1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 #include <linux/sched.h>
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/buffer_head.h>
13 #include <linux/delay.h>
14 #include <linux/sort.h>
15 #include <linux/hash.h>
16 #include <linux/jhash.h>
17 #include <linux/kallsyms.h>
18 #include <linux/gfs2_ondisk.h>
19 #include <linux/list.h>
20 #include <linux/wait.h>
21 #include <linux/module.h>
22 #include <linux/uaccess.h>
23 #include <linux/seq_file.h>
24 #include <linux/debugfs.h>
25 #include <linux/kthread.h>
26 #include <linux/freezer.h>
27 #include <linux/workqueue.h>
28 #include <linux/jiffies.h>
29 #include <linux/rcupdate.h>
30 #include <linux/rculist_bl.h>
31 #include <linux/bit_spinlock.h>
32 #include <linux/percpu.h>
33 #include <linux/list_sort.h>
34 #include <linux/lockref.h>
35 #include <linux/rhashtable.h>
48 #define CREATE_TRACE_POINTS
49 #include "trace_gfs2.h"
51 struct gfs2_glock_iter {
52 struct gfs2_sbd *sdp; /* incore superblock */
53 struct rhashtable_iter hti; /* rhashtable iterator */
54 struct gfs2_glock *gl; /* current glock struct */
55 loff_t last_pos; /* last position */
58 typedef void (*glock_examiner) (struct gfs2_glock * gl);
60 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
61 static void __gfs2_glock_dq(struct gfs2_holder *gh);
63 static struct dentry *gfs2_root;
64 static struct workqueue_struct *glock_workqueue;
65 struct workqueue_struct *gfs2_delete_workqueue;
66 static LIST_HEAD(lru_list);
67 static atomic_t lru_count = ATOMIC_INIT(0);
68 static DEFINE_SPINLOCK(lru_lock);
70 #define GFS2_GL_HASH_SHIFT 15
71 #define GFS2_GL_HASH_SIZE BIT(GFS2_GL_HASH_SHIFT)
73 static const struct rhashtable_params ht_parms = {
74 .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4,
75 .key_len = offsetofend(struct lm_lockname, ln_type),
76 .key_offset = offsetof(struct gfs2_glock, gl_name),
77 .head_offset = offsetof(struct gfs2_glock, gl_node),
80 static struct rhashtable gl_hash_table;
82 #define GLOCK_WAIT_TABLE_BITS 12
83 #define GLOCK_WAIT_TABLE_SIZE (1 << GLOCK_WAIT_TABLE_BITS)
84 static wait_queue_head_t glock_wait_table[GLOCK_WAIT_TABLE_SIZE] __cacheline_aligned;
86 struct wait_glock_queue {
87 struct lm_lockname *name;
88 wait_queue_entry_t wait;
91 static int glock_wake_function(wait_queue_entry_t *wait, unsigned int mode,
94 struct wait_glock_queue *wait_glock =
95 container_of(wait, struct wait_glock_queue, wait);
96 struct lm_lockname *wait_name = wait_glock->name;
97 struct lm_lockname *wake_name = key;
99 if (wake_name->ln_sbd != wait_name->ln_sbd ||
100 wake_name->ln_number != wait_name->ln_number ||
101 wake_name->ln_type != wait_name->ln_type)
103 return autoremove_wake_function(wait, mode, sync, key);
106 static wait_queue_head_t *glock_waitqueue(struct lm_lockname *name)
108 u32 hash = jhash2((u32 *)name, ht_parms.key_len / 4, 0);
110 return glock_wait_table + hash_32(hash, GLOCK_WAIT_TABLE_BITS);
114 * wake_up_glock - Wake up waiters on a glock
117 static void wake_up_glock(struct gfs2_glock *gl)
119 wait_queue_head_t *wq = glock_waitqueue(&gl->gl_name);
121 if (waitqueue_active(wq))
122 __wake_up(wq, TASK_NORMAL, 1, &gl->gl_name);
125 static void gfs2_glock_dealloc(struct rcu_head *rcu)
127 struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
129 kfree(gl->gl_lksb.sb_lvbptr);
130 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
131 struct gfs2_glock_aspace *gla =
132 container_of(gl, struct gfs2_glock_aspace, glock);
133 kmem_cache_free(gfs2_glock_aspace_cachep, gla);
135 kmem_cache_free(gfs2_glock_cachep, gl);
139 * glock_blocked_by_withdraw - determine if we can still use a glock
142 * We need to allow some glocks to be enqueued, dequeued, promoted, and demoted
143 * when we're withdrawn. For example, to maintain metadata integrity, we should
144 * disallow the use of inode and rgrp glocks when withdrawn. Other glocks, like
145 * iopen or the transaction glocks may be safely used because none of their
146 * metadata goes through the journal. So in general, we should disallow all
147 * glocks that are journaled, and allow all the others. One exception is:
148 * we need to allow our active journal to be promoted and demoted so others
149 * may recover it and we can reacquire it when they're done.
151 static bool glock_blocked_by_withdraw(struct gfs2_glock *gl)
153 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
155 if (likely(!gfs2_withdrawn(sdp)))
157 if (gl->gl_ops->go_flags & GLOF_NONDISK)
159 if (!sdp->sd_jdesc ||
160 gl->gl_name.ln_number == sdp->sd_jdesc->jd_no_addr)
165 void gfs2_glock_free(struct gfs2_glock *gl)
167 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
169 gfs2_glock_assert_withdraw(gl, atomic_read(&gl->gl_revokes) == 0);
170 rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
173 call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
174 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
175 wake_up(&sdp->sd_glock_wait);
179 * gfs2_glock_hold() - increment reference count on glock
180 * @gl: The glock to hold
184 void gfs2_glock_hold(struct gfs2_glock *gl)
186 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
187 lockref_get(&gl->gl_lockref);
191 * demote_ok - Check to see if it's ok to unlock a glock
194 * Returns: 1 if it's ok
197 static int demote_ok(const struct gfs2_glock *gl)
199 const struct gfs2_glock_operations *glops = gl->gl_ops;
201 if (gl->gl_state == LM_ST_UNLOCKED)
204 * Note that demote_ok is used for the lru process of disposing of
205 * glocks. For this purpose, we don't care if the glock's holders
206 * have the HIF_MAY_DEMOTE flag set or not. If someone is using
207 * them, don't demote.
209 if (!list_empty(&gl->gl_holders))
211 if (glops->go_demote_ok)
212 return glops->go_demote_ok(gl);
217 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
219 if (!(gl->gl_ops->go_flags & GLOF_LRU))
222 spin_lock(&lru_lock);
224 list_move_tail(&gl->gl_lru, &lru_list);
226 if (!test_bit(GLF_LRU, &gl->gl_flags)) {
227 set_bit(GLF_LRU, &gl->gl_flags);
228 atomic_inc(&lru_count);
231 spin_unlock(&lru_lock);
234 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
236 if (!(gl->gl_ops->go_flags & GLOF_LRU))
239 spin_lock(&lru_lock);
240 if (test_bit(GLF_LRU, &gl->gl_flags)) {
241 list_del_init(&gl->gl_lru);
242 atomic_dec(&lru_count);
243 clear_bit(GLF_LRU, &gl->gl_flags);
245 spin_unlock(&lru_lock);
249 * Enqueue the glock on the work queue. Passes one glock reference on to the
252 static void __gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
253 if (!queue_delayed_work(glock_workqueue, &gl->gl_work, delay)) {
255 * We are holding the lockref spinlock, and the work was still
256 * queued above. The queued work (glock_work_func) takes that
257 * spinlock before dropping its glock reference(s), so it
258 * cannot have dropped them in the meantime.
260 GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2);
261 gl->gl_lockref.count--;
265 static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
266 spin_lock(&gl->gl_lockref.lock);
267 __gfs2_glock_queue_work(gl, delay);
268 spin_unlock(&gl->gl_lockref.lock);
271 static void __gfs2_glock_put(struct gfs2_glock *gl)
273 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
274 struct address_space *mapping = gfs2_glock2aspace(gl);
276 lockref_mark_dead(&gl->gl_lockref);
278 gfs2_glock_remove_from_lru(gl);
279 spin_unlock(&gl->gl_lockref.lock);
280 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
282 truncate_inode_pages_final(mapping);
283 if (!gfs2_withdrawn(sdp))
284 GLOCK_BUG_ON(gl, !mapping_empty(mapping));
286 trace_gfs2_glock_put(gl);
287 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
291 * Cause the glock to be put in work queue context.
