2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 #include <linux/sched.h>
13 #include <linux/slab.h>
14 #include <linux/spinlock.h>
15 #include <linux/buffer_head.h>
16 #include <linux/delay.h>
17 #include <linux/sort.h>
18 #include <linux/jhash.h>
19 #include <linux/kallsyms.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <linux/list.h>
22 #include <linux/wait.h>
23 #include <linux/module.h>
24 #include <asm/uaccess.h>
25 #include <linux/seq_file.h>
26 #include <linux/debugfs.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/workqueue.h>
30 #include <linux/jiffies.h>
31 #include <linux/rcupdate.h>
32 #include <linux/rculist_bl.h>
33 #include <linux/bit_spinlock.h>
34 #include <linux/percpu.h>
35 #include <linux/list_sort.h>
36 #include <linux/lockref.h>
37 #include <linux/rhashtable.h>
50 #define CREATE_TRACE_POINTS
51 #include "trace_gfs2.h"
53 struct gfs2_glock_iter {
54 struct gfs2_sbd *sdp; /* incore superblock */
55 struct rhashtable_iter hti; /* rhashtable iterator */
56 struct gfs2_glock *gl; /* current glock struct */
57 loff_t last_pos; /* last position */
60 typedef void (*glock_examiner) (struct gfs2_glock * gl);
62 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
64 static struct dentry *gfs2_root;
65 static struct workqueue_struct *glock_workqueue;
66 struct workqueue_struct *gfs2_delete_workqueue;
67 static LIST_HEAD(lru_list);
68 static atomic_t lru_count = ATOMIC_INIT(0);
69 static DEFINE_SPINLOCK(lru_lock);
71 #define GFS2_GL_HASH_SHIFT 15
72 #define GFS2_GL_HASH_SIZE BIT(GFS2_GL_HASH_SHIFT)
74 static struct rhashtable_params ht_parms = {
75 .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4,
76 .key_len = sizeof(struct lm_lockname),
77 .key_offset = offsetof(struct gfs2_glock, gl_name),
78 .head_offset = offsetof(struct gfs2_glock, gl_node),
81 static struct rhashtable gl_hash_table;
83 static void gfs2_glock_dealloc(struct rcu_head *rcu)
85 struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
87 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
88 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
90 kfree(gl->gl_lksb.sb_lvbptr);
91 kmem_cache_free(gfs2_glock_cachep, gl);
95 void gfs2_glock_free(struct gfs2_glock *gl)
97 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
99 call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
100 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
101 wake_up(&sdp->sd_glock_wait);
105 * gfs2_glock_hold() - increment reference count on glock
106 * @gl: The glock to hold
110 static void gfs2_glock_hold(struct gfs2_glock *gl)
112 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
113 lockref_get(&gl->gl_lockref);
117 * demote_ok - Check to see if it's ok to unlock a glock
120 * Returns: 1 if it's ok
123 static int demote_ok(const struct gfs2_glock *gl)
125 const struct gfs2_glock_operations *glops = gl->gl_ops;
127 if (gl->gl_state == LM_ST_UNLOCKED)
129 if (!list_empty(&gl->gl_holders))
131 if (glops->go_demote_ok)
132 return glops->go_demote_ok(gl);
137 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
139 if (!(gl->gl_ops->go_flags & GLOF_LRU))
142 spin_lock(&lru_lock);
144 list_del(&gl->gl_lru);
145 list_add_tail(&gl->gl_lru, &lru_list);
147 if (!test_bit(GLF_LRU, &gl->gl_flags)) {
148 set_bit(GLF_LRU, &gl->gl_flags);
149 atomic_inc(&lru_count);
152 spin_unlock(&lru_lock);
155 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
157 spin_lock(&lru_lock);
158 if (test_bit(GLF_LRU, &gl->gl_flags)) {
159 list_del_init(&gl->gl_lru);
160 atomic_dec(&lru_count);
161 clear_bit(GLF_LRU, &gl->gl_flags);
163 spin_unlock(&lru_lock);
167 * gfs2_glock_put() - Decrement reference count on glock
168 * @gl: The glock to put
172 void gfs2_glock_put(struct gfs2_glock *gl)
174 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
175 struct address_space *mapping = gfs2_glock2aspace(gl);
177 if (lockref_put_or_lock(&gl->gl_lockref))
180 lockref_mark_dead(&gl->gl_lockref);
182 gfs2_glock_remove_from_lru(gl);
183 spin_unlock(&gl->gl_lockref.lock);
184 rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
185 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
186 GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
187 trace_gfs2_glock_put(gl);
188 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
192 * may_grant - check if its ok to grant a new lock
194 * @gh: The lock request which we wish to grant
196 * Returns: true if its ok to grant the lock
199 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
201 const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
202 if ((gh->gh_state == LM_ST_EXCLUSIVE ||
203 gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
205 if (gl->gl_state == gh->gh_state)
207 if (gh->gh_flags & GL_EXACT)
209 if (gl->gl_state == LM_ST_EXCLUSIVE) {
210 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
212 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
215 if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
220 static void gfs2_holder_wake(struct gfs2_holder *gh)
222 clear_bit(HIF_WAIT, &gh->gh_iflags);
223 smp_mb__after_atomic();
224 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
228 * do_error - Something unexpected has happened during a lock request
232 static void do_error(struct gfs2_glock *gl, const int ret)
234 struct gfs2_holder *gh, *tmp;
236 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
237 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
239 if (ret & LM_OUT_ERROR)
241 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
242 gh->gh_error = GLR_TRYFAILED;
245 list_del_init(&gh->gh_list);
246 trace_gfs2_glock_queue(gh, 0);
247 gfs2_holder_wake(gh);
252 * do_promote - promote as many requests as possible on the current queue
255 * Returns: 1 if there is a blocked holder at the head of the list, or 2
256 * if a type specific operation is underway.
