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>
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 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);
297 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
298 * @sdp: The GFS2 superblock
299 * @number: the lock number
300 * @glops: The glock_operations to use
301 * @create: If 0, don't create the glock if it doesn't exist
302 * @glp: the glock is returned here
304 * This does not lock a glock, just finds/creates structures for one.
309 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
310 const struct gfs2_glock_operations *glops, int create,
311 struct gfs2_glock **glp)
313 struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
314 struct gfs2_glock *gl, *tmp;
315 unsigned int hash = gl_hash(sdp, &name);
318 read_lock(gl_lock_addr(hash));
319 gl = search_bucket(hash, sdp, &name);
320 read_unlock(gl_lock_addr(hash));
327 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
333 atomic_set(&gl->gl_ref, 1);
334 gl->gl_state = LM_ST_UNLOCKED;
335 gl->gl_demote_state = LM_ST_EXCLUSIVE;
337 gl->gl_owner_pid = NULL;
340 gl->gl_req_gh = NULL;
341 gl->gl_req_bh = NULL;
343 gl->gl_stamp = jiffies;
344 gl->gl_tchange = jiffies;
345 gl->gl_object = NULL;
347 gl->gl_aspace = NULL;
348 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
350 /* If this glock protects actual on-disk data or metadata blocks,
351 create a VFS inode to manage the pages/buffers holding them. */
352 if (glops == &gfs2_inode_glops || glops == &gfs2_rgrp_glops) {
353 gl->gl_aspace = gfs2_aspace_get(sdp);
354 if (!gl->gl_aspace) {
360 error = gfs2_lm_get_lock(sdp, &name, &gl->gl_lock);
364 write_lock(gl_lock_addr(hash));
365 tmp = search_bucket(hash, sdp, &name);
367 write_unlock(gl_lock_addr(hash));
371 hlist_add_head(&gl->gl_list, &gl_hash_table[hash].hb_list);
372 write_unlock(gl_lock_addr(hash));
381 gfs2_aspace_put(gl->gl_aspace);
383 kmem_cache_free(gfs2_glock_cachep, gl);
388 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
390 * @state: the state we're requesting
391 * @flags: the modifier flags
392 * @gh: the holder structure
396 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
397 struct gfs2_holder *gh)
399 INIT_LIST_HEAD(&gh->gh_list);
401 gh->gh_ip = (unsigned long)__builtin_return_address(0);
402 gh->gh_owner_pid = get_pid(task_pid(current));
403 gh->gh_state = state;
404 gh->gh_flags = flags;
411 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
412 * @state: the state we're requesting
413 * @flags: the modifier flags
414 * @gh: the holder structure
416 * Don't mess with the glock.
420 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
422 gh->gh_state = state;
423 gh->gh_flags = flags;
425 gh->gh_ip = (unsigned long)__builtin_return_address(0);
429 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
430 * @gh: the holder structure
434 void gfs2_holder_uninit(struct gfs2_holder *gh)
436 put_pid(gh->gh_owner_pid);
437 gfs2_glock_put(gh->gh_gl);
442 static void gfs2_holder_wake(struct gfs2_holder *gh)
444 clear_bit(HIF_WAIT, &gh->gh_iflags);
445 smp_mb__after_clear_bit();
446 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
449 static int just_schedule(void *word)
455 static void wait_on_holder(struct gfs2_holder *gh)
458 wait_on_bit(&gh->gh_iflags, HIF_WAIT, just_schedule, TASK_UNINTERRUPTIBLE);
461 static void gfs2_demote_wake(struct gfs2_glock *gl)
463 gl->gl_demote_state = LM_ST_EXCLUSIVE;
464 clear_bit(GLF_DEMOTE, &gl->gl_flags);
465 smp_mb__after_clear_bit();
466 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
469 static void wait_on_demote(struct gfs2_glock *gl)
472 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, just_schedule, TASK_UNINTERRUPTIBLE);
476 * rq_mutex - process a mutex request in the queue
477 * @gh: the glock holder
479 * Returns: 1 if the queue is blocked
482 static int rq_mutex(struct gfs2_holder *gh)
484 struct gfs2_glock *gl = gh->gh_gl;
486 list_del_init(&gh->gh_list);
487 /* gh->gh_error never examined. */
488 set_bit(GLF_LOCK, &gl->gl_flags);
489 clear_bit(HIF_WAIT, &gh->gh_iflags);
491 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
497 * rq_promote - process a promote request in the queue
498 * @gh: the glock holder
500 * Acquire a new inter-node lock, or change a lock state to more restrictive.
