2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2007 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;
49 int hash; /* hash bucket index */
50 struct gfs2_sbd *sdp; /* incore superblock */
51 struct gfs2_glock *gl; /* current glock struct */
52 struct seq_file *seq; /* sequence file for debugfs */
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 glock_iter *gi, struct gfs2_glock *gl);
60 static void gfs2_glock_xmote_th(struct gfs2_glock *gl, struct gfs2_holder *gh);
61 static void gfs2_glock_drop_th(struct gfs2_glock *gl);
62 static void run_queue(struct gfs2_glock *gl);
64 static DECLARE_RWSEM(gfs2_umount_flush_sem);
65 static struct dentry *gfs2_root;
66 static struct task_struct *scand_process;
67 static unsigned int scand_secs = 5;
68 static struct workqueue_struct *glock_workqueue;
70 #define GFS2_GL_HASH_SHIFT 15
71 #define GFS2_GL_HASH_SIZE (1 << GFS2_GL_HASH_SHIFT)
72 #define GFS2_GL_HASH_MASK (GFS2_GL_HASH_SIZE - 1)
74 static struct gfs2_gl_hash_bucket gl_hash_table[GFS2_GL_HASH_SIZE];
75 static struct dentry *gfs2_root;
78 * Despite what you might think, the numbers below are not arbitrary :-)
79 * They are taken from the ipv4 routing hash code, which is well tested
80 * and thus should be nearly optimal. Later on we might tweek the numbers
81 * but for now this should be fine.
83 * The reason for putting the locks in a separate array from the list heads
84 * is that we can have fewer locks than list heads and save memory. We use
85 * the same hash function for both, but with a different hash mask.
87 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
88 defined(CONFIG_PROVE_LOCKING)
91 # define GL_HASH_LOCK_SZ 256
94 # define GL_HASH_LOCK_SZ 4096
96 # define GL_HASH_LOCK_SZ 2048
98 # define GL_HASH_LOCK_SZ 1024
100 # define GL_HASH_LOCK_SZ 512
102 # define GL_HASH_LOCK_SZ 256
106 /* We never want more locks than chains */
107 #if GFS2_GL_HASH_SIZE < GL_HASH_LOCK_SZ
108 # undef GL_HASH_LOCK_SZ
109 # define GL_HASH_LOCK_SZ GFS2_GL_HASH_SIZE
112 static rwlock_t gl_hash_locks[GL_HASH_LOCK_SZ];
114 static inline rwlock_t *gl_lock_addr(unsigned int x)
116 return &gl_hash_locks[x & (GL_HASH_LOCK_SZ-1)];
118 #else /* not SMP, so no spinlocks required */
119 static inline rwlock_t *gl_lock_addr(unsigned int x)
126 * relaxed_state_ok - is a requested lock compatible with the current lock mode?
127 * @actual: the current state of the lock
128 * @requested: the lock state that was requested by the caller
129 * @flags: the modifier flags passed in by the caller
131 * Returns: 1 if the locks are compatible, 0 otherwise
134 static inline int relaxed_state_ok(unsigned int actual, unsigned requested,
137 if (actual == requested)
140 if (flags & GL_EXACT)
143 if (actual == LM_ST_EXCLUSIVE && requested == LM_ST_SHARED)
146 if (actual != LM_ST_UNLOCKED && (flags & LM_FLAG_ANY))
153 * gl_hash() - Turn glock number into hash bucket number
154 * @lock: The glock number
156 * Returns: The number of the corresponding hash bucket
159 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
160 const struct lm_lockname *name)
164 h = jhash(&name->ln_number, sizeof(u64), 0);
165 h = jhash(&name->ln_type, sizeof(unsigned int), h);
166 h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
167 h &= GFS2_GL_HASH_MASK;
173 * glock_free() - Perform a few checks and then release struct gfs2_glock
174 * @gl: The glock to release
176 * Also calls lock module to release its internal structure for this glock.
180 static void glock_free(struct gfs2_glock *gl)
182 struct gfs2_sbd *sdp = gl->gl_sbd;
183 struct inode *aspace = gl->gl_aspace;
185 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
186 sdp->sd_lockstruct.ls_ops->lm_put_lock(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 static 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 gfs2_assert(sdp, gl->gl_state == LM_ST_UNLOCKED);
221 gfs2_assert(sdp, list_empty(&gl->gl_reclaim));
222 gfs2_assert(sdp, list_empty(&gl->gl_holders));
223 gfs2_assert(sdp, list_empty(&gl->gl_waiters1));
224 gfs2_assert(sdp, list_empty(&gl->gl_waiters3));
229 write_unlock(gl_lock_addr(gl->gl_hash));
235 * search_bucket() - Find struct gfs2_glock by lock number
236 * @bucket: the bucket to search
237 * @name: The lock name
239 * Returns: NULL, or the struct gfs2_glock with the requested number
242 static struct gfs2_glock *search_bucket(unsigned int hash,
243 const struct gfs2_sbd *sdp,
244 const struct lm_lockname *name)
246 struct gfs2_glock *gl;
247 struct hlist_node *h;
249 hlist_for_each_entry(gl, h, &gl_hash_table[hash].hb_list, gl_list) {
250 if (!lm_name_equal(&gl->gl_name, name))
252 if (gl->gl_sbd != sdp)
255 atomic_inc(&gl->gl_ref);
264 * gfs2_glock_find() - Find glock by lock number
265 * @sdp: The GFS2 superblock
266 * @name: The lock name
268 * Returns: NULL, or the struct gfs2_glock with the requested number
271 static struct gfs2_glock *gfs2_glock_find(const struct gfs2_sbd *sdp,
272 const struct lm_lockname *name)
274 unsigned int hash = gl_hash(sdp, name);
275 struct gfs2_glock *gl;
277 read_lock(gl_lock_addr(hash));
278 gl = search_bucket(hash, sdp, name);
279 read_unlock(gl_lock_addr(hash));
284 static void glock_work_func(struct work_struct *work)
286 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
288 spin_lock(&gl->gl_spin);
289 if (test_and_clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags))
290 set_bit(GLF_DEMOTE, &gl->gl_flags);
292 spin_unlock(&gl->gl_spin);
296 static int gfs2_lm_get_lock(struct gfs2_sbd *sdp, struct lm_lockname *name,
300 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
301 error = sdp->sd_lockstruct.ls_ops->lm_get_lock(
302 sdp->sd_lockstruct.ls_lockspace, name, lockp);
307 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
308 * @sdp: The GFS2 superblock
309 * @number: the lock number
310 * @glops: The glock_operations to use
311 * @create: If 0, don't create the glock if it doesn't exist
312 * @glp: the glock is returned here
314 * This does not lock a glock, just finds/creates structures for one.
