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 #include <linux/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/buffer_head.h>
14 #include <linux/delay.h>
15 #include <linux/sort.h>
16 #include <linux/jhash.h>
17 #include <linux/kallsyms.h>
18 #include <linux/gfs2_ondisk.h>
19 #include <linux/list.h>
20 #include <linux/wait.h>
21 #include <linux/module.h>
22 #include <asm/uaccess.h>
23 #include <linux/seq_file.h>
24 #include <linux/debugfs.h>
25 #include <linux/kthread.h>
26 #include <linux/freezer.h>
27 #include <linux/workqueue.h>
28 #include <linux/jiffies.h>
41 #define CREATE_TRACE_POINTS
42 #include "trace_gfs2.h"
44 struct gfs2_gl_hash_bucket {
45 struct hlist_head hb_list;
48 struct gfs2_glock_iter {
49 int hash; /* hash bucket index */
50 struct gfs2_sbd *sdp; /* incore superblock */
51 struct gfs2_glock *gl; /* current glock struct */
52 char string[512]; /* scratch space */
55 typedef void (*glock_examiner) (struct gfs2_glock * gl);
57 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp);
58 static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl);
59 #define GLOCK_BUG_ON(gl,x) do { if (unlikely(x)) { __dump_glock(NULL, gl); BUG(); } } while(0)
60 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
62 static struct dentry *gfs2_root;
63 static struct workqueue_struct *glock_workqueue;
64 struct workqueue_struct *gfs2_delete_workqueue;
65 static LIST_HEAD(lru_list);
66 static atomic_t lru_count = ATOMIC_INIT(0);
67 static DEFINE_SPINLOCK(lru_lock);
69 #define GFS2_GL_HASH_SHIFT 15
70 #define GFS2_GL_HASH_SIZE (1 << GFS2_GL_HASH_SHIFT)
71 #define GFS2_GL_HASH_MASK (GFS2_GL_HASH_SIZE - 1)
73 static struct gfs2_gl_hash_bucket gl_hash_table[GFS2_GL_HASH_SIZE];
74 static struct dentry *gfs2_root;
77 * Despite what you might think, the numbers below are not arbitrary :-)
78 * They are taken from the ipv4 routing hash code, which is well tested
79 * and thus should be nearly optimal. Later on we might tweek the numbers
80 * but for now this should be fine.
82 * The reason for putting the locks in a separate array from the list heads
83 * is that we can have fewer locks than list heads and save memory. We use
84 * the same hash function for both, but with a different hash mask.
86 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
87 defined(CONFIG_PROVE_LOCKING)
90 # define GL_HASH_LOCK_SZ 256
93 # define GL_HASH_LOCK_SZ 4096
95 # define GL_HASH_LOCK_SZ 2048
97 # define GL_HASH_LOCK_SZ 1024
99 # define GL_HASH_LOCK_SZ 512
101 # define GL_HASH_LOCK_SZ 256
105 /* We never want more locks than chains */
106 #if GFS2_GL_HASH_SIZE < GL_HASH_LOCK_SZ
107 # undef GL_HASH_LOCK_SZ
108 # define GL_HASH_LOCK_SZ GFS2_GL_HASH_SIZE
111 static rwlock_t gl_hash_locks[GL_HASH_LOCK_SZ];
113 static inline rwlock_t *gl_lock_addr(unsigned int x)
115 return &gl_hash_locks[x & (GL_HASH_LOCK_SZ-1)];
117 #else /* not SMP, so no spinlocks required */
118 static inline rwlock_t *gl_lock_addr(unsigned int x)
125 * gl_hash() - Turn glock number into hash bucket number
126 * @lock: The glock number
128 * Returns: The number of the corresponding hash bucket
131 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
132 const struct lm_lockname *name)
136 h = jhash(&name->ln_number, sizeof(u64), 0);
137 h = jhash(&name->ln_type, sizeof(unsigned int), h);
138 h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
139 h &= GFS2_GL_HASH_MASK;
145 * glock_free() - Perform a few checks and then release struct gfs2_glock
146 * @gl: The glock to release
148 * Also calls lock module to release its internal structure for this glock.
152 static void glock_free(struct gfs2_glock *gl)
154 struct gfs2_sbd *sdp = gl->gl_sbd;
155 struct address_space *mapping = gfs2_glock2aspace(gl);
156 struct kmem_cache *cachep = gfs2_glock_cachep;
158 GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
159 trace_gfs2_glock_put(gl);
161 cachep = gfs2_glock_aspace_cachep;
162 sdp->sd_lockstruct.ls_ops->lm_put_lock(cachep, gl);
166 * gfs2_glock_hold() - increment reference count on glock
167 * @gl: The glock to hold
171 void gfs2_glock_hold(struct gfs2_glock *gl)
173 GLOCK_BUG_ON(gl, atomic_read(&gl->gl_ref) == 0);
174 atomic_inc(&gl->gl_ref);
178 * demote_ok - Check to see if it's ok to unlock a glock
181 * Returns: 1 if it's ok
184 static int demote_ok(const struct gfs2_glock *gl)
186 const struct gfs2_glock_operations *glops = gl->gl_ops;
188 if (gl->gl_state == LM_ST_UNLOCKED)
190 if (!list_empty(&gl->gl_holders))
192 if (glops->go_demote_ok)
193 return glops->go_demote_ok(gl);
198 * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
203 static void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
206 may_reclaim = (demote_ok(gl) &&
207 (atomic_read(&gl->gl_ref) == 1 ||
208 (gl->gl_name.ln_type == LM_TYPE_INODE &&
209 atomic_read(&gl->gl_ref) <= 2)));
210 spin_lock(&lru_lock);
211 if (list_empty(&gl->gl_lru) && may_reclaim) {
212 list_add_tail(&gl->gl_lru, &lru_list);
213 atomic_inc(&lru_count);
215 spin_unlock(&lru_lock);
219 * gfs2_glock_put_nolock() - Decrement reference count on glock
220 * @gl: The glock to put
222 * This function should only be used if the caller has its own reference
223 * to the glock, in addition to the one it is dropping.
