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/completion.h>
14 #include <linux/buffer_head.h>
15 #include <linux/delay.h>
16 #include <linux/sort.h>
17 #include <linux/jhash.h>
18 #include <linux/kallsyms.h>
19 #include <linux/gfs2_ondisk.h>
20 #include <linux/list.h>
21 #include <linux/lm_interface.h>
22 #include <linux/wait.h>
23 #include <linux/module.h>
24 #include <linux/rwsem.h>
25 #include <asm/uaccess.h>
26 #include <linux/seq_file.h>
27 #include <linux/debugfs.h>
28 #include <linux/kthread.h>
29 #include <linux/freezer.h>
30 #include <linux/workqueue.h>
31 #include <linux/jiffies.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 DECLARE_RWSEM(gfs2_umount_flush_sem);
63 static struct dentry *gfs2_root;
64 static struct task_struct *scand_process;
65 static unsigned int scand_secs = 5;
66 static struct workqueue_struct *glock_workqueue;
68 #define GFS2_GL_HASH_SHIFT 15
69 #define GFS2_GL_HASH_SIZE (1 << GFS2_GL_HASH_SHIFT)
70 #define GFS2_GL_HASH_MASK (GFS2_GL_HASH_SIZE - 1)
72 static struct gfs2_gl_hash_bucket gl_hash_table[GFS2_GL_HASH_SIZE];
73 static struct dentry *gfs2_root;
76 * Despite what you might think, the numbers below are not arbitrary :-)
77 * They are taken from the ipv4 routing hash code, which is well tested
78 * and thus should be nearly optimal. Later on we might tweek the numbers
79 * but for now this should be fine.
81 * The reason for putting the locks in a separate array from the list heads
82 * is that we can have fewer locks than list heads and save memory. We use
83 * the same hash function for both, but with a different hash mask.
85 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
86 defined(CONFIG_PROVE_LOCKING)
89 # define GL_HASH_LOCK_SZ 256
92 # define GL_HASH_LOCK_SZ 4096
94 # define GL_HASH_LOCK_SZ 2048
96 # define GL_HASH_LOCK_SZ 1024
98 # define GL_HASH_LOCK_SZ 512
100 # define GL_HASH_LOCK_SZ 256
104 /* We never want more locks than chains */
105 #if GFS2_GL_HASH_SIZE < GL_HASH_LOCK_SZ
106 # undef GL_HASH_LOCK_SZ
107 # define GL_HASH_LOCK_SZ GFS2_GL_HASH_SIZE
110 static rwlock_t gl_hash_locks[GL_HASH_LOCK_SZ];
112 static inline rwlock_t *gl_lock_addr(unsigned int x)
114 return &gl_hash_locks[x & (GL_HASH_LOCK_SZ-1)];
116 #else /* not SMP, so no spinlocks required */
117 static inline rwlock_t *gl_lock_addr(unsigned int x)
124 * gl_hash() - Turn glock number into hash bucket number
125 * @lock: The glock number
127 * Returns: The number of the corresponding hash bucket
130 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
131 const struct lm_lockname *name)
135 h = jhash(&name->ln_number, sizeof(u64), 0);
136 h = jhash(&name->ln_type, sizeof(unsigned int), h);
137 h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
138 h &= GFS2_GL_HASH_MASK;
144 * glock_free() - Perform a few checks and then release struct gfs2_glock
145 * @gl: The glock to release
147 * Also calls lock module to release its internal structure for this glock.
151 static void glock_free(struct gfs2_glock *gl)
153 struct gfs2_sbd *sdp = gl->gl_sbd;
154 struct inode *aspace = gl->gl_aspace;
156 if (sdp->sd_lockstruct.ls_ops->lm_put_lock)
157 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl->gl_lock);
160 gfs2_aspace_put(aspace);
162 kmem_cache_free(gfs2_glock_cachep, gl);
166 * gfs2_glock_hold() - increment reference count on glock
167 * @gl: The glock to hold
171 static void gfs2_glock_hold(struct gfs2_glock *gl)
173 atomic_inc(&gl->gl_ref);
177 * gfs2_glock_put() - Decrement reference count on glock
178 * @gl: The glock to put
182 int gfs2_glock_put(struct gfs2_glock *gl)
186 write_lock(gl_lock_addr(gl->gl_hash));
187 if (atomic_dec_and_test(&gl->gl_ref)) {
188 hlist_del(&gl->gl_list);
189 write_unlock(gl_lock_addr(gl->gl_hash));
190 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_UNLOCKED);
191 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_reclaim));
192 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
197 write_unlock(gl_lock_addr(gl->gl_hash));
203 * search_bucket() - Find struct gfs2_glock by lock number
204 * @bucket: the bucket to search
205 * @name: The lock name
207 * Returns: NULL, or the struct gfs2_glock with the requested number
210 static struct gfs2_glock *search_bucket(unsigned int hash,
211 const struct gfs2_sbd *sdp,
212 const struct lm_lockname *name)
214 struct gfs2_glock *gl;
215 struct hlist_node *h;
217 hlist_for_each_entry(gl, h, &gl_hash_table[hash].hb_list, gl_list) {
218 if (!lm_name_equal(&gl->gl_name, name))
220 if (gl->gl_sbd != sdp)
223 atomic_inc(&gl->gl_ref);
232 * gfs2_glock_find() - Find glock by lock number
233 * @sdp: The GFS2 superblock
234 * @name: The lock name
236 * Returns: NULL, or the struct gfs2_glock with the requested number
239 static struct gfs2_glock *gfs2_glock_find(const struct gfs2_sbd *sdp,
240 const struct lm_lockname *name)
242 unsigned int hash = gl_hash(sdp, name);
