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(gh);
512 clear_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags);
513 clear_and_wake_up_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags);
518 * do_promote - promote as many requests as possible on the current queue
521 * Returns: 1 if there is a blocked holder at the head of the list, or 2
522 * if a type specific operation is underway.
525 static int do_promote(struct gfs2_glock *gl)
527 struct gfs2_holder *gh, *tmp, *first_gh;
528 bool incompat_holders_demoted = false;
530 first_gh = find_first_strong_holder(gl);
531 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
532 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
534 if (!may_grant(gl, first_gh, gh)) {
536 * If we get here, it means we may not grant this holder for
537 * some reason. If this holder is the head of the list, it
538 * means we have a blocked holder at the head, so return 1.
540 if (list_is_first(&gh->gh_list, &gl->gl_holders))
545 if (!incompat_holders_demoted) {
546 demote_incompat_holders(gl, first_gh);
547 incompat_holders_demoted = true;
550 set_bit(HIF_HOLDER, &gh->gh_iflags);
551 trace_gfs2_promote(gh);
552 gfs2_holder_wake(gh);
558 * find_first_waiter - find the first gh that's waiting for the glock
562 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
564 struct gfs2_holder *gh;
566 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
567 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
574 * state_change - record that the glock is now in a different state
576 * @new_state: the new state
579 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
583 held1 = (gl->gl_state != LM_ST_UNLOCKED);
584 held2 = (new_state != LM_ST_UNLOCKED);
586 if (held1 != held2) {
587 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
589 gl->gl_lockref.count++;
591 gl->gl_lockref.count--;
593 if (new_state != gl->gl_target)
594 /* shorten our minimum hold time */
595 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
597 gl->gl_state = new_state;
598 gl->gl_tchange = jiffies;
601 static void gfs2_set_demote(struct gfs2_glock *gl)
603 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
605 set_bit(GLF_DEMOTE, &gl->gl_flags);
607 wake_up(&sdp->sd_async_glock_wait);
610 static void gfs2_demote_wake(struct gfs2_glock *gl)
612 gl->gl_demote_state = LM_ST_EXCLUSIVE;
613 clear_bit(GLF_DEMOTE, &gl->gl_flags);
614 smp_mb__after_atomic();
615 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
619 * finish_xmote - The DLM has replied to one of our lock requests
621 * @ret: The status from the DLM
625 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
627 const struct gfs2_glock_operations *glops = gl->gl_ops;
628 struct gfs2_holder *gh;
629 unsigned state = ret & LM_OUT_ST_MASK;
632 spin_lock(&gl->gl_lockref.lock);
633 trace_gfs2_glock_state_change(gl, state);
634 state_change(gl, state);
635 gh = find_first_waiter(gl);
637 /* Demote to UN request arrived during demote to SH or DF */
638 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
639 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
640 gl->gl_target = LM_ST_UNLOCKED;
642 /* Check for state != intended state */
643 if (unlikely(state != gl->gl_target)) {
644 if (gh && (ret & LM_OUT_CANCELED))
645 gfs2_holder_wake(gh);
646 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
647 /* move to back of queue and try next entry */
648 if (ret & LM_OUT_CANCELED) {
649 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
650 list_move_tail(&gh->gh_list, &gl->gl_holders);
651 gh = find_first_waiter(gl);
652 gl->gl_target = gh->gh_state;
655 /* Some error or failed "try lock" - report it */
656 if ((ret & LM_OUT_ERROR) ||
657 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
658 gl->gl_target = gl->gl_state;
664 /* Unlocked due to conversion deadlock, try again */
667 do_xmote(gl, gh, gl->gl_target);
669 /* Conversion fails, unlock and try again */
672 do_xmote(gl, gh, LM_ST_UNLOCKED);
674 default: /* Everything else */
675 fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n",
676 gl->gl_target, state);
679 spin_unlock(&gl->gl_lockref.lock);
683 /* Fast path - we got what we asked for */
684 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
685 gfs2_demote_wake(gl);
686 if (state != LM_ST_UNLOCKED) {
687 if (glops->go_xmote_bh) {
688 spin_unlock(&gl->gl_lockref.lock);
689 rv = glops->go_xmote_bh(gl);
690 spin_lock(&gl->gl_lockref.lock);
701 clear_bit(GLF_LOCK, &gl->gl_flags);
703 spin_unlock(&gl->gl_lockref.lock);
706 static bool is_system_glock(struct gfs2_glock *gl)
708 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
709 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
711 if (gl == m_ip->i_gl)
717 * do_xmote - Calls the DLM to change the state of a lock
718 * @gl: The lock state
719 * @gh: The holder (only for promotes)
720 * @target: The target lock state
724 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
725 __releases(&gl->gl_lockref.lock)
726 __acquires(&gl->gl_lockref.lock)
728 const struct gfs2_glock_operations *glops = gl->gl_ops;
729 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
730 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
733 if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) &&
734 gh && !(gh->gh_flags & LM_FLAG_NOEXP))
736 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
738 GLOCK_BUG_ON(gl, gl->gl_state == target);
739 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
740 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
743 * If another process is already doing the invalidate, let that
744 * finish first. The glock state machine will get back to this
745 * holder again later.
