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 spin_lock(&lru_lock);
179 if (!list_empty(&gl->gl_lru)) {
180 list_del_init(&gl->gl_lru);
181 atomic_dec(&lru_count);
182 clear_bit(GLF_LRU, &gl->gl_flags);
184 spin_unlock(&lru_lock);
188 * gfs2_glock_put() - Decrement reference count on glock
189 * @gl: The glock to put
193 void gfs2_glock_put(struct gfs2_glock *gl)
195 struct gfs2_sbd *sdp = gl->gl_sbd;
196 struct address_space *mapping = gfs2_glock2aspace(gl);
198 if (lockref_put_or_lock(&gl->gl_lockref))
201 lockref_mark_dead(&gl->gl_lockref);
203 gfs2_glock_remove_from_lru(gl);
204 spin_unlock(&gl->gl_lockref.lock);
205 spin_lock_bucket(gl->gl_hash);
206 hlist_bl_del_rcu(&gl->gl_list);
207 spin_unlock_bucket(gl->gl_hash);
208 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
209 GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
210 trace_gfs2_glock_put(gl);
211 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
215 * search_bucket() - Find struct gfs2_glock by lock number
216 * @bucket: the bucket to search
217 * @name: The lock name
219 * Returns: NULL, or the struct gfs2_glock with the requested number
222 static struct gfs2_glock *search_bucket(unsigned int hash,
223 const struct gfs2_sbd *sdp,
224 const struct lm_lockname *name)
226 struct gfs2_glock *gl;
227 struct hlist_bl_node *h;
229 hlist_bl_for_each_entry_rcu(gl, h, &gl_hash_table[hash], gl_list) {
230 if (!lm_name_equal(&gl->gl_name, name))
232 if (gl->gl_sbd != sdp)
234 if (lockref_get_not_dead(&gl->gl_lockref))
242 * may_grant - check if its ok to grant a new lock
244 * @gh: The lock request which we wish to grant
246 * Returns: true if its ok to grant the lock
249 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
251 const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
252 if ((gh->gh_state == LM_ST_EXCLUSIVE ||
253 gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
255 if (gl->gl_state == gh->gh_state)
257 if (gh->gh_flags & GL_EXACT)
259 if (gl->gl_state == LM_ST_EXCLUSIVE) {
260 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
262 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
265 if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
270 static void gfs2_holder_wake(struct gfs2_holder *gh)
272 clear_bit(HIF_WAIT, &gh->gh_iflags);
273 smp_mb__after_atomic();
274 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
278 * do_error - Something unexpected has happened during a lock request
282 static inline void do_error(struct gfs2_glock *gl, const int ret)
284 struct gfs2_holder *gh, *tmp;
286 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
287 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
289 if (ret & LM_OUT_ERROR)
291 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
292 gh->gh_error = GLR_TRYFAILED;
295 list_del_init(&gh->gh_list);
296 trace_gfs2_glock_queue(gh, 0);
297 gfs2_holder_wake(gh);
302 * do_promote - promote as many requests as possible on the current queue
305 * Returns: 1 if there is a blocked holder at the head of the list, or 2
306 * if a type specific operation is underway.
309 static int do_promote(struct gfs2_glock *gl)
310 __releases(&gl->gl_spin)
311 __acquires(&gl->gl_spin)
313 const struct gfs2_glock_operations *glops = gl->gl_ops;
314 struct gfs2_holder *gh, *tmp;
318 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
319 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
321 if (may_grant(gl, gh)) {
322 if (gh->gh_list.prev == &gl->gl_holders &&
324 spin_unlock(&gl->gl_spin);
325 /* FIXME: eliminate this eventually */
326 ret = glops->go_lock(gh);
327 spin_lock(&gl->gl_spin);
332 list_del_init(&gh->gh_list);
333 trace_gfs2_glock_queue(gh, 0);
334 gfs2_holder_wake(gh);
337 set_bit(HIF_HOLDER, &gh->gh_iflags);
338 trace_gfs2_promote(gh, 1);
339 gfs2_holder_wake(gh);
342 set_bit(HIF_HOLDER, &gh->gh_iflags);
343 trace_gfs2_promote(gh, 0);
344 gfs2_holder_wake(gh);
347 if (gh->gh_list.prev == &gl->gl_holders)
356 * find_first_waiter - find the first gh that's waiting for the glock
360 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
362 struct gfs2_holder *gh;
364 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
365 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
372 * state_change - record that the glock is now in a different state
374 * @new_state the new state
378 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
382 held1 = (gl->gl_state != LM_ST_UNLOCKED);
383 held2 = (new_state != LM_ST_UNLOCKED);
385 if (held1 != held2) {
386 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
388 gl->gl_lockref.count++;
390 gl->gl_lockref.count--;
392 if (held1 && held2 && list_empty(&gl->gl_holders))
393 clear_bit(GLF_QUEUED, &gl->gl_flags);
395 if (new_state != gl->gl_target)
396 /* shorten our minimum hold time */
397 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
399 gl->gl_state = new_state;
400 gl->gl_tchange = jiffies;
403 static void gfs2_demote_wake(struct gfs2_glock *gl)
405 gl->gl_demote_state = LM_ST_EXCLUSIVE;
406 clear_bit(GLF_DEMOTE, &gl->gl_flags);
407 smp_mb__after_atomic();
408 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
412 * finish_xmote - The DLM has replied to one of our lock requests
414 * @ret: The status from the DLM
418 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
420 const struct gfs2_glock_operations *glops = gl->gl_ops;
421 struct gfs2_holder *gh;
422 unsigned state = ret & LM_OUT_ST_MASK;
425 spin_lock(&gl->gl_spin);
426 trace_gfs2_glock_state_change(gl, state);
427 state_change(gl, state);
428 gh = find_first_waiter(gl);
430 /* Demote to UN request arrived during demote to SH or DF */
431 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
432 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
433 gl->gl_target = LM_ST_UNLOCKED;
435 /* Check for state != intended state */
436 if (unlikely(state != gl->gl_target)) {