319 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
320 const struct gfs2_glock_operations *glops, int create,
321 struct gfs2_glock **glp)
323 struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
324 struct gfs2_glock *gl, *tmp;
325 unsigned int hash = gl_hash(sdp, &name);
328 read_lock(gl_lock_addr(hash));
329 gl = search_bucket(hash, sdp, &name);
330 read_unlock(gl_lock_addr(hash));
337 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
343 atomic_set(&gl->gl_ref, 1);
344 gl->gl_state = LM_ST_UNLOCKED;
345 gl->gl_demote_state = LM_ST_EXCLUSIVE;
347 gl->gl_owner_pid = NULL;
350 gl->gl_req_gh = NULL;
352 gl->gl_stamp = jiffies;
353 gl->gl_tchange = jiffies;
354 gl->gl_object = NULL;
356 gl->gl_aspace = NULL;
357 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
359 /* If this glock protects actual on-disk data or metadata blocks,
360 create a VFS inode to manage the pages/buffers holding them. */
361 if (glops == &gfs2_inode_glops || glops == &gfs2_rgrp_glops) {
362 gl->gl_aspace = gfs2_aspace_get(sdp);
363 if (!gl->gl_aspace) {
369 error = gfs2_lm_get_lock(sdp, &name, &gl->gl_lock);
373 write_lock(gl_lock_addr(hash));
374 tmp = search_bucket(hash, sdp, &name);
376 write_unlock(gl_lock_addr(hash));
380 hlist_add_head(&gl->gl_list, &gl_hash_table[hash].hb_list);
381 write_unlock(gl_lock_addr(hash));
390 gfs2_aspace_put(gl->gl_aspace);
392 kmem_cache_free(gfs2_glock_cachep, gl);
397 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
399 * @state: the state we're requesting
400 * @flags: the modifier flags
401 * @gh: the holder structure
405 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
406 struct gfs2_holder *gh)
408 INIT_LIST_HEAD(&gh->gh_list);
410 gh->gh_ip = (unsigned long)__builtin_return_address(0);
411 gh->gh_owner_pid = get_pid(task_pid(current));
412 gh->gh_state = state;
413 gh->gh_flags = flags;
420 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
421 * @state: the state we're requesting
422 * @flags: the modifier flags
423 * @gh: the holder structure
425 * Don't mess with the glock.
429 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
431 gh->gh_state = state;
432 gh->gh_flags = flags;
434 gh->gh_ip = (unsigned long)__builtin_return_address(0);
438 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
439 * @gh: the holder structure
443 void gfs2_holder_uninit(struct gfs2_holder *gh)
445 put_pid(gh->gh_owner_pid);
446 gfs2_glock_put(gh->gh_gl);
451 static void gfs2_holder_wake(struct gfs2_holder *gh)
453 clear_bit(HIF_WAIT, &gh->gh_iflags);
454 smp_mb__after_clear_bit();
455 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
458 static int just_schedule(void *word)
464 static void wait_on_holder(struct gfs2_holder *gh)
467 wait_on_bit(&gh->gh_iflags, HIF_WAIT, just_schedule, TASK_UNINTERRUPTIBLE);
470 static void gfs2_demote_wake(struct gfs2_glock *gl)
472 gl->gl_demote_state = LM_ST_EXCLUSIVE;
473 clear_bit(GLF_DEMOTE, &gl->gl_flags);
474 smp_mb__after_clear_bit();
475 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
478 static void wait_on_demote(struct gfs2_glock *gl)
481 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, just_schedule, TASK_UNINTERRUPTIBLE);
485 * rq_mutex - process a mutex request in the queue
486 * @gh: the glock holder
488 * Returns: 1 if the queue is blocked
491 static int rq_mutex(struct gfs2_holder *gh)
493 struct gfs2_glock *gl = gh->gh_gl;
495 list_del_init(&gh->gh_list);
496 /* gh->gh_error never examined. */
497 set_bit(GLF_LOCK, &gl->gl_flags);
498 clear_bit(HIF_WAIT, &gh->gh_iflags);
500 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
506 * rq_promote - process a promote request in the queue
507 * @gh: the glock holder
509 * Acquire a new inter-node lock, or change a lock state to more restrictive.
511 * Returns: 1 if the queue is blocked
514 static int rq_promote(struct gfs2_holder *gh)
516 struct gfs2_glock *gl = gh->gh_gl;
518 if (!relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
519 if (list_empty(&gl->gl_holders)) {
521 set_bit(GLF_LOCK, &gl->gl_flags);
522 spin_unlock(&gl->gl_spin);
523 gfs2_glock_xmote_th(gh->gh_gl, gh);
524 spin_lock(&gl->gl_spin);
529 if (list_empty(&gl->gl_holders)) {
530 set_bit(HIF_FIRST, &gh->gh_iflags);
531 set_bit(GLF_LOCK, &gl->gl_flags);
533 struct gfs2_holder *next_gh;
534 if (gh->gh_state == LM_ST_EXCLUSIVE)
536 next_gh = list_entry(gl->gl_holders.next, struct gfs2_holder,
538 if (next_gh->gh_state == LM_ST_EXCLUSIVE)
542 list_move_tail(&gh->gh_list, &gl->gl_holders);
544 set_bit(HIF_HOLDER, &gh->gh_iflags);
546 gfs2_holder_wake(gh);
552 * rq_demote - process a demote request in the queue
553 * @gh: the glock holder
555 * Returns: 1 if the queue is blocked
558 static int rq_demote(struct gfs2_glock *gl)
560 if (!list_empty(&gl->gl_holders))
563 if (gl->gl_state == gl->gl_demote_state ||
564 gl->gl_state == LM_ST_UNLOCKED) {
565 gfs2_demote_wake(gl);
569 set_bit(GLF_LOCK, &gl->gl_flags);
570 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
572 if (gl->gl_demote_state == LM_ST_UNLOCKED ||
573 gl->gl_state != LM_ST_EXCLUSIVE) {
574 spin_unlock(&gl->gl_spin);
575 gfs2_glock_drop_th(gl);
577 spin_unlock(&gl->gl_spin);
578 gfs2_glock_xmote_th(gl, NULL);
581 spin_lock(&gl->gl_spin);
582 clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
588 * run_queue - process holder structures on a glock
592 static void run_queue(struct gfs2_glock *gl)
594 struct gfs2_holder *gh;
598 if (test_bit(GLF_LOCK, &gl->gl_flags))
601 if (!list_empty(&gl->gl_waiters1)) {
602 gh = list_entry(gl->gl_waiters1.next,
603 struct gfs2_holder, gh_list);
604 blocked = rq_mutex(gh);
605 } else if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
606 blocked = rq_demote(gl);
607 if (test_bit(GLF_WAITERS2, &gl->gl_flags) &&
609 set_bit(GLF_DEMOTE, &gl->gl_flags);
610 gl->gl_demote_state = LM_ST_UNLOCKED;
612 clear_bit(GLF_WAITERS2, &gl->gl_flags);
613 } else if (!list_empty(&gl->gl_waiters3)) {
614 gh = list_entry(gl->gl_waiters3.next,
615 struct gfs2_holder, gh_list);
616 blocked = rq_promote(gh);
626 * gfs2_glmutex_lock - acquire a local lock on a glock
629 * Gives caller exclusive access to manipulate a glock structure.