226 void gfs2_glock_put_nolock(struct gfs2_glock *gl)
228 if (atomic_dec_and_test(&gl->gl_ref))
230 gfs2_glock_schedule_for_reclaim(gl);
234 * gfs2_glock_put() - Decrement reference count on glock
235 * @gl: The glock to put
239 int gfs2_glock_put(struct gfs2_glock *gl)
243 write_lock(gl_lock_addr(gl->gl_hash));
244 if (atomic_dec_and_lock(&gl->gl_ref, &lru_lock)) {
245 hlist_del(&gl->gl_list);
246 if (!list_empty(&gl->gl_lru)) {
247 list_del_init(&gl->gl_lru);
248 atomic_dec(&lru_count);
250 spin_unlock(&lru_lock);
251 write_unlock(gl_lock_addr(gl->gl_hash));
252 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
257 spin_lock(&gl->gl_spin);
258 gfs2_glock_schedule_for_reclaim(gl);
259 spin_unlock(&gl->gl_spin);
260 write_unlock(gl_lock_addr(gl->gl_hash));
266 * search_bucket() - Find struct gfs2_glock by lock number
267 * @bucket: the bucket to search
268 * @name: The lock name
270 * Returns: NULL, or the struct gfs2_glock with the requested number
273 static struct gfs2_glock *search_bucket(unsigned int hash,
274 const struct gfs2_sbd *sdp,
275 const struct lm_lockname *name)
277 struct gfs2_glock *gl;
278 struct hlist_node *h;
280 hlist_for_each_entry(gl, h, &gl_hash_table[hash].hb_list, gl_list) {
281 if (!lm_name_equal(&gl->gl_name, name))
283 if (gl->gl_sbd != sdp)
286 atomic_inc(&gl->gl_ref);
295 * may_grant - check if its ok to grant a new lock
297 * @gh: The lock request which we wish to grant
299 * Returns: true if its ok to grant the lock
302 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
304 const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
305 if ((gh->gh_state == LM_ST_EXCLUSIVE ||
306 gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
308 if (gl->gl_state == gh->gh_state)
310 if (gh->gh_flags & GL_EXACT)
312 if (gl->gl_state == LM_ST_EXCLUSIVE) {
313 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
315 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
318 if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
323 static void gfs2_holder_wake(struct gfs2_holder *gh)
325 clear_bit(HIF_WAIT, &gh->gh_iflags);
326 smp_mb__after_clear_bit();
327 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
331 * do_error - Something unexpected has happened during a lock request
335 static inline void do_error(struct gfs2_glock *gl, const int ret)
337 struct gfs2_holder *gh, *tmp;
339 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
340 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
342 if (ret & LM_OUT_ERROR)
344 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
345 gh->gh_error = GLR_TRYFAILED;
348 list_del_init(&gh->gh_list);
349 trace_gfs2_glock_queue(gh, 0);
350 gfs2_holder_wake(gh);
355 * do_promote - promote as many requests as possible on the current queue
358 * Returns: 1 if there is a blocked holder at the head of the list, or 2
359 * if a type specific operation is underway.
362 static int do_promote(struct gfs2_glock *gl)
363 __releases(&gl->gl_spin)
364 __acquires(&gl->gl_spin)
366 const struct gfs2_glock_operations *glops = gl->gl_ops;
367 struct gfs2_holder *gh, *tmp;
371 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
372 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
374 if (may_grant(gl, gh)) {
375 if (gh->gh_list.prev == &gl->gl_holders &&
377 spin_unlock(&gl->gl_spin);
378 /* FIXME: eliminate this eventually */
379 ret = glops->go_lock(gh);
380 spin_lock(&gl->gl_spin);
385 list_del_init(&gh->gh_list);
386 trace_gfs2_glock_queue(gh, 0);
387 gfs2_holder_wake(gh);
390 set_bit(HIF_HOLDER, &gh->gh_iflags);
391 trace_gfs2_promote(gh, 1);
392 gfs2_holder_wake(gh);
395 set_bit(HIF_HOLDER, &gh->gh_iflags);
396 trace_gfs2_promote(gh, 0);
397 gfs2_holder_wake(gh);
400 if (gh->gh_list.prev == &gl->gl_holders)
409 * find_first_waiter - find the first gh that's waiting for the glock
413 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
415 struct gfs2_holder *gh;
417 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
418 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
425 * state_change - record that the glock is now in a different state
427 * @new_state the new state
431 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
435 held1 = (gl->gl_state != LM_ST_UNLOCKED);
436 held2 = (new_state != LM_ST_UNLOCKED);
438 if (held1 != held2) {
442 gfs2_glock_put_nolock(gl);
444 if (held1 && held2 && list_empty(&gl->gl_holders))
445 clear_bit(GLF_QUEUED, &gl->gl_flags);
447 gl->gl_state = new_state;
448 gl->gl_tchange = jiffies;
451 static void gfs2_demote_wake(struct gfs2_glock *gl)
453 gl->gl_demote_state = LM_ST_EXCLUSIVE;
454 clear_bit(GLF_DEMOTE, &gl->gl_flags);
455 smp_mb__after_clear_bit();
456 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
460 * finish_xmote - The DLM has replied to one of our lock requests
462 * @ret: The status from the DLM
466 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
468 const struct gfs2_glock_operations *glops = gl->gl_ops;
469 struct gfs2_holder *gh;
470 unsigned state = ret & LM_OUT_ST_MASK;
473 spin_lock(&gl->gl_spin);
474 trace_gfs2_glock_state_change(gl, state);
475 state_change(gl, state);
476 gh = find_first_waiter(gl);
478 /* Demote to UN request arrived during demote to SH or DF */
479 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
480 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
481 gl->gl_target = LM_ST_UNLOCKED;
483 /* Check for state != intended state */
484 if (unlikely(state != gl->gl_target)) {
485 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
486 /* move to back of queue and try next entry */
487 if (ret & LM_OUT_CANCELED) {
488 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
