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, 2 if type specific operation is
482 * underway, or error.
484 int gfs2_instantiate(struct gfs2_holder *gh)
486 struct gfs2_glock *gl = gh->gh_gl;
487 const struct gfs2_glock_operations *glops = gl->gl_ops;
491 if (!test_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags))
495 * Since we unlock the lockref lock, we set a flag to indicate
496 * instantiate is in progress.
498 if (test_and_set_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags)) {
499 wait_on_bit(&gl->gl_flags, GLF_INSTANTIATE_IN_PROG,
500 TASK_UNINTERRUPTIBLE);
502 * Here we just waited for a different instantiate to finish.
503 * But that may not have been successful, as when a process
504 * locks an inode glock _before_ it has an actual inode to
505 * instantiate into. So we check again. This process might
506 * have an inode to instantiate, so might be successful.
511 ret = glops->go_instantiate(gh);
513 clear_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags);
514 clear_and_wake_up_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags);
519 * do_promote - promote as many requests as possible on the current queue
522 * Returns: 1 if there is a blocked holder at the head of the list, or 2
523 * if a type specific operation is underway.
526 static int do_promote(struct gfs2_glock *gl)
527 __releases(&gl->gl_lockref.lock)
528 __acquires(&gl->gl_lockref.lock)
530 struct gfs2_holder *gh, *tmp, *first_gh;
531 bool incompat_holders_demoted = false;
536 first_gh = find_first_strong_holder(gl);
537 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
538 lock_released = false;
539 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
541 if (!may_grant(gl, first_gh, gh)) {
543 * If we get here, it means we may not grant this holder for
544 * some reason. If this holder is the head of the list, it
545 * means we have a blocked holder at the head, so return 1.
547 if (list_is_first(&gh->gh_list, &gl->gl_holders))
552 if (!incompat_holders_demoted) {
553 demote_incompat_holders(gl, first_gh);
554 incompat_holders_demoted = true;
557 if (test_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags) &&
558 !(gh->gh_flags & GL_SKIP) && gl->gl_ops->go_instantiate) {
559 lock_released = true;
560 spin_unlock(&gl->gl_lockref.lock);
561 ret = gfs2_instantiate(gh);
562 spin_lock(&gl->gl_lockref.lock);
567 list_del_init(&gh->gh_list);
568 trace_gfs2_glock_queue(gh, 0);
569 gfs2_holder_wake(gh);
573 set_bit(HIF_HOLDER, &gh->gh_iflags);
574 trace_gfs2_promote(gh);
575 gfs2_holder_wake(gh);
577 * If we released the gl_lockref.lock the holders list may have
578 * changed. For that reason, we start again at the start of
588 * find_first_waiter - find the first gh that's waiting for the glock
592 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
594 struct gfs2_holder *gh;
596 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
597 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
604 * state_change - record that the glock is now in a different state
606 * @new_state: the new state
609 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
613 held1 = (gl->gl_state != LM_ST_UNLOCKED);
614 held2 = (new_state != LM_ST_UNLOCKED);
616 if (held1 != held2) {
617 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
619 gl->gl_lockref.count++;
621 gl->gl_lockref.count--;
623 if (new_state != gl->gl_target)
624 /* shorten our minimum hold time */
625 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
627 gl->gl_state = new_state;
628 gl->gl_tchange = jiffies;
631 static void gfs2_set_demote(struct gfs2_glock *gl)
633 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
635 set_bit(GLF_DEMOTE, &gl->gl_flags);
637 wake_up(&sdp->sd_async_glock_wait);
640 static void gfs2_demote_wake(struct gfs2_glock *gl)
642 gl->gl_demote_state = LM_ST_EXCLUSIVE;
643 clear_bit(GLF_DEMOTE, &gl->gl_flags);
644 smp_mb__after_atomic();
645 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
649 * finish_xmote - The DLM has replied to one of our lock requests
651 * @ret: The status from the DLM
655 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
657 const struct gfs2_glock_operations *glops = gl->gl_ops;
658 struct gfs2_holder *gh;
659 unsigned state = ret & LM_OUT_ST_MASK;
662 spin_lock(&gl->gl_lockref.lock);
663 trace_gfs2_glock_state_change(gl, state);
664 state_change(gl, state);
665 gh = find_first_waiter(gl);
667 /* Demote to UN request arrived during demote to SH or DF */
668 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
669 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
670 gl->gl_target = LM_ST_UNLOCKED;
672 /* Check for state != intended state */
673 if (unlikely(state != gl->gl_target)) {
674 if (gh && (ret & LM_OUT_CANCELED))
675 gfs2_holder_wake(gh);
676 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
677 /* move to back of queue and try next entry */
678 if (ret & LM_OUT_CANCELED) {
679 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
680 list_move_tail(&gh->gh_list, &gl->gl_holders);
681 gh = find_first_waiter(gl);
682 gl->gl_target = gh->gh_state;
685 /* Some error or failed "try lock" - report it */
686 if ((ret & LM_OUT_ERROR) ||
687 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
688 gl->gl_target = gl->gl_state;
694 /* Unlocked due to conversion deadlock, try again */
697 do_xmote(gl, gh, gl->gl_target);
699 /* Conversion fails, unlock and try again */
702 do_xmote(gl, gh, LM_ST_UNLOCKED);
704 default: /* Everything else */
705 fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n",
706 gl->gl_target, state);
709 spin_unlock(&gl->gl_lockref.lock);
713 /* Fast path - we got what we asked for */
714 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
715 gfs2_demote_wake(gl);
716 if (state != LM_ST_UNLOCKED) {
717 if (glops->go_xmote_bh) {
718 spin_unlock(&gl->gl_lockref.lock);
719 rv = glops->go_xmote_bh(gl);
720 spin_lock(&gl->gl_lockref.lock);
731 clear_bit(GLF_LOCK, &gl->gl_flags);
733 spin_unlock(&gl->gl_lockref.lock);
736 static bool is_system_glock(struct gfs2_glock *gl)
738 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
739 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
741 if (gl == m_ip->i_gl)
747 * do_xmote - Calls the DLM to change the state of a lock
748 * @gl: The lock state
749 * @gh: The holder (only for promotes)
750 * @target: The target lock state
754 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
755 __releases(&gl->gl_lockref.lock)
756 __acquires(&gl->gl_lockref.lock)
758 const struct gfs2_glock_operations *glops = gl->gl_ops;
759 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
760 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
763 if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) &&
764 gh && !(gh->gh_flags & LM_FLAG_NOEXP))
766 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
768 GLOCK_BUG_ON(gl, gl->gl_state == target);
769 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
770 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
773 * If another process is already doing the invalidate, let that
774 * finish first. The glock state machine will get back to this
775 * holder again later.
