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>
49 #define CREATE_TRACE_POINTS
50 #include "trace_gfs2.h"
52 struct gfs2_glock_iter {
53 int hash; /* hash bucket index */
54 unsigned nhash; /* Index within current bucket */
55 struct gfs2_sbd *sdp; /* incore superblock */
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 (1 << GFS2_GL_HASH_SHIFT)
73 #define GFS2_GL_HASH_MASK (GFS2_GL_HASH_SIZE - 1)
75 static struct hlist_bl_head gl_hash_table[GFS2_GL_HASH_SIZE];
76 static struct dentry *gfs2_root;
79 * gl_hash() - Turn glock number into hash bucket number
80 * @lock: The glock number
82 * Returns: The number of the corresponding hash bucket
85 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
86 const struct lm_lockname *name)
90 h = jhash(&name->ln_number, sizeof(u64), 0);
91 h = jhash(&name->ln_type, sizeof(unsigned int), h);
92 h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
93 h &= GFS2_GL_HASH_MASK;
98 static inline void spin_lock_bucket(unsigned int hash)
100 hlist_bl_lock(&gl_hash_table[hash]);
103 static inline void spin_unlock_bucket(unsigned int hash)
105 hlist_bl_unlock(&gl_hash_table[hash]);
108 static void gfs2_glock_dealloc(struct rcu_head *rcu)
110 struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
112 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
113 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
115 kfree(gl->gl_lksb.sb_lvbptr);
116 kmem_cache_free(gfs2_glock_cachep, gl);
120 void gfs2_glock_free(struct gfs2_glock *gl)
122 struct gfs2_sbd *sdp = gl->gl_sbd;
124 call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
125 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
126 wake_up(&sdp->sd_glock_wait);
130 * gfs2_glock_hold() - increment reference count on glock
131 * @gl: The glock to hold
135 static void gfs2_glock_hold(struct gfs2_glock *gl)
137 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
138 lockref_get(&gl->gl_lockref);
142 * demote_ok - Check to see if it's ok to unlock a glock
145 * Returns: 1 if it's ok
148 static int demote_ok(const struct gfs2_glock *gl)
150 const struct gfs2_glock_operations *glops = gl->gl_ops;
152 if (gl->gl_state == LM_ST_UNLOCKED)
154 if (!list_empty(&gl->gl_holders))
156 if (glops->go_demote_ok)
157 return glops->go_demote_ok(gl);
162 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
164 spin_lock(&lru_lock);
166 if (!list_empty(&gl->gl_lru))
167 list_del_init(&gl->gl_lru);
169 atomic_inc(&lru_count);
171 list_add_tail(&gl->gl_lru, &lru_list);
172 set_bit(GLF_LRU, &gl->gl_flags);
173 spin_unlock(&lru_lock);
176 static void __gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
178 if (!list_empty(&gl->gl_lru)) {
179 list_del_init(&gl->gl_lru);
180 atomic_dec(&lru_count);
181 clear_bit(GLF_LRU, &gl->gl_flags);
185 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
187 spin_lock(&lru_lock);
188 __gfs2_glock_remove_from_lru(gl);
189 spin_unlock(&lru_lock);
193 * gfs2_glock_put() - Decrement reference count on glock
194 * @gl: The glock to put
198 void gfs2_glock_put(struct gfs2_glock *gl)
200 struct gfs2_sbd *sdp = gl->gl_sbd;
201 struct address_space *mapping = gfs2_glock2aspace(gl);
203 if (lockref_put_or_lock(&gl->gl_lockref))
206 lockref_mark_dead(&gl->gl_lockref);
208 spin_lock(&lru_lock);
209 __gfs2_glock_remove_from_lru(gl);
210 spin_unlock(&lru_lock);
211 spin_unlock(&gl->gl_lockref.lock);
212 spin_lock_bucket(gl->gl_hash);
213 hlist_bl_del_rcu(&gl->gl_list);
214 spin_unlock_bucket(gl->gl_hash);
215 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
216 GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
217 trace_gfs2_glock_put(gl);
218 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
222 * search_bucket() - Find struct gfs2_glock by lock number
223 * @bucket: the bucket to search
224 * @name: The lock name
226 * Returns: NULL, or the struct gfs2_glock with the requested number
229 static struct gfs2_glock *search_bucket(unsigned int hash,
230 const struct gfs2_sbd *sdp,
231 const struct lm_lockname *name)
233 struct gfs2_glock *gl;
234 struct hlist_bl_node *h;
236 hlist_bl_for_each_entry_rcu(gl, h, &gl_hash_table[hash], gl_list) {
237 if (!lm_name_equal(&gl->gl_name, name))
239 if (gl->gl_sbd != sdp)
241 if (lockref_get_not_dead(&gl->gl_lockref))
249 * may_grant - check if its ok to grant a new lock
251 * @gh: The lock request which we wish to grant
253 * Returns: true if its ok to grant the lock
256 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
258 const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
259 if ((gh->gh_state == LM_ST_EXCLUSIVE ||
260 gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
262 if (gl->gl_state == gh->gh_state)
264 if (gh->gh_flags & GL_EXACT)
266 if (gl->gl_state == LM_ST_EXCLUSIVE) {
267 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
269 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
272 if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
277 static void gfs2_holder_wake(struct gfs2_holder *gh)
279 clear_bit(HIF_WAIT, &gh->gh_iflags);
280 smp_mb__after_atomic();
281 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
285 * do_error - Something unexpected has happened during a lock request
289 static inline void do_error(struct gfs2_glock *gl, const int ret)
291 struct gfs2_holder *gh, *tmp;
293 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
294 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
296 if (ret & LM_OUT_ERROR)
298 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
299 gh->gh_error = GLR_TRYFAILED;
302 list_del_init(&gh->gh_list);
303 trace_gfs2_glock_queue(gh, 0);
304 gfs2_holder_wake(gh);
309 * do_promote - promote as many requests as possible on the current queue
312 * Returns: 1 if there is a blocked holder at the head of the list, or 2
313 * if a type specific operation is underway.