489 list_move_tail(&gh->gh_list, &gl->gl_holders);
490 gh = find_first_waiter(gl);
491 gl->gl_target = gh->gh_state;
494 /* Some error or failed "try lock" - report it */
495 if ((ret & LM_OUT_ERROR) ||
496 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
497 gl->gl_target = gl->gl_state;
503 /* Unlocked due to conversion deadlock, try again */
506 do_xmote(gl, gh, gl->gl_target);
508 /* Conversion fails, unlock and try again */
511 do_xmote(gl, gh, LM_ST_UNLOCKED);
513 default: /* Everything else */
514 printk(KERN_ERR "GFS2: wanted %u got %u\n", gl->gl_target, state);
517 spin_unlock(&gl->gl_spin);
521 /* Fast path - we got what we asked for */
522 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
523 gfs2_demote_wake(gl);
524 if (state != LM_ST_UNLOCKED) {
525 if (glops->go_xmote_bh) {
526 spin_unlock(&gl->gl_spin);
527 rv = glops->go_xmote_bh(gl, gh);
528 spin_lock(&gl->gl_spin);
539 clear_bit(GLF_LOCK, &gl->gl_flags);
541 spin_unlock(&gl->gl_spin);
545 * do_xmote - Calls the DLM to change the state of a lock
546 * @gl: The lock state
547 * @gh: The holder (only for promotes)
548 * @target: The target lock state
552 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
553 __releases(&gl->gl_spin)
554 __acquires(&gl->gl_spin)
556 const struct gfs2_glock_operations *glops = gl->gl_ops;
557 struct gfs2_sbd *sdp = gl->gl_sbd;
558 unsigned int lck_flags = gh ? gh->gh_flags : 0;
561 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
563 GLOCK_BUG_ON(gl, gl->gl_state == target);
564 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
565 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
567 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
568 do_error(gl, 0); /* Fail queued try locks */
571 spin_unlock(&gl->gl_spin);
572 if (glops->go_xmote_th)
573 glops->go_xmote_th(gl);
574 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
575 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
576 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
579 if (target != LM_ST_UNLOCKED && (gl->gl_state == LM_ST_SHARED ||
580 gl->gl_state == LM_ST_DEFERRED) &&
581 !(lck_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
582 lck_flags |= LM_FLAG_TRY_1CB;
584 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
586 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
587 GLOCK_BUG_ON(gl, ret);
588 } else { /* lock_nolock */
589 finish_xmote(gl, target);
590 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
594 spin_lock(&gl->gl_spin);
598 * find_first_holder - find the first "holder" gh
602 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
604 struct gfs2_holder *gh;
606 if (!list_empty(&gl->gl_holders)) {
607 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
608 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
615 * run_queue - do all outstanding tasks related to a glock
616 * @gl: The glock in question
617 * @nonblock: True if we must not block in run_queue
621 static void run_queue(struct gfs2_glock *gl, const int nonblock)
622 __releases(&gl->gl_spin)
623 __acquires(&gl->gl_spin)
625 struct gfs2_holder *gh = NULL;
628 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
631 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
633 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
634 gl->gl_demote_state != gl->gl_state) {
635 if (find_first_holder(gl))
639 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
640 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
641 gl->gl_target = gl->gl_demote_state;
643 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
644 gfs2_demote_wake(gl);
645 ret = do_promote(gl);
650 gh = find_first_waiter(gl);
651 gl->gl_target = gh->gh_state;
652 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
653 do_error(gl, 0); /* Fail queued try locks */
655 do_xmote(gl, gh, gl->gl_target);
660 clear_bit(GLF_LOCK, &gl->gl_flags);
661 smp_mb__after_clear_bit();
663 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
664 gfs2_glock_put_nolock(gl);
668 clear_bit(GLF_LOCK, &gl->gl_flags);
669 smp_mb__after_clear_bit();
673 static void delete_work_func(struct work_struct *work)
675 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
676 struct gfs2_sbd *sdp = gl->gl_sbd;
677 struct gfs2_inode *ip;
679 u64 no_addr = gl->gl_name.ln_number;
682 /* Note: Unsafe to dereference ip as we don't hold right refs/locks */
685 inode = gfs2_ilookup(sdp->sd_vfs, no_addr);
687 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
688 if (inode && !IS_ERR(inode)) {
689 d_prune_aliases(inode);
695 static void glock_work_func(struct work_struct *work)
697 unsigned long delay = 0;
698 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
701 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
702 finish_xmote(gl, gl->gl_reply);
705 spin_lock(&gl->gl_spin);
706 if (test_and_clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
707 gl->gl_state != LM_ST_UNLOCKED &&
708 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
709 unsigned long holdtime, now = jiffies;
710 holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
711 if (time_before(now, holdtime))
712 delay = holdtime - now;
713 set_bit(delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE, &gl->gl_flags);
716 spin_unlock(&gl->gl_spin);
718 queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
725 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
726 * @sdp: The GFS2 superblock
727 * @number: the lock number
728 * @glops: The glock_operations to use
729 * @create: If 0, don't create the glock if it doesn't exist
730 * @glp: the glock is returned here
732 * This does not lock a glock, just finds/creates structures for one.