777 if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
780 do_error(gl, 0); /* Fail queued try locks */
783 set_bit(GLF_BLOCKING, &gl->gl_flags);
784 if ((gl->gl_req == LM_ST_UNLOCKED) ||
785 (gl->gl_state == LM_ST_EXCLUSIVE) ||
786 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
787 clear_bit(GLF_BLOCKING, &gl->gl_flags);
788 spin_unlock(&gl->gl_lockref.lock);
789 if (glops->go_sync) {
790 ret = glops->go_sync(gl);
791 /* If we had a problem syncing (due to io errors or whatever,
792 * we should not invalidate the metadata or tell dlm to
793 * release the glock to other nodes.
796 if (cmpxchg(&sdp->sd_log_error, 0, ret)) {
797 fs_err(sdp, "Error %d syncing glock \n", ret);
798 gfs2_dump_glock(NULL, gl, true);
803 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) {
805 * The call to go_sync should have cleared out the ail list.
806 * If there are still items, we have a problem. We ought to
807 * withdraw, but we can't because the withdraw code also uses
808 * glocks. Warn about the error, dump the glock, then fall
809 * through and wait for logd to do the withdraw for us.
811 if ((atomic_read(&gl->gl_ail_count) != 0) &&
812 (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) {
813 gfs2_glock_assert_warn(gl,
814 !atomic_read(&gl->gl_ail_count));
815 gfs2_dump_glock(NULL, gl, true);
817 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
818 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
824 * Check for an error encountered since we called go_sync and go_inval.
825 * If so, we can't withdraw from the glock code because the withdraw
826 * code itself uses glocks (see function signal_our_withdraw) to
827 * change the mount to read-only. Most importantly, we must not call
828 * dlm to unlock the glock until the journal is in a known good state
829 * (after journal replay) otherwise other nodes may use the object
830 * (rgrp or dinode) and then later, journal replay will corrupt the
831 * file system. The best we can do here is wait for the logd daemon
832 * to see sd_log_error and withdraw, and in the meantime, requeue the
835 * We make a special exception for some system glocks, such as the
836 * system statfs inode glock, which needs to be granted before the
837 * gfs2_quotad daemon can exit, and that exit needs to finish before
838 * we can unmount the withdrawn file system.
840 * However, if we're just unlocking the lock (say, for unmount, when
841 * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
842 * then it's okay to tell dlm to unlock it.
844 if (unlikely(sdp->sd_log_error && !gfs2_withdrawn(sdp)))
845 gfs2_withdraw_delayed(sdp);
846 if (glock_blocked_by_withdraw(gl) &&
847 (target != LM_ST_UNLOCKED ||
848 test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags))) {
849 if (!is_system_glock(gl)) {
850 gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD);
853 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
857 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
859 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
860 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
861 target == LM_ST_UNLOCKED &&
862 test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
863 finish_xmote(gl, target);
864 gfs2_glock_queue_work(gl, 0);
866 fs_err(sdp, "lm_lock ret %d\n", ret);
867 GLOCK_BUG_ON(gl, !gfs2_withdrawn(sdp));
869 } else { /* lock_nolock */
870 finish_xmote(gl, target);
871 gfs2_glock_queue_work(gl, 0);
874 spin_lock(&gl->gl_lockref.lock);
878 * run_queue - do all outstanding tasks related to a glock
879 * @gl: The glock in question
880 * @nonblock: True if we must not block in run_queue
884 static void run_queue(struct gfs2_glock *gl, const int nonblock)
885 __releases(&gl->gl_lockref.lock)
886 __acquires(&gl->gl_lockref.lock)
888 struct gfs2_holder *gh = NULL;
891 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
894 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
896 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
897 gl->gl_demote_state != gl->gl_state) {
898 if (find_first_holder(gl))
902 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
903 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
904 gl->gl_target = gl->gl_demote_state;
906 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
907 gfs2_demote_wake(gl);
908 ret = do_promote(gl);
913 gh = find_first_waiter(gl);
914 gl->gl_target = gh->gh_state;
915 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
916 do_error(gl, 0); /* Fail queued try locks */
918 do_xmote(gl, gh, gl->gl_target);
923 clear_bit(GLF_LOCK, &gl->gl_flags);
924 smp_mb__after_atomic();
925 gl->gl_lockref.count++;
926 __gfs2_glock_queue_work(gl, 0);
930 clear_bit(GLF_LOCK, &gl->gl_flags);
931 smp_mb__after_atomic();
935 void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation)
937 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
939 if (ri->ri_magic == 0)
940 ri->ri_magic = cpu_to_be32(GFS2_MAGIC);
941 if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC))
942 ri->ri_generation_deleted = cpu_to_be64(generation);
945 bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation)
947 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
949 if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC))
951 return generation <= be64_to_cpu(ri->ri_generation_deleted);
954 static void gfs2_glock_poke(struct gfs2_glock *gl)
956 int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP;
957 struct gfs2_holder gh;
960 __gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh, _RET_IP_);
961 error = gfs2_glock_nq(&gh);
964 gfs2_holder_uninit(&gh);
967 static bool gfs2_try_evict(struct gfs2_glock *gl)
969 struct gfs2_inode *ip;
970 bool evicted = false;
973 * If there is contention on the iopen glock and we have an inode, try
974 * to grab and release the inode so that it can be evicted. This will
975 * allow the remote node to go ahead and delete the inode without us
976 * having to do it, which will avoid rgrp glock thrashing.
978 * The remote node is likely still holding the corresponding inode
979 * glock, so it will run before we get to verify that the delete has
982 spin_lock(&gl->gl_lockref.lock);
984 if (ip && !igrab(&ip->i_inode))
986 spin_unlock(&gl->gl_lockref.lock);
988 struct gfs2_glock *inode_gl = NULL;
990 gl->gl_no_formal_ino = ip->i_no_formal_ino;
991 set_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
992 d_prune_aliases(&ip->i_inode);
995 /* If the inode was evicted, gl->gl_object will now be NULL. */
996 spin_lock(&gl->gl_lockref.lock);
1000 lockref_get(&inode_gl->gl_lockref);
1001 clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
1003 spin_unlock(&gl->gl_lockref.lock);
1005 gfs2_glock_poke(inode_gl);
1006 gfs2_glock_put(inode_gl);
1013 static void delete_work_func(struct work_struct *work)
1015 struct delayed_work *dwork = to_delayed_work(work);
1016 struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete);
1017 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1018 struct inode *inode;
1019 u64 no_addr = gl->gl_name.ln_number;
1021 spin_lock(&gl->gl_lockref.lock);
1022 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1023 spin_unlock(&gl->gl_lockref.lock);
1025 if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
1027 * If we can evict the inode, give the remote node trying to
1028 * delete the inode some time before verifying that the delete
1029 * has happened. Otherwise, if we cause contention on the inode glock
1030 * immediately, the remote node will think that we still have
1031 * the inode in use, and so it will give up waiting.