316 static int do_promote(struct gfs2_glock *gl)
317 __releases(&gl->gl_spin)
318 __acquires(&gl->gl_spin)
320 const struct gfs2_glock_operations *glops = gl->gl_ops;
321 struct gfs2_holder *gh, *tmp;
325 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
326 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
328 if (may_grant(gl, gh)) {
329 if (gh->gh_list.prev == &gl->gl_holders &&
331 spin_unlock(&gl->gl_spin);
332 /* FIXME: eliminate this eventually */
333 ret = glops->go_lock(gh);
334 spin_lock(&gl->gl_spin);
339 list_del_init(&gh->gh_list);
340 trace_gfs2_glock_queue(gh, 0);
341 gfs2_holder_wake(gh);
344 set_bit(HIF_HOLDER, &gh->gh_iflags);
345 trace_gfs2_promote(gh, 1);
346 gfs2_holder_wake(gh);
349 set_bit(HIF_HOLDER, &gh->gh_iflags);
350 trace_gfs2_promote(gh, 0);
351 gfs2_holder_wake(gh);
354 if (gh->gh_list.prev == &gl->gl_holders)
363 * find_first_waiter - find the first gh that's waiting for the glock
367 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
369 struct gfs2_holder *gh;
371 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
372 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
379 * state_change - record that the glock is now in a different state
381 * @new_state the new state
385 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
389 held1 = (gl->gl_state != LM_ST_UNLOCKED);
390 held2 = (new_state != LM_ST_UNLOCKED);
392 if (held1 != held2) {
393 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
395 gl->gl_lockref.count++;
397 gl->gl_lockref.count--;
399 if (held1 && held2 && list_empty(&gl->gl_holders))
400 clear_bit(GLF_QUEUED, &gl->gl_flags);
402 if (new_state != gl->gl_target)
403 /* shorten our minimum hold time */
404 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
406 gl->gl_state = new_state;
407 gl->gl_tchange = jiffies;
410 static void gfs2_demote_wake(struct gfs2_glock *gl)
412 gl->gl_demote_state = LM_ST_EXCLUSIVE;
413 clear_bit(GLF_DEMOTE, &gl->gl_flags);
414 smp_mb__after_atomic();
415 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
419 * finish_xmote - The DLM has replied to one of our lock requests
421 * @ret: The status from the DLM
425 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
427 const struct gfs2_glock_operations *glops = gl->gl_ops;
428 struct gfs2_holder *gh;
429 unsigned state = ret & LM_OUT_ST_MASK;
432 spin_lock(&gl->gl_spin);
433 trace_gfs2_glock_state_change(gl, state);
434 state_change(gl, state);
435 gh = find_first_waiter(gl);
437 /* Demote to UN request arrived during demote to SH or DF */
438 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
439 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
440 gl->gl_target = LM_ST_UNLOCKED;
442 /* Check for state != intended state */
443 if (unlikely(state != gl->gl_target)) {
444 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
445 /* move to back of queue and try next entry */
446 if (ret & LM_OUT_CANCELED) {
447 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
448 list_move_tail(&gh->gh_list, &gl->gl_holders);
449 gh = find_first_waiter(gl);
450 gl->gl_target = gh->gh_state;
453 /* Some error or failed "try lock" - report it */
454 if ((ret & LM_OUT_ERROR) ||
455 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
456 gl->gl_target = gl->gl_state;
462 /* Unlocked due to conversion deadlock, try again */
465 do_xmote(gl, gh, gl->gl_target);
467 /* Conversion fails, unlock and try again */
470 do_xmote(gl, gh, LM_ST_UNLOCKED);
472 default: /* Everything else */
473 pr_err("wanted %u got %u\n", gl->gl_target, state);
476 spin_unlock(&gl->gl_spin);
480 /* Fast path - we got what we asked for */
481 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
482 gfs2_demote_wake(gl);
483 if (state != LM_ST_UNLOCKED) {
484 if (glops->go_xmote_bh) {
485 spin_unlock(&gl->gl_spin);
486 rv = glops->go_xmote_bh(gl, gh);
487 spin_lock(&gl->gl_spin);
498 clear_bit(GLF_LOCK, &gl->gl_flags);
500 spin_unlock(&gl->gl_spin);
504 * do_xmote - Calls the DLM to change the state of a lock
505 * @gl: The lock state
506 * @gh: The holder (only for promotes)
507 * @target: The target lock state
511 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
512 __releases(&gl->gl_spin)
513 __acquires(&gl->gl_spin)
515 const struct gfs2_glock_operations *glops = gl->gl_ops;
516 struct gfs2_sbd *sdp = gl->gl_sbd;
517 unsigned int lck_flags = gh ? gh->gh_flags : 0;
520 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
522 GLOCK_BUG_ON(gl, gl->gl_state == target);
523 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
524 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
526 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
527 do_error(gl, 0); /* Fail queued try locks */
530 set_bit(GLF_BLOCKING, &gl->gl_flags);
531 if ((gl->gl_req == LM_ST_UNLOCKED) ||
532 (gl->gl_state == LM_ST_EXCLUSIVE) ||
533 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
534 clear_bit(GLF_BLOCKING, &gl->gl_flags);
535 spin_unlock(&gl->gl_spin);
538 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
539 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
540 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
543 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
545 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
547 pr_err("lm_lock ret %d\n", ret);
550 } else { /* lock_nolock */
551 finish_xmote(gl, target);
552 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
556 spin_lock(&gl->gl_spin);
560 * find_first_holder - find the first "holder" gh
564 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
566 struct gfs2_holder *gh;
568 if (!list_empty(&gl->gl_holders)) {
569 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
