struct delayed_work *dwork = to_delayed_work(work);
984 struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete);
985 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
987 u64 no_addr = gl->gl_name.ln_number;
989 spin_lock(&gl->gl_lockref.lock);
990 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
991 spin_unlock(&gl->gl_lockref.lock);
993 if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
995 * If we can evict the inode, give the remote node trying to
996 * delete the inode some time before verifying that the delete
997 * has happened. Otherwise, if we cause contention on the inode glock
998 * immediately, the remote node will think that we still have
999 * the inode in use, and so it will give up waiting.
1001 * If we can't evict the inode, signal to the remote node that
1002 * the inode is still in use. We'll later try to delete the
1003 * inode locally in gfs2_evict_inode.
1005 * FIXME: We only need to verify that the remote node has
1006 * deleted the inode because nodes before this remote delete
1007 * rework won't cooperate. At a later time, when we no longer
1008 * care about compatibility with such nodes, we can skip this
1011 if (gfs2_try_evict(gl)) {
1012 if (gfs2_queue_delete_work(gl, 5 * HZ))
1018 inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino,
1019 GFS2_BLKST_UNLINKED);
1020 if (!IS_ERR_OR_NULL(inode)) {
1021 d_prune_aliases(inode);
1028 static void glock_work_func(struct work_struct *work)
1030 unsigned long delay = 0;
1031 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
1032 unsigned int drop_refs = 1;
1034 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
1035 finish_xmote(gl, gl->gl_reply);
1038 spin_lock(&gl->gl_lockref.lock);
1039 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1040 gl->gl_state != LM_ST_UNLOCKED &&
1041 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
1042 unsigned long holdtime, now = jiffies;
1044 holdtime = gl->gl_tchange + gl->gl_hold_time;
1045 if (time_before(now, holdtime))
1046 delay = holdtime - now;
1049 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1050 gfs2_set_demote(gl);
1055 /* Keep one glock reference for the work we requeue. */
1057 if (gl->gl_name.ln_type != LM_TYPE_INODE)
1059 __gfs2_glock_queue_work(gl, delay);
1063 * Drop the remaining glock references manually here. (Mind that
1064 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
1067 gl->gl_lockref.count -= drop_refs;
1068 if (!gl->gl_lockref.count) {
1069 __gfs2_glock_put(gl);
1072 spin_unlock(&gl->gl_lockref.lock);
1075 static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
1076 struct gfs2_glock *new)
1078 struct wait_glock_queue wait;
1079 wait_queue_head_t *wq = glock_waitqueue(name);
1080 struct gfs2_glock *gl;
1083 init_wait(&wait.wait);
1084 wait.wait.func = glock_wake_function;
1087 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1090 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
1091 &new->gl_node, ht_parms);
1095 gl = rhashtable_lookup_fast(&gl_hash_table,
1098 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
1105 finish_wait(wq, &wait.wait);
1110 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
1111 * @sdp: The GFS2 superblock
1112 * @number: the lock number
1113 * @glops: The glock_operations to use
1114 * @create: If 0, don't create the glock if it doesn't exist
1115 * @glp: the glock is returned here
1117 * This does not lock a glock, just finds/creates structures for one.
1122 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
1123 const struct gfs2_glock_operations *glops, int create,
1124 struct gfs2_glock **glp)
1126 struct super_block *s = sdp->sd_vfs;
1127 struct lm_lockname name = { .ln_number = number,
1128 .ln_type = glops->go_type,
1130 struct gfs2_glock *gl, *tmp;
1131 struct address_space *mapping;
1134 gl = find_insert_glock(&name, NULL);
1142 if (glops->go_flags & GLOF_ASPACE) {
1143 struct gfs2_glock_aspace *gla =
1144 kmem_cache_alloc(gfs2_glock_aspace_cachep, GFP_NOFS);
1149 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_NOFS);
1153 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
1156 if (glops->go_flags & GLOF_LVB) {
1157 gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1158 if (!gl->gl_lksb.sb_lvbptr) {
1159 gfs2_glock_dealloc(&gl->gl_rcu);
1164 atomic_inc(&sdp->sd_glock_disposal);
1165 gl->gl_node.next = NULL;
1166 gl->gl_flags = glops->go_instantiate ? BIT(GLF_INSTANTIATE_NEEDED) : 0;
1168 lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass);
1169 gl->gl_lockref.count = 1;
1170 gl->gl_state = LM_ST_UNLOCKED;
1171 gl->gl_target = LM_ST_UNLOCKED;
1172 gl->gl_demote_state = LM_ST_EXCLUSIVE;
1175 /* We use the global stats to estimate the initial per-glock stats */
1176 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
1178 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
1179 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
1180 gl->gl_tchange = jiffies;
1181 gl->gl_object = NULL;
1182 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
1183 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
1184 if (gl->gl_name.ln_type == LM_TYPE_IOPEN)
1185 INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func);
1187 mapping = gfs2_glock2aspace(gl);
1189 mapping->a_ops = &gfs2_meta_aops;
1190 mapping->host = s->s_bdev->bd_inode;
1192 mapping_set_gfp_mask(mapping, GFP_NOFS);
1193 mapping->private_data = NULL;
1194 mapping->writeback_index = 0;
1197 tmp = find_insert_glock(&name, gl);
1209 gfs2_glock_dealloc(&gl->gl_rcu);
1210 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
1211 wake_up(&sdp->sd_glock_wait);
1218 * __gfs2_holder_init - initialize a struct gfs2_holder in the default way
1220 * @state: the state we're requesting
1221 * @flags: the modifier flags
1222 * @gh: the holder structure
1226 void __gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
1227 struct gfs2_holder *gh, unsigned long ip)
1229 INIT_LIST_HEAD(&gh->gh_list);
1232 gh->gh_owner_pid = get_pid(task_pid(current));
1233 gh->gh_state = state;
1234 gh->gh_flags = flags;
1236 gfs2_glock_hold(gl);
1240 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1241 * @state: the state we're requesting
1242 * @flags: the modifier flags
1243 * @gh: the holder structure
1245 * Don't mess with the glock.