1033 * If we can't evict the inode, signal to the remote node that
1034 * the inode is still in use. We'll later try to delete the
1035 * inode locally in gfs2_evict_inode.
1037 * FIXME: We only need to verify that the remote node has
1038 * deleted the inode because nodes before this remote delete
1039 * rework won't cooperate. At a later time, when we no longer
1040 * care about compatibility with such nodes, we can skip this
1043 if (gfs2_try_evict(gl)) {
1044 if (gfs2_queue_delete_work(gl, 5 * HZ))
1050 inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino,
1051 GFS2_BLKST_UNLINKED);
1052 if (!IS_ERR_OR_NULL(inode)) {
1053 d_prune_aliases(inode);
1060 static void glock_work_func(struct work_struct *work)
1062 unsigned long delay = 0;
1063 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
1064 unsigned int drop_refs = 1;
1066 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
1067 finish_xmote(gl, gl->gl_reply);
1070 spin_lock(&gl->gl_lockref.lock);
1071 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1072 gl->gl_state != LM_ST_UNLOCKED &&
1073 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
1074 unsigned long holdtime, now = jiffies;
1076 holdtime = gl->gl_tchange + gl->gl_hold_time;
1077 if (time_before(now, holdtime))
1078 delay = holdtime - now;
1081 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1082 gfs2_set_demote(gl);
1087 /* Keep one glock reference for the work we requeue. */
1089 if (gl->gl_name.ln_type != LM_TYPE_INODE)
1091 __gfs2_glock_queue_work(gl, delay);
1095 * Drop the remaining glock references manually here. (Mind that
1096 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
1099 gl->gl_lockref.count -= drop_refs;
1100 if (!gl->gl_lockref.count) {
1101 __gfs2_glock_put(gl);
1104 spin_unlock(&gl->gl_lockref.lock);
1107 static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
1108 struct gfs2_glock *new)
1110 struct wait_glock_queue wait;
1111 wait_queue_head_t *wq = glock_waitqueue(name);
1112 struct gfs2_glock *gl;
1115 init_wait(&wait.wait);
1116 wait.wait.func = glock_wake_function;
1119 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1122 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
1123 &new->gl_node, ht_parms);
1127 gl = rhashtable_lookup_fast(&gl_hash_table,
1130 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
1137 finish_wait(wq, &wait.wait);
1142 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
1143 * @sdp: The GFS2 superblock
1144 * @number: the lock number
1145 * @glops: The glock_operations to use
1146 * @create: If 0, don't create the glock if it doesn't exist
1147 * @glp: the glock is returned here
1149 * This does not lock a glock, just finds/creates structures for one.
1154 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
1155 const struct gfs2_glock_operations *glops, int create,
1156 struct gfs2_glock **glp)
1158 struct super_block *s = sdp->sd_vfs;
1159 struct lm_lockname name = { .ln_number = number,
1160 .ln_type = glops->go_type,
1162 struct gfs2_glock *gl, *tmp;
1163 struct address_space *mapping;
1166 gl = find_insert_glock(&name, NULL);
1174 if (glops->go_flags & GLOF_ASPACE) {
1175 struct gfs2_glock_aspace *gla =
1176 kmem_cache_alloc(gfs2_glock_aspace_cachep, GFP_NOFS);
1181 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_NOFS);
1185 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
1188 if (glops->go_flags & GLOF_LVB) {
1189 gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1190 if (!gl->gl_lksb.sb_lvbptr) {
1191 gfs2_glock_dealloc(&gl->gl_rcu);
1196 atomic_inc(&sdp->sd_glock_disposal);
1197 gl->gl_node.next = NULL;
1198 gl->gl_flags = glops->go_instantiate ? BIT(GLF_INSTANTIATE_NEEDED) : 0;
1200 lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass);
1201 gl->gl_lockref.count = 1;
1202 gl->gl_state = LM_ST_UNLOCKED;
1203 gl->gl_target = LM_ST_UNLOCKED;
1204 gl->gl_demote_state = LM_ST_EXCLUSIVE;
1207 /* We use the global stats to estimate the initial per-glock stats */
1208 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
1210 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
1211 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
1212 gl->gl_tchange = jiffies;
1213 gl->gl_object = NULL;
1214 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
1215 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
1216 if (gl->gl_name.ln_type == LM_TYPE_IOPEN)
1217 INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func);
1219 mapping = gfs2_glock2aspace(gl);
1221 mapping->a_ops = &gfs2_meta_aops;
1222 mapping->host = s->s_bdev->bd_inode;
1224 mapping_set_gfp_mask(mapping, GFP_NOFS);
1225 mapping->private_data = NULL;
1226 mapping->writeback_index = 0;
1229 tmp = find_insert_glock(&name, gl);
1241 gfs2_glock_dealloc(&gl->gl_rcu);
1242 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
1243 wake_up(&sdp->sd_glock_wait);
1250 * __gfs2_holder_init - initialize a struct gfs2_holder in the default way
1252 * @state: the state we're requesting
1253 * @flags: the modifier flags
1254 * @gh: the holder structure
1258 void __gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
1259 struct gfs2_holder *gh, unsigned long ip)
1261 INIT_LIST_HEAD(&gh->gh_list);
1264 gh->gh_owner_pid = get_pid(task_pid(current));
1265 gh->gh_state = state;
1266 gh->gh_flags = flags;
1268 gfs2_glock_hold(gl);
1272 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1273 * @state: the state we're requesting
1274 * @flags: the modifier flags
1275 * @gh: the holder structure
1277 * Don't mess with the glock.
1281 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
1283 gh->gh_state = state;
1284 gh->gh_flags = flags;
1286 gh->gh_ip = _RET_IP_;
1287 put_pid(gh->gh_owner_pid);
1288 gh->gh_owner_pid = get_pid(task_pid(current));
1292 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1293 * @gh: the holder structure
1297 void gfs2_holder_uninit(struct gfs2_holder *gh)
1299 put_pid(gh->gh_owner_pid);
1300 gfs2_glock_put(gh->gh_gl);
1301 gfs2_holder_mark_uninitialized(gh);
1305 static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
1306 unsigned long start_time)
1308 /* Have we waited longer that a second? */
1309 if (time_after(jiffies, start_time + HZ)) {
1310 /* Lengthen the minimum hold time. */
1311 gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
1317 * gfs2_glock_wait - wait on a glock acquisition
1318 * @gh: the glock holder
1320 * Returns: 0 on success
1323 int gfs2_glock_wait(struct gfs2_holder *gh)
1325 unsigned long start_time = jiffies;
1328 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1329 gfs2_glock_update_hold_time(gh->gh_gl, start_time);
1330 return gh->gh_error;
1333 static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
1337 for (i = 0; i < num_gh; i++)
1338 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
1344 * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1345 * @num_gh: the number of holders in the array
1346 * @ghs: the glock holder array
1348 * Returns: 0 on success, meaning all glocks have been granted and are held.