502 * Returns: 1 if the queue is blocked
505 static int rq_promote(struct gfs2_holder *gh)
507 struct gfs2_glock *gl = gh->gh_gl;
509 if (!relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
510 if (list_empty(&gl->gl_holders)) {
512 set_bit(GLF_LOCK, &gl->gl_flags);
513 spin_unlock(&gl->gl_spin);
514 gfs2_glock_xmote_th(gh->gh_gl, gh);
515 spin_lock(&gl->gl_spin);
520 if (list_empty(&gl->gl_holders)) {
521 set_bit(HIF_FIRST, &gh->gh_iflags);
522 set_bit(GLF_LOCK, &gl->gl_flags);
524 struct gfs2_holder *next_gh;
525 if (gh->gh_state == LM_ST_EXCLUSIVE)
527 next_gh = list_entry(gl->gl_holders.next, struct gfs2_holder,
529 if (next_gh->gh_state == LM_ST_EXCLUSIVE)
533 list_move_tail(&gh->gh_list, &gl->gl_holders);
535 set_bit(HIF_HOLDER, &gh->gh_iflags);
537 gfs2_holder_wake(gh);
543 * rq_demote - process a demote request in the queue
544 * @gh: the glock holder
546 * Returns: 1 if the queue is blocked
549 static int rq_demote(struct gfs2_glock *gl)
551 if (!list_empty(&gl->gl_holders))
554 if (gl->gl_state == gl->gl_demote_state ||
555 gl->gl_state == LM_ST_UNLOCKED) {
556 gfs2_demote_wake(gl);
560 set_bit(GLF_LOCK, &gl->gl_flags);
561 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
563 if (gl->gl_demote_state == LM_ST_UNLOCKED ||
564 gl->gl_state != LM_ST_EXCLUSIVE) {
565 spin_unlock(&gl->gl_spin);
566 gfs2_glock_drop_th(gl);
568 spin_unlock(&gl->gl_spin);
569 gfs2_glock_xmote_th(gl, NULL);
572 spin_lock(&gl->gl_spin);
573 clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
579 * run_queue - process holder structures on a glock
583 static void run_queue(struct gfs2_glock *gl)
585 struct gfs2_holder *gh;
589 if (test_bit(GLF_LOCK, &gl->gl_flags))
592 if (!list_empty(&gl->gl_waiters1)) {
593 gh = list_entry(gl->gl_waiters1.next,
594 struct gfs2_holder, gh_list);
595 blocked = rq_mutex(gh);
596 } else if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
597 blocked = rq_demote(gl);
598 if (test_bit(GLF_WAITERS2, &gl->gl_flags) &&
600 set_bit(GLF_DEMOTE, &gl->gl_flags);
601 gl->gl_demote_state = LM_ST_UNLOCKED;
603 clear_bit(GLF_WAITERS2, &gl->gl_flags);
604 } else if (!list_empty(&gl->gl_waiters3)) {
605 gh = list_entry(gl->gl_waiters3.next,
606 struct gfs2_holder, gh_list);
607 blocked = rq_promote(gh);
617 * gfs2_glmutex_lock - acquire a local lock on a glock
620 * Gives caller exclusive access to manipulate a glock structure.