437 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
438 /* move to back of queue and try next entry */
439 if (ret & LM_OUT_CANCELED) {
440 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
441 list_move_tail(&gh->gh_list, &gl->gl_holders);
442 gh = find_first_waiter(gl);
443 gl->gl_target = gh->gh_state;
446 /* Some error or failed "try lock" - report it */
447 if ((ret & LM_OUT_ERROR) ||
448 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
449 gl->gl_target = gl->gl_state;
455 /* Unlocked due to conversion deadlock, try again */
458 do_xmote(gl, gh, gl->gl_target);
460 /* Conversion fails, unlock and try again */
463 do_xmote(gl, gh, LM_ST_UNLOCKED);
465 default: /* Everything else */
466 pr_err("wanted %u got %u\n", gl->gl_target, state);
469 spin_unlock(&gl->gl_spin);
473 /* Fast path - we got what we asked for */
474 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
475 gfs2_demote_wake(gl);
476 if (state != LM_ST_UNLOCKED) {
477 if (glops->go_xmote_bh) {
478 spin_unlock(&gl->gl_spin);
479 rv = glops->go_xmote_bh(gl, gh);
480 spin_lock(&gl->gl_spin);
491 clear_bit(GLF_LOCK, &gl->gl_flags);
493 spin_unlock(&gl->gl_spin);
497 * do_xmote - Calls the DLM to change the state of a lock
498 * @gl: The lock state
499 * @gh: The holder (only for promotes)
500 * @target: The target lock state
504 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
505 __releases(&gl->gl_spin)
506 __acquires(&gl->gl_spin)
508 const struct gfs2_glock_operations *glops = gl->gl_ops;
509 struct gfs2_sbd *sdp = gl->gl_sbd;
510 unsigned int lck_flags = gh ? gh->gh_flags : 0;
513 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
515 GLOCK_BUG_ON(gl, gl->gl_state == target);
516 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
517 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
519 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
520 do_error(gl, 0); /* Fail queued try locks */
523 set_bit(GLF_BLOCKING, &gl->gl_flags);
524 if ((gl->gl_req == LM_ST_UNLOCKED) ||
525 (gl->gl_state == LM_ST_EXCLUSIVE) ||
526 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
527 clear_bit(GLF_BLOCKING, &gl->gl_flags);
528 spin_unlock(&gl->gl_spin);
531 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
532 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
533 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
536 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
538 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
540 pr_err("lm_lock ret %d\n", ret);
543 } else { /* lock_nolock */
544 finish_xmote(gl, target);
545 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
549 spin_lock(&gl->gl_spin);
553 * find_first_holder - find the first "holder" gh
557 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
559 struct gfs2_holder *gh;
561 if (!list_empty(&gl->gl_holders)) {
562 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
563 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
570 * run_queue - do all outstanding tasks related to a glock
571 * @gl: The glock in question
572 * @nonblock: True if we must not block in run_queue
576 static void run_queue(struct gfs2_glock *gl, const int nonblock)
577 __releases(&gl->gl_spin)
578 __acquires(&gl->gl_spin)
580 struct gfs2_holder *gh = NULL;
583 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
586 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
588 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
589 gl->gl_demote_state != gl->gl_state) {
590 if (find_first_holder(gl))
594 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
595 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
596 gl->gl_target = gl->gl_demote_state;
598 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
599 gfs2_demote_wake(gl);
600 ret = do_promote(gl);
605 gh = find_first_waiter(gl);
606 gl->gl_target = gh->gh_state;
607 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
608 do_error(gl, 0); /* Fail queued try locks */
610 do_xmote(gl, gh, gl->gl_target);
615 clear_bit(GLF_LOCK, &gl->gl_flags);
616 smp_mb__after_atomic();
617 gl->gl_lockref.count++;
618 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
619 gl->gl_lockref.count--;
623 clear_bit(GLF_LOCK, &gl->gl_flags);
624 smp_mb__after_atomic();
628 static void delete_work_func(struct work_struct *work)
630 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
631 struct gfs2_sbd *sdp = gl->gl_sbd;
632 struct gfs2_inode *ip;
634 u64 no_addr = gl->gl_name.ln_number;
637 /* Note: Unsafe to dereference ip as we don't hold right refs/locks */
640 inode = gfs2_ilookup(sdp->sd_vfs, no_addr, 1);
642 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
643 if (inode && !IS_ERR(inode)) {
644 d_prune_aliases(inode);
650 static void glock_work_func(struct work_struct *work)
652 unsigned long delay = 0;
653 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
656 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
657 finish_xmote(gl, gl->gl_reply);
660 spin_lock(&gl->gl_spin);
661 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
662 gl->gl_state != LM_ST_UNLOCKED &&
663 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
664 unsigned long holdtime, now = jiffies;
666 holdtime = gl->gl_tchange + gl->gl_hold_time;
667 if (time_before(now, holdtime))
668 delay = holdtime - now;
671 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
672 set_bit(GLF_DEMOTE, &gl->gl_flags);
676 spin_unlock(&gl->gl_spin);
680 if (gl->gl_name.ln_type != LM_TYPE_INODE)
682 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
690 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
691 * @sdp: The GFS2 superblock
692 * @number: the lock number
693 * @glops: The glock_operations to use
694 * @create: If 0, don't create the glock if it doesn't exist
695 * @glp: the glock is returned here
697 * This does not lock a glock, just finds/creates structures for one.