1249 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
1251 gh->gh_state = state;
1252 gh->gh_flags = flags;
1254 gh->gh_ip = _RET_IP_;
1255 put_pid(gh->gh_owner_pid);
1256 gh->gh_owner_pid = get_pid(task_pid(current));
1260 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1261 * @gh: the holder structure
1265 void gfs2_holder_uninit(struct gfs2_holder *gh)
1267 put_pid(gh->gh_owner_pid);
1268 gfs2_glock_put(gh->gh_gl);
1269 gfs2_holder_mark_uninitialized(gh);
1273 static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
1274 unsigned long start_time)
1276 /* Have we waited longer that a second? */
1277 if (time_after(jiffies, start_time + HZ)) {
1278 /* Lengthen the minimum hold time. */
1279 gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
1285 * gfs2_glock_holder_ready - holder is ready and its error code can be collected
1286 * @gh: the glock holder
1288 * Called when a glock holder no longer needs to be waited for because it is
1289 * now either held (HIF_HOLDER set; gh_error == 0), or acquiring the lock has
1290 * failed (gh_error != 0).
1293 int gfs2_glock_holder_ready(struct gfs2_holder *gh)
1295 if (gh->gh_error || (gh->gh_flags & GL_SKIP))
1296 return gh->gh_error;
1297 gh->gh_error = gfs2_instantiate(gh);
1300 return gh->gh_error;
1304 * gfs2_glock_wait - wait on a glock acquisition
1305 * @gh: the glock holder
1307 * Returns: 0 on success
1310 int gfs2_glock_wait(struct gfs2_holder *gh)
1312 unsigned long start_time = jiffies;
1315 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1316 gfs2_glock_update_hold_time(gh->gh_gl, start_time);
1317 return gfs2_glock_holder_ready(gh);
1320 static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
1324 for (i = 0; i < num_gh; i++)
1325 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
1331 * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1332 * @num_gh: the number of holders in the array
1333 * @ghs: the glock holder array
1335 * Returns: 0 on success, meaning all glocks have been granted and are held.
1336 * -ESTALE if the request timed out, meaning all glocks were released,
1337 * and the caller should retry the operation.
1340 int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
1342 struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
1343 int i, ret = 0, timeout = 0;
1344 unsigned long start_time = jiffies;
1348 * Total up the (minimum hold time * 2) of all glocks and use that to
1349 * determine the max amount of time we should wait.
1351 for (i = 0; i < num_gh; i++)
1352 timeout += ghs[i].gh_gl->gl_hold_time << 1;
1354 if (!wait_event_timeout(sdp->sd_async_glock_wait,
1355 !glocks_pending(num_gh, ghs), timeout)) {
1356 ret = -ESTALE; /* request timed out. */
1360 for (i = 0; i < num_gh; i++) {
1361 struct gfs2_holder *gh = &ghs[i];
1364 if (test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1365 gfs2_glock_update_hold_time(gh->gh_gl,
1368 ret2 = gfs2_glock_holder_ready(gh);
1375 for (i = 0; i < num_gh; i++) {
1376 struct gfs2_holder *gh = &ghs[i];
1385 * handle_callback - process a demote request
1387 * @state: the state the caller wants us to change to
1388 * @delay: zero to demote immediately; otherwise pending demote
1389 * @remote: true if this came from a different cluster node
1391 * There are only two requests that we are going to see in actual
1392 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1395 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
1396 unsigned long delay, bool remote)
1399 set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1401 gfs2_set_demote(gl);
1402 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
1403 gl->gl_demote_state = state;
1404 gl->gl_demote_time = jiffies;
1405 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
1406 gl->gl_demote_state != state) {
1407 gl->gl_demote_state = LM_ST_UNLOCKED;
1409 if (gl->gl_ops->go_callback)
1410 gl->gl_ops->go_callback(gl, remote);
1411 trace_gfs2_demote_rq(gl, remote);
1414 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
1416 struct va_format vaf;
1419 va_start(args, fmt);
1422 seq_vprintf(seq, fmt, args);
1427 pr_err("%pV", &vaf);
1434 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1435 * @gh: the holder structure to add
1437 * Eventually we should move the recursive locking trap to a
1438 * debugging option or something like that. This is the fast
1439 * path and needs to have the minimum number of distractions.
