2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 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>
45 struct gfs2_gl_hash_bucket {
46 struct hlist_head hb_list;
50 int hash; /* hash bucket index */
51 struct gfs2_sbd *sdp; /* incore superblock */
52 struct gfs2_glock *gl; /* current glock struct */
53 struct seq_file *seq; /* sequence file for debugfs */
54 char string[512]; /* scratch space */
57 typedef void (*glock_examiner) (struct gfs2_glock * gl);
59 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp);
60 static int dump_glock(struct glock_iter *gi, struct gfs2_glock *gl);
61 static void gfs2_glock_xmote_th(struct gfs2_glock *gl, struct gfs2_holder *gh);
62 static void gfs2_glock_drop_th(struct gfs2_glock *gl);
63 static void run_queue(struct gfs2_glock *gl);
65 static DECLARE_RWSEM(gfs2_umount_flush_sem);
66 static struct dentry *gfs2_root;
67 static struct task_struct *scand_process;
68 static unsigned int scand_secs = 5;
69 static struct workqueue_struct *glock_workqueue;
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 * relaxed_state_ok - is a requested lock compatible with the current lock mode?
128 * @actual: the current state of the lock
129 * @requested: the lock state that was requested by the caller
130 * @flags: the modifier flags passed in by the caller
132 * Returns: 1 if the locks are compatible, 0 otherwise
135 static inline int relaxed_state_ok(unsigned int actual, unsigned requested,
138 if (actual == requested)
141 if (flags & GL_EXACT)
144 if (actual == LM_ST_EXCLUSIVE && requested == LM_ST_SHARED)
147 if (actual != LM_ST_UNLOCKED && (flags & LM_FLAG_ANY))
154 * gl_hash() - Turn glock number into hash bucket number
155 * @lock: The glock number
157 * Returns: The number of the corresponding hash bucket
160 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
161 const struct lm_lockname *name)
165 h = jhash(&name->ln_number, sizeof(u64), 0);
166 h = jhash(&name->ln_type, sizeof(unsigned int), h);
167 h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
168 h &= GFS2_GL_HASH_MASK;
174 * glock_free() - Perform a few checks and then release struct gfs2_glock
175 * @gl: The glock to release
177 * Also calls lock module to release its internal structure for this glock.
181 static void glock_free(struct gfs2_glock *gl)
183 struct gfs2_sbd *sdp = gl->gl_sbd;
184 struct inode *aspace = gl->gl_aspace;
186 gfs2_lm_put_lock(sdp, gl->gl_lock);
189 gfs2_aspace_put(aspace);
191 kmem_cache_free(gfs2_glock_cachep, gl);
195 * gfs2_glock_hold() - increment reference count on glock
196 * @gl: The glock to hold
200 void gfs2_glock_hold(struct gfs2_glock *gl)
202 atomic_inc(&gl->gl_ref);
206 * gfs2_glock_put() - Decrement reference count on glock
207 * @gl: The glock to put
211 int gfs2_glock_put(struct gfs2_glock *gl)
214 struct gfs2_sbd *sdp = gl->gl_sbd;
216 write_lock(gl_lock_addr(gl->gl_hash));
217 if (atomic_dec_and_test(&gl->gl_ref)) {
218 hlist_del(&gl->gl_list);
219 write_unlock(gl_lock_addr(gl->gl_hash));
220 BUG_ON(spin_is_locked(&gl->gl_spin));
221 gfs2_assert(sdp, gl->gl_state == LM_ST_UNLOCKED);
222 gfs2_assert(sdp, list_empty(&gl->gl_reclaim));
223 gfs2_assert(sdp, list_empty(&gl->gl_holders));
224 gfs2_assert(sdp, list_empty(&gl->gl_waiters1));
225 gfs2_assert(sdp, list_empty(&gl->gl_waiters3));
230 write_unlock(gl_lock_addr(gl->gl_hash));
236 * search_bucket() - Find struct gfs2_glock by lock number
237 * @bucket: the bucket to search
238 * @name: The lock name
240 * Returns: NULL, or the struct gfs2_glock with the requested number
243 static struct gfs2_glock *search_bucket(unsigned int hash,
244 const struct gfs2_sbd *sdp,
245 const struct lm_lockname *name)
247 struct gfs2_glock *gl;
248 struct hlist_node *h;
250 hlist_for_each_entry(gl, h, &gl_hash_table[hash].hb_list, gl_list) {
251 if (!lm_name_equal(&gl->gl_name, name))
253 if (gl->gl_sbd != sdp)
256 atomic_inc(&gl->gl_ref);
265 * gfs2_glock_find() - Find glock by lock number
266 * @sdp: The GFS2 superblock
267 * @name: The lock name
269 * Returns: NULL, or the struct gfs2_glock with the requested number
272 static struct gfs2_glock *gfs2_glock_find(const struct gfs2_sbd *sdp,
273 const struct lm_lockname *name)
275 unsigned int hash = gl_hash(sdp, name);
276 struct gfs2_glock *gl;
278 read_lock(gl_lock_addr(hash));
279 gl = search_bucket(hash, sdp, name);
280 read_unlock(gl_lock_addr(hash));
285 static void glock_work_func(struct work_struct *work)
287 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
289 spin_lock(&gl->gl_spin);
290 if (test_and_clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags))
291 set_bit(GLF_DEMOTE, &gl->gl_flags);
293 spin_unlock(&gl->gl_spin);
298 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
299 * @sdp: The GFS2 superblock
300 * @number: the lock number
301 * @glops: The glock_operations to use
302 * @create: If 0, don't create the glock if it doesn't exist
303 * @glp: the glock is returned here
305 * This does not lock a glock, just finds/creates structures for one.
