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 <linux/rwsem.h>
23 #include <asm/uaccess.h>
24 #include <linux/seq_file.h>
25 #include <linux/debugfs.h>
26 #include <linux/kthread.h>
27 #include <linux/freezer.h>
28 #include <linux/workqueue.h>
29 #include <linux/jiffies.h>
42 #define CREATE_TRACE_POINTS
43 #include "trace_gfs2.h"
45 struct gfs2_gl_hash_bucket {
46 struct hlist_head hb_list;
49 struct gfs2_glock_iter {
50 int hash; /* hash bucket index */
51 struct gfs2_sbd *sdp; /* incore superblock */
52 struct gfs2_glock *gl; /* current glock struct */
53 char string[512]; /* scratch space */
56 typedef void (*glock_examiner) (struct gfs2_glock * gl);
58 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp);
59 static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl);
60 #define GLOCK_BUG_ON(gl,x) do { if (unlikely(x)) { __dump_glock(NULL, gl); BUG(); } } while(0)
61 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
63 static DECLARE_RWSEM(gfs2_umount_flush_sem);
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 gfs2_gl_hash_bucket gl_hash_table[GFS2_GL_HASH_SIZE];
76 static struct dentry *gfs2_root;
79 * Despite what you might think, the numbers below are not arbitrary :-)
80 * They are taken from the ipv4 routing hash code, which is well tested
81 * and thus should be nearly optimal. Later on we might tweek the numbers
82 * but for now this should be fine.
84 * The reason for putting the locks in a separate array from the list heads
85 * is that we can have fewer locks than list heads and save memory. We use
86 * the same hash function for both, but with a different hash mask.
88 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
89 defined(CONFIG_PROVE_LOCKING)
92 # define GL_HASH_LOCK_SZ 256
95 # define GL_HASH_LOCK_SZ 4096
97 # define GL_HASH_LOCK_SZ 2048
99 # define GL_HASH_LOCK_SZ 1024
101 # define GL_HASH_LOCK_SZ 512
103 # define GL_HASH_LOCK_SZ 256
107 /* We never want more locks than chains */
108 #if GFS2_GL_HASH_SIZE < GL_HASH_LOCK_SZ
109 # undef GL_HASH_LOCK_SZ
110 # define GL_HASH_LOCK_SZ GFS2_GL_HASH_SIZE
113 static rwlock_t gl_hash_locks[GL_HASH_LOCK_SZ];
115 static inline rwlock_t *gl_lock_addr(unsigned int x)
117 return &gl_hash_locks[x & (GL_HASH_LOCK_SZ-1)];
119 #else /* not SMP, so no spinlocks required */
120 static inline rwlock_t *gl_lock_addr(unsigned int x)
127 * gl_hash() - Turn glock number into hash bucket number
128 * @lock: The glock number
130 * Returns: The number of the corresponding hash bucket
133 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
134 const struct lm_lockname *name)
138 h = jhash(&name->ln_number, sizeof(u64), 0);
139 h = jhash(&name->ln_type, sizeof(unsigned int), h);
140 h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
141 h &= GFS2_GL_HASH_MASK;
147 * glock_free() - Perform a few checks and then release struct gfs2_glock
148 * @gl: The glock to release
150 * Also calls lock module to release its internal structure for this glock.
154 static void glock_free(struct gfs2_glock *gl)
156 struct gfs2_sbd *sdp = gl->gl_sbd;
157 struct inode *aspace = gl->gl_aspace;
160 gfs2_aspace_put(aspace);
161 trace_gfs2_glock_put(gl);
162 sdp->sd_lockstruct.ls_ops->lm_put_lock(gfs2_glock_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_promote - promote as many requests as possible on the current queue
334 * Returns: 1 if there is a blocked holder at the head of the list, or 2
335 * if a type specific operation is underway.