293 void gfs2_glock_queue_put(struct gfs2_glock *gl)
295 gfs2_glock_queue_work(gl, 0);
299 * gfs2_glock_put() - Decrement reference count on glock
300 * @gl: The glock to put
304 void gfs2_glock_put(struct gfs2_glock *gl)
306 if (lockref_put_or_lock(&gl->gl_lockref))
309 __gfs2_glock_put(gl);
313 * may_grant - check if it's ok to grant a new lock
315 * @current_gh: One of the current holders of @gl
316 * @gh: The lock request which we wish to grant
318 * With our current compatibility rules, if a glock has one or more active
319 * holders (HIF_HOLDER flag set), any of those holders can be passed in as
320 * @current_gh; they are all the same as far as compatibility with the new @gh
323 * Returns true if it's ok to grant the lock.
326 static inline bool may_grant(struct gfs2_glock *gl,
327 struct gfs2_holder *current_gh,
328 struct gfs2_holder *gh)
331 GLOCK_BUG_ON(gl, !test_bit(HIF_HOLDER, ¤t_gh->gh_iflags));
333 switch(current_gh->gh_state) {
334 case LM_ST_EXCLUSIVE:
336 * Here we make a special exception to grant holders
337 * who agree to share the EX lock with other holders
338 * who also have the bit set. If the original holder
339 * has the LM_FLAG_NODE_SCOPE bit set, we grant more
340 * holders with the bit set.
342 return gh->gh_state == LM_ST_EXCLUSIVE &&
343 (current_gh->gh_flags & LM_FLAG_NODE_SCOPE) &&
344 (gh->gh_flags & LM_FLAG_NODE_SCOPE);
348 return gh->gh_state == current_gh->gh_state;
355 if (gl->gl_state == gh->gh_state)
357 if (gh->gh_flags & GL_EXACT)
359 if (gl->gl_state == LM_ST_EXCLUSIVE) {
360 return gh->gh_state == LM_ST_SHARED ||
361 gh->gh_state == LM_ST_DEFERRED;
363 if (gh->gh_flags & LM_FLAG_ANY)
364 return gl->gl_state != LM_ST_UNLOCKED;
368 static void gfs2_holder_wake(struct gfs2_holder *gh)
370 clear_bit(HIF_WAIT, &gh->gh_iflags);
371 smp_mb__after_atomic();
372 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
373 if (gh->gh_flags & GL_ASYNC) {
374 struct gfs2_sbd *sdp = gh->gh_gl->gl_name.ln_sbd;
376 wake_up(&sdp->sd_async_glock_wait);
381 * do_error - Something unexpected has happened during a lock request
383 * @ret: The status from the DLM
386 static void do_error(struct gfs2_glock *gl, const int ret)
388 struct gfs2_holder *gh, *tmp;
390 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
391 if (!test_bit(HIF_WAIT, &gh->gh_iflags))
393 if (ret & LM_OUT_ERROR)
395 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
396 gh->gh_error = GLR_TRYFAILED;
399 list_del_init(&gh->gh_list);
400 trace_gfs2_glock_queue(gh, 0);
401 gfs2_holder_wake(gh);
406 * demote_incompat_holders - demote incompatible demoteable holders
407 * @gl: the glock we want to promote
408 * @current_gh: the newly promoted holder
410 * We're passing the newly promoted holder in @current_gh, but actually, any of
411 * the strong holders would do.
413 static void demote_incompat_holders(struct gfs2_glock *gl,
414 struct gfs2_holder *current_gh)
416 struct gfs2_holder *gh, *tmp;
419 * Demote incompatible holders before we make ourselves eligible.
420 * (This holder may or may not allow auto-demoting, but we don't want
421 * to demote the new holder before it's even granted.)
423 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
425 * Since holders are at the front of the list, we stop when we
426 * find the first non-holder.
428 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
430 if (gh == current_gh)
432 if (test_bit(HIF_MAY_DEMOTE, &gh->gh_iflags) &&
433 !may_grant(gl, current_gh, gh)) {
435 * We should not recurse into do_promote because
436 * __gfs2_glock_dq only calls handle_callback,
437 * gfs2_glock_add_to_lru and __gfs2_glock_queue_work.
445 * find_first_holder - find the first "holder" gh
449 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
451 struct gfs2_holder *gh;
453 if (!list_empty(&gl->gl_holders)) {
454 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder,
456 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
463 * find_first_strong_holder - find the first non-demoteable holder
466 * Find the first holder that doesn't have the HIF_MAY_DEMOTE flag set.
468 static inline struct gfs2_holder *
469 find_first_strong_holder(struct gfs2_glock *gl)
471 struct gfs2_holder *gh;
473 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
474 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
476 if (!test_bit(HIF_MAY_DEMOTE, &gh->gh_iflags))
483 * gfs2_instantiate - Call the glops instantiate function
484 * @gh: The glock holder
486 * Returns: 0 if instantiate was successful, or error.
488 int gfs2_instantiate(struct gfs2_holder *gh)
490 struct gfs2_glock *gl = gh->gh_gl;
491 const struct gfs2_glock_operations *glops = gl->gl_ops;
495 if (!test_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags))
499 * Since we unlock the lockref lock, we set a flag to indicate
500 * instantiate is in progress.
502 if (test_and_set_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags)) {
503 wait_on_bit(&gl->gl_flags, GLF_INSTANTIATE_IN_PROG,
504 TASK_UNINTERRUPTIBLE);
506 * Here we just waited for a different instantiate to finish.
507 * But that may not have been successful, as when a process
508 * locks an inode glock _before_ it has an actual inode to
509 * instantiate into. So we check again. This process might
510 * have an inode to instantiate, so might be successful.
515 ret = glops->go_instantiate(gl);
517 clear_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags);
518 clear_and_wake_up_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags);
524 return glops->go_held(gh);
529 * do_promote - promote as many requests as possible on the current queue
532 * Returns: 1 if there is a blocked holder at the head of the list
535 static int do_promote(struct gfs2_glock *gl)
537 struct gfs2_holder *gh, *tmp, *current_gh;
538 bool incompat_holders_demoted = false;
540 current_gh = find_first_strong_holder(gl);
541 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
542 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
544 if (!may_grant(gl, current_gh, gh)) {
546 * If we get here, it means we may not grant this holder for
547 * some reason. If this holder is the head of the list, it
548 * means we have a blocked holder at the head, so return 1.