243 struct gfs2_glock *gl;
245 read_lock(gl_lock_addr(hash));
246 gl = search_bucket(hash, sdp, name);
247 read_unlock(gl_lock_addr(hash));
253 * may_grant - check if its ok to grant a new lock
255 * @gh: The lock request which we wish to grant
257 * Returns: true if its ok to grant the lock
260 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
262 const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
263 if ((gh->gh_state == LM_ST_EXCLUSIVE ||
264 gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
266 if (gl->gl_state == gh->gh_state)
268 if (gh->gh_flags & GL_EXACT)
270 if (gl->gl_state == LM_ST_EXCLUSIVE) {
271 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
273 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
276 if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
281 static void gfs2_holder_wake(struct gfs2_holder *gh)
283 clear_bit(HIF_WAIT, &gh->gh_iflags);
284 smp_mb__after_clear_bit();
285 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
289 * do_promote - promote as many requests as possible on the current queue
292 * Returns: true if there is a blocked holder at the head of the list
295 static int do_promote(struct gfs2_glock *gl)
296 __releases(&gl->gl_spin)
297 __acquires(&gl->gl_spin)
299 const struct gfs2_glock_operations *glops = gl->gl_ops;
300 struct gfs2_holder *gh, *tmp;
304 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
305 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
307 if (may_grant(gl, gh)) {
308 if (gh->gh_list.prev == &gl->gl_holders &&
310 spin_unlock(&gl->gl_spin);
311 /* FIXME: eliminate this eventually */
312 ret = glops->go_lock(gh);
313 spin_lock(&gl->gl_spin);
316 list_del_init(&gh->gh_list);
317 gfs2_holder_wake(gh);
320 set_bit(HIF_HOLDER, &gh->gh_iflags);
321 gfs2_holder_wake(gh);
324 set_bit(HIF_HOLDER, &gh->gh_iflags);
325 gfs2_holder_wake(gh);
328 if (gh->gh_list.prev == &gl->gl_holders)
336 * do_error - Something unexpected has happened during a lock request
340 static inline void do_error(struct gfs2_glock *gl, const int ret)
342 struct gfs2_holder *gh, *tmp;
344 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
345 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
347 if (ret & LM_OUT_ERROR)
349 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
350 gh->gh_error = GLR_TRYFAILED;
353 list_del_init(&gh->gh_list);
354 gfs2_holder_wake(gh);
359 * find_first_waiter - find the first gh that's waiting for the glock
363 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
365 struct gfs2_holder *gh;
367 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
368 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
375 * state_change - record that the glock is now in a different state
377 * @new_state the new state
381 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
385 held1 = (gl->gl_state != LM_ST_UNLOCKED);
386 held2 = (new_state != LM_ST_UNLOCKED);
388 if (held1 != held2) {
395 gl->gl_state = new_state;
396 gl->gl_tchange = jiffies;
399 static void gfs2_demote_wake(struct gfs2_glock *gl)
401 gl->gl_demote_state = LM_ST_EXCLUSIVE;
402 clear_bit(GLF_DEMOTE, &gl->gl_flags);
403 smp_mb__after_clear_bit();
404 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
408 * finish_xmote - The DLM has replied to one of our lock requests
410 * @ret: The status from the DLM
414 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
416 const struct gfs2_glock_operations *glops = gl->gl_ops;
417 struct gfs2_holder *gh;
418 unsigned state = ret & LM_OUT_ST_MASK;
420 spin_lock(&gl->gl_spin);
421 state_change(gl, state);
422 gh = find_first_waiter(gl);
424 /* Demote to UN request arrived during demote to SH or DF */
425 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
426 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
427 gl->gl_target = LM_ST_UNLOCKED;
429 /* Check for state != intended state */
430 if (unlikely(state != gl->gl_target)) {
431 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
432 /* move to back of queue and try next entry */
433 if (ret & LM_OUT_CANCELED) {
434 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
435 list_move_tail(&gh->gh_list, &gl->gl_holders);
436 gh = find_first_waiter(gl);
437 gl->gl_target = gh->gh_state;
440 /* Some error or failed "try lock" - report it */
441 if ((ret & LM_OUT_ERROR) ||
442 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
443 gl->gl_target = gl->gl_state;
449 /* Unlocked due to conversion deadlock, try again */
452 do_xmote(gl, gh, gl->gl_target);
454 /* Conversion fails, unlock and try again */
457 do_xmote(gl, gh, LM_ST_UNLOCKED);
459 default: /* Everything else */
460 printk(KERN_ERR "GFS2: wanted %u got %u\n", gl->gl_target, state);
463 spin_unlock(&gl->gl_spin);
468 /* Fast path - we got what we asked for */
469 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
470 gfs2_demote_wake(gl);
471 if (state != LM_ST_UNLOCKED) {