747 if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
750 do_error(gl, 0); /* Fail queued try locks */
753 set_bit(GLF_BLOCKING, &gl->gl_flags);
754 if ((gl->gl_req == LM_ST_UNLOCKED) ||
755 (gl->gl_state == LM_ST_EXCLUSIVE) ||
756 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
757 clear_bit(GLF_BLOCKING, &gl->gl_flags);
758 spin_unlock(&gl->gl_lockref.lock);
759 if (glops->go_sync) {
760 ret = glops->go_sync(gl);
761 /* If we had a problem syncing (due to io errors or whatever,
762 * we should not invalidate the metadata or tell dlm to
763 * release the glock to other nodes.
766 if (cmpxchg(&sdp->sd_log_error, 0, ret)) {
767 fs_err(sdp, "Error %d syncing glock \n", ret);
768 gfs2_dump_glock(NULL, gl, true);
773 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) {
775 * The call to go_sync should have cleared out the ail list.
776 * If there are still items, we have a problem. We ought to
777 * withdraw, but we can't because the withdraw code also uses
778 * glocks. Warn about the error, dump the glock, then fall
779 * through and wait for logd to do the withdraw for us.
781 if ((atomic_read(&gl->gl_ail_count) != 0) &&
782 (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) {
783 gfs2_glock_assert_warn(gl,
784 !atomic_read(&gl->gl_ail_count));
785 gfs2_dump_glock(NULL, gl, true);
787 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
788 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
794 * Check for an error encountered since we called go_sync and go_inval.
795 * If so, we can't withdraw from the glock code because the withdraw
796 * code itself uses glocks (see function signal_our_withdraw) to
797 * change the mount to read-only. Most importantly, we must not call
798 * dlm to unlock the glock until the journal is in a known good state
799 * (after journal replay) otherwise other nodes may use the object
800 * (rgrp or dinode) and then later, journal replay will corrupt the
801 * file system. The best we can do here is wait for the logd daemon
802 * to see sd_log_error and withdraw, and in the meantime, requeue the
805 * We make a special exception for some system glocks, such as the
806 * system statfs inode glock, which needs to be granted before the
807 * gfs2_quotad daemon can exit, and that exit needs to finish before
808 * we can unmount the withdrawn file system.
810 * However, if we're just unlocking the lock (say, for unmount, when
811 * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
812 * then it's okay to tell dlm to unlock it.
814 if (unlikely(sdp->sd_log_error && !gfs2_withdrawn(sdp)))
815 gfs2_withdraw_delayed(sdp);
816 if (glock_blocked_by_withdraw(gl) &&
817 (target != LM_ST_UNLOCKED ||
818 test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags))) {
819 if (!is_system_glock(gl)) {
820 gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD);
823 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
827 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
829 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
830 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
831 target == LM_ST_UNLOCKED &&
832 test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
833 finish_xmote(gl, target);
834 gfs2_glock_queue_work(gl, 0);
836 fs_err(sdp, "lm_lock ret %d\n", ret);
837 GLOCK_BUG_ON(gl, !gfs2_withdrawn(sdp));
839 } else { /* lock_nolock */
840 finish_xmote(gl, target);
841 gfs2_glock_queue_work(gl, 0);
844 spin_lock(&gl->gl_lockref.lock);
848 * run_queue - do all outstanding tasks related to a glock
849 * @gl: The glock in question
850 * @nonblock: True if we must not block in run_queue
854 static void run_queue(struct gfs2_glock *gl, const int nonblock)
855 __releases(&gl->gl_lockref.lock)
856 __acquires(&gl->gl_lockref.lock)
858 struct gfs2_holder *gh = NULL;
861 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
864 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
866 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
867 gl->gl_demote_state != gl->gl_state) {
868 if (find_first_holder(gl))
872 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
873 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
874 gl->gl_target = gl->gl_demote_state;
876 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
877 gfs2_demote_wake(gl);
878 ret = do_promote(gl);
883 gh = find_first_waiter(gl);
884 gl->gl_target = gh->gh_state;
885 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
886 do_error(gl, 0); /* Fail queued try locks */
888 do_xmote(gl, gh, gl->gl_target);
893 clear_bit(GLF_LOCK, &gl->gl_flags);
894 smp_mb__after_atomic();
895 gl->gl_lockref.count++;
896 __gfs2_glock_queue_work(gl, 0);
900 clear_bit(GLF_LOCK, &gl->gl_flags);
901 smp_mb__after_atomic();
905 void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation)
907 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
909 if (ri->ri_magic == 0)
910 ri->ri_magic = cpu_to_be32(GFS2_MAGIC);
911 if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC))
912 ri->ri_generation_deleted = cpu_to_be64(generation);
915 bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation)
917 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
919 if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC))
921 return generation <= be64_to_cpu(ri->ri_generation_deleted);
924 static void gfs2_glock_poke(struct gfs2_glock *gl)
926 int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP;
927 struct gfs2_holder gh;
930 __gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh, _RET_IP_);
931 error = gfs2_glock_nq(&gh);
934 gfs2_holder_uninit(&gh);
937 static bool gfs2_try_evict(struct gfs2_glock *gl)
939 struct gfs2_inode *ip;
940 bool evicted = false;
943 * If there is contention on the iopen glock and we have an inode, try
944 * to grab and release the inode so that it can be evicted. This will
945 * allow the remote node to go ahead and delete the inode without us
946 * having to do it, which will avoid rgrp glock thrashing.