550 if (list_is_first(&gh->gh_list, &gl->gl_holders))
555 set_bit(HIF_HOLDER, &gh->gh_iflags);
556 trace_gfs2_promote(gh);
557 gfs2_holder_wake(gh);
558 if (!incompat_holders_demoted) {
560 demote_incompat_holders(gl, current_gh);
561 incompat_holders_demoted = true;
568 * find_first_waiter - find the first gh that's waiting for the glock
572 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
574 struct gfs2_holder *gh;
576 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
577 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
584 * state_change - record that the glock is now in a different state
586 * @new_state: the new state
589 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
593 held1 = (gl->gl_state != LM_ST_UNLOCKED);
594 held2 = (new_state != LM_ST_UNLOCKED);
596 if (held1 != held2) {
597 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
599 gl->gl_lockref.count++;
601 gl->gl_lockref.count--;
603 if (new_state != gl->gl_target)
604 /* shorten our minimum hold time */
605 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
607 gl->gl_state = new_state;
608 gl->gl_tchange = jiffies;
611 static void gfs2_set_demote(struct gfs2_glock *gl)
613 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
615 set_bit(GLF_DEMOTE, &gl->gl_flags);
617 wake_up(&sdp->sd_async_glock_wait);
620 static void gfs2_demote_wake(struct gfs2_glock *gl)
622 gl->gl_demote_state = LM_ST_EXCLUSIVE;
623 clear_bit(GLF_DEMOTE, &gl->gl_flags);
624 smp_mb__after_atomic();
625 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
629 * finish_xmote - The DLM has replied to one of our lock requests
631 * @ret: The status from the DLM
635 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
637 const struct gfs2_glock_operations *glops = gl->gl_ops;
638 struct gfs2_holder *gh;
639 unsigned state = ret & LM_OUT_ST_MASK;
641 spin_lock(&gl->gl_lockref.lock);
642 trace_gfs2_glock_state_change(gl, state);
643 state_change(gl, state);
644 gh = find_first_waiter(gl);
646 /* Demote to UN request arrived during demote to SH or DF */
647 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
648 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
649 gl->gl_target = LM_ST_UNLOCKED;
651 /* Check for state != intended state */
652 if (unlikely(state != gl->gl_target)) {
653 if (gh && (ret & LM_OUT_CANCELED))
654 gfs2_holder_wake(gh);
655 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
656 /* move to back of queue and try next entry */
657 if (ret & LM_OUT_CANCELED) {
658 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
659 list_move_tail(&gh->gh_list, &gl->gl_holders);
660 gh = find_first_waiter(gl);
661 gl->gl_target = gh->gh_state;
664 /* Some error or failed "try lock" - report it */
665 if ((ret & LM_OUT_ERROR) ||
666 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
667 gl->gl_target = gl->gl_state;
673 /* Unlocked due to conversion deadlock, try again */
676 do_xmote(gl, gh, gl->gl_target);
678 /* Conversion fails, unlock and try again */
681 do_xmote(gl, gh, LM_ST_UNLOCKED);
683 default: /* Everything else */
684 fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n",
685 gl->gl_target, state);
688 spin_unlock(&gl->gl_lockref.lock);
692 /* Fast path - we got what we asked for */
693 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
694 gfs2_demote_wake(gl);
695 if (state != LM_ST_UNLOCKED) {
696 if (glops->go_xmote_bh) {
699 spin_unlock(&gl->gl_lockref.lock);
700 rv = glops->go_xmote_bh(gl);
701 spin_lock(&gl->gl_lockref.lock);
710 clear_bit(GLF_LOCK, &gl->gl_flags);
711 spin_unlock(&gl->gl_lockref.lock);
714 static bool is_system_glock(struct gfs2_glock *gl)
716 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
717 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
719 if (gl == m_ip->i_gl)
725 * do_xmote - Calls the DLM to change the state of a lock
726 * @gl: The lock state
727 * @gh: The holder (only for promotes)
728 * @target: The target lock state
732 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
733 __releases(&gl->gl_lockref.lock)
734 __acquires(&gl->gl_lockref.lock)
736 const struct gfs2_glock_operations *glops = gl->gl_ops;
737 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
738 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
741 if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) &&
742 gh && !(gh->gh_flags & LM_FLAG_NOEXP))
744 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
746 GLOCK_BUG_ON(gl, gl->gl_state == target);
747 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
748 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
751 * If another process is already doing the invalidate, let that
752 * finish first. The glock state machine will get back to this
753 * holder again later.
755 if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
758 do_error(gl, 0); /* Fail queued try locks */
761 set_bit(GLF_BLOCKING, &gl->gl_flags);
762 if ((gl->gl_req == LM_ST_UNLOCKED) ||
763 (gl->gl_state == LM_ST_EXCLUSIVE) ||
764 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
765 clear_bit(GLF_BLOCKING, &gl->gl_flags);
766 spin_unlock(&gl->gl_lockref.lock);
767 if (glops->go_sync) {
768 ret = glops->go_sync(gl);
769 /* If we had a problem syncing (due to io errors or whatever,
770 * we should not invalidate the metadata or tell dlm to
771 * release the glock to other nodes.
774 if (cmpxchg(&sdp->sd_log_error, 0, ret)) {
775 fs_err(sdp, "Error %d syncing glock \n", ret);
776 gfs2_dump_glock(NULL, gl, true);
781 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) {
783 * The call to go_sync should have cleared out the ail list.
784 * If there are still items, we have a problem. We ought to
785 * withdraw, but we can't because the withdraw code also uses
786 * glocks. Warn about the error, dump the glock, then fall
787 * through and wait for logd to do the withdraw for us.
789 if ((atomic_read(&gl->gl_ail_count) != 0) &&
790 (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) {
791 gfs2_glock_assert_warn(gl,
792 !atomic_read(&gl->gl_ail_count));
793 gfs2_dump_glock(NULL, gl, true);
795 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
796 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
802 * Check for an error encountered since we called go_sync and go_inval.
803 * If so, we can't withdraw from the glock code because the withdraw
804 * code itself uses glocks (see function signal_our_withdraw) to
805 * change the mount to read-only. Most importantly, we must not call
806 * dlm to unlock the glock until the journal is in a known good state
807 * (after journal replay) otherwise other nodes may use the object
808 * (rgrp or dinode) and then later, journal replay will corrupt the
809 * file system. The best we can do here is wait for the logd daemon
810 * to see sd_log_error and withdraw, and in the meantime, requeue the
813 * We make a special exception for some system glocks, such as the
814 * system statfs inode glock, which needs to be granted before the
815 * gfs2_quotad daemon can exit, and that exit needs to finish before
816 * we can unmount the withdrawn file system.
818 * However, if we're just unlocking the lock (say, for unmount, when
819 * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
820 * then it's okay to tell dlm to unlock it.
822 if (unlikely(sdp->sd_log_error && !gfs2_withdrawn(sdp)))
823 gfs2_withdraw_delayed(sdp);
824 if (glock_blocked_by_withdraw(gl) &&
825 (target != LM_ST_UNLOCKED ||
826 test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags))) {
827 if (!is_system_glock(gl)) {
828 gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD);
831 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
835 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
837 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
838 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
839 target == LM_ST_UNLOCKED &&
840 test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
841 finish_xmote(gl, target);
842 gfs2_glock_queue_work(gl, 0);
844 fs_err(sdp, "lm_lock ret %d\n", ret);
845 GLOCK_BUG_ON(gl, !gfs2_withdrawn(sdp));
847 } else { /* lock_nolock */
848 finish_xmote(gl, target);
849 gfs2_glock_queue_work(gl, 0);
852 spin_lock(&gl->gl_lockref.lock);
856 * run_queue - do all outstanding tasks related to a glock
857 * @gl: The glock in question
858 * @nonblock: True if we must not block in run_queue
862 static void run_queue(struct gfs2_glock *gl, const int nonblock)
863 __releases(&gl->gl_lockref.lock)
864 __acquires(&gl->gl_lockref.lock)
866 struct gfs2_holder *gh = NULL;
868 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
871 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
873 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
874 gl->gl_demote_state != gl->gl_state) {
875 if (find_first_holder(gl))
879 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
880 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
881 gl->gl_target = gl->gl_demote_state;
883 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
884 gfs2_demote_wake(gl);
885 if (do_promote(gl) == 0)
887 gh = find_first_waiter(gl);
888 gl->gl_target = gh->gh_state;
889 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
890 do_error(gl, 0); /* Fail queued try locks */
892 do_xmote(gl, gh, gl->gl_target);
896 clear_bit(GLF_LOCK, &gl->gl_flags);
897 smp_mb__after_atomic();
898 gl->gl_lockref.count++;
899 __gfs2_glock_queue_work(gl, 0);
903 clear_bit(GLF_LOCK, &gl->gl_flags);
904 smp_mb__after_atomic();
908 void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation)
910 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
912 if (ri->ri_magic == 0)
913 ri->ri_magic = cpu_to_be32(GFS2_MAGIC);
914 if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC))
915 ri->ri_generation_deleted = cpu_to_be64(generation);
918 bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation)
920 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
922 if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC))
924 return generation <= be64_to_cpu(ri->ri_generation_deleted);
927 static void gfs2_glock_poke(struct gfs2_glock *gl)
929 int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP;
930 struct gfs2_holder gh;
933 __gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh, _RET_IP_);
934 error = gfs2_glock_nq(&gh);
937 gfs2_holder_uninit(&gh);
940 static bool gfs2_try_evict(struct gfs2_glock *gl)
942 struct gfs2_inode *ip;
943 bool evicted = false;
946 * If there is contention on the iopen glock and we have an inode, try
947 * to grab and release the inode so that it can be evicted. This will
948 * allow the remote node to go ahead and delete the inode without us
949 * having to do it, which will avoid rgrp glock thrashing.