472 if (glops->go_xmote_bh) {
474 spin_unlock(&gl->gl_spin);
475 rv = glops->go_xmote_bh(gl, gh);
478 spin_lock(&gl->gl_spin);
487 clear_bit(GLF_LOCK, &gl->gl_flags);
488 spin_unlock(&gl->gl_spin);
492 static unsigned int gfs2_lm_lock(struct gfs2_sbd *sdp, void *lock,
493 unsigned int cur_state, unsigned int req_state,
496 int ret = LM_OUT_ERROR;
498 if (!sdp->sd_lockstruct.ls_ops->lm_lock)
499 return req_state == LM_ST_UNLOCKED ? 0 : req_state;
501 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
502 ret = sdp->sd_lockstruct.ls_ops->lm_lock(lock, cur_state,
508 * do_xmote - Calls the DLM to change the state of a lock
509 * @gl: The lock state
510 * @gh: The holder (only for promotes)
511 * @target: The target lock state
515 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
516 __releases(&gl->gl_spin)
517 __acquires(&gl->gl_spin)
519 const struct gfs2_glock_operations *glops = gl->gl_ops;
520 struct gfs2_sbd *sdp = gl->gl_sbd;
521 unsigned int lck_flags = gh ? gh->gh_flags : 0;
524 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
526 BUG_ON(gl->gl_state == target);
527 BUG_ON(gl->gl_state == gl->gl_target);
528 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
530 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
531 do_error(gl, 0); /* Fail queued try locks */
533 spin_unlock(&gl->gl_spin);
534 if (glops->go_xmote_th)
535 glops->go_xmote_th(gl);
536 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
537 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
538 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
541 if (target != LM_ST_UNLOCKED && (gl->gl_state == LM_ST_SHARED ||
542 gl->gl_state == LM_ST_DEFERRED) &&
543 !(lck_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
544 lck_flags |= LM_FLAG_TRY_1CB;
545 ret = gfs2_lm_lock(sdp, gl->gl_lock, gl->gl_state, target, lck_flags);
547 if (!(ret & LM_OUT_ASYNC)) {
548 finish_xmote(gl, ret);
550 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
553 GLOCK_BUG_ON(gl, ret != LM_OUT_ASYNC);
555 spin_lock(&gl->gl_spin);
559 * find_first_holder - find the first "holder" gh
563 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
565 struct gfs2_holder *gh;
567 if (!list_empty(&gl->gl_holders)) {
568 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
569 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
576 * run_queue - do all outstanding tasks related to a glock
577 * @gl: The glock in question
578 * @nonblock: True if we must not block in run_queue
582 static void run_queue(struct gfs2_glock *gl, const int nonblock)
583 __releases(&gl->gl_spin)
584 __acquires(&gl->gl_spin)
586 struct gfs2_holder *gh = NULL;
588 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
591 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
593 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
594 gl->gl_demote_state != gl->gl_state) {
595 if (find_first_holder(gl))
599 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
600 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
601 gl->gl_target = gl->gl_demote_state;
603 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
604 gfs2_demote_wake(gl);
605 if (do_promote(gl) == 0)
607 gh = find_first_waiter(gl);
608 gl->gl_target = gh->gh_state;
609 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
610 do_error(gl, 0); /* Fail queued try locks */
612 do_xmote(gl, gh, gl->gl_target);
617 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
620 clear_bit(GLF_LOCK, &gl->gl_flags);
623 static void glock_work_func(struct work_struct *work)
625 unsigned long delay = 0;
626 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
628 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags))
629 finish_xmote(gl, gl->gl_reply);
630 spin_lock(&gl->gl_spin);
631 if (test_and_clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
632 gl->gl_state != LM_ST_UNLOCKED &&
633 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
634 unsigned long holdtime, now = jiffies;
635 holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
636 if (time_before(now, holdtime))
637 delay = holdtime - now;
638 set_bit(delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE, &gl->gl_flags);
641 spin_unlock(&gl->gl_spin);
643 queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
647 static int gfs2_lm_get_lock(struct gfs2_sbd *sdp, struct lm_lockname *name,
651 if (!sdp->sd_lockstruct.ls_ops->lm_get_lock)
653 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
654 error = sdp->sd_lockstruct.ls_ops->lm_get_lock(
655 sdp->sd_lockstruct.ls_lockspace, name, lockp);
660 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
661 * @sdp: The GFS2 superblock
662 * @number: the lock number
663 * @glops: The glock_operations to use
664 * @create: If 0, don't create the glock if it doesn't exist
665 * @glp: the glock is returned here
667 * This does not lock a glock, just finds/creates structures for one.