570 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
577 * run_queue - do all outstanding tasks related to a glock
578 * @gl: The glock in question
579 * @nonblock: True if we must not block in run_queue
583 static void run_queue(struct gfs2_glock *gl, const int nonblock)
584 __releases(&gl->gl_spin)
585 __acquires(&gl->gl_spin)
587 struct gfs2_holder *gh = NULL;
590 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
593 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
595 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
596 gl->gl_demote_state != gl->gl_state) {
597 if (find_first_holder(gl))
601 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
602 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
603 gl->gl_target = gl->gl_demote_state;
605 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
606 gfs2_demote_wake(gl);
607 ret = do_promote(gl);
612 gh = find_first_waiter(gl);
613 gl->gl_target = gh->gh_state;
614 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
615 do_error(gl, 0); /* Fail queued try locks */
617 do_xmote(gl, gh, gl->gl_target);
622 clear_bit(GLF_LOCK, &gl->gl_flags);
623 smp_mb__after_atomic();
624 gl->gl_lockref.count++;
625 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
626 gl->gl_lockref.count--;
630 clear_bit(GLF_LOCK, &gl->gl_flags);
631 smp_mb__after_atomic();
635 static void delete_work_func(struct work_struct *work)
637 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
638 struct gfs2_sbd *sdp = gl->gl_sbd;
639 struct gfs2_inode *ip;
641 u64 no_addr = gl->gl_name.ln_number;
644 /* Note: Unsafe to dereference ip as we don't hold right refs/locks */
647 inode = gfs2_ilookup(sdp->sd_vfs, no_addr, 1);
649 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
650 if (inode && !IS_ERR(inode)) {
651 d_prune_aliases(inode);
657 static void glock_work_func(struct work_struct *work)
659 unsigned long delay = 0;
660 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
663 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
664 finish_xmote(gl, gl->gl_reply);
667 spin_lock(&gl->gl_spin);
668 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
669 gl->gl_state != LM_ST_UNLOCKED &&
670 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
671 unsigned long holdtime, now = jiffies;
673 holdtime = gl->gl_tchange + gl->gl_hold_time;
674 if (time_before(now, holdtime))
675 delay = holdtime - now;
678 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
679 set_bit(GLF_DEMOTE, &gl->gl_flags);
683 spin_unlock(&gl->gl_spin);
687 if (gl->gl_name.ln_type != LM_TYPE_INODE)
689 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
697 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
698 * @sdp: The GFS2 superblock
699 * @number: the lock number
700 * @glops: The glock_operations to use
701 * @create: If 0, don't create the glock if it doesn't exist
702 * @glp: the glock is returned here
704 * This does not lock a glock, just finds/creates structures for one.
709 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
710 const struct gfs2_glock_operations *glops, int create,
711 struct gfs2_glock **glp)
713 struct super_block *s = sdp->sd_vfs;
714 struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
715 struct gfs2_glock *gl, *tmp;
716 unsigned int hash = gl_hash(sdp, &name);
717 struct address_space *mapping;
718 struct kmem_cache *cachep;
721 gl = search_bucket(hash, sdp, &name);
730 if (glops->go_flags & GLOF_ASPACE)
731 cachep = gfs2_glock_aspace_cachep;
733 cachep = gfs2_glock_cachep;
734 gl = kmem_cache_alloc(cachep, GFP_NOFS);
738 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
740 if (glops->go_flags & GLOF_LVB) {
741 gl->gl_lksb.sb_lvbptr = kzalloc(GFS2_MIN_LVB_SIZE, GFP_NOFS);
742 if (!gl->gl_lksb.sb_lvbptr) {
743 kmem_cache_free(cachep, gl);
748 atomic_inc(&sdp->sd_glock_disposal);
752 gl->gl_lockref.count = 1;
753 gl->gl_state = LM_ST_UNLOCKED;
754 gl->gl_target = LM_ST_UNLOCKED;
755 gl->gl_demote_state = LM_ST_EXCLUSIVE;
758 gl->gl_dstamp = ktime_set(0, 0);
760 /* We use the global stats to estimate the initial per-glock stats */
761 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
763 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
764 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
765 gl->gl_tchange = jiffies;
766 gl->gl_object = NULL;
767 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
768 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
769 INIT_WORK(&gl->gl_delete, delete_work_func);
771 mapping = gfs2_glock2aspace(gl);
773 mapping->a_ops = &gfs2_meta_aops;
774 mapping->host = s->s_bdev->bd_inode;
776 mapping_set_gfp_mask(mapping, GFP_NOFS);
777 mapping->private_data = NULL;
778 mapping->backing_dev_info = s->s_bdi;
779 mapping->writeback_index = 0;
782 spin_lock_bucket(hash);
783 tmp = search_bucket(hash, sdp, &name);
785 spin_unlock_bucket(hash);
786 kfree(gl->gl_lksb.sb_lvbptr);
787 kmem_cache_free(cachep, gl);
788 atomic_dec(&sdp->sd_glock_disposal);
791 hlist_bl_add_head_rcu(&gl->gl_list, &gl_hash_table[hash]);
792 spin_unlock_bucket(hash);
801 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
803 * @state: the state we're requesting
804 * @flags: the modifier flags
805 * @gh: the holder structure
809 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
810 struct gfs2_holder *gh)
812 INIT_LIST_HEAD(&gh->gh_list);
814 gh->gh_ip = (unsigned long)__builtin_return_address(0);
815 gh->gh_owner_pid = get_pid(task_pid(current));
816 gh->gh_state = state;
817 gh->gh_flags = flags;
824 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
825 * @state: the state we're requesting
826 * @flags: the modifier flags
827 * @gh: the holder structure
829 * Don't mess with the glock.