338 static int do_promote(struct gfs2_glock *gl)
339 __releases(&gl->gl_spin)
340 __acquires(&gl->gl_spin)
342 const struct gfs2_glock_operations *glops = gl->gl_ops;
343 struct gfs2_holder *gh, *tmp;
347 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
348 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
350 if (may_grant(gl, gh)) {
351 if (gh->gh_list.prev == &gl->gl_holders &&
353 spin_unlock(&gl->gl_spin);
354 /* FIXME: eliminate this eventually */
355 ret = glops->go_lock(gh);
356 spin_lock(&gl->gl_spin);
361 list_del_init(&gh->gh_list);
362 trace_gfs2_glock_queue(gh, 0);
363 gfs2_holder_wake(gh);
366 set_bit(HIF_HOLDER, &gh->gh_iflags);
367 trace_gfs2_promote(gh, 1);
368 gfs2_holder_wake(gh);
371 set_bit(HIF_HOLDER, &gh->gh_iflags);
372 trace_gfs2_promote(gh, 0);
373 gfs2_holder_wake(gh);
376 if (gh->gh_list.prev == &gl->gl_holders)
384 * do_error - Something unexpected has happened during a lock request
388 static inline void do_error(struct gfs2_glock *gl, const int ret)
390 struct gfs2_holder *gh, *tmp;
392 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
393 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
395 if (ret & LM_OUT_ERROR)
397 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
398 gh->gh_error = GLR_TRYFAILED;
401 list_del_init(&gh->gh_list);
402 trace_gfs2_glock_queue(gh, 0);
403 gfs2_holder_wake(gh);
408 * find_first_waiter - find the first gh that's waiting for the glock
412 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
414 struct gfs2_holder *gh;
416 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
417 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
424 * state_change - record that the glock is now in a different state
426 * @new_state the new state
430 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
434 held1 = (gl->gl_state != LM_ST_UNLOCKED);
435 held2 = (new_state != LM_ST_UNLOCKED);
437 if (held1 != held2) {
441 gfs2_glock_put_nolock(gl);
444 gl->gl_state = new_state;
445 gl->gl_tchange = jiffies;
448 static void gfs2_demote_wake(struct gfs2_glock *gl)
450 gl->gl_demote_state = LM_ST_EXCLUSIVE;
451 clear_bit(GLF_DEMOTE, &gl->gl_flags);
452 smp_mb__after_clear_bit();
453 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
457 * finish_xmote - The DLM has replied to one of our lock requests
459 * @ret: The status from the DLM
463 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
465 const struct gfs2_glock_operations *glops = gl->gl_ops;
466 struct gfs2_holder *gh;
467 unsigned state = ret & LM_OUT_ST_MASK;
470 spin_lock(&gl->gl_spin);
471 trace_gfs2_glock_state_change(gl, state);
472 state_change(gl, state);
473 gh = find_first_waiter(gl);
475 /* Demote to UN request arrived during demote to SH or DF */
476 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
477 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
478 gl->gl_target = LM_ST_UNLOCKED;
480 /* Check for state != intended state */
481 if (unlikely(state != gl->gl_target)) {
482 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
483 /* move to back of queue and try next entry */
484 if (ret & LM_OUT_CANCELED) {
485 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
486 list_move_tail(&gh->gh_list, &gl->gl_holders);
487 gh = find_first_waiter(gl);
488 gl->gl_target = gh->gh_state;
491 /* Some error or failed "try lock" - report it */
492 if ((ret & LM_OUT_ERROR) ||
493 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
494 gl->gl_target = gl->gl_state;
500 /* Unlocked due to conversion deadlock, try again */
503 do_xmote(gl, gh, gl->gl_target);
505 /* Conversion fails, unlock and try again */
508 do_xmote(gl, gh, LM_ST_UNLOCKED);
510 default: /* Everything else */
511 printk(KERN_ERR "GFS2: wanted %u got %u\n", gl->gl_target, state);
514 spin_unlock(&gl->gl_spin);
518 /* Fast path - we got what we asked for */
519 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
520 gfs2_demote_wake(gl);
521 if (state != LM_ST_UNLOCKED) {
522 if (glops->go_xmote_bh) {
523 spin_unlock(&gl->gl_spin);
524 rv = glops->go_xmote_bh(gl, gh);
525 spin_lock(&gl->gl_spin);
536 clear_bit(GLF_LOCK, &gl->gl_flags);
538 spin_unlock(&gl->gl_spin);
541 static unsigned int gfs2_lm_lock(struct gfs2_sbd *sdp, void *lock,
542 unsigned int req_state,
545 int ret = LM_OUT_ERROR;
547 if (!sdp->sd_lockstruct.ls_ops->lm_lock)
548 return req_state == LM_ST_UNLOCKED ? 0 : req_state;
550 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
551 ret = sdp->sd_lockstruct.ls_ops->lm_lock(lock,
557 * do_xmote - Calls the DLM to change the state of a lock
558 * @gl: The lock state
559 * @gh: The holder (only for promotes)
560 * @target: The target lock state
564 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
565 __releases(&gl->gl_spin)
566 __acquires(&gl->gl_spin)
568 const struct gfs2_glock_operations *glops = gl->gl_ops;
569 struct gfs2_sbd *sdp = gl->gl_sbd;
570 unsigned int lck_flags = gh ? gh->gh_flags : 0;
573 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
575 BUG_ON(gl->gl_state == target);
576 BUG_ON(gl->gl_state == gl->gl_target);
577 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
579 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
580 do_error(gl, 0); /* Fail queued try locks */
582 spin_unlock(&gl->gl_spin);
583 if (glops->go_xmote_th)
584 glops->go_xmote_th(gl);
585 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
586 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
587 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
590 if (target != LM_ST_UNLOCKED && (gl->gl_state == LM_ST_SHARED ||
591 gl->gl_state == LM_ST_DEFERRED) &&
592 !(lck_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