259 static int do_promote(struct gfs2_glock *gl)
260 __releases(&gl->gl_lockref.lock)
261 __acquires(&gl->gl_lockref.lock)
263 const struct gfs2_glock_operations *glops = gl->gl_ops;
264 struct gfs2_holder *gh, *tmp;
268 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
269 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
271 if (may_grant(gl, gh)) {
272 if (gh->gh_list.prev == &gl->gl_holders &&
274 spin_unlock(&gl->gl_lockref.lock);
275 /* FIXME: eliminate this eventually */
276 ret = glops->go_lock(gh);
277 spin_lock(&gl->gl_lockref.lock);
282 list_del_init(&gh->gh_list);
283 trace_gfs2_glock_queue(gh, 0);
284 gfs2_holder_wake(gh);
287 set_bit(HIF_HOLDER, &gh->gh_iflags);
288 trace_gfs2_promote(gh, 1);
289 gfs2_holder_wake(gh);
292 set_bit(HIF_HOLDER, &gh->gh_iflags);
293 trace_gfs2_promote(gh, 0);
294 gfs2_holder_wake(gh);
297 if (gh->gh_list.prev == &gl->gl_holders)
306 * find_first_waiter - find the first gh that's waiting for the glock
310 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
312 struct gfs2_holder *gh;
314 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
315 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
322 * state_change - record that the glock is now in a different state
324 * @new_state the new state
328 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
332 held1 = (gl->gl_state != LM_ST_UNLOCKED);
333 held2 = (new_state != LM_ST_UNLOCKED);
335 if (held1 != held2) {
336 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
338 gl->gl_lockref.count++;
340 gl->gl_lockref.count--;
342 if (held1 && held2 && list_empty(&gl->gl_holders))
343 clear_bit(GLF_QUEUED, &gl->gl_flags);
345 if (new_state != gl->gl_target)
346 /* shorten our minimum hold time */
347 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
349 gl->gl_state = new_state;
350 gl->gl_tchange = jiffies;
353 static void gfs2_demote_wake(struct gfs2_glock *gl)
355 gl->gl_demote_state = LM_ST_EXCLUSIVE;
356 clear_bit(GLF_DEMOTE, &gl->gl_flags);
357 smp_mb__after_atomic();
358 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
362 * finish_xmote - The DLM has replied to one of our lock requests
364 * @ret: The status from the DLM
368 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
370 const struct gfs2_glock_operations *glops = gl->gl_ops;
371 struct gfs2_holder *gh;
372 unsigned state = ret & LM_OUT_ST_MASK;
375 spin_lock(&gl->gl_lockref.lock);
376 trace_gfs2_glock_state_change(gl, state);
377 state_change(gl, state);
378 gh = find_first_waiter(gl);
380 /* Demote to UN request arrived during demote to SH or DF */
381 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
382 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
383 gl->gl_target = LM_ST_UNLOCKED;
385 /* Check for state != intended state */
386 if (unlikely(state != gl->gl_target)) {
387 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
388 /* move to back of queue and try next entry */
389 if (ret & LM_OUT_CANCELED) {
390 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
391 list_move_tail(&gh->gh_list, &gl->gl_holders);
392 gh = find_first_waiter(gl);
393 gl->gl_target = gh->gh_state;
396 /* Some error or failed "try lock" - report it */
397 if ((ret & LM_OUT_ERROR) ||
398 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
399 gl->gl_target = gl->gl_state;
405 /* Unlocked due to conversion deadlock, try again */
408 do_xmote(gl, gh, gl->gl_target);
410 /* Conversion fails, unlock and try again */
413 do_xmote(gl, gh, LM_ST_UNLOCKED);
415 default: /* Everything else */
416 pr_err("wanted %u got %u\n", gl->gl_target, state);
419 spin_unlock(&gl->gl_lockref.lock);
423 /* Fast path - we got what we asked for */
424 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
425 gfs2_demote_wake(gl);
426 if (state != LM_ST_UNLOCKED) {
427 if (glops->go_xmote_bh) {
428 spin_unlock(&gl->gl_lockref.lock);
429 rv = glops->go_xmote_bh(gl, gh);
430 spin_lock(&gl->gl_lockref.lock);
441 clear_bit(GLF_LOCK, &gl->gl_flags);
443 spin_unlock(&gl->gl_lockref.lock);
447 * do_xmote - Calls the DLM to change the state of a lock
448 * @gl: The lock state
449 * @gh: The holder (only for promotes)
450 * @target: The target lock state
454 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
455 __releases(&gl->gl_lockref.lock)
456 __acquires(&gl->gl_lockref.lock)
458 const struct gfs2_glock_operations *glops = gl->gl_ops;
459 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
460 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
463 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
465 GLOCK_BUG_ON(gl, gl->gl_state == target);
466 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
467 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
469 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
470 do_error(gl, 0); /* Fail queued try locks */
473 set_bit(GLF_BLOCKING, &gl->gl_flags);
474 if ((gl->gl_req == LM_ST_UNLOCKED) ||
475 (gl->gl_state == LM_ST_EXCLUSIVE) ||
476 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
477 clear_bit(GLF_BLOCKING, &gl->gl_flags);
478 spin_unlock(&gl->gl_lockref.lock);
481 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
482 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
483 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
486 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
488 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
489 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
490 target == LM_ST_UNLOCKED &&
491 test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
492 finish_xmote(gl, target);
493 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
497 pr_err("lm_lock ret %d\n", ret);
500 } else { /* lock_nolock */
501 finish_xmote(gl, target);
502 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
506 spin_lock(&gl->gl_lockref.lock);
510 * find_first_holder - find the first "holder" gh
514 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
516 struct gfs2_holder *gh;
518 if (!list_empty(&gl->gl_holders)) {