951 * The remote node is likely still holding the corresponding inode
952 * glock, so it will run before we get to verify that the delete has
955 spin_lock(&gl->gl_lockref.lock);
957 if (ip && !igrab(&ip->i_inode))
959 spin_unlock(&gl->gl_lockref.lock);
961 struct gfs2_glock *inode_gl = NULL;
963 gl->gl_no_formal_ino = ip->i_no_formal_ino;
964 set_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
965 d_prune_aliases(&ip->i_inode);
968 /* If the inode was evicted, gl->gl_object will now be NULL. */
969 spin_lock(&gl->gl_lockref.lock);
973 lockref_get(&inode_gl->gl_lockref);
974 clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
976 spin_unlock(&gl->gl_lockref.lock);
978 gfs2_glock_poke(inode_gl);
979 gfs2_glock_put(inode_gl);
986 static void delete_work_func(struct work_struct *work)
988 struct delayed_work *dwork = to_delayed_work(work);
989 struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete);
990 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
992 u64 no_addr = gl->gl_name.ln_number;
994 spin_lock(&gl->gl_lockref.lock);
995 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
996 spin_unlock(&gl->gl_lockref.lock);
998 if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
1000 * If we can evict the inode, give the remote node trying to
1001 * delete the inode some time before verifying that the delete
1002 * has happened. Otherwise, if we cause contention on the inode glock
1003 * immediately, the remote node will think that we still have
1004 * the inode in use, and so it will give up waiting.
1006 * If we can't evict the inode, signal to the remote node that
1007 * the inode is still in use. We'll later try to delete the
1008 * inode locally in gfs2_evict_inode.
1010 * FIXME: We only need to verify that the remote node has
1011 * deleted the inode because nodes before this remote delete
1012 * rework won't cooperate. At a later time, when we no longer
1013 * care about compatibility with such nodes, we can skip this
1016 if (gfs2_try_evict(gl)) {
1017 if (gfs2_queue_delete_work(gl, 5 * HZ))
1023 inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino,
1024 GFS2_BLKST_UNLINKED);
1025 if (!IS_ERR_OR_NULL(inode)) {
1026 d_prune_aliases(inode);
1033 static void glock_work_func(struct work_struct *work)
1035 unsigned long delay = 0;
1036 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
1037 unsigned int drop_refs = 1;
1039 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
1040 finish_xmote(gl, gl->gl_reply);
1043 spin_lock(&gl->gl_lockref.lock);
1044 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1045 gl->gl_state != LM_ST_UNLOCKED &&
1046 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
1047 unsigned long holdtime, now = jiffies;
1049 holdtime = gl->gl_tchange + gl->gl_hold_time;
1050 if (time_before(now, holdtime))
1051 delay = holdtime - now;
1054 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1055 gfs2_set_demote(gl);
1060 /* Keep one glock reference for the work we requeue. */
1062 if (gl->gl_name.ln_type != LM_TYPE_INODE)
1064 __gfs2_glock_queue_work(gl, delay);
1068 * Drop the remaining glock references manually here. (Mind that
1069 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
1072 gl->gl_lockref.count -= drop_refs;
1073 if (!gl->gl_lockref.count) {
1074 __gfs2_glock_put(gl);
1077 spin_unlock(&gl->gl_lockref.lock);
1080 static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
1081 struct gfs2_glock *new)
1083 struct wait_glock_queue wait;
1084 wait_queue_head_t *wq = glock_waitqueue(name);
1085 struct gfs2_glock *gl;
1088 init_wait(&wait.wait);
1089 wait.wait.func = glock_wake_function;
1092 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1095 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
1096 &new->gl_node, ht_parms);
1100 gl = rhashtable_lookup_fast(&gl_hash_table,
1103 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
1110 finish_wait(wq, &wait.wait);
1115 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
1116 * @sdp: The GFS2 superblock
1117 * @number: the lock number
1118 * @glops: The glock_operations to use
1119 * @create: If 0, don't create the glock if it doesn't exist
1120 * @glp: the glock is returned here
1122 * This does not lock a glock, just finds/creates structures for one.
1127 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
1128 const struct gfs2_glock_operations *glops, int create,
1129 struct gfs2_glock **glp)
1131 struct super_block *s = sdp->sd_vfs;
1132 struct lm_lockname name = { .ln_number = number,
1133 .ln_type = glops->go_type,
1135 struct gfs2_glock *gl, *tmp;
1136 struct address_space *mapping;
1139 gl = find_insert_glock(&name, NULL);
1147 if (glops->go_flags & GLOF_ASPACE) {
1148 struct gfs2_glock_aspace *gla =
1149 kmem_cache_alloc(gfs2_glock_aspace_cachep, GFP_NOFS);
1154 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_NOFS);
1158 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
1161 if (glops->go_flags & GLOF_LVB) {
1162 gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1163 if (!gl->gl_lksb.sb_lvbptr) {
1164 gfs2_glock_dealloc(&gl->gl_rcu);
1169 atomic_inc(&sdp->sd_glock_disposal);
1170 gl->gl_node.next = NULL;
1171 gl->gl_flags = glops->go_instantiate ? BIT(GLF_INSTANTIATE_NEEDED) : 0;
1173 lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass);
1174 gl->gl_lockref.count = 1;
1175 gl->gl_state = LM_ST_UNLOCKED;
1176 gl->gl_target = LM_ST_UNLOCKED;
1177 gl->gl_demote_state = LM_ST_EXCLUSIVE;
1180 /* We use the global stats to estimate the initial per-glock stats */
1181 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
1183 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
1184 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
1185 gl->gl_tchange = jiffies;
1186 gl->gl_object = NULL;
1187 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
1188 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
1189 if (gl->gl_name.ln_type == LM_TYPE_IOPEN)
1190 INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func);
1192 mapping = gfs2_glock2aspace(gl);
1194 mapping->a_ops = &gfs2_meta_aops;
1195 mapping->host = s->s_bdev->bd_inode;
1197 mapping_set_gfp_mask(mapping, GFP_NOFS);
1198 mapping->private_data = NULL;
1199 mapping->writeback_index = 0;
1202 tmp = find_insert_glock(&name, gl);
1214 gfs2_glock_dealloc(&gl->gl_rcu);
1215 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
1216 wake_up(&sdp->sd_glock_wait);
1223 * __gfs2_holder_init - initialize a struct gfs2_holder in the default way
1225 * @state: the state we're requesting
1226 * @flags: the modifier flags
1227 * @gh: the holder structure
1231 void __gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
1232 struct gfs2_holder *gh, unsigned long ip)
1234 INIT_LIST_HEAD(&gh->gh_list);
1237 gh->gh_owner_pid = get_pid(task_pid(current));
1238 gh->gh_state = state;
1239 gh->gh_flags = flags;
1241 gfs2_glock_hold(gl);
1245 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1246 * @state: the state we're requesting
1247 * @flags: the modifier flags
1248 * @gh: the holder structure
1250 * Don't mess with the glock.
1254 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
1256 gh->gh_state = state;
1257 gh->gh_flags = flags;
1259 gh->gh_ip = _RET_IP_;
1260 put_pid(gh->gh_owner_pid);
1261 gh->gh_owner_pid = get_pid(task_pid(current));
1265 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1266 * @gh: the holder structure
1270 void gfs2_holder_uninit(struct gfs2_holder *gh)
1272 put_pid(gh->gh_owner_pid);
1273 gfs2_glock_put(gh->gh_gl);
1274 gfs2_holder_mark_uninitialized(gh);
1278 static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
1279 unsigned long start_time)
1281 /* Have we waited longer that a second? */
1282 if (time_after(jiffies, start_time + HZ)) {
1283 /* Lengthen the minimum hold time. */
1284 gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
1290 * gfs2_glock_holder_ready - holder is ready and its error code can be collected
1291 * @gh: the glock holder
1293 * Called when a glock holder no longer needs to be waited for because it is
1294 * now either held (HIF_HOLDER set; gh_error == 0), or acquiring the lock has
1295 * failed (gh_error != 0).