1349 * -ESTALE if the request timed out, meaning all glocks were released,
1350 * and the caller should retry the operation.
1353 int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
1355 struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
1356 int i, ret = 0, timeout = 0;
1357 unsigned long start_time = jiffies;
1362 * Total up the (minimum hold time * 2) of all glocks and use that to
1363 * determine the max amount of time we should wait.
1365 for (i = 0; i < num_gh; i++)
1366 timeout += ghs[i].gh_gl->gl_hold_time << 1;
1369 if (!wait_event_timeout(sdp->sd_async_glock_wait,
1370 !glocks_pending(num_gh, ghs), timeout))
1371 ret = -ESTALE; /* request timed out. */
1374 * If dlm granted all our requests, we need to adjust the glock
1375 * minimum hold time values according to how long we waited.
1377 * If our request timed out, we need to repeatedly release any held
1378 * glocks we acquired thus far to allow dlm to acquire the remaining
1379 * glocks without deadlocking. We cannot currently cancel outstanding
1380 * glock acquisitions.
1382 * The HIF_WAIT bit tells us which requests still need a response from
1385 * If dlm sent us any errors, we return the first error we find.
1387 keep_waiting = false;
1388 for (i = 0; i < num_gh; i++) {
1389 /* Skip holders we have already dequeued below. */
1390 if (!gfs2_holder_queued(&ghs[i]))
1392 /* Skip holders with a pending DLM response. */
1393 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags)) {
1394 keep_waiting = true;
1398 if (test_bit(HIF_HOLDER, &ghs[i].gh_iflags)) {
1400 gfs2_glock_dq(&ghs[i]);
1402 gfs2_glock_update_hold_time(ghs[i].gh_gl,
1406 ret = ghs[i].gh_error;
1413 * At this point, we've either acquired all locks or released them all.
1419 * handle_callback - process a demote request
1421 * @state: the state the caller wants us to change to
1422 * @delay: zero to demote immediately; otherwise pending demote
1423 * @remote: true if this came from a different cluster node
1425 * There are only two requests that we are going to see in actual
1426 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1429 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
1430 unsigned long delay, bool remote)
1433 set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1435 gfs2_set_demote(gl);
1436 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
1437 gl->gl_demote_state = state;
1438 gl->gl_demote_time = jiffies;
1439 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
1440 gl->gl_demote_state != state) {
1441 gl->gl_demote_state = LM_ST_UNLOCKED;
1443 if (gl->gl_ops->go_callback)
1444 gl->gl_ops->go_callback(gl, remote);
1445 trace_gfs2_demote_rq(gl, remote);
1448 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
1450 struct va_format vaf;
1453 va_start(args, fmt);
1456 seq_vprintf(seq, fmt, args);
1461 pr_err("%pV", &vaf);
1468 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1469 * @gh: the holder structure to add
1471 * Eventually we should move the recursive locking trap to a
1472 * debugging option or something like that. This is the fast
1473 * path and needs to have the minimum number of distractions.
1477 static inline void add_to_queue(struct gfs2_holder *gh)
1478 __releases(&gl->gl_lockref.lock)
1479 __acquires(&gl->gl_lockref.lock)
1481 struct gfs2_glock *gl = gh->gh_gl;
1482 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1483 struct list_head *insert_pt = NULL;
1484 struct gfs2_holder *gh2;
1487 GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1488 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1489 GLOCK_BUG_ON(gl, true);
1491 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1492 if (test_bit(GLF_LOCK, &gl->gl_flags)) {
1493 struct gfs2_holder *first_gh;
1495 first_gh = find_first_strong_holder(gl);
1496 try_futile = !may_grant(gl, first_gh, gh);
1498 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1502 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1503 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
1504 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK) &&
1505 !test_bit(HIF_MAY_DEMOTE, &gh2->gh_iflags)))
1506 goto trap_recursive;
1508 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1510 gh->gh_error = GLR_TRYFAILED;
1511 gfs2_holder_wake(gh);
1514 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1516 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1517 insert_pt = &gh2->gh_list;
1519 trace_gfs2_glock_queue(gh, 1);
1520 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1521 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1522 if (likely(insert_pt == NULL)) {
1523 list_add_tail(&gh->gh_list, &gl->gl_holders);
1524 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1528 list_add_tail(&gh->gh_list, insert_pt);
1530 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
1531 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1532 spin_unlock(&gl->gl_lockref.lock);
1533 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1534 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1535 spin_lock(&gl->gl_lockref.lock);
1540 fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1541 fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1542 fs_err(sdp, "lock type: %d req lock state : %d\n",
1543 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1544 fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1545 fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1546 fs_err(sdp, "lock type: %d req lock state : %d\n",
1547 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1548 gfs2_dump_glock(NULL, gl, true);
1553 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1554 * @gh: the holder structure
1556 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1558 * Returns: 0, GLR_TRYFAILED, or errno on failure
1561 int gfs2_glock_nq(struct gfs2_holder *gh)
1563 struct gfs2_glock *gl = gh->gh_gl;
1566 if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP))
1569 if (test_bit(GLF_LRU, &gl->gl_flags))
1570 gfs2_glock_remove_from_lru(gl);
1573 spin_lock(&gl->gl_lockref.lock);
1575 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1576 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1577 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1578 gl->gl_lockref.count++;
1579 __gfs2_glock_queue_work(gl, 0);
1582 spin_unlock(&gl->gl_lockref.lock);
1584 if (!(gh->gh_flags & GL_ASYNC))
1585 error = gfs2_glock_wait(gh);
1591 * gfs2_glock_poll - poll to see if an async request has been completed
1594 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1597 int gfs2_glock_poll(struct gfs2_holder *gh)
1599 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1602 static inline bool needs_demote(struct gfs2_glock *gl)
1604 return (test_bit(GLF_DEMOTE, &gl->gl_flags) ||
1605 test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags));
1608 static void __gfs2_glock_dq(struct gfs2_holder *gh)
1610 struct gfs2_glock *gl = gh->gh_gl;
1611 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1616 * This while loop is similar to function demote_incompat_holders:
1617 * If the glock is due to be demoted (which may be from another node
1618 * or even if this holder is GL_NOCACHE), the weak holders are
1619 * demoted as well, allowing the glock to be demoted.