672 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
673 const struct gfs2_glock_operations *glops, int create,
674 struct gfs2_glock **glp)
676 struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
677 struct gfs2_glock *gl, *tmp;
678 unsigned int hash = gl_hash(sdp, &name);
681 read_lock(gl_lock_addr(hash));
682 gl = search_bucket(hash, sdp, &name);
683 read_unlock(gl_lock_addr(hash));
690 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
696 atomic_set(&gl->gl_ref, 1);
697 gl->gl_state = LM_ST_UNLOCKED;
698 gl->gl_target = LM_ST_UNLOCKED;
699 gl->gl_demote_state = LM_ST_EXCLUSIVE;
702 gl->gl_stamp = jiffies;
703 gl->gl_tchange = jiffies;
704 gl->gl_object = NULL;
706 gl->gl_aspace = NULL;
707 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
709 /* If this glock protects actual on-disk data or metadata blocks,
710 create a VFS inode to manage the pages/buffers holding them. */
711 if (glops == &gfs2_inode_glops || glops == &gfs2_rgrp_glops) {
712 gl->gl_aspace = gfs2_aspace_get(sdp);
713 if (!gl->gl_aspace) {
719 error = gfs2_lm_get_lock(sdp, &name, &gl->gl_lock);
723 write_lock(gl_lock_addr(hash));
724 tmp = search_bucket(hash, sdp, &name);
726 write_unlock(gl_lock_addr(hash));
730 hlist_add_head(&gl->gl_list, &gl_hash_table[hash].hb_list);
731 write_unlock(gl_lock_addr(hash));
740 gfs2_aspace_put(gl->gl_aspace);
742 kmem_cache_free(gfs2_glock_cachep, gl);
747 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
749 * @state: the state we're requesting
750 * @flags: the modifier flags
751 * @gh: the holder structure
755 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
756 struct gfs2_holder *gh)
758 INIT_LIST_HEAD(&gh->gh_list);
760 gh->gh_ip = (unsigned long)__builtin_return_address(0);
761 gh->gh_owner_pid = get_pid(task_pid(current));
762 gh->gh_state = state;
763 gh->gh_flags = flags;
770 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
771 * @state: the state we're requesting
772 * @flags: the modifier flags
773 * @gh: the holder structure
775 * Don't mess with the glock.
779 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
781 gh->gh_state = state;
782 gh->gh_flags = flags;
784 gh->gh_ip = (unsigned long)__builtin_return_address(0);
788 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
789 * @gh: the holder structure
793 void gfs2_holder_uninit(struct gfs2_holder *gh)
795 put_pid(gh->gh_owner_pid);
796 gfs2_glock_put(gh->gh_gl);
801 static int just_schedule(void *word)
807 static void wait_on_holder(struct gfs2_holder *gh)
810 wait_on_bit(&gh->gh_iflags, HIF_WAIT, just_schedule, TASK_UNINTERRUPTIBLE);
813 static void wait_on_demote(struct gfs2_glock *gl)
816 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, just_schedule, TASK_UNINTERRUPTIBLE);
820 * handle_callback - process a demote request
822 * @state: the state the caller wants us to change to
824 * There are only two requests that we are going to see in actual
825 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
828 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
829 int remote, unsigned long delay)
831 int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
833 set_bit(bit, &gl->gl_flags);
834 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
835 gl->gl_demote_state = state;
836 gl->gl_demote_time = jiffies;
837 if (remote && gl->gl_ops->go_type == LM_TYPE_IOPEN &&
839 gfs2_glock_schedule_for_reclaim(gl);
840 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
841 gl->gl_demote_state != state) {
842 gl->gl_demote_state = LM_ST_UNLOCKED;
847 * gfs2_glock_wait - wait on a glock acquisition
848 * @gh: the glock holder
850 * Returns: 0 on success
853 int gfs2_glock_wait(struct gfs2_holder *gh)
859 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
865 struct gfs2_glock_iter *gi = seq->private;
866 vsprintf(gi->string, fmt, args);
867 seq_printf(seq, gi->string);
869 printk(KERN_ERR " ");
876 * add_to_queue - Add a holder to the wait queue (but look for recursion)
877 * @gh: the holder structure to add
879 * Eventually we should move the recursive locking trap to a
880 * debugging option or something like that. This is the fast
881 * path and needs to have the minimum number of distractions.
885 static inline void add_to_queue(struct gfs2_holder *gh)
886 __releases(&gl->gl_spin)
887 __acquires(&gl->gl_spin)
889 struct gfs2_glock *gl = gh->gh_gl;
890 struct gfs2_sbd *sdp = gl->gl_sbd;
891 struct list_head *insert_pt = NULL;
892 struct gfs2_holder *gh2;
895 BUG_ON(gh->gh_owner_pid == NULL);
896 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
899 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
900 if (test_bit(GLF_LOCK, &gl->gl_flags))
902 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
906 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
907 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
908 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
911 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) &&
912 !may_grant(gl, gh)) {
914 gh->gh_error = GLR_TRYFAILED;
915 gfs2_holder_wake(gh);
918 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
920 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
921 insert_pt = &gh2->gh_list;
923 if (likely(insert_pt == NULL)) {
924 list_add_tail(&gh->gh_list, &gl->gl_holders);
925 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
929 list_add_tail(&gh->gh_list, insert_pt);
931 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
932 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
933 spin_unlock(&gl->gl_spin);
934 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
935 sdp->sd_lockstruct.ls_ops->lm_cancel(gl->gl_lock);
936 spin_lock(&gl->gl_spin);
941 print_symbol(KERN_ERR "original: %s\n", gh2->gh_ip);
942 printk(KERN_ERR "pid: %d\n", pid_nr(gh2->gh_owner_pid));
943 printk(KERN_ERR "lock type: %d req lock state : %d\n",
944 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
945 print_symbol(KERN_ERR "new: %s\n", gh->gh_ip);
946 printk(KERN_ERR "pid: %d\n", pid_nr(gh->gh_owner_pid));
947 printk(KERN_ERR "lock type: %d req lock state : %d\n",
948 gh->gh_gl->gl_name.ln_type, gh->gh_state);
949 __dump_glock(NULL, gl);
954 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
955 * @gh: the holder structure
957 * if (gh->gh_flags & GL_ASYNC), this never returns an error
959 * Returns: 0, GLR_TRYFAILED, or errno on failure
962 int gfs2_glock_nq(struct gfs2_holder *gh)
964 struct gfs2_glock *gl = gh->gh_gl;
965 struct gfs2_sbd *sdp = gl->gl_sbd;
968 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