1298 int gfs2_glock_holder_ready(struct gfs2_holder *gh)
1300 if (gh->gh_error || (gh->gh_flags & GL_SKIP))
1301 return gh->gh_error;
1302 gh->gh_error = gfs2_instantiate(gh);
1305 return gh->gh_error;
1309 * gfs2_glock_wait - wait on a glock acquisition
1310 * @gh: the glock holder
1312 * Returns: 0 on success
1315 int gfs2_glock_wait(struct gfs2_holder *gh)
1317 unsigned long start_time = jiffies;
1320 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1321 gfs2_glock_update_hold_time(gh->gh_gl, start_time);
1322 return gfs2_glock_holder_ready(gh);
1325 static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
1329 for (i = 0; i < num_gh; i++)
1330 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
1336 * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1337 * @num_gh: the number of holders in the array
1338 * @ghs: the glock holder array
1340 * Returns: 0 on success, meaning all glocks have been granted and are held.
1341 * -ESTALE if the request timed out, meaning all glocks were released,
1342 * and the caller should retry the operation.
1345 int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
1347 struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
1348 int i, ret = 0, timeout = 0;
1349 unsigned long start_time = jiffies;
1353 * Total up the (minimum hold time * 2) of all glocks and use that to
1354 * determine the max amount of time we should wait.
1356 for (i = 0; i < num_gh; i++)
1357 timeout += ghs[i].gh_gl->gl_hold_time << 1;
1359 if (!wait_event_timeout(sdp->sd_async_glock_wait,
1360 !glocks_pending(num_gh, ghs), timeout)) {
1361 ret = -ESTALE; /* request timed out. */
1365 for (i = 0; i < num_gh; i++) {
1366 struct gfs2_holder *gh = &ghs[i];
1369 if (test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1370 gfs2_glock_update_hold_time(gh->gh_gl,
1373 ret2 = gfs2_glock_holder_ready(gh);
1380 for (i = 0; i < num_gh; i++) {
1381 struct gfs2_holder *gh = &ghs[i];
1390 * handle_callback - process a demote request
1392 * @state: the state the caller wants us to change to
1393 * @delay: zero to demote immediately; otherwise pending demote
1394 * @remote: true if this came from a different cluster node
1396 * There are only two requests that we are going to see in actual
1397 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1400 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
1401 unsigned long delay, bool remote)
1404 set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1406 gfs2_set_demote(gl);
1407 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
1408 gl->gl_demote_state = state;
1409 gl->gl_demote_time = jiffies;
1410 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
1411 gl->gl_demote_state != state) {
1412 gl->gl_demote_state = LM_ST_UNLOCKED;
1414 if (gl->gl_ops->go_callback)
1415 gl->gl_ops->go_callback(gl, remote);
1416 trace_gfs2_demote_rq(gl, remote);
1419 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
1421 struct va_format vaf;
1424 va_start(args, fmt);
1427 seq_vprintf(seq, fmt, args);
1432 pr_err("%pV", &vaf);
1439 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1440 * @gh: the holder structure to add
1442 * Eventually we should move the recursive locking trap to a
1443 * debugging option or something like that. This is the fast
1444 * path and needs to have the minimum number of distractions.
1448 static inline void add_to_queue(struct gfs2_holder *gh)
1449 __releases(&gl->gl_lockref.lock)
1450 __acquires(&gl->gl_lockref.lock)
1452 struct gfs2_glock *gl = gh->gh_gl;
1453 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1454 struct list_head *insert_pt = NULL;
1455 struct gfs2_holder *gh2;
1458 GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1459 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1460 GLOCK_BUG_ON(gl, true);
1462 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1463 if (test_bit(GLF_LOCK, &gl->gl_flags)) {
1464 struct gfs2_holder *current_gh;
1466 current_gh = find_first_strong_holder(gl);
1467 try_futile = !may_grant(gl, current_gh, gh);
1469 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1473 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1474 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
1475 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK) &&
1476 !test_bit(HIF_MAY_DEMOTE, &gh2->gh_iflags)))
1477 goto trap_recursive;
1479 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1481 gh->gh_error = GLR_TRYFAILED;
1482 gfs2_holder_wake(gh);
1485 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1487 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1488 insert_pt = &gh2->gh_list;
1490 trace_gfs2_glock_queue(gh, 1);
1491 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1492 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1493 if (likely(insert_pt == NULL)) {
1494 list_add_tail(&gh->gh_list, &gl->gl_holders);
1495 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1499 list_add_tail(&gh->gh_list, insert_pt);
1501 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
1502 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1503 spin_unlock(&gl->gl_lockref.lock);
1504 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1505 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1506 spin_lock(&gl->gl_lockref.lock);
1511 fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1512 fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1513 fs_err(sdp, "lock type: %d req lock state : %d\n",
1514 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1515 fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1516 fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1517 fs_err(sdp, "lock type: %d req lock state : %d\n",
1518 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1519 gfs2_dump_glock(NULL, gl, true);
1524 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1525 * @gh: the holder structure
1527 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1529 * Returns: 0, GLR_TRYFAILED, or errno on failure
1532 int gfs2_glock_nq(struct gfs2_holder *gh)
1534 struct gfs2_glock *gl = gh->gh_gl;
1537 if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP))
1540 if (test_bit(GLF_LRU, &gl->gl_flags))
1541 gfs2_glock_remove_from_lru(gl);
1544 spin_lock(&gl->gl_lockref.lock);
1546 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1547 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1548 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1549 gl->gl_lockref.count++;
1550 __gfs2_glock_queue_work(gl, 0);
1553 spin_unlock(&gl->gl_lockref.lock);
1555 if (!(gh->gh_flags & GL_ASYNC))
1556 error = gfs2_glock_wait(gh);
1562 * gfs2_glock_poll - poll to see if an async request has been completed
1565 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1568 int gfs2_glock_poll(struct gfs2_holder *gh)
1570 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1573 static inline bool needs_demote(struct gfs2_glock *gl)
1575 return (test_bit(GLF_DEMOTE, &gl->gl_flags) ||
1576 test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags));
1579 static void __gfs2_glock_dq(struct gfs2_holder *gh)
1581 struct gfs2_glock *gl = gh->gh_gl;
1582 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1587 * This while loop is similar to function demote_incompat_holders:
1588 * If the glock is due to be demoted (which may be from another node
1589 * or even if this holder is GL_NOCACHE), the weak holders are
1590 * demoted as well, allowing the glock to be demoted.
1594 * If we're in the process of file system withdraw, we cannot
1595 * just dequeue any glocks until our journal is recovered, lest
1596 * we introduce file system corruption. We need two exceptions
1597 * to this rule: We need to allow unlocking of nondisk glocks
1598 * and the glock for our own journal that needs recovery.
1600 if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) &&
1601 glock_blocked_by_withdraw(gl) &&
1602 gh->gh_gl != sdp->sd_jinode_gl) {
1603 sdp->sd_glock_dqs_held++;
1604 spin_unlock(&gl->gl_lockref.lock);
1606 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
1607 TASK_UNINTERRUPTIBLE);
1608 spin_lock(&gl->gl_lockref.lock);
1612 * This holder should not be cached, so mark it for demote.
1613 * Note: this should be done before the check for needs_demote
1616 if (gh->gh_flags & GL_NOCACHE)
1617 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1619 list_del_init(&gh->gh_list);
1620 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1621 trace_gfs2_glock_queue(gh, 0);
1624 * If there hasn't been a demote request we are done.
1625 * (Let the remaining holders, if any, keep holding it.)
1627 if (!needs_demote(gl)) {
1628 if (list_empty(&gl->gl_holders))
1633 * If we have another strong holder (we cannot auto-demote)
1634 * we are done. It keeps holding it until it is done.
1636 if (find_first_strong_holder(gl))
1640 * If we have a weak holder at the head of the list, it
1641 * (and all others like it) must be auto-demoted. If there
1642 * are no more weak holders, we exit the while loop.