1443 static inline void add_to_queue(struct gfs2_holder *gh)
1444 __releases(&gl->gl_lockref.lock)
1445 __acquires(&gl->gl_lockref.lock)
1447 struct gfs2_glock *gl = gh->gh_gl;
1448 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1449 struct list_head *insert_pt = NULL;
1450 struct gfs2_holder *gh2;
1453 GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1454 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1455 GLOCK_BUG_ON(gl, true);
1457 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1458 if (test_bit(GLF_LOCK, &gl->gl_flags)) {
1459 struct gfs2_holder *first_gh;
1461 first_gh = find_first_strong_holder(gl);
1462 try_futile = !may_grant(gl, first_gh, gh);
1464 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1468 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1469 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
1470 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK) &&
1471 !test_bit(HIF_MAY_DEMOTE, &gh2->gh_iflags)))
1472 goto trap_recursive;
1474 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1476 gh->gh_error = GLR_TRYFAILED;
1477 gfs2_holder_wake(gh);
1480 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1482 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1483 insert_pt = &gh2->gh_list;
1485 trace_gfs2_glock_queue(gh, 1);
1486 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1487 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1488 if (likely(insert_pt == NULL)) {
1489 list_add_tail(&gh->gh_list, &gl->gl_holders);
1490 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1494 list_add_tail(&gh->gh_list, insert_pt);
1496 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
1497 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1498 spin_unlock(&gl->gl_lockref.lock);
1499 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1500 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1501 spin_lock(&gl->gl_lockref.lock);
1506 fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1507 fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1508 fs_err(sdp, "lock type: %d req lock state : %d\n",
1509 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1510 fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1511 fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1512 fs_err(sdp, "lock type: %d req lock state : %d\n",
1513 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1514 gfs2_dump_glock(NULL, gl, true);
1519 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1520 * @gh: the holder structure
1522 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1524 * Returns: 0, GLR_TRYFAILED, or errno on failure
1527 int gfs2_glock_nq(struct gfs2_holder *gh)
1529 struct gfs2_glock *gl = gh->gh_gl;
1532 if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP))
1535 if (test_bit(GLF_LRU, &gl->gl_flags))
1536 gfs2_glock_remove_from_lru(gl);
1539 spin_lock(&gl->gl_lockref.lock);
1541 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1542 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1543 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1544 gl->gl_lockref.count++;
1545 __gfs2_glock_queue_work(gl, 0);
1548 spin_unlock(&gl->gl_lockref.lock);
1550 if (!(gh->gh_flags & GL_ASYNC))
1551 error = gfs2_glock_wait(gh);
1557 * gfs2_glock_poll - poll to see if an async request has been completed
1560 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1563 int gfs2_glock_poll(struct gfs2_holder *gh)
1565 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1568 static inline bool needs_demote(struct gfs2_glock *gl)
1570 return (test_bit(GLF_DEMOTE, &gl->gl_flags) ||
1571 test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags));
1574 static void __gfs2_glock_dq(struct gfs2_holder *gh)
1576 struct gfs2_glock *gl = gh->gh_gl;
1577 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1582 * This while loop is similar to function demote_incompat_holders:
1583 * If the glock is due to be demoted (which may be from another node
1584 * or even if this holder is GL_NOCACHE), the weak holders are
1585 * demoted as well, allowing the glock to be demoted.
1589 * If we're in the process of file system withdraw, we cannot
1590 * just dequeue any glocks until our journal is recovered, lest
1591 * we introduce file system corruption. We need two exceptions
1592 * to this rule: We need to allow unlocking of nondisk glocks
1593 * and the glock for our own journal that needs recovery.
1595 if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) &&
1596 glock_blocked_by_withdraw(gl) &&
1597 gh->gh_gl != sdp->sd_jinode_gl) {
1598 sdp->sd_glock_dqs_held++;
1599 spin_unlock(&gl->gl_lockref.lock);
1601 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
1602 TASK_UNINTERRUPTIBLE);
1603 spin_lock(&gl->gl_lockref.lock);
1607 * This holder should not be cached, so mark it for demote.
1608 * Note: this should be done before the check for needs_demote
1611 if (gh->gh_flags & GL_NOCACHE)
1612 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1614 list_del_init(&gh->gh_list);
1615 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1616 trace_gfs2_glock_queue(gh, 0);
1619 * If there hasn't been a demote request we are done.
1620 * (Let the remaining holders, if any, keep holding it.)
1622 if (!needs_demote(gl)) {
1623 if (list_empty(&gl->gl_holders))
1628 * If we have another strong holder (we cannot auto-demote)
1629 * we are done. It keeps holding it until it is done.
1631 if (find_first_strong_holder(gl))
1635 * If we have a weak holder at the head of the list, it
1636 * (and all others like it) must be auto-demoted. If there
1637 * are no more weak holders, we exit the while loop.