623 static void gfs2_glmutex_lock(struct gfs2_glock *gl)
625 spin_lock(&gl->gl_spin);
626 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
627 struct gfs2_holder gh;
629 gfs2_holder_init(gl, 0, 0, &gh);
630 set_bit(HIF_WAIT, &gh.gh_iflags);
631 list_add_tail(&gh.gh_list, &gl->gl_waiters1);
632 spin_unlock(&gl->gl_spin);
634 gfs2_holder_uninit(&gh);
636 gl->gl_owner_pid = get_pid(task_pid(current));
637 gl->gl_ip = (unsigned long)__builtin_return_address(0);
638 spin_unlock(&gl->gl_spin);
643 * gfs2_glmutex_trylock - try to acquire a local lock on a glock
646 * Returns: 1 if the glock is acquired
649 static int gfs2_glmutex_trylock(struct gfs2_glock *gl)
653 spin_lock(&gl->gl_spin);
654 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
657 gl->gl_owner_pid = get_pid(task_pid(current));
658 gl->gl_ip = (unsigned long)__builtin_return_address(0);
660 spin_unlock(&gl->gl_spin);
666 * gfs2_glmutex_unlock - release a local lock on a glock
671 static void gfs2_glmutex_unlock(struct gfs2_glock *gl)
675 spin_lock(&gl->gl_spin);
676 clear_bit(GLF_LOCK, &gl->gl_flags);
677 pid = gl->gl_owner_pid;
678 gl->gl_owner_pid = NULL;
681 spin_unlock(&gl->gl_spin);
687 * handle_callback - process a demote request
689 * @state: the state the caller wants us to change to
691 * There are only two requests that we are going to see in actual
692 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
695 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
696 int remote, unsigned long delay)
698 int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
700 spin_lock(&gl->gl_spin);
701 set_bit(bit, &gl->gl_flags);
702 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
703 gl->gl_demote_state = state;
704 gl->gl_demote_time = jiffies;
705 if (remote && gl->gl_ops->go_type == LM_TYPE_IOPEN &&
707 gfs2_glock_schedule_for_reclaim(gl);
708 spin_unlock(&gl->gl_spin);
711 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
712 gl->gl_demote_state != state) {
713 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
714 set_bit(GLF_WAITERS2, &gl->gl_flags);
716 gl->gl_demote_state = LM_ST_UNLOCKED;
718 spin_unlock(&gl->gl_spin);
722 * state_change - record that the glock is now in a different state
724 * @new_state the new state
728 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
732 held1 = (gl->gl_state != LM_ST_UNLOCKED);
733 held2 = (new_state != LM_ST_UNLOCKED);
735 if (held1 != held2) {
742 gl->gl_state = new_state;
743 gl->gl_tchange = jiffies;
747 * xmote_bh - Called after the lock module is done acquiring a lock
748 * @gl: The glock in question
749 * @ret: the int returned from the lock module
753 static void xmote_bh(struct gfs2_glock *gl, unsigned int ret)
755 struct gfs2_sbd *sdp = gl->gl_sbd;
756 const struct gfs2_glock_operations *glops = gl->gl_ops;
757 struct gfs2_holder *gh = gl->gl_req_gh;
758 int prev_state = gl->gl_state;
761 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
762 gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
763 gfs2_assert_warn(sdp, !(ret & LM_OUT_ASYNC));
765 state_change(gl, ret & LM_OUT_ST_MASK);
767 if (prev_state != LM_ST_UNLOCKED && !(ret & LM_OUT_CACHEABLE)) {
769 glops->go_inval(gl, DIO_METADATA);
770 } else if (gl->gl_state == LM_ST_DEFERRED) {
771 /* We might not want to do this here.
772 Look at moving to the inode glops. */
774 glops->go_inval(gl, 0);
777 /* Deal with each possible exit condition */
780 gl->gl_stamp = jiffies;
781 if (ret & LM_OUT_CANCELED) {
784 spin_lock(&gl->gl_spin);
785 if (gl->gl_state != gl->gl_demote_state) {
786 gl->gl_req_bh = NULL;
787 spin_unlock(&gl->gl_spin);
788 gfs2_glock_drop_th(gl);
792 gfs2_demote_wake(gl);
793 spin_unlock(&gl->gl_spin);
796 spin_lock(&gl->gl_spin);
797 list_del_init(&gh->gh_list);