971 spin_lock(&gl->gl_spin);
974 spin_unlock(&gl->gl_spin);
976 if (!(gh->gh_flags & GL_ASYNC))
977 error = gfs2_glock_wait(gh);
983 * gfs2_glock_poll - poll to see if an async request has been completed
986 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
989 int gfs2_glock_poll(struct gfs2_holder *gh)
991 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
995 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
996 * @gh: the glock holder
1000 void gfs2_glock_dq(struct gfs2_holder *gh)
1002 struct gfs2_glock *gl = gh->gh_gl;
1003 const struct gfs2_glock_operations *glops = gl->gl_ops;
1007 spin_lock(&gl->gl_spin);
1008 if (gh->gh_flags & GL_NOCACHE)
1009 handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1011 list_del_init(&gh->gh_list);
1012 if (find_first_holder(gl) == NULL) {
1013 if (glops->go_unlock) {
1014 GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1015 spin_unlock(&gl->gl_spin);
1016 glops->go_unlock(gh);
1017 spin_lock(&gl->gl_spin);
1018 clear_bit(GLF_LOCK, &gl->gl_flags);
1020 gl->gl_stamp = jiffies;
1021 if (list_empty(&gl->gl_holders) &&
1022 !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1023 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1026 spin_unlock(&gl->gl_spin);
1027 if (likely(fast_path))
1030 gfs2_glock_hold(gl);
1031 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1032 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1033 delay = gl->gl_ops->go_min_hold_time;
1034 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1038 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1040 struct gfs2_glock *gl = gh->gh_gl;
1046 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1047 * @gh: the holder structure
1051 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1054 gfs2_holder_uninit(gh);
1058 * gfs2_glock_nq_num - acquire a glock based on lock number
1059 * @sdp: the filesystem
1060 * @number: the lock number
1061 * @glops: the glock operations for the type of glock
1062 * @state: the state to acquire the glock in
1063 * @flags: modifier flags for the aquisition
1064 * @gh: the struct gfs2_holder
1069 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1070 const struct gfs2_glock_operations *glops,
1071 unsigned int state, int flags, struct gfs2_holder *gh)
1073 struct gfs2_glock *gl;
1076 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1078 error = gfs2_glock_nq_init(gl, state, flags, gh);
1086 * glock_compare - Compare two struct gfs2_glock structures for sorting
1087 * @arg_a: the first structure
1088 * @arg_b: the second structure
1092 static int glock_compare(const void *arg_a, const void *arg_b)
1094 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1095 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1096 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1097 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1099 if (a->ln_number > b->ln_number)
1101 if (a->ln_number < b->ln_number)
1103 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1108 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1109 * @num_gh: the number of structures
1110 * @ghs: an array of struct gfs2_holder structures
1112 * Returns: 0 on success (all glocks acquired),
1113 * errno on failure (no glocks acquired)
1116 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1117 struct gfs2_holder **p)
1122 for (x = 0; x < num_gh; x++)
1125 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1127 for (x = 0; x < num_gh; x++) {
1128 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1130 error = gfs2_glock_nq(p[x]);
1133 gfs2_glock_dq(p[x]);
1142 * gfs2_glock_nq_m - acquire multiple glocks
1143 * @num_gh: the number of structures
1144 * @ghs: an array of struct gfs2_holder structures
1147 * Returns: 0 on success (all glocks acquired),
1148 * errno on failure (no glocks acquired)
1151 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1153 struct gfs2_holder *tmp[4];
1154 struct gfs2_holder **pph = tmp;
1161 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1162 return gfs2_glock_nq(ghs);
1166 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1171 error = nq_m_sync(num_gh, ghs, pph);
1180 * gfs2_glock_dq_m - release multiple glocks
1181 * @num_gh: the number of structures
1182 * @ghs: an array of struct gfs2_holder structures
1186 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1190 for (x = 0; x < num_gh; x++)
1191 gfs2_glock_dq(&ghs[x]);
1195 * gfs2_glock_dq_uninit_m - release multiple glocks
1196 * @num_gh: the number of structures
1197 * @ghs: an array of struct gfs2_holder structures
1201 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1205 for (x = 0; x < num_gh; x++)
1206 gfs2_glock_dq_uninit(&ghs[x]);
1209 static int gfs2_lm_hold_lvb(struct gfs2_sbd *sdp, void *lock, char **lvbp)
1212 if (!sdp->sd_lockstruct.ls_ops->lm_hold_lvb)
1214 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1215 error = sdp->sd_lockstruct.ls_ops->lm_hold_lvb(lock, lvbp);
1220 * gfs2_lvb_hold - attach a LVB from a glock
1221 * @gl: The glock in question
1225 int gfs2_lvb_hold(struct gfs2_glock *gl)
1229 if (!atomic_read(&gl->gl_lvb_count)) {
1230 error = gfs2_lm_hold_lvb(gl->gl_sbd, gl->gl_lock, &gl->gl_lvb);
1233 gfs2_glock_hold(gl);
1235 atomic_inc(&gl->gl_lvb_count);
1241 * gfs2_lvb_unhold - detach a LVB from a glock
1242 * @gl: The glock in question
1246 void gfs2_lvb_unhold(struct gfs2_glock *gl)
1248 struct gfs2_sbd *sdp = gl->gl_sbd;
1250 gfs2_glock_hold(gl);
1251 gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count) > 0);
1252 if (atomic_dec_and_test(&gl->gl_lvb_count)) {
1253 if (sdp->sd_lockstruct.ls_ops->lm_unhold_lvb)
1254 sdp->sd_lockstruct.ls_ops->lm_unhold_lvb(gl->gl_lock, gl->gl_lvb);
1261 static void blocking_cb(struct gfs2_sbd *sdp, struct lm_lockname *name,
1264 struct gfs2_glock *gl;
1265 unsigned long delay = 0;
1266 unsigned long holdtime;
1267 unsigned long now = jiffies;
1269 gl = gfs2_glock_find(sdp, name);
1273 holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
1274 if (time_before(now, holdtime))
1275 delay = holdtime - now;
1276 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1277 delay = gl->gl_ops->go_min_hold_time;
1279 spin_lock(&gl->gl_spin);
1280 handle_callback(gl, state, 1, delay);
1281 spin_unlock(&gl->gl_spin);
1282 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1287 * gfs2_glock_cb - Callback used by locking module
1288 * @sdp: Pointer to the superblock
1289 * @type: Type of callback
1290 * @data: Type dependent data pointer
1292 * Called by the locking module when it wants to tell us something.