1639 gh = find_first_holder(gl);
1642 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1643 gfs2_glock_add_to_lru(gl);
1645 if (unlikely(!fast_path)) {
1646 gl->gl_lockref.count++;
1647 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1648 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1649 gl->gl_name.ln_type == LM_TYPE_INODE)
1650 delay = gl->gl_hold_time;
1651 __gfs2_glock_queue_work(gl, delay);
1656 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1657 * @gh: the glock holder
1660 void gfs2_glock_dq(struct gfs2_holder *gh)
1662 struct gfs2_glock *gl = gh->gh_gl;
1664 spin_lock(&gl->gl_lockref.lock);
1665 if (list_is_first(&gh->gh_list, &gl->gl_holders) &&
1666 !test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1667 spin_unlock(&gl->gl_lockref.lock);
1668 gl->gl_name.ln_sbd->sd_lockstruct.ls_ops->lm_cancel(gl);
1669 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1670 spin_lock(&gl->gl_lockref.lock);
1673 __gfs2_glock_dq(gh);
1674 spin_unlock(&gl->gl_lockref.lock);
1677 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1679 struct gfs2_glock *gl = gh->gh_gl;
1682 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1686 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1687 * @gh: the holder structure
1691 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1694 gfs2_holder_uninit(gh);
1698 * gfs2_glock_nq_num - acquire a glock based on lock number
1699 * @sdp: the filesystem
1700 * @number: the lock number
1701 * @glops: the glock operations for the type of glock
1702 * @state: the state to acquire the glock in
1703 * @flags: modifier flags for the acquisition
1704 * @gh: the struct gfs2_holder
1709 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1710 const struct gfs2_glock_operations *glops,
1711 unsigned int state, u16 flags, struct gfs2_holder *gh)
1713 struct gfs2_glock *gl;
1716 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1718 error = gfs2_glock_nq_init(gl, state, flags, gh);
1726 * glock_compare - Compare two struct gfs2_glock structures for sorting
1727 * @arg_a: the first structure
1728 * @arg_b: the second structure
1732 static int glock_compare(const void *arg_a, const void *arg_b)
1734 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1735 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1736 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1737 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1739 if (a->ln_number > b->ln_number)
1741 if (a->ln_number < b->ln_number)
1743 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1748 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1749 * @num_gh: the number of structures
1750 * @ghs: an array of struct gfs2_holder structures
1751 * @p: placeholder for the holder structure to pass back
1753 * Returns: 0 on success (all glocks acquired),
1754 * errno on failure (no glocks acquired)
1757 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1758 struct gfs2_holder **p)
1763 for (x = 0; x < num_gh; x++)
1766 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1768 for (x = 0; x < num_gh; x++) {
1769 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1771 error = gfs2_glock_nq(p[x]);
1774 gfs2_glock_dq(p[x]);
1783 * gfs2_glock_nq_m - acquire multiple glocks
1784 * @num_gh: the number of structures
1785 * @ghs: an array of struct gfs2_holder structures
1788 * Returns: 0 on success (all glocks acquired),
1789 * errno on failure (no glocks acquired)
1792 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1794 struct gfs2_holder *tmp[4];
1795 struct gfs2_holder **pph = tmp;
1802 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1803 return gfs2_glock_nq(ghs);
1807 pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *),
1813 error = nq_m_sync(num_gh, ghs, pph);
1822 * gfs2_glock_dq_m - release multiple glocks
1823 * @num_gh: the number of structures
1824 * @ghs: an array of struct gfs2_holder structures
1828 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1831 gfs2_glock_dq(&ghs[num_gh]);
1834 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1836 unsigned long delay = 0;
1837 unsigned long holdtime;
1838 unsigned long now = jiffies;
1840 gfs2_glock_hold(gl);
1841 spin_lock(&gl->gl_lockref.lock);
1842 holdtime = gl->gl_tchange + gl->gl_hold_time;
1843 if (!list_empty(&gl->gl_holders) &&
1844 gl->gl_name.ln_type == LM_TYPE_INODE) {
1845 if (time_before(now, holdtime))
1846 delay = holdtime - now;
1847 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1848 delay = gl->gl_hold_time;
1851 * Note 1: We cannot call demote_incompat_holders from handle_callback
1852 * or gfs2_set_demote due to recursion problems like: gfs2_glock_dq ->
1853 * handle_callback -> demote_incompat_holders -> gfs2_glock_dq
1854 * Plus, we only want to demote the holders if the request comes from
1855 * a remote cluster node because local holder conflicts are resolved
1858 * Note 2: if a remote node wants this glock in EX mode, lock_dlm will
1859 * request that we set our state to UNLOCKED. Here we mock up a holder
1860 * to make it look like someone wants the lock EX locally. Any SH
1861 * and DF requests should be able to share the lock without demoting.
1863 * Note 3: We only want to demote the demoteable holders when there
1864 * are no more strong holders. The demoteable holders might as well
1865 * keep the glock until the last strong holder is done with it.
1867 if (!find_first_strong_holder(gl)) {
1868 struct gfs2_holder mock_gh = {
1870 .gh_state = (state == LM_ST_UNLOCKED) ?
