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
544 * holder for some reason. If this holder is at the
545 * head of the list, it means we have a blocked holder
546 * at the head, so return 1.
548 if (list_is_first(&gh->gh_list, &gl->gl_holders))
553 if (!incompat_holders_demoted) {
554 demote_incompat_holders(gl, first_gh);
555 incompat_holders_demoted = true;
558 if (test_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags) &&
559 !(gh->gh_flags & GL_SKIP) && gl->gl_ops->go_instantiate) {
560 lock_released = true;
561 spin_unlock(&gl->gl_lockref.lock);
562 ret = gfs2_instantiate(gh);
563 spin_lock(&gl->gl_lockref.lock);
568 list_del_init(&gh->gh_list);
569 trace_gfs2_glock_queue(gh, 0);
570 gfs2_holder_wake(gh);
574 set_bit(HIF_HOLDER, &gh->gh_iflags);
575 trace_gfs2_promote(gh);
576 gfs2_holder_wake(gh);
578 * If we released the gl_lockref.lock the holders list may have
579 * changed. For that reason, we start again at the start of
589 * find_first_waiter - find the first gh that's waiting for the glock
593 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
595 struct gfs2_holder *gh;
597 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
598 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
605 * state_change - record that the glock is now in a different state
607 * @new_state: the new state
610 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
614 held1 = (gl->gl_state != LM_ST_UNLOCKED);
615 held2 = (new_state != LM_ST_UNLOCKED);
617 if (held1 != held2) {
618 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
620 gl->gl_lockref.count++;
622 gl->gl_lockref.count--;
624 if (new_state != gl->gl_target)
625 /* shorten our minimum hold time */
626 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
628 gl->gl_state = new_state;
629 gl->gl_tchange = jiffies;
632 static void gfs2_set_demote(struct gfs2_glock *gl)
634 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
636 set_bit(GLF_DEMOTE, &gl->gl_flags);
638 wake_up(&sdp->sd_async_glock_wait);
641 static void gfs2_demote_wake(struct gfs2_glock *gl)
643 gl->gl_demote_state = LM_ST_EXCLUSIVE;
644 clear_bit(GLF_DEMOTE, &gl->gl_flags);
645 smp_mb__after_atomic();
646 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
650 * finish_xmote - The DLM has replied to one of our lock requests
652 * @ret: The status from the DLM
656 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
658 const struct gfs2_glock_operations *glops = gl->gl_ops;
659 struct gfs2_holder *gh;
660 unsigned state = ret & LM_OUT_ST_MASK;
663 spin_lock(&gl->gl_lockref.lock);
664 trace_gfs2_glock_state_change(gl, state);
665 state_change(gl, state);
666 gh = find_first_waiter(gl);
668 /* Demote to UN request arrived during demote to SH or DF */
669 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
670 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
671 gl->gl_target = LM_ST_UNLOCKED;
673 /* Check for state != intended state */
674 if (unlikely(state != gl->gl_target)) {
675 if (gh && (ret & LM_OUT_CANCELED))
676 gfs2_holder_wake(gh);
677 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
678 /* move to back of queue and try next entry */
679 if (ret & LM_OUT_CANCELED) {
680 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
681 list_move_tail(&gh->gh_list, &gl->gl_holders);
682 gh = find_first_waiter(gl);
683 gl->gl_target = gh->gh_state;
686 /* Some error or failed "try lock" - report it */
687 if ((ret & LM_OUT_ERROR) ||
688 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
689 gl->gl_target = gl->gl_state;
695 /* Unlocked due to conversion deadlock, try again */
698 do_xmote(gl, gh, gl->gl_target);
700 /* Conversion fails, unlock and try again */
703 do_xmote(gl, gh, LM_ST_UNLOCKED);
705 default: /* Everything else */
706 fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n",
707 gl->gl_target, state);
710 spin_unlock(&gl->gl_lockref.lock);
714 /* Fast path - we got what we asked for */
715 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
716 gfs2_demote_wake(gl);
717 if (state != LM_ST_UNLOCKED) {
718 if (glops->go_xmote_bh) {
719 spin_unlock(&gl->gl_lockref.lock);
720 rv = glops->go_xmote_bh(gl);
721 spin_lock(&gl->gl_lockref.lock);
732 clear_bit(GLF_LOCK, &gl->gl_flags);
734 spin_unlock(&gl->gl_lockref.lock);
737 static bool is_system_glock(struct gfs2_glock *gl)
739 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
740 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
742 if (gl == m_ip->i_gl)
748 * do_xmote - Calls the DLM to change the state of a lock
749 * @gl: The lock state
750 * @gh: The holder (only for promotes)
751 * @target: The target lock state
755 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
756 __releases(&gl->gl_lockref.lock)
757 __acquires(&gl->gl_lockref.lock)
759 const struct gfs2_glock_operations *glops = gl->gl_ops;
760 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
761 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
764 if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) &&
765 gh && !(gh->gh_flags & LM_FLAG_NOEXP))
767 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
769 GLOCK_BUG_ON(gl, gl->gl_state == target);
770 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
771 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
774 * If another process is already doing the invalidate, let that
775 * finish first. The glock state machine will get back to this
776 * holder again later.
778 if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
781 do_error(gl, 0); /* Fail queued try locks */
784 set_bit(GLF_BLOCKING, &gl->gl_flags);
785 if ((gl->gl_req == LM_ST_UNLOCKED) ||
786 (gl->gl_state == LM_ST_EXCLUSIVE) ||
787 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
788 clear_bit(GLF_BLOCKING, &gl->gl_flags);
789 spin_unlock(&gl->gl_lockref.lock);
790 if (glops->go_sync) {
791 ret = glops->go_sync(gl);
792 /* If we had a problem syncing (due to io errors or whatever,
793 * we should not invalidate the metadata or tell dlm to
794 * release the glock to other nodes.
797 if (cmpxchg(&sdp->sd_log_error, 0, ret)) {
798 fs_err(sdp, "Error %d syncing glock \n", ret);
799 gfs2_dump_glock(NULL, gl, true);
804 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) {
806 * The call to go_sync should have cleared out the ail list.
807 * If there are still items, we have a problem. We ought to
808 * withdraw, but we can't because the withdraw code also uses
809 * glocks. Warn about the error, dump the glock, then fall
810 * through and wait for logd to do the withdraw for us.
812 if ((atomic_read(&gl->gl_ail_count) != 0) &&
813 (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) {
814 gfs2_glock_assert_warn(gl,
815 !atomic_read(&gl->gl_ail_count));
816 gfs2_dump_glock(NULL, gl, true);
818 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
819 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
825 * Check for an error encountered since we called go_sync and go_inval.