702 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
703 const struct gfs2_glock_operations *glops, int create,
704 struct gfs2_glock **glp)
706 struct super_block *s = sdp->sd_vfs;
707 struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
708 struct gfs2_glock *gl, *tmp;
709 unsigned int hash = gl_hash(sdp, &name);
710 struct address_space *mapping;
711 struct kmem_cache *cachep;
714 gl = search_bucket(hash, sdp, &name);
723 if (glops->go_flags & GLOF_ASPACE)
724 cachep = gfs2_glock_aspace_cachep;
726 cachep = gfs2_glock_cachep;
727 gl = kmem_cache_alloc(cachep, GFP_NOFS);
731 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
733 if (glops->go_flags & GLOF_LVB) {
734 gl->gl_lksb.sb_lvbptr = kzalloc(GFS2_MIN_LVB_SIZE, GFP_NOFS);
735 if (!gl->gl_lksb.sb_lvbptr) {
736 kmem_cache_free(cachep, gl);
741 atomic_inc(&sdp->sd_glock_disposal);
745 gl->gl_lockref.count = 1;
746 gl->gl_state = LM_ST_UNLOCKED;
747 gl->gl_target = LM_ST_UNLOCKED;
748 gl->gl_demote_state = LM_ST_EXCLUSIVE;
751 gl->gl_dstamp = ktime_set(0, 0);
753 /* We use the global stats to estimate the initial per-glock stats */
754 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
756 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
757 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
758 gl->gl_tchange = jiffies;
759 gl->gl_object = NULL;
760 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
761 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
762 INIT_WORK(&gl->gl_delete, delete_work_func);
764 mapping = gfs2_glock2aspace(gl);
766 mapping->a_ops = &gfs2_meta_aops;
767 mapping->host = s->s_bdev->bd_inode;
769 mapping_set_gfp_mask(mapping, GFP_NOFS);
770 mapping->private_data = NULL;
771 mapping->writeback_index = 0;
774 spin_lock_bucket(hash);
775 tmp = search_bucket(hash, sdp, &name);
777 spin_unlock_bucket(hash);
778 kfree(gl->gl_lksb.sb_lvbptr);
779 kmem_cache_free(cachep, gl);
780 atomic_dec(&sdp->sd_glock_disposal);
783 hlist_bl_add_head_rcu(&gl->gl_list, &gl_hash_table[hash]);
784 spin_unlock_bucket(hash);
793 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
795 * @state: the state we're requesting
796 * @flags: the modifier flags
797 * @gh: the holder structure
801 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
802 struct gfs2_holder *gh)
804 INIT_LIST_HEAD(&gh->gh_list);
806 gh->gh_ip = _RET_IP_;
807 gh->gh_owner_pid = get_pid(task_pid(current));
808 gh->gh_state = state;
809 gh->gh_flags = flags;
816 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
817 * @state: the state we're requesting
818 * @flags: the modifier flags
819 * @gh: the holder structure
821 * Don't mess with the glock.
825 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
827 gh->gh_state = state;
828 gh->gh_flags = flags;
830 gh->gh_ip = _RET_IP_;
831 put_pid(gh->gh_owner_pid);
832 gh->gh_owner_pid = get_pid(task_pid(current));
836 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
837 * @gh: the holder structure
841 void gfs2_holder_uninit(struct gfs2_holder *gh)
843 put_pid(gh->gh_owner_pid);
844 gfs2_glock_put(gh->gh_gl);
850 * gfs2_glock_wait - wait on a glock acquisition
851 * @gh: the glock holder
853 * Returns: 0 on success
856 int gfs2_glock_wait(struct gfs2_holder *gh)
858 unsigned long time1 = jiffies;
861 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
862 if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
863 /* Lengthen the minimum hold time. */
864 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
871 * handle_callback - process a demote request
873 * @state: the state the caller wants us to change to
875 * There are only two requests that we are going to see in actual
876 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
879 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
880 unsigned long delay, bool remote)
882 int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
884 set_bit(bit, &gl->gl_flags);
885 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
886 gl->gl_demote_state = state;
887 gl->gl_demote_time = jiffies;
888 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
889 gl->gl_demote_state != state) {
890 gl->gl_demote_state = LM_ST_UNLOCKED;
892 if (gl->gl_ops->go_callback)
893 gl->gl_ops->go_callback(gl, remote);
894 trace_gfs2_demote_rq(gl, remote);
897 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
899 struct va_format vaf;
905 seq_vprintf(seq, fmt, args);
917 * add_to_queue - Add a holder to the wait queue (but look for recursion)
918 * @gh: the holder structure to add
920 * Eventually we should move the recursive locking trap to a
921 * debugging option or something like that. This is the fast
922 * path and needs to have the minimum number of distractions.