632 static void gfs2_glmutex_lock(struct gfs2_glock *gl)
634 spin_lock(&gl->gl_spin);
635 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
636 struct gfs2_holder gh;
638 gfs2_holder_init(gl, 0, 0, &gh);
639 set_bit(HIF_WAIT, &gh.gh_iflags);
640 list_add_tail(&gh.gh_list, &gl->gl_waiters1);
641 spin_unlock(&gl->gl_spin);
643 gfs2_holder_uninit(&gh);
645 gl->gl_owner_pid = get_pid(task_pid(current));
646 gl->gl_ip = (unsigned long)__builtin_return_address(0);
647 spin_unlock(&gl->gl_spin);
652 * gfs2_glmutex_trylock - try to acquire a local lock on a glock
655 * Returns: 1 if the glock is acquired
658 static int gfs2_glmutex_trylock(struct gfs2_glock *gl)
662 spin_lock(&gl->gl_spin);
663 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
666 gl->gl_owner_pid = get_pid(task_pid(current));
667 gl->gl_ip = (unsigned long)__builtin_return_address(0);
669 spin_unlock(&gl->gl_spin);
675 * gfs2_glmutex_unlock - release a local lock on a glock
680 static void gfs2_glmutex_unlock(struct gfs2_glock *gl)
684 spin_lock(&gl->gl_spin);
685 clear_bit(GLF_LOCK, &gl->gl_flags);
686 pid = gl->gl_owner_pid;
687 gl->gl_owner_pid = NULL;
690 spin_unlock(&gl->gl_spin);
696 * handle_callback - process a demote request
698 * @state: the state the caller wants us to change to
700 * There are only two requests that we are going to see in actual
701 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
704 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
705 int remote, unsigned long delay)
707 int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
709 spin_lock(&gl->gl_spin);
710 set_bit(bit, &gl->gl_flags);
711 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
712 gl->gl_demote_state = state;
713 gl->gl_demote_time = jiffies;
714 if (remote && gl->gl_ops->go_type == LM_TYPE_IOPEN &&
716 gfs2_glock_schedule_for_reclaim(gl);
717 spin_unlock(&gl->gl_spin);
720 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
721 gl->gl_demote_state != state) {
722 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
723 set_bit(GLF_WAITERS2, &gl->gl_flags);
725 gl->gl_demote_state = LM_ST_UNLOCKED;
727 spin_unlock(&gl->gl_spin);
731 * state_change - record that the glock is now in a different state
733 * @new_state the new state
737 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
741 held1 = (gl->gl_state != LM_ST_UNLOCKED);
742 held2 = (new_state != LM_ST_UNLOCKED);
744 if (held1 != held2) {
751 gl->gl_state = new_state;
752 gl->gl_tchange = jiffies;
756 * drop_bh - Called after a lock module unlock completes
758 * @ret: the return status
760 * Doesn't wake up the process waiting on the struct gfs2_holder (if any)
761 * Doesn't drop the reference on the glock the top half took out
765 static void drop_bh(struct gfs2_glock *gl, unsigned int ret)
767 struct gfs2_sbd *sdp = gl->gl_sbd;
768 const struct gfs2_glock_operations *glops = gl->gl_ops;
769 struct gfs2_holder *gh = gl->gl_req_gh;
771 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
772 gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
773 gfs2_assert_warn(sdp, !ret);
775 state_change(gl, LM_ST_UNLOCKED);
778 glops->go_inval(gl, DIO_METADATA);
781 spin_lock(&gl->gl_spin);
782 list_del_init(&gh->gh_list);
784 spin_unlock(&gl->gl_spin);
787 spin_lock(&gl->gl_spin);
788 gfs2_demote_wake(gl);
789 gl->gl_req_gh = NULL;
790 clear_bit(GLF_LOCK, &gl->gl_flags);
791 spin_unlock(&gl->gl_spin);
796 gfs2_holder_wake(gh);
800 * xmote_bh - Called after the lock module is done acquiring a lock
801 * @gl: The glock in question
802 * @ret: the int returned from the lock module
806 static void xmote_bh(struct gfs2_glock *gl, unsigned int ret)
808 struct gfs2_sbd *sdp = gl->gl_sbd;
809 const struct gfs2_glock_operations *glops = gl->gl_ops;
810 struct gfs2_holder *gh = gl->gl_req_gh;
811 int prev_state = gl->gl_state;
814 if ((ret & LM_OUT_ST_MASK) == LM_ST_UNLOCKED) {
819 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
820 gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
821 gfs2_assert_warn(sdp, !(ret & LM_OUT_ASYNC));
823 state_change(gl, ret & LM_OUT_ST_MASK);
825 if (prev_state != LM_ST_UNLOCKED && !(ret & LM_OUT_CACHEABLE)) {
827 glops->go_inval(gl, DIO_METADATA);
828 } else if (gl->gl_state == LM_ST_DEFERRED) {
829 /* We might not want to do this here.