1623 * If we're in the process of file system withdraw, we cannot
1624 * just dequeue any glocks until our journal is recovered, lest
1625 * we introduce file system corruption. We need two exceptions
1626 * to this rule: We need to allow unlocking of nondisk glocks
1627 * and the glock for our own journal that needs recovery.
1629 if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) &&
1630 glock_blocked_by_withdraw(gl) &&
1631 gh->gh_gl != sdp->sd_jinode_gl) {
1632 sdp->sd_glock_dqs_held++;
1633 spin_unlock(&gl->gl_lockref.lock);
1635 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
1636 TASK_UNINTERRUPTIBLE);
1637 spin_lock(&gl->gl_lockref.lock);
1641 * This holder should not be cached, so mark it for demote.
1642 * Note: this should be done before the check for needs_demote
1645 if (gh->gh_flags & GL_NOCACHE)
1646 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1648 list_del_init(&gh->gh_list);
1649 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1650 trace_gfs2_glock_queue(gh, 0);
1653 * If there hasn't been a demote request we are done.
1654 * (Let the remaining holders, if any, keep holding it.)
1656 if (!needs_demote(gl)) {
1657 if (list_empty(&gl->gl_holders))
1662 * If we have another strong holder (we cannot auto-demote)
1663 * we are done. It keeps holding it until it is done.
1665 if (find_first_strong_holder(gl))
1669 * If we have a weak holder at the head of the list, it
1670 * (and all others like it) must be auto-demoted. If there
1671 * are no more weak holders, we exit the while loop.
1673 gh = find_first_holder(gl);
1676 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1677 gfs2_glock_add_to_lru(gl);
1679 if (unlikely(!fast_path)) {
1680 gl->gl_lockref.count++;
1681 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1682 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1683 gl->gl_name.ln_type == LM_TYPE_INODE)
1684 delay = gl->gl_hold_time;
1685 __gfs2_glock_queue_work(gl, delay);
1690 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1691 * @gh: the glock holder
1694 void gfs2_glock_dq(struct gfs2_holder *gh)
1696 struct gfs2_glock *gl = gh->gh_gl;
1698 spin_lock(&gl->gl_lockref.lock);
1699 if (list_is_first(&gh->gh_list, &gl->gl_holders) &&
1700 !test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1701 spin_unlock(&gl->gl_lockref.lock);
1702 gl->gl_name.ln_sbd->sd_lockstruct.ls_ops->lm_cancel(gl);
1703 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1704 spin_lock(&gl->gl_lockref.lock);
1707 __gfs2_glock_dq(gh);
1708 spin_unlock(&gl->gl_lockref.lock);
1711 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1713 struct gfs2_glock *gl = gh->gh_gl;
1716 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1720 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1721 * @gh: the holder structure
1725 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1728 gfs2_holder_uninit(gh);
1732 * gfs2_glock_nq_num - acquire a glock based on lock number
1733 * @sdp: the filesystem
1734 * @number: the lock number
1735 * @glops: the glock operations for the type of glock
1736 * @state: the state to acquire the glock in
1737 * @flags: modifier flags for the acquisition
1738 * @gh: the struct gfs2_holder
1743 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1744 const struct gfs2_glock_operations *glops,
1745 unsigned int state, u16 flags, struct gfs2_holder *gh)
1747 struct gfs2_glock *gl;
1750 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1752 error = gfs2_glock_nq_init(gl, state, flags, gh);
1760 * glock_compare - Compare two struct gfs2_glock structures for sorting
1761 * @arg_a: the first structure
1762 * @arg_b: the second structure
1766 static int glock_compare(const void *arg_a, const void *arg_b)
1768 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1769 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1770 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1771 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1773 if (a->ln_number > b->ln_number)
1775 if (a->ln_number < b->ln_number)
1777 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1782 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1783 * @num_gh: the number of structures
1784 * @ghs: an array of struct gfs2_holder structures
1785 * @p: placeholder for the holder structure to pass back
1787 * Returns: 0 on success (all glocks acquired),
1788 * errno on failure (no glocks acquired)
1791 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1792 struct gfs2_holder **p)
1797 for (x = 0; x < num_gh; x++)
1800 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1802 for (x = 0; x < num_gh; x++) {
1803 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1805 error = gfs2_glock_nq(p[x]);
1808 gfs2_glock_dq(p[x]);
1817 * gfs2_glock_nq_m - acquire multiple glocks
1818 * @num_gh: the number of structures
1819 * @ghs: an array of struct gfs2_holder structures
1822 * Returns: 0 on success (all glocks acquired),
1823 * errno on failure (no glocks acquired)
1826 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1828 struct gfs2_holder *tmp[4];
1829 struct gfs2_holder **pph = tmp;
1836 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1837 return gfs2_glock_nq(ghs);
1841 pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *),
1847 error = nq_m_sync(num_gh, ghs, pph);
1856 * gfs2_glock_dq_m - release multiple glocks
1857 * @num_gh: the number of structures
1858 * @ghs: an array of struct gfs2_holder structures
1862 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1865 gfs2_glock_dq(&ghs[num_gh]);
1868 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1870 unsigned long delay = 0;
1871 unsigned long holdtime;
1872 unsigned long now = jiffies;
1874 gfs2_glock_hold(gl);
1875 spin_lock(&gl->gl_lockref.lock);
1876 holdtime = gl->gl_tchange + gl->gl_hold_time;
1877 if (!list_empty(&gl->gl_holders) &&
1878 gl->gl_name.ln_type == LM_TYPE_INODE) {
1879 if (time_before(now, holdtime))
1880 delay = holdtime - now;
1881 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1882 delay = gl->gl_hold_time;
1885 * Note 1: We cannot call demote_incompat_holders from handle_callback
1886 * or gfs2_set_demote due to recursion problems like: gfs2_glock_dq ->
1887 * handle_callback -> demote_incompat_holders -> gfs2_glock_dq
1888 * Plus, we only want to demote the holders if the request comes from
1889 * a remote cluster node because local holder conflicts are resolved
1892 * Note 2: if a remote node wants this glock in EX mode, lock_dlm will
1893 * request that we set our state to UNLOCKED. Here we mock up a holder
1894 * to make it look like someone wants the lock EX locally. Any SH
1895 * and DF requests should be able to share the lock without demoting.
1897 * Note 3: We only want to demote the demoteable holders when there
1898 * are no more strong holders. The demoteable holders might as well
1899 * keep the glock until the last strong holder is done with it.
1901 if (!find_first_strong_holder(gl)) {
1902 struct gfs2_holder mock_gh = {
1904 .gh_state = (state == LM_ST_UNLOCKED) ?