926 static inline void add_to_queue(struct gfs2_holder *gh)
927 __releases(&gl->gl_spin)
928 __acquires(&gl->gl_spin)
930 struct gfs2_glock *gl = gh->gh_gl;
931 struct gfs2_sbd *sdp = gl->gl_sbd;
932 struct list_head *insert_pt = NULL;
933 struct gfs2_holder *gh2;
936 BUG_ON(gh->gh_owner_pid == NULL);
937 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
940 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
941 if (test_bit(GLF_LOCK, &gl->gl_flags))
942 try_futile = !may_grant(gl, gh);
943 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
947 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
948 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
949 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
952 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
954 gh->gh_error = GLR_TRYFAILED;
955 gfs2_holder_wake(gh);
958 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
960 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
961 insert_pt = &gh2->gh_list;
963 set_bit(GLF_QUEUED, &gl->gl_flags);
964 trace_gfs2_glock_queue(gh, 1);
965 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
966 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
967 if (likely(insert_pt == NULL)) {
968 list_add_tail(&gh->gh_list, &gl->gl_holders);
969 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
973 list_add_tail(&gh->gh_list, insert_pt);
975 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
976 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
977 spin_unlock(&gl->gl_spin);
978 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
979 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
980 spin_lock(&gl->gl_spin);
985 pr_err("original: %pSR\n", (void *)gh2->gh_ip);
986 pr_err("pid: %d\n", pid_nr(gh2->gh_owner_pid));
987 pr_err("lock type: %d req lock state : %d\n",
988 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
989 pr_err("new: %pSR\n", (void *)gh->gh_ip);
990 pr_err("pid: %d\n", pid_nr(gh->gh_owner_pid));
991 pr_err("lock type: %d req lock state : %d\n",
992 gh->gh_gl->gl_name.ln_type, gh->gh_state);
993 gfs2_dump_glock(NULL, gl);
998 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
999 * @gh: the holder structure
1001 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1003 * Returns: 0, GLR_TRYFAILED, or errno on failure
1006 int gfs2_glock_nq(struct gfs2_holder *gh)
1008 struct gfs2_glock *gl = gh->gh_gl;
1009 struct gfs2_sbd *sdp = gl->gl_sbd;
1012 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1015 if (test_bit(GLF_LRU, &gl->gl_flags))
1016 gfs2_glock_remove_from_lru(gl);
1018 spin_lock(&gl->gl_spin);
1020 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1021 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1022 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1023 gl->gl_lockref.count++;
1024 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1025 gl->gl_lockref.count--;
1028 spin_unlock(&gl->gl_spin);
1030 if (!(gh->gh_flags & GL_ASYNC))
1031 error = gfs2_glock_wait(gh);
1037 * gfs2_glock_poll - poll to see if an async request has been completed
1040 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1043 int gfs2_glock_poll(struct gfs2_holder *gh)
1045 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1049 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1050 * @gh: the glock holder
1054 void gfs2_glock_dq(struct gfs2_holder *gh)
1056 struct gfs2_glock *gl = gh->gh_gl;
1057 const struct gfs2_glock_operations *glops = gl->gl_ops;
1061 spin_lock(&gl->gl_spin);
1062 if (gh->gh_flags & GL_NOCACHE)
1063 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1065 list_del_init(&gh->gh_list);
1066 if (find_first_holder(gl) == NULL) {
1067 if (glops->go_unlock) {
1068 GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1069 spin_unlock(&gl->gl_spin);
1070 glops->go_unlock(gh);
1071 spin_lock(&gl->gl_spin);
1072 clear_bit(GLF_LOCK, &gl->gl_flags);
1074 if (list_empty(&gl->gl_holders) &&
1075 !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1076 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1079 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl) &&
1080 (glops->go_flags & GLOF_LRU))
1081 gfs2_glock_add_to_lru(gl);
1083 trace_gfs2_glock_queue(gh, 0);
1084 spin_unlock(&gl->gl_spin);
1085 if (likely(fast_path))
1088 gfs2_glock_hold(gl);
1089 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1090 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1091 gl->gl_name.ln_type == LM_TYPE_INODE)
1092 delay = gl->gl_hold_time;
1093 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1097 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1099 struct gfs2_glock *gl = gh->gh_gl;
1102 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1106 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1107 * @gh: the holder structure
1111 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1114 gfs2_holder_uninit(gh);
1118 * gfs2_glock_nq_num - acquire a glock based on lock number
1119 * @sdp: the filesystem
1120 * @number: the lock number
1121 * @glops: the glock operations for the type of glock
1122 * @state: the state to acquire the glock in
1123 * @flags: modifier flags for the acquisition
1124 * @gh: the struct gfs2_holder
1129 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1130 const struct gfs2_glock_operations *glops,
1131 unsigned int state, int flags, struct gfs2_holder *gh)
1133 struct gfs2_glock *gl;
1136 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1138 error = gfs2_glock_nq_init(gl, state, flags, gh);
1146 * glock_compare - Compare two struct gfs2_glock structures for sorting
1147 * @arg_a: the first structure
1148 * @arg_b: the second structure
1152 static int glock_compare(const void *arg_a, const void *arg_b)