830 Look at moving to the inode glops. */
832 glops->go_inval(gl, 0);
835 /* Deal with each possible exit condition */
838 gl->gl_stamp = jiffies;
839 if (ret & LM_OUT_CANCELED) {
842 spin_lock(&gl->gl_spin);
843 if (gl->gl_state != gl->gl_demote_state) {
844 spin_unlock(&gl->gl_spin);
845 gfs2_glock_drop_th(gl);
849 gfs2_demote_wake(gl);
850 spin_unlock(&gl->gl_spin);
853 spin_lock(&gl->gl_spin);
854 list_del_init(&gh->gh_list);
856 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
858 gh->gh_error = GLR_CANCELED;
859 if (ret & LM_OUT_CANCELED)
861 if (relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
862 list_add_tail(&gh->gh_list, &gl->gl_holders);
864 set_bit(HIF_HOLDER, &gh->gh_iflags);
865 set_bit(HIF_FIRST, &gh->gh_iflags);
869 gh->gh_error = GLR_TRYFAILED;
870 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
872 gh->gh_error = -EINVAL;
873 if (gfs2_assert_withdraw(sdp, 0) == -1)
874 fs_err(sdp, "ret = 0x%.8X\n", ret);
876 spin_unlock(&gl->gl_spin);
879 if (glops->go_xmote_bh)
880 glops->go_xmote_bh(gl);
883 spin_lock(&gl->gl_spin);
884 gl->gl_req_gh = NULL;
885 clear_bit(GLF_LOCK, &gl->gl_flags);
886 spin_unlock(&gl->gl_spin);
892 gfs2_holder_wake(gh);
895 static unsigned int gfs2_lm_lock(struct gfs2_sbd *sdp, void *lock,
896 unsigned int cur_state, unsigned int req_state,
900 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
901 ret = sdp->sd_lockstruct.ls_ops->lm_lock(lock, cur_state,
907 * gfs2_glock_xmote_th - Call into the lock module to acquire or change a glock
908 * @gl: The glock in question
909 * @state: the requested state
910 * @flags: modifier flags to the lock call
914 static void gfs2_glock_xmote_th(struct gfs2_glock *gl, struct gfs2_holder *gh)
916 struct gfs2_sbd *sdp = gl->gl_sbd;
917 int flags = gh ? gh->gh_flags : 0;
918 unsigned state = gh ? gh->gh_state : gl->gl_demote_state;
919 const struct gfs2_glock_operations *glops = gl->gl_ops;
920 int lck_flags = flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB |
921 LM_FLAG_NOEXP | LM_FLAG_ANY |
923 unsigned int lck_ret;
925 if (glops->go_xmote_th)
926 glops->go_xmote_th(gl);
928 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
929 gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
930 gfs2_assert_warn(sdp, state != LM_ST_UNLOCKED);
931 gfs2_assert_warn(sdp, state != gl->gl_state);
935 lck_ret = gfs2_lm_lock(sdp, gl->gl_lock, gl->gl_state, state, lck_flags);
937 if (gfs2_assert_withdraw(sdp, !(lck_ret & LM_OUT_ERROR)))
940 if (lck_ret & LM_OUT_ASYNC)
941 gfs2_assert_warn(sdp, lck_ret == LM_OUT_ASYNC);
943 xmote_bh(gl, lck_ret);
946 static unsigned int gfs2_lm_unlock(struct gfs2_sbd *sdp, void *lock,
947 unsigned int cur_state)
950 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
951 ret = sdp->sd_lockstruct.ls_ops->lm_unlock(lock, cur_state);
956 * gfs2_glock_drop_th - call into the lock module to unlock a lock
961 static void gfs2_glock_drop_th(struct gfs2_glock *gl)
963 struct gfs2_sbd *sdp = gl->gl_sbd;
964 const struct gfs2_glock_operations *glops = gl->gl_ops;
967 if (glops->go_xmote_th)
968 glops->go_xmote_th(gl);
970 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
971 gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
972 gfs2_assert_warn(sdp, gl->gl_state != LM_ST_UNLOCKED);
976 ret = gfs2_lm_unlock(sdp, gl->gl_lock, gl->gl_state);
978 if (gfs2_assert_withdraw(sdp, !(ret & LM_OUT_ERROR)))
984 gfs2_assert_warn(sdp, ret == LM_OUT_ASYNC);
988 * do_cancels - cancel requests for locks stuck waiting on an expire flag
989 * @gh: the LM_FLAG_PRIORITY holder waiting to acquire the lock
991 * Don't cancel GL_NOCANCEL requests.
994 static void do_cancels(struct gfs2_holder *gh)
996 struct gfs2_glock *gl = gh->gh_gl;
997 struct gfs2_sbd *sdp = gl->gl_sbd;
999 spin_lock(&gl->gl_spin);
1001 while (gl->gl_req_gh != gh &&
1002 !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1003 !list_empty(&gh->gh_list)) {
1004 if (!(gl->gl_req_gh && (gl->gl_req_gh->gh_flags & GL_NOCANCEL))) {
1005 spin_unlock(&gl->gl_spin);
1006 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1007 sdp->sd_lockstruct.ls_ops->lm_cancel(gl->gl_lock);
1009 spin_lock(&gl->gl_spin);
1011 spin_unlock(&gl->gl_spin);
1013 spin_lock(&gl->gl_spin);
1017 spin_unlock(&gl->gl_spin);
1021 * glock_wait_internal - wait on a glock acquisition
1022 * @gh: the glock holder
1024 * Returns: 0 on success
1027 static int glock_wait_internal(struct gfs2_holder *gh)
1029 struct gfs2_glock *gl = gh->gh_gl;
1030 struct gfs2_sbd *sdp = gl->gl_sbd;
1031 const struct gfs2_glock_operations *glops = gl->gl_ops;
1033 if (test_bit(HIF_ABORTED, &gh->gh_iflags))
1036 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1037 spin_lock(&gl->gl_spin);
1038 if (gl->gl_req_gh != gh &&
1039 !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1040 !list_empty(&gh->gh_list)) {
1041 list_del_init(&gh->gh_list);
1042 gh->gh_error = GLR_TRYFAILED;
1044 spin_unlock(&gl->gl_spin);
1045 return gh->gh_error;
1047 spin_unlock(&gl->gl_spin);
1050 if (gh->gh_flags & LM_FLAG_PRIORITY)
1055 return gh->gh_error;
1057 gfs2_assert_withdraw(sdp, test_bit(HIF_HOLDER, &gh->gh_iflags));
1058 gfs2_assert_withdraw(sdp, relaxed_state_ok(gl->gl_state, gh->gh_state,
1061 if (test_bit(HIF_FIRST, &gh->gh_iflags)) {
1062 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1064 if (glops->go_lock) {
1065 gh->gh_error = glops->go_lock(gh);
1067 spin_lock(&gl->gl_spin);
1068 list_del_init(&gh->gh_list);
1069 spin_unlock(&gl->gl_spin);
1073 spin_lock(&gl->gl_spin);
1074 gl->gl_req_gh = NULL;
1075 clear_bit(GLF_LOCK, &gl->gl_flags);
1077 spin_unlock(&gl->gl_spin);
1080 return gh->gh_error;
1083 static inline struct gfs2_holder *
1084 find_holder_by_owner(struct list_head *head, struct pid *pid)
1086 struct gfs2_holder *gh;
1088 list_for_each_entry(gh, head, gh_list) {
1089 if (gh->gh_owner_pid == pid)
1096 static void print_dbg(struct glock_iter *gi, const char *fmt, ...)