519 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
520 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
527 * run_queue - do all outstanding tasks related to a glock
528 * @gl: The glock in question
529 * @nonblock: True if we must not block in run_queue
533 static void run_queue(struct gfs2_glock *gl, const int nonblock)
534 __releases(&gl->gl_lockref.lock)
535 __acquires(&gl->gl_lockref.lock)
537 struct gfs2_holder *gh = NULL;
540 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
543 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
545 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
546 gl->gl_demote_state != gl->gl_state) {
547 if (find_first_holder(gl))
551 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
552 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
553 gl->gl_target = gl->gl_demote_state;
555 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
556 gfs2_demote_wake(gl);
557 ret = do_promote(gl);
562 gh = find_first_waiter(gl);
563 gl->gl_target = gh->gh_state;
564 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
565 do_error(gl, 0); /* Fail queued try locks */
567 do_xmote(gl, gh, gl->gl_target);
572 clear_bit(GLF_LOCK, &gl->gl_flags);
573 smp_mb__after_atomic();
574 gl->gl_lockref.count++;
575 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
576 gl->gl_lockref.count--;
580 clear_bit(GLF_LOCK, &gl->gl_flags);
581 smp_mb__after_atomic();
585 static void delete_work_func(struct work_struct *work)
587 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
588 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
590 u64 no_addr = gl->gl_name.ln_number;
592 /* If someone's using this glock to create a new dinode, the block must
593 have been freed by another node, then re-used, in which case our
594 iopen callback is too late after the fact. Ignore it. */
595 if (test_bit(GLF_INODE_CREATING, &gl->gl_flags))
598 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
599 if (inode && !IS_ERR(inode)) {
600 d_prune_aliases(inode);
607 static void glock_work_func(struct work_struct *work)
609 unsigned long delay = 0;
610 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
613 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
614 finish_xmote(gl, gl->gl_reply);
617 spin_lock(&gl->gl_lockref.lock);
618 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
619 gl->gl_state != LM_ST_UNLOCKED &&
620 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
621 unsigned long holdtime, now = jiffies;
623 holdtime = gl->gl_tchange + gl->gl_hold_time;
624 if (time_before(now, holdtime))
625 delay = holdtime - now;
628 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
629 set_bit(GLF_DEMOTE, &gl->gl_flags);
633 spin_unlock(&gl->gl_lockref.lock);
637 if (gl->gl_name.ln_type != LM_TYPE_INODE)
639 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
647 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
648 * @sdp: The GFS2 superblock
649 * @number: the lock number
650 * @glops: The glock_operations to use
651 * @create: If 0, don't create the glock if it doesn't exist
652 * @glp: the glock is returned here
654 * This does not lock a glock, just finds/creates structures for one.
659 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
660 const struct gfs2_glock_operations *glops, int create,
661 struct gfs2_glock **glp)
663 struct super_block *s = sdp->sd_vfs;
664 struct lm_lockname name = { .ln_number = number,
665 .ln_type = glops->go_type,
667 struct gfs2_glock *gl, *tmp = NULL;
668 struct address_space *mapping;
669 struct kmem_cache *cachep;
673 gl = rhashtable_lookup_fast(&gl_hash_table, &name, ht_parms);
674 if (gl && !lockref_get_not_dead(&gl->gl_lockref))
684 if (glops->go_flags & GLOF_ASPACE)
685 cachep = gfs2_glock_aspace_cachep;
687 cachep = gfs2_glock_cachep;
688 gl = kmem_cache_alloc(cachep, GFP_NOFS);
692 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
694 if (glops->go_flags & GLOF_LVB) {
695 gl->gl_lksb.sb_lvbptr = kzalloc(GFS2_MIN_LVB_SIZE, GFP_NOFS);
696 if (!gl->gl_lksb.sb_lvbptr) {
697 kmem_cache_free(cachep, gl);
702 atomic_inc(&sdp->sd_glock_disposal);
703 gl->gl_node.next = NULL;
706 gl->gl_lockref.count = 1;
707 gl->gl_state = LM_ST_UNLOCKED;
708 gl->gl_target = LM_ST_UNLOCKED;
709 gl->gl_demote_state = LM_ST_EXCLUSIVE;
711 gl->gl_dstamp = ktime_set(0, 0);
713 /* We use the global stats to estimate the initial per-glock stats */
714 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
716 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
717 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
718 gl->gl_tchange = jiffies;
719 gl->gl_object = NULL;
720 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
721 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
722 INIT_WORK(&gl->gl_delete, delete_work_func);
724 mapping = gfs2_glock2aspace(gl);
726 mapping->a_ops = &gfs2_meta_aops;
727 mapping->host = s->s_bdev->bd_inode;
729 mapping_set_gfp_mask(mapping, GFP_NOFS);
730 mapping->private_data = NULL;
731 mapping->writeback_index = 0;
735 ret = rhashtable_lookup_insert_fast(&gl_hash_table, &gl->gl_node,
742 if (ret == -EEXIST) {
745 tmp = rhashtable_lookup_fast(&gl_hash_table, &name, ht_parms);
746 if (tmp == NULL || !lockref_get_not_dead(&tmp->gl_lockref)) {
759 kfree(gl->gl_lksb.sb_lvbptr);
760 kmem_cache_free(cachep, gl);
761 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
762 wake_up(&sdp->sd_glock_wait);
769 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
771 * @state: the state we're requesting
772 * @flags: the modifier flags
773 * @gh: the holder structure
777 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
778 struct gfs2_holder *gh)
780 INIT_LIST_HEAD(&gh->gh_list);
782 gh->gh_ip = _RET_IP_;
783 gh->gh_owner_pid = get_pid(task_pid(current));
784 gh->gh_state = state;
785 gh->gh_flags = flags;
792 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
793 * @state: the state we're requesting
794 * @flags: the modifier flags
795 * @gh: the holder structure
797 * Don't mess with the glock.