1154 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1155 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1156 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1157 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1159 if (a->ln_number > b->ln_number)
1161 if (a->ln_number < b->ln_number)
1163 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1168 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1169 * @num_gh: the number of structures
1170 * @ghs: an array of struct gfs2_holder structures
1172 * Returns: 0 on success (all glocks acquired),
1173 * errno on failure (no glocks acquired)
1176 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1177 struct gfs2_holder **p)
1182 for (x = 0; x < num_gh; x++)
1185 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1187 for (x = 0; x < num_gh; x++) {
1188 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1190 error = gfs2_glock_nq(p[x]);
1193 gfs2_glock_dq(p[x]);
1202 * gfs2_glock_nq_m - acquire multiple glocks
1203 * @num_gh: the number of structures
1204 * @ghs: an array of struct gfs2_holder structures
1207 * Returns: 0 on success (all glocks acquired),
1208 * errno on failure (no glocks acquired)
1211 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1213 struct gfs2_holder *tmp[4];
1214 struct gfs2_holder **pph = tmp;
1221 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1222 return gfs2_glock_nq(ghs);
1226 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1231 error = nq_m_sync(num_gh, ghs, pph);
1240 * gfs2_glock_dq_m - release multiple glocks
1241 * @num_gh: the number of structures
1242 * @ghs: an array of struct gfs2_holder structures
1246 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1249 gfs2_glock_dq(&ghs[num_gh]);
1252 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1254 unsigned long delay = 0;
1255 unsigned long holdtime;
1256 unsigned long now = jiffies;
1258 gfs2_glock_hold(gl);
1259 holdtime = gl->gl_tchange + gl->gl_hold_time;
1260 if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1261 gl->gl_name.ln_type == LM_TYPE_INODE) {
1262 if (time_before(now, holdtime))
1263 delay = holdtime - now;
1264 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1265 delay = gl->gl_hold_time;
1268 spin_lock(&gl->gl_spin);
1269 handle_callback(gl, state, delay, true);
1270 spin_unlock(&gl->gl_spin);
1271 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1276 * gfs2_should_freeze - Figure out if glock should be frozen
1277 * @gl: The glock in question
1279 * Glocks are not frozen if (a) the result of the dlm operation is
1280 * an error, (b) the locking operation was an unlock operation or
1281 * (c) if there is a "noexp" flagged request anywhere in the queue
1283 * Returns: 1 if freezing should occur, 0 otherwise
1286 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1288 const struct gfs2_holder *gh;
1290 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1292 if (gl->gl_target == LM_ST_UNLOCKED)
1295 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1296 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1298 if (LM_FLAG_NOEXP & gh->gh_flags)
1306 * gfs2_glock_complete - Callback used by locking
1307 * @gl: Pointer to the glock
1308 * @ret: The return value from the dlm
1310 * The gl_reply field is under the gl_spin lock so that it is ok
1311 * to use a bitfield shared with other glock state fields.
1314 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1316 struct lm_lockstruct *ls = &gl->gl_sbd->sd_lockstruct;
1318 spin_lock(&gl->gl_spin);
1321 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1322 if (gfs2_should_freeze(gl)) {
1323 set_bit(GLF_FROZEN, &gl->gl_flags);
1324 spin_unlock(&gl->gl_spin);
1329 gl->gl_lockref.count++;
1330 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1331 spin_unlock(&gl->gl_spin);
1333 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1337 static int glock_cmp(void *priv, struct list_head *a, struct list_head *b)
1339 struct gfs2_glock *gla, *glb;
1341 gla = list_entry(a, struct gfs2_glock, gl_lru);
1342 glb = list_entry(b, struct gfs2_glock, gl_lru);
1344 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1346 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1353 * gfs2_dispose_glock_lru - Demote a list of glocks
1354 * @list: The list to dispose of
1356 * Disposing of glocks may involve disk accesses, so that here we sort
1357 * the glocks by number (i.e. disk location of the inodes) so that if
1358 * there are any such accesses, they'll be sent in order (mostly).
1360 * Must be called under the lru_lock, but may drop and retake this
1361 * lock. While the lru_lock is dropped, entries may vanish from the
1362 * list, but no new entries will appear on the list (since it is
1366 static void gfs2_dispose_glock_lru(struct list_head *list)
1367 __releases(&lru_lock)
1368 __acquires(&lru_lock)
1370 struct gfs2_glock *gl;
1372 list_sort(NULL, list, glock_cmp);
1374 while(!list_empty(list)) {
1375 gl = list_entry(list->next, struct gfs2_glock, gl_lru);
1376 list_del_init(&gl->gl_lru);
1377 if (!spin_trylock(&gl->gl_spin)) {
1379 list_add(&gl->gl_lru, &lru_list);
1380 atomic_inc(&lru_count);
1383 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1384 spin_unlock(&gl->gl_spin);
1385 goto add_back_to_lru;
1387 clear_bit(GLF_LRU, &gl->gl_flags);
1388 gl->gl_lockref.count++;
1390 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1391 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1392 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1393 gl->gl_lockref.count--;
1394 spin_unlock(&gl->gl_spin);
1395 cond_resched_lock(&lru_lock);
1400 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1401 * @nr: The number of entries to scan
1403 * This function selects the entries on the LRU which are able to
1404 * be demoted, and then kicks off the process by calling
1405 * gfs2_dispose_glock_lru() above.