593 lck_flags |= LM_FLAG_TRY_1CB;
594 ret = gfs2_lm_lock(sdp, gl, target, lck_flags);
596 if (!(ret & LM_OUT_ASYNC)) {
597 finish_xmote(gl, ret);
598 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
601 GLOCK_BUG_ON(gl, ret != LM_OUT_ASYNC);
603 spin_lock(&gl->gl_spin);
607 * find_first_holder - find the first "holder" gh
611 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
613 struct gfs2_holder *gh;
615 if (!list_empty(&gl->gl_holders)) {
616 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
617 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
624 * run_queue - do all outstanding tasks related to a glock
625 * @gl: The glock in question
626 * @nonblock: True if we must not block in run_queue
630 static void run_queue(struct gfs2_glock *gl, const int nonblock)
631 __releases(&gl->gl_spin)
632 __acquires(&gl->gl_spin)
634 struct gfs2_holder *gh = NULL;
637 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
640 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
642 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
643 gl->gl_demote_state != gl->gl_state) {
644 if (find_first_holder(gl))
648 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
649 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
650 gl->gl_target = gl->gl_demote_state;
652 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
653 gfs2_demote_wake(gl);
654 ret = do_promote(gl);
659 gh = find_first_waiter(gl);
660 gl->gl_target = gh->gh_state;
661 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
662 do_error(gl, 0); /* Fail queued try locks */
664 do_xmote(gl, gh, gl->gl_target);
669 clear_bit(GLF_LOCK, &gl->gl_flags);
670 smp_mb__after_clear_bit();
672 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
673 gfs2_glock_put_nolock(gl);
677 clear_bit(GLF_LOCK, &gl->gl_flags);
678 smp_mb__after_clear_bit();
682 static void delete_work_func(struct work_struct *work)
684 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
685 struct gfs2_sbd *sdp = gl->gl_sbd;
686 struct gfs2_inode *ip = NULL;
690 spin_lock(&gl->gl_spin);
691 ip = (struct gfs2_inode *)gl->gl_object;
693 no_addr = ip->i_no_addr;
694 spin_unlock(&gl->gl_spin);
696 inode = gfs2_ilookup(sdp->sd_vfs, no_addr);
698 d_prune_aliases(inode);
705 static void glock_work_func(struct work_struct *work)
707 unsigned long delay = 0;
708 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
711 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
712 finish_xmote(gl, gl->gl_reply);
715 down_read(&gfs2_umount_flush_sem);
716 spin_lock(&gl->gl_spin);
717 if (test_and_clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
718 gl->gl_state != LM_ST_UNLOCKED &&
719 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
720 unsigned long holdtime, now = jiffies;
721 holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
722 if (time_before(now, holdtime))
723 delay = holdtime - now;
724 set_bit(delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE, &gl->gl_flags);
727 spin_unlock(&gl->gl_spin);
728 up_read(&gfs2_umount_flush_sem);
730 queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
737 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
738 * @sdp: The GFS2 superblock
739 * @number: the lock number
740 * @glops: The glock_operations to use
741 * @create: If 0, don't create the glock if it doesn't exist
742 * @glp: the glock is returned here
744 * This does not lock a glock, just finds/creates structures for one.
749 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
750 const struct gfs2_glock_operations *glops, int create,
751 struct gfs2_glock **glp)
753 struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
754 struct gfs2_glock *gl, *tmp;
755 unsigned int hash = gl_hash(sdp, &name);
758 read_lock(gl_lock_addr(hash));
759 gl = search_bucket(hash, sdp, &name);
760 read_unlock(gl_lock_addr(hash));
768 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
774 atomic_set(&gl->gl_ref, 1);
775 gl->gl_state = LM_ST_UNLOCKED;
776 gl->gl_target = LM_ST_UNLOCKED;
777 gl->gl_demote_state = LM_ST_EXCLUSIVE;
780 snprintf(gl->gl_strname, GDLM_STRNAME_BYTES, "%8x%16llx", name.ln_type, (unsigned long long)number);
781 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
782 gl->gl_lksb.sb_lvbptr = gl->gl_lvb;
783 gl->gl_tchange = jiffies;
784 gl->gl_object = NULL;
786 gl->gl_aspace = NULL;
787 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
788 INIT_WORK(&gl->gl_delete, delete_work_func);
790 /* If this glock protects actual on-disk data or metadata blocks,
791 create a VFS inode to manage the pages/buffers holding them. */
792 if (glops == &gfs2_inode_glops || glops == &gfs2_rgrp_glops) {
793 gl->gl_aspace = gfs2_aspace_get(sdp);
794 if (!gl->gl_aspace) {
800 write_lock(gl_lock_addr(hash));
801 tmp = search_bucket(hash, sdp, &name);
803 write_unlock(gl_lock_addr(hash));
807 hlist_add_head(&gl->gl_list, &gl_hash_table[hash].hb_list);
808 write_unlock(gl_lock_addr(hash));
816 kmem_cache_free(gfs2_glock_cachep, gl);
821 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
823 * @state: the state we're requesting
824 * @flags: the modifier flags
825 * @gh: the holder structure
829 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
830 struct gfs2_holder *gh)
832 INIT_LIST_HEAD(&gh->gh_list);
834 gh->gh_ip = (unsigned long)__builtin_return_address(0);
835 gh->gh_owner_pid = get_pid(task_pid(current));
836 gh->gh_state = state;
837 gh->gh_flags = flags;
844 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
845 * @state: the state we're requesting
846 * @flags: the modifier flags
847 * @gh: the holder structure
849 * Don't mess with the glock.