737 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
738 const struct gfs2_glock_operations *glops, int create,
739 struct gfs2_glock **glp)
741 struct super_block *s = sdp->sd_vfs;
742 struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
743 struct gfs2_glock *gl, *tmp;
744 unsigned int hash = gl_hash(sdp, &name);
745 struct address_space *mapping;
747 read_lock(gl_lock_addr(hash));
748 gl = search_bucket(hash, sdp, &name);
749 read_unlock(gl_lock_addr(hash));
757 if (glops->go_flags & GLOF_ASPACE)
758 gl = kmem_cache_alloc(gfs2_glock_aspace_cachep, GFP_KERNEL);
760 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
764 atomic_inc(&sdp->sd_glock_disposal);
767 atomic_set(&gl->gl_ref, 1);
768 gl->gl_state = LM_ST_UNLOCKED;
769 gl->gl_target = LM_ST_UNLOCKED;
770 gl->gl_demote_state = LM_ST_EXCLUSIVE;
773 snprintf(gl->gl_strname, GDLM_STRNAME_BYTES, "%8x%16llx", name.ln_type, (unsigned long long)number);
774 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
775 gl->gl_lksb.sb_lvbptr = gl->gl_lvb;
776 gl->gl_tchange = jiffies;
777 gl->gl_object = NULL;
779 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
780 INIT_WORK(&gl->gl_delete, delete_work_func);
782 mapping = gfs2_glock2aspace(gl);
784 mapping->a_ops = &gfs2_meta_aops;
785 mapping->host = s->s_bdev->bd_inode;
787 mapping_set_gfp_mask(mapping, GFP_NOFS);
788 mapping->assoc_mapping = NULL;
789 mapping->backing_dev_info = s->s_bdi;
790 mapping->writeback_index = 0;
793 write_lock(gl_lock_addr(hash));
794 tmp = search_bucket(hash, sdp, &name);
796 write_unlock(gl_lock_addr(hash));
800 hlist_add_head(&gl->gl_list, &gl_hash_table[hash].hb_list);
801 write_unlock(gl_lock_addr(hash));
810 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
812 * @state: the state we're requesting
813 * @flags: the modifier flags
814 * @gh: the holder structure
818 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
819 struct gfs2_holder *gh)
821 INIT_LIST_HEAD(&gh->gh_list);
823 gh->gh_ip = (unsigned long)__builtin_return_address(0);
824 gh->gh_owner_pid = get_pid(task_pid(current));
825 gh->gh_state = state;
826 gh->gh_flags = flags;
833 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
834 * @state: the state we're requesting
835 * @flags: the modifier flags
836 * @gh: the holder structure
838 * Don't mess with the glock.
842 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
844 gh->gh_state = state;
845 gh->gh_flags = flags;
847 gh->gh_ip = (unsigned long)__builtin_return_address(0);
848 if (gh->gh_owner_pid)
849 put_pid(gh->gh_owner_pid);
850 gh->gh_owner_pid = get_pid(task_pid(current));
854 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
855 * @gh: the holder structure
859 void gfs2_holder_uninit(struct gfs2_holder *gh)
861 put_pid(gh->gh_owner_pid);
862 gfs2_glock_put(gh->gh_gl);
868 * gfs2_glock_holder_wait
871 * This function and gfs2_glock_demote_wait both show up in the WCHAN
872 * field. Thus I've separated these otherwise identical functions in
873 * order to be more informative to the user.
876 static int gfs2_glock_holder_wait(void *word)
882 static int gfs2_glock_demote_wait(void *word)
888 static void wait_on_holder(struct gfs2_holder *gh)
891 wait_on_bit(&gh->gh_iflags, HIF_WAIT, gfs2_glock_holder_wait, TASK_UNINTERRUPTIBLE);
894 static void wait_on_demote(struct gfs2_glock *gl)
897 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, gfs2_glock_demote_wait, TASK_UNINTERRUPTIBLE);
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,
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);
924 trace_gfs2_demote_rq(gl);
928 * gfs2_glock_wait - wait on a glock acquisition
929 * @gh: the glock holder
931 * Returns: 0 on success
934 int gfs2_glock_wait(struct gfs2_holder *gh)
940 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
942 struct va_format vaf;
948 struct gfs2_glock_iter *gi = seq->private;
949 vsprintf(gi->string, fmt, args);
950 seq_printf(seq, gi->string);
955 printk(KERN_ERR " %pV", &vaf);
962 * add_to_queue - Add a holder to the wait queue (but look for recursion)
963 * @gh: the holder structure to add
965 * Eventually we should move the recursive locking trap to a
966 * debugging option or something like that. This is the fast
967 * path and needs to have the minimum number of distractions.