1408 static long gfs2_scan_glock_lru(int nr)
1410 struct gfs2_glock *gl;
1415 spin_lock(&lru_lock);
1416 while ((nr-- >= 0) && !list_empty(&lru_list)) {
1417 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1419 /* Test for being demotable */
1420 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1421 list_move(&gl->gl_lru, &dispose);
1422 atomic_dec(&lru_count);
1427 list_move(&gl->gl_lru, &skipped);
1429 list_splice(&skipped, &lru_list);
1430 if (!list_empty(&dispose))
1431 gfs2_dispose_glock_lru(&dispose);
1432 spin_unlock(&lru_lock);
1437 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1438 struct shrink_control *sc)
1440 if (!(sc->gfp_mask & __GFP_FS))
1442 return gfs2_scan_glock_lru(sc->nr_to_scan);
1445 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1446 struct shrink_control *sc)
1448 return vfs_pressure_ratio(atomic_read(&lru_count));
1451 static struct shrinker glock_shrinker = {
1452 .seeks = DEFAULT_SEEKS,
1453 .count_objects = gfs2_glock_shrink_count,
1454 .scan_objects = gfs2_glock_shrink_scan,
1458 * examine_bucket - Call a function for glock in a hash bucket
1459 * @examiner: the function
1460 * @sdp: the filesystem
1461 * @bucket: the bucket
1465 static void examine_bucket(glock_examiner examiner, const struct gfs2_sbd *sdp,
1468 struct gfs2_glock *gl;
1469 struct hlist_bl_head *head = &gl_hash_table[hash];
1470 struct hlist_bl_node *pos;
1473 hlist_bl_for_each_entry_rcu(gl, pos, head, gl_list) {
1474 if ((gl->gl_sbd == sdp) && lockref_get_not_dead(&gl->gl_lockref))
1481 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1485 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1486 examine_bucket(examiner, sdp, x);
1491 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1492 * @gl: The glock to thaw
1496 static void thaw_glock(struct gfs2_glock *gl)
1498 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1500 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1501 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0) {
1508 * clear_glock - look at a glock and see if we can free it from glock cache
1509 * @gl: the glock to look at
1513 static void clear_glock(struct gfs2_glock *gl)
1515 gfs2_glock_remove_from_lru(gl);
1517 spin_lock(&gl->gl_spin);
1518 if (gl->gl_state != LM_ST_UNLOCKED)
1519 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1520 spin_unlock(&gl->gl_spin);
1521 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1526 * gfs2_glock_thaw - Thaw any frozen glocks
1527 * @sdp: The super block
1531 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1533 glock_hash_walk(thaw_glock, sdp);
1536 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1538 spin_lock(&gl->gl_spin);
1539 gfs2_dump_glock(seq, gl);
1540 spin_unlock(&gl->gl_spin);
1543 static void dump_glock_func(struct gfs2_glock *gl)
1545 dump_glock(NULL, gl);
1549 * gfs2_gl_hash_clear - Empty out the glock hash table
1550 * @sdp: the filesystem
1551 * @wait: wait until it's all gone
1553 * Called when unmounting the filesystem.
1556 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1558 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1559 flush_workqueue(glock_workqueue);
1560 glock_hash_walk(clear_glock, sdp);
1561 flush_workqueue(glock_workqueue);
1562 wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
1563 glock_hash_walk(dump_glock_func, sdp);
1566 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1568 struct gfs2_glock *gl = ip->i_gl;
1571 ret = gfs2_truncatei_resume(ip);
1572 gfs2_assert_withdraw(gl->gl_sbd, ret == 0);
1574 spin_lock(&gl->gl_spin);
1575 clear_bit(GLF_LOCK, &gl->gl_flags);
1577 spin_unlock(&gl->gl_spin);
1580 static const char *state2str(unsigned state)
1583 case LM_ST_UNLOCKED:
1587 case LM_ST_DEFERRED:
1589 case LM_ST_EXCLUSIVE:
1595 static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1598 if (flags & LM_FLAG_TRY)
1600 if (flags & LM_FLAG_TRY_1CB)
1602 if (flags & LM_FLAG_NOEXP)
1604 if (flags & LM_FLAG_ANY)
1606 if (flags & LM_FLAG_PRIORITY)
1608 if (flags & GL_ASYNC)
1610 if (flags & GL_EXACT)
1612 if (flags & GL_NOCACHE)
1614 if (test_bit(HIF_HOLDER, &iflags))
1616 if (test_bit(HIF_WAIT, &iflags))
1618 if (test_bit(HIF_FIRST, &iflags))
1625 * dump_holder - print information about a glock holder
1626 * @seq: the seq_file struct
1627 * @gh: the glock holder
1631 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1633 struct task_struct *gh_owner = NULL;
1637 if (gh->gh_owner_pid)
1638 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1639 gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1640 state2str(gh->gh_state),
1641 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1643 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1644 gh_owner ? gh_owner->comm : "(ended)",
1649 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1651 const unsigned long *gflags = &gl->gl_flags;
1654 if (test_bit(GLF_LOCK, gflags))
1656 if (test_bit(GLF_DEMOTE, gflags))
1658 if (test_bit(GLF_PENDING_DEMOTE, gflags))
1660 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1662 if (test_bit(GLF_DIRTY, gflags))
1664 if (test_bit(GLF_LFLUSH, gflags))
1666 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1668 if (test_bit(GLF_REPLY_PENDING, gflags))
1670 if (test_bit(GLF_INITIAL, gflags))
1672 if (test_bit(GLF_FROZEN, gflags))
1674 if (test_bit(GLF_QUEUED, gflags))
1676 if (test_bit(GLF_LRU, gflags))
1680 if (test_bit(GLF_BLOCKING, gflags))
1687 * gfs2_dump_glock - print information about a glock
1688 * @seq: The seq_file struct
1691 * The file format is as follows:
1692 * One line per object, capital letters are used to indicate objects
1693 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1694 * other objects are indented by a single space and follow the glock to
1695 * which they are related. Fields are indicated by lower case letters
1696 * followed by a colon and the field value, except for strings which are in
1697 * [] so that its possible to see if they are composed of spaces for
1698 * example. The field's are n = number (id of the object), f = flags,
1699 * t = type, s = state, r = refcount, e = error, p = pid.