853 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
855 gh->gh_state = state;
856 gh->gh_flags = flags;
858 gh->gh_ip = (unsigned long)__builtin_return_address(0);
862 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
863 * @gh: the holder structure
867 void gfs2_holder_uninit(struct gfs2_holder *gh)
869 put_pid(gh->gh_owner_pid);
870 gfs2_glock_put(gh->gh_gl);
876 * gfs2_glock_holder_wait
879 * This function and gfs2_glock_demote_wait both show up in the WCHAN
880 * field. Thus I've separated these otherwise identical functions in
881 * order to be more informative to the user.
884 static int gfs2_glock_holder_wait(void *word)
890 static int gfs2_glock_demote_wait(void *word)
896 static void wait_on_holder(struct gfs2_holder *gh)
899 wait_on_bit(&gh->gh_iflags, HIF_WAIT, gfs2_glock_holder_wait, TASK_UNINTERRUPTIBLE);
902 static void wait_on_demote(struct gfs2_glock *gl)
905 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, gfs2_glock_demote_wait, TASK_UNINTERRUPTIBLE);
909 * handle_callback - process a demote request
911 * @state: the state the caller wants us to change to
913 * There are only two requests that we are going to see in actual
914 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
917 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
920 int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
922 set_bit(bit, &gl->gl_flags);
923 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
924 gl->gl_demote_state = state;
925 gl->gl_demote_time = jiffies;
926 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
927 gl->gl_demote_state != state) {
928 gl->gl_demote_state = LM_ST_UNLOCKED;
930 if (gl->gl_ops->go_callback)
931 gl->gl_ops->go_callback(gl);
932 trace_gfs2_demote_rq(gl);
936 * gfs2_glock_wait - wait on a glock acquisition
937 * @gh: the glock holder
939 * Returns: 0 on success
942 int gfs2_glock_wait(struct gfs2_holder *gh)
948 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
954 struct gfs2_glock_iter *gi = seq->private;
955 vsprintf(gi->string, fmt, args);
956 seq_printf(seq, gi->string);
958 printk(KERN_ERR " ");
965 * add_to_queue - Add a holder to the wait queue (but look for recursion)
966 * @gh: the holder structure to add
968 * Eventually we should move the recursive locking trap to a
969 * debugging option or something like that. This is the fast
970 * path and needs to have the minimum number of distractions.
974 static inline void add_to_queue(struct gfs2_holder *gh)
975 __releases(&gl->gl_spin)
976 __acquires(&gl->gl_spin)
978 struct gfs2_glock *gl = gh->gh_gl;
979 struct gfs2_sbd *sdp = gl->gl_sbd;
980 struct list_head *insert_pt = NULL;
981 struct gfs2_holder *gh2;
984 BUG_ON(gh->gh_owner_pid == NULL);
985 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
988 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
989 if (test_bit(GLF_LOCK, &gl->gl_flags))
991 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
995 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
996 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
997 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
1000 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) &&
1001 !may_grant(gl, gh)) {
1003 gh->gh_error = GLR_TRYFAILED;
1004 gfs2_holder_wake(gh);
1007 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1009 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1010 insert_pt = &gh2->gh_list;
1012 if (likely(insert_pt == NULL)) {
1013 list_add_tail(&gh->gh_list, &gl->gl_holders);
1014 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1018 trace_gfs2_glock_queue(gh, 1);
1019 list_add_tail(&gh->gh_list, insert_pt);
1021 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
1022 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1023 spin_unlock(&gl->gl_spin);
1024 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1025 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1026 spin_lock(&gl->gl_spin);
1031 print_symbol(KERN_ERR "original: %s\n", gh2->gh_ip);
1032 printk(KERN_ERR "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1033 printk(KERN_ERR "lock type: %d req lock state : %d\n",
1034 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1035 print_symbol(KERN_ERR "new: %s\n", gh->gh_ip);
1036 printk(KERN_ERR "pid: %d\n", pid_nr(gh->gh_owner_pid));
1037 printk(KERN_ERR "lock type: %d req lock state : %d\n",
1038 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1039 __dump_glock(NULL, gl);
1044 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1045 * @gh: the holder structure
1047 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1049 * Returns: 0, GLR_TRYFAILED, or errno on failure
1052 int gfs2_glock_nq(struct gfs2_holder *gh)
1054 struct gfs2_glock *gl = gh->gh_gl;
1055 struct gfs2_sbd *sdp = gl->gl_sbd;