971 static inline void add_to_queue(struct gfs2_holder *gh)
972 __releases(&gl->gl_spin)
973 __acquires(&gl->gl_spin)
975 struct gfs2_glock *gl = gh->gh_gl;
976 struct gfs2_sbd *sdp = gl->gl_sbd;
977 struct list_head *insert_pt = NULL;
978 struct gfs2_holder *gh2;
981 BUG_ON(gh->gh_owner_pid == NULL);
982 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
985 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
986 if (test_bit(GLF_LOCK, &gl->gl_flags))
988 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
992 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
993 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
994 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
997 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) &&
998 !may_grant(gl, gh)) {
1000 gh->gh_error = GLR_TRYFAILED;
1001 gfs2_holder_wake(gh);
1004 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1006 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1007 insert_pt = &gh2->gh_list;
1009 set_bit(GLF_QUEUED, &gl->gl_flags);
1010 if (likely(insert_pt == NULL)) {
1011 list_add_tail(&gh->gh_list, &gl->gl_holders);
1012 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1016 trace_gfs2_glock_queue(gh, 1);
1017 list_add_tail(&gh->gh_list, insert_pt);
1019 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
1020 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1021 spin_unlock(&gl->gl_spin);
1022 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1023 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1024 spin_lock(&gl->gl_spin);
1029 print_symbol(KERN_ERR "original: %s\n", gh2->gh_ip);
1030 printk(KERN_ERR "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1031 printk(KERN_ERR "lock type: %d req lock state : %d\n",
1032 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1033 print_symbol(KERN_ERR "new: %s\n", gh->gh_ip);
1034 printk(KERN_ERR "pid: %d\n", pid_nr(gh->gh_owner_pid));
1035 printk(KERN_ERR "lock type: %d req lock state : %d\n",
1036 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1037 __dump_glock(NULL, gl);
1042 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1043 * @gh: the holder structure
1045 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1047 * Returns: 0, GLR_TRYFAILED, or errno on failure
1050 int gfs2_glock_nq(struct gfs2_holder *gh)
1052 struct gfs2_glock *gl = gh->gh_gl;
1053 struct gfs2_sbd *sdp = gl->gl_sbd;
1056 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1059 spin_lock(&gl->gl_spin);
1061 if ((LM_FLAG_NOEXP & gh->gh_flags) &&
1062 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1063 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1065 spin_unlock(&gl->gl_spin);
1067 if (!(gh->gh_flags & GL_ASYNC))
1068 error = gfs2_glock_wait(gh);
1074 * gfs2_glock_poll - poll to see if an async request has been completed
1077 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1080 int gfs2_glock_poll(struct gfs2_holder *gh)
1082 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1086 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1087 * @gh: the glock holder
1091 void gfs2_glock_dq(struct gfs2_holder *gh)
1093 struct gfs2_glock *gl = gh->gh_gl;
1094 const struct gfs2_glock_operations *glops = gl->gl_ops;
1098 spin_lock(&gl->gl_spin);
1099 if (gh->gh_flags & GL_NOCACHE)
1100 handle_callback(gl, LM_ST_UNLOCKED, 0);
1102 list_del_init(&gh->gh_list);
1103 if (find_first_holder(gl) == NULL) {
1104 if (glops->go_unlock) {
1105 GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1106 spin_unlock(&gl->gl_spin);
1107 glops->go_unlock(gh);
1108 spin_lock(&gl->gl_spin);
1109 clear_bit(GLF_LOCK, &gl->gl_flags);
1111 if (list_empty(&gl->gl_holders) &&
1112 !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1113 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1116 trace_gfs2_glock_queue(gh, 0);
1117 spin_unlock(&gl->gl_spin);
1118 if (likely(fast_path))
1121 gfs2_glock_hold(gl);
1122 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1123 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1124 delay = gl->gl_ops->go_min_hold_time;
1125 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1129 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1131 struct gfs2_glock *gl = gh->gh_gl;
1137 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1138 * @gh: the holder structure
1142 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1145 gfs2_holder_uninit(gh);
1149 * gfs2_glock_nq_num - acquire a glock based on lock number
1150 * @sdp: the filesystem
1151 * @number: the lock number
1152 * @glops: the glock operations for the type of glock
1153 * @state: the state to acquire the glock in
1154 * @flags: modifier flags for the aquisition
1155 * @gh: the struct gfs2_holder
1160 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1161 const struct gfs2_glock_operations *glops,
1162 unsigned int state, int flags, struct gfs2_holder *gh)
1164 struct gfs2_glock *gl;
1167 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1169 error = gfs2_glock_nq_init(gl, state, flags, gh);
1177 * glock_compare - Compare two struct gfs2_glock structures for sorting
1178 * @arg_a: the first structure
1179 * @arg_b: the second structure
1183 static int glock_compare(const void *arg_a, const void *arg_b)
1185 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1186 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1187 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1188 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1190 if (a->ln_number > b->ln_number)
1192 if (a->ln_number < b->ln_number)
1194 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1199 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1200 * @num_gh: the number of structures
1201 * @ghs: an array of struct gfs2_holder structures
1203 * Returns: 0 on success (all glocks acquired),
1204 * errno on failure (no glocks acquired)