1100 va_start(args, fmt);
1102 vsprintf(gi->string, fmt, args);
1103 seq_printf(gi->seq, gi->string);
1111 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1112 * @gh: the holder structure to add
1116 static void add_to_queue(struct gfs2_holder *gh)
1118 struct gfs2_glock *gl = gh->gh_gl;
1119 struct gfs2_holder *existing;
1121 BUG_ON(gh->gh_owner_pid == NULL);
1122 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1125 if (!(gh->gh_flags & GL_FLOCK)) {
1126 existing = find_holder_by_owner(&gl->gl_holders,
1129 print_symbol(KERN_WARNING "original: %s\n",
1131 printk(KERN_INFO "pid : %d\n",
1132 pid_nr(existing->gh_owner_pid));
1133 printk(KERN_INFO "lock type : %d lock state : %d\n",
1134 existing->gh_gl->gl_name.ln_type,
1135 existing->gh_gl->gl_state);
1136 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1137 printk(KERN_INFO "pid : %d\n",
1138 pid_nr(gh->gh_owner_pid));
1139 printk(KERN_INFO "lock type : %d lock state : %d\n",
1140 gl->gl_name.ln_type, gl->gl_state);
1144 existing = find_holder_by_owner(&gl->gl_waiters3,
1147 print_symbol(KERN_WARNING "original: %s\n",
1149 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1154 if (gh->gh_flags & LM_FLAG_PRIORITY)
1155 list_add(&gh->gh_list, &gl->gl_waiters3);
1157 list_add_tail(&gh->gh_list, &gl->gl_waiters3);
1161 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1162 * @gh: the holder structure
1164 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1166 * Returns: 0, GLR_TRYFAILED, or errno on failure
1169 int gfs2_glock_nq(struct gfs2_holder *gh)
1171 struct gfs2_glock *gl = gh->gh_gl;
1172 struct gfs2_sbd *sdp = gl->gl_sbd;
1176 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1177 set_bit(HIF_ABORTED, &gh->gh_iflags);
1181 spin_lock(&gl->gl_spin);
1184 spin_unlock(&gl->gl_spin);
1186 if (!(gh->gh_flags & GL_ASYNC)) {
1187 error = glock_wait_internal(gh);
1188 if (error == GLR_CANCELED) {
1198 * gfs2_glock_poll - poll to see if an async request has been completed
1201 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1204 int gfs2_glock_poll(struct gfs2_holder *gh)
1206 struct gfs2_glock *gl = gh->gh_gl;
1209 spin_lock(&gl->gl_spin);
1211 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1213 else if (list_empty(&gh->gh_list)) {
1214 if (gh->gh_error == GLR_CANCELED) {
1215 spin_unlock(&gl->gl_spin);
1217 if (gfs2_glock_nq(gh))
1224 spin_unlock(&gl->gl_spin);
1230 * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
1231 * @gh: the holder structure
1233 * Returns: 0, GLR_TRYFAILED, or errno on failure
1236 int gfs2_glock_wait(struct gfs2_holder *gh)
1240 error = glock_wait_internal(gh);
1241 if (error == GLR_CANCELED) {
1243 gh->gh_flags &= ~GL_ASYNC;
1244 error = gfs2_glock_nq(gh);
1251 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1252 * @gh: the glock holder
1256 void gfs2_glock_dq(struct gfs2_holder *gh)
1258 struct gfs2_glock *gl = gh->gh_gl;
1259 const struct gfs2_glock_operations *glops = gl->gl_ops;
1262 if (gh->gh_flags & GL_NOCACHE)
1263 handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1265 gfs2_glmutex_lock(gl);
1267 spin_lock(&gl->gl_spin);
1268 list_del_init(&gh->gh_list);
1270 if (list_empty(&gl->gl_holders)) {
1271 if (glops->go_unlock) {
1272 spin_unlock(&gl->gl_spin);
1273 glops->go_unlock(gh);
1274 spin_lock(&gl->gl_spin);
1276 gl->gl_stamp = jiffies;
1279 clear_bit(GLF_LOCK, &gl->gl_flags);
1280 spin_unlock(&gl->gl_spin);
1282 gfs2_glock_hold(gl);
1283 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1284 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1285 delay = gl->gl_ops->go_min_hold_time;
1286 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1290 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1292 struct gfs2_glock *gl = gh->gh_gl;
1298 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1299 * @gh: the holder structure
1303 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1306 gfs2_holder_uninit(gh);
1310 * gfs2_glock_nq_num - acquire a glock based on lock number
1311 * @sdp: the filesystem
1312 * @number: the lock number
1313 * @glops: the glock operations for the type of glock
1314 * @state: the state to acquire the glock in
1315 * @flags: modifier flags for the aquisition
1316 * @gh: the struct gfs2_holder
1321 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1322 const struct gfs2_glock_operations *glops,
1323 unsigned int state, int flags, struct gfs2_holder *gh)
1325 struct gfs2_glock *gl;
1328 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1330 error = gfs2_glock_nq_init(gl, state, flags, gh);
1338 * glock_compare - Compare two struct gfs2_glock structures for sorting
1339 * @arg_a: the first structure
1340 * @arg_b: the second structure
1344 static int glock_compare(const void *arg_a, const void *arg_b)
1346 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1347 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1348 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1349 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1351 if (a->ln_number > b->ln_number)
1353 if (a->ln_number < b->ln_number)
1355 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1360 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1361 * @num_gh: the number of structures
1362 * @ghs: an array of struct gfs2_holder structures
1364 * Returns: 0 on success (all glocks acquired),
1365 * errno on failure (no glocks acquired)
1368 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1369 struct gfs2_holder **p)
1374 for (x = 0; x < num_gh; x++)
1377 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1379 for (x = 0; x < num_gh; x++) {
1380 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1382 error = gfs2_glock_nq(p[x]);
1385 gfs2_glock_dq(p[x]);
1394 * gfs2_glock_nq_m - acquire multiple glocks
1395 * @num_gh: the number of structures
1396 * @ghs: an array of struct gfs2_holder structures
1399 * Returns: 0 on success (all glocks acquired),
1400 * errno on failure (no glocks acquired)
1403 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1405 struct gfs2_holder *tmp[4];
1406 struct gfs2_holder **pph = tmp;
1413 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1414 return gfs2_glock_nq(ghs);
1418 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1423 error = nq_m_sync(num_gh, ghs, pph);
1432 * gfs2_glock_dq_m - release multiple glocks
1433 * @num_gh: the number of structures
1434 * @ghs: an array of struct gfs2_holder structures
1438 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1442 for (x = 0; x < num_gh; x++)
1443 gfs2_glock_dq(&ghs[x]);
1447 * gfs2_glock_dq_uninit_m - release multiple glocks
1448 * @num_gh: the number of structures
1449 * @ghs: an array of struct gfs2_holder structures
1453 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1457 for (x = 0; x < num_gh; x++)
1458 gfs2_glock_dq_uninit(&ghs[x]);
1461 static int gfs2_lm_hold_lvb(struct gfs2_sbd *sdp, void *lock, char **lvbp)
1464 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1465 error = sdp->sd_lockstruct.ls_ops->lm_hold_lvb(lock, lvbp);
1470 * gfs2_lvb_hold - attach a LVB from a glock
1471 * @gl: The glock in question
1475 int gfs2_lvb_hold(struct gfs2_glock *gl)
1479 gfs2_glmutex_lock(gl);
1481 if (!atomic_read(&gl->gl_lvb_count)) {
1482 error = gfs2_lm_hold_lvb(gl->gl_sbd, gl->gl_lock, &gl->gl_lvb);
1484 gfs2_glmutex_unlock(gl);
1487 gfs2_glock_hold(gl);
1489 atomic_inc(&gl->gl_lvb_count);
1491 gfs2_glmutex_unlock(gl);
1497 * gfs2_lvb_unhold - detach a LVB from a glock
1498 * @gl: The glock in question
1502 void gfs2_lvb_unhold(struct gfs2_glock *gl)
1504 struct gfs2_sbd *sdp = gl->gl_sbd;
1506 gfs2_glock_hold(gl);
1507 gfs2_glmutex_lock(gl);
1509 gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count) > 0);
1510 if (atomic_dec_and_test(&gl->gl_lvb_count)) {
1511 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1512 sdp->sd_lockstruct.ls_ops->lm_unhold_lvb(gl->gl_lock, gl->gl_lvb);
1517 gfs2_glmutex_unlock(gl);
1521 static void blocking_cb(struct gfs2_sbd *sdp, struct lm_lockname *name,
1524 struct gfs2_glock *gl;
1525 unsigned long delay = 0;
1526 unsigned long holdtime;
1527 unsigned long now = jiffies;
1529 gl = gfs2_glock_find(sdp, name);
1533 holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
1534 if (time_before(now, holdtime))
1535 delay = holdtime - now;
1537 handle_callback(gl, state, 1, delay);
1538 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1543 * gfs2_glock_cb - Callback used by locking module
1544 * @sdp: Pointer to the superblock
1545 * @type: Type of callback
1546 * @data: Type dependent data pointer
1548 * Called by the locking module when it wants to tell us something.