1644 gh = find_first_holder(gl);
1647 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1648 gfs2_glock_add_to_lru(gl);
1650 if (unlikely(!fast_path)) {
1651 gl->gl_lockref.count++;
1652 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1653 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1654 gl->gl_name.ln_type == LM_TYPE_INODE)
1655 delay = gl->gl_hold_time;
1656 __gfs2_glock_queue_work(gl, delay);
1661 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1662 * @gh: the glock holder
1665 void gfs2_glock_dq(struct gfs2_holder *gh)
1667 struct gfs2_glock *gl = gh->gh_gl;
1669 spin_lock(&gl->gl_lockref.lock);
1670 if (list_is_first(&gh->gh_list, &gl->gl_holders) &&
1671 !test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1672 spin_unlock(&gl->gl_lockref.lock);
1673 gl->gl_name.ln_sbd->sd_lockstruct.ls_ops->lm_cancel(gl);
1674 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1675 spin_lock(&gl->gl_lockref.lock);
1678 __gfs2_glock_dq(gh);
1679 spin_unlock(&gl->gl_lockref.lock);
1682 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1684 struct gfs2_glock *gl = gh->gh_gl;
1687 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1691 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1692 * @gh: the holder structure
1696 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1699 gfs2_holder_uninit(gh);
1703 * gfs2_glock_nq_num - acquire a glock based on lock number
1704 * @sdp: the filesystem
1705 * @number: the lock number
1706 * @glops: the glock operations for the type of glock
1707 * @state: the state to acquire the glock in
1708 * @flags: modifier flags for the acquisition
1709 * @gh: the struct gfs2_holder
1714 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1715 const struct gfs2_glock_operations *glops,
1716 unsigned int state, u16 flags, struct gfs2_holder *gh)
1718 struct gfs2_glock *gl;
1721 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1723 error = gfs2_glock_nq_init(gl, state, flags, gh);
1731 * glock_compare - Compare two struct gfs2_glock structures for sorting
1732 * @arg_a: the first structure
1733 * @arg_b: the second structure
1737 static int glock_compare(const void *arg_a, const void *arg_b)
1739 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1740 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1741 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1742 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1744 if (a->ln_number > b->ln_number)
1746 if (a->ln_number < b->ln_number)
1748 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1753 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1754 * @num_gh: the number of structures
1755 * @ghs: an array of struct gfs2_holder structures
1756 * @p: placeholder for the holder structure to pass back
1758 * Returns: 0 on success (all glocks acquired),
1759 * errno on failure (no glocks acquired)
1762 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1763 struct gfs2_holder **p)
1768 for (x = 0; x < num_gh; x++)
1771 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1773 for (x = 0; x < num_gh; x++) {
1774 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1776 error = gfs2_glock_nq(p[x]);
1779 gfs2_glock_dq(p[x]);
1788 * gfs2_glock_nq_m - acquire multiple glocks
1789 * @num_gh: the number of structures
1790 * @ghs: an array of struct gfs2_holder structures
1793 * Returns: 0 on success (all glocks acquired),
1794 * errno on failure (no glocks acquired)
1797 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1799 struct gfs2_holder *tmp[4];
1800 struct gfs2_holder **pph = tmp;
1807 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1808 return gfs2_glock_nq(ghs);
1812 pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *),
1818 error = nq_m_sync(num_gh, ghs, pph);
1827 * gfs2_glock_dq_m - release multiple glocks
1828 * @num_gh: the number of structures
1829 * @ghs: an array of struct gfs2_holder structures
1833 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1836 gfs2_glock_dq(&ghs[num_gh]);
1839 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1841 unsigned long delay = 0;
1842 unsigned long holdtime;
1843 unsigned long now = jiffies;
1845 gfs2_glock_hold(gl);
1846 spin_lock(&gl->gl_lockref.lock);
1847 holdtime = gl->gl_tchange + gl->gl_hold_time;
1848 if (!list_empty(&gl->gl_holders) &&
1849 gl->gl_name.ln_type == LM_TYPE_INODE) {
1850 if (time_before(now, holdtime))
1851 delay = holdtime - now;
1852 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1853 delay = gl->gl_hold_time;
1856 * Note 1: We cannot call demote_incompat_holders from handle_callback
1857 * or gfs2_set_demote due to recursion problems like: gfs2_glock_dq ->
1858 * handle_callback -> demote_incompat_holders -> gfs2_glock_dq
1859 * Plus, we only want to demote the holders if the request comes from
1860 * a remote cluster node because local holder conflicts are resolved
1863 * Note 2: if a remote node wants this glock in EX mode, lock_dlm will
1864 * request that we set our state to UNLOCKED. Here we mock up a holder
1865 * to make it look like someone wants the lock EX locally. Any SH
1866 * and DF requests should be able to share the lock without demoting.
1868 * Note 3: We only want to demote the demoteable holders when there
1869 * are no more strong holders. The demoteable holders might as well
1870 * keep the glock until the last strong holder is done with it.
1872 if (!find_first_strong_holder(gl)) {
1873 struct gfs2_holder mock_gh = {
1875 .gh_state = (state == LM_ST_UNLOCKED) ?
1876 LM_ST_EXCLUSIVE : state,
1877 .gh_iflags = BIT(HIF_HOLDER)
1880 demote_incompat_holders(gl, &mock_gh);
1882 handle_callback(gl, state, delay, true);
1883 __gfs2_glock_queue_work(gl, delay);
1884 spin_unlock(&gl->gl_lockref.lock);
1888 * gfs2_should_freeze - Figure out if glock should be frozen
1889 * @gl: The glock in question
1891 * Glocks are not frozen if (a) the result of the dlm operation is
1892 * an error, (b) the locking operation was an unlock operation or
1893 * (c) if there is a "noexp" flagged request anywhere in the queue
1895 * Returns: 1 if freezing should occur, 0 otherwise
1898 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1900 const struct gfs2_holder *gh;
1902 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1904 if (gl->gl_target == LM_ST_UNLOCKED)
1907 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1908 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1910 if (LM_FLAG_NOEXP & gh->gh_flags)
1918 * gfs2_glock_complete - Callback used by locking
1919 * @gl: Pointer to the glock
1920 * @ret: The return value from the dlm
1922 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1923 * to use a bitfield shared with other glock state fields.
1926 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1928 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1930 spin_lock(&gl->gl_lockref.lock);
1933 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1934 if (gfs2_should_freeze(gl)) {
1935 set_bit(GLF_FROZEN, &gl->gl_flags);
1936 spin_unlock(&gl->gl_lockref.lock);
1941 gl->gl_lockref.count++;
1942 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1943 __gfs2_glock_queue_work(gl, 0);
1944 spin_unlock(&gl->gl_lockref.lock);
1947 static int glock_cmp(void *priv, const struct list_head *a,
1948 const struct list_head *b)
1950 struct gfs2_glock *gla, *glb;
1952 gla = list_entry(a, struct gfs2_glock, gl_lru);
1953 glb = list_entry(b, struct gfs2_glock, gl_lru);
1955 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1957 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1964 * gfs2_dispose_glock_lru - Demote a list of glocks
1965 * @list: The list to dispose of
1967 * Disposing of glocks may involve disk accesses, so that here we sort
1968 * the glocks by number (i.e. disk location of the inodes) so that if
1969 * there are any such accesses, they'll be sent in order (mostly).
1971 * Must be called under the lru_lock, but may drop and retake this
1972 * lock. While the lru_lock is dropped, entries may vanish from the
1973 * list, but no new entries will appear on the list (since it is
1977 static void gfs2_dispose_glock_lru(struct list_head *list)
1978 __releases(&lru_lock)
1979 __acquires(&lru_lock)
1981 struct gfs2_glock *gl;
1983 list_sort(NULL, list, glock_cmp);
1985 while(!list_empty(list)) {
1986 gl = list_first_entry(list, struct gfs2_glock, gl_lru);
1987 list_del_init(&gl->gl_lru);
1988 clear_bit(GLF_LRU, &gl->gl_flags);
1989 if (!spin_trylock(&gl->gl_lockref.lock)) {
1991 list_add(&gl->gl_lru, &lru_list);
1992 set_bit(GLF_LRU, &gl->gl_flags);
1993 atomic_inc(&lru_count);
1996 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1997 spin_unlock(&gl->gl_lockref.lock);
1998 goto add_back_to_lru;
2000 gl->gl_lockref.count++;
2002 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2003 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
2004 __gfs2_glock_queue_work(gl, 0);
2005 spin_unlock(&gl->gl_lockref.lock);
2006 cond_resched_lock(&lru_lock);
2011 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
2012 * @nr: The number of entries to scan
2014 * This function selects the entries on the LRU which are able to
2015 * be demoted, and then kicks off the process by calling
2016 * gfs2_dispose_glock_lru() above.