1207 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1208 struct gfs2_holder **p)
1213 for (x = 0; x < num_gh; x++)
1216 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1218 for (x = 0; x < num_gh; x++) {
1219 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1221 error = gfs2_glock_nq(p[x]);
1224 gfs2_glock_dq(p[x]);
1233 * gfs2_glock_nq_m - acquire multiple glocks
1234 * @num_gh: the number of structures
1235 * @ghs: an array of struct gfs2_holder structures
1238 * Returns: 0 on success (all glocks acquired),
1239 * errno on failure (no glocks acquired)
1242 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1244 struct gfs2_holder *tmp[4];
1245 struct gfs2_holder **pph = tmp;
1252 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1253 return gfs2_glock_nq(ghs);
1257 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1262 error = nq_m_sync(num_gh, ghs, pph);
1271 * gfs2_glock_dq_m - release multiple glocks
1272 * @num_gh: the number of structures
1273 * @ghs: an array of struct gfs2_holder structures
1277 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1281 for (x = 0; x < num_gh; x++)
1282 gfs2_glock_dq(&ghs[x]);
1286 * gfs2_glock_dq_uninit_m - release multiple glocks
1287 * @num_gh: the number of structures
1288 * @ghs: an array of struct gfs2_holder structures
1292 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1296 for (x = 0; x < num_gh; x++)
1297 gfs2_glock_dq_uninit(&ghs[x]);
1300 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1302 unsigned long delay = 0;
1303 unsigned long holdtime;
1304 unsigned long now = jiffies;
1306 gfs2_glock_hold(gl);
1307 holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
1308 if (test_bit(GLF_QUEUED, &gl->gl_flags)) {
1309 if (time_before(now, holdtime))
1310 delay = holdtime - now;
1311 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1312 delay = gl->gl_ops->go_min_hold_time;
1315 spin_lock(&gl->gl_spin);
1316 handle_callback(gl, state, delay);
1317 spin_unlock(&gl->gl_spin);
1318 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1323 * gfs2_should_freeze - Figure out if glock should be frozen
1324 * @gl: The glock in question
1326 * Glocks are not frozen if (a) the result of the dlm operation is
1327 * an error, (b) the locking operation was an unlock operation or
1328 * (c) if there is a "noexp" flagged request anywhere in the queue
1330 * Returns: 1 if freezing should occur, 0 otherwise
1333 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1335 const struct gfs2_holder *gh;
1337 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1339 if (gl->gl_target == LM_ST_UNLOCKED)
1342 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1343 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1345 if (LM_FLAG_NOEXP & gh->gh_flags)
1353 * gfs2_glock_complete - Callback used by locking
1354 * @gl: Pointer to the glock
1355 * @ret: The return value from the dlm
1357 * The gl_reply field is under the gl_spin lock so that it is ok
1358 * to use a bitfield shared with other glock state fields.
1361 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1363 struct lm_lockstruct *ls = &gl->gl_sbd->sd_lockstruct;
1365 spin_lock(&gl->gl_spin);
1368 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_flags))) {
1369 if (gfs2_should_freeze(gl)) {
1370 set_bit(GLF_FROZEN, &gl->gl_flags);
1371 spin_unlock(&gl->gl_spin);
1376 spin_unlock(&gl->gl_spin);
1377 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1379 gfs2_glock_hold(gl);
1380 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1385 static int gfs2_shrink_glock_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
1387 struct gfs2_glock *gl;
1395 if (!(gfp_mask & __GFP_FS))
1398 spin_lock(&lru_lock);
1399 while(nr && !list_empty(&lru_list)) {
1400 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1401 list_del_init(&gl->gl_lru);
1402 atomic_dec(&lru_count);
1404 /* Test for being demotable */
1405 if (!test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1406 gfs2_glock_hold(gl);
1407 spin_unlock(&lru_lock);
1408 spin_lock(&gl->gl_spin);
1409 may_demote = demote_ok(gl);
1411 handle_callback(gl, LM_ST_UNLOCKED, 0);
1414 clear_bit(GLF_LOCK, &gl->gl_flags);
1415 smp_mb__after_clear_bit();
1416 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1417 gfs2_glock_put_nolock(gl);
1418 spin_unlock(&gl->gl_spin);
1419 spin_lock(&lru_lock);
1423 list_add(&gl->gl_lru, &skipped);
1425 list_splice(&skipped, &lru_list);
1426 atomic_add(nr_skipped, &lru_count);
1427 spin_unlock(&lru_lock);
1429 return (atomic_read(&lru_count) / 100) * sysctl_vfs_cache_pressure;
1432 static struct shrinker glock_shrinker = {
1433 .shrink = gfs2_shrink_glock_memory,
1434 .seeks = DEFAULT_SEEKS,
1438 * examine_bucket - Call a function for glock in a hash bucket
1439 * @examiner: the function
1440 * @sdp: the filesystem
1441 * @bucket: the bucket
1443 * Returns: 1 if the bucket has entries
1446 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1449 struct gfs2_glock *gl, *prev = NULL;
1450 int has_entries = 0;
1451 struct hlist_head *head = &gl_hash_table[hash].hb_list;
1453 read_lock(gl_lock_addr(hash));
1454 /* Can't use hlist_for_each_entry - don't want prefetch here */
1455 if (hlist_empty(head))
1457 gl = list_entry(head->first, struct gfs2_glock, gl_list);
1459 if (!sdp || gl->gl_sbd == sdp) {
1460 gfs2_glock_hold(gl);
1461 read_unlock(gl_lock_addr(hash));
1463 gfs2_glock_put(prev);
1467 read_lock(gl_lock_addr(hash));
1469 if (gl->gl_list.next == NULL)
1471 gl = list_entry(gl->gl_list.next, struct gfs2_glock, gl_list);
1474 read_unlock(gl_lock_addr(hash));
1476 gfs2_glock_put(prev);
1483 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1484 * @gl: The glock to thaw
1486 * N.B. When we freeze a glock, we leave a ref to the glock outstanding,
1487 * so this has to result in the ref count being dropped by one.