1549 * Either we need to drop a lock, one of our ASYNC requests completed, or
1550 * a journal from another client needs to be recovered.
1553 void gfs2_glock_cb(void *cb_data, unsigned int type, void *data)
1555 struct gfs2_sbd *sdp = cb_data;
1559 blocking_cb(sdp, data, LM_ST_UNLOCKED);
1563 blocking_cb(sdp, data, LM_ST_DEFERRED);
1567 blocking_cb(sdp, data, LM_ST_SHARED);
1571 struct lm_async_cb *async = data;
1572 struct gfs2_glock *gl;
1574 down_read(&gfs2_umount_flush_sem);
1575 gl = gfs2_glock_find(sdp, &async->lc_name);
1576 if (gfs2_assert_warn(sdp, gl))
1578 xmote_bh(gl, async->lc_ret);
1579 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1581 up_read(&gfs2_umount_flush_sem);
1585 case LM_CB_NEED_RECOVERY:
1586 gfs2_jdesc_make_dirty(sdp, *(unsigned int *)data);
1587 if (sdp->sd_recoverd_process)
1588 wake_up_process(sdp->sd_recoverd_process);
1591 case LM_CB_DROPLOCKS:
1592 gfs2_gl_hash_clear(sdp, NO_WAIT);
1593 gfs2_quota_scan(sdp);
1597 gfs2_assert_warn(sdp, 0);
1603 * demote_ok - Check to see if it's ok to unlock a glock
1606 * Returns: 1 if it's ok
1609 static int demote_ok(struct gfs2_glock *gl)
1611 const struct gfs2_glock_operations *glops = gl->gl_ops;
1614 if (test_bit(GLF_STICKY, &gl->gl_flags))
1616 else if (glops->go_demote_ok)
1617 demote = glops->go_demote_ok(gl);
1623 * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1628 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
1630 struct gfs2_sbd *sdp = gl->gl_sbd;
1632 spin_lock(&sdp->sd_reclaim_lock);
1633 if (list_empty(&gl->gl_reclaim)) {
1634 gfs2_glock_hold(gl);
1635 list_add(&gl->gl_reclaim, &sdp->sd_reclaim_list);
1636 atomic_inc(&sdp->sd_reclaim_count);
1637 spin_unlock(&sdp->sd_reclaim_lock);
1638 wake_up(&sdp->sd_reclaim_wq);
1640 spin_unlock(&sdp->sd_reclaim_lock);
1644 * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1645 * @sdp: the filesystem
1647 * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1648 * different glock and we notice that there are a lot of glocks in the
1653 void gfs2_reclaim_glock(struct gfs2_sbd *sdp)
1655 struct gfs2_glock *gl;
1657 spin_lock(&sdp->sd_reclaim_lock);
1658 if (list_empty(&sdp->sd_reclaim_list)) {
1659 spin_unlock(&sdp->sd_reclaim_lock);
1662 gl = list_entry(sdp->sd_reclaim_list.next,
1663 struct gfs2_glock, gl_reclaim);
1664 list_del_init(&gl->gl_reclaim);
1665 spin_unlock(&sdp->sd_reclaim_lock);
1667 atomic_dec(&sdp->sd_reclaim_count);
1668 atomic_inc(&sdp->sd_reclaimed);
1670 if (gfs2_glmutex_trylock(gl)) {
1671 if (list_empty(&gl->gl_holders) &&
1672 gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1673 handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1674 gfs2_glmutex_unlock(gl);
1681 * examine_bucket - Call a function for glock in a hash bucket
1682 * @examiner: the function
1683 * @sdp: the filesystem
1684 * @bucket: the bucket
1686 * Returns: 1 if the bucket has entries
1689 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1692 struct gfs2_glock *gl, *prev = NULL;
1693 int has_entries = 0;
1694 struct hlist_head *head = &gl_hash_table[hash].hb_list;
1696 read_lock(gl_lock_addr(hash));
1697 /* Can't use hlist_for_each_entry - don't want prefetch here */
1698 if (hlist_empty(head))
1700 gl = list_entry(head->first, struct gfs2_glock, gl_list);
1702 if (!sdp || gl->gl_sbd == sdp) {
1703 gfs2_glock_hold(gl);
1704 read_unlock(gl_lock_addr(hash));
1706 gfs2_glock_put(prev);
1710 read_lock(gl_lock_addr(hash));
1712 if (gl->gl_list.next == NULL)
1714 gl = list_entry(gl->gl_list.next, struct gfs2_glock, gl_list);
1717 read_unlock(gl_lock_addr(hash));
1719 gfs2_glock_put(prev);
1725 * scan_glock - look at a glock and see if we can reclaim it
1726 * @gl: the glock to look at
1730 static void scan_glock(struct gfs2_glock *gl)
1732 if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object)
1735 if (gfs2_glmutex_trylock(gl)) {
1736 if (list_empty(&gl->gl_holders) &&
1737 gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1739 gfs2_glmutex_unlock(gl);
1744 gfs2_glmutex_unlock(gl);
1745 gfs2_glock_schedule_for_reclaim(gl);
1749 * clear_glock - look at a glock and see if we can free it from glock cache
1750 * @gl: the glock to look at
1754 static void clear_glock(struct gfs2_glock *gl)
1756 struct gfs2_sbd *sdp = gl->gl_sbd;
1759 spin_lock(&sdp->sd_reclaim_lock);
1760 if (!list_empty(&gl->gl_reclaim)) {
1761 list_del_init(&gl->gl_reclaim);
1762 atomic_dec(&sdp->sd_reclaim_count);
1763 spin_unlock(&sdp->sd_reclaim_lock);
1764 released = gfs2_glock_put(gl);
1765 gfs2_assert(sdp, !released);
1767 spin_unlock(&sdp->sd_reclaim_lock);
1770 if (gfs2_glmutex_trylock(gl)) {
1771 if (list_empty(&gl->gl_holders) &&
1772 gl->gl_state != LM_ST_UNLOCKED)
1773 handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1774 gfs2_glmutex_unlock(gl);
1779 * gfs2_gl_hash_clear - Empty out the glock hash table
1780 * @sdp: the filesystem
1781 * @wait: wait until it's all gone
1783 * Called when unmounting the filesystem, or when inter-node lock manager
1784 * requests DROPLOCKS because it is running out of capacity.