2019 static long gfs2_scan_glock_lru(int nr)
2021 struct gfs2_glock *gl;
2026 spin_lock(&lru_lock);
2027 while ((nr-- >= 0) && !list_empty(&lru_list)) {
2028 gl = list_first_entry(&lru_list, struct gfs2_glock, gl_lru);
2030 /* Test for being demotable */
2031 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
2032 list_move(&gl->gl_lru, &dispose);
2033 atomic_dec(&lru_count);
2038 list_move(&gl->gl_lru, &skipped);
2040 list_splice(&skipped, &lru_list);
2041 if (!list_empty(&dispose))
2042 gfs2_dispose_glock_lru(&dispose);
2043 spin_unlock(&lru_lock);
2048 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
2049 struct shrink_control *sc)
2051 if (!(sc->gfp_mask & __GFP_FS))
2053 return gfs2_scan_glock_lru(sc->nr_to_scan);
2056 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
2057 struct shrink_control *sc)
2059 return vfs_pressure_ratio(atomic_read(&lru_count));
2062 static struct shrinker glock_shrinker = {
2063 .seeks = DEFAULT_SEEKS,
2064 .count_objects = gfs2_glock_shrink_count,
2065 .scan_objects = gfs2_glock_shrink_scan,
2069 * glock_hash_walk - Call a function for glock in a hash bucket
2070 * @examiner: the function
2071 * @sdp: the filesystem
2073 * Note that the function can be called multiple times on the same
2074 * object. So the user must ensure that the function can cope with
2078 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
2080 struct gfs2_glock *gl;
2081 struct rhashtable_iter iter;
2083 rhashtable_walk_enter(&gl_hash_table, &iter);
2086 rhashtable_walk_start(&iter);
2088 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl)) {
2089 if (gl->gl_name.ln_sbd == sdp)
2093 rhashtable_walk_stop(&iter);
2094 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
2096 rhashtable_walk_exit(&iter);
2099 bool gfs2_queue_delete_work(struct gfs2_glock *gl, unsigned long delay)
2103 spin_lock(&gl->gl_lockref.lock);
2104 queued = queue_delayed_work(gfs2_delete_workqueue,
2105 &gl->gl_delete, delay);
2107 set_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2108 spin_unlock(&gl->gl_lockref.lock);
2112 void gfs2_cancel_delete_work(struct gfs2_glock *gl)
2114 if (cancel_delayed_work(&gl->gl_delete)) {
2115 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2120 bool gfs2_delete_work_queued(const struct gfs2_glock *gl)
2122 return test_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2125 static void flush_delete_work(struct gfs2_glock *gl)
2127 if (gl->gl_name.ln_type == LM_TYPE_IOPEN) {
2128 if (cancel_delayed_work(&gl->gl_delete)) {
2129 queue_delayed_work(gfs2_delete_workqueue,
2135 void gfs2_flush_delete_work(struct gfs2_sbd *sdp)
2137 glock_hash_walk(flush_delete_work, sdp);
2138 flush_workqueue(gfs2_delete_workqueue);
2142 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
2143 * @gl: The glock to thaw
2147 static void thaw_glock(struct gfs2_glock *gl)
2149 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
2151 if (!lockref_get_not_dead(&gl->gl_lockref))
2153 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
2154 gfs2_glock_queue_work(gl, 0);
2158 * clear_glock - look at a glock and see if we can free it from glock cache
2159 * @gl: the glock to look at
2163 static void clear_glock(struct gfs2_glock *gl)
2165 gfs2_glock_remove_from_lru(gl);
2167 spin_lock(&gl->gl_lockref.lock);
2168 if (!__lockref_is_dead(&gl->gl_lockref)) {
2169 gl->gl_lockref.count++;
2170 if (gl->gl_state != LM_ST_UNLOCKED)
2171 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2172 __gfs2_glock_queue_work(gl, 0);
2174 spin_unlock(&gl->gl_lockref.lock);
2178 * gfs2_glock_thaw - Thaw any frozen glocks
2179 * @sdp: The super block
2183 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
2185 glock_hash_walk(thaw_glock, sdp);
2188 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2190 spin_lock(&gl->gl_lockref.lock);
2191 gfs2_dump_glock(seq, gl, fsid);
2192 spin_unlock(&gl->gl_lockref.lock);
2195 static void dump_glock_func(struct gfs2_glock *gl)
2197 dump_glock(NULL, gl, true);
2201 * gfs2_gl_hash_clear - Empty out the glock hash table
2202 * @sdp: the filesystem
2204 * Called when unmounting the filesystem.
2207 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
2209 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
2210 flush_workqueue(glock_workqueue);
2211 glock_hash_walk(clear_glock, sdp);
2212 flush_workqueue(glock_workqueue);
2213 wait_event_timeout(sdp->sd_glock_wait,
2214 atomic_read(&sdp->sd_glock_disposal) == 0,
2216 glock_hash_walk(dump_glock_func, sdp);
2219 static const char *state2str(unsigned state)
2222 case LM_ST_UNLOCKED:
2226 case LM_ST_DEFERRED:
2228 case LM_ST_EXCLUSIVE:
2234 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
2237 if (flags & LM_FLAG_TRY)
2239 if (flags & LM_FLAG_TRY_1CB)
2241 if (flags & LM_FLAG_NOEXP)
2243 if (flags & LM_FLAG_ANY)
2245 if (flags & LM_FLAG_PRIORITY)
2247 if (flags & LM_FLAG_NODE_SCOPE)
2249 if (flags & GL_ASYNC)
2251 if (flags & GL_EXACT)
2253 if (flags & GL_NOCACHE)
2255 if (test_bit(HIF_HOLDER, &iflags))
2257 if (test_bit(HIF_WAIT, &iflags))
2259 if (test_bit(HIF_MAY_DEMOTE, &iflags))
2261 if (flags & GL_SKIP)
2268 * dump_holder - print information about a glock holder
2269 * @seq: the seq_file struct
2270 * @gh: the glock holder
2271 * @fs_id_buf: pointer to file system id (if requested)
2275 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
2276 const char *fs_id_buf)
2278 struct task_struct *gh_owner = NULL;
2282 if (gh->gh_owner_pid)
2283 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
2284 gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2285 fs_id_buf, state2str(gh->gh_state),
2286 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
2288 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
2289 gh_owner ? gh_owner->comm : "(ended)",
2294 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
2296 const unsigned long *gflags = &gl->gl_flags;
2299 if (test_bit(GLF_LOCK, gflags))
2301 if (test_bit(GLF_DEMOTE, gflags))
2303 if (test_bit(GLF_PENDING_DEMOTE, gflags))
2305 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
2307 if (test_bit(GLF_DIRTY, gflags))
2309 if (test_bit(GLF_LFLUSH, gflags))
2311 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
2313 if (test_bit(GLF_REPLY_PENDING, gflags))
2315 if (test_bit(GLF_INITIAL, gflags))
2317 if (test_bit(GLF_FROZEN, gflags))
2319 if (!list_empty(&gl->gl_holders))
2321 if (test_bit(GLF_LRU, gflags))
2325 if (test_bit(GLF_BLOCKING, gflags))
2327 if (test_bit(GLF_PENDING_DELETE, gflags))
2329 if (test_bit(GLF_FREEING, gflags))
2331 if (test_bit(GLF_INSTANTIATE_NEEDED, gflags))
2333 if (test_bit(GLF_INSTANTIATE_IN_PROG, gflags))
2340 * gfs2_dump_glock - print information about a glock
2341 * @seq: The seq_file struct
2343 * @fsid: If true, also dump the file system id
2345 * The file format is as follows:
2346 * One line per object, capital letters are used to indicate objects
2347 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2348 * other objects are indented by a single space and follow the glock to
2349 * which they are related. Fields are indicated by lower case letters
2350 * followed by a colon and the field value, except for strings which are in
2351 * [] so that its possible to see if they are composed of spaces for
2352 * example. The field's are n = number (id of the object), f = flags,
2353 * t = type, s = state, r = refcount, e = error, p = pid.