833 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
835 gh->gh_state = state;
836 gh->gh_flags = flags;
838 gh->gh_ip = (unsigned long)__builtin_return_address(0);
839 if (gh->gh_owner_pid)
840 put_pid(gh->gh_owner_pid);
841 gh->gh_owner_pid = get_pid(task_pid(current));
845 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
846 * @gh: the holder structure
850 void gfs2_holder_uninit(struct gfs2_holder *gh)
852 put_pid(gh->gh_owner_pid);
853 gfs2_glock_put(gh->gh_gl);
859 * gfs2_glock_holder_wait
862 * This function and gfs2_glock_demote_wait both show up in the WCHAN
863 * field. Thus I've separated these otherwise identical functions in
864 * order to be more informative to the user.
867 static int gfs2_glock_holder_wait(void *word)
873 static int gfs2_glock_demote_wait(void *word)
880 * gfs2_glock_wait - wait on a glock acquisition
881 * @gh: the glock holder
883 * Returns: 0 on success
886 int gfs2_glock_wait(struct gfs2_holder *gh)
888 unsigned long time1 = jiffies;
891 wait_on_bit(&gh->gh_iflags, HIF_WAIT, gfs2_glock_holder_wait, TASK_UNINTERRUPTIBLE);
892 if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
893 /* Lengthen the minimum hold time. */
894 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
901 * handle_callback - process a demote request
903 * @state: the state the caller wants us to change to
905 * There are only two requests that we are going to see in actual
906 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
909 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
910 unsigned long delay, bool remote)
912 int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
914 set_bit(bit, &gl->gl_flags);
915 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
916 gl->gl_demote_state = state;
917 gl->gl_demote_time = jiffies;
918 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
919 gl->gl_demote_state != state) {
920 gl->gl_demote_state = LM_ST_UNLOCKED;
922 if (gl->gl_ops->go_callback)
923 gl->gl_ops->go_callback(gl, remote);
924 trace_gfs2_demote_rq(gl, remote);
927 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
929 struct va_format vaf;
935 seq_vprintf(seq, fmt, args);
947 * add_to_queue - Add a holder to the wait queue (but look for recursion)
948 * @gh: the holder structure to add
950 * Eventually we should move the recursive locking trap to a
951 * debugging option or something like that. This is the fast
952 * path and needs to have the minimum number of distractions.
956 static inline void add_to_queue(struct gfs2_holder *gh)
957 __releases(&gl->gl_spin)
958 __acquires(&gl->gl_spin)
960 struct gfs2_glock *gl = gh->gh_gl;
961 struct gfs2_sbd *sdp = gl->gl_sbd;
962 struct list_head *insert_pt = NULL;
963 struct gfs2_holder *gh2;
966 BUG_ON(gh->gh_owner_pid == NULL);
967 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
970 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
971 if (test_bit(GLF_LOCK, &gl->gl_flags))
972 try_futile = !may_grant(gl, gh);
973 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
977 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
978 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
979 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
982 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
984 gh->gh_error = GLR_TRYFAILED;
985 gfs2_holder_wake(gh);
988 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
990 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
991 insert_pt = &gh2->gh_list;
993 set_bit(GLF_QUEUED, &gl->gl_flags);
994 trace_gfs2_glock_queue(gh, 1);
995 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
996 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
997 if (likely(insert_pt == NULL)) {
998 list_add_tail(&gh->gh_list, &gl->gl_holders);
999 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1003 list_add_tail(&gh->gh_list, insert_pt);
1005 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
1006 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1007 spin_unlock(&gl->gl_spin);
1008 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1009 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1010 spin_lock(&gl->gl_spin);
1015 pr_err("original: %pSR\n", (void *)gh2->gh_ip);
1016 pr_err("pid: %d\n", pid_nr(gh2->gh_owner_pid));
1017 pr_err("lock type: %d req lock state : %d\n",
1018 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1019 pr_err("new: %pSR\n", (void *)gh->gh_ip);
1020 pr_err("pid: %d\n", pid_nr(gh->gh_owner_pid));
1021 pr_err("lock type: %d req lock state : %d\n",
1022 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1023 gfs2_dump_glock(NULL, gl);
1028 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1029 * @gh: the holder structure
1031 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1033 * Returns: 0, GLR_TRYFAILED, or errno on failure
1036 int gfs2_glock_nq(struct gfs2_holder *gh)
1038 struct gfs2_glock *gl = gh->gh_gl;
1039 struct gfs2_sbd *sdp = gl->gl_sbd;
1042 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1045 if (test_bit(GLF_LRU, &gl->gl_flags))
1046 gfs2_glock_remove_from_lru(gl);
1048 spin_lock(&gl->gl_spin);
1050 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1051 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1052 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1053 gl->gl_lockref.count++;
1054 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1055 gl->gl_lockref.count--;
1058 spin_unlock(&gl->gl_spin);
1060 if (!(gh->gh_flags & GL_ASYNC))
1061 error = gfs2_glock_wait(gh);
1067 * gfs2_glock_poll - poll to see if an async request has been completed
1070 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1073 int gfs2_glock_poll(struct gfs2_holder *gh)
1075 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1079 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1080 * @gh: the glock holder
1084 void gfs2_glock_dq(struct gfs2_holder *gh)
1086 struct gfs2_glock *gl = gh->gh_gl;
1087 const struct gfs2_glock_operations *glops = gl->gl_ops;
1091 spin_lock(&gl->gl_spin);
1092 if (gh->gh_flags & GL_NOCACHE)
1093 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1095 list_del_init(&gh->gh_list);
1096 if (find_first_holder(gl) == NULL) {
1097 if (glops->go_unlock) {
1098 GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1099 spin_unlock(&gl->gl_spin);
1100 glops->go_unlock(gh);
1101 spin_lock(&gl->gl_spin);
1102 clear_bit(GLF_LOCK, &gl->gl_flags);