1703 void gfs2_dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1705 const struct gfs2_glock_operations *glops = gl->gl_ops;
1706 unsigned long long dtime;
1707 const struct gfs2_holder *gh;
1708 char gflags_buf[32];
1710 dtime = jiffies - gl->gl_demote_time;
1711 dtime *= 1000000/HZ; /* demote time in uSec */
1712 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1714 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",
1715 state2str(gl->gl_state),
1716 gl->gl_name.ln_type,
1717 (unsigned long long)gl->gl_name.ln_number,
1718 gflags2str(gflags_buf, gl),
1719 state2str(gl->gl_target),
1720 state2str(gl->gl_demote_state), dtime,
1721 atomic_read(&gl->gl_ail_count),
1722 atomic_read(&gl->gl_revokes),
1723 (int)gl->gl_lockref.count, gl->gl_hold_time);
1725 list_for_each_entry(gh, &gl->gl_holders, gh_list)
1726 dump_holder(seq, gh);
1728 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1729 glops->go_dump(seq, gl);
1732 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
1734 struct gfs2_glock *gl = iter_ptr;
1736 seq_printf(seq, "G: n:%u/%llx rtt:%lld/%lld rttb:%lld/%lld irt:%lld/%lld dcnt: %lld qcnt: %lld\n",
1737 gl->gl_name.ln_type,
1738 (unsigned long long)gl->gl_name.ln_number,
1739 (long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
1740 (long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
1741 (long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
1742 (long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
1743 (long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
1744 (long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
1745 (long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
1746 (long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
1750 static const char *gfs2_gltype[] = {
1764 static const char *gfs2_stype[] = {
1765 [GFS2_LKS_SRTT] = "srtt",
1766 [GFS2_LKS_SRTTVAR] = "srttvar",
1767 [GFS2_LKS_SRTTB] = "srttb",
1768 [GFS2_LKS_SRTTVARB] = "srttvarb",
1769 [GFS2_LKS_SIRT] = "sirt",
1770 [GFS2_LKS_SIRTVAR] = "sirtvar",
1771 [GFS2_LKS_DCOUNT] = "dlm",
1772 [GFS2_LKS_QCOUNT] = "queue",
1775 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1777 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
1779 struct gfs2_glock_iter *gi = seq->private;
1780 struct gfs2_sbd *sdp = gi->sdp;
1781 unsigned index = gi->hash >> 3;
1782 unsigned subindex = gi->hash & 0x07;
1786 if (index == 0 && subindex != 0)
1789 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
1790 (index == 0) ? "cpu": gfs2_stype[subindex]);
1792 for_each_possible_cpu(i) {
1793 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
1797 value = lkstats->lkstats[index - 1].stats[subindex];
1799 seq_printf(seq, " %15lld", (long long)value);
1801 seq_putc(seq, '\n');
1805 int __init gfs2_glock_init(void)
1808 for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
1809 INIT_HLIST_BL_HEAD(&gl_hash_table[i]);
1812 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1813 WQ_HIGHPRI | WQ_FREEZABLE, 0);
1814 if (!glock_workqueue)
1816 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1817 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1819 if (!gfs2_delete_workqueue) {
1820 destroy_workqueue(glock_workqueue);
1824 register_shrinker(&glock_shrinker);
1829 void gfs2_glock_exit(void)
1831 unregister_shrinker(&glock_shrinker);
1832 destroy_workqueue(glock_workqueue);
1833 destroy_workqueue(gfs2_delete_workqueue);
1836 static inline struct gfs2_glock *glock_hash_chain(unsigned hash)
1838 return hlist_bl_entry(hlist_bl_first_rcu(&gl_hash_table[hash]),
1839 struct gfs2_glock, gl_list);
1842 static inline struct gfs2_glock *glock_hash_next(struct gfs2_glock *gl)
1844 return hlist_bl_entry(rcu_dereference(gl->gl_list.next),
1845 struct gfs2_glock, gl_list);
1848 static int gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1850 struct gfs2_glock *gl;
1855 gi->gl = glock_hash_next(gl);
1858 if (gi->hash >= GFS2_GL_HASH_SIZE) {
1862 gi->gl = glock_hash_chain(gi->hash);
1865 while (gi->gl == NULL) {
1867 if (gi->hash >= GFS2_GL_HASH_SIZE) {
1871 gi->gl = glock_hash_chain(gi->hash);
1874 /* Skip entries for other sb and dead entries */
1875 } while (gi->sdp != gi->gl->gl_sbd ||
1876 __lockref_is_dead(&gi->gl->gl_lockref));
1881 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1883 struct gfs2_glock_iter *gi = seq->private;
1886 if (gi->last_pos <= *pos)
1887 n = gi->nhash + (*pos - gi->last_pos);
1895 if (gfs2_glock_iter_next(gi))
1899 gi->last_pos = *pos;
1903 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1906 struct gfs2_glock_iter *gi = seq->private;