826 * If so, we can't withdraw from the glock code because the withdraw
827 * code itself uses glocks (see function signal_our_withdraw) to
828 * change the mount to read-only. Most importantly, we must not call
829 * dlm to unlock the glock until the journal is in a known good state
830 * (after journal replay) otherwise other nodes may use the object
831 * (rgrp or dinode) and then later, journal replay will corrupt the
832 * file system. The best we can do here is wait for the logd daemon
833 * to see sd_log_error and withdraw, and in the meantime, requeue the
836 * We make a special exception for some system glocks, such as the
837 * system statfs inode glock, which needs to be granted before the
838 * gfs2_quotad daemon can exit, and that exit needs to finish before
839 * we can unmount the withdrawn file system.
841 * However, if we're just unlocking the lock (say, for unmount, when
842 * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
843 * then it's okay to tell dlm to unlock it.
845 if (unlikely(sdp->sd_log_error && !gfs2_withdrawn(sdp)))
846 gfs2_withdraw_delayed(sdp);
847 if (glock_blocked_by_withdraw(gl) &&
848 (target != LM_ST_UNLOCKED ||
849 test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags))) {
850 if (!is_system_glock(gl)) {
851 gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD);
854 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
858 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
860 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
861 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
862 target == LM_ST_UNLOCKED &&
863 test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
864 finish_xmote(gl, target);
865 gfs2_glock_queue_work(gl, 0);
867 fs_err(sdp, "lm_lock ret %d\n", ret);
868 GLOCK_BUG_ON(gl, !gfs2_withdrawn(sdp));
870 } else { /* lock_nolock */
871 finish_xmote(gl, target);
872 gfs2_glock_queue_work(gl, 0);
875 spin_lock(&gl->gl_lockref.lock);
879 * run_queue - do all outstanding tasks related to a glock
880 * @gl: The glock in question
881 * @nonblock: True if we must not block in run_queue
885 static void run_queue(struct gfs2_glock *gl, const int nonblock)
886 __releases(&gl->gl_lockref.lock)
887 __acquires(&gl->gl_lockref.lock)
889 struct gfs2_holder *gh = NULL;
892 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
895 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
897 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
898 gl->gl_demote_state != gl->gl_state) {
899 if (find_first_holder(gl))
903 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
904 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
905 gl->gl_target = gl->gl_demote_state;
907 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
908 gfs2_demote_wake(gl);
909 ret = do_promote(gl);
914 gh = find_first_waiter(gl);
915 gl->gl_target = gh->gh_state;
916 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
917 do_error(gl, 0); /* Fail queued try locks */
919 do_xmote(gl, gh, gl->gl_target);
924 clear_bit(GLF_LOCK, &gl->gl_flags);
925 smp_mb__after_atomic();
926 gl->gl_lockref.count++;
927 __gfs2_glock_queue_work(gl, 0);
931 clear_bit(GLF_LOCK, &gl->gl_flags);
932 smp_mb__after_atomic();
936 void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation)
938 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
940 if (ri->ri_magic == 0)
941 ri->ri_magic = cpu_to_be32(GFS2_MAGIC);
942 if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC))
943 ri->ri_generation_deleted = cpu_to_be64(generation);
946 bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation)
948 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
950 if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC))
952 return generation <= be64_to_cpu(ri->ri_generation_deleted);
955 static void gfs2_glock_poke(struct gfs2_glock *gl)
957 int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP;
958 struct gfs2_holder gh;
961 __gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh, _RET_IP_);
962 error = gfs2_glock_nq(&gh);
965 gfs2_holder_uninit(&gh);
968 static bool gfs2_try_evict(struct gfs2_glock *gl)
970 struct gfs2_inode *ip;
971 bool evicted = false;
974 * If there is contention on the iopen glock and we have an inode, try
975 * to grab and release the inode so that it can be evicted. This will
976 * allow the remote node to go ahead and delete the inode without us
977 * having to do it, which will avoid rgrp glock thrashing.
979 * The remote node is likely still holding the corresponding inode
980 * glock, so it will run before we get to verify that the delete has
983 spin_lock(&gl->gl_lockref.lock);
985 if (ip && !igrab(&ip->i_inode))
987 spin_unlock(&gl->gl_lockref.lock);
989 struct gfs2_glock *inode_gl = NULL;
991 gl->gl_no_formal_ino = ip->i_no_formal_ino;
992 set_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
993 d_prune_aliases(&ip->i_inode);
996 /* If the inode was evicted, gl->gl_object will now be NULL. */
997 spin_lock(&gl->gl_lockref.lock);
1000 inode_gl = ip->i_gl;
1001 lockref_get(&inode_gl->gl_lockref);
1002 clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
1004 spin_unlock(&gl->gl_lockref.lock);
1006 gfs2_glock_poke(inode_gl);
1007 gfs2_glock_put(inode_gl);
1014 static void delete_work_func(struct work_struct *work)
1016 struct delayed_work *dwork = to_delayed_work(work);
1017 struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete);
1018 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1019 struct inode *inode;
1020 u64 no_addr = gl->gl_name.ln_number;
1022 spin_lock(&gl->gl_lockref.lock);
1023 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1024 spin_unlock(&gl->gl_lockref.lock);
1026 if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
1028 * If we can evict the inode, give the remote node trying to
1029 * delete the inode some time before verifying that the delete
1030 * has happened. Otherwise, if we cause contention on the inode glock
1031 * immediately, the remote node will think that we still have
1032 * the inode in use, and so it will give up waiting.
1034 * If we can't evict the inode, signal to the remote node that
1035 * the inode is still in use. We'll later try to delete the
1036 * inode locally in gfs2_evict_inode.