1293 * Either we need to drop a lock, one of our ASYNC requests completed, or
1294 * a journal from another client needs to be recovered.
1297 void gfs2_glock_cb(void *cb_data, unsigned int type, void *data)
1299 struct gfs2_sbd *sdp = cb_data;
1303 blocking_cb(sdp, data, LM_ST_UNLOCKED);
1307 blocking_cb(sdp, data, LM_ST_DEFERRED);
1311 blocking_cb(sdp, data, LM_ST_SHARED);
1315 struct lm_async_cb *async = data;
1316 struct gfs2_glock *gl;
1318 down_read(&gfs2_umount_flush_sem);
1319 gl = gfs2_glock_find(sdp, &async->lc_name);
1320 if (gfs2_assert_warn(sdp, gl))
1322 gl->gl_reply = async->lc_ret;
1323 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1324 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1326 up_read(&gfs2_umount_flush_sem);
1330 case LM_CB_NEED_RECOVERY:
1331 gfs2_jdesc_make_dirty(sdp, *(unsigned int *)data);
1332 if (sdp->sd_recoverd_process)
1333 wake_up_process(sdp->sd_recoverd_process);
1337 gfs2_assert_warn(sdp, 0);
1343 * demote_ok - Check to see if it's ok to unlock a glock
1346 * Returns: 1 if it's ok
1349 static int demote_ok(struct gfs2_glock *gl)
1351 const struct gfs2_glock_operations *glops = gl->gl_ops;
1354 if (test_bit(GLF_STICKY, &gl->gl_flags))
1356 else if (glops->go_demote_ok)
1357 demote = glops->go_demote_ok(gl);
1363 * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1368 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
1370 struct gfs2_sbd *sdp = gl->gl_sbd;
1372 spin_lock(&sdp->sd_reclaim_lock);
1373 if (list_empty(&gl->gl_reclaim)) {
1374 gfs2_glock_hold(gl);
1375 list_add(&gl->gl_reclaim, &sdp->sd_reclaim_list);
1376 atomic_inc(&sdp->sd_reclaim_count);
1377 spin_unlock(&sdp->sd_reclaim_lock);
1378 wake_up(&sdp->sd_reclaim_wq);
1380 spin_unlock(&sdp->sd_reclaim_lock);
1384 * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1385 * @sdp: the filesystem
1387 * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1388 * different glock and we notice that there are a lot of glocks in the
1393 void gfs2_reclaim_glock(struct gfs2_sbd *sdp)
1395 struct gfs2_glock *gl;
1396 int done_callback = 0;
1398 spin_lock(&sdp->sd_reclaim_lock);
1399 if (list_empty(&sdp->sd_reclaim_list)) {
1400 spin_unlock(&sdp->sd_reclaim_lock);
1403 gl = list_entry(sdp->sd_reclaim_list.next,
1404 struct gfs2_glock, gl_reclaim);
1405 list_del_init(&gl->gl_reclaim);
1406 spin_unlock(&sdp->sd_reclaim_lock);
1408 atomic_dec(&sdp->sd_reclaim_count);
1409 atomic_inc(&sdp->sd_reclaimed);
1411 spin_lock(&gl->gl_spin);
1412 if (find_first_holder(gl) == NULL &&
1413 gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl)) {
1414 handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1417 spin_unlock(&gl->gl_spin);
1418 if (!done_callback ||
1419 queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1424 * examine_bucket - Call a function for glock in a hash bucket
1425 * @examiner: the function
1426 * @sdp: the filesystem
1427 * @bucket: the bucket
1429 * Returns: 1 if the bucket has entries
1432 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1435 struct gfs2_glock *gl, *prev = NULL;
1436 int has_entries = 0;
1437 struct hlist_head *head = &gl_hash_table[hash].hb_list;
1439 read_lock(gl_lock_addr(hash));
1440 /* Can't use hlist_for_each_entry - don't want prefetch here */
1441 if (hlist_empty(head))
1443 gl = list_entry(head->first, struct gfs2_glock, gl_list);
1445 if (!sdp || gl->gl_sbd == sdp) {
1446 gfs2_glock_hold(gl);
1447 read_unlock(gl_lock_addr(hash));
1449 gfs2_glock_put(prev);
1453 read_lock(gl_lock_addr(hash));
1455 if (gl->gl_list.next == NULL)
1457 gl = list_entry(gl->gl_list.next, struct gfs2_glock, gl_list);
1460 read_unlock(gl_lock_addr(hash));
1462 gfs2_glock_put(prev);
1468 * scan_glock - look at a glock and see if we can reclaim it
1469 * @gl: the glock to look at
1473 static void scan_glock(struct gfs2_glock *gl)
1475 if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object)
1477 if (test_bit(GLF_LOCK, &gl->gl_flags))
1480 spin_lock(&gl->gl_spin);
1481 if (find_first_holder(gl) == NULL &&