1909 gi->last_pos = *pos;
1910 if (gfs2_glock_iter_next(gi))
1916 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1918 struct gfs2_glock_iter *gi = seq->private;
1925 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1927 dump_glock(seq, iter_ptr);
1931 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
1933 struct gfs2_glock_iter *gi = seq->private;
1936 if (*pos >= GFS2_NR_SBSTATS)
1939 return SEQ_START_TOKEN;
1942 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
1945 struct gfs2_glock_iter *gi = seq->private;
1948 if (gi->hash >= GFS2_NR_SBSTATS) {
1952 return SEQ_START_TOKEN;
1955 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
1960 static const struct seq_operations gfs2_glock_seq_ops = {
1961 .start = gfs2_glock_seq_start,
1962 .next = gfs2_glock_seq_next,
1963 .stop = gfs2_glock_seq_stop,
1964 .show = gfs2_glock_seq_show,
1967 static const struct seq_operations gfs2_glstats_seq_ops = {
1968 .start = gfs2_glock_seq_start,
1969 .next = gfs2_glock_seq_next,
1970 .stop = gfs2_glock_seq_stop,
1971 .show = gfs2_glstats_seq_show,
1974 static const struct seq_operations gfs2_sbstats_seq_ops = {
1975 .start = gfs2_sbstats_seq_start,
1976 .next = gfs2_sbstats_seq_next,
1977 .stop = gfs2_sbstats_seq_stop,
1978 .show = gfs2_sbstats_seq_show,
1981 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
1983 static int gfs2_glocks_open(struct inode *inode, struct file *file)
1985 int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1986 sizeof(struct gfs2_glock_iter));
1988 struct seq_file *seq = file->private_data;
1989 struct gfs2_glock_iter *gi = seq->private;
1990 gi->sdp = inode->i_private;
1991 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
1993 seq->size = GFS2_SEQ_GOODSIZE;
1998 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2000 int ret = seq_open_private(file, &gfs2_glstats_seq_ops,
2001 sizeof(struct gfs2_glock_iter));
2003 struct seq_file *seq = file->private_data;
2004 struct gfs2_glock_iter *gi = seq->private;
2005 gi->sdp = inode->i_private;
2006 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2008 seq->size = GFS2_SEQ_GOODSIZE;
2013 static int gfs2_sbstats_open(struct inode *inode, struct file *file)
2015 int ret = seq_open_private(file, &gfs2_sbstats_seq_ops,
2016 sizeof(struct gfs2_glock_iter));
2018 struct seq_file *seq = file->private_data;
2019 struct gfs2_glock_iter *gi = seq->private;
2020 gi->sdp = inode->i_private;
2025 static const struct file_operations gfs2_glocks_fops = {
2026 .owner = THIS_MODULE,
2027 .open = gfs2_glocks_open,
2029 .llseek = seq_lseek,
2030 .release = seq_release_private,
2033 static const struct file_operations gfs2_glstats_fops = {
2034 .owner = THIS_MODULE,
2035 .open = gfs2_glstats_open,
2037 .llseek = seq_lseek,
2038 .release = seq_release_private,
2041 static const struct file_operations gfs2_sbstats_fops = {
2042 .owner = THIS_MODULE,
2043 .open = gfs2_sbstats_open,
2045 .llseek = seq_lseek,
2046 .release = seq_release_private,
2049 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2051 struct dentry *dent;
2053 dent = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2054 if (IS_ERR_OR_NULL(dent))
2056 sdp->debugfs_dir = dent;
2058 dent = debugfs_create_file("glocks",
2060 sdp->debugfs_dir, sdp,
2062 if (IS_ERR_OR_NULL(dent))
2064 sdp->debugfs_dentry_glocks = dent;
2066 dent = debugfs_create_file("glstats",
2068 sdp->debugfs_dir, sdp,
2069 &gfs2_glstats_fops);
2070 if (IS_ERR_OR_NULL(dent))
2072 sdp->debugfs_dentry_glstats = dent;
2074 dent = debugfs_create_file("sbstats",
2076 sdp->debugfs_dir, sdp,
2077 &gfs2_sbstats_fops);
2078 if (IS_ERR_OR_NULL(dent))
2080 sdp->debugfs_dentry_sbstats = dent;
2084 gfs2_delete_debugfs_file(sdp);
2085 return dent ? PTR_ERR(dent) : -ENOMEM;
2088 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2090 if (sdp->debugfs_dir) {
2091 if (sdp->debugfs_dentry_glocks) {
2092 debugfs_remove(sdp->debugfs_dentry_glocks);
2093 sdp->debugfs_dentry_glocks = NULL;
2095 if (sdp->debugfs_dentry_glstats) {
2096 debugfs_remove(sdp->debugfs_dentry_glstats);
2097 sdp->debugfs_dentry_glstats = NULL;
2099 if (sdp->debugfs_dentry_sbstats) {
2100 debugfs_remove(sdp->debugfs_dentry_sbstats);
2101 sdp->debugfs_dentry_sbstats = NULL;
2103 debugfs_remove(sdp->debugfs_dir);
2104 sdp->debugfs_dir = NULL;
2108 int gfs2_register_debugfs(void)
2110 gfs2_root = debugfs_create_dir("gfs2", NULL);
2111 if (IS_ERR(gfs2_root))
2112 return PTR_ERR(gfs2_root);
2113 return gfs2_root ? 0 : -ENOMEM;
2116 void gfs2_unregister_debugfs(void)
2118 debugfs_remove(gfs2_root);