1871 LM_ST_EXCLUSIVE : state,
1872 .gh_iflags = BIT(HIF_HOLDER)
1875 demote_incompat_holders(gl, &mock_gh);
1877 handle_callback(gl, state, delay, true);
1878 __gfs2_glock_queue_work(gl, delay);
1879 spin_unlock(&gl->gl_lockref.lock);
1883 * gfs2_should_freeze - Figure out if glock should be frozen
1884 * @gl: The glock in question
1886 * Glocks are not frozen if (a) the result of the dlm operation is
1887 * an error, (b) the locking operation was an unlock operation or
1888 * (c) if there is a "noexp" flagged request anywhere in the queue
1890 * Returns: 1 if freezing should occur, 0 otherwise
1893 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1895 const struct gfs2_holder *gh;
1897 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1899 if (gl->gl_target == LM_ST_UNLOCKED)
1902 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1903 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1905 if (LM_FLAG_NOEXP & gh->gh_flags)
1913 * gfs2_glock_complete - Callback used by locking
1914 * @gl: Pointer to the glock
1915 * @ret: The return value from the dlm
1917 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1918 * to use a bitfield shared with other glock state fields.
1921 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1923 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1925 spin_lock(&gl->gl_lockref.lock);
1928 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1929 if (gfs2_should_freeze(gl)) {
1930 set_bit(GLF_FROZEN, &gl->gl_flags);
1931 spin_unlock(&gl->gl_lockref.lock);
1936 gl->gl_lockref.count++;
1937 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1938 __gfs2_glock_queue_work(gl, 0);
1939 spin_unlock(&gl->gl_lockref.lock);
1942 static int glock_cmp(void *priv, const struct list_head *a,
1943 const struct list_head *b)
1945 struct gfs2_glock *gla, *glb;
1947 gla = list_entry(a, struct gfs2_glock, gl_lru);
1948 glb = list_entry(b, struct gfs2_glock, gl_lru);
1950 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1952 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1959 * gfs2_dispose_glock_lru - Demote a list of glocks
1960 * @list: The list to dispose of
1962 * Disposing of glocks may involve disk accesses, so that here we sort
1963 * the glocks by number (i.e. disk location of the inodes) so that if
1964 * there are any such accesses, they'll be sent in order (mostly).
1966 * Must be called under the lru_lock, but may drop and retake this
1967 * lock. While the lru_lock is dropped, entries may vanish from the
1968 * list, but no new entries will appear on the list (since it is
1972 static void gfs2_dispose_glock_lru(struct list_head *list)
1973 __releases(&lru_lock)
1974 __acquires(&lru_lock)
1976 struct gfs2_glock *gl;
1978 list_sort(NULL, list, glock_cmp);
1980 while(!list_empty(list)) {
1981 gl = list_first_entry(list, struct gfs2_glock, gl_lru);
1982 list_del_init(&gl->gl_lru);
1983 clear_bit(GLF_LRU, &gl->gl_flags);
1984 if (!spin_trylock(&gl->gl_lockref.lock)) {
1986 list_add(&gl->gl_lru, &lru_list);
1987 set_bit(GLF_LRU, &gl->gl_flags);
1988 atomic_inc(&lru_count);
1991 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1992 spin_unlock(&gl->gl_lockref.lock);
1993 goto add_back_to_lru;
1995 gl->gl_lockref.count++;
1997 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1998 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1999 __gfs2_glock_queue_work(gl, 0);
2000 spin_unlock(&gl->gl_lockref.lock);
2001 cond_resched_lock(&lru_lock);
2006 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
2007 * @nr: The number of entries to scan
2009 * This function selects the entries on the LRU which are able to
2010 * be demoted, and then kicks off the process by calling
2011 * gfs2_dispose_glock_lru() above.