1038 * FIXME: We only need to verify that the remote node has
1039 * deleted the inode because nodes before this remote delete
1040 * rework won't cooperate. At a later time, when we no longer
1041 * care about compatibility with such nodes, we can skip this
1044 if (gfs2_try_evict(gl)) {
1045 if (gfs2_queue_delete_work(gl, 5 * HZ))
1051 inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino,
1052 GFS2_BLKST_UNLINKED);
1053 if (!IS_ERR_OR_NULL(inode)) {
1054 d_prune_aliases(inode);
1061 static void glock_work_func(struct work_struct *work)
1063 unsigned long delay = 0;
1064 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
1065 unsigned int drop_refs = 1;
1067 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
1068 finish_xmote(gl, gl->gl_reply);
1071 spin_lock(&gl->gl_lockref.lock);
1072 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1073 gl->gl_state != LM_ST_UNLOCKED &&
1074 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
1075 unsigned long holdtime, now = jiffies;
1077 holdtime = gl->gl_tchange + gl->gl_hold_time;
1078 if (time_before(now, holdtime))
1079 delay = holdtime - now;
1082 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1083 gfs2_set_demote(gl);
1088 /* Keep one glock reference for the work we requeue. */
1090 if (gl->gl_name.ln_type != LM_TYPE_INODE)
1092 __gfs2_glock_queue_work(gl, delay);
1096 * Drop the remaining glock references manually here. (Mind that
1097 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
1100 gl->gl_lockref.count -= drop_refs;
1101 if (!gl->gl_lockref.count) {
1102 __gfs2_glock_put(gl);
1105 spin_unlock(&gl->gl_lockref.lock);
1108 static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
1109 struct gfs2_glock *new)
1111 struct wait_glock_queue wait;
1112 wait_queue_head_t *wq = glock_waitqueue(name);
1113 struct gfs2_glock *gl;
1116 init_wait(&wait.wait);
1117 wait.wait.func = glock_wake_function;
1120 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1123 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
1124 &new->gl_node, ht_parms);
1128 gl = rhashtable_lookup_fast(&gl_hash_table,
1131 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
1138 finish_wait(wq, &wait.wait);
1143 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
1144 * @sdp: The GFS2 superblock
1145 * @number: the lock number
1146 * @glops: The glock_operations to use
1147 * @create: If 0, don't create the glock if it doesn't exist
1148 * @glp: the glock is returned here
1150 * This does not lock a glock, just finds/creates structures for one.
1155 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
1156 const struct gfs2_glock_operations *glops, int create,
1157 struct gfs2_glock **glp)
1159 struct super_block *s = sdp->sd_vfs;
1160 struct lm_lockname name = { .ln_number = number,
1161 .ln_type = glops->go_type,
1163 struct gfs2_glock *gl, *tmp;
1164 struct address_space *mapping;
1167 gl = find_insert_glock(&name, NULL);
1175 if (glops->go_flags & GLOF_ASPACE) {
1176 struct gfs2_glock_aspace *gla =
1177 kmem_cache_alloc(gfs2_glock_aspace_cachep, GFP_NOFS);
1182 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_NOFS);
1186 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
1189 if (glops->go_flags & GLOF_LVB) {
1190 gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1191 if (!gl->gl_lksb.sb_lvbptr) {
1192 gfs2_glock_dealloc(&gl->gl_rcu);
1197 atomic_inc(&sdp->sd_glock_disposal);
1198 gl->gl_node.next = NULL;
1199 gl->gl_flags = glops->go_instantiate ? BIT(GLF_INSTANTIATE_NEEDED) : 0;
1201 lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass);
1202 gl->gl_lockref.count = 1;
1203 gl->gl_state = LM_ST_UNLOCKED;
1204 gl->gl_target = LM_ST_UNLOCKED;
1205 gl->gl_demote_state = LM_ST_EXCLUSIVE;
1208 /* We use the global stats to estimate the initial per-glock stats */
1209 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
1211 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
1212 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
1213 gl->gl_tchange = jiffies;
1214 gl->gl_object = NULL;
1215 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
1216 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
1217 if (gl->gl_name.ln_type == LM_TYPE_IOPEN)
1218 INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func);
1220 mapping = gfs2_glock2aspace(gl);
1222 mapping->a_ops = &gfs2_meta_aops;
1223 mapping->host = s->s_bdev->bd_inode;
1225 mapping_set_gfp_mask(mapping, GFP_NOFS);
1226 mapping->private_data = NULL;
1227 mapping->writeback_index = 0;
1230 tmp = find_insert_glock(&name, gl);
1242 gfs2_glock_dealloc(&gl->gl_rcu);
1243 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
1244 wake_up(&sdp->sd_glock_wait);
1251 * __gfs2_holder_init - initialize a struct gfs2_holder in the default way
1253 * @state: the state we're requesting
1254 * @flags: the modifier flags
1255 * @gh: the holder structure
1259 void __gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
1260 struct gfs2_holder *gh, unsigned long ip)
1262 INIT_LIST_HEAD(&gh->gh_list);
1265 gh->gh_owner_pid = get_pid(task_pid(current));
1266 gh->gh_state = state;
1267 gh->gh_flags = flags;
1269 gfs2_glock_hold(gl);
1273 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1274 * @state: the state we're requesting
1275 * @flags: the modifier flags
1276 * @gh: the holder structure
1278 * Don't mess with the glock.
1282 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
1284 gh->gh_state = state;
1285 gh->gh_flags = flags;
1287 gh->gh_ip = _RET_IP_;
1288 put_pid(gh->gh_owner_pid);
1289 gh->gh_owner_pid = get_pid(task_pid(current));
1293 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1294 * @gh: the holder structure
1298 void gfs2_holder_uninit(struct gfs2_holder *gh)
1300 put_pid(gh->gh_owner_pid);
1301 gfs2_glock_put(gh->gh_gl);
1302 gfs2_holder_mark_uninitialized(gh);
1306 static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
1307 unsigned long start_time)
1309 /* Have we waited longer that a second? */
1310 if (time_after(jiffies, start_time + HZ)) {
1311 /* Lengthen the minimum hold time. */
1312 gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
1318 * gfs2_glock_wait - wait on a glock acquisition
1319 * @gh: the glock holder
1321 * Returns: 0 on success
1324 int gfs2_glock_wait(struct gfs2_holder *gh)
1326 unsigned long start_time = jiffies;
1329 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1330 gfs2_glock_update_hold_time(gh->gh_gl, start_time);
1331 return gh->gh_error;
1334 static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
1338 for (i = 0; i < num_gh; i++)
1339 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
1345 * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1346 * @num_gh: the number of holders in the array
1347 * @ghs: the glock holder array
1349 * Returns: 0 on success, meaning all glocks have been granted and are held.
1350 * -ESTALE if the request timed out, meaning all glocks were released,
1351 * and the caller should retry the operation.
1354 int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
1356 struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
1357 int i, ret = 0, timeout = 0;
1358 unsigned long start_time = jiffies;
1363 * Total up the (minimum hold time * 2) of all glocks and use that to
1364 * determine the max amount of time we should wait.
1366 for (i = 0; i < num_gh; i++)
1367 timeout += ghs[i].gh_gl->gl_hold_time << 1;
1370 if (!wait_event_timeout(sdp->sd_async_glock_wait,
1371 !glocks_pending(num_gh, ghs), timeout))
1372 ret = -ESTALE; /* request timed out. */
1375 * If dlm granted all our requests, we need to adjust the glock
1376 * minimum hold time values according to how long we waited.