1490 static void thaw_glock(struct gfs2_glock *gl)
1492 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1494 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1495 gfs2_glock_hold(gl);
1496 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1501 * clear_glock - look at a glock and see if we can free it from glock cache
1502 * @gl: the glock to look at
1506 static void clear_glock(struct gfs2_glock *gl)
1508 spin_lock(&lru_lock);
1509 if (!list_empty(&gl->gl_lru)) {
1510 list_del_init(&gl->gl_lru);
1511 atomic_dec(&lru_count);
1513 spin_unlock(&lru_lock);
1515 spin_lock(&gl->gl_spin);
1516 if (gl->gl_state != LM_ST_UNLOCKED)
1517 handle_callback(gl, LM_ST_UNLOCKED, 0);
1518 spin_unlock(&gl->gl_spin);
1519 gfs2_glock_hold(gl);
1520 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1525 * gfs2_glock_thaw - Thaw any frozen glocks
1526 * @sdp: The super block
1530 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1534 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1535 examine_bucket(thaw_glock, sdp, x);
1539 * gfs2_gl_hash_clear - Empty out the glock hash table
1540 * @sdp: the filesystem
1541 * @wait: wait until it's all gone
1543 * Called when unmounting the filesystem.
1546 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1550 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1551 examine_bucket(clear_glock, sdp, x);
1552 flush_workqueue(glock_workqueue);
1553 wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
1554 gfs2_dump_lockstate(sdp);
1557 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1559 struct gfs2_glock *gl = ip->i_gl;
1562 ret = gfs2_truncatei_resume(ip);
1563 gfs2_assert_withdraw(gl->gl_sbd, ret == 0);
1565 spin_lock(&gl->gl_spin);
1566 clear_bit(GLF_LOCK, &gl->gl_flags);
1568 spin_unlock(&gl->gl_spin);
1571 static const char *state2str(unsigned state)
1574 case LM_ST_UNLOCKED:
1578 case LM_ST_DEFERRED:
1580 case LM_ST_EXCLUSIVE:
1586 static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1589 if (flags & LM_FLAG_TRY)
1591 if (flags & LM_FLAG_TRY_1CB)
1593 if (flags & LM_FLAG_NOEXP)
1595 if (flags & LM_FLAG_ANY)
1597 if (flags & LM_FLAG_PRIORITY)
1599 if (flags & GL_ASYNC)
1601 if (flags & GL_EXACT)
1603 if (flags & GL_NOCACHE)
1605 if (test_bit(HIF_HOLDER, &iflags))
1607 if (test_bit(HIF_WAIT, &iflags))
1609 if (test_bit(HIF_FIRST, &iflags))
1616 * dump_holder - print information about a glock holder
1617 * @seq: the seq_file struct
1618 * @gh: the glock holder
1620 * Returns: 0 on success, -ENOBUFS when we run out of space
1623 static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1625 struct task_struct *gh_owner = NULL;
1628 if (gh->gh_owner_pid)
1629 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1630 gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1631 state2str(gh->gh_state),
1632 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1634 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1635 gh_owner ? gh_owner->comm : "(ended)",
1640 static const char *gflags2str(char *buf, const unsigned long *gflags)
1643 if (test_bit(GLF_LOCK, gflags))
1645 if (test_bit(GLF_DEMOTE, gflags))
1647 if (test_bit(GLF_PENDING_DEMOTE, gflags))
1649 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1651 if (test_bit(GLF_DIRTY, gflags))
1653 if (test_bit(GLF_LFLUSH, gflags))
1655 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1657 if (test_bit(GLF_REPLY_PENDING, gflags))
1659 if (test_bit(GLF_INITIAL, gflags))
1661 if (test_bit(GLF_FROZEN, gflags))
1663 if (test_bit(GLF_QUEUED, gflags))
1670 * __dump_glock - print information about a glock
1671 * @seq: The seq_file struct
1674 * The file format is as follows:
1675 * One line per object, capital letters are used to indicate objects
1676 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1677 * other objects are indented by a single space and follow the glock to
1678 * which they are related. Fields are indicated by lower case letters
1679 * followed by a colon and the field value, except for strings which are in
1680 * [] so that its possible to see if they are composed of spaces for
1681 * example. The field's are n = number (id of the object), f = flags,
1682 * t = type, s = state, r = refcount, e = error, p = pid.