801 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
803 gh->gh_state = state;
804 gh->gh_flags = flags;
806 gh->gh_ip = _RET_IP_;
807 put_pid(gh->gh_owner_pid);
808 gh->gh_owner_pid = get_pid(task_pid(current));
812 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
813 * @gh: the holder structure
817 void gfs2_holder_uninit(struct gfs2_holder *gh)
819 put_pid(gh->gh_owner_pid);
820 gfs2_glock_put(gh->gh_gl);
821 gfs2_holder_mark_uninitialized(gh);
826 * gfs2_glock_wait - wait on a glock acquisition
827 * @gh: the glock holder
829 * Returns: 0 on success
832 int gfs2_glock_wait(struct gfs2_holder *gh)
834 unsigned long time1 = jiffies;
837 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
838 if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
839 /* Lengthen the minimum hold time. */
840 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
847 * handle_callback - process a demote request
849 * @state: the state the caller wants us to change to
851 * There are only two requests that we are going to see in actual
852 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
855 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
856 unsigned long delay, bool remote)
858 int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
860 set_bit(bit, &gl->gl_flags);
861 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
862 gl->gl_demote_state = state;
863 gl->gl_demote_time = jiffies;
864 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
865 gl->gl_demote_state != state) {
866 gl->gl_demote_state = LM_ST_UNLOCKED;
868 if (gl->gl_ops->go_callback)
869 gl->gl_ops->go_callback(gl, remote);
870 trace_gfs2_demote_rq(gl, remote);
873 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
875 struct va_format vaf;
881 seq_vprintf(seq, fmt, args);
893 * add_to_queue - Add a holder to the wait queue (but look for recursion)
894 * @gh: the holder structure to add
896 * Eventually we should move the recursive locking trap to a
897 * debugging option or something like that. This is the fast
898 * path and needs to have the minimum number of distractions.
902 static inline void add_to_queue(struct gfs2_holder *gh)
903 __releases(&gl->gl_lockref.lock)
904 __acquires(&gl->gl_lockref.lock)
906 struct gfs2_glock *gl = gh->gh_gl;
907 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
908 struct list_head *insert_pt = NULL;
909 struct gfs2_holder *gh2;
912 BUG_ON(gh->gh_owner_pid == NULL);
913 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
916 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
917 if (test_bit(GLF_LOCK, &gl->gl_flags))
918 try_futile = !may_grant(gl, gh);
919 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
923 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
924 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
925 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
928 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
930 gh->gh_error = GLR_TRYFAILED;
931 gfs2_holder_wake(gh);
934 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
936 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
937 insert_pt = &gh2->gh_list;
939 set_bit(GLF_QUEUED, &gl->gl_flags);
940 trace_gfs2_glock_queue(gh, 1);
941 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
942 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
943 if (likely(insert_pt == NULL)) {
944 list_add_tail(&gh->gh_list, &gl->gl_holders);
945 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
949 list_add_tail(&gh->gh_list, insert_pt);
951 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
952 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
953 spin_unlock(&gl->gl_lockref.lock);
954 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
955 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
956 spin_lock(&gl->gl_lockref.lock);
961 pr_err("original: %pSR\n", (void *)gh2->gh_ip);
962 pr_err("pid: %d\n", pid_nr(gh2->gh_owner_pid));
963 pr_err("lock type: %d req lock state : %d\n",
964 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
965 pr_err("new: %pSR\n", (void *)gh->gh_ip);
966 pr_err("pid: %d\n", pid_nr(gh->gh_owner_pid));
967 pr_err("lock type: %d req lock state : %d\n",
968 gh->gh_gl->gl_name.ln_type, gh->gh_state);
969 gfs2_dump_glock(NULL, gl);
974 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
975 * @gh: the holder structure
977 * if (gh->gh_flags & GL_ASYNC), this never returns an error
979 * Returns: 0, GLR_TRYFAILED, or errno on failure
982 int gfs2_glock_nq(struct gfs2_holder *gh)
984 struct gfs2_glock *gl = gh->gh_gl;
985 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
988 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
991 if (test_bit(GLF_LRU, &gl->gl_flags))
992 gfs2_glock_remove_from_lru(gl);
994 spin_lock(&gl->gl_lockref.lock);
996 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
997 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
998 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
999 gl->gl_lockref.count++;
1000 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1001 gl->gl_lockref.count--;
1004 spin_unlock(&gl->gl_lockref.lock);
1006 if (!(gh->gh_flags & GL_ASYNC))
1007 error = gfs2_glock_wait(gh);
1013 * gfs2_glock_poll - poll to see if an async request has been completed
1016 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1019 int gfs2_glock_poll(struct gfs2_holder *gh)
1021 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1025 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1026 * @gh: the glock holder
1030 void gfs2_glock_dq(struct gfs2_holder *gh)
1032 struct gfs2_glock *gl = gh->gh_gl;
1033 const struct gfs2_glock_operations *glops = gl->gl_ops;
1037 spin_lock(&gl->gl_lockref.lock);
1038 if (gh->gh_flags & GL_NOCACHE)
1039 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1041 list_del_init(&gh->gh_list);
1042 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1043 if (find_first_holder(gl) == NULL) {
1044 if (glops->go_unlock) {
1045 GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1046 spin_unlock(&gl->gl_lockref.lock);
1047 glops->go_unlock(gh);
1048 spin_lock(&gl->gl_lockref.lock);
1049 clear_bit(GLF_LOCK, &gl->gl_flags);
1051 if (list_empty(&gl->gl_holders) &&