948 * The remote node is likely still holding the corresponding inode
949 * glock, so it will run before we get to verify that the delete has
952 spin_lock(&gl->gl_lockref.lock);
954 if (ip && !igrab(&ip->i_inode))
956 spin_unlock(&gl->gl_lockref.lock);
958 struct gfs2_glock *inode_gl = NULL;
960 gl->gl_no_formal_ino = ip->i_no_formal_ino;
961 set_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
962 d_prune_aliases(&ip->i_inode);
965 /* If the inode was evicted, gl->gl_object will now be NULL. */
966 spin_lock(&gl->gl_lockref.lock);
970 lockref_get(&inode_gl->gl_lockref);
971 clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
973 spin_unlock(&gl->gl_lockref.lock);
975 gfs2_glock_poke(inode_gl);
976 gfs2_glock_put(inode_gl);
983 static void delete_work_func(struct work_struct *work)
985 struct delayed_work *dwork = to_delayed_work(work);
986 struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete);
987 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
989 u64 no_addr = gl->gl_name.ln_number;
991 spin_lock(&gl->gl_lockref.lock);
992 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
993 spin_unlock(&gl->gl_lockref.lock);
995 if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
997 * If we can evict the inode, give the remote node trying to
998 * delete the inode some time before verifying that the delete
999 * has happened. Otherwise, if we cause contention on the inode glock
1000 * immediately, the remote node will think that we still have
1001 * the inode in use, and so it will give up waiting.
1003 * If we can't evict the inode, signal to the remote node that
1004 * the inode is still in use. We'll later try to delete the
1005 * inode locally in gfs2_evict_inode.
1007 * FIXME: We only need to verify that the remote node has
1008 * deleted the inode because nodes before this remote delete
1009 * rework won't cooperate. At a later time, when we no longer
1010 * care about compatibility with such nodes, we can skip this
1013 if (gfs2_try_evict(gl)) {
1014 if (gfs2_queue_delete_work(gl, 5 * HZ))
1020 inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino,
1021 GFS2_BLKST_UNLINKED);
1022 if (!IS_ERR_OR_NULL(inode)) {
1023 d_prune_aliases(inode);
1030 static void glock_work_func(struct work_struct *work)
1032 unsigned long delay = 0;
1033 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
1034 unsigned int drop_refs = 1;
1036 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
1037 finish_xmote(gl, gl->gl_reply);
1040 spin_lock(&gl->gl_lockref.lock);
1041 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1042 gl->gl_state != LM_ST_UNLOCKED &&
1043 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
1044 unsigned long holdtime, now = jiffies;
1046 holdtime = gl->gl_tchange + gl->gl_hold_time;
1047 if (time_before(now, holdtime))
1048 delay = holdtime - now;
1051 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1052 gfs2_set_demote(gl);
1057 /* Keep one glock reference for the work we requeue. */
1059 if (gl->gl_name.ln_type != LM_TYPE_INODE)
1061 __gfs2_glock_queue_work(gl, delay);
1065 * Drop the remaining glock references manually here. (Mind that
1066 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
1069 gl->gl_lockref.count -= drop_refs;
1070 if (!gl->gl_lockref.count) {
1071 __gfs2_glock_put(gl);
1074 spin_unlock(&gl->gl_lockref.lock);
1077 static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
1078 struct gfs2_glock *new)
1080 struct wait_glock_queue wait;
1081 wait_queue_head_t *wq = glock_waitqueue(name);
1082 struct gfs2_glock *gl;
1085 init_wait(&wait.wait);
1086 wait.wait.func = glock_wake_function;
1089 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1092 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
1093 &new->gl_node, ht_parms);
1097 gl = rhashtable_lookup_fast(&gl_hash_table,
1100 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
1107 finish_wait(wq, &wait.wait);
1112 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
1113 * @sdp: The GFS2 superblock
1114 * @number: the lock number
1115 * @glops: The glock_operations to use
1116 * @create: If 0, don't create the glock if it doesn't exist
1117 * @glp: the glock is returned here
1119 * This does not lock a glock, just finds/creates structures for one.