310 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
311 const struct gfs2_glock_operations *glops, int create,
312 struct gfs2_glock **glp)
314 struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
315 struct gfs2_glock *gl, *tmp;
316 unsigned int hash = gl_hash(sdp, &name);
319 read_lock(gl_lock_addr(hash));
320 gl = search_bucket(hash, sdp, &name);
321 read_unlock(gl_lock_addr(hash));
328 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
334 atomic_set(&gl->gl_ref, 1);
335 gl->gl_state = LM_ST_UNLOCKED;
336 gl->gl_demote_state = LM_ST_EXCLUSIVE;
338 gl->gl_owner_pid = 0;
341 gl->gl_req_gh = NULL;
342 gl->gl_req_bh = NULL;
344 gl->gl_stamp = jiffies;
345 gl->gl_tchange = jiffies;
346 gl->gl_object = NULL;
348 gl->gl_aspace = NULL;
349 lops_init_le(&gl->gl_le, &gfs2_glock_lops);
350 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
352 /* If this glock protects actual on-disk data or metadata blocks,
353 create a VFS inode to manage the pages/buffers holding them. */
354 if (glops == &gfs2_inode_glops || glops == &gfs2_rgrp_glops) {
355 gl->gl_aspace = gfs2_aspace_get(sdp);
356 if (!gl->gl_aspace) {
362 error = gfs2_lm_get_lock(sdp, &name, &gl->gl_lock);
366 write_lock(gl_lock_addr(hash));
367 tmp = search_bucket(hash, sdp, &name);
369 write_unlock(gl_lock_addr(hash));
373 hlist_add_head(&gl->gl_list, &gl_hash_table[hash].hb_list);
374 write_unlock(gl_lock_addr(hash));
383 gfs2_aspace_put(gl->gl_aspace);
385 kmem_cache_free(gfs2_glock_cachep, gl);
390 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
392 * @state: the state we're requesting
393 * @flags: the modifier flags
394 * @gh: the holder structure
398 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
399 struct gfs2_holder *gh)
401 INIT_LIST_HEAD(&gh->gh_list);
403 gh->gh_ip = (unsigned long)__builtin_return_address(0);
404 gh->gh_owner_pid = current->pid;
405 gh->gh_state = state;
406 gh->gh_flags = flags;
413 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
414 * @state: the state we're requesting
415 * @flags: the modifier flags
416 * @gh: the holder structure
418 * Don't mess with the glock.
422 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
424 gh->gh_state = state;
425 gh->gh_flags = flags;
427 gh->gh_ip = (unsigned long)__builtin_return_address(0);
431 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
432 * @gh: the holder structure
436 void gfs2_holder_uninit(struct gfs2_holder *gh)
438 gfs2_glock_put(gh->gh_gl);
443 static void gfs2_holder_wake(struct gfs2_holder *gh)
445 clear_bit(HIF_WAIT, &gh->gh_iflags);
446 smp_mb__after_clear_bit();
447 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
450 static int just_schedule(void *word)
456 static void wait_on_holder(struct gfs2_holder *gh)
459 wait_on_bit(&gh->gh_iflags, HIF_WAIT, just_schedule, TASK_UNINTERRUPTIBLE);
462 static void gfs2_demote_wake(struct gfs2_glock *gl)
464 BUG_ON(!spin_is_locked(&gl->gl_spin));
465 gl->gl_demote_state = LM_ST_EXCLUSIVE;
466 clear_bit(GLF_DEMOTE, &gl->gl_flags);
467 smp_mb__after_clear_bit();
468 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
471 static void wait_on_demote(struct gfs2_glock *gl)
474 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, just_schedule, TASK_UNINTERRUPTIBLE);
478 * rq_mutex - process a mutex request in the queue
479 * @gh: the glock holder
481 * Returns: 1 if the queue is blocked
484 static int rq_mutex(struct gfs2_holder *gh)
486 struct gfs2_glock *gl = gh->gh_gl;
488 list_del_init(&gh->gh_list);
489 /* gh->gh_error never examined. */
490 set_bit(GLF_LOCK, &gl->gl_flags);
491 clear_bit(HIF_WAIT, &gh->gh_iflags);
493 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
499 * rq_promote - process a promote request in the queue
500 * @gh: the glock holder
502 * Acquire a new inter-node lock, or change a lock state to more restrictive.
504 * Returns: 1 if the queue is blocked
507 static int rq_promote(struct gfs2_holder *gh)
509 struct gfs2_glock *gl = gh->gh_gl;
510 struct gfs2_sbd *sdp = gl->gl_sbd;
512 if (!relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
513 if (list_empty(&gl->gl_holders)) {
515 set_bit(GLF_LOCK, &gl->gl_flags);
516 spin_unlock(&gl->gl_spin);
518 if (atomic_read(&sdp->sd_reclaim_count) >
519 gfs2_tune_get(sdp, gt_reclaim_limit) &&
520 !(gh->gh_flags & LM_FLAG_PRIORITY)) {
521 gfs2_reclaim_glock(sdp);
522 gfs2_reclaim_glock(sdp);
525 gfs2_glock_xmote_th(gh->gh_gl, gh);
526 spin_lock(&gl->gl_spin);
531 if (list_empty(&gl->gl_holders)) {
532 set_bit(HIF_FIRST, &gh->gh_iflags);
533 set_bit(GLF_LOCK, &gl->gl_flags);
535 struct gfs2_holder *next_gh;
536 if (gh->gh_state == LM_ST_EXCLUSIVE)
538 next_gh = list_entry(gl->gl_holders.next, struct gfs2_holder,
540 if (next_gh->gh_state == LM_ST_EXCLUSIVE)
544 list_move_tail(&gh->gh_list, &gl->gl_holders);
546 set_bit(HIF_HOLDER, &gh->gh_iflags);
548 gfs2_holder_wake(gh);
554 * rq_demote - process a demote request in the queue
555 * @gh: the glock holder
557 * Returns: 1 if the queue is blocked
560 static int rq_demote(struct gfs2_glock *gl)
562 if (!list_empty(&gl->gl_holders))
565 if (gl->gl_state == gl->gl_demote_state ||
566 gl->gl_state == LM_ST_UNLOCKED) {
567 gfs2_demote_wake(gl);
571 set_bit(GLF_LOCK, &gl->gl_flags);
572 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
574 if (gl->gl_demote_state == LM_ST_UNLOCKED ||
575 gl->gl_state != LM_ST_EXCLUSIVE) {
576 spin_unlock(&gl->gl_spin);
577 gfs2_glock_drop_th(gl);
579 spin_unlock(&gl->gl_spin);
580 gfs2_glock_xmote_th(gl, NULL);
583 spin_lock(&gl->gl_spin);
584 clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
590 * run_queue - process holder structures on a glock
594 static void run_queue(struct gfs2_glock *gl)
596 struct gfs2_holder *gh;
600 if (test_bit(GLF_LOCK, &gl->gl_flags))
603 if (!list_empty(&gl->gl_waiters1)) {
604 gh = list_entry(gl->gl_waiters1.next,
605 struct gfs2_holder, gh_list);
607 if (test_bit(HIF_MUTEX, &gh->gh_iflags))
608 blocked = rq_mutex(gh);
610 gfs2_assert_warn(gl->gl_sbd, 0);
612 } else if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
613 blocked = rq_demote(gl);
614 if (gl->gl_waiters2 && !blocked) {
615 set_bit(GLF_DEMOTE, &gl->gl_flags);
616 gl->gl_demote_state = LM_ST_UNLOCKED;
619 } else if (!list_empty(&gl->gl_waiters3)) {
620 gh = list_entry(gl->gl_waiters3.next,
621 struct gfs2_holder, gh_list);
623 if (test_bit(HIF_PROMOTE, &gh->gh_iflags))
624 blocked = rq_promote(gh);
626 gfs2_assert_warn(gl->gl_sbd, 0);
637 * gfs2_glmutex_lock - acquire a local lock on a glock
640 * Gives caller exclusive access to manipulate a glock structure.