2014 static long gfs2_scan_glock_lru(int nr)
2016 struct gfs2_glock *gl;
2021 spin_lock(&lru_lock);
2022 while ((nr-- >= 0) && !list_empty(&lru_list)) {
2023 gl = list_first_entry(&lru_list, struct gfs2_glock, gl_lru);
2025 /* Test for being demotable */
2026 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
2027 list_move(&gl->gl_lru, &dispose);
2028 atomic_dec(&lru_count);
2033 list_move(&gl->gl_lru, &skipped);
2035 list_splice(&skipped, &lru_list);
2036 if (!list_empty(&dispose))
2037 gfs2_dispose_glock_lru(&dispose);
2038 spin_unlock(&lru_lock);
2043 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
2044 struct shrink_control *sc)
2046 if (!(sc->gfp_mask & __GFP_FS))
2048 return gfs2_scan_glock_lru(sc->nr_to_scan);
2051 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
2052 struct shrink_control *sc)
2054 return vfs_pressure_ratio(atomic_read(&lru_count));
2057 static struct shrinker glock_shrinker = {
2058 .seeks = DEFAULT_SEEKS,
2059 .count_objects = gfs2_glock_shrink_count,
2060 .scan_objects = gfs2_glock_shrink_scan,
2064 * glock_hash_walk - Call a function for glock in a hash bucket
2065 * @examiner: the function
2066 * @sdp: the filesystem
2068 * Note that the function can be called multiple times on the same
2069 * object. So the user must ensure that the function can cope with
2073 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
2075 struct gfs2_glock *gl;
2076 struct rhashtable_iter iter;
2078 rhashtable_walk_enter(&gl_hash_table, &iter);
2081 rhashtable_walk_start(&iter);
2083 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl)) {
2084 if (gl->gl_name.ln_sbd == sdp)
2088 rhashtable_walk_stop(&iter);
2089 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
2091 rhashtable_walk_exit(&iter);
2094 bool gfs2_queue_delete_work(struct gfs2_glock *gl, unsigned long delay)
2098 spin_lock(&gl->gl_lockref.lock);
2099 queued = queue_delayed_work(gfs2_delete_workqueue,
2100 &gl->gl_delete, delay);
2102 set_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2103 spin_unlock(&gl->gl_lockref.lock);
2107 void gfs2_cancel_delete_work(struct gfs2_glock *gl)
2109 if (cancel_delayed_work(&gl->gl_delete)) {
2110 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2115 bool gfs2_delete_work_queued(const struct gfs2_glock *gl)
2117 return test_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2120 static void flush_delete_work(struct gfs2_glock *gl)
2122 if (gl->gl_name.ln_type == LM_TYPE_IOPEN) {
2123 if (cancel_delayed_work(&gl->gl_delete)) {
2124 queue_delayed_work(gfs2_delete_workqueue,
2130 void gfs2_flush_delete_work(struct gfs2_sbd *sdp)
2132 glock_hash_walk(flush_delete_work, sdp);
2133 flush_workqueue(gfs2_delete_workqueue);
2137 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
2138 * @gl: The glock to thaw
2142 static void thaw_glock(struct gfs2_glock *gl)
2144 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
2146 if (!lockref_get_not_dead(&gl->gl_lockref))
2148 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
2149 gfs2_glock_queue_work(gl, 0);
2153 * clear_glock - look at a glock and see if we can free it from glock cache
2154 * @gl: the glock to look at
2158 static void clear_glock(struct gfs2_glock *gl)
2160 gfs2_glock_remove_from_lru(gl);
2162 spin_lock(&gl->gl_lockref.lock);
2163 if (!__lockref_is_dead(&gl->gl_lockref)) {
2164 gl->gl_lockref.count++;
2165 if (gl->gl_state != LM_ST_UNLOCKED)
2166 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2167 __gfs2_glock_queue_work(gl, 0);
2169 spin_unlock(&gl->gl_lockref.lock);
2173 * gfs2_glock_thaw - Thaw any frozen glocks
2174 * @sdp: The super block
2178 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
2180 glock_hash_walk(thaw_glock, sdp);
2183 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2185 spin_lock(&gl->gl_lockref.lock);
2186 gfs2_dump_glock(seq, gl, fsid);
2187 spin_unlock(&gl->gl_lockref.lock);
2190 static void dump_glock_func(struct gfs2_glock *gl)
2192 dump_glock(NULL, gl, true);
2196 * gfs2_gl_hash_clear - Empty out the glock hash table
2197 * @sdp: the filesystem
2199 * Called when unmounting the filesystem.
2202 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
2204 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
2205 flush_workqueue(glock_workqueue);
2206 glock_hash_walk(clear_glock, sdp);
2207 flush_workqueue(glock_workqueue);
2208 wait_event_timeout(sdp->sd_glock_wait,
2209 atomic_read(&sdp->sd_glock_disposal) == 0,
2211 glock_hash_walk(dump_glock_func, sdp);
2214 static const char *state2str(unsigned state)
2217 case LM_ST_UNLOCKED:
2221 case LM_ST_DEFERRED:
2223 case LM_ST_EXCLUSIVE:
2229 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
2232 if (flags & LM_FLAG_TRY)
2234 if (flags & LM_FLAG_TRY_1CB)
2236 if (flags & LM_FLAG_NOEXP)
2238 if (flags & LM_FLAG_ANY)
2240 if (flags & LM_FLAG_PRIORITY)
2242 if (flags & LM_FLAG_NODE_SCOPE)
2244 if (flags & GL_ASYNC)
2246 if (flags & GL_EXACT)
2248 if (flags & GL_NOCACHE)
2250 if (test_bit(HIF_HOLDER, &iflags))
2252 if (test_bit(HIF_WAIT, &iflags))
2254 if (test_bit(HIF_MAY_DEMOTE, &iflags))
2256 if (flags & GL_SKIP)
2263 * dump_holder - print information about a glock holder
2264 * @seq: the seq_file struct
2265 * @gh: the glock holder
2266 * @fs_id_buf: pointer to file system id (if requested)