1124 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
1125 const struct gfs2_glock_operations *glops, int create,
1126 struct gfs2_glock **glp)
1128 struct super_block *s = sdp->sd_vfs;
1129 struct lm_lockname name = { .ln_number = number,
1130 .ln_type = glops->go_type,
1132 struct gfs2_glock *gl, *tmp;
1133 struct address_space *mapping;
1136 gl = find_insert_glock(&name, NULL);
1144 if (glops->go_flags & GLOF_ASPACE) {
1145 struct gfs2_glock_aspace *gla =
1146 kmem_cache_alloc(gfs2_glock_aspace_cachep, GFP_NOFS);
1151 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_NOFS);
1155 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
1158 if (glops->go_flags & GLOF_LVB) {
1159 gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1160 if (!gl->gl_lksb.sb_lvbptr) {
1161 gfs2_glock_dealloc(&gl->gl_rcu);
1166 atomic_inc(&sdp->sd_glock_disposal);
1167 gl->gl_node.next = NULL;
1168 gl->gl_flags = glops->go_instantiate ? BIT(GLF_INSTANTIATE_NEEDED) : 0;
1170 lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass);
1171 gl->gl_lockref.count = 1;
1172 gl->gl_state = LM_ST_UNLOCKED;
1173 gl->gl_target = LM_ST_UNLOCKED;
1174 gl->gl_demote_state = LM_ST_EXCLUSIVE;
1177 /* We use the global stats to estimate the initial per-glock stats */
1178 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
1180 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
1181 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
1182 gl->gl_tchange = jiffies;
1183 gl->gl_object = NULL;
1184 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
1185 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
1186 if (gl->gl_name.ln_type == LM_TYPE_IOPEN)
1187 INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func);
1189 mapping = gfs2_glock2aspace(gl);
1191 mapping->a_ops = &gfs2_meta_aops;
1192 mapping->host = s->s_bdev->bd_inode;
1194 mapping_set_gfp_mask(mapping, GFP_NOFS);
1195 mapping->private_data = NULL;
1196 mapping->writeback_index = 0;
1199 tmp = find_insert_glock(&name, gl);
1211 gfs2_glock_dealloc(&gl->gl_rcu);
1212 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
1213 wake_up(&sdp->sd_glock_wait);
1220 * __gfs2_holder_init - initialize a struct gfs2_holder in the default way
1222 * @state: the state we're requesting
1223 * @flags: the modifier flags
1224 * @gh: the holder structure
1228 void __gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
1229 struct gfs2_holder *gh, unsigned long ip)
1231 INIT_LIST_HEAD(&gh->gh_list);
1234 gh->gh_owner_pid = get_pid(task_pid(current));
1235 gh->gh_state = state;
1236 gh->gh_flags = flags;
1238 gfs2_glock_hold(gl);
1242 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1243 * @state: the state we're requesting
1244 * @flags: the modifier flags
1245 * @gh: the holder structure
1247 * Don't mess with the glock.
1251 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
1253 gh->gh_state = state;
1254 gh->gh_flags = flags;
1256 gh->gh_ip = _RET_IP_;
1257 put_pid(gh->gh_owner_pid);
1258 gh->gh_owner_pid = get_pid(task_pid(current));
1262 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1263 * @gh: the holder structure
1267 void gfs2_holder_uninit(struct gfs2_holder *gh)
1269 put_pid(gh->gh_owner_pid);
1270 gfs2_glock_put(gh->gh_gl);
1271 gfs2_holder_mark_uninitialized(gh);
1275 static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
1276 unsigned long start_time)
1278 /* Have we waited longer that a second? */
1279 if (time_after(jiffies, start_time + HZ)) {
1280 /* Lengthen the minimum hold time. */
1281 gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
1287 * gfs2_glock_holder_ready - holder is ready and its error code can be collected
1288 * @gh: the glock holder
1290 * Called when a glock holder no longer needs to be waited for because it is
1291 * now either held (HIF_HOLDER set; gh_error == 0), or acquiring the lock has
1292 * failed (gh_error != 0).
1295 int gfs2_glock_holder_ready(struct gfs2_holder *gh)
1297 if (gh->gh_error || (gh->gh_flags & GL_SKIP))
1298 return gh->gh_error;
1299 gh->gh_error = gfs2_instantiate(gh);
1302 return gh->gh_error;
1306 * gfs2_glock_wait - wait on a glock acquisition
1307 * @gh: the glock holder
1309 * Returns: 0 on success
1312 int gfs2_glock_wait(struct gfs2_holder *gh)
1314 unsigned long start_time = jiffies;
1317 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1318 gfs2_glock_update_hold_time(gh->gh_gl, start_time);
1319 return gfs2_glock_holder_ready(gh);
1322 static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
1326 for (i = 0; i < num_gh; i++)
1327 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
1333 * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1334 * @num_gh: the number of holders in the array
1335 * @ghs: the glock holder array
1337 * Returns: 0 on success, meaning all glocks have been granted and are held.
1338 * -ESTALE if the request timed out, meaning all glocks were released,
1339 * and the caller should retry the operation.
1342 int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
1344 struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
1345 int i, ret = 0, timeout = 0;
1346 unsigned long start_time = jiffies;
1350 * Total up the (minimum hold time * 2) of all glocks and use that to
1351 * determine the max amount of time we should wait.
1353 for (i = 0; i < num_gh; i++)
1354 timeout += ghs[i].gh_gl->gl_hold_time << 1;
1356 if (!wait_event_timeout(sdp->sd_async_glock_wait,
1357 !glocks_pending(num_gh, ghs), timeout)) {
1358 ret = -ESTALE; /* request timed out. */
1362 for (i = 0; i < num_gh; i++) {
1363 struct gfs2_holder *gh = &ghs[i];
1366 if (test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1367 gfs2_glock_update_hold_time(gh->gh_gl,
1370 ret2 = gfs2_glock_holder_ready(gh);
1377 for (i = 0; i < num_gh; i++) {
1378 struct gfs2_holder *gh = &ghs[i];
1387 * handle_callback - process a demote request
1389 * @state: the state the caller wants us to change to
1390 * @delay: zero to demote immediately; otherwise pending demote
1391 * @remote: true if this came from a different cluster node
1393 * There are only two requests that we are going to see in actual
1394 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1397 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
1398 unsigned long delay, bool remote)
1401 set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1403 gfs2_set_demote(gl);
1404 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
1405 gl->gl_demote_state = state;
1406 gl->gl_demote_time = jiffies;
1407 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
1408 gl->gl_demote_state != state) {
1409 gl->gl_demote_state = LM_ST_UNLOCKED;
1411 if (gl->gl_ops->go_callback)
1412 gl->gl_ops->go_callback(gl, remote);
1413 trace_gfs2_demote_rq(gl, remote);
1416 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
1418 struct va_format vaf;
1421 va_start(args, fmt);
1424 seq_vprintf(seq, fmt, args);
1429 pr_err("%pV", &vaf);
1436 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1437 * @gh: the holder structure to add
1439 * Eventually we should move the recursive locking trap to a
1440 * debugging option or something like that. This is the fast
1441 * path and needs to have the minimum number of distractions.