1058 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1061 spin_lock(&gl->gl_spin);
1064 spin_unlock(&gl->gl_spin);
1066 if (!(gh->gh_flags & GL_ASYNC))
1067 error = gfs2_glock_wait(gh);
1073 * gfs2_glock_poll - poll to see if an async request has been completed
1076 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1079 int gfs2_glock_poll(struct gfs2_holder *gh)
1081 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1085 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1086 * @gh: the glock holder
1090 void gfs2_glock_dq(struct gfs2_holder *gh)
1092 struct gfs2_glock *gl = gh->gh_gl;
1093 const struct gfs2_glock_operations *glops = gl->gl_ops;
1097 spin_lock(&gl->gl_spin);
1098 if (gh->gh_flags & GL_NOCACHE)
1099 handle_callback(gl, LM_ST_UNLOCKED, 0);
1101 list_del_init(&gh->gh_list);
1102 if (find_first_holder(gl) == NULL) {
1103 if (glops->go_unlock) {
1104 GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1105 spin_unlock(&gl->gl_spin);
1106 glops->go_unlock(gh);
1107 spin_lock(&gl->gl_spin);
1108 clear_bit(GLF_LOCK, &gl->gl_flags);
1110 if (list_empty(&gl->gl_holders) &&
1111 !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1112 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1115 trace_gfs2_glock_queue(gh, 0);
1116 spin_unlock(&gl->gl_spin);
1117 if (likely(fast_path))
1120 gfs2_glock_hold(gl);
1121 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1122 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1123 delay = gl->gl_ops->go_min_hold_time;
1124 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1128 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1130 struct gfs2_glock *gl = gh->gh_gl;
1136 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1137 * @gh: the holder structure
1141 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1144 gfs2_holder_uninit(gh);
1148 * gfs2_glock_nq_num - acquire a glock based on lock number
1149 * @sdp: the filesystem
1150 * @number: the lock number
1151 * @glops: the glock operations for the type of glock
1152 * @state: the state to acquire the glock in
1153 * @flags: modifier flags for the aquisition
1154 * @gh: the struct gfs2_holder
1159 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1160 const struct gfs2_glock_operations *glops,
1161 unsigned int state, int flags, struct gfs2_holder *gh)
1163 struct gfs2_glock *gl;
1166 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1168 error = gfs2_glock_nq_init(gl, state, flags, gh);
1176 * glock_compare - Compare two struct gfs2_glock structures for sorting
1177 * @arg_a: the first structure
1178 * @arg_b: the second structure
1182 static int glock_compare(const void *arg_a, const void *arg_b)
1184 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1185 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1186 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1187 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1189 if (a->ln_number > b->ln_number)
1191 if (a->ln_number < b->ln_number)
1193 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1198 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1199 * @num_gh: the number of structures
1200 * @ghs: an array of struct gfs2_holder structures
1202 * Returns: 0 on success (all glocks acquired),
1203 * errno on failure (no glocks acquired)
1206 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1207 struct gfs2_holder **p)
1212 for (x = 0; x < num_gh; x++)
1215 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1217 for (x = 0; x < num_gh; x++) {
1218 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1220 error = gfs2_glock_nq(p[x]);
1223 gfs2_glock_dq(p[x]);
1232 * gfs2_glock_nq_m - acquire multiple glocks
1233 * @num_gh: the number of structures
1234 * @ghs: an array of struct gfs2_holder structures
1237 * Returns: 0 on success (all glocks acquired),
1238 * errno on failure (no glocks acquired)
1241 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1243 struct gfs2_holder *tmp[4];
1244 struct gfs2_holder **pph = tmp;
1251 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1252 return gfs2_glock_nq(ghs);
1256 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1261 error = nq_m_sync(num_gh, ghs, pph);
1270 * gfs2_glock_dq_m - release multiple glocks
1271 * @num_gh: the number of structures
1272 * @ghs: an array of struct gfs2_holder structures
1276 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1280 for (x = 0; x < num_gh; x++)
1281 gfs2_glock_dq(&ghs[x]);
1285 * gfs2_glock_dq_uninit_m - release multiple glocks
1286 * @num_gh: the number of structures
1287 * @ghs: an array of struct gfs2_holder structures
1291 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1295 for (x = 0; x < num_gh; x++)
1296 gfs2_glock_dq_uninit(&ghs[x]);
1299 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1301 unsigned long delay = 0;