799 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
801 gh->gh_error = GLR_CANCELED;
802 if (ret & LM_OUT_CANCELED)
804 if (relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
805 list_add_tail(&gh->gh_list, &gl->gl_holders);
807 set_bit(HIF_HOLDER, &gh->gh_iflags);
808 set_bit(HIF_FIRST, &gh->gh_iflags);
812 gh->gh_error = GLR_TRYFAILED;
813 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
815 gh->gh_error = -EINVAL;
816 if (gfs2_assert_withdraw(sdp, 0) == -1)
817 fs_err(sdp, "ret = 0x%.8X\n", ret);
819 spin_unlock(&gl->gl_spin);
822 if (glops->go_xmote_bh)
823 glops->go_xmote_bh(gl);
826 spin_lock(&gl->gl_spin);
827 gl->gl_req_gh = NULL;
828 gl->gl_req_bh = NULL;
829 clear_bit(GLF_LOCK, &gl->gl_flags);
830 spin_unlock(&gl->gl_spin);
836 gfs2_holder_wake(gh);
840 * gfs2_glock_xmote_th - Call into the lock module to acquire or change a glock
841 * @gl: The glock in question
842 * @state: the requested state
843 * @flags: modifier flags to the lock call
847 static void gfs2_glock_xmote_th(struct gfs2_glock *gl, struct gfs2_holder *gh)
849 struct gfs2_sbd *sdp = gl->gl_sbd;
850 int flags = gh ? gh->gh_flags : 0;
851 unsigned state = gh ? gh->gh_state : gl->gl_demote_state;
852 const struct gfs2_glock_operations *glops = gl->gl_ops;
853 int lck_flags = flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB |
854 LM_FLAG_NOEXP | LM_FLAG_ANY |
856 unsigned int lck_ret;
858 if (glops->go_xmote_th)
859 glops->go_xmote_th(gl);
861 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
862 gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
863 gfs2_assert_warn(sdp, state != LM_ST_UNLOCKED);
864 gfs2_assert_warn(sdp, state != gl->gl_state);
867 gl->gl_req_bh = xmote_bh;
869 lck_ret = gfs2_lm_lock(sdp, gl->gl_lock, gl->gl_state, state, lck_flags);
871 if (gfs2_assert_withdraw(sdp, !(lck_ret & LM_OUT_ERROR)))
874 if (lck_ret & LM_OUT_ASYNC)
875 gfs2_assert_warn(sdp, lck_ret == LM_OUT_ASYNC);
877 xmote_bh(gl, lck_ret);
881 * drop_bh - Called after a lock module unlock completes
883 * @ret: the return status
885 * Doesn't wake up the process waiting on the struct gfs2_holder (if any)
886 * Doesn't drop the reference on the glock the top half took out
890 static void drop_bh(struct gfs2_glock *gl, unsigned int ret)
892 struct gfs2_sbd *sdp = gl->gl_sbd;
893 const struct gfs2_glock_operations *glops = gl->gl_ops;
894 struct gfs2_holder *gh = gl->gl_req_gh;
896 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
897 gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
898 gfs2_assert_warn(sdp, !ret);
900 state_change(gl, LM_ST_UNLOCKED);
903 glops->go_inval(gl, DIO_METADATA);
906 spin_lock(&gl->gl_spin);
907 list_del_init(&gh->gh_list);
909 spin_unlock(&gl->gl_spin);
912 spin_lock(&gl->gl_spin);
913 gfs2_demote_wake(gl);
914 gl->gl_req_gh = NULL;
915 gl->gl_req_bh = NULL;
916 clear_bit(GLF_LOCK, &gl->gl_flags);
917 spin_unlock(&gl->gl_spin);
922 gfs2_holder_wake(gh);
926 * gfs2_glock_drop_th - call into the lock module to unlock a lock
931 static void gfs2_glock_drop_th(struct gfs2_glock *gl)
933 struct gfs2_sbd *sdp = gl->gl_sbd;
934 const struct gfs2_glock_operations *glops = gl->gl_ops;
937 if (glops->go_xmote_th)
938 glops->go_xmote_th(gl);
940 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
941 gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
942 gfs2_assert_warn(sdp, gl->gl_state != LM_ST_UNLOCKED);
945 gl->gl_req_bh = drop_bh;
947 ret = gfs2_lm_unlock(sdp, gl->gl_lock, gl->gl_state);
949 if (gfs2_assert_withdraw(sdp, !(ret & LM_OUT_ERROR)))
955 gfs2_assert_warn(sdp, ret == LM_OUT_ASYNC);
959 * do_cancels - cancel requests for locks stuck waiting on an expire flag
960 * @gh: the LM_FLAG_PRIORITY holder waiting to acquire the lock
962 * Don't cancel GL_NOCANCEL requests.