1905 LM_ST_EXCLUSIVE : state,
1906 .gh_iflags = BIT(HIF_HOLDER)
1909 demote_incompat_holders(gl, &mock_gh);
1911 handle_callback(gl, state, delay, true);
1912 __gfs2_glock_queue_work(gl, delay);
1913 spin_unlock(&gl->gl_lockref.lock);
1917 * gfs2_should_freeze - Figure out if glock should be frozen
1918 * @gl: The glock in question
1920 * Glocks are not frozen if (a) the result of the dlm operation is
1921 * an error, (b) the locking operation was an unlock operation or
1922 * (c) if there is a "noexp" flagged request anywhere in the queue
1924 * Returns: 1 if freezing should occur, 0 otherwise
1927 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1929 const struct gfs2_holder *gh;
1931 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1933 if (gl->gl_target == LM_ST_UNLOCKED)
1936 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1937 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1939 if (LM_FLAG_NOEXP & gh->gh_flags)
1947 * gfs2_glock_complete - Callback used by locking
1948 * @gl: Pointer to the glock
1949 * @ret: The return value from the dlm
1951 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1952 * to use a bitfield shared with other glock state fields.
1955 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1957 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1959 spin_lock(&gl->gl_lockref.lock);
1962 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1963 if (gfs2_should_freeze(gl)) {
1964 set_bit(GLF_FROZEN, &gl->gl_flags);
1965 spin_unlock(&gl->gl_lockref.lock);
1970 gl->gl_lockref.count++;
1971 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1972 __gfs2_glock_queue_work(gl, 0);
1973 spin_unlock(&gl->gl_lockref.lock);
1976 static int glock_cmp(void *priv, const struct list_head *a,
1977 const struct list_head *b)
1979 struct gfs2_glock *gla, *glb;
1981 gla = list_entry(a, struct gfs2_glock, gl_lru);
1982 glb = list_entry(b, struct gfs2_glock, gl_lru);
1984 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1986 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1993 * gfs2_dispose_glock_lru - Demote a list of glocks
1994 * @list: The list to dispose of
1996 * Disposing of glocks may involve disk accesses, so that here we sort
1997 * the glocks by number (i.e. disk location of the inodes) so that if
1998 * there are any such accesses, they'll be sent in order (mostly).
2000 * Must be called under the lru_lock, but may drop and retake this
2001 * lock. While the lru_lock is dropped, entries may vanish from the
2002 * list, but no new entries will appear on the list (since it is
2006 static void gfs2_dispose_glock_lru(struct list_head *list)
2007 __releases(&lru_lock)
2008 __acquires(&lru_lock)
2010 struct gfs2_glock *gl;
2012 list_sort(NULL, list, glock_cmp);
2014 while(!list_empty(list)) {
2015 gl = list_first_entry(list, struct gfs2_glock, gl_lru);
2016 list_del_init(&gl->gl_lru);
2017 clear_bit(GLF_LRU, &gl->gl_flags);
2018 if (!spin_trylock(&gl->gl_lockref.lock)) {
2020 list_add(&gl->gl_lru, &lru_list);
2021 set_bit(GLF_LRU, &gl->gl_flags);
2022 atomic_inc(&lru_count);
2025 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
2026 spin_unlock(&gl->gl_lockref.lock);
2027 goto add_back_to_lru;
2029 gl->gl_lockref.count++;
2031 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2032 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
2033 __gfs2_glock_queue_work(gl, 0);
2034 spin_unlock(&gl->gl_lockref.lock);
2035 cond_resched_lock(&lru_lock);
2040 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
2041 * @nr: The number of entries to scan
2043 * This function selects the entries on the LRU which are able to
2044 * be demoted, and then kicks off the process by calling
2045 * gfs2_dispose_glock_lru() above.
2048 static long gfs2_scan_glock_lru(int nr)
2050 struct gfs2_glock *gl;
2055 spin_lock(&lru_lock);
2056 while ((nr-- >= 0) && !list_empty(&lru_list)) {
2057 gl = list_first_entry(&lru_list, struct gfs2_glock, gl_lru);
2059 /* Test for being demotable */
2060 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
2061 list_move(&gl->gl_lru, &dispose);
2062 atomic_dec(&lru_count);
2067 list_move(&gl->gl_lru, &skipped);
2069 list_splice(&skipped, &lru_list);
2070 if (!list_empty(&dispose))
2071 gfs2_dispose_glock_lru(&dispose);
2072 spin_unlock(&lru_lock);
2077 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
2078 struct shrink_control *sc)
2080 if (!(sc->gfp_mask & __GFP_FS))
2082 return gfs2_scan_glock_lru(sc->nr_to_scan);
2085 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
2086 struct shrink_control *sc)
2088 return vfs_pressure_ratio(atomic_read(&lru_count));
2091 static struct shrinker glock_shrinker = {
2092 .seeks = DEFAULT_SEEKS,
2093 .count_objects = gfs2_glock_shrink_count,
2094 .scan_objects = gfs2_glock_shrink_scan,
2098 * glock_hash_walk - Call a function for glock in a hash bucket
2099 * @examiner: the function
2100 * @sdp: the filesystem
2102 * Note that the function can be called multiple times on the same
2103 * object. So the user must ensure that the function can cope with
2107 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
2109 struct gfs2_glock *gl;
2110 struct rhashtable_iter iter;
2112 rhashtable_walk_enter(&gl_hash_table, &iter);
2115 rhashtable_walk_start(&iter);
2117 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl)) {
2118 if (gl->gl_name.ln_sbd == sdp)
2122 rhashtable_walk_stop(&iter);
2123 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
2125 rhashtable_walk_exit(&iter);
2128 bool gfs2_queue_delete_work(struct gfs2_glock *gl, unsigned long delay)
2132 spin_lock(&gl->gl_lockref.lock);
2133 queued = queue_delayed_work(gfs2_delete_workqueue,
2134 &gl->gl_delete, delay);
2136 set_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2137 spin_unlock(&gl->gl_lockref.lock);
2141 void gfs2_cancel_delete_work(struct gfs2_glock *gl)
2143 if (cancel_delayed_work(&gl->gl_delete)) {
2144 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2149 bool gfs2_delete_work_queued(const struct gfs2_glock *gl)
2151 return test_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2154 static void flush_delete_work(struct gfs2_glock *gl)
2156 if (gl->gl_name.ln_type == LM_TYPE_IOPEN) {
2157 if (cancel_delayed_work(&gl->gl_delete)) {
2158 queue_delayed_work(gfs2_delete_workqueue,
2164 void gfs2_flush_delete_work(struct gfs2_sbd *sdp)
2166 glock_hash_walk(flush_delete_work, sdp);
2167 flush_workqueue(gfs2_delete_workqueue);
2171 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
2172 * @gl: The glock to thaw
2176 static void thaw_glock(struct gfs2_glock *gl)
2178 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
2180 if (!lockref_get_not_dead(&gl->gl_lockref))
2182 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
2183 gfs2_glock_queue_work(gl, 0);
2187 * clear_glock - look at a glock and see if we can free it from glock cache
2188 * @gl: the glock to look at
2192 static void clear_glock(struct gfs2_glock *gl)
2194 gfs2_glock_remove_from_lru(gl);
2196 spin_lock(&gl->gl_lockref.lock);
2197 if (!__lockref_is_dead(&gl->gl_lockref)) {
2198 gl->gl_lockref.count++;
2199 if (gl->gl_state != LM_ST_UNLOCKED)
2200 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2201 __gfs2_glock_queue_work(gl, 0);
2203 spin_unlock(&gl->gl_lockref.lock);
2207 * gfs2_glock_thaw - Thaw any frozen glocks
2208 * @sdp: The super block
2212 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
2214 glock_hash_walk(thaw_glock, sdp);
2217 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2219 spin_lock(&gl->gl_lockref.lock);
2220 gfs2_dump_glock(seq, gl, fsid);
2221 spin_unlock(&gl->gl_lockref.lock);
2224 static void dump_glock_func(struct gfs2_glock *gl)
2226 dump_glock(NULL, gl, true);
2230 * gfs2_gl_hash_clear - Empty out the glock hash table
2231 * @sdp: the filesystem
2233 * Called when unmounting the filesystem.