1378 * If our request timed out, we need to repeatedly release any held
1379 * glocks we acquired thus far to allow dlm to acquire the remaining
1380 * glocks without deadlocking. We cannot currently cancel outstanding
1381 * glock acquisitions.
1383 * The HIF_WAIT bit tells us which requests still need a response from
1386 * If dlm sent us any errors, we return the first error we find.
1388 keep_waiting = false;
1389 for (i = 0; i < num_gh; i++) {
1390 /* Skip holders we have already dequeued below. */
1391 if (!gfs2_holder_queued(&ghs[i]))
1393 /* Skip holders with a pending DLM response. */
1394 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags)) {
1395 keep_waiting = true;
1399 if (test_bit(HIF_HOLDER, &ghs[i].gh_iflags)) {
1401 gfs2_glock_dq(&ghs[i]);
1403 gfs2_glock_update_hold_time(ghs[i].gh_gl,
1407 ret = ghs[i].gh_error;
1414 * At this point, we've either acquired all locks or released them all.
1420 * handle_callback - process a demote request
1422 * @state: the state the caller wants us to change to
1423 * @delay: zero to demote immediately; otherwise pending demote
1424 * @remote: true if this came from a different cluster node
1426 * There are only two requests that we are going to see in actual
1427 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1430 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
1431 unsigned long delay, bool remote)
1434 set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1436 gfs2_set_demote(gl);
1437 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
1438 gl->gl_demote_state = state;
1439 gl->gl_demote_time = jiffies;
1440 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
1441 gl->gl_demote_state != state) {
1442 gl->gl_demote_state = LM_ST_UNLOCKED;
1444 if (gl->gl_ops->go_callback)
1445 gl->gl_ops->go_callback(gl, remote);
1446 trace_gfs2_demote_rq(gl, remote);
1449 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
1451 struct va_format vaf;
1454 va_start(args, fmt);
1457 seq_vprintf(seq, fmt, args);
1462 pr_err("%pV", &vaf);
1469 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1470 * @gh: the holder structure to add
1472 * Eventually we should move the recursive locking trap to a
1473 * debugging option or something like that. This is the fast
1474 * path and needs to have the minimum number of distractions.
1478 static inline void add_to_queue(struct gfs2_holder *gh)
1479 __releases(&gl->gl_lockref.lock)
1480 __acquires(&gl->gl_lockref.lock)
1482 struct gfs2_glock *gl = gh->gh_gl;
1483 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1484 struct list_head *insert_pt = NULL;
1485 struct gfs2_holder *gh2;
1488 GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1489 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1490 GLOCK_BUG_ON(gl, true);
1492 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1493 if (test_bit(GLF_LOCK, &gl->gl_flags)) {
1494 struct gfs2_holder *first_gh;
1496 first_gh = find_first_strong_holder(gl);
1497 try_futile = !may_grant(gl, first_gh, gh);
1499 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1503 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1504 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
1505 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK) &&
1506 !test_bit(HIF_MAY_DEMOTE, &gh2->gh_iflags)))
1507 goto trap_recursive;
1509 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1511 gh->gh_error = GLR_TRYFAILED;
1512 gfs2_holder_wake(gh);
1515 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1517 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1518 insert_pt = &gh2->gh_list;
1520 trace_gfs2_glock_queue(gh, 1);
1521 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1522 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1523 if (likely(insert_pt == NULL)) {
1524 list_add_tail(&gh->gh_list, &gl->gl_holders);
1525 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1529 list_add_tail(&gh->gh_list, insert_pt);
1531 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
1532 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1533 spin_unlock(&gl->gl_lockref.lock);
1534 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1535 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1536 spin_lock(&gl->gl_lockref.lock);
1541 fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1542 fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1543 fs_err(sdp, "lock type: %d req lock state : %d\n",
1544 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1545 fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1546 fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1547 fs_err(sdp, "lock type: %d req lock state : %d\n",
1548 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1549 gfs2_dump_glock(NULL, gl, true);
1554 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1555 * @gh: the holder structure
1557 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1559 * Returns: 0, GLR_TRYFAILED, or errno on failure
1562 int gfs2_glock_nq(struct gfs2_holder *gh)
1564 struct gfs2_glock *gl = gh->gh_gl;
1567 if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP))
1570 if (test_bit(GLF_LRU, &gl->gl_flags))
1571 gfs2_glock_remove_from_lru(gl);
1574 spin_lock(&gl->gl_lockref.lock);
1576 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1577 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1578 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1579 gl->gl_lockref.count++;
1580 __gfs2_glock_queue_work(gl, 0);
1583 spin_unlock(&gl->gl_lockref.lock);
1585 if (!(gh->gh_flags & GL_ASYNC))
1586 error = gfs2_glock_wait(gh);
1592 * gfs2_glock_poll - poll to see if an async request has been completed
1595 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1598 int gfs2_glock_poll(struct gfs2_holder *gh)
1600 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1603 static inline bool needs_demote(struct gfs2_glock *gl)
1605 return (test_bit(GLF_DEMOTE, &gl->gl_flags) ||
1606 test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags));
1609 static void __gfs2_glock_dq(struct gfs2_holder *gh)
1611 struct gfs2_glock *gl = gh->gh_gl;
1612 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1617 * This while loop is similar to function demote_incompat_holders:
1618 * If the glock is due to be demoted (which may be from another node
1619 * or even if this holder is GL_NOCACHE), the weak holders are
1620 * demoted as well, allowing the glock to be demoted.
1624 * If we're in the process of file system withdraw, we cannot
1625 * just dequeue any glocks until our journal is recovered, lest
1626 * we introduce file system corruption. We need two exceptions
1627 * to this rule: We need to allow unlocking of nondisk glocks
1628 * and the glock for our own journal that needs recovery.
1630 if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) &&
1631 glock_blocked_by_withdraw(gl) &&
1632 gh->gh_gl != sdp->sd_jinode_gl) {
1633 sdp->sd_glock_dqs_held++;
1634 spin_unlock(&gl->gl_lockref.lock);
1636 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
1637 TASK_UNINTERRUPTIBLE);
1638 spin_lock(&gl->gl_lockref.lock);
1642 * This holder should not be cached, so mark it for demote.
1643 * Note: this should be done before the check for needs_demote
1646 if (gh->gh_flags & GL_NOCACHE)
1647 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1649 list_del_init(&gh->gh_list);
1650 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1651 trace_gfs2_glock_queue(gh, 0);
1654 * If there hasn't been a demote request we are done.