643 static void gfs2_glmutex_lock(struct gfs2_glock *gl)
645 struct gfs2_holder gh;
647 gfs2_holder_init(gl, 0, 0, &gh);
648 set_bit(HIF_MUTEX, &gh.gh_iflags);
649 if (test_and_set_bit(HIF_WAIT, &gh.gh_iflags))
652 spin_lock(&gl->gl_spin);
653 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
654 list_add_tail(&gh.gh_list, &gl->gl_waiters1);
656 gl->gl_owner_pid = current->pid;
657 gl->gl_ip = (unsigned long)__builtin_return_address(0);
658 clear_bit(HIF_WAIT, &gh.gh_iflags);
660 wake_up_bit(&gh.gh_iflags, HIF_WAIT);
662 spin_unlock(&gl->gl_spin);
665 gfs2_holder_uninit(&gh);
669 * gfs2_glmutex_trylock - try to acquire a local lock on a glock
672 * Returns: 1 if the glock is acquired
675 static int gfs2_glmutex_trylock(struct gfs2_glock *gl)
679 spin_lock(&gl->gl_spin);
680 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
683 gl->gl_owner_pid = current->pid;
684 gl->gl_ip = (unsigned long)__builtin_return_address(0);
686 spin_unlock(&gl->gl_spin);
692 * gfs2_glmutex_unlock - release a local lock on a glock
697 static void gfs2_glmutex_unlock(struct gfs2_glock *gl)
699 spin_lock(&gl->gl_spin);
700 clear_bit(GLF_LOCK, &gl->gl_flags);
701 gl->gl_owner_pid = 0;
704 BUG_ON(!spin_is_locked(&gl->gl_spin));
705 spin_unlock(&gl->gl_spin);
709 * handle_callback - process a demote request
711 * @state: the state the caller wants us to change to
713 * There are only two requests that we are going to see in actual
714 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
717 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
718 int remote, unsigned long delay)
720 int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
722 spin_lock(&gl->gl_spin);
723 set_bit(bit, &gl->gl_flags);
724 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
725 gl->gl_demote_state = state;
726 gl->gl_demote_time = jiffies;
727 if (remote && gl->gl_ops->go_type == LM_TYPE_IOPEN &&
729 gfs2_glock_schedule_for_reclaim(gl);
730 spin_unlock(&gl->gl_spin);
733 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
734 gl->gl_demote_state != state) {
735 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
738 gl->gl_demote_state = LM_ST_UNLOCKED;
740 spin_unlock(&gl->gl_spin);
744 * state_change - record that the glock is now in a different state
746 * @new_state the new state
750 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
754 held1 = (gl->gl_state != LM_ST_UNLOCKED);
755 held2 = (new_state != LM_ST_UNLOCKED);
757 if (held1 != held2) {
764 gl->gl_state = new_state;
765 gl->gl_tchange = jiffies;
769 * xmote_bh - Called after the lock module is done acquiring a lock
770 * @gl: The glock in question
771 * @ret: the int returned from the lock module
775 static void xmote_bh(struct gfs2_glock *gl, unsigned int ret)
777 struct gfs2_sbd *sdp = gl->gl_sbd;
778 const struct gfs2_glock_operations *glops = gl->gl_ops;
779 struct gfs2_holder *gh = gl->gl_req_gh;
780 int prev_state = gl->gl_state;
783 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
784 gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
785 gfs2_assert_warn(sdp, !(ret & LM_OUT_ASYNC));
787 state_change(gl, ret & LM_OUT_ST_MASK);
789 if (prev_state != LM_ST_UNLOCKED && !(ret & LM_OUT_CACHEABLE)) {
791 glops->go_inval(gl, DIO_METADATA);
792 } else if (gl->gl_state == LM_ST_DEFERRED) {
793 /* We might not want to do this here.
794 Look at moving to the inode glops. */
796 glops->go_inval(gl, 0);
799 /* Deal with each possible exit condition */
802 gl->gl_stamp = jiffies;
803 if (ret & LM_OUT_CANCELED) {
806 spin_lock(&gl->gl_spin);
807 if (gl->gl_state != gl->gl_demote_state) {
808 gl->gl_req_bh = NULL;
809 spin_unlock(&gl->gl_spin);
810 gfs2_glock_drop_th(gl);
814 gfs2_demote_wake(gl);
815 spin_unlock(&gl->gl_spin);
818 spin_lock(&gl->gl_spin);
819 list_del_init(&gh->gh_list);
821 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
823 gh->gh_error = GLR_CANCELED;
824 if (ret & LM_OUT_CANCELED)
826 if (relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
827 list_add_tail(&gh->gh_list, &gl->gl_holders);
829 set_bit(HIF_HOLDER, &gh->gh_iflags);
830 set_bit(HIF_FIRST, &gh->gh_iflags);
834 gh->gh_error = GLR_TRYFAILED;
835 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
837 gh->gh_error = -EINVAL;
838 if (gfs2_assert_withdraw(sdp, 0) == -1)
839 fs_err(sdp, "ret = 0x%.8X\n", ret);
841 spin_unlock(&gl->gl_spin);
844 if (glops->go_xmote_bh)
845 glops->go_xmote_bh(gl);
848 spin_lock(&gl->gl_spin);
849 gl->gl_req_gh = NULL;
850 gl->gl_req_bh = NULL;
851 clear_bit(GLF_LOCK, &gl->gl_flags);
852 spin_unlock(&gl->gl_spin);
858 gfs2_holder_wake(gh);
862 * gfs2_glock_xmote_th - Call into the lock module to acquire or change a glock
863 * @gl: The glock in question
864 * @state: the requested state
865 * @flags: modifier flags to the lock call
869 static void gfs2_glock_xmote_th(struct gfs2_glock *gl, struct gfs2_holder *gh)
871 struct gfs2_sbd *sdp = gl->gl_sbd;