965 static void do_cancels(struct gfs2_holder *gh)
967 struct gfs2_glock *gl = gh->gh_gl;
969 spin_lock(&gl->gl_spin);
971 while (gl->gl_req_gh != gh &&
972 !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
973 !list_empty(&gh->gh_list)) {
974 if (gl->gl_req_bh && !(gl->gl_req_gh &&
975 (gl->gl_req_gh->gh_flags & GL_NOCANCEL))) {
976 spin_unlock(&gl->gl_spin);
977 gfs2_lm_cancel(gl->gl_sbd, gl->gl_lock);
979 spin_lock(&gl->gl_spin);
981 spin_unlock(&gl->gl_spin);
983 spin_lock(&gl->gl_spin);
987 spin_unlock(&gl->gl_spin);
991 * glock_wait_internal - wait on a glock acquisition
992 * @gh: the glock holder
994 * Returns: 0 on success
997 static int glock_wait_internal(struct gfs2_holder *gh)
999 struct gfs2_glock *gl = gh->gh_gl;
1000 struct gfs2_sbd *sdp = gl->gl_sbd;
1001 const struct gfs2_glock_operations *glops = gl->gl_ops;
1003 if (test_bit(HIF_ABORTED, &gh->gh_iflags))
1006 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1007 spin_lock(&gl->gl_spin);
1008 if (gl->gl_req_gh != gh &&
1009 !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1010 !list_empty(&gh->gh_list)) {
1011 list_del_init(&gh->gh_list);
1012 gh->gh_error = GLR_TRYFAILED;
1014 spin_unlock(&gl->gl_spin);
1015 return gh->gh_error;
1017 spin_unlock(&gl->gl_spin);
1020 if (gh->gh_flags & LM_FLAG_PRIORITY)
1025 return gh->gh_error;
1027 gfs2_assert_withdraw(sdp, test_bit(HIF_HOLDER, &gh->gh_iflags));
1028 gfs2_assert_withdraw(sdp, relaxed_state_ok(gl->gl_state, gh->gh_state,
1031 if (test_bit(HIF_FIRST, &gh->gh_iflags)) {
1032 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1034 if (glops->go_lock) {
1035 gh->gh_error = glops->go_lock(gh);
1037 spin_lock(&gl->gl_spin);
1038 list_del_init(&gh->gh_list);
1039 spin_unlock(&gl->gl_spin);
1043 spin_lock(&gl->gl_spin);
1044 gl->gl_req_gh = NULL;
1045 gl->gl_req_bh = NULL;
1046 clear_bit(GLF_LOCK, &gl->gl_flags);
1048 spin_unlock(&gl->gl_spin);
1051 return gh->gh_error;
1054 static inline struct gfs2_holder *
1055 find_holder_by_owner(struct list_head *head, struct pid *pid)
1057 struct gfs2_holder *gh;
1059 list_for_each_entry(gh, head, gh_list) {
1060 if (gh->gh_owner_pid == pid)
1067 static void print_dbg(struct glock_iter *gi, const char *fmt, ...)
1071 va_start(args, fmt);
1073 vsprintf(gi->string, fmt, args);
1074 seq_printf(gi->seq, gi->string);
1082 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1083 * @gh: the holder structure to add
1087 static void add_to_queue(struct gfs2_holder *gh)
1089 struct gfs2_glock *gl = gh->gh_gl;
1090 struct gfs2_holder *existing;
1092 BUG_ON(gh->gh_owner_pid == NULL);
1093 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1096 if (!(gh->gh_flags & GL_FLOCK)) {
1097 existing = find_holder_by_owner(&gl->gl_holders,
1100 print_symbol(KERN_WARNING "original: %s\n",
1102 printk(KERN_INFO "pid : %d\n",
1103 pid_nr(existing->gh_owner_pid));
1104 printk(KERN_INFO "lock type : %d lock state : %d\n",
1105 existing->gh_gl->gl_name.ln_type,
1106 existing->gh_gl->gl_state);
1107 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1108 printk(KERN_INFO "pid : %d\n",
1109 pid_nr(gh->gh_owner_pid));
1110 printk(KERN_INFO "lock type : %d lock state : %d\n",
1111 gl->gl_name.ln_type, gl->gl_state);
1115 existing = find_holder_by_owner(&gl->gl_waiters3,
1118 print_symbol(KERN_WARNING "original: %s\n",
1120 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1125 if (gh->gh_flags & LM_FLAG_PRIORITY)
1126 list_add(&gh->gh_list, &gl->gl_waiters3);
1128 list_add_tail(&gh->gh_list, &gl->gl_waiters3);
1132 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1133 * @gh: the holder structure
1135 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1137 * Returns: 0, GLR_TRYFAILED, or errno on failure
1140 int gfs2_glock_nq(struct gfs2_holder *gh)