1655 * (Let the remaining holders, if any, keep holding it.)
1657 if (!needs_demote(gl)) {
1658 if (list_empty(&gl->gl_holders))
1663 * If we have another strong holder (we cannot auto-demote)
1664 * we are done. It keeps holding it until it is done.
1666 if (find_first_strong_holder(gl))
1670 * If we have a weak holder at the head of the list, it
1671 * (and all others like it) must be auto-demoted. If there
1672 * are no more weak holders, we exit the while loop.
1674 gh = find_first_holder(gl);
1677 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1678 gfs2_glock_add_to_lru(gl);
1680 if (unlikely(!fast_path)) {
1681 gl->gl_lockref.count++;
1682 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1683 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1684 gl->gl_name.ln_type == LM_TYPE_INODE)
1685 delay = gl->gl_hold_time;
1686 __gfs2_glock_queue_work(gl, delay);
1691 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1692 * @gh: the glock holder
1695 void gfs2_glock_dq(struct gfs2_holder *gh)
1697 struct gfs2_glock *gl = gh->gh_gl;
1699 spin_lock(&gl->gl_lockref.lock);
1700 if (list_is_first(&gh->gh_list, &gl->gl_holders) &&
1701 !test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1702 spin_unlock(&gl->gl_lockref.lock);
1703 gl->gl_name.ln_sbd->sd_lockstruct.ls_ops->lm_cancel(gl);
1704 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1705 spin_lock(&gl->gl_lockref.lock);
1708 __gfs2_glock_dq(gh);
1709 spin_unlock(&gl->gl_lockref.lock);
1712 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1714 struct gfs2_glock *gl = gh->gh_gl;
1717 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1721 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1722 * @gh: the holder structure
1726 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1729 gfs2_holder_uninit(gh);
1733 * gfs2_glock_nq_num - acquire a glock based on lock number
1734 * @sdp: the filesystem
1735 * @number: the lock number
1736 * @glops: the glock operations for the type of glock
1737 * @state: the state to acquire the glock in
1738 * @flags: modifier flags for the acquisition
1739 * @gh: the struct gfs2_holder
1744 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1745 const struct gfs2_glock_operations *glops,
1746 unsigned int state, u16 flags, struct gfs2_holder *gh)
1748 struct gfs2_glock *gl;
1751 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1753 error = gfs2_glock_nq_init(gl, state, flags, gh);
1761 * glock_compare - Compare two struct gfs2_glock structures for sorting
1762 * @arg_a: the first structure
1763 * @arg_b: the second structure
1767 static int glock_compare(const void *arg_a, const void *arg_b)
1769 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1770 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1771 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1772 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1774 if (a->ln_number > b->ln_number)
1776 if (a->ln_number < b->ln_number)
1778 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1783 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1784 * @num_gh: the number of structures
1785 * @ghs: an array of struct gfs2_holder structures
1786 * @p: placeholder for the holder structure to pass back
1788 * Returns: 0 on success (all glocks acquired),
1789 * errno on failure (no glocks acquired)
1792 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1793 struct gfs2_holder **p)
1798 for (x = 0; x < num_gh; x++)
1801 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1803 for (x = 0; x < num_gh; x++) {
1804 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1806 error = gfs2_glock_nq(p[x]);
1809 gfs2_glock_dq(p[x]);
1818 * gfs2_glock_nq_m - acquire multiple glocks
1819 * @num_gh: the number of structures
1820 * @ghs: an array of struct gfs2_holder structures
1823 * Returns: 0 on success (all glocks acquired),
1824 * errno on failure (no glocks acquired)
1827 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1829 struct gfs2_holder *tmp[4];
1830 struct gfs2_holder **pph = tmp;
1837 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1838 return gfs2_glock_nq(ghs);
1842 pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *),
1848 error = nq_m_sync(num_gh, ghs, pph);
1857 * gfs2_glock_dq_m - release multiple glocks
1858 * @num_gh: the number of structures
1859 * @ghs: an array of struct gfs2_holder structures
1863 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1866 gfs2_glock_dq(&ghs[num_gh]);
1869 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1871 unsigned long delay = 0;
1872 unsigned long holdtime;
1873 unsigned long now = jiffies;
1875 gfs2_glock_hold(gl);
1876 spin_lock(&gl->gl_lockref.lock);
1877 holdtime = gl->gl_tchange + gl->gl_hold_time;
1878 if (!list_empty(&gl->gl_holders) &&
1879 gl->gl_name.ln_type == LM_TYPE_INODE) {
1880 if (time_before(now, holdtime))
1881 delay = holdtime - now;
1882 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1883 delay = gl->gl_hold_time;
1886 * Note 1: We cannot call demote_incompat_holders from handle_callback
1887 * or gfs2_set_demote due to recursion problems like: gfs2_glock_dq ->
1888 * handle_callback -> demote_incompat_holders -> gfs2_glock_dq
1889 * Plus, we only want to demote the holders if the request comes from
1890 * a remote cluster node because local holder conflicts are resolved
1893 * Note 2: if a remote node wants this glock in EX mode, lock_dlm will
1894 * request that we set our state to UNLOCKED. Here we mock up a holder
1895 * to make it look like someone wants the lock EX locally. Any SH
1896 * and DF requests should be able to share the lock without demoting.
1898 * Note 3: We only want to demote the demoteable holders when there
1899 * are no more strong holders. The demoteable holders might as well
1900 * keep the glock until the last strong holder is done with it.
1902 if (!find_first_strong_holder(gl)) {
1903 struct gfs2_holder mock_gh = {
1905 .gh_state = (state == LM_ST_UNLOCKED) ?
1906 LM_ST_EXCLUSIVE : state,
1907 .gh_iflags = BIT(HIF_HOLDER)
1910 demote_incompat_holders(gl, &mock_gh);
1912 handle_callback(gl, state, delay, true);
1913 __gfs2_glock_queue_work(gl, delay);
1914 spin_unlock(&gl->gl_lockref.lock);
1918 * gfs2_should_freeze - Figure out if glock should be frozen
1919 * @gl: The glock in question
1921 * Glocks are not frozen if (a) the result of the dlm operation is
1922 * an error, (b) the locking operation was an unlock operation or
1923 * (c) if there is a "noexp" flagged request anywhere in the queue
1925 * Returns: 1 if freezing should occur, 0 otherwise
1928 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1930 const struct gfs2_holder *gh;
1932 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1934 if (gl->gl_target == LM_ST_UNLOCKED)
1937 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1938 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1940 if (LM_FLAG_NOEXP & gh->gh_flags)
1948 * gfs2_glock_complete - Callback used by locking
1949 * @gl: Pointer to the glock
1950 * @ret: The return value from the dlm
1952 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1953 * to use a bitfield shared with other glock state fields.