872 int flags = gh ? gh->gh_flags : 0;
873 unsigned state = gh ? gh->gh_state : gl->gl_demote_state;
874 const struct gfs2_glock_operations *glops = gl->gl_ops;
875 int lck_flags = flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB |
876 LM_FLAG_NOEXP | LM_FLAG_ANY |
878 unsigned int lck_ret;
880 if (glops->go_xmote_th)
881 glops->go_xmote_th(gl);
883 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
884 gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
885 gfs2_assert_warn(sdp, state != LM_ST_UNLOCKED);
886 gfs2_assert_warn(sdp, state != gl->gl_state);
889 gl->gl_req_bh = xmote_bh;
891 lck_ret = gfs2_lm_lock(sdp, gl->gl_lock, gl->gl_state, state, lck_flags);
893 if (gfs2_assert_withdraw(sdp, !(lck_ret & LM_OUT_ERROR)))
896 if (lck_ret & LM_OUT_ASYNC)
897 gfs2_assert_warn(sdp, lck_ret == LM_OUT_ASYNC);
899 xmote_bh(gl, lck_ret);
903 * drop_bh - Called after a lock module unlock completes
905 * @ret: the return status
907 * Doesn't wake up the process waiting on the struct gfs2_holder (if any)
908 * Doesn't drop the reference on the glock the top half took out
912 static void drop_bh(struct gfs2_glock *gl, unsigned int ret)
914 struct gfs2_sbd *sdp = gl->gl_sbd;
915 const struct gfs2_glock_operations *glops = gl->gl_ops;
916 struct gfs2_holder *gh = gl->gl_req_gh;
918 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
919 gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
920 gfs2_assert_warn(sdp, !ret);
922 state_change(gl, LM_ST_UNLOCKED);
925 glops->go_inval(gl, DIO_METADATA);
928 spin_lock(&gl->gl_spin);
929 list_del_init(&gh->gh_list);
931 spin_unlock(&gl->gl_spin);
934 spin_lock(&gl->gl_spin);
935 gfs2_demote_wake(gl);
936 gl->gl_req_gh = NULL;
937 gl->gl_req_bh = NULL;
938 clear_bit(GLF_LOCK, &gl->gl_flags);
939 spin_unlock(&gl->gl_spin);
944 gfs2_holder_wake(gh);
948 * gfs2_glock_drop_th - call into the lock module to unlock a lock
953 static void gfs2_glock_drop_th(struct gfs2_glock *gl)
955 struct gfs2_sbd *sdp = gl->gl_sbd;
956 const struct gfs2_glock_operations *glops = gl->gl_ops;
959 if (glops->go_drop_th)
960 glops->go_drop_th(gl);
962 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
963 gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
964 gfs2_assert_warn(sdp, gl->gl_state != LM_ST_UNLOCKED);
967 gl->gl_req_bh = drop_bh;
969 ret = gfs2_lm_unlock(sdp, gl->gl_lock, gl->gl_state);
971 if (gfs2_assert_withdraw(sdp, !(ret & LM_OUT_ERROR)))
977 gfs2_assert_warn(sdp, ret == LM_OUT_ASYNC);
981 * do_cancels - cancel requests for locks stuck waiting on an expire flag
982 * @gh: the LM_FLAG_PRIORITY holder waiting to acquire the lock
984 * Don't cancel GL_NOCANCEL requests.
987 static void do_cancels(struct gfs2_holder *gh)
989 struct gfs2_glock *gl = gh->gh_gl;
991 spin_lock(&gl->gl_spin);
993 while (gl->gl_req_gh != gh &&
994 !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
995 !list_empty(&gh->gh_list)) {
996 if (gl->gl_req_bh && !(gl->gl_req_gh &&
997 (gl->gl_req_gh->gh_flags & GL_NOCANCEL))) {
998 spin_unlock(&gl->gl_spin);
999 gfs2_lm_cancel(gl->gl_sbd, gl->gl_lock);
1001 spin_lock(&gl->gl_spin);
1003 spin_unlock(&gl->gl_spin);
1005 spin_lock(&gl->gl_spin);
1009 spin_unlock(&gl->gl_spin);
1013 * glock_wait_internal - wait on a glock acquisition
1014 * @gh: the glock holder
1016 * Returns: 0 on success
1019 static int glock_wait_internal(struct gfs2_holder *gh)
1021 struct gfs2_glock *gl = gh->gh_gl;
1022 struct gfs2_sbd *sdp = gl->gl_sbd;
1023 const struct gfs2_glock_operations *glops = gl->gl_ops;
1025 if (test_bit(HIF_ABORTED, &gh->gh_iflags))
1028 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1029 spin_lock(&gl->gl_spin);
1030 if (gl->gl_req_gh != gh &&
1031 !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1032 !list_empty(&gh->gh_list)) {
1033 list_del_init(&gh->gh_list);
1034 gh->gh_error = GLR_TRYFAILED;
1036 spin_unlock(&gl->gl_spin);
1037 return gh->gh_error;
1039 spin_unlock(&gl->gl_spin);
1042 if (gh->gh_flags & LM_FLAG_PRIORITY)
1047 return gh->gh_error;
1049 gfs2_assert_withdraw(sdp, test_bit(HIF_HOLDER, &gh->gh_iflags));
1050 gfs2_assert_withdraw(sdp, relaxed_state_ok(gl->gl_state, gh->gh_state,
1053 if (test_bit(HIF_FIRST, &gh->gh_iflags)) {
1054 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1056 if (glops->go_lock) {
1057 gh->gh_error = glops->go_lock(gh);
1059 spin_lock(&gl->gl_spin);
1060 list_del_init(&gh->gh_list);
1061 spin_unlock(&gl->gl_spin);
1065 spin_lock(&gl->gl_spin);
1066 gl->gl_req_gh = NULL;
1067 gl->gl_req_bh = NULL;
1068 clear_bit(GLF_LOCK, &gl->gl_flags);
1070 spin_unlock(&gl->gl_spin);
1073 return gh->gh_error;
1076 static inline struct gfs2_holder *
1077 find_holder_by_owner(struct list_head *head, pid_t pid)
1079 struct gfs2_holder *gh;
1081 list_for_each_entry(gh, head, gh_list) {
1082 if (gh->gh_owner_pid == pid)
1089 static void print_dbg(struct glock_iter *gi, const char *fmt, ...)