1052 !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1053 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1056 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1057 gfs2_glock_add_to_lru(gl);
1059 trace_gfs2_glock_queue(gh, 0);
1060 spin_unlock(&gl->gl_lockref.lock);
1061 if (likely(fast_path))
1064 gfs2_glock_hold(gl);
1065 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1066 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1067 gl->gl_name.ln_type == LM_TYPE_INODE)
1068 delay = gl->gl_hold_time;
1069 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1073 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1075 struct gfs2_glock *gl = gh->gh_gl;
1078 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1082 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1083 * @gh: the holder structure
1087 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1090 gfs2_holder_uninit(gh);
1094 * gfs2_glock_nq_num - acquire a glock based on lock number
1095 * @sdp: the filesystem
1096 * @number: the lock number
1097 * @glops: the glock operations for the type of glock
1098 * @state: the state to acquire the glock in
1099 * @flags: modifier flags for the acquisition
1100 * @gh: the struct gfs2_holder
1105 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1106 const struct gfs2_glock_operations *glops,
1107 unsigned int state, u16 flags, struct gfs2_holder *gh)
1109 struct gfs2_glock *gl;
1112 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1114 error = gfs2_glock_nq_init(gl, state, flags, gh);
1122 * glock_compare - Compare two struct gfs2_glock structures for sorting
1123 * @arg_a: the first structure
1124 * @arg_b: the second structure
1128 static int glock_compare(const void *arg_a, const void *arg_b)
1130 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1131 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1132 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1133 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1135 if (a->ln_number > b->ln_number)
1137 if (a->ln_number < b->ln_number)
1139 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1144 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1145 * @num_gh: the number of structures
1146 * @ghs: an array of struct gfs2_holder structures
1148 * Returns: 0 on success (all glocks acquired),
1149 * errno on failure (no glocks acquired)
1152 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1153 struct gfs2_holder **p)
1158 for (x = 0; x < num_gh; x++)
1161 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1163 for (x = 0; x < num_gh; x++) {
1164 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1166 error = gfs2_glock_nq(p[x]);
1169 gfs2_glock_dq(p[x]);
1178 * gfs2_glock_nq_m - acquire multiple glocks
1179 * @num_gh: the number of structures
1180 * @ghs: an array of struct gfs2_holder structures
1183 * Returns: 0 on success (all glocks acquired),
1184 * errno on failure (no glocks acquired)
1187 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1189 struct gfs2_holder *tmp[4];
1190 struct gfs2_holder **pph = tmp;
1197 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1198 return gfs2_glock_nq(ghs);
1202 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1207 error = nq_m_sync(num_gh, ghs, pph);
1216 * gfs2_glock_dq_m - release multiple glocks
1217 * @num_gh: the number of structures
1218 * @ghs: an array of struct gfs2_holder structures
1222 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1225 gfs2_glock_dq(&ghs[num_gh]);
1228 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1230 unsigned long delay = 0;
1231 unsigned long holdtime;
1232 unsigned long now = jiffies;
1234 gfs2_glock_hold(gl);
1235 holdtime = gl->gl_tchange + gl->gl_hold_time;
1236 if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1237 gl->gl_name.ln_type == LM_TYPE_INODE) {
1238 if (time_before(now, holdtime))
1239 delay = holdtime - now;
1240 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1241 delay = gl->gl_hold_time;
1244 spin_lock(&gl->gl_lockref.lock);
1245 handle_callback(gl, state, delay, true);
1246 spin_unlock(&gl->gl_lockref.lock);
1247 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1252 * gfs2_should_freeze - Figure out if glock should be frozen
1253 * @gl: The glock in question
1255 * Glocks are not frozen if (a) the result of the dlm operation is
1256 * an error, (b) the locking operation was an unlock operation or
1257 * (c) if there is a "noexp" flagged request anywhere in the queue
1259 * Returns: 1 if freezing should occur, 0 otherwise
1262 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1264 const struct gfs2_holder *gh;
1266 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1268 if (gl->gl_target == LM_ST_UNLOCKED)
1271 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1272 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1274 if (LM_FLAG_NOEXP & gh->gh_flags)
1282 * gfs2_glock_complete - Callback used by locking
1283 * @gl: Pointer to the glock
1284 * @ret: The return value from the dlm
1286 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1287 * to use a bitfield shared with other glock state fields.
1290 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1292 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1294 spin_lock(&gl->gl_lockref.lock);
1297 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1298 if (gfs2_should_freeze(gl)) {
1299 set_bit(GLF_FROZEN, &gl->gl_flags);
1300 spin_unlock(&gl->gl_lockref.lock);
1305 gl->gl_lockref.count++;
1306 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1307 spin_unlock(&gl->gl_lockref.lock);
1309 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1313 static int glock_cmp(void *priv, struct list_head *a, struct list_head *b)
1315 struct gfs2_glock *gla, *glb;
1317 gla = list_entry(a, struct gfs2_glock, gl_lru);
1318 glb = list_entry(b, struct gfs2_glock, gl_lru);
1320 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1322 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1329 * gfs2_dispose_glock_lru - Demote a list of glocks
1330 * @list: The list to dispose of
1332 * Disposing of glocks may involve disk accesses, so that here we sort
1333 * the glocks by number (i.e. disk location of the inodes) so that if
1334 * there are any such accesses, they'll be sent in order (mostly).