1104 if (list_empty(&gl->gl_holders) &&
1105 !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1106 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1109 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1110 gfs2_glock_add_to_lru(gl);
1112 trace_gfs2_glock_queue(gh, 0);
1113 spin_unlock(&gl->gl_spin);
1114 if (likely(fast_path))
1117 gfs2_glock_hold(gl);
1118 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1119 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1120 gl->gl_name.ln_type == LM_TYPE_INODE)
1121 delay = gl->gl_hold_time;
1122 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1126 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1128 struct gfs2_glock *gl = gh->gh_gl;
1131 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, gfs2_glock_demote_wait, TASK_UNINTERRUPTIBLE);
1135 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1136 * @gh: the holder structure
1140 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1143 gfs2_holder_uninit(gh);
1147 * gfs2_glock_nq_num - acquire a glock based on lock number
1148 * @sdp: the filesystem
1149 * @number: the lock number
1150 * @glops: the glock operations for the type of glock
1151 * @state: the state to acquire the glock in
1152 * @flags: modifier flags for the acquisition
1153 * @gh: the struct gfs2_holder
1158 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1159 const struct gfs2_glock_operations *glops,
1160 unsigned int state, int flags, struct gfs2_holder *gh)
1162 struct gfs2_glock *gl;
1165 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1167 error = gfs2_glock_nq_init(gl, state, flags, gh);
1175 * glock_compare - Compare two struct gfs2_glock structures for sorting
1176 * @arg_a: the first structure
1177 * @arg_b: the second structure
1181 static int glock_compare(const void *arg_a, const void *arg_b)
1183 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1184 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1185 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1186 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1188 if (a->ln_number > b->ln_number)
1190 if (a->ln_number < b->ln_number)
1192 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1197 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1198 * @num_gh: the number of structures
1199 * @ghs: an array of struct gfs2_holder structures
1201 * Returns: 0 on success (all glocks acquired),
1202 * errno on failure (no glocks acquired)
1205 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1206 struct gfs2_holder **p)
1211 for (x = 0; x < num_gh; x++)
1214 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1216 for (x = 0; x < num_gh; x++) {
1217 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1219 error = gfs2_glock_nq(p[x]);
1222 gfs2_glock_dq(p[x]);
1231 * gfs2_glock_nq_m - acquire multiple glocks
1232 * @num_gh: the number of structures
1233 * @ghs: an array of struct gfs2_holder structures
1236 * Returns: 0 on success (all glocks acquired),
1237 * errno on failure (no glocks acquired)
1240 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1242 struct gfs2_holder *tmp[4];
1243 struct gfs2_holder **pph = tmp;
1250 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1251 return gfs2_glock_nq(ghs);
1255 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1260 error = nq_m_sync(num_gh, ghs, pph);
1269 * gfs2_glock_dq_m - release multiple glocks
1270 * @num_gh: the number of structures
1271 * @ghs: an array of struct gfs2_holder structures
1275 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1278 gfs2_glock_dq(&ghs[num_gh]);
1281 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1283 unsigned long delay = 0;
1284 unsigned long holdtime;
1285 unsigned long now = jiffies;
1287 gfs2_glock_hold(gl);
1288 holdtime = gl->gl_tchange + gl->gl_hold_time;
1289 if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1290 gl->gl_name.ln_type == LM_TYPE_INODE) {
1291 if (time_before(now, holdtime))
1292 delay = holdtime - now;
1293 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1294 delay = gl->gl_hold_time;
1297 spin_lock(&gl->gl_spin);
1298 handle_callback(gl, state, delay, true);
1299 spin_unlock(&gl->gl_spin);
1300 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1305 * gfs2_should_freeze - Figure out if glock should be frozen
1306 * @gl: The glock in question
1308 * Glocks are not frozen if (a) the result of the dlm operation is
1309 * an error, (b) the locking operation was an unlock operation or
1310 * (c) if there is a "noexp" flagged request anywhere in the queue
1312 * Returns: 1 if freezing should occur, 0 otherwise
1315 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1317 const struct gfs2_holder *gh;
1319 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1321 if (gl->gl_target == LM_ST_UNLOCKED)
1324 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1325 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1327 if (LM_FLAG_NOEXP & gh->gh_flags)
1335 * gfs2_glock_complete - Callback used by locking
1336 * @gl: Pointer to the glock
1337 * @ret: The return value from the dlm
1339 * The gl_reply field is under the gl_spin lock so that it is ok
1340 * to use a bitfield shared with other glock state fields.
1343 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1345 struct lm_lockstruct *ls = &gl->gl_sbd->sd_lockstruct;
1347 spin_lock(&gl->gl_spin);
1350 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1351 if (gfs2_should_freeze(gl)) {
1352 set_bit(GLF_FROZEN, &gl->gl_flags);
1353 spin_unlock(&gl->gl_spin);
1358 gl->gl_lockref.count++;
1359 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1360 spin_unlock(&gl->gl_spin);
1362 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1366 static int glock_cmp(void *priv, struct list_head *a, struct list_head *b)
1368 struct gfs2_glock *gla, *glb;
1370 gla = list_entry(a, struct gfs2_glock, gl_lru);
1371 glb = list_entry(b, struct gfs2_glock, gl_lru);
1373 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1375 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1382 * gfs2_dispose_glock_lru - Demote a list of glocks
1383 * @list: The list to dispose of
1385 * Disposing of glocks may involve disk accesses, so that here we sort
1386 * the glocks by number (i.e. disk location of the inodes) so that if
1387 * there are any such accesses, they'll be sent in order (mostly).