1093 va_start(args, fmt);
1095 vsprintf(gi->string, fmt, args);
1096 seq_printf(gi->seq, gi->string);
1104 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1105 * @gh: the holder structure to add
1109 static void add_to_queue(struct gfs2_holder *gh)
1111 struct gfs2_glock *gl = gh->gh_gl;
1112 struct gfs2_holder *existing;
1114 BUG_ON(!gh->gh_owner_pid);
1115 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1118 if (!(gh->gh_flags & GL_FLOCK)) {
1119 existing = find_holder_by_owner(&gl->gl_holders,
1122 print_symbol(KERN_WARNING "original: %s\n",
1124 printk(KERN_INFO "pid : %d\n", existing->gh_owner_pid);
1125 printk(KERN_INFO "lock type : %d lock state : %d\n",
1126 existing->gh_gl->gl_name.ln_type,
1127 existing->gh_gl->gl_state);
1128 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1129 printk(KERN_INFO "pid : %d\n", gh->gh_owner_pid);
1130 printk(KERN_INFO "lock type : %d lock state : %d\n",
1131 gl->gl_name.ln_type, gl->gl_state);
1135 existing = find_holder_by_owner(&gl->gl_waiters3,
1138 print_symbol(KERN_WARNING "original: %s\n",
1140 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1145 if (gh->gh_flags & LM_FLAG_PRIORITY)
1146 list_add(&gh->gh_list, &gl->gl_waiters3);
1148 list_add_tail(&gh->gh_list, &gl->gl_waiters3);
1152 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1153 * @gh: the holder structure
1155 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1157 * Returns: 0, GLR_TRYFAILED, or errno on failure
1160 int gfs2_glock_nq(struct gfs2_holder *gh)
1162 struct gfs2_glock *gl = gh->gh_gl;
1163 struct gfs2_sbd *sdp = gl->gl_sbd;
1167 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1168 set_bit(HIF_ABORTED, &gh->gh_iflags);
1172 set_bit(HIF_PROMOTE, &gh->gh_iflags);
1174 spin_lock(&gl->gl_spin);
1177 spin_unlock(&gl->gl_spin);
1179 if (!(gh->gh_flags & GL_ASYNC)) {
1180 error = glock_wait_internal(gh);
1181 if (error == GLR_CANCELED) {
1191 * gfs2_glock_poll - poll to see if an async request has been completed
1194 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1197 int gfs2_glock_poll(struct gfs2_holder *gh)
1199 struct gfs2_glock *gl = gh->gh_gl;
1202 spin_lock(&gl->gl_spin);
1204 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1206 else if (list_empty(&gh->gh_list)) {
1207 if (gh->gh_error == GLR_CANCELED) {
1208 spin_unlock(&gl->gl_spin);
1210 if (gfs2_glock_nq(gh))
1217 spin_unlock(&gl->gl_spin);
1223 * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
1224 * @gh: the holder structure
1226 * Returns: 0, GLR_TRYFAILED, or errno on failure
1229 int gfs2_glock_wait(struct gfs2_holder *gh)
1233 error = glock_wait_internal(gh);
1234 if (error == GLR_CANCELED) {
1236 gh->gh_flags &= ~GL_ASYNC;
1237 error = gfs2_glock_nq(gh);
1244 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1245 * @gh: the glock holder
1249 void gfs2_glock_dq(struct gfs2_holder *gh)
1251 struct gfs2_glock *gl = gh->gh_gl;
1252 const struct gfs2_glock_operations *glops = gl->gl_ops;
1255 if (gh->gh_flags & GL_NOCACHE)
1256 handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1258 gfs2_glmutex_lock(gl);
1260 spin_lock(&gl->gl_spin);
1261 list_del_init(&gh->gh_list);
1263 if (list_empty(&gl->gl_holders)) {
1264 spin_unlock(&gl->gl_spin);
1266 if (glops->go_unlock)
1267 glops->go_unlock(gh);
1269 spin_lock(&gl->gl_spin);
1270 gl->gl_stamp = jiffies;
1273 clear_bit(GLF_LOCK, &gl->gl_flags);
1274 spin_unlock(&gl->gl_spin);
1276 gfs2_glock_hold(gl);
1277 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1278 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1279 delay = gl->gl_ops->go_min_hold_time;
1280 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1284 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1286 struct gfs2_glock *gl = gh->gh_gl;
1292 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1293 * @gh: the holder structure
1297 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1300 gfs2_holder_uninit(gh);
1304 * gfs2_glock_nq_num - acquire a glock based on lock number
1305 * @sdp: the filesystem
1306 * @number: the lock number
1307 * @glops: the glock operations for the type of glock
1308 * @state: the state to acquire the glock in
1309 * @flags: modifier flags for the aquisition
1310 * @gh: the struct gfs2_holder
1315 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1316 const struct gfs2_glock_operations *glops,
1317 unsigned int state, int flags, struct gfs2_holder *gh)
1319 struct gfs2_glock *gl;
1322 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1324 error = gfs2_glock_nq_init(gl, state, flags, gh);
1332 * glock_compare - Compare two struct gfs2_glock structures for sorting
1333 * @arg_a: the first structure
1334 * @arg_b: the second structure
1338 static int glock_compare(const void *arg_a, const void *arg_b)
1340 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1341 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1342 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1343 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1345 if (a->ln_number > b->ln_number)
1347 if (a->ln_number < b->ln_number)
1349 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1354 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1355 * @num_gh: the number of structures
1356 * @ghs: an array of struct gfs2_holder structures
1358 * Returns: 0 on success (all glocks acquired),
1359 * errno on failure (no glocks acquired)
1362 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1363 struct gfs2_holder **p)
1368 for (x = 0; x < num_gh; x++)
1371 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1373 for (x = 0; x < num_gh; x++) {
1374 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1376 error = gfs2_glock_nq(p[x]);
1379 gfs2_glock_dq(p[x]);
1388 * gfs2_glock_nq_m - acquire multiple glocks
1389 * @num_gh: the number of structures
1390 * @ghs: an array of struct gfs2_holder structures
1393 * Returns: 0 on success (all glocks acquired),
1394 * errno on failure (no glocks acquired)
1397 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1399 struct gfs2_holder *tmp[4];
1400 struct gfs2_holder **pph = tmp;
1407 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1408 return gfs2_glock_nq(ghs);
1412 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1417 error = nq_m_sync(num_gh, ghs, pph);
1426 * gfs2_glock_dq_m - release multiple glocks
1427 * @num_gh: the number of structures
1428 * @ghs: an array of struct gfs2_holder structures
1432 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1436 for (x = 0; x < num_gh; x++)
1437 gfs2_glock_dq(&ghs[x]);
1441 * gfs2_glock_dq_uninit_m - release multiple glocks
1442 * @num_gh: the number of structures
1443 * @ghs: an array of struct gfs2_holder structures
1447 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1451 for (x = 0; x < num_gh; x++)
1452 gfs2_glock_dq_uninit(&ghs[x]);
1456 * gfs2_lvb_hold - attach a LVB from a glock
1457 * @gl: The glock in question
1461 int gfs2_lvb_hold(struct gfs2_glock *gl)
1465 gfs2_glmutex_lock(gl);
1467 if (!atomic_read(&gl->gl_lvb_count)) {
1468 error = gfs2_lm_hold_lvb(gl->gl_sbd, gl->gl_lock, &gl->gl_lvb);
1470 gfs2_glmutex_unlock(gl);
1473 gfs2_glock_hold(gl);
1475 atomic_inc(&gl->gl_lvb_count);
1477 gfs2_glmutex_unlock(gl);
1483 * gfs2_lvb_unhold - detach a LVB from a glock
1484 * @gl: The glock in question
1488 void gfs2_lvb_unhold(struct gfs2_glock *gl)
1490 gfs2_glock_hold(gl);
1491 gfs2_glmutex_lock(gl);
1493 gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count) > 0);
1494 if (atomic_dec_and_test(&gl->gl_lvb_count)) {
1495 gfs2_lm_unhold_lvb(gl->gl_sbd, gl->gl_lock, gl->gl_lvb);
1500 gfs2_glmutex_unlock(gl);
1504 static void blocking_cb(struct gfs2_sbd *sdp, struct lm_lockname *name,
1507 struct gfs2_glock *gl;
1508 unsigned long delay = 0;
1509 unsigned long holdtime;
1510 unsigned long now = jiffies;
1512 gl = gfs2_glock_find(sdp, name);
1516 holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
1517 if (time_before(now, holdtime))
1518 delay = holdtime - now;
1520 handle_callback(gl, state, 1, delay);
1521 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1526 * gfs2_glock_cb - Callback used by locking module
1527 * @sdp: Pointer to the superblock
1528 * @type: Type of callback
1529 * @data: Type dependent data pointer
1531 * Called by the locking module when it wants to tell us something.