1302 unsigned long holdtime;
1303 unsigned long now = jiffies;
1305 gfs2_glock_hold(gl);
1306 holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
1307 if (time_before(now, holdtime))
1308 delay = holdtime - now;
1309 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1310 delay = gl->gl_ops->go_min_hold_time;
1312 spin_lock(&gl->gl_spin);
1313 handle_callback(gl, state, delay);
1314 spin_unlock(&gl->gl_spin);
1315 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1320 * gfs2_glock_complete - Callback used by locking
1321 * @gl: Pointer to the glock
1322 * @ret: The return value from the dlm
1326 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1328 struct lm_lockstruct *ls = &gl->gl_sbd->sd_lockstruct;
1330 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_flags))) {
1331 struct gfs2_holder *gh;
1332 spin_lock(&gl->gl_spin);
1333 gh = find_first_waiter(gl);
1334 if ((!(gh && (gh->gh_flags & LM_FLAG_NOEXP)) &&
1335 (gl->gl_target != LM_ST_UNLOCKED)) ||
1336 ((ret & ~LM_OUT_ST_MASK) != 0))
1337 set_bit(GLF_FROZEN, &gl->gl_flags);
1338 spin_unlock(&gl->gl_spin);
1339 if (test_bit(GLF_FROZEN, &gl->gl_flags))
1342 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1343 gfs2_glock_hold(gl);
1344 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1349 static int gfs2_shrink_glock_memory(int nr, gfp_t gfp_mask)
1351 struct gfs2_glock *gl;
1359 if (!(gfp_mask & __GFP_FS))
1362 spin_lock(&lru_lock);
1363 while(nr && !list_empty(&lru_list)) {
1364 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1365 list_del_init(&gl->gl_lru);
1366 atomic_dec(&lru_count);
1368 /* Test for being demotable */
1369 if (!test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1370 gfs2_glock_hold(gl);
1371 spin_unlock(&lru_lock);
1372 spin_lock(&gl->gl_spin);
1373 may_demote = demote_ok(gl);
1375 handle_callback(gl, LM_ST_UNLOCKED, 0);
1378 clear_bit(GLF_LOCK, &gl->gl_flags);
1379 smp_mb__after_clear_bit();
1380 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1381 gfs2_glock_put_nolock(gl);
1382 spin_unlock(&gl->gl_spin);
1383 spin_lock(&lru_lock);
1387 list_add(&gl->gl_lru, &skipped);
1389 list_splice(&skipped, &lru_list);
1390 atomic_add(nr_skipped, &lru_count);
1391 spin_unlock(&lru_lock);
1393 return (atomic_read(&lru_count) / 100) * sysctl_vfs_cache_pressure;
1396 static struct shrinker glock_shrinker = {
1397 .shrink = gfs2_shrink_glock_memory,
1398 .seeks = DEFAULT_SEEKS,
1402 * examine_bucket - Call a function for glock in a hash bucket
1403 * @examiner: the function
1404 * @sdp: the filesystem
1405 * @bucket: the bucket
1407 * Returns: 1 if the bucket has entries
1410 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1413 struct gfs2_glock *gl, *prev = NULL;
1414 int has_entries = 0;
1415 struct hlist_head *head = &gl_hash_table[hash].hb_list;
1417 read_lock(gl_lock_addr(hash));
1418 /* Can't use hlist_for_each_entry - don't want prefetch here */
1419 if (hlist_empty(head))
1421 gl = list_entry(head->first, struct gfs2_glock, gl_list);
1423 if (!sdp || gl->gl_sbd == sdp) {
1424 gfs2_glock_hold(gl);
1425 read_unlock(gl_lock_addr(hash));
1427 gfs2_glock_put(prev);
1431 read_lock(gl_lock_addr(hash));
1433 if (gl->gl_list.next == NULL)
1435 gl = list_entry(gl->gl_list.next, struct gfs2_glock, gl_list);
1438 read_unlock(gl_lock_addr(hash));
1440 gfs2_glock_put(prev);
1447 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1448 * @gl: The glock to thaw
1450 * N.B. When we freeze a glock, we leave a ref to the glock outstanding,
1451 * so this has to result in the ref count being dropped by one.
1454 static void thaw_glock(struct gfs2_glock *gl)
1456 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1458 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1459 gfs2_glock_hold(gl);
1460 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1465 * clear_glock - look at a glock and see if we can free it from glock cache
1466 * @gl: the glock to look at
1470 static void clear_glock(struct gfs2_glock *gl)
1472 spin_lock(&lru_lock);
1473 if (!list_empty(&gl->gl_lru)) {
1474 list_del_init(&gl->gl_lru);
1475 atomic_dec(&lru_count);
1477 spin_unlock(&lru_lock);
1479 spin_lock(&gl->gl_spin);
1480 if (find_first_holder(gl) == NULL && gl->gl_state != LM_ST_UNLOCKED)
1481 handle_callback(gl, LM_ST_UNLOCKED, 0);
1482 spin_unlock(&gl->gl_spin);
1483 gfs2_glock_hold(gl);
1484 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1489 * gfs2_glock_thaw - Thaw any frozen glocks
1490 * @sdp: The super block
1494 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1498 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1499 examine_bucket(thaw_glock, sdp, x);
1503 * gfs2_gl_hash_clear - Empty out the glock hash table
1504 * @sdp: the filesystem
1505 * @wait: wait until it's all gone
1507 * Called when unmounting the filesystem.