1956 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1958 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1960 spin_lock(&gl->gl_lockref.lock);
1963 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1964 if (gfs2_should_freeze(gl)) {
1965 set_bit(GLF_FROZEN, &gl->gl_flags);
1966 spin_unlock(&gl->gl_lockref.lock);
1971 gl->gl_lockref.count++;
1972 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1973 __gfs2_glock_queue_work(gl, 0);
1974 spin_unlock(&gl->gl_lockref.lock);
1977 static int glock_cmp(void *priv, const struct list_head *a,
1978 const struct list_head *b)
1980 struct gfs2_glock *gla, *glb;
1982 gla = list_entry(a, struct gfs2_glock, gl_lru);
1983 glb = list_entry(b, struct gfs2_glock, gl_lru);
1985 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1987 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1994 * gfs2_dispose_glock_lru - Demote a list of glocks
1995 * @list: The list to dispose of
1997 * Disposing of glocks may involve disk accesses, so that here we sort
1998 * the glocks by number (i.e. disk location of the inodes) so that if
1999 * there are any such accesses, they'll be sent in order (mostly).
2001 * Must be called under the lru_lock, but may drop and retake this
2002 * lock. While the lru_lock is dropped, entries may vanish from the
2003 * list, but no new entries will appear on the list (since it is
2007 static void gfs2_dispose_glock_lru(struct list_head *list)
2008 __releases(&lru_lock)
2009 __acquires(&lru_lock)
2011 struct gfs2_glock *gl;
2013 list_sort(NULL, list, glock_cmp);
2015 while(!list_empty(list)) {
2016 gl = list_first_entry(list, struct gfs2_glock, gl_lru);
2017 list_del_init(&gl->gl_lru);
2018 clear_bit(GLF_LRU, &gl->gl_flags);
2019 if (!spin_trylock(&gl->gl_lockref.lock)) {
2021 list_add(&gl->gl_lru, &lru_list);
2022 set_bit(GLF_LRU, &gl->gl_flags);
2023 atomic_inc(&lru_count);
2026 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
2027 spin_unlock(&gl->gl_lockref.lock);
2028 goto add_back_to_lru;
2030 gl->gl_lockref.count++;
2032 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2033 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
2034 __gfs2_glock_queue_work(gl, 0);
2035 spin_unlock(&gl->gl_lockref.lock);
2036 cond_resched_lock(&lru_lock);
2041 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
2042 * @nr: The number of entries to scan
2044 * This function selects the entries on the LRU which are able to
2045 * be demoted, and then kicks off the process by calling
2046 * gfs2_dispose_glock_lru() above.
2049 static long gfs2_scan_glock_lru(int nr)
2051 struct gfs2_glock *gl;
2056 spin_lock(&lru_lock);
2057 while ((nr-- >= 0) && !list_empty(&lru_list)) {
2058 gl = list_first_entry(&lru_list, struct gfs2_glock, gl_lru);
2060 /* Test for being demotable */
2061 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
2062 list_move(&gl->gl_lru, &dispose);
2063 atomic_dec(&lru_count);
2068 list_move(&gl->gl_lru, &skipped);
2070 list_splice(&skipped, &lru_list);
2071 if (!list_empty(&dispose))
2072 gfs2_dispose_glock_lru(&dispose);
2073 spin_unlock(&lru_lock);
2078 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
2079 struct shrink_control *sc)
2081 if (!(sc->gfp_mask & __GFP_FS))
2083 return gfs2_scan_glock_lru(sc->nr_to_scan);
2086 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
2087 struct shrink_control *sc)
2089 return vfs_pressure_ratio(atomic_read(&lru_count));
2092 static struct shrinker glock_shrinker = {
2093 .seeks = DEFAULT_SEEKS,
2094 .count_objects = gfs2_glock_shrink_count,
2095 .scan_objects = gfs2_glock_shrink_scan,
2099 * glock_hash_walk - Call a function for glock in a hash bucket
2100 * @examiner: the function
2101 * @sdp: the filesystem
2103 * Note that the function can be called multiple times on the same
2104 * object. So the user must ensure that the function can cope with
2108 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
2110 struct gfs2_glock *gl;
2111 struct rhashtable_iter iter;
2113 rhashtable_walk_enter(&gl_hash_table, &iter);
2116 rhashtable_walk_start(&iter);
2118 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl)) {
2119 if (gl->gl_name.ln_sbd == sdp)
2123 rhashtable_walk_stop(&iter);
2124 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
2126 rhashtable_walk_exit(&iter);
2129 bool gfs2_queue_delete_work(struct gfs2_glock *gl, unsigned long delay)
2133 spin_lock(&gl->gl_lockref.lock);
2134 queued = queue_delayed_work(gfs2_delete_workqueue,
2135 &gl->gl_delete, delay);
2137 set_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2138 spin_unlock(&gl->gl_lockref.lock);
2142 void gfs2_cancel_delete_work(struct gfs2_glock *gl)
2144 if (cancel_delayed_work(&gl->gl_delete)) {
2145 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2150 bool gfs2_delete_work_queued(const struct gfs2_glock *gl)
2152 return test_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2155 static void flush_delete_work(struct gfs2_glock *gl)
2157 if (gl->gl_name.ln_type == LM_TYPE_IOPEN) {
2158 if (cancel_delayed_work(&gl->gl_delete)) {
2159 queue_delayed_work(gfs2_delete_workqueue,
2165 void gfs2_flush_delete_work(struct gfs2_sbd *sdp)
2167 glock_hash_walk(flush_delete_work, sdp);
2168 flush_workqueue(gfs2_delete_workqueue);
2172 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
2173 * @gl: The glock to thaw
2177 static void thaw_glock(struct gfs2_glock *gl)
2179 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
2181 if (!lockref_get_not_dead(&gl->gl_lockref))
2183 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
2184 gfs2_glock_queue_work(gl, 0);
2188 * clear_glock - look at a glock and see if we can free it from glock cache
2189 * @gl: the glock to look at
2193 static void clear_glock(struct gfs2_glock *gl)
2195 gfs2_glock_remove_from_lru(gl);
2197 spin_lock(&gl->gl_lockref.lock);
2198 if (!__lockref_is_dead(&gl->gl_lockref)) {
2199 gl->gl_lockref.count++;
2200 if (gl->gl_state != LM_ST_UNLOCKED)
2201 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2202 __gfs2_glock_queue_work(gl, 0);
2204 spin_unlock(&gl->gl_lockref.lock);
2208 * gfs2_glock_thaw - Thaw any frozen glocks
2209 * @sdp: The super block
2213 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
2215 glock_hash_walk(thaw_glock, sdp);
2218 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2220 spin_lock(&gl->gl_lockref.lock);
2221 gfs2_dump_glock(seq, gl, fsid);
2222 spin_unlock(&gl->gl_lockref.lock);
2225 static void dump_glock_func(struct gfs2_glock *gl)
2227 dump_glock(NULL, gl, true);
2231 * gfs2_gl_hash_clear - Empty out the glock hash table
2232 * @sdp: the filesystem
2234 * Called when unmounting the filesystem.