1532 * Either we need to drop a lock, one of our ASYNC requests completed, or
1533 * a journal from another client needs to be recovered.
1536 void gfs2_glock_cb(void *cb_data, unsigned int type, void *data)
1538 struct gfs2_sbd *sdp = cb_data;
1542 blocking_cb(sdp, data, LM_ST_UNLOCKED);
1546 blocking_cb(sdp, data, LM_ST_DEFERRED);
1550 blocking_cb(sdp, data, LM_ST_SHARED);
1554 struct lm_async_cb *async = data;
1555 struct gfs2_glock *gl;
1557 down_read(&gfs2_umount_flush_sem);
1558 gl = gfs2_glock_find(sdp, &async->lc_name);
1559 if (gfs2_assert_warn(sdp, gl))
1561 if (!gfs2_assert_warn(sdp, gl->gl_req_bh))
1562 gl->gl_req_bh(gl, async->lc_ret);
1563 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1565 up_read(&gfs2_umount_flush_sem);
1569 case LM_CB_NEED_RECOVERY:
1570 gfs2_jdesc_make_dirty(sdp, *(unsigned int *)data);
1571 if (sdp->sd_recoverd_process)
1572 wake_up_process(sdp->sd_recoverd_process);
1575 case LM_CB_DROPLOCKS:
1576 gfs2_gl_hash_clear(sdp, NO_WAIT);
1577 gfs2_quota_scan(sdp);
1581 gfs2_assert_warn(sdp, 0);
1587 * demote_ok - Check to see if it's ok to unlock a glock
1590 * Returns: 1 if it's ok
1593 static int demote_ok(struct gfs2_glock *gl)
1595 const struct gfs2_glock_operations *glops = gl->gl_ops;
1598 if (test_bit(GLF_STICKY, &gl->gl_flags))
1600 else if (glops->go_demote_ok)
1601 demote = glops->go_demote_ok(gl);
1607 * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1612 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
1614 struct gfs2_sbd *sdp = gl->gl_sbd;
1616 spin_lock(&sdp->sd_reclaim_lock);
1617 if (list_empty(&gl->gl_reclaim)) {
1618 gfs2_glock_hold(gl);
1619 list_add(&gl->gl_reclaim, &sdp->sd_reclaim_list);
1620 atomic_inc(&sdp->sd_reclaim_count);
1622 spin_unlock(&sdp->sd_reclaim_lock);
1624 wake_up(&sdp->sd_reclaim_wq);
1628 * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1629 * @sdp: the filesystem
1631 * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1632 * different glock and we notice that there are a lot of glocks in the
1637 void gfs2_reclaim_glock(struct gfs2_sbd *sdp)
1639 struct gfs2_glock *gl;
1641 spin_lock(&sdp->sd_reclaim_lock);
1642 if (list_empty(&sdp->sd_reclaim_list)) {
1643 spin_unlock(&sdp->sd_reclaim_lock);
1646 gl = list_entry(sdp->sd_reclaim_list.next,
1647 struct gfs2_glock, gl_reclaim);
1648 list_del_init(&gl->gl_reclaim);
1649 spin_unlock(&sdp->sd_reclaim_lock);
1651 atomic_dec(&sdp->sd_reclaim_count);
1652 atomic_inc(&sdp->sd_reclaimed);
1654 if (gfs2_glmutex_trylock(gl)) {
1655 if (list_empty(&gl->gl_holders) &&
1656 gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1657 handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1658 gfs2_glmutex_unlock(gl);
1665 * examine_bucket - Call a function for glock in a hash bucket
1666 * @examiner: the function
1667 * @sdp: the filesystem
1668 * @bucket: the bucket
1670 * Returns: 1 if the bucket has entries
1673 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1676 struct gfs2_glock *gl, *prev = NULL;
1677 int has_entries = 0;
1678 struct hlist_head *head = &gl_hash_table[hash].hb_list;
1680 read_lock(gl_lock_addr(hash));
1681 /* Can't use hlist_for_each_entry - don't want prefetch here */
1682 if (hlist_empty(head))
1684 gl = list_entry(head->first, struct gfs2_glock, gl_list);
1686 if (!sdp || gl->gl_sbd == sdp) {
1687 gfs2_glock_hold(gl);
1688 read_unlock(gl_lock_addr(hash));
1690 gfs2_glock_put(prev);
1694 read_lock(gl_lock_addr(hash));
1696 if (gl->gl_list.next == NULL)
1698 gl = list_entry(gl->gl_list.next, struct gfs2_glock, gl_list);
1701 read_unlock(gl_lock_addr(hash));
1703 gfs2_glock_put(prev);
1709 * scan_glock - look at a glock and see if we can reclaim it
1710 * @gl: the glock to look at
1714 static void scan_glock(struct gfs2_glock *gl)
1716 if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object)
1719 if (gfs2_glmutex_trylock(gl)) {
1720 if (list_empty(&gl->gl_holders) &&
1721 gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1723 gfs2_glmutex_unlock(gl);
1728 gfs2_glmutex_unlock(gl);
1729 gfs2_glock_schedule_for_reclaim(gl);
1733 * clear_glock - look at a glock and see if we can free it from glock cache
1734 * @gl: the glock to look at
1738 static void clear_glock(struct gfs2_glock *gl)
1740 struct gfs2_sbd *sdp = gl->gl_sbd;
1743 spin_lock(&sdp->sd_reclaim_lock);
1744 if (!list_empty(&gl->gl_reclaim)) {
1745 list_del_init(&gl->gl_reclaim);
1746 atomic_dec(&sdp->sd_reclaim_count);
1747 spin_unlock(&sdp->sd_reclaim_lock);
1748 released = gfs2_glock_put(gl);
1749 gfs2_assert(sdp, !released);
1751 spin_unlock(&sdp->sd_reclaim_lock);
1754 if (gfs2_glmutex_trylock(gl)) {
1755 if (list_empty(&gl->gl_holders) &&
1756 gl->gl_state != LM_ST_UNLOCKED)
1757 handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1758 gfs2_glmutex_unlock(gl);
1763 * gfs2_gl_hash_clear - Empty out the glock hash table
1764 * @sdp: the filesystem
1765 * @wait: wait until it's all gone
1767 * Called when unmounting the filesystem, or when inter-node lock manager
1768 * requests DROPLOCKS because it is running out of capacity.