1510 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1520 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1521 if (examine_bucket(clear_glock, sdp, x))
1528 if (time_after_eq(jiffies,
1529 t + gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
1530 fs_warn(sdp, "Unmount seems to be stalled. "
1531 "Dumping lock state...\n");
1532 gfs2_dump_lockstate(sdp);
1536 down_write(&gfs2_umount_flush_sem);
1537 invalidate_inodes(sdp->sd_vfs);
1538 up_write(&gfs2_umount_flush_sem);
1543 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1545 struct gfs2_glock *gl = ip->i_gl;
1548 ret = gfs2_truncatei_resume(ip);
1549 gfs2_assert_withdraw(gl->gl_sbd, ret == 0);
1551 spin_lock(&gl->gl_spin);
1552 clear_bit(GLF_LOCK, &gl->gl_flags);
1554 spin_unlock(&gl->gl_spin);
1557 static const char *state2str(unsigned state)
1560 case LM_ST_UNLOCKED:
1564 case LM_ST_DEFERRED:
1566 case LM_ST_EXCLUSIVE:
1572 static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1575 if (flags & LM_FLAG_TRY)
1577 if (flags & LM_FLAG_TRY_1CB)
1579 if (flags & LM_FLAG_NOEXP)
1581 if (flags & LM_FLAG_ANY)
1583 if (flags & LM_FLAG_PRIORITY)
1585 if (flags & GL_ASYNC)
1587 if (flags & GL_EXACT)
1589 if (flags & GL_NOCACHE)
1591 if (test_bit(HIF_HOLDER, &iflags))
1593 if (test_bit(HIF_WAIT, &iflags))
1595 if (test_bit(HIF_FIRST, &iflags))
1602 * dump_holder - print information about a glock holder
1603 * @seq: the seq_file struct
1604 * @gh: the glock holder
1606 * Returns: 0 on success, -ENOBUFS when we run out of space
1609 static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1611 struct task_struct *gh_owner = NULL;
1612 char buffer[KSYM_SYMBOL_LEN];
1615 sprint_symbol(buffer, gh->gh_ip);
1616 if (gh->gh_owner_pid)
1617 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1618 gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %s\n",
1619 state2str(gh->gh_state),
1620 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1622 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1623 gh_owner ? gh_owner->comm : "(ended)", buffer);
1627 static const char *gflags2str(char *buf, const unsigned long *gflags)
1630 if (test_bit(GLF_LOCK, gflags))
1632 if (test_bit(GLF_DEMOTE, gflags))
1634 if (test_bit(GLF_PENDING_DEMOTE, gflags))
1636 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1638 if (test_bit(GLF_DIRTY, gflags))
1640 if (test_bit(GLF_LFLUSH, gflags))
1642 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1644 if (test_bit(GLF_REPLY_PENDING, gflags))
1646 if (test_bit(GLF_INITIAL, gflags))
1648 if (test_bit(GLF_FROZEN, gflags))
1655 * __dump_glock - print information about a glock
1656 * @seq: The seq_file struct
1659 * The file format is as follows:
1660 * One line per object, capital letters are used to indicate objects
1661 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1662 * other objects are indented by a single space and follow the glock to
1663 * which they are related. Fields are indicated by lower case letters
1664 * followed by a colon and the field value, except for strings which are in
1665 * [] so that its possible to see if they are composed of spaces for
1666 * example. The field's are n = number (id of the object), f = flags,
1667 * t = type, s = state, r = refcount, e = error, p = pid.
1669 * Returns: 0 on success, -ENOBUFS when we run out of space
1672 static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1674 const struct gfs2_glock_operations *glops = gl->gl_ops;
1675 unsigned long long dtime;
1676 const struct gfs2_holder *gh;
1677 char gflags_buf[32];
1680 dtime = jiffies - gl->gl_demote_time;
1681 dtime *= 1000000/HZ; /* demote time in uSec */
1682 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1684 gfs2_print_dbg(seq, "G: s:%s n:%u/%llu f:%s t:%s d:%s/%llu a:%d r:%d\n",
1685 state2str(gl->gl_state),
1686 gl->gl_name.ln_type,
1687 (unsigned long long)gl->gl_name.ln_number,
1688 gflags2str(gflags_buf, &gl->gl_flags),
1689 state2str(gl->gl_target),
1690 state2str(gl->gl_demote_state), dtime,
1691 atomic_read(&gl->gl_ail_count),
1692 atomic_read(&gl->gl_ref));
1694 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1695 error = dump_holder(seq, gh);
1699 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1700 error = glops->go_dump(seq, gl);
1705 static int dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1708 spin_lock(&gl->gl_spin);
1709 ret = __dump_glock(seq, gl);
1710 spin_unlock(&gl->gl_spin);
1715 * gfs2_dump_lockstate - print out the current lockstate
1716 * @sdp: the filesystem
1717 * @ub: the buffer to copy the information into
1719 * If @ub is NULL, dump the lockstate to the console.