2236 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
2238 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
2239 flush_workqueue(glock_workqueue);
2240 glock_hash_walk(clear_glock, sdp);
2241 flush_workqueue(glock_workqueue);
2242 wait_event_timeout(sdp->sd_glock_wait,
2243 atomic_read(&sdp->sd_glock_disposal) == 0,
2245 glock_hash_walk(dump_glock_func, sdp);
2248 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
2250 struct gfs2_glock *gl = ip->i_gl;
2253 ret = gfs2_truncatei_resume(ip);
2254 gfs2_glock_assert_withdraw(gl, ret == 0);
2256 spin_lock(&gl->gl_lockref.lock);
2257 clear_bit(GLF_LOCK, &gl->gl_flags);
2259 spin_unlock(&gl->gl_lockref.lock);
2262 static const char *state2str(unsigned state)
2265 case LM_ST_UNLOCKED:
2269 case LM_ST_DEFERRED:
2271 case LM_ST_EXCLUSIVE:
2277 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
2280 if (flags & LM_FLAG_TRY)
2282 if (flags & LM_FLAG_TRY_1CB)
2284 if (flags & LM_FLAG_NOEXP)
2286 if (flags & LM_FLAG_ANY)
2288 if (flags & LM_FLAG_PRIORITY)
2290 if (flags & LM_FLAG_NODE_SCOPE)
2292 if (flags & GL_ASYNC)
2294 if (flags & GL_EXACT)
2296 if (flags & GL_NOCACHE)
2298 if (test_bit(HIF_HOLDER, &iflags))
2300 if (test_bit(HIF_WAIT, &iflags))
2302 if (test_bit(HIF_MAY_DEMOTE, &iflags))
2304 if (flags & GL_SKIP)
2311 * dump_holder - print information about a glock holder
2312 * @seq: the seq_file struct
2313 * @gh: the glock holder
2314 * @fs_id_buf: pointer to file system id (if requested)
2318 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
2319 const char *fs_id_buf)
2321 struct task_struct *gh_owner = NULL;
2325 if (gh->gh_owner_pid)
2326 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
2327 gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2328 fs_id_buf, state2str(gh->gh_state),
2329 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
2331 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
2332 gh_owner ? gh_owner->comm : "(ended)",
2337 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
2339 const unsigned long *gflags = &gl->gl_flags;
2342 if (test_bit(GLF_LOCK, gflags))
2344 if (test_bit(GLF_DEMOTE, gflags))
2346 if (test_bit(GLF_PENDING_DEMOTE, gflags))
2348 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
2350 if (test_bit(GLF_DIRTY, gflags))
2352 if (test_bit(GLF_LFLUSH, gflags))
2354 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
2356 if (test_bit(GLF_REPLY_PENDING, gflags))
2358 if (test_bit(GLF_INITIAL, gflags))
2360 if (test_bit(GLF_FROZEN, gflags))
2362 if (!list_empty(&gl->gl_holders))
2364 if (test_bit(GLF_LRU, gflags))
2368 if (test_bit(GLF_BLOCKING, gflags))
2370 if (test_bit(GLF_PENDING_DELETE, gflags))
2372 if (test_bit(GLF_FREEING, gflags))
2374 if (test_bit(GLF_INSTANTIATE_NEEDED, gflags))
2376 if (test_bit(GLF_INSTANTIATE_IN_PROG, gflags))
2383 * gfs2_dump_glock - print information about a glock
2384 * @seq: The seq_file struct
2386 * @fsid: If true, also dump the file system id
2388 * The file format is as follows:
2389 * One line per object, capital letters are used to indicate objects
2390 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2391 * other objects are indented by a single space and follow the glock to
2392 * which they are related. Fields are indicated by lower case letters
2393 * followed by a colon and the field value, except for strings which are in
2394 * [] so that its possible to see if they are composed of spaces for
2395 * example. The field's are n = number (id of the object), f = flags,
2396 * t = type, s = state, r = refcount, e = error, p = pid.