1142 struct gfs2_glock *gl = gh->gh_gl;
1143 struct gfs2_sbd *sdp = gl->gl_sbd;
1147 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1148 set_bit(HIF_ABORTED, &gh->gh_iflags);
1152 spin_lock(&gl->gl_spin);
1155 spin_unlock(&gl->gl_spin);
1157 if (!(gh->gh_flags & GL_ASYNC)) {
1158 error = glock_wait_internal(gh);
1159 if (error == GLR_CANCELED) {
1169 * gfs2_glock_poll - poll to see if an async request has been completed
1172 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1175 int gfs2_glock_poll(struct gfs2_holder *gh)
1177 struct gfs2_glock *gl = gh->gh_gl;
1180 spin_lock(&gl->gl_spin);
1182 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1184 else if (list_empty(&gh->gh_list)) {
1185 if (gh->gh_error == GLR_CANCELED) {
1186 spin_unlock(&gl->gl_spin);
1188 if (gfs2_glock_nq(gh))
1195 spin_unlock(&gl->gl_spin);
1201 * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
1202 * @gh: the holder structure
1204 * Returns: 0, GLR_TRYFAILED, or errno on failure
1207 int gfs2_glock_wait(struct gfs2_holder *gh)
1211 error = glock_wait_internal(gh);
1212 if (error == GLR_CANCELED) {
1214 gh->gh_flags &= ~GL_ASYNC;
1215 error = gfs2_glock_nq(gh);
1222 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1223 * @gh: the glock holder
1227 void gfs2_glock_dq(struct gfs2_holder *gh)
1229 struct gfs2_glock *gl = gh->gh_gl;
1230 const struct gfs2_glock_operations *glops = gl->gl_ops;
1233 if (gh->gh_flags & GL_NOCACHE)
1234 handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1236 gfs2_glmutex_lock(gl);
1238 spin_lock(&gl->gl_spin);
1239 list_del_init(&gh->gh_list);
1241 if (list_empty(&gl->gl_holders)) {
1242 if (glops->go_unlock) {
1243 spin_unlock(&gl->gl_spin);
1244 glops->go_unlock(gh);
1245 spin_lock(&gl->gl_spin);
1247 gl->gl_stamp = jiffies;
1250 clear_bit(GLF_LOCK, &gl->gl_flags);
1251 spin_unlock(&gl->gl_spin);
1253 gfs2_glock_hold(gl);
1254 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1255 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1256 delay = gl->gl_ops->go_min_hold_time;
1257 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1261 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1263 struct gfs2_glock *gl = gh->gh_gl;
1269 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1270 * @gh: the holder structure
1274 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1277 gfs2_holder_uninit(gh);
1281 * gfs2_glock_nq_num - acquire a glock based on lock number
1282 * @sdp: the filesystem
1283 * @number: the lock number
1284 * @glops: the glock operations for the type of glock
1285 * @state: the state to acquire the glock in
1286 * @flags: modifier flags for the aquisition
1287 * @gh: the struct gfs2_holder
1292 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1293 const struct gfs2_glock_operations *glops,
1294 unsigned int state, int flags, struct gfs2_holder *gh)
1296 struct gfs2_glock *gl;
1299 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1301 error = gfs2_glock_nq_init(gl, state, flags, gh);
1309 * glock_compare - Compare two struct gfs2_glock structures for sorting
1310 * @arg_a: the first structure
1311 * @arg_b: the second structure
1315 static int glock_compare(const void *arg_a, const void *arg_b)
1317 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1318 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1319 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1320 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1322 if (a->ln_number > b->ln_number)
1324 if (a->ln_number < b->ln_number)
1326 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1331 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1332 * @num_gh: the number of structures
1333 * @ghs: an array of struct gfs2_holder structures
1335 * Returns: 0 on success (all glocks acquired),