1336 * Must be called under the lru_lock, but may drop and retake this
1337 * lock. While the lru_lock is dropped, entries may vanish from the
1338 * list, but no new entries will appear on the list (since it is
1342 static void gfs2_dispose_glock_lru(struct list_head *list)
1343 __releases(&lru_lock)
1344 __acquires(&lru_lock)
1346 struct gfs2_glock *gl;
1348 list_sort(NULL, list, glock_cmp);
1350 while(!list_empty(list)) {
1351 gl = list_entry(list->next, struct gfs2_glock, gl_lru);
1352 list_del_init(&gl->gl_lru);
1353 clear_bit(GLF_LRU, &gl->gl_flags);
1354 if (!spin_trylock(&gl->gl_lockref.lock)) {
1356 list_add(&gl->gl_lru, &lru_list);
1357 set_bit(GLF_LRU, &gl->gl_flags);
1358 atomic_inc(&lru_count);
1361 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1362 spin_unlock(&gl->gl_lockref.lock);
1363 goto add_back_to_lru;
1365 gl->gl_lockref.count++;
1367 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1368 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1369 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1370 gl->gl_lockref.count--;
1371 spin_unlock(&gl->gl_lockref.lock);
1372 cond_resched_lock(&lru_lock);
1377 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1378 * @nr: The number of entries to scan
1380 * This function selects the entries on the LRU which are able to
1381 * be demoted, and then kicks off the process by calling
1382 * gfs2_dispose_glock_lru() above.
1385 static long gfs2_scan_glock_lru(int nr)
1387 struct gfs2_glock *gl;
1392 spin_lock(&lru_lock);
1393 while ((nr-- >= 0) && !list_empty(&lru_list)) {
1394 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1396 /* Test for being demotable */
1397 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1398 list_move(&gl->gl_lru, &dispose);
1399 atomic_dec(&lru_count);
1404 list_move(&gl->gl_lru, &skipped);
1406 list_splice(&skipped, &lru_list);
1407 if (!list_empty(&dispose))
1408 gfs2_dispose_glock_lru(&dispose);
1409 spin_unlock(&lru_lock);
1414 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1415 struct shrink_control *sc)
1417 if (!(sc->gfp_mask & __GFP_FS))
1419 return gfs2_scan_glock_lru(sc->nr_to_scan);
1422 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1423 struct shrink_control *sc)
1425 return vfs_pressure_ratio(atomic_read(&lru_count));
1428 static struct shrinker glock_shrinker = {
1429 .seeks = DEFAULT_SEEKS,
1430 .count_objects = gfs2_glock_shrink_count,
1431 .scan_objects = gfs2_glock_shrink_scan,
1435 * examine_bucket - Call a function for glock in a hash bucket
1436 * @examiner: the function
1437 * @sdp: the filesystem
1438 * @bucket: the bucket
1440 * Note that the function can be called multiple times on the same
1441 * object. So the user must ensure that the function can cope with
1445 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1447 struct gfs2_glock *gl;
1448 struct rhashtable_iter iter;
1450 rhashtable_walk_enter(&gl_hash_table, &iter);
1453 gl = ERR_PTR(rhashtable_walk_start(&iter));
1457 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl))
1458 if ((gl->gl_name.ln_sbd == sdp) &&
1459 lockref_get_not_dead(&gl->gl_lockref))
1462 rhashtable_walk_stop(&iter);
1463 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
1465 rhashtable_walk_exit(&iter);
1469 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1470 * @gl: The glock to thaw
1474 static void thaw_glock(struct gfs2_glock *gl)
1476 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1478 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1479 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0) {
1486 * clear_glock - look at a glock and see if we can free it from glock cache
1487 * @gl: the glock to look at
1491 static void clear_glock(struct gfs2_glock *gl)
1493 gfs2_glock_remove_from_lru(gl);
1495 spin_lock(&gl->gl_lockref.lock);
1496 if (gl->gl_state != LM_ST_UNLOCKED)
1497 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1498 spin_unlock(&gl->gl_lockref.lock);
1499 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1504 * gfs2_glock_thaw - Thaw any frozen glocks
1505 * @sdp: The super block
1509 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1511 glock_hash_walk(thaw_glock, sdp);
1514 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1516 spin_lock(&gl->gl_lockref.lock);
1517 gfs2_dump_glock(seq, gl);
1518 spin_unlock(&gl->gl_lockref.lock);
1521 static void dump_glock_func(struct gfs2_glock *gl)
1523 dump_glock(NULL, gl);
1527 * gfs2_gl_hash_clear - Empty out the glock hash table
1528 * @sdp: the filesystem
1529 * @wait: wait until it's all gone
1531 * Called when unmounting the filesystem.
1534 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1536 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1537 flush_workqueue(glock_workqueue);
1538 glock_hash_walk(clear_glock, sdp);
1539 flush_workqueue(glock_workqueue);
1540 wait_event_timeout(sdp->sd_glock_wait,
1541 atomic_read(&sdp->sd_glock_disposal) == 0,
1543 glock_hash_walk(dump_glock_func, sdp);
1546 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1548 struct gfs2_glock *gl = ip->i_gl;
1551 ret = gfs2_truncatei_resume(ip);
1552 gfs2_assert_withdraw(gl->gl_name.ln_sbd, ret == 0);
1554 spin_lock(&gl->gl_lockref.lock);
1555 clear_bit(GLF_LOCK, &gl->gl_flags);
1557 spin_unlock(&gl->gl_lockref.lock);
1560 static const char *state2str(unsigned state)
1563 case LM_ST_UNLOCKED:
1567 case LM_ST_DEFERRED:
1569 case LM_ST_EXCLUSIVE:
1575 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
1578 if (flags & LM_FLAG_TRY)
1580 if (flags & LM_FLAG_TRY_1CB)
1582 if (flags & LM_FLAG_NOEXP)
1584 if (flags & LM_FLAG_ANY)
1586 if (flags & LM_FLAG_PRIORITY)
1588 if (flags & GL_ASYNC)
1590 if (flags & GL_EXACT)
1592 if (flags & GL_NOCACHE)
1594 if (test_bit(HIF_HOLDER, &iflags))
1596 if (test_bit(HIF_WAIT, &iflags))
1598 if (test_bit(HIF_FIRST, &iflags))
1605 * dump_holder - print information about a glock holder
1606 * @seq: the seq_file struct
1607 * @gh: the glock holder
1611 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1613 struct task_struct *gh_owner = NULL;
1617 if (gh->gh_owner_pid)
1618 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1619 gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1620 state2str(gh->gh_state),
1621 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1623 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1624 gh_owner ? gh_owner->comm : "(ended)",
1629 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1631 const unsigned long *gflags = &gl->gl_flags;
1634 if (test_bit(GLF_LOCK, gflags))
1636 if (test_bit(GLF_DEMOTE, gflags))
1638 if (test_bit(GLF_PENDING_DEMOTE, gflags))
1640 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1642 if (test_bit(GLF_DIRTY, gflags))
1644 if (test_bit(GLF_LFLUSH, gflags))
1646 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1648 if (test_bit(GLF_REPLY_PENDING, gflags))
1650 if (test_bit(GLF_INITIAL, gflags))
1652 if (test_bit(GLF_FROZEN, gflags))
1654 if (test_bit(GLF_QUEUED, gflags))
1656 if (test_bit(GLF_LRU, gflags))
1660 if (test_bit(GLF_BLOCKING, gflags))
1667 * gfs2_dump_glock - print information about a glock
1668 * @seq: The seq_file struct
1671 * The file format is as follows:
1672 * One line per object, capital letters are used to indicate objects
1673 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1674 * other objects are indented by a single space and follow the glock to
1675 * which they are related. Fields are indicated by lower case letters
1676 * followed by a colon and the field value, except for strings which are in
1677 * [] so that its possible to see if they are composed of spaces for
1678 * example. The field's are n = number (id of the object), f = flags,
1679 * t = type, s = state, r = refcount, e = error, p = pid.