1771 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp, int wait)
1781 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1782 if (examine_bucket(clear_glock, sdp, x))
1789 if (time_after_eq(jiffies,
1790 t + gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
1791 fs_warn(sdp, "Unmount seems to be stalled. "
1792 "Dumping lock state...\n");
1793 gfs2_dump_lockstate(sdp);
1797 down_write(&gfs2_umount_flush_sem);
1798 invalidate_inodes(sdp->sd_vfs);
1799 up_write(&gfs2_umount_flush_sem);
1805 * Diagnostic routines to help debug distributed deadlock
1808 static void gfs2_print_symbol(struct glock_iter *gi, const char *fmt,
1809 unsigned long address)
1811 char buffer[KSYM_SYMBOL_LEN];
1813 sprint_symbol(buffer, address);
1814 print_dbg(gi, fmt, buffer);
1818 * dump_holder - print information about a glock holder
1819 * @str: a string naming the type of holder
1820 * @gh: the glock holder
1822 * Returns: 0 on success, -ENOBUFS when we run out of space
1825 static int dump_holder(struct glock_iter *gi, char *str,
1826 struct gfs2_holder *gh)
1829 struct task_struct *gh_owner;
1831 print_dbg(gi, " %s\n", str);
1832 if (gh->gh_owner_pid) {
1833 print_dbg(gi, " owner = %ld ", (long)gh->gh_owner_pid);
1834 gh_owner = find_task_by_pid(gh->gh_owner_pid);
1836 print_dbg(gi, "(%s)\n", gh_owner->comm);
1838 print_dbg(gi, "(ended)\n");
1840 print_dbg(gi, " owner = -1\n");
1841 print_dbg(gi, " gh_state = %u\n", gh->gh_state);
1842 print_dbg(gi, " gh_flags =");
1843 for (x = 0; x < 32; x++)
1844 if (gh->gh_flags & (1 << x))
1845 print_dbg(gi, " %u", x);
1846 print_dbg(gi, " \n");
1847 print_dbg(gi, " error = %d\n", gh->gh_error);
1848 print_dbg(gi, " gh_iflags =");
1849 for (x = 0; x < 32; x++)
1850 if (test_bit(x, &gh->gh_iflags))
1851 print_dbg(gi, " %u", x);
1852 print_dbg(gi, " \n");
1853 gfs2_print_symbol(gi, " initialized at: %s\n", gh->gh_ip);
1859 * dump_inode - print information about an inode
1862 * Returns: 0 on success, -ENOBUFS when we run out of space
1865 static int dump_inode(struct glock_iter *gi, struct gfs2_inode *ip)
1869 print_dbg(gi, " Inode:\n");
1870 print_dbg(gi, " num = %llu/%llu\n",
1871 (unsigned long long)ip->i_no_formal_ino,
1872 (unsigned long long)ip->i_no_addr);
1873 print_dbg(gi, " type = %u\n", IF2DT(ip->i_inode.i_mode));
1874 print_dbg(gi, " i_flags =");
1875 for (x = 0; x < 32; x++)
1876 if (test_bit(x, &ip->i_flags))
1877 print_dbg(gi, " %u", x);
1878 print_dbg(gi, " \n");
1883 * dump_glock - print information about a glock
1885 * @count: where we are in the buffer
1887 * Returns: 0 on success, -ENOBUFS when we run out of space
1890 static int dump_glock(struct glock_iter *gi, struct gfs2_glock *gl)
1892 struct gfs2_holder *gh;
1894 int error = -ENOBUFS;
1895 struct task_struct *gl_owner;
1897 spin_lock(&gl->gl_spin);
1899 print_dbg(gi, "Glock 0x%p (%u, 0x%llx)\n", gl, gl->gl_name.ln_type,
1900 (unsigned long long)gl->gl_name.ln_number);
1901 print_dbg(gi, " gl_flags =");
1902 for (x = 0; x < 32; x++) {
1903 if (test_bit(x, &gl->gl_flags))
1904 print_dbg(gi, " %u", x);
1906 if (!test_bit(GLF_LOCK, &gl->gl_flags))
1907 print_dbg(gi, " (unlocked)");
1908 print_dbg(gi, " \n");
1909 print_dbg(gi, " gl_ref = %d\n", atomic_read(&gl->gl_ref));
1910 print_dbg(gi, " gl_state = %u\n", gl->gl_state);
1911 if (gl->gl_owner_pid) {
1912 gl_owner = find_task_by_pid(gl->gl_owner_pid);
1914 print_dbg(gi, " gl_owner = pid %d (%s)\n",
1915 gl->gl_owner_pid, gl_owner->comm);
1917 print_dbg(gi, " gl_owner = %d (ended)\n",
1920 print_dbg(gi, " gl_owner = -1\n");
1921 print_dbg(gi, " gl_ip = %lu\n", gl->gl_ip);
1922 print_dbg(gi, " req_gh = %s\n", (gl->gl_req_gh) ? "yes" : "no");
1923 print_dbg(gi, " req_bh = %s\n", (gl->gl_req_bh) ? "yes" : "no");
1924 print_dbg(gi, " lvb_count = %d\n", atomic_read(&gl->gl_lvb_count));
1925 print_dbg(gi, " object = %s\n", (gl->gl_object) ? "yes" : "no");
1926 print_dbg(gi, " le = %s\n",
1927 (list_empty(&gl->gl_le.le_list)) ? "no" : "yes");
1928 print_dbg(gi, " reclaim = %s\n",
1929 (list_empty(&gl->gl_reclaim)) ? "no" : "yes");
1931 print_dbg(gi, " aspace = 0x%p nrpages = %lu\n", gl->gl_aspace,
1932 gl->gl_aspace->i_mapping->nrpages);
1934 print_dbg(gi, " aspace = no\n");
1935 print_dbg(gi, " ail = %d\n", atomic_read(&gl->gl_ail_count));
1936 if (gl->gl_req_gh) {
1937 error = dump_holder(gi, "Request", gl->gl_req_gh);
1941 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1942 error = dump_holder(gi, "Holder", gh);
1946 list_for_each_entry(gh, &gl->gl_waiters1, gh_list) {
1947 error = dump_holder(gi, "Waiter1", gh);
1951 list_for_each_entry(gh, &gl->gl_waiters3, gh_list) {
1952 error = dump_holder(gi, "Waiter3", gh);
1956 if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
1957 print_dbg(gi, " Demotion req to state %u (%llu uS ago)\n",
1958 gl->gl_demote_state, (unsigned long long)
1959 (jiffies - gl->gl_demote_time)*(1000000/HZ));
1961 if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object) {
1962 if (!test_bit(GLF_LOCK, &gl->gl_flags) &&
1963 list_empty(&gl->gl_holders)) {
1964 error = dump_inode(gi, gl->gl_object);
1969 print_dbg(gi, " Inode: busy\n");
1976 spin_unlock(&gl->gl_spin);
1981 * gfs2_dump_lockstate - print out the current lockstate
1982 * @sdp: the filesystem
1983 * @ub: the buffer to copy the information into
1985 * If @ub is NULL, dump the lockstate to the console.