1482 gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1483 gfs2_glock_schedule_for_reclaim(gl);
1484 spin_unlock(&gl->gl_spin);
1488 * clear_glock - look at a glock and see if we can free it from glock cache
1489 * @gl: the glock to look at
1493 static void clear_glock(struct gfs2_glock *gl)
1495 struct gfs2_sbd *sdp = gl->gl_sbd;
1498 spin_lock(&sdp->sd_reclaim_lock);
1499 if (!list_empty(&gl->gl_reclaim)) {
1500 list_del_init(&gl->gl_reclaim);
1501 atomic_dec(&sdp->sd_reclaim_count);
1502 spin_unlock(&sdp->sd_reclaim_lock);
1503 released = gfs2_glock_put(gl);
1504 gfs2_assert(sdp, !released);
1506 spin_unlock(&sdp->sd_reclaim_lock);
1509 spin_lock(&gl->gl_spin);
1510 if (find_first_holder(gl) == NULL && gl->gl_state != LM_ST_UNLOCKED)
1511 handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1512 spin_unlock(&gl->gl_spin);
1513 gfs2_glock_hold(gl);
1514 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1519 * gfs2_gl_hash_clear - Empty out the glock hash table
1520 * @sdp: the filesystem
1521 * @wait: wait until it's all gone
1523 * Called when unmounting the filesystem.
1526 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1536 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1537 if (examine_bucket(clear_glock, sdp, x))
1544 if (time_after_eq(jiffies,
1545 t + gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
1546 fs_warn(sdp, "Unmount seems to be stalled. "
1547 "Dumping lock state...\n");
1548 gfs2_dump_lockstate(sdp);
1552 down_write(&gfs2_umount_flush_sem);
1553 invalidate_inodes(sdp->sd_vfs);
1554 up_write(&gfs2_umount_flush_sem);
1559 static const char *state2str(unsigned state)
1562 case LM_ST_UNLOCKED:
1566 case LM_ST_DEFERRED:
1568 case LM_ST_EXCLUSIVE:
1574 static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1577 if (flags & LM_FLAG_TRY)
1579 if (flags & LM_FLAG_TRY_1CB)
1581 if (flags & LM_FLAG_NOEXP)
1583 if (flags & LM_FLAG_ANY)
1585 if (flags & LM_FLAG_PRIORITY)
1587 if (flags & GL_ASYNC)
1589 if (flags & GL_EXACT)
1591 if (flags & GL_NOCACHE)
1593 if (test_bit(HIF_HOLDER, &iflags))
1595 if (test_bit(HIF_WAIT, &iflags))
1597 if (test_bit(HIF_FIRST, &iflags))
1604 * dump_holder - print information about a glock holder
1605 * @seq: the seq_file struct
1606 * @gh: the glock holder
1608 * Returns: 0 on success, -ENOBUFS when we run out of space
1611 static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1613 struct task_struct *gh_owner = NULL;
1614 char buffer[KSYM_SYMBOL_LEN];
1617 sprint_symbol(buffer, gh->gh_ip);
1618 if (gh->gh_owner_pid)
1619 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1620 gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %s\n",
1621 state2str(gh->gh_state),
1622 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1624 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1625 gh_owner ? gh_owner->comm : "(ended)", buffer);
1629 static const char *gflags2str(char *buf, const unsigned long *gflags)
1632 if (test_bit(GLF_LOCK, gflags))
1634 if (test_bit(GLF_STICKY, gflags))
1636 if (test_bit(GLF_DEMOTE, gflags))
1638 if (test_bit(GLF_PENDING_DEMOTE, gflags))
1640 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1642 if (test_bit(GLF_DIRTY, gflags))
1644 if (test_bit(GLF_LFLUSH, gflags))
1646 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1648 if (test_bit(GLF_REPLY_PENDING, gflags))
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 l:%d 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_lvb_count),
1692 atomic_read(&gl->gl_ail_count),
1693 atomic_read(&gl->gl_ref));
1695 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1696 error = dump_holder(seq, gh);
1700 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1701 error = glops->go_dump(seq, gl);
1706 static int dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1709 spin_lock(&gl->gl_spin);
1710 ret = __dump_glock(seq, gl);
1711 spin_unlock(&gl->gl_spin);
1716 * gfs2_dump_lockstate - print out the current lockstate
1717 * @sdp: the filesystem
1718 * @ub: the buffer to copy the information into
1720 * If @ub is NULL, dump the lockstate to the console.