1389 * Must be called under the lru_lock, but may drop and retake this
1390 * lock. While the lru_lock is dropped, entries may vanish from the
1391 * list, but no new entries will appear on the list (since it is
1395 static void gfs2_dispose_glock_lru(struct list_head *list)
1396 __releases(&lru_lock)
1397 __acquires(&lru_lock)
1399 struct gfs2_glock *gl;
1401 list_sort(NULL, list, glock_cmp);
1403 while(!list_empty(list)) {
1404 gl = list_entry(list->next, struct gfs2_glock, gl_lru);
1405 list_del_init(&gl->gl_lru);
1406 if (!spin_trylock(&gl->gl_spin)) {
1408 list_add(&gl->gl_lru, &lru_list);
1409 atomic_inc(&lru_count);
1412 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1413 spin_unlock(&gl->gl_spin);
1414 goto add_back_to_lru;
1416 clear_bit(GLF_LRU, &gl->gl_flags);
1417 gl->gl_lockref.count++;
1419 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1420 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1421 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1422 gl->gl_lockref.count--;
1423 spin_unlock(&gl->gl_spin);
1424 cond_resched_lock(&lru_lock);
1429 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1430 * @nr: The number of entries to scan
1432 * This function selects the entries on the LRU which are able to
1433 * be demoted, and then kicks off the process by calling
1434 * gfs2_dispose_glock_lru() above.
1437 static long gfs2_scan_glock_lru(int nr)
1439 struct gfs2_glock *gl;
1444 spin_lock(&lru_lock);
1445 while ((nr-- >= 0) && !list_empty(&lru_list)) {
1446 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1448 /* Test for being demotable */
1449 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1450 list_move(&gl->gl_lru, &dispose);
1451 atomic_dec(&lru_count);
1456 list_move(&gl->gl_lru, &skipped);
1458 list_splice(&skipped, &lru_list);
1459 if (!list_empty(&dispose))
1460 gfs2_dispose_glock_lru(&dispose);
1461 spin_unlock(&lru_lock);
1466 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1467 struct shrink_control *sc)
1469 if (!(sc->gfp_mask & __GFP_FS))
1471 return gfs2_scan_glock_lru(sc->nr_to_scan);
1474 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1475 struct shrink_control *sc)
1477 return vfs_pressure_ratio(atomic_read(&lru_count));
1480 static struct shrinker glock_shrinker = {
1481 .seeks = DEFAULT_SEEKS,
1482 .count_objects = gfs2_glock_shrink_count,
1483 .scan_objects = gfs2_glock_shrink_scan,
1487 * examine_bucket - Call a function for glock in a hash bucket
1488 * @examiner: the function
1489 * @sdp: the filesystem
1490 * @bucket: the bucket
1494 static void examine_bucket(glock_examiner examiner, const struct gfs2_sbd *sdp,
1497 struct gfs2_glock *gl;
1498 struct hlist_bl_head *head = &gl_hash_table[hash];
1499 struct hlist_bl_node *pos;
1502 hlist_bl_for_each_entry_rcu(gl, pos, head, gl_list) {
1503 if ((gl->gl_sbd == sdp) && lockref_get_not_dead(&gl->gl_lockref))
1510 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1514 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1515 examine_bucket(examiner, sdp, x);
1520 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1521 * @gl: The glock to thaw
1525 static void thaw_glock(struct gfs2_glock *gl)
1527 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1529 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1530 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0) {
1537 * clear_glock - look at a glock and see if we can free it from glock cache
1538 * @gl: the glock to look at
1542 static void clear_glock(struct gfs2_glock *gl)
1544 gfs2_glock_remove_from_lru(gl);
1546 spin_lock(&gl->gl_spin);
1547 if (gl->gl_state != LM_ST_UNLOCKED)
1548 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1549 spin_unlock(&gl->gl_spin);
1550 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1555 * gfs2_glock_thaw - Thaw any frozen glocks
1556 * @sdp: The super block
1560 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1562 glock_hash_walk(thaw_glock, sdp);
1565 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1567 spin_lock(&gl->gl_spin);
1568 gfs2_dump_glock(seq, gl);
1569 spin_unlock(&gl->gl_spin);
1572 static void dump_glock_func(struct gfs2_glock *gl)
1574 dump_glock(NULL, gl);
1578 * gfs2_gl_hash_clear - Empty out the glock hash table
1579 * @sdp: the filesystem
1580 * @wait: wait until it's all gone
1582 * Called when unmounting the filesystem.