1989 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
1991 struct gfs2_glock *gl;
1992 struct hlist_node *h;
1996 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1998 read_lock(gl_lock_addr(x));
2000 hlist_for_each_entry(gl, h, &gl_hash_table[x].hb_list, gl_list) {
2001 if (gl->gl_sbd != sdp)
2004 error = dump_glock(NULL, gl);
2009 read_unlock(gl_lock_addr(x));
2020 * gfs2_scand - Look for cached glocks and inodes to toss from memory
2021 * @sdp: Pointer to GFS2 superblock
2023 * One of these daemons runs, finding candidates to add to sd_reclaim_list.
2027 static int gfs2_scand(void *data)
2032 while (!kthread_should_stop()) {
2033 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
2034 examine_bucket(scan_glock, NULL, x);
2035 if (freezing(current))
2040 schedule_timeout_interruptible(delay * HZ);
2048 int __init gfs2_glock_init(void)
2051 for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
2052 INIT_HLIST_HEAD(&gl_hash_table[i].hb_list);
2054 #ifdef GL_HASH_LOCK_SZ
2055 for(i = 0; i < GL_HASH_LOCK_SZ; i++) {
2056 rwlock_init(&gl_hash_locks[i]);
2060 scand_process = kthread_run(gfs2_scand, NULL, "gfs2_scand");
2061 if (IS_ERR(scand_process))
2062 return PTR_ERR(scand_process);
2064 glock_workqueue = create_workqueue("glock_workqueue");
2065 if (IS_ERR(glock_workqueue)) {
2066 kthread_stop(scand_process);
2067 return PTR_ERR(glock_workqueue);
2073 void gfs2_glock_exit(void)
2075 destroy_workqueue(glock_workqueue);
2076 kthread_stop(scand_process);
2079 module_param(scand_secs, uint, S_IRUGO|S_IWUSR);
2080 MODULE_PARM_DESC(scand_secs, "The number of seconds between scand runs");
2082 static int gfs2_glock_iter_next(struct glock_iter *gi)
2084 struct gfs2_glock *gl;
2087 read_lock(gl_lock_addr(gi->hash));
2090 gi->gl = hlist_entry(gl->gl_list.next,
2091 struct gfs2_glock, gl_list);
2093 gfs2_glock_hold(gi->gl);
2095 read_unlock(gl_lock_addr(gi->hash));
2098 if (gl && gi->gl == NULL)
2100 while(gi->gl == NULL) {
2101 if (gi->hash >= GFS2_GL_HASH_SIZE)
2103 read_lock(gl_lock_addr(gi->hash));
2104 gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
2105 struct gfs2_glock, gl_list);
2107 gfs2_glock_hold(gi->gl);
2108 read_unlock(gl_lock_addr(gi->hash));
2112 if (gi->sdp != gi->gl->gl_sbd)
2118 static void gfs2_glock_iter_free(struct glock_iter *gi)
2121 gfs2_glock_put(gi->gl);
2125 static struct glock_iter *gfs2_glock_iter_init(struct gfs2_sbd *sdp)
2127 struct glock_iter *gi;
2129 gi = kmalloc(sizeof (*gi), GFP_KERNEL);
2137 memset(gi->string, 0, sizeof(gi->string));
2139 if (gfs2_glock_iter_next(gi)) {
2140 gfs2_glock_iter_free(gi);
2147 static void *gfs2_glock_seq_start(struct seq_file *file, loff_t *pos)
2149 struct glock_iter *gi;
2152 gi = gfs2_glock_iter_init(file->private);
2157 if (gfs2_glock_iter_next(gi)) {
2158 gfs2_glock_iter_free(gi);
2166 static void *gfs2_glock_seq_next(struct seq_file *file, void *iter_ptr,
2169 struct glock_iter *gi = iter_ptr;
2173 if (gfs2_glock_iter_next(gi)) {
2174 gfs2_glock_iter_free(gi);
2181 static void gfs2_glock_seq_stop(struct seq_file *file, void *iter_ptr)
2183 struct glock_iter *gi = iter_ptr;
2185 gfs2_glock_iter_free(gi);
2188 static int gfs2_glock_seq_show(struct seq_file *file, void *iter_ptr)
2190 struct glock_iter *gi = iter_ptr;
2193 dump_glock(gi, gi->gl);
2198 static const struct seq_operations gfs2_glock_seq_ops = {
2199 .start = gfs2_glock_seq_start,
2200 .next = gfs2_glock_seq_next,
2201 .stop = gfs2_glock_seq_stop,
2202 .show = gfs2_glock_seq_show,
2205 static int gfs2_debugfs_open(struct inode *inode, struct file *file)
2207 struct seq_file *seq;
2210 ret = seq_open(file, &gfs2_glock_seq_ops);
2214 seq = file->private_data;
2215 seq->private = inode->i_private;
2220 static const struct file_operations gfs2_debug_fops = {
2221 .owner = THIS_MODULE,
2222 .open = gfs2_debugfs_open,
2224 .llseek = seq_lseek,
2225 .release = seq_release
2228 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2230 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2231 if (!sdp->debugfs_dir)
2233 sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
2235 sdp->debugfs_dir, sdp,
2237 if (!sdp->debugfs_dentry_glocks)
2243 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2245 if (sdp && sdp->debugfs_dir) {
2246 if (sdp->debugfs_dentry_glocks) {
2247 debugfs_remove(sdp->debugfs_dentry_glocks);
2248 sdp->debugfs_dentry_glocks = NULL;
2250 debugfs_remove(sdp->debugfs_dir);
2251 sdp->debugfs_dir = NULL;
2255 int gfs2_register_debugfs(void)
2257 gfs2_root = debugfs_create_dir("gfs2", NULL);
2258 return gfs2_root ? 0 : -ENOMEM;
2261 void gfs2_unregister_debugfs(void)
2263 debugfs_remove(gfs2_root);