1724 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
1726 struct gfs2_glock *gl;
1727 struct hlist_node *h;
1731 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1733 read_lock(gl_lock_addr(x));
1735 hlist_for_each_entry(gl, h, &gl_hash_table[x].hb_list, gl_list) {
1736 if (gl->gl_sbd != sdp)
1739 error = dump_glock(NULL, gl);
1744 read_unlock(gl_lock_addr(x));
1755 * gfs2_scand - Look for cached glocks and inodes to toss from memory
1756 * @sdp: Pointer to GFS2 superblock
1758 * One of these daemons runs, finding candidates to add to sd_reclaim_list.
1762 static int gfs2_scand(void *data)
1767 while (!kthread_should_stop()) {
1768 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1769 examine_bucket(scan_glock, NULL, x);
1770 if (freezing(current))
1775 schedule_timeout_interruptible(delay * HZ);
1783 int __init gfs2_glock_init(void)
1786 for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
1787 INIT_HLIST_HEAD(&gl_hash_table[i].hb_list);
1789 #ifdef GL_HASH_LOCK_SZ
1790 for(i = 0; i < GL_HASH_LOCK_SZ; i++) {
1791 rwlock_init(&gl_hash_locks[i]);
1795 scand_process = kthread_run(gfs2_scand, NULL, "gfs2_scand");
1796 if (IS_ERR(scand_process))
1797 return PTR_ERR(scand_process);
1799 glock_workqueue = create_workqueue("glock_workqueue");
1800 if (IS_ERR(glock_workqueue)) {
1801 kthread_stop(scand_process);
1802 return PTR_ERR(glock_workqueue);
1808 void gfs2_glock_exit(void)
1810 destroy_workqueue(glock_workqueue);
1811 kthread_stop(scand_process);
1814 module_param(scand_secs, uint, S_IRUGO|S_IWUSR);
1815 MODULE_PARM_DESC(scand_secs, "The number of seconds between scand runs");
1817 static int gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1819 struct gfs2_glock *gl;
1822 read_lock(gl_lock_addr(gi->hash));
1825 gi->gl = hlist_entry(gl->gl_list.next,
1826 struct gfs2_glock, gl_list);
1828 gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
1829 struct gfs2_glock, gl_list);
1832 gfs2_glock_hold(gi->gl);
1833 read_unlock(gl_lock_addr(gi->hash));
1836 while (gi->gl == NULL) {
1838 if (gi->hash >= GFS2_GL_HASH_SIZE)
1840 read_lock(gl_lock_addr(gi->hash));
1841 gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
1842 struct gfs2_glock, gl_list);
1844 gfs2_glock_hold(gi->gl);
1845 read_unlock(gl_lock_addr(gi->hash));
1848 if (gi->sdp != gi->gl->gl_sbd)
1854 static void gfs2_glock_iter_free(struct gfs2_glock_iter *gi)
1857 gfs2_glock_put(gi->gl);
1861 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1863 struct gfs2_glock_iter *gi = seq->private;
1869 if (gfs2_glock_iter_next(gi)) {
1870 gfs2_glock_iter_free(gi);
1878 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1881 struct gfs2_glock_iter *gi = seq->private;
1885 if (gfs2_glock_iter_next(gi)) {
1886 gfs2_glock_iter_free(gi);
1893 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1895 struct gfs2_glock_iter *gi = seq->private;
1896 gfs2_glock_iter_free(gi);
1899 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1901 return dump_glock(seq, iter_ptr);
1904 static const struct seq_operations gfs2_glock_seq_ops = {
1905 .start = gfs2_glock_seq_start,
1906 .next = gfs2_glock_seq_next,
1907 .stop = gfs2_glock_seq_stop,
1908 .show = gfs2_glock_seq_show,
1911 static int gfs2_debugfs_open(struct inode *inode, struct file *file)
1913 int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1914 sizeof(struct gfs2_glock_iter));
1916 struct seq_file *seq = file->private_data;
1917 struct gfs2_glock_iter *gi = seq->private;
1918 gi->sdp = inode->i_private;
1923 static const struct file_operations gfs2_debug_fops = {
1924 .owner = THIS_MODULE,
1925 .open = gfs2_debugfs_open,
1927 .llseek = seq_lseek,
1928 .release = seq_release_private,
1931 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
1933 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
1934 if (!sdp->debugfs_dir)
1936 sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
1938 sdp->debugfs_dir, sdp,
1940 if (!sdp->debugfs_dentry_glocks)
1946 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
1948 if (sdp && sdp->debugfs_dir) {
1949 if (sdp->debugfs_dentry_glocks) {
1950 debugfs_remove(sdp->debugfs_dentry_glocks);
1951 sdp->debugfs_dentry_glocks = NULL;
1953 debugfs_remove(sdp->debugfs_dir);
1954 sdp->debugfs_dir = NULL;
1958 int gfs2_register_debugfs(void)
1960 gfs2_root = debugfs_create_dir("gfs2", NULL);
1961 return gfs2_root ? 0 : -ENOMEM;
1964 void gfs2_unregister_debugfs(void)
1966 debugfs_remove(gfs2_root);