1723 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
1725 struct gfs2_glock *gl;
1726 struct hlist_node *h;
1730 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1732 read_lock(gl_lock_addr(x));
1734 hlist_for_each_entry(gl, h, &gl_hash_table[x].hb_list, gl_list) {
1735 if (gl->gl_sbd != sdp)
1738 error = dump_glock(NULL, gl);
1743 read_unlock(gl_lock_addr(x));
1754 int __init gfs2_glock_init(void)
1757 for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
1758 INIT_HLIST_HEAD(&gl_hash_table[i].hb_list);
1760 #ifdef GL_HASH_LOCK_SZ
1761 for(i = 0; i < GL_HASH_LOCK_SZ; i++) {
1762 rwlock_init(&gl_hash_locks[i]);
1766 glock_workqueue = create_workqueue("glock_workqueue");
1767 if (IS_ERR(glock_workqueue))
1768 return PTR_ERR(glock_workqueue);
1769 gfs2_delete_workqueue = create_workqueue("delete_workqueue");
1770 if (IS_ERR(gfs2_delete_workqueue)) {
1771 destroy_workqueue(glock_workqueue);
1772 return PTR_ERR(gfs2_delete_workqueue);
1775 register_shrinker(&glock_shrinker);
1780 void gfs2_glock_exit(void)
1782 unregister_shrinker(&glock_shrinker);
1783 destroy_workqueue(glock_workqueue);
1784 destroy_workqueue(gfs2_delete_workqueue);
1787 static int gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1789 struct gfs2_glock *gl;
1792 read_lock(gl_lock_addr(gi->hash));
1795 gi->gl = hlist_entry(gl->gl_list.next,
1796 struct gfs2_glock, gl_list);
1798 gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
1799 struct gfs2_glock, gl_list);
1802 gfs2_glock_hold(gi->gl);
1803 read_unlock(gl_lock_addr(gi->hash));
1806 while (gi->gl == NULL) {
1808 if (gi->hash >= GFS2_GL_HASH_SIZE)
1810 read_lock(gl_lock_addr(gi->hash));
1811 gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
1812 struct gfs2_glock, gl_list);
1814 gfs2_glock_hold(gi->gl);
1815 read_unlock(gl_lock_addr(gi->hash));
1818 if (gi->sdp != gi->gl->gl_sbd)
1824 static void gfs2_glock_iter_free(struct gfs2_glock_iter *gi)
1827 gfs2_glock_put(gi->gl);
1831 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1833 struct gfs2_glock_iter *gi = seq->private;
1839 if (gfs2_glock_iter_next(gi)) {
1840 gfs2_glock_iter_free(gi);
1848 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1851 struct gfs2_glock_iter *gi = seq->private;
1855 if (gfs2_glock_iter_next(gi)) {
1856 gfs2_glock_iter_free(gi);
1863 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1865 struct gfs2_glock_iter *gi = seq->private;
1866 gfs2_glock_iter_free(gi);
1869 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1871 return dump_glock(seq, iter_ptr);
1874 static const struct seq_operations gfs2_glock_seq_ops = {
1875 .start = gfs2_glock_seq_start,
1876 .next = gfs2_glock_seq_next,
1877 .stop = gfs2_glock_seq_stop,
1878 .show = gfs2_glock_seq_show,
1881 static int gfs2_debugfs_open(struct inode *inode, struct file *file)
1883 int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1884 sizeof(struct gfs2_glock_iter));
1886 struct seq_file *seq = file->private_data;
1887 struct gfs2_glock_iter *gi = seq->private;
1888 gi->sdp = inode->i_private;
1893 static const struct file_operations gfs2_debug_fops = {
1894 .owner = THIS_MODULE,
1895 .open = gfs2_debugfs_open,
1897 .llseek = seq_lseek,
1898 .release = seq_release_private,
1901 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
1903 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
1904 if (!sdp->debugfs_dir)
1906 sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
1908 sdp->debugfs_dir, sdp,
1910 if (!sdp->debugfs_dentry_glocks)
1916 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
1918 if (sdp && sdp->debugfs_dir) {
1919 if (sdp->debugfs_dentry_glocks) {
1920 debugfs_remove(sdp->debugfs_dentry_glocks);
1921 sdp->debugfs_dentry_glocks = NULL;
1923 debugfs_remove(sdp->debugfs_dir);
1924 sdp->debugfs_dir = NULL;
1928 int gfs2_register_debugfs(void)
1930 gfs2_root = debugfs_create_dir("gfs2", NULL);
1931 return gfs2_root ? 0 : -ENOMEM;
1934 void gfs2_unregister_debugfs(void)
1936 debugfs_remove(gfs2_root);