2357 void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2359 const struct gfs2_glock_operations *glops = gl->gl_ops;
2360 unsigned long long dtime;
2361 const struct gfs2_holder *gh;
2362 char gflags_buf[32];
2363 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2364 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2365 unsigned long nrpages = 0;
2367 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
2368 struct address_space *mapping = gfs2_glock2aspace(gl);
2370 nrpages = mapping->nrpages;
2372 memset(fs_id_buf, 0, sizeof(fs_id_buf));
2373 if (fsid && sdp) /* safety precaution */
2374 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2375 dtime = jiffies - gl->gl_demote_time;
2376 dtime *= 1000000/HZ; /* demote time in uSec */
2377 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
2379 gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2380 "v:%d r:%d m:%ld p:%lu\n",
2381 fs_id_buf, state2str(gl->gl_state),
2382 gl->gl_name.ln_type,
2383 (unsigned long long)gl->gl_name.ln_number,
2384 gflags2str(gflags_buf, gl),
2385 state2str(gl->gl_target),
2386 state2str(gl->gl_demote_state), dtime,
2387 atomic_read(&gl->gl_ail_count),
2388 atomic_read(&gl->gl_revokes),
2389 (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages);
2391 list_for_each_entry(gh, &gl->gl_holders, gh_list)
2392 dump_holder(seq, gh, fs_id_buf);
2394 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
2395 glops->go_dump(seq, gl, fs_id_buf);
2398 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
2400 struct gfs2_glock *gl = iter_ptr;
2402 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2403 gl->gl_name.ln_type,
2404 (unsigned long long)gl->gl_name.ln_number,
2405 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
2406 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
2407 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
2408 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
2409 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
2410 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
2411 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
2412 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
2416 static const char *gfs2_gltype[] = {
2430 static const char *gfs2_stype[] = {
2431 [GFS2_LKS_SRTT] = "srtt",
2432 [GFS2_LKS_SRTTVAR] = "srttvar",
2433 [GFS2_LKS_SRTTB] = "srttb",
2434 [GFS2_LKS_SRTTVARB] = "srttvarb",
2435 [GFS2_LKS_SIRT] = "sirt",
2436 [GFS2_LKS_SIRTVAR] = "sirtvar",
2437 [GFS2_LKS_DCOUNT] = "dlm",
2438 [GFS2_LKS_QCOUNT] = "queue",
2441 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2443 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
2445 struct gfs2_sbd *sdp = seq->private;
2446 loff_t pos = *(loff_t *)iter_ptr;
2447 unsigned index = pos >> 3;
2448 unsigned subindex = pos & 0x07;
2451 if (index == 0 && subindex != 0)
2454 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
2455 (index == 0) ? "cpu": gfs2_stype[subindex]);
2457 for_each_possible_cpu(i) {
2458 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
2461 seq_printf(seq, " %15u", i);
2463 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
2464 lkstats[index - 1].stats[subindex]);
2466 seq_putc(seq, '\n');
2470 int __init gfs2_glock_init(void)
2474 ret = rhashtable_init(&gl_hash_table, &ht_parms);
2478 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
2479 WQ_HIGHPRI | WQ_FREEZABLE, 0);
2480 if (!glock_workqueue) {
2481 rhashtable_destroy(&gl_hash_table);
2484 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
2485 WQ_MEM_RECLAIM | WQ_FREEZABLE,
2487 if (!gfs2_delete_workqueue) {
2488 destroy_workqueue(glock_workqueue);
2489 rhashtable_destroy(&gl_hash_table);
2493 ret = register_shrinker(&glock_shrinker);
2495 destroy_workqueue(gfs2_delete_workqueue);
2496 destroy_workqueue(glock_workqueue);
2497 rhashtable_destroy(&gl_hash_table);
2501 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2502 init_waitqueue_head(glock_wait_table + i);
2507 void gfs2_glock_exit(void)
2509 unregister_shrinker(&glock_shrinker);
2510 rhashtable_destroy(&gl_hash_table);
2511 destroy_workqueue(glock_workqueue);
2512 destroy_workqueue(gfs2_delete_workqueue);
2515 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2517 struct gfs2_glock *gl = gi->gl;
2522 if (!lockref_put_not_zero(&gl->gl_lockref))
2523 gfs2_glock_queue_put(gl);
2526 gl = rhashtable_walk_next(&gi->hti);
2527 if (IS_ERR_OR_NULL(gl)) {
2528 if (gl == ERR_PTR(-EAGAIN)) {
2535 if (gl->gl_name.ln_sbd != gi->sdp)
2538 if (!lockref_get_not_dead(&gl->gl_lockref))
2542 if (__lockref_is_dead(&gl->gl_lockref))
2550 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2553 struct gfs2_glock_iter *gi = seq->private;
2557 * We can either stay where we are, skip to the next hash table
2558 * entry, or start from the beginning.
2560 if (*pos < gi->last_pos) {
2561 rhashtable_walk_exit(&gi->hti);
2562 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2565 n = *pos - gi->last_pos;
2568 rhashtable_walk_start(&gi->hti);
2570 gfs2_glock_iter_next(gi, n);
2571 gi->last_pos = *pos;
2575 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2578 struct gfs2_glock_iter *gi = seq->private;
2581 gi->last_pos = *pos;
2582 gfs2_glock_iter_next(gi, 1);
2586 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2589 struct gfs2_glock_iter *gi = seq->private;
2591 rhashtable_walk_stop(&gi->hti);
2594 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2596 dump_glock(seq, iter_ptr, false);
2600 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2603 if (*pos >= GFS2_NR_SBSTATS)
2608 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2612 if (*pos >= GFS2_NR_SBSTATS)
2617 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2622 static const struct seq_operations gfs2_glock_seq_ops = {
2623 .start = gfs2_glock_seq_start,
2624 .next = gfs2_glock_seq_next,
2625 .stop = gfs2_glock_seq_stop,
2626 .show = gfs2_glock_seq_show,
2629 static const struct seq_operations gfs2_glstats_seq_ops = {
2630 .start = gfs2_glock_seq_start,
2631 .next = gfs2_glock_seq_next,
2632 .stop = gfs2_glock_seq_stop,
2633 .show = gfs2_glstats_seq_show,
2636 static const struct seq_operations gfs2_sbstats_sops = {
2637 .start = gfs2_sbstats_seq_start,
2638 .next = gfs2_sbstats_seq_next,
2639 .stop = gfs2_sbstats_seq_stop,
2640 .show = gfs2_sbstats_seq_show,
2643 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2645 static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2646 const struct seq_operations *ops)
2648 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2650 struct seq_file *seq = file->private_data;
2651 struct gfs2_glock_iter *gi = seq->private;
2653 gi->sdp = inode->i_private;
2654 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2656 seq->size = GFS2_SEQ_GOODSIZE;
2658 * Initially, we are "before" the first hash table entry; the
2659 * first call to rhashtable_walk_next gets us the first entry.
2663 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2668 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2670 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2673 static int gfs2_glocks_release(struct inode *inode, struct file *file)
2675 struct seq_file *seq = file->private_data;
2676 struct gfs2_glock_iter *gi = seq->private;
2679 gfs2_glock_put(gi->gl);
2680 rhashtable_walk_exit(&gi->hti);
2681 return seq_release_private(inode, file);
2684 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2686 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2689 static const struct file_operations gfs2_glocks_fops = {
2690 .owner = THIS_MODULE,
2691 .open = gfs2_glocks_open,
2693 .llseek = seq_lseek,
2694 .release = gfs2_glocks_release,
2697 static const struct file_operations gfs2_glstats_fops = {
2698 .owner = THIS_MODULE,
2699 .open = gfs2_glstats_open,
2701 .llseek = seq_lseek,
2702 .release = gfs2_glocks_release,
2705 DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats);
2707 void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2709 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2711 debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2714 debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2715 &gfs2_glstats_fops);
2717 debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2718 &gfs2_sbstats_fops);
2721 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2723 debugfs_remove_recursive(sdp->debugfs_dir);
2724 sdp->debugfs_dir = NULL;
2727 void gfs2_register_debugfs(void)
2729 gfs2_root = debugfs_create_dir("gfs2", NULL);
2732 void gfs2_unregister_debugfs(void)
2734 debugfs_remove(gfs2_root);