2400 void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2402 const struct gfs2_glock_operations *glops = gl->gl_ops;
2403 unsigned long long dtime;
2404 const struct gfs2_holder *gh;
2405 char gflags_buf[32];
2406 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2407 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2408 unsigned long nrpages = 0;
2410 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
2411 struct address_space *mapping = gfs2_glock2aspace(gl);
2413 nrpages = mapping->nrpages;
2415 memset(fs_id_buf, 0, sizeof(fs_id_buf));
2416 if (fsid && sdp) /* safety precaution */
2417 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2418 dtime = jiffies - gl->gl_demote_time;
2419 dtime *= 1000000/HZ; /* demote time in uSec */
2420 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
2422 gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2423 "v:%d r:%d m:%ld p:%lu\n",
2424 fs_id_buf, state2str(gl->gl_state),
2425 gl->gl_name.ln_type,
2426 (unsigned long long)gl->gl_name.ln_number,
2427 gflags2str(gflags_buf, gl),
2428 state2str(gl->gl_target),
2429 state2str(gl->gl_demote_state), dtime,
2430 atomic_read(&gl->gl_ail_count),
2431 atomic_read(&gl->gl_revokes),
2432 (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages);
2434 list_for_each_entry(gh, &gl->gl_holders, gh_list)
2435 dump_holder(seq, gh, fs_id_buf);
2437 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
2438 glops->go_dump(seq, gl, fs_id_buf);
2441 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
2443 struct gfs2_glock *gl = iter_ptr;
2445 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2446 gl->gl_name.ln_type,
2447 (unsigned long long)gl->gl_name.ln_number,
2448 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
2449 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
2450 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
2451 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
2452 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
2453 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
2454 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
2455 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
2459 static const char *gfs2_gltype[] = {
2473 static const char *gfs2_stype[] = {
2474 [GFS2_LKS_SRTT] = "srtt",
2475 [GFS2_LKS_SRTTVAR] = "srttvar",
2476 [GFS2_LKS_SRTTB] = "srttb",
2477 [GFS2_LKS_SRTTVARB] = "srttvarb",
2478 [GFS2_LKS_SIRT] = "sirt",
2479 [GFS2_LKS_SIRTVAR] = "sirtvar",
2480 [GFS2_LKS_DCOUNT] = "dlm",
2481 [GFS2_LKS_QCOUNT] = "queue",
2484 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2486 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
2488 struct gfs2_sbd *sdp = seq->private;
2489 loff_t pos = *(loff_t *)iter_ptr;
2490 unsigned index = pos >> 3;
2491 unsigned subindex = pos & 0x07;
2494 if (index == 0 && subindex != 0)
2497 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
2498 (index == 0) ? "cpu": gfs2_stype[subindex]);
2500 for_each_possible_cpu(i) {
2501 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
2504 seq_printf(seq, " %15u", i);
2506 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
2507 lkstats[index - 1].stats[subindex]);
2509 seq_putc(seq, '\n');
2513 int __init gfs2_glock_init(void)
2517 ret = rhashtable_init(&gl_hash_table, &ht_parms);
2521 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
2522 WQ_HIGHPRI | WQ_FREEZABLE, 0);
2523 if (!glock_workqueue) {
2524 rhashtable_destroy(&gl_hash_table);
2527 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
2528 WQ_MEM_RECLAIM | WQ_FREEZABLE,
2530 if (!gfs2_delete_workqueue) {
2531 destroy_workqueue(glock_workqueue);
2532 rhashtable_destroy(&gl_hash_table);
2536 ret = register_shrinker(&glock_shrinker);
2538 destroy_workqueue(gfs2_delete_workqueue);
2539 destroy_workqueue(glock_workqueue);
2540 rhashtable_destroy(&gl_hash_table);
2544 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2545 init_waitqueue_head(glock_wait_table + i);
2550 void gfs2_glock_exit(void)
2552 unregister_shrinker(&glock_shrinker);
2553 rhashtable_destroy(&gl_hash_table);
2554 destroy_workqueue(glock_workqueue);
2555 destroy_workqueue(gfs2_delete_workqueue);
2558 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2560 struct gfs2_glock *gl = gi->gl;
2565 if (!lockref_put_not_zero(&gl->gl_lockref))
2566 gfs2_glock_queue_put(gl);
2569 gl = rhashtable_walk_next(&gi->hti);
2570 if (IS_ERR_OR_NULL(gl)) {
2571 if (gl == ERR_PTR(-EAGAIN)) {
2578 if (gl->gl_name.ln_sbd != gi->sdp)
2581 if (!lockref_get_not_dead(&gl->gl_lockref))
2585 if (__lockref_is_dead(&gl->gl_lockref))
2593 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2596 struct gfs2_glock_iter *gi = seq->private;
2600 * We can either stay where we are, skip to the next hash table
2601 * entry, or start from the beginning.
2603 if (*pos < gi->last_pos) {
2604 rhashtable_walk_exit(&gi->hti);
2605 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2608 n = *pos - gi->last_pos;
2611 rhashtable_walk_start(&gi->hti);
2613 gfs2_glock_iter_next(gi, n);
2614 gi->last_pos = *pos;
2618 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2621 struct gfs2_glock_iter *gi = seq->private;
2624 gi->last_pos = *pos;
2625 gfs2_glock_iter_next(gi, 1);
2629 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2632 struct gfs2_glock_iter *gi = seq->private;
2634 rhashtable_walk_stop(&gi->hti);
2637 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2639 dump_glock(seq, iter_ptr, false);
2643 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2646 if (*pos >= GFS2_NR_SBSTATS)
2651 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2655 if (*pos >= GFS2_NR_SBSTATS)
2660 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2665 static const struct seq_operations gfs2_glock_seq_ops = {
2666 .start = gfs2_glock_seq_start,
2667 .next = gfs2_glock_seq_next,
2668 .stop = gfs2_glock_seq_stop,
2669 .show = gfs2_glock_seq_show,
2672 static const struct seq_operations gfs2_glstats_seq_ops = {
2673 .start = gfs2_glock_seq_start,
2674 .next = gfs2_glock_seq_next,
2675 .stop = gfs2_glock_seq_stop,
2676 .show = gfs2_glstats_seq_show,
2679 static const struct seq_operations gfs2_sbstats_sops = {
2680 .start = gfs2_sbstats_seq_start,
2681 .next = gfs2_sbstats_seq_next,
2682 .stop = gfs2_sbstats_seq_stop,
2683 .show = gfs2_sbstats_seq_show,
2686 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2688 static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2689 const struct seq_operations *ops)
2691 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2693 struct seq_file *seq = file->private_data;
2694 struct gfs2_glock_iter *gi = seq->private;
2696 gi->sdp = inode->i_private;
2697 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2699 seq->size = GFS2_SEQ_GOODSIZE;
2701 * Initially, we are "before" the first hash table entry; the
2702 * first call to rhashtable_walk_next gets us the first entry.
2706 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2711 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2713 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2716 static int gfs2_glocks_release(struct inode *inode, struct file *file)
2718 struct seq_file *seq = file->private_data;
2719 struct gfs2_glock_iter *gi = seq->private;
2722 gfs2_glock_put(gi->gl);
2723 rhashtable_walk_exit(&gi->hti);
2724 return seq_release_private(inode, file);
2727 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2729 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2732 static const struct file_operations gfs2_glocks_fops = {
2733 .owner = THIS_MODULE,
2734 .open = gfs2_glocks_open,
2736 .llseek = seq_lseek,
2737 .release = gfs2_glocks_release,
2740 static const struct file_operations gfs2_glstats_fops = {
2741 .owner = THIS_MODULE,
2742 .open = gfs2_glstats_open,
2744 .llseek = seq_lseek,
2745 .release = gfs2_glocks_release,
2748 DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats);
2750 void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2752 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2754 debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2757 debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2758 &gfs2_glstats_fops);
2760 debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2761 &gfs2_sbstats_fops);
2764 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2766 debugfs_remove_recursive(sdp->debugfs_dir);
2767 sdp->debugfs_dir = NULL;
2770 void gfs2_register_debugfs(void)
2772 gfs2_root = debugfs_create_dir("gfs2", NULL);
2775 void gfs2_unregister_debugfs(void)
2777 debugfs_remove(gfs2_root);