1787 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp, int wait)
1797 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1798 if (examine_bucket(clear_glock, sdp, x))
1805 if (time_after_eq(jiffies,
1806 t + gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
1807 fs_warn(sdp, "Unmount seems to be stalled. "
1808 "Dumping lock state...\n");
1809 gfs2_dump_lockstate(sdp);
1813 down_write(&gfs2_umount_flush_sem);
1814 invalidate_inodes(sdp->sd_vfs);
1815 up_write(&gfs2_umount_flush_sem);
1821 * Diagnostic routines to help debug distributed deadlock
1824 static void gfs2_print_symbol(struct glock_iter *gi, const char *fmt,
1825 unsigned long address)
1827 char buffer[KSYM_SYMBOL_LEN];
1829 sprint_symbol(buffer, address);
1830 print_dbg(gi, fmt, buffer);
1834 * dump_holder - print information about a glock holder
1835 * @str: a string naming the type of holder
1836 * @gh: the glock holder
1838 * Returns: 0 on success, -ENOBUFS when we run out of space
1841 static int dump_holder(struct glock_iter *gi, char *str,
1842 struct gfs2_holder *gh)
1845 struct task_struct *gh_owner;
1847 print_dbg(gi, " %s\n", str);
1848 if (gh->gh_owner_pid) {
1849 print_dbg(gi, " owner = %ld ",
1850 (long)pid_nr(gh->gh_owner_pid));
1851 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1853 print_dbg(gi, "(%s)\n", gh_owner->comm);
1855 print_dbg(gi, "(ended)\n");
1857 print_dbg(gi, " owner = -1\n");
1858 print_dbg(gi, " gh_state = %u\n", gh->gh_state);
1859 print_dbg(gi, " gh_flags =");
1860 for (x = 0; x < 32; x++)
1861 if (gh->gh_flags & (1 << x))
1862 print_dbg(gi, " %u", x);
1863 print_dbg(gi, " \n");
1864 print_dbg(gi, " error = %d\n", gh->gh_error);
1865 print_dbg(gi, " gh_iflags =");
1866 for (x = 0; x < 32; x++)
1867 if (test_bit(x, &gh->gh_iflags))
1868 print_dbg(gi, " %u", x);
1869 print_dbg(gi, " \n");
1870 gfs2_print_symbol(gi, " initialized at: %s\n", gh->gh_ip);
1876 * dump_inode - print information about an inode
1879 * Returns: 0 on success, -ENOBUFS when we run out of space
1882 static int dump_inode(struct glock_iter *gi, struct gfs2_inode *ip)
1886 print_dbg(gi, " Inode:\n");
1887 print_dbg(gi, " num = %llu/%llu\n",
1888 (unsigned long long)ip->i_no_formal_ino,
1889 (unsigned long long)ip->i_no_addr);
1890 print_dbg(gi, " type = %u\n", IF2DT(ip->i_inode.i_mode));
1891 print_dbg(gi, " i_flags =");
1892 for (x = 0; x < 32; x++)
1893 if (test_bit(x, &ip->i_flags))
1894 print_dbg(gi, " %u", x);
1895 print_dbg(gi, " \n");
1900 * dump_glock - print information about a glock
1902 * @count: where we are in the buffer
1904 * Returns: 0 on success, -ENOBUFS when we run out of space
1907 static int dump_glock(struct glock_iter *gi, struct gfs2_glock *gl)
1909 struct gfs2_holder *gh;
1911 int error = -ENOBUFS;
1912 struct task_struct *gl_owner;
1914 spin_lock(&gl->gl_spin);
1916 print_dbg(gi, "Glock 0x%p (%u, 0x%llx)\n", gl, gl->gl_name.ln_type,
1917 (unsigned long long)gl->gl_name.ln_number);
1918 print_dbg(gi, " gl_flags =");
1919 for (x = 0; x < 32; x++) {
1920 if (test_bit(x, &gl->gl_flags))
1921 print_dbg(gi, " %u", x);
1923 if (!test_bit(GLF_LOCK, &gl->gl_flags))
1924 print_dbg(gi, " (unlocked)");
1925 print_dbg(gi, " \n");
1926 print_dbg(gi, " gl_ref = %d\n", atomic_read(&gl->gl_ref));
1927 print_dbg(gi, " gl_state = %u\n", gl->gl_state);
1928 if (gl->gl_owner_pid) {
1929 gl_owner = pid_task(gl->gl_owner_pid, PIDTYPE_PID);
1931 print_dbg(gi, " gl_owner = pid %d (%s)\n",
1932 pid_nr(gl->gl_owner_pid), gl_owner->comm);
1934 print_dbg(gi, " gl_owner = %d (ended)\n",
1935 pid_nr(gl->gl_owner_pid));
1937 print_dbg(gi, " gl_owner = -1\n");
1938 print_dbg(gi, " gl_ip = %lu\n", gl->gl_ip);
1939 print_dbg(gi, " req_gh = %s\n", (gl->gl_req_gh) ? "yes" : "no");
1940 print_dbg(gi, " lvb_count = %d\n", atomic_read(&gl->gl_lvb_count));
1941 print_dbg(gi, " object = %s\n", (gl->gl_object) ? "yes" : "no");
1942 print_dbg(gi, " reclaim = %s\n",
1943 (list_empty(&gl->gl_reclaim)) ? "no" : "yes");
1945 print_dbg(gi, " aspace = 0x%p nrpages = %lu\n", gl->gl_aspace,
1946 gl->gl_aspace->i_mapping->nrpages);
1948 print_dbg(gi, " aspace = no\n");
1949 print_dbg(gi, " ail = %d\n", atomic_read(&gl->gl_ail_count));
1950 if (gl->gl_req_gh) {
1951 error = dump_holder(gi, "Request", gl->gl_req_gh);
1955 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1956 error = dump_holder(gi, "Holder", gh);
1960 list_for_each_entry(gh, &gl->gl_waiters1, gh_list) {
1961 error = dump_holder(gi, "Waiter1", gh);
1965 list_for_each_entry(gh, &gl->gl_waiters3, gh_list) {
1966 error = dump_holder(gi, "Waiter3", gh);
1970 if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
1971 print_dbg(gi, " Demotion req to state %u (%llu uS ago)\n",
1972 gl->gl_demote_state, (unsigned long long)
1973 (jiffies - gl->gl_demote_time)*(1000000/HZ));
1975 if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object) {
1976 if (!test_bit(GLF_LOCK, &gl->gl_flags) &&
1977 list_empty(&gl->gl_holders)) {
1978 error = dump_inode(gi, gl->gl_object);
1983 print_dbg(gi, " Inode: busy\n");
1990 spin_unlock(&gl->gl_spin);
1995 * gfs2_dump_lockstate - print out the current lockstate
1996 * @sdp: the filesystem
1997 * @ub: the buffer to copy the information into
1999 * If @ub is NULL, dump the lockstate to the console.