2237 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
2239 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
2240 flush_workqueue(glock_workqueue);
2241 glock_hash_walk(clear_glock, sdp);
2242 flush_workqueue(glock_workqueue);
2243 wait_event_timeout(sdp->sd_glock_wait,
2244 atomic_read(&sdp->sd_glock_disposal) == 0,
2246 glock_hash_walk(dump_glock_func, sdp);
2249 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
2251 struct gfs2_glock *gl = ip->i_gl;
2254 ret = gfs2_truncatei_resume(ip);
2255 gfs2_glock_assert_withdraw(gl, ret == 0);
2257 spin_lock(&gl->gl_lockref.lock);
2258 clear_bit(GLF_LOCK, &gl->gl_flags);
2260 spin_unlock(&gl->gl_lockref.lock);
2263 static const char *state2str(unsigned state)
2266 case LM_ST_UNLOCKED:
2270 case LM_ST_DEFERRED:
2272 case LM_ST_EXCLUSIVE:
2278 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
2281 if (flags & LM_FLAG_TRY)
2283 if (flags & LM_FLAG_TRY_1CB)
2285 if (flags & LM_FLAG_NOEXP)
2287 if (flags & LM_FLAG_ANY)
2289 if (flags & LM_FLAG_PRIORITY)
2291 if (flags & LM_FLAG_NODE_SCOPE)
2293 if (flags & GL_ASYNC)
2295 if (flags & GL_EXACT)
2297 if (flags & GL_NOCACHE)
2299 if (test_bit(HIF_HOLDER, &iflags))
2301 if (test_bit(HIF_WAIT, &iflags))
2303 if (test_bit(HIF_MAY_DEMOTE, &iflags))
2305 if (flags & GL_SKIP)
2312 * dump_holder - print information about a glock holder
2313 * @seq: the seq_file struct
2314 * @gh: the glock holder
2315 * @fs_id_buf: pointer to file system id (if requested)
2319 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
2320 const char *fs_id_buf)
2322 struct task_struct *gh_owner = NULL;
2326 if (gh->gh_owner_pid)
2327 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
2328 gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2329 fs_id_buf, state2str(gh->gh_state),
2330 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
2332 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
2333 gh_owner ? gh_owner->comm : "(ended)",
2338 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
2340 const unsigned long *gflags = &gl->gl_flags;
2343 if (test_bit(GLF_LOCK, gflags))
2345 if (test_bit(GLF_DEMOTE, gflags))
2347 if (test_bit(GLF_PENDING_DEMOTE, gflags))
2349 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
2351 if (test_bit(GLF_DIRTY, gflags))
2353 if (test_bit(GLF_LFLUSH, gflags))
2355 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
2357 if (test_bit(GLF_REPLY_PENDING, gflags))
2359 if (test_bit(GLF_INITIAL, gflags))
2361 if (test_bit(GLF_FROZEN, gflags))
2363 if (!list_empty(&gl->gl_holders))
2365 if (test_bit(GLF_LRU, gflags))
2369 if (test_bit(GLF_BLOCKING, gflags))
2371 if (test_bit(GLF_PENDING_DELETE, gflags))
2373 if (test_bit(GLF_FREEING, gflags))
2375 if (test_bit(GLF_INSTANTIATE_NEEDED, gflags))
2377 if (test_bit(GLF_INSTANTIATE_IN_PROG, gflags))
2384 * gfs2_dump_glock - print information about a glock
2385 * @seq: The seq_file struct
2387 * @fsid: If true, also dump the file system id
2389 * The file format is as follows:
2390 * One line per object, capital letters are used to indicate objects
2391 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2392 * other objects are indented by a single space and follow the glock to
2393 * which they are related. Fields are indicated by lower case letters
2394 * followed by a colon and the field value, except for strings which are in
2395 * [] so that its possible to see if they are composed of spaces for
2396 * example. The field's are n = number (id of the object), f = flags,
2397 * t = type, s = state, r = refcount, e = error, p = pid.