1683 void gfs2_dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1685 const struct gfs2_glock_operations *glops = gl->gl_ops;
1686 unsigned long long dtime;
1687 const struct gfs2_holder *gh;
1688 char gflags_buf[32];
1690 dtime = jiffies - gl->gl_demote_time;
1691 dtime *= 1000000/HZ; /* demote time in uSec */
1692 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1694 gfs2_print_dbg(seq, "G: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d v:%d r:%d m:%ld\n",
1695 state2str(gl->gl_state),
1696 gl->gl_name.ln_type,
1697 (unsigned long long)gl->gl_name.ln_number,
1698 gflags2str(gflags_buf, gl),
1699 state2str(gl->gl_target),
1700 state2str(gl->gl_demote_state), dtime,
1701 atomic_read(&gl->gl_ail_count),
1702 atomic_read(&gl->gl_revokes),
1703 (int)gl->gl_lockref.count, gl->gl_hold_time);
1705 list_for_each_entry(gh, &gl->gl_holders, gh_list)
1706 dump_holder(seq, gh);
1708 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1709 glops->go_dump(seq, gl);
1712 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
1714 struct gfs2_glock *gl = iter_ptr;
1716 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
1717 gl->gl_name.ln_type,
1718 (unsigned long long)gl->gl_name.ln_number,
1719 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
1720 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
1721 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
1722 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
1723 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
1724 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
1725 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
1726 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
1730 static const char *gfs2_gltype[] = {
1744 static const char *gfs2_stype[] = {
1745 [GFS2_LKS_SRTT] = "srtt",
1746 [GFS2_LKS_SRTTVAR] = "srttvar",
1747 [GFS2_LKS_SRTTB] = "srttb",
1748 [GFS2_LKS_SRTTVARB] = "srttvarb",
1749 [GFS2_LKS_SIRT] = "sirt",
1750 [GFS2_LKS_SIRTVAR] = "sirtvar",
1751 [GFS2_LKS_DCOUNT] = "dlm",
1752 [GFS2_LKS_QCOUNT] = "queue",
1755 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1757 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
1759 struct gfs2_sbd *sdp = seq->private;
1760 loff_t pos = *(loff_t *)iter_ptr;
1761 unsigned index = pos >> 3;
1762 unsigned subindex = pos & 0x07;
1765 if (index == 0 && subindex != 0)
1768 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
1769 (index == 0) ? "cpu": gfs2_stype[subindex]);
1771 for_each_possible_cpu(i) {
1772 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
1775 seq_printf(seq, " %15u", i);
1777 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
1778 lkstats[index - 1].stats[subindex]);
1780 seq_putc(seq, '\n');
1784 int __init gfs2_glock_init(void)
1788 ret = rhashtable_init(&gl_hash_table, &ht_parms);
1792 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1793 WQ_HIGHPRI | WQ_FREEZABLE, 0);
1794 if (!glock_workqueue) {
1795 rhashtable_destroy(&gl_hash_table);
1798 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1799 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1801 if (!gfs2_delete_workqueue) {
1802 destroy_workqueue(glock_workqueue);
1803 rhashtable_destroy(&gl_hash_table);
1807 ret = register_shrinker(&glock_shrinker);
1809 destroy_workqueue(gfs2_delete_workqueue);
1810 destroy_workqueue(glock_workqueue);
1811 rhashtable_destroy(&gl_hash_table);
1818 void gfs2_glock_exit(void)
1820 unregister_shrinker(&glock_shrinker);
1821 rhashtable_destroy(&gl_hash_table);
1822 destroy_workqueue(glock_workqueue);
1823 destroy_workqueue(gfs2_delete_workqueue);
1826 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1828 while ((gi->gl = rhashtable_walk_next(&gi->hti))) {
1829 if (IS_ERR(gi->gl)) {
1830 if (PTR_ERR(gi->gl) == -EAGAIN)
1835 /* Skip entries for other sb and dead entries */
1836 if (gi->sdp == gi->gl->gl_name.ln_sbd &&
1837 !__lockref_is_dead(&gi->gl->gl_lockref))
1842 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1844 struct gfs2_glock_iter *gi = seq->private;
1847 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
1848 if (rhashtable_walk_start(&gi->hti) != 0)
1852 gfs2_glock_iter_next(gi);
1853 } while (gi->gl && n--);
1855 gi->last_pos = *pos;
1860 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1863 struct gfs2_glock_iter *gi = seq->private;
1866 gi->last_pos = *pos;
1867 gfs2_glock_iter_next(gi);
1872 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1874 struct gfs2_glock_iter *gi = seq->private;