1585 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1587 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1588 flush_workqueue(glock_workqueue);
1589 glock_hash_walk(clear_glock, sdp);
1590 flush_workqueue(glock_workqueue);
1591 wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
1592 glock_hash_walk(dump_glock_func, sdp);
1595 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1597 struct gfs2_glock *gl = ip->i_gl;
1600 ret = gfs2_truncatei_resume(ip);
1601 gfs2_assert_withdraw(gl->gl_sbd, ret == 0);
1603 spin_lock(&gl->gl_spin);
1604 clear_bit(GLF_LOCK, &gl->gl_flags);
1606 spin_unlock(&gl->gl_spin);
1609 static const char *state2str(unsigned state)
1612 case LM_ST_UNLOCKED:
1616 case LM_ST_DEFERRED:
1618 case LM_ST_EXCLUSIVE:
1624 static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1627 if (flags & LM_FLAG_TRY)
1629 if (flags & LM_FLAG_TRY_1CB)
1631 if (flags & LM_FLAG_NOEXP)
1633 if (flags & LM_FLAG_ANY)
1635 if (flags & LM_FLAG_PRIORITY)
1637 if (flags & GL_ASYNC)
1639 if (flags & GL_EXACT)
1641 if (flags & GL_NOCACHE)
1643 if (test_bit(HIF_HOLDER, &iflags))
1645 if (test_bit(HIF_WAIT, &iflags))
1647 if (test_bit(HIF_FIRST, &iflags))
1654 * dump_holder - print information about a glock holder
1655 * @seq: the seq_file struct
1656 * @gh: the glock holder
1660 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1662 struct task_struct *gh_owner = NULL;
1666 if (gh->gh_owner_pid)
1667 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1668 gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1669 state2str(gh->gh_state),
1670 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1672 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1673 gh_owner ? gh_owner->comm : "(ended)",
1678 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1680 const unsigned long *gflags = &gl->gl_flags;
1683 if (test_bit(GLF_LOCK, gflags))
1685 if (test_bit(GLF_DEMOTE, gflags))
1687 if (test_bit(GLF_PENDING_DEMOTE, gflags))
1689 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1691 if (test_bit(GLF_DIRTY, gflags))
1693 if (test_bit(GLF_LFLUSH, gflags))
1695 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1697 if (test_bit(GLF_REPLY_PENDING, gflags))
1699 if (test_bit(GLF_INITIAL, gflags))
1701 if (test_bit(GLF_FROZEN, gflags))
1703 if (test_bit(GLF_QUEUED, gflags))
1705 if (test_bit(GLF_LRU, gflags))
1709 if (test_bit(GLF_BLOCKING, gflags))
1716 * gfs2_dump_glock - print information about a glock
1717 * @seq: The seq_file struct
1720 * The file format is as follows:
1721 * One line per object, capital letters are used to indicate objects
1722 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1723 * other objects are indented by a single space and follow the glock to
1724 * which they are related. Fields are indicated by lower case letters
1725 * followed by a colon and the field value, except for strings which are in
1726 * [] so that its possible to see if they are composed of spaces for
1727 * example. The field's are n = number (id of the object), f = flags,
1728 * t = type, s = state, r = refcount, e = error, p = pid.
1732 void gfs2_dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1734 const struct gfs2_glock_operations *glops = gl->gl_ops;
1735 unsigned long long dtime;
1736 const struct gfs2_holder *gh;
1737 char gflags_buf[32];
1739 dtime = jiffies - gl->gl_demote_time;
1740 dtime *= 1000000/HZ; /* demote time in uSec */
1741 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1743 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",
1744 state2str(gl->gl_state),
1745 gl->gl_name.ln_type,
1746 (unsigned long long)gl->gl_name.ln_number,
1747 gflags2str(gflags_buf, gl),
1748 state2str(gl->gl_target),
1749 state2str(gl->gl_demote_state), dtime,
1750 atomic_read(&gl->gl_ail_count),
1751 atomic_read(&gl->gl_revokes),
1752 (int)gl->gl_lockref.count, gl->gl_hold_time);
1754 list_for_each_entry(gh, &gl->gl_holders, gh_list)
1755 dump_holder(seq, gh);
1757 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1758 glops->go_dump(seq, gl);
1761 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
1763 struct gfs2_glock *gl = iter_ptr;
1765 seq_printf(seq, "G: n:%u/%llx rtt:%lld/%lld rttb:%lld/%lld irt:%lld/%lld dcnt: %lld qcnt: %lld\n",
1766 gl->gl_name.ln_type,
1767 (unsigned long long)gl->gl_name.ln_number,
1768 (long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
1769 (long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
1770 (long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
1771 (long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
1772 (long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
1773 (long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
1774 (long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
1775 (long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
1779 static const char *gfs2_gltype[] = {
1793 static const char *gfs2_stype[] = {
1794 [GFS2_LKS_SRTT] = "srtt",
1795 [GFS2_LKS_SRTTVAR] = "srttvar",
1796 [GFS2_LKS_SRTTB] = "srttb",
1797 [GFS2_LKS_SRTTVARB] = "srttvarb",
1798 [GFS2_LKS_SIRT] = "sirt",
1799 [GFS2_LKS_SIRTVAR] = "sirtvar",
1800 [GFS2_LKS_DCOUNT] = "dlm",
1801 [GFS2_LKS_QCOUNT] = "queue",
1804 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1806 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
1808 struct gfs2_glock_iter *gi = seq->private;
1809 struct gfs2_sbd *sdp = gi->sdp;
1810 unsigned index = gi->hash >> 3;
1811 unsigned subindex = gi->hash & 0x07;
1815 if (index == 0 && subindex != 0)
1818 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
1819 (index == 0) ? "cpu": gfs2_stype[subindex]);
1821 for_each_possible_cpu(i) {
1822 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
1826 value = lkstats->lkstats[index - 1].stats[subindex];
1828 seq_printf(seq, " %15lld", (long long)value);
1830 seq_putc(seq, '\n');
1834 int __init gfs2_glock_init(void)
1837 for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
1838 INIT_HLIST_BL_HEAD(&gl_hash_table[i]);