1684 * Returns: 0 on success, -ENOBUFS when we run out of space
1687 static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1689 const struct gfs2_glock_operations *glops = gl->gl_ops;
1690 unsigned long long dtime;
1691 const struct gfs2_holder *gh;
1692 char gflags_buf[32];
1695 dtime = jiffies - gl->gl_demote_time;
1696 dtime *= 1000000/HZ; /* demote time in uSec */
1697 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1699 gfs2_print_dbg(seq, "G: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d r:%d\n",
1700 state2str(gl->gl_state),
1701 gl->gl_name.ln_type,
1702 (unsigned long long)gl->gl_name.ln_number,
1703 gflags2str(gflags_buf, &gl->gl_flags),
1704 state2str(gl->gl_target),
1705 state2str(gl->gl_demote_state), dtime,
1706 atomic_read(&gl->gl_ail_count),
1707 atomic_read(&gl->gl_ref));
1709 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1710 error = dump_holder(seq, gh);
1714 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1715 error = glops->go_dump(seq, gl);
1720 static int dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1723 spin_lock(&gl->gl_spin);
1724 ret = __dump_glock(seq, gl);
1725 spin_unlock(&gl->gl_spin);
1730 * gfs2_dump_lockstate - print out the current lockstate
1731 * @sdp: the filesystem
1732 * @ub: the buffer to copy the information into
1734 * If @ub is NULL, dump the lockstate to the console.
1738 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
1740 struct gfs2_glock *gl;
1741 struct hlist_node *h;
1745 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1747 read_lock(gl_lock_addr(x));
1749 hlist_for_each_entry(gl, h, &gl_hash_table[x].hb_list, gl_list) {
1750 if (gl->gl_sbd != sdp)
1753 error = dump_glock(NULL, gl);
1758 read_unlock(gl_lock_addr(x));
1769 int __init gfs2_glock_init(void)
1772 for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
1773 INIT_HLIST_HEAD(&gl_hash_table[i].hb_list);
1775 #ifdef GL_HASH_LOCK_SZ
1776 for(i = 0; i < GL_HASH_LOCK_SZ; i++) {
1777 rwlock_init(&gl_hash_locks[i]);
1781 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1782 WQ_HIGHPRI | WQ_FREEZEABLE, 0);
1783 if (IS_ERR(glock_workqueue))
1784 return PTR_ERR(glock_workqueue);
1785 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1786 WQ_MEM_RECLAIM | WQ_FREEZEABLE,
1788 if (IS_ERR(gfs2_delete_workqueue)) {
1789 destroy_workqueue(glock_workqueue);
1790 return PTR_ERR(gfs2_delete_workqueue);
1793 register_shrinker(&glock_shrinker);
1798 void gfs2_glock_exit(void)
1800 unregister_shrinker(&glock_shrinker);
1801 destroy_workqueue(glock_workqueue);
1802 destroy_workqueue(gfs2_delete_workqueue);
1805 static int gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1807 struct gfs2_glock *gl;
1810 read_lock(gl_lock_addr(gi->hash));
1813 gi->gl = hlist_entry(gl->gl_list.next,
1814 struct gfs2_glock, gl_list);
1816 gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
1817 struct gfs2_glock, gl_list);
1820 gfs2_glock_hold(gi->gl);
1821 read_unlock(gl_lock_addr(gi->hash));
1824 while (gi->gl == NULL) {
1826 if (gi->hash >= GFS2_GL_HASH_SIZE)
1828 read_lock(gl_lock_addr(gi->hash));
1829 gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
1830 struct gfs2_glock, gl_list);
1832 gfs2_glock_hold(gi->gl);
1833 read_unlock(gl_lock_addr(gi->hash));
1836 if (gi->sdp != gi->gl->gl_sbd)
1842 static void gfs2_glock_iter_free(struct gfs2_glock_iter *gi)
1845 gfs2_glock_put(gi->gl);
1849 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1851 struct gfs2_glock_iter *gi = seq->private;
1857 if (gfs2_glock_iter_next(gi)) {
1858 gfs2_glock_iter_free(gi);
1866 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1869 struct gfs2_glock_iter *gi = seq->private;
1873 if (gfs2_glock_iter_next(gi)) {
1874 gfs2_glock_iter_free(gi);
1881 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1883 struct gfs2_glock_iter *gi = seq->private;
1884 gfs2_glock_iter_free(gi);
1887 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1889 return dump_glock(seq, iter_ptr);
1892 static const struct seq_operations gfs2_glock_seq_ops = {
1893 .start = gfs2_glock_seq_start,
1894 .next = gfs2_glock_seq_next,
1895 .stop = gfs2_glock_seq_stop,
1896 .show = gfs2_glock_seq_show,
1899 static int gfs2_debugfs_open(struct inode *inode, struct file *file)
1901 int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1902 sizeof(struct gfs2_glock_iter));
1904 struct seq_file *seq = file->private_data;
1905 struct gfs2_glock_iter *gi = seq->private;
1906 gi->sdp = inode->i_private;
1911 static const struct file_operations gfs2_debug_fops = {
1912 .owner = THIS_MODULE,
1913 .open = gfs2_debugfs_open,
1915 .llseek = seq_lseek,
1916 .release = seq_release_private,
1919 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
1921 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
1922 if (!sdp->debugfs_dir)
1924 sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
1926 sdp->debugfs_dir, sdp,
1928 if (!sdp->debugfs_dentry_glocks)
1934 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
1936 if (sdp && sdp->debugfs_dir) {
1937 if (sdp->debugfs_dentry_glocks) {
1938 debugfs_remove(sdp->debugfs_dentry_glocks);
1939 sdp->debugfs_dentry_glocks = NULL;
1941 debugfs_remove(sdp->debugfs_dir);
1942 sdp->debugfs_dir = NULL;
1946 int gfs2_register_debugfs(void)
1948 gfs2_root = debugfs_create_dir("gfs2", NULL);
1949 return gfs2_root ? 0 : -ENOMEM;
1952 void gfs2_unregister_debugfs(void)
1954 debugfs_remove(gfs2_root);