1877 rhashtable_walk_stop(&gi->hti);
1878 rhashtable_walk_exit(&gi->hti);
1881 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1883 dump_glock(seq, iter_ptr);
1887 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
1890 if (*pos >= GFS2_NR_SBSTATS)
1895 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
1899 if (*pos >= GFS2_NR_SBSTATS)
1904 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
1909 static const struct seq_operations gfs2_glock_seq_ops = {
1910 .start = gfs2_glock_seq_start,
1911 .next = gfs2_glock_seq_next,
1912 .stop = gfs2_glock_seq_stop,
1913 .show = gfs2_glock_seq_show,
1916 static const struct seq_operations gfs2_glstats_seq_ops = {
1917 .start = gfs2_glock_seq_start,
1918 .next = gfs2_glock_seq_next,
1919 .stop = gfs2_glock_seq_stop,
1920 .show = gfs2_glstats_seq_show,
1923 static const struct seq_operations gfs2_sbstats_seq_ops = {
1924 .start = gfs2_sbstats_seq_start,
1925 .next = gfs2_sbstats_seq_next,
1926 .stop = gfs2_sbstats_seq_stop,
1927 .show = gfs2_sbstats_seq_show,
1930 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
1932 static int gfs2_glocks_open(struct inode *inode, struct file *file)
1934 int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1935 sizeof(struct gfs2_glock_iter));
1937 struct seq_file *seq = file->private_data;
1938 struct gfs2_glock_iter *gi = seq->private;
1940 gi->sdp = inode->i_private;
1941 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
1943 seq->size = GFS2_SEQ_GOODSIZE;
1949 static int gfs2_glocks_release(struct inode *inode, struct file *file)
1951 struct seq_file *seq = file->private_data;
1952 struct gfs2_glock_iter *gi = seq->private;
1955 return seq_release_private(inode, file);
1958 static int gfs2_glstats_open(struct inode *inode, struct file *file)
1960 int ret = seq_open_private(file, &gfs2_glstats_seq_ops,
1961 sizeof(struct gfs2_glock_iter));
1963 struct seq_file *seq = file->private_data;
1964 struct gfs2_glock_iter *gi = seq->private;
1965 gi->sdp = inode->i_private;
1966 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
1968 seq->size = GFS2_SEQ_GOODSIZE;
1974 static int gfs2_sbstats_open(struct inode *inode, struct file *file)
1976 int ret = seq_open(file, &gfs2_sbstats_seq_ops);
1978 struct seq_file *seq = file->private_data;
1979 seq->private = inode->i_private; /* sdp */
1984 static const struct file_operations gfs2_glocks_fops = {
1985 .owner = THIS_MODULE,
1986 .open = gfs2_glocks_open,
1988 .llseek = seq_lseek,
1989 .release = gfs2_glocks_release,
1992 static const struct file_operations gfs2_glstats_fops = {
1993 .owner = THIS_MODULE,
1994 .open = gfs2_glstats_open,
1996 .llseek = seq_lseek,
1997 .release = gfs2_glocks_release,
2000 static const struct file_operations gfs2_sbstats_fops = {
2001 .owner = THIS_MODULE,
2002 .open = gfs2_sbstats_open,
2004 .llseek = seq_lseek,
2005 .release = seq_release,
2008 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2010 struct dentry *dent;
2012 dent = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2013 if (IS_ERR_OR_NULL(dent))
2015 sdp->debugfs_dir = dent;
2017 dent = debugfs_create_file("glocks",
2019 sdp->debugfs_dir, sdp,
2021 if (IS_ERR_OR_NULL(dent))
2023 sdp->debugfs_dentry_glocks = dent;
2025 dent = debugfs_create_file("glstats",
2027 sdp->debugfs_dir, sdp,
2028 &gfs2_glstats_fops);
2029 if (IS_ERR_OR_NULL(dent))
2031 sdp->debugfs_dentry_glstats = dent;
2033 dent = debugfs_create_file("sbstats",
2035 sdp->debugfs_dir, sdp,
2036 &gfs2_sbstats_fops);
2037 if (IS_ERR_OR_NULL(dent))
2039 sdp->debugfs_dentry_sbstats = dent;
2043 gfs2_delete_debugfs_file(sdp);
2044 return dent ? PTR_ERR(dent) : -ENOMEM;
2047 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2049 if (sdp->debugfs_dir) {
2050 if (sdp->debugfs_dentry_glocks) {
2051 debugfs_remove(sdp->debugfs_dentry_glocks);
2052 sdp->debugfs_dentry_glocks = NULL;
2054 if (sdp->debugfs_dentry_glstats) {
2055 debugfs_remove(sdp->debugfs_dentry_glstats);
2056 sdp->debugfs_dentry_glstats = NULL;
2058 if (sdp->debugfs_dentry_sbstats) {
2059 debugfs_remove(sdp->debugfs_dentry_sbstats);
2060 sdp->debugfs_dentry_sbstats = NULL;
2062 debugfs_remove(sdp->debugfs_dir);
2063 sdp->debugfs_dir = NULL;
2067 int gfs2_register_debugfs(void)
2069 gfs2_root = debugfs_create_dir("gfs2", NULL);
2070 if (IS_ERR(gfs2_root))
2071 return PTR_ERR(gfs2_root);
2072 return gfs2_root ? 0 : -ENOMEM;
2075 void gfs2_unregister_debugfs(void)
2077 debugfs_remove(gfs2_root);