2270 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
2271 const char *fs_id_buf)
2273 struct task_struct *gh_owner = NULL;
2277 if (gh->gh_owner_pid)
2278 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
2279 gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2280 fs_id_buf, state2str(gh->gh_state),
2281 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
2283 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
2284 gh_owner ? gh_owner->comm : "(ended)",
2289 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
2291 const unsigned long *gflags = &gl->gl_flags;
2294 if (test_bit(GLF_LOCK, gflags))
2296 if (test_bit(GLF_DEMOTE, gflags))
2298 if (test_bit(GLF_PENDING_DEMOTE, gflags))
2300 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
2302 if (test_bit(GLF_DIRTY, gflags))
2304 if (test_bit(GLF_LFLUSH, gflags))
2306 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
2308 if (test_bit(GLF_REPLY_PENDING, gflags))
2310 if (test_bit(GLF_INITIAL, gflags))
2312 if (test_bit(GLF_FROZEN, gflags))
2314 if (!list_empty(&gl->gl_holders))
2316 if (test_bit(GLF_LRU, gflags))
2320 if (test_bit(GLF_BLOCKING, gflags))
2322 if (test_bit(GLF_PENDING_DELETE, gflags))
2324 if (test_bit(GLF_FREEING, gflags))
2326 if (test_bit(GLF_INSTANTIATE_NEEDED, gflags))
2328 if (test_bit(GLF_INSTANTIATE_IN_PROG, gflags))
2335 * gfs2_dump_glock - print information about a glock
2336 * @seq: The seq_file struct
2338 * @fsid: If true, also dump the file system id
2340 * The file format is as follows:
2341 * One line per object, capital letters are used to indicate objects
2342 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2343 * other objects are indented by a single space and follow the glock to
2344 * which they are related. Fields are indicated by lower case letters
2345 * followed by a colon and the field value, except for strings which are in
2346 * [] so that its possible to see if they are composed of spaces for
2347 * example. The field's are n = number (id of the object), f = flags,
2348 * t = type, s = state, r = refcount, e = error, p = pid.
2352 void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2354 const struct gfs2_glock_operations *glops = gl->gl_ops;
2355 unsigned long long dtime;
2356 const struct gfs2_holder *gh;
2357 char gflags_buf[32];
2358 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2359 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2360 unsigned long nrpages = 0;
2362 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
2363 struct address_space *mapping = gfs2_glock2aspace(gl);
2365 nrpages = mapping->nrpages;
2367 memset(fs_id_buf, 0, sizeof(fs_id_buf));
2368 if (fsid && sdp) /* safety precaution */
2369 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2370 dtime = jiffies - gl->gl_demote_time;
2371 dtime *= 1000000/HZ; /* demote time in uSec */
2372 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
2374 gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2375 "v:%d r:%d m:%ld p:%lu\n",
2376 fs_id_buf, state2str(gl->gl_state),
2377 gl->gl_name.ln_type,
2378 (unsigned long long)gl->gl_name.ln_number,
2379 gflags2str(gflags_buf, gl),
2380 state2str(gl->gl_target),
2381 state2str(gl->gl_demote_state), dtime,
2382 atomic_read(&gl->gl_ail_count),
2383 atomic_read(&gl->gl_revokes),
2384 (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages);
2386 list_for_each_entry(gh, &gl->gl_holders, gh_list)
2387 dump_holder(seq, gh, fs_id_buf);
2389 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
2390 glops->go_dump(seq, gl, fs_id_buf);
2393 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
2395 struct gfs2_glock *gl = iter_ptr;
2397 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2398 gl->gl_name.ln_type,
2399 (unsigned long long)gl->gl_name.ln_number,
2400 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
2401 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
2402 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
2403 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
2404 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
2405 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
2406 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
2407 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
2411 static const char *gfs2_gltype[] = {
2425 static const char *gfs2_stype[] = {
2426 [GFS2_LKS_SRTT] = "srtt",
2427 [GFS2_LKS_SRTTVAR] = "srttvar",
2428 [GFS2_LKS_SRTTB] = "srttb",
2429 [GFS2_LKS_SRTTVARB] = "srttvarb",
2430 [GFS2_LKS_SIRT] = "sirt",
2431 [GFS2_LKS_SIRTVAR] = "sirtvar",
2432 [GFS2_LKS_DCOUNT] = "dlm",
2433 [GFS2_LKS_QCOUNT] = "queue",
2436 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2438 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
2440 struct gfs2_sbd *sdp = seq->private;
2441 loff_t pos = *(loff_t *)iter_ptr;
2442 unsigned index = pos >> 3;
2443 unsigned subindex = pos & 0x07;
2446 if (index == 0 && subindex != 0)
2449 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
2450 (index == 0) ? "cpu": gfs2_stype[subindex]);
2452 for_each_possible_cpu(i) {
2453 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
2456 seq_printf(seq, " %15u", i);
2458 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
2459 lkstats[index - 1].stats[subindex]);
2461 seq_putc(seq, '\n');
2465 int __init gfs2_glock_init(void)