1445 static inline void add_to_queue(struct gfs2_holder *gh)
1446 __releases(&gl->gl_lockref.lock)
1447 __acquires(&gl->gl_lockref.lock)
1449 struct gfs2_glock *gl = gh->gh_gl;
1450 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1451 struct list_head *insert_pt = NULL;
1452 struct gfs2_holder *gh2;
1455 GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1456 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1457 GLOCK_BUG_ON(gl, true);
1459 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1460 if (test_bit(GLF_LOCK, &gl->gl_flags)) {
1461 struct gfs2_holder *first_gh;
1463 first_gh = find_first_strong_holder(gl);
1464 try_futile = !may_grant(gl, first_gh, gh);
1466 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1470 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1471 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
1472 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK) &&
1473 !test_bit(HIF_MAY_DEMOTE, &gh2->gh_iflags)))
1474 goto trap_recursive;
1476 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1478 gh->gh_error = GLR_TRYFAILED;
1479 gfs2_holder_wake(gh);
1482 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1484 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1485 insert_pt = &gh2->gh_list;
1487 trace_gfs2_glock_queue(gh, 1);
1488 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1489 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1490 if (likely(insert_pt == NULL)) {
1491 list_add_tail(&gh->gh_list, &gl->gl_holders);
1492 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1496 list_add_tail(&gh->gh_list, insert_pt);
1498 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
1499 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1500 spin_unlock(&gl->gl_lockref.lock);
1501 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1502 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1503 spin_lock(&gl->gl_lockref.lock);
1508 fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1509 fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1510 fs_err(sdp, "lock type: %d req lock state : %d\n",
1511 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1512 fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1513 fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1514 fs_err(sdp, "lock type: %d req lock state : %d\n",
1515 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1516 gfs2_dump_glock(NULL, gl, true);
1521 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1522 * @gh: the holder structure
1524 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1526 * Returns: 0, GLR_TRYFAILED, or errno on failure
1529 int gfs2_glock_nq(struct gfs2_holder *gh)
1531 struct gfs2_glock *gl = gh->gh_gl;
1534 if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP))
1537 if (test_bit(GLF_LRU, &gl->gl_flags))
1538 gfs2_glock_remove_from_lru(gl);
1541 spin_lock(&gl->gl_lockref.lock);
1543 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1544 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1545 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1546 gl->gl_lockref.count++;
1547 __gfs2_glock_queue_work(gl, 0);
1550 spin_unlock(&gl->gl_lockref.lock);
1552 if (!(gh->gh_flags & GL_ASYNC))
1553 error = gfs2_glock_wait(gh);
1559 * gfs2_glock_poll - poll to see if an async request has been completed
1562 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1565 int gfs2_glock_poll(struct gfs2_holder *gh)
1567 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1570 static inline bool needs_demote(struct gfs2_glock *gl)
1572 return (test_bit(GLF_DEMOTE, &gl->gl_flags) ||
1573 test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags));
1576 static void __gfs2_glock_dq(struct gfs2_holder *gh)
1578 struct gfs2_glock *gl = gh->gh_gl;
1579 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1584 * This while loop is similar to function demote_incompat_holders:
1585 * If the glock is due to be demoted (which may be from another node
1586 * or even if this holder is GL_NOCACHE), the weak holders are
1587 * demoted as well, allowing the glock to be demoted.
1591 * If we're in the process of file system withdraw, we cannot
1592 * just dequeue any glocks until our journal is recovered, lest
1593 * we introduce file system corruption. We need two exceptions
1594 * to this rule: We need to allow unlocking of nondisk glocks
1595 * and the glock for our own journal that needs recovery.
1597 if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) &&
1598 glock_blocked_by_withdraw(gl) &&
1599 gh->gh_gl != sdp->sd_jinode_gl) {
1600 sdp->sd_glock_dqs_held++;
1601 spin_unlock(&gl->gl_lockref.lock);
1603 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
1604 TASK_UNINTERRUPTIBLE);
1605 spin_lock(&gl->gl_lockref.lock);
1609 * This holder should not be cached, so mark it for demote.
1610 * Note: this should be done before the check for needs_demote
1613 if (gh->gh_flags & GL_NOCACHE)
1614 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1616 list_del_init(&gh->gh_list);
1617 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1618 trace_gfs2_glock_queue(gh, 0);
1621 * If there hasn't been a demote request we are done.
1622 * (Let the remaining holders, if any, keep holding it.)