1841 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1842 WQ_HIGHPRI | WQ_FREEZABLE, 0);
1843 if (!glock_workqueue)
1845 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1846 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1848 if (!gfs2_delete_workqueue) {
1849 destroy_workqueue(glock_workqueue);
1853 register_shrinker(&glock_shrinker);
1858 void gfs2_glock_exit(void)
1860 unregister_shrinker(&glock_shrinker);
1861 destroy_workqueue(glock_workqueue);
1862 destroy_workqueue(gfs2_delete_workqueue);
1865 static inline struct gfs2_glock *glock_hash_chain(unsigned hash)
1867 return hlist_bl_entry(hlist_bl_first_rcu(&gl_hash_table[hash]),
1868 struct gfs2_glock, gl_list);
1871 static inline struct gfs2_glock *glock_hash_next(struct gfs2_glock *gl)
1873 return hlist_bl_entry(rcu_dereference(gl->gl_list.next),
1874 struct gfs2_glock, gl_list);
1877 static int gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1879 struct gfs2_glock *gl;
1884 gi->gl = glock_hash_next(gl);
1887 if (gi->hash >= GFS2_GL_HASH_SIZE) {
1891 gi->gl = glock_hash_chain(gi->hash);
1894 while (gi->gl == NULL) {
1896 if (gi->hash >= GFS2_GL_HASH_SIZE) {
1900 gi->gl = glock_hash_chain(gi->hash);
1903 /* Skip entries for other sb and dead entries */
1904 } while (gi->sdp != gi->gl->gl_sbd ||
1905 __lockref_is_dead(&gi->gl->gl_lockref));
1910 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1912 struct gfs2_glock_iter *gi = seq->private;
1915 if (gi->last_pos <= *pos)
1916 n = gi->nhash + (*pos - gi->last_pos);
1924 if (gfs2_glock_iter_next(gi))
1928 gi->last_pos = *pos;
1932 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1935 struct gfs2_glock_iter *gi = seq->private;
1938 gi->last_pos = *pos;
1939 if (gfs2_glock_iter_next(gi))
1945 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1947 struct gfs2_glock_iter *gi = seq->private;
1954 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1956 dump_glock(seq, iter_ptr);
1960 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
1962 struct gfs2_glock_iter *gi = seq->private;
1965 if (*pos >= GFS2_NR_SBSTATS)
1968 return SEQ_START_TOKEN;
1971 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
1974 struct gfs2_glock_iter *gi = seq->private;
1977 if (gi->hash >= GFS2_NR_SBSTATS) {
1981 return SEQ_START_TOKEN;
1984 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
1989 static const struct seq_operations gfs2_glock_seq_ops = {
1990 .start = gfs2_glock_seq_start,
1991 .next = gfs2_glock_seq_next,
1992 .stop = gfs2_glock_seq_stop,
1993 .show = gfs2_glock_seq_show,
1996 static const struct seq_operations gfs2_glstats_seq_ops = {
1997 .start = gfs2_glock_seq_start,
1998 .next = gfs2_glock_seq_next,
1999 .stop = gfs2_glock_seq_stop,
2000 .show = gfs2_glstats_seq_show,
2003 static const struct seq_operations gfs2_sbstats_seq_ops = {
2004 .start = gfs2_sbstats_seq_start,
2005 .next = gfs2_sbstats_seq_next,
2006 .stop = gfs2_sbstats_seq_stop,
2007 .show = gfs2_sbstats_seq_show,
2010 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2012 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2014 int ret = seq_open_private(file, &gfs2_glock_seq_ops,
2015 sizeof(struct gfs2_glock_iter));
2017 struct seq_file *seq = file->private_data;
2018 struct gfs2_glock_iter *gi = seq->private;
2019 gi->sdp = inode->i_private;
2020 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2022 seq->size = GFS2_SEQ_GOODSIZE;
2027 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2029 int ret = seq_open_private(file, &gfs2_glstats_seq_ops,
2030 sizeof(struct gfs2_glock_iter));
2032 struct seq_file *seq = file->private_data;
2033 struct gfs2_glock_iter *gi = seq->private;
2034 gi->sdp = inode->i_private;
2035 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2037 seq->size = GFS2_SEQ_GOODSIZE;
2042 static int gfs2_sbstats_open(struct inode *inode, struct file *file)
2044 int ret = seq_open_private(file, &gfs2_sbstats_seq_ops,
2045 sizeof(struct gfs2_glock_iter));
2047 struct seq_file *seq = file->private_data;
2048 struct gfs2_glock_iter *gi = seq->private;
2049 gi->sdp = inode->i_private;
2054 static const struct file_operations gfs2_glocks_fops = {
2055 .owner = THIS_MODULE,
2056 .open = gfs2_glocks_open,
2058 .llseek = seq_lseek,
2059 .release = seq_release_private,
2062 static const struct file_operations gfs2_glstats_fops = {
2063 .owner = THIS_MODULE,
2064 .open = gfs2_glstats_open,
2066 .llseek = seq_lseek,
2067 .release = seq_release_private,
2070 static const struct file_operations gfs2_sbstats_fops = {
2071 .owner = THIS_MODULE,
2072 .open = gfs2_sbstats_open,
2074 .llseek = seq_lseek,
2075 .release = seq_release_private,
2078 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2080 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2081 if (!sdp->debugfs_dir)
2083 sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
2085 sdp->debugfs_dir, sdp,
2087 if (!sdp->debugfs_dentry_glocks)
2090 sdp->debugfs_dentry_glstats = debugfs_create_file("glstats",
2092 sdp->debugfs_dir, sdp,
2093 &gfs2_glstats_fops);
2094 if (!sdp->debugfs_dentry_glstats)
2097 sdp->debugfs_dentry_sbstats = debugfs_create_file("sbstats",
2099 sdp->debugfs_dir, sdp,
2100 &gfs2_sbstats_fops);
2101 if (!sdp->debugfs_dentry_sbstats)
2106 gfs2_delete_debugfs_file(sdp);
2110 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2112 if (sdp->debugfs_dir) {
2113 if (sdp->debugfs_dentry_glocks) {
2114 debugfs_remove(sdp->debugfs_dentry_glocks);
2115 sdp->debugfs_dentry_glocks = NULL;
2117 if (sdp->debugfs_dentry_glstats) {
2118 debugfs_remove(sdp->debugfs_dentry_glstats);
2119 sdp->debugfs_dentry_glstats = NULL;
2121 if (sdp->debugfs_dentry_sbstats) {
2122 debugfs_remove(sdp->debugfs_dentry_sbstats);
2123 sdp->debugfs_dentry_sbstats = NULL;
2125 debugfs_remove(sdp->debugfs_dir);
2126 sdp->debugfs_dir = NULL;
2130 int gfs2_register_debugfs(void)
2132 gfs2_root = debugfs_create_dir("gfs2", NULL);
2133 return gfs2_root ? 0 : -ENOMEM;
2136 void gfs2_unregister_debugfs(void)
2138 debugfs_remove(gfs2_root);