2401 void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2403 const struct gfs2_glock_operations *glops = gl->gl_ops;
2404 unsigned long long dtime;
2405 const struct gfs2_holder *gh;
2406 char gflags_buf[32];
2407 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2408 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2409 unsigned long nrpages = 0;
2411 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
2412 struct address_space *mapping = gfs2_glock2aspace(gl);
2414 nrpages = mapping->nrpages;
2416 memset(fs_id_buf, 0, sizeof(fs_id_buf));
2417 if (fsid && sdp) /* safety precaution */
2418 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2419 dtime = jiffies - gl->gl_demote_time;
2420 dtime *= 1000000/HZ; /* demote time in uSec */
2421 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
2423 gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2424 "v:%d r:%d m:%ld p:%lu\n",
2425 fs_id_buf, state2str(gl->gl_state),
2426 gl->gl_name.ln_type,
2427 (unsigned long long)gl->gl_name.ln_number,
2428 gflags2str(gflags_buf, gl),
2429 state2str(gl->gl_target),
2430 state2str(gl->gl_demote_state), dtime,
2431 atomic_read(&gl->gl_ail_count),
2432 atomic_read(&gl->gl_revokes),
2433 (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages);
2435 list_for_each_entry(gh, &gl->gl_holders, gh_list)
2436 dump_holder(seq, gh, fs_id_buf);
2438 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
2439 glops->go_dump(seq, gl, fs_id_buf);
2442 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
2444 struct gfs2_glock *gl = iter_ptr;
2446 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2447 gl->gl_name.ln_type,
2448 (unsigned long long)gl->gl_name.ln_number,
2449 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
2450 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
2451 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
2452 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
2453 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
2454 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
2455 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
2456 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
2460 static const char *gfs2_gltype[] = {
2474 static const char *gfs2_stype[] = {
2475 [GFS2_LKS_SRTT] = "srtt",
2476 [GFS2_LKS_SRTTVAR] = "srttvar",
2477 [GFS2_LKS_SRTTB] = "srttb",
2478 [GFS2_LKS_SRTTVARB] = "srttvarb",
2479 [GFS2_LKS_SIRT] = "sirt",
2480 [GFS2_LKS_SIRTVAR] = "sirtvar",
2481 [GFS2_LKS_DCOUNT] = "dlm",
2482 [GFS2_LKS_QCOUNT] = "queue",
2485 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2487 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
2489 struct gfs2_sbd *sdp = seq->private;
2490 loff_t pos = *(loff_t *)iter_ptr;
2491 unsigned index = pos >> 3;
2492 unsigned subindex = pos & 0x07;
2495 if (index == 0 && subindex != 0)
2498 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
2499 (index == 0) ? "cpu": gfs2_stype[subindex]);
2501 for_each_possible_cpu(i) {
2502 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
2505 seq_printf(seq, " %15u", i);
2507 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
2508 lkstats[index - 1].stats[subindex]);
2510 seq_putc(seq, '\n');
2514 int __init gfs2_glock_init(void)
2518 ret = rhashtable_init(&gl_hash_table, &ht_parms);
2522 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
2523 WQ_HIGHPRI | WQ_FREEZABLE, 0);
2524 if (!glock_workqueue) {
2525 rhashtable_destroy(&gl_hash_table);
2528 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
2529 WQ_MEM_RECLAIM | WQ_FREEZABLE,
2531 if (!gfs2_delete_workqueue) {
2532 destroy_workqueue(glock_workqueue);
2533 rhashtable_destroy(&gl_hash_table);
2537 ret = register_shrinker(&glock_shrinker);
2539 destroy_workqueue(gfs2_delete_workqueue);
2540 destroy_workqueue(glock_workqueue);
2541 rhashtable_destroy(&gl_hash_table);
2545 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2546 init_waitqueue_head(glock_wait_table + i);
2551 void gfs2_glock_exit(void)
2553 unregister_shrinker(&glock_shrinker);
2554 rhashtable_destroy(&gl_hash_table);
2555 destroy_workqueue(glock_workqueue);
2556 destroy_workqueue(gfs2_delete_workqueue);
2559 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2561 struct gfs2_glock *gl = gi->gl;
2566 if (!lockref_put_not_zero(&gl->gl_lockref))
2567 gfs2_glock_queue_put(gl);
2570 gl = rhashtable_walk_next(&gi->hti);
2571 if (IS_ERR_OR_NULL(gl)) {
2572 if (gl == ERR_PTR(-EAGAIN)) {
2579 if (gl->gl_name.ln_sbd != gi->sdp)
2582 if (!lockref_get_not_dead(&gl->gl_lockref))
2586 if (__lockref_is_dead(&gl->gl_lockref))
2594 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2597 struct gfs2_glock_iter *gi = seq->private;
2601 * We can either stay where we are, skip to the next hash table
2602 * entry, or start from the beginning.
2604 if (*pos < gi->last_pos) {
2605 rhashtable_walk_exit(&gi->hti);
2606 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2609 n = *pos - gi->last_pos;
2612 rhashtable_walk_start(&gi->hti);
2614 gfs2_glock_iter_next(gi, n);
2615 gi->last_pos = *pos;
2619 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2622 struct gfs2_glock_iter *gi = seq->private;
2625 gi->last_pos = *pos;
2626 gfs2_glock_iter_next(gi, 1);
2630 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2633 struct gfs2_glock_iter *gi = seq->private;
2635 rhashtable_walk_stop(&gi->hti);
2638 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2640 dump_glock(seq, iter_ptr, false);
2644 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2647 if (*pos >= GFS2_NR_SBSTATS)
2652 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2656 if (*pos >= GFS2_NR_SBSTATS)
2661 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2666 static const struct seq_operations gfs2_glock_seq_ops = {
2667 .start = gfs2_glock_seq_start,
2668 .next = gfs2_glock_seq_next,
2669 .stop = gfs2_glock_seq_stop,
2670 .show = gfs2_glock_seq_show,
2673 static const struct seq_operations gfs2_glstats_seq_ops = {
2674 .start = gfs2_glock_seq_start,
2675 .next = gfs2_glock_seq_next,
2676 .stop = gfs2_glock_seq_stop,
2677 .show = gfs2_glstats_seq_show,
2680 static const struct seq_operations gfs2_sbstats_sops = {
2681 .start = gfs2_sbstats_seq_start,
2682 .next = gfs2_sbstats_seq_next,
2683 .stop = gfs2_sbstats_seq_stop,
2684 .show = gfs2_sbstats_seq_show,
2687 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2689 static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2690 const struct seq_operations *ops)
2692 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2694 struct seq_file *seq = file->private_data;
2695 struct gfs2_glock_iter *gi = seq->private;
2697 gi->sdp = inode->i_private;
2698 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2700 seq->size = GFS2_SEQ_GOODSIZE;
2702 * Initially, we are "before" the first hash table entry; the
2703 * first call to rhashtable_walk_next gets us the first entry.
2707 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2712 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2714 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2717 static int gfs2_glocks_release(struct inode *inode, struct file *file)
2719 struct seq_file *seq = file->private_data;
2720 struct gfs2_glock_iter *gi = seq->private;
2723 gfs2_glock_put(gi->gl);
2724 rhashtable_walk_exit(&gi->hti);
2725 return seq_release_private(inode, file);
2728 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2730 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2733 static const struct file_operations gfs2_glocks_fops = {
2734 .owner = THIS_MODULE,
2735 .open = gfs2_glocks_open,
2737 .llseek = seq_lseek,
2738 .release = gfs2_glocks_release,
2741 static const struct file_operations gfs2_glstats_fops = {
2742 .owner = THIS_MODULE,
2743 .open = gfs2_glstats_open,
2745 .llseek = seq_lseek,
2746 .release = gfs2_glocks_release,
2749 DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats);
2751 void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2753 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2755 debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2758 debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2759 &gfs2_glstats_fops);
2761 debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2762 &gfs2_sbstats_fops);
2765 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2767 debugfs_remove_recursive(sdp->debugfs_dir);
2768 sdp->debugfs_dir = NULL;
2771 void gfs2_register_debugfs(void)
2773 gfs2_root = debugfs_create_dir("gfs2", NULL);
2776 void gfs2_unregister_debugfs(void)
2778 debugfs_remove(gfs2_root);