2003 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
2005 struct gfs2_glock *gl;
2006 struct hlist_node *h;
2010 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
2012 read_lock(gl_lock_addr(x));
2014 hlist_for_each_entry(gl, h, &gl_hash_table[x].hb_list, gl_list) {
2015 if (gl->gl_sbd != sdp)
2018 error = dump_glock(NULL, gl);
2023 read_unlock(gl_lock_addr(x));
2034 * gfs2_scand - Look for cached glocks and inodes to toss from memory
2035 * @sdp: Pointer to GFS2 superblock
2037 * One of these daemons runs, finding candidates to add to sd_reclaim_list.
2041 static int gfs2_scand(void *data)
2046 while (!kthread_should_stop()) {
2047 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
2048 examine_bucket(scan_glock, NULL, x);
2049 if (freezing(current))
2054 schedule_timeout_interruptible(delay * HZ);
2062 int __init gfs2_glock_init(void)
2065 for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
2066 INIT_HLIST_HEAD(&gl_hash_table[i].hb_list);
2068 #ifdef GL_HASH_LOCK_SZ
2069 for(i = 0; i < GL_HASH_LOCK_SZ; i++) {
2070 rwlock_init(&gl_hash_locks[i]);
2074 scand_process = kthread_run(gfs2_scand, NULL, "gfs2_scand");
2075 if (IS_ERR(scand_process))
2076 return PTR_ERR(scand_process);
2078 glock_workqueue = create_workqueue("glock_workqueue");
2079 if (IS_ERR(glock_workqueue)) {
2080 kthread_stop(scand_process);
2081 return PTR_ERR(glock_workqueue);
2087 void gfs2_glock_exit(void)
2089 destroy_workqueue(glock_workqueue);
2090 kthread_stop(scand_process);
2093 module_param(scand_secs, uint, S_IRUGO|S_IWUSR);
2094 MODULE_PARM_DESC(scand_secs, "The number of seconds between scand runs");
2096 static int gfs2_glock_iter_next(struct glock_iter *gi)
2098 struct gfs2_glock *gl;
2101 read_lock(gl_lock_addr(gi->hash));
2104 gi->gl = hlist_entry(gl->gl_list.next,
2105 struct gfs2_glock, gl_list);
2107 gfs2_glock_hold(gi->gl);
2109 read_unlock(gl_lock_addr(gi->hash));
2112 if (gl && gi->gl == NULL)
2114 while(gi->gl == NULL) {
2115 if (gi->hash >= GFS2_GL_HASH_SIZE)
2117 read_lock(gl_lock_addr(gi->hash));
2118 gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
2119 struct gfs2_glock, gl_list);
2121 gfs2_glock_hold(gi->gl);
2122 read_unlock(gl_lock_addr(gi->hash));
2126 if (gi->sdp != gi->gl->gl_sbd)
2132 static void gfs2_glock_iter_free(struct glock_iter *gi)
2135 gfs2_glock_put(gi->gl);
2139 static struct glock_iter *gfs2_glock_iter_init(struct gfs2_sbd *sdp)
2141 struct glock_iter *gi;
2143 gi = kmalloc(sizeof (*gi), GFP_KERNEL);
2151 memset(gi->string, 0, sizeof(gi->string));
2153 if (gfs2_glock_iter_next(gi)) {
2154 gfs2_glock_iter_free(gi);
2161 static void *gfs2_glock_seq_start(struct seq_file *file, loff_t *pos)
2163 struct glock_iter *gi;
2166 gi = gfs2_glock_iter_init(file->private);
2171 if (gfs2_glock_iter_next(gi)) {
2172 gfs2_glock_iter_free(gi);
2180 static void *gfs2_glock_seq_next(struct seq_file *file, void *iter_ptr,
2183 struct glock_iter *gi = iter_ptr;
2187 if (gfs2_glock_iter_next(gi)) {
2188 gfs2_glock_iter_free(gi);
2195 static void gfs2_glock_seq_stop(struct seq_file *file, void *iter_ptr)
2197 struct glock_iter *gi = iter_ptr;
2199 gfs2_glock_iter_free(gi);
2202 static int gfs2_glock_seq_show(struct seq_file *file, void *iter_ptr)
2204 struct glock_iter *gi = iter_ptr;
2207 dump_glock(gi, gi->gl);
2212 static const struct seq_operations gfs2_glock_seq_ops = {
2213 .start = gfs2_glock_seq_start,
2214 .next = gfs2_glock_seq_next,
2215 .stop = gfs2_glock_seq_stop,
2216 .show = gfs2_glock_seq_show,
2219 static int gfs2_debugfs_open(struct inode *inode, struct file *file)
2221 struct seq_file *seq;
2224 ret = seq_open(file, &gfs2_glock_seq_ops);
2228 seq = file->private_data;
2229 seq->private = inode->i_private;
2234 static const struct file_operations gfs2_debug_fops = {
2235 .owner = THIS_MODULE,
2236 .open = gfs2_debugfs_open,
2238 .llseek = seq_lseek,
2239 .release = seq_release
2242 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2244 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2245 if (!sdp->debugfs_dir)
2247 sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
2249 sdp->debugfs_dir, sdp,
2251 if (!sdp->debugfs_dentry_glocks)
2257 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2259 if (sdp && sdp->debugfs_dir) {
2260 if (sdp->debugfs_dentry_glocks) {
2261 debugfs_remove(sdp->debugfs_dentry_glocks);
2262 sdp->debugfs_dentry_glocks = NULL;
2264 debugfs_remove(sdp->debugfs_dir);
2265 sdp->debugfs_dir = NULL;
2269 int gfs2_register_debugfs(void)
2271 gfs2_root = debugfs_create_dir("gfs2", NULL);
2272 return gfs2_root ? 0 : -ENOMEM;
2275 void gfs2_unregister_debugfs(void)
2277 debugfs_remove(gfs2_root);