1624 if (!needs_demote(gl)) {
1625 if (list_empty(&gl->gl_holders))
1630 * If we have another strong holder (we cannot auto-demote)
1631 * we are done. It keeps holding it until it is done.
1633 if (find_first_strong_holder(gl))
1637 * If we have a weak holder at the head of the list, it
1638 * (and all others like it) must be auto-demoted. If there
1639 * are no more weak holders, we exit the while loop.
1641 gh = find_first_holder(gl);
1644 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1645 gfs2_glock_add_to_lru(gl);
1647 if (unlikely(!fast_path)) {
1648 gl->gl_lockref.count++;
1649 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1650 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1651 gl->gl_name.ln_type == LM_TYPE_INODE)
1652 delay = gl->gl_hold_time;
1653 __gfs2_glock_queue_work(gl, delay);
1658 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1659 * @gh: the glock holder
1662 void gfs2_glock_dq(struct gfs2_holder *gh)
1664 struct gfs2_glock *gl = gh->gh_gl;
1666 spin_lock(&gl->gl_lockref.lock);
1667 if (list_is_first(&gh->gh_list, &gl->gl_holders) &&
1668 !test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1669 spin_unlock(&gl->gl_lockref.lock);
1670 gl->gl_name.ln_sbd->sd_lockstruct.ls_ops->lm_cancel(gl);
1671 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1672 spin_lock(&gl->gl_lockref.lock);
1675 __gfs2_glock_dq(gh);
1676 spin_unlock(&gl->gl_lockref.lock);
1679 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1681 struct gfs2_glock *gl = gh->gh_gl;
1684 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1688 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1689 * @gh: the holder structure
1693 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1696 gfs2_holder_uninit(gh);
1700 * gfs2_glock_nq_num - acquire a glock based on lock number
1701 * @sdp: the filesystem
1702 * @number: the lock number
1703 * @glops: the glock operations for the type of glock
1704 * @state: the state to acquire the glock in
1705 * @flags: modifier flags for the acquisition
1706 * @gh: the struct gfs2_holder
1711 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1712 const struct gfs2_glock_operations *glops,
1713 unsigned int state, u16 flags, struct gfs2_holder *gh)
1715 struct gfs2_glock *gl;
1718 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1720 error = gfs2_glock_nq_init(gl, state, flags, gh);
1728 * glock_compare - Compare two struct gfs2_glock structures for sorting
1729 * @arg_a: the first structure
1730 * @arg_b: the second structure
1734 static int glock_compare(const void *arg_a, const void *arg_b)
1736 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1737 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1738 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1739 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1741 if (a->ln_number > b->ln_number)
1743 if (a->ln_number < b->ln_number)
1745 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1750 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1751 * @num_gh: the number of structures
1752 * @ghs: an array of struct gfs2_holder structures
1753 * @p: placeholder for the holder structure to pass back
1755 * Returns: 0 on success (all glocks acquired),
1756 * errno on failure (no glocks acquired)
1759 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1760 struct gfs2_holder **p)
1765 for (x = 0; x < num_gh; x++)
1768 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1770 for (x = 0; x < num_gh; x++) {
1771 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1773 error = gfs2_glock_nq(p[x]);
1776 gfs2_glock_dq(p[x]);
1785 * gfs2_glock_nq_m - acquire multiple glocks
1786 * @num_gh: the number of structures
1787 * @ghs: an array of struct gfs2_holder structures
1790 * Returns: 0 on success (all glocks acquired),
1791 * errno on failure (no glocks acquired)
1794 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1796 struct gfs2_holder *tmp[4];
1797 struct gfs2_holder **pph = tmp;
1804 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
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 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
2218 struct gfs2_glock *gl = ip->i_gl;
2221 ret = gfs2_truncatei_resume(ip);
2222 gfs2_glock_assert_withdraw(gl, ret == 0);
2224 spin_lock(&gl->gl_lockref.lock);
2225 clear_bit(GLF_LOCK, &gl->gl_flags);
2228 spin_unlock(&gl->gl_lockref.lock);
2231 void gfs2_wait_truncate(struct gfs2_inode *ip)
2233 struct gfs2_glock *gl = ip->i_gl;
2234 wait_queue_head_t *wq = glock_waitqueue(&gl->gl_name);
2236 wait_event(*wq, !(ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG));
2239 static const char *state2str(unsigned state)
2242 case LM_ST_UNLOCKED:
2246 case LM_ST_DEFERRED:
2248 case LM_ST_EXCLUSIVE:
2254 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
2257 if (flags & LM_FLAG_TRY)
2259 if (flags & LM_FLAG_TRY_1CB)
2261 if (flags & LM_FLAG_NOEXP)
2263 if (flags & LM_FLAG_ANY)
2265 if (flags & LM_FLAG_PRIORITY)
2267 if (flags & LM_FLAG_NODE_SCOPE)
2269 if (flags & GL_ASYNC)
2271 if (flags & GL_EXACT)
2273 if (flags & GL_NOCACHE)
2275 if (test_bit(HIF_HOLDER, &iflags))
2277 if (test_bit(HIF_WAIT, &iflags))
2279 if (test_bit(HIF_MAY_DEMOTE, &iflags))
2281 if (flags & GL_SKIP)
2288 * dump_holder - print information about a glock holder
2289 * @seq: the seq_file struct
2290 * @gh: the glock holder