2469 ret = rhashtable_init(&gl_hash_table, &ht_parms);
2473 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
2474 WQ_HIGHPRI | WQ_FREEZABLE, 0);
2475 if (!glock_workqueue) {
2476 rhashtable_destroy(&gl_hash_table);
2479 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
2480 WQ_MEM_RECLAIM | WQ_FREEZABLE,
2482 if (!gfs2_delete_workqueue) {
2483 destroy_workqueue(glock_workqueue);
2484 rhashtable_destroy(&gl_hash_table);
2488 ret = register_shrinker(&glock_shrinker);
2490 destroy_workqueue(gfs2_delete_workqueue);
2491 destroy_workqueue(glock_workqueue);
2492 rhashtable_destroy(&gl_hash_table);
2496 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2497 init_waitqueue_head(glock_wait_table + i);
2502 void gfs2_glock_exit(void)
2504 unregister_shrinker(&glock_shrinker);
2505 rhashtable_destroy(&gl_hash_table);
2506 destroy_workqueue(glock_workqueue);
2507 destroy_workqueue(gfs2_delete_workqueue);
2510 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2512 struct gfs2_glock *gl = gi->gl;
2517 if (!lockref_put_not_zero(&gl->gl_lockref))
2518 gfs2_glock_queue_put(gl);
2521 gl = rhashtable_walk_next(&gi->hti);
2522 if (IS_ERR_OR_NULL(gl)) {
2523 if (gl == ERR_PTR(-EAGAIN)) {
2530 if (gl->gl_name.ln_sbd != gi->sdp)
2533 if (!lockref_get_not_dead(&gl->gl_lockref))
2537 if (__lockref_is_dead(&gl->gl_lockref))
2545 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2548 struct gfs2_glock_iter *gi = seq->private;
2552 * We can either stay where we are, skip to the next hash table
2553 * entry, or start from the beginning.
2555 if (*pos < gi->last_pos) {
2556 rhashtable_walk_exit(&gi->hti);
2557 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2560 n = *pos - gi->last_pos;
2563 rhashtable_walk_start(&gi->hti);
2565 gfs2_glock_iter_next(gi, n);
2566 gi->last_pos = *pos;
2570 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2573 struct gfs2_glock_iter *gi = seq->private;
2576 gi->last_pos = *pos;
2577 gfs2_glock_iter_next(gi, 1);
2581 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2584 struct gfs2_glock_iter *gi = seq->private;
2586 rhashtable_walk_stop(&gi->hti);
2589 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2591 dump_glock(seq, iter_ptr, false);
2595 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2598 if (*pos >= GFS2_NR_SBSTATS)
2603 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2607 if (*pos >= GFS2_NR_SBSTATS)
2612 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2617 static const struct seq_operations gfs2_glock_seq_ops = {
2618 .start = gfs2_glock_seq_start,
2619 .next = gfs2_glock_seq_next,
2620 .stop = gfs2_glock_seq_stop,
2621 .show = gfs2_glock_seq_show,
2624 static const struct seq_operations gfs2_glstats_seq_ops = {
2625 .start = gfs2_glock_seq_start,
2626 .next = gfs2_glock_seq_next,
2627 .stop = gfs2_glock_seq_stop,
2628 .show = gfs2_glstats_seq_show,
2631 static const struct seq_operations gfs2_sbstats_sops = {
2632 .start = gfs2_sbstats_seq_start,
2633 .next = gfs2_sbstats_seq_next,
2634 .stop = gfs2_sbstats_seq_stop,
2635 .show = gfs2_sbstats_seq_show,
2638 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2640 static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2641 const struct seq_operations *ops)
2643 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2645 struct seq_file *seq = file->private_data;
2646 struct gfs2_glock_iter *gi = seq->private;
2648 gi->sdp = inode->i_private;
2649 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2651 seq->size = GFS2_SEQ_GOODSIZE;
2653 * Initially, we are "before" the first hash table entry; the
2654 * first call to rhashtable_walk_next gets us the first entry.
2658 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2663 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2665 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2668 static int gfs2_glocks_release(struct inode *inode, struct file *file)
2670 struct seq_file *seq = file->private_data;
2671 struct gfs2_glock_iter *gi = seq->private;
2674 gfs2_glock_put(gi->gl);
2675 rhashtable_walk_exit(&gi->hti);
2676 return seq_release_private(inode, file);
2679 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2681 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2684 static const struct file_operations gfs2_glocks_fops = {
2685 .owner = THIS_MODULE,
2686 .open = gfs2_glocks_open,
2688 .llseek = seq_lseek,
2689 .release = gfs2_glocks_release,
2692 static const struct file_operations gfs2_glstats_fops = {
2693 .owner = THIS_MODULE,
2694 .open = gfs2_glstats_open,
2696 .llseek = seq_lseek,
2697 .release = gfs2_glocks_release,
2700 DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats);
2702 void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2704 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2706 debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2709 debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2710 &gfs2_glstats_fops);
2712 debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2713 &gfs2_sbstats_fops);
2716 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2718 debugfs_remove_recursive(sdp->debugfs_dir);
2719 sdp->debugfs_dir = NULL;
2722 void gfs2_register_debugfs(void)
2724 gfs2_root = debugfs_create_dir("gfs2", NULL);
2727 void gfs2_unregister_debugfs(void)
2729 debugfs_remove(gfs2_root);