2291 * @fs_id_buf: pointer to file system id (if requested)
2295 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
2296 const char *fs_id_buf)
2298 struct task_struct *gh_owner = NULL;
2302 if (gh->gh_owner_pid)
2303 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
2304 gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2305 fs_id_buf, state2str(gh->gh_state),
2306 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
2308 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
2309 gh_owner ? gh_owner->comm : "(ended)",
2314 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
2316 const unsigned long *gflags = &gl->gl_flags;
2319 if (test_bit(GLF_LOCK, gflags))
2321 if (test_bit(GLF_DEMOTE, gflags))
2323 if (test_bit(GLF_PENDING_DEMOTE, gflags))
2325 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
2327 if (test_bit(GLF_DIRTY, gflags))
2329 if (test_bit(GLF_LFLUSH, gflags))
2331 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
2333 if (test_bit(GLF_REPLY_PENDING, gflags))
2335 if (test_bit(GLF_INITIAL, gflags))
2337 if (test_bit(GLF_FROZEN, gflags))
2339 if (!list_empty(&gl->gl_holders))
2341 if (test_bit(GLF_LRU, gflags))
2345 if (test_bit(GLF_BLOCKING, gflags))
2347 if (test_bit(GLF_PENDING_DELETE, gflags))
2349 if (test_bit(GLF_FREEING, gflags))
2351 if (test_bit(GLF_INSTANTIATE_NEEDED, gflags))
2353 if (test_bit(GLF_INSTANTIATE_IN_PROG, gflags))
2360 * gfs2_dump_glock - print information about a glock
2361 * @seq: The seq_file struct
2363 * @fsid: If true, also dump the file system id
2365 * The file format is as follows:
2366 * One line per object, capital letters are used to indicate objects
2367 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2368 * other objects are indented by a single space and follow the glock to
2369 * which they are related. Fields are indicated by lower case letters
2370 * followed by a colon and the field value, except for strings which are in
2371 * [] so that its possible to see if they are composed of spaces for
2372 * example. The field's are n = number (id of the object), f = flags,
2373 * t = type, s = state, r = refcount, e = error, p = pid.
2377 void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2379 const struct gfs2_glock_operations *glops = gl->gl_ops;
2380 unsigned long long dtime;
2381 const struct gfs2_holder *gh;
2382 char gflags_buf[32];
2383 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2384 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2385 unsigned long nrpages = 0;
2387 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
2388 struct address_space *mapping = gfs2_glock2aspace(gl);
2390 nrpages = mapping->nrpages;
2392 memset(fs_id_buf, 0, sizeof(fs_id_buf));
2393 if (fsid && sdp) /* safety precaution */
2394 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2395 dtime = jiffies - gl->gl_demote_time;
2396 dtime *= 1000000/HZ; /* demote time in uSec */
2397 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
2399 gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2400 "v:%d r:%d m:%ld p:%lu\n",
2401 fs_id_buf, state2str(gl->gl_state),
2402 gl->gl_name.ln_type,
2403 (unsigned long long)gl->gl_name.ln_number,
2404 gflags2str(gflags_buf, gl),
2405 state2str(gl->gl_target),
2406 state2str(gl->gl_demote_state), dtime,
2407 atomic_read(&gl->gl_ail_count),
2408 atomic_read(&gl->gl_revokes),
2409 (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages);
2411 list_for_each_entry(gh, &gl->gl_holders, gh_list)
2412 dump_holder(seq, gh, fs_id_buf);
2414 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
2415 glops->go_dump(seq, gl, fs_id_buf);
2418 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
2420 struct gfs2_glock *gl = iter_ptr;
2422 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2423 gl->gl_name.ln_type,
2424 (unsigned long long)gl->gl_name.ln_number,
2425 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
2426 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
2427 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
2428 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
2429 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
2430 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
2431 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
2432 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
2436 static const char *gfs2_gltype[] = {
2450 static const char *gfs2_stype[] = {
2451 [GFS2_LKS_SRTT] = "srtt",
2452 [GFS2_LKS_SRTTVAR] = "srttvar",
2453 [GFS2_LKS_SRTTB] = "srttb",
2454 [GFS2_LKS_SRTTVARB] = "srttvarb",
2455 [GFS2_LKS_SIRT] = "sirt",
2456 [GFS2_LKS_SIRTVAR] = "sirtvar",
2457 [GFS2_LKS_DCOUNT] = "dlm",
2458 [GFS2_LKS_QCOUNT] = "queue",
2461 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2463 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
2465 struct gfs2_sbd *sdp = seq->private;
2466 loff_t pos = *(loff_t *)iter_ptr;
2467 unsigned index = pos >> 3;
2468 unsigned subindex = pos & 0x07;
2471 if (index == 0 && subindex != 0)
2474 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
2475 (index == 0) ? "cpu": gfs2_stype[subindex]);
2477 for_each_possible_cpu(i) {
2478 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
2481 seq_printf(seq, " %15u", i);
2483 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
2484 lkstats[index - 1].stats[subindex]);
2486 seq_putc(seq, '\n');
2490 int __init gfs2_glock_init(void)
2494 ret = rhashtable_init(&gl_hash_table, &ht_parms);
2498 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
2499 WQ_HIGHPRI | WQ_FREEZABLE, 0);
2500 if (!glock_workqueue) {
2501 rhashtable_destroy(&gl_hash_table);
2504 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
2505 WQ_MEM_RECLAIM | WQ_FREEZABLE,
2507 if (!gfs2_delete_workqueue) {
2508 destroy_workqueue(glock_workqueue);
2509 rhashtable_destroy(&gl_hash_table);
2513 ret = register_shrinker(&glock_shrinker);
2515 destroy_workqueue(gfs2_delete_workqueue);
2516 destroy_workqueue(glock_workqueue);
2517 rhashtable_destroy(&gl_hash_table);
2521 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2522 init_waitqueue_head(glock_wait_table + i);
2527 void gfs2_glock_exit(void)
2529 unregister_shrinker(&glock_shrinker);
2530 rhashtable_destroy(&gl_hash_table);
2531 destroy_workqueue(glock_workqueue);
2532 destroy_workqueue(gfs2_delete_workqueue);
2535 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2537 struct gfs2_glock *gl = gi->gl;
2542 if (!lockref_put_not_zero(&gl->gl_lockref))
2543 gfs2_glock_queue_put(gl);
2546 gl = rhashtable_walk_next(&gi->hti);
2547 if (IS_ERR_OR_NULL(gl)) {
2548 if (gl == ERR_PTR(-EAGAIN)) {
2555 if (gl->gl_name.ln_sbd != gi->sdp)
2558 if (!lockref_get_not_dead(&gl->gl_lockref))
2562 if (__lockref_is_dead(&gl->gl_lockref))
2570 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2573 struct gfs2_glock_iter *gi = seq->private;
2577 * We can either stay where we are, skip to the next hash table
2578 * entry, or start from the beginning.
2580 if (*pos < gi->last_pos) {
2581 rhashtable_walk_exit(&gi->hti);
2582 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2585 n = *pos - gi->last_pos;
2588 rhashtable_walk_start(&gi->hti);
2590 gfs2_glock_iter_next(gi, n);
2591 gi->last_pos = *pos;
2595 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2598 struct gfs2_glock_iter *gi = seq->private;
2601 gi->last_pos = *pos;
2602 gfs2_glock_iter_next(gi, 1);
2606 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2609 struct gfs2_glock_iter *gi = seq->private;
2611 rhashtable_walk_stop(&gi->hti);
2614 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2616 dump_glock(seq, iter_ptr, false);
2620 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2623 if (*pos >= GFS2_NR_SBSTATS)
2628 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2632 if (*pos >= GFS2_NR_SBSTATS)
2637 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2642 static const struct seq_operations gfs2_glock_seq_ops = {
2643 .start = gfs2_glock_seq_start,
2644 .next = gfs2_glock_seq_next,
2645 .stop = gfs2_glock_seq_stop,
2646 .show = gfs2_glock_seq_show,
2649 static const struct seq_operations gfs2_glstats_seq_ops = {
2650 .start = gfs2_glock_seq_start,
2651 .next = gfs2_glock_seq_next,
2652 .stop = gfs2_glock_seq_stop,
2653 .show = gfs2_glstats_seq_show,
2656 static const struct seq_operations gfs2_sbstats_sops = {
2657 .start = gfs2_sbstats_seq_start,
2658 .next = gfs2_sbstats_seq_next,
2659 .stop = gfs2_sbstats_seq_stop,
2660 .show = gfs2_sbstats_seq_show,
2663 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2665 static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2666 const struct seq_operations *ops)
2668 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2670 struct seq_file *seq = file->private_data;
2671 struct gfs2_glock_iter *gi = seq->private;
2673 gi->sdp = inode->i_private;
2674 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2676 seq->size = GFS2_SEQ_GOODSIZE;
2678 * Initially, we are "before" the first hash table entry; the
2679 * first call to rhashtable_walk_next gets us the first entry.
2683 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2688 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2690 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2693 static int gfs2_glocks_release(struct inode *inode, struct file *file)
2695 struct seq_file *seq = file->private_data;
2696 struct gfs2_glock_iter *gi = seq->private;
2699 gfs2_glock_put(gi->gl);
2700 rhashtable_walk_exit(&gi->hti);
2701 return seq_release_private(inode, file);
2704 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2706 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2709 static const struct file_operations gfs2_glocks_fops = {
2710 .owner = THIS_MODULE,
2711 .open = gfs2_glocks_open,
2713 .llseek = seq_lseek,
2714 .release = gfs2_glocks_release,
2717 static const struct file_operations gfs2_glstats_fops = {
2718 .owner = THIS_MODULE,
2719 .open = gfs2_glstats_open,
2721 .llseek = seq_lseek,
2722 .release = gfs2_glocks_release,
2725 DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats);
2727 void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2729 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2731 debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2734 debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2735 &gfs2_glstats_fops);
2737 debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2738 &gfs2_sbstats_fops);
2741 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2743 debugfs_remove_recursive(sdp->debugfs_dir);
2744 sdp->debugfs_dir = NULL;
2747 void gfs2_register_debugfs(void)
2749 gfs2_root = debugfs_create_dir("gfs2", NULL);
2752 void gfs2_unregister_debugfs(void)
2754 debugfs_remove(gfs2_root);