1336 * errno on failure (no glocks acquired)
1339 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1340 struct gfs2_holder **p)
1345 for (x = 0; x < num_gh; x++)
1348 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1350 for (x = 0; x < num_gh; x++) {
1351 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1353 error = gfs2_glock_nq(p[x]);
1356 gfs2_glock_dq(p[x]);
1365 * gfs2_glock_nq_m - acquire multiple glocks
1366 * @num_gh: the number of structures
1367 * @ghs: an array of struct gfs2_holder structures
1370 * Returns: 0 on success (all glocks acquired),
1371 * errno on failure (no glocks acquired)
1374 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1376 struct gfs2_holder *tmp[4];
1377 struct gfs2_holder **pph = tmp;
1384 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1385 return gfs2_glock_nq(ghs);
1389 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1394 error = nq_m_sync(num_gh, ghs, pph);
1403 * gfs2_glock_dq_m - release multiple glocks
1404 * @num_gh: the number of structures
1405 * @ghs: an array of struct gfs2_holder structures
1409 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1413 for (x = 0; x < num_gh; x++)
1414 gfs2_glock_dq(&ghs[x]);
1418 * gfs2_glock_dq_uninit_m - release multiple glocks
1419 * @num_gh: the number of structures
1420 * @ghs: an array of struct gfs2_holder structures
1424 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1428 for (x = 0; x < num_gh; x++)
1429 gfs2_glock_dq_uninit(&ghs[x]);
1433 * gfs2_lvb_hold - attach a LVB from a glock
1434 * @gl: The glock in question
1438 int gfs2_lvb_hold(struct gfs2_glock *gl)
1442 gfs2_glmutex_lock(gl);
1444 if (!atomic_read(&gl->gl_lvb_count)) {
1445 error = gfs2_lm_hold_lvb(gl->gl_sbd, gl->gl_lock, &gl->gl_lvb);
1447 gfs2_glmutex_unlock(gl);
1450 gfs2_glock_hold(gl);
1452 atomic_inc(&gl->gl_lvb_count);
1454 gfs2_glmutex_unlock(gl);
1460 * gfs2_lvb_unhold - detach a LVB from a glock
1461 * @gl: The glock in question
1465 void gfs2_lvb_unhold(struct gfs2_glock *gl)
1467 gfs2_glock_hold(gl);
1468 gfs2_glmutex_lock(gl);
1470 gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count) > 0);
1471 if (atomic_dec_and_test(&gl->gl_lvb_count)) {
1472 gfs2_lm_unhold_lvb(gl->gl_sbd, gl->gl_lock, gl->gl_lvb);
1477 gfs2_glmutex_unlock(gl);
1481 static void blocking_cb(struct gfs2_sbd *sdp, struct lm_lockname *name,
1484 struct gfs2_glock *gl;
1485 unsigned long delay = 0;
1486 unsigned long holdtime;
1487 unsigned long now = jiffies;
1489 gl = gfs2_glock_find(sdp, name);
1493 holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
1494 if (time_before(now, holdtime))
1495 delay = holdtime - now;
1497 handle_callback(gl, state, 1, delay);
1498 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1503 * gfs2_glock_cb - Callback used by locking module
1504 * @sdp: Pointer to the superblock
1505 * @type: Type of callback
1506 * @data: Type dependent data pointer
1508 * Called by the locking module when it wants to tell us something.
1509 * Either we need to drop a lock, one of our ASYNC requests completed, or
1510 * a journal from another client needs to be recovered.
1513 void gfs2_glock_cb(void *cb_data, unsigned int type, void *data)
1515 struct gfs2_sbd *sdp = cb_data;
1519 blocking_cb(sdp, data, LM_ST_UNLOCKED);
1523 blocking_cb(sdp, data, LM_ST_DEFERRED);
1527 blocking_cb(sdp, data, LM_ST_SHARED);
1531 struct lm_async_cb *async = data;
1532 struct gfs2_glock *gl;
1534 down_read(&gfs2_umount_flush_sem);
1535 gl = gfs2_glock_find(sdp, &async->lc_name);
1536 if (gfs2_assert_warn(sdp, gl))
1538 if (!gfs2_assert_warn(sdp, gl->gl_req_bh))
1539 gl->gl_req_bh(gl, async->lc_ret);
1540 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1542 up_read(&gfs2_umount_flush_sem);
1546 case LM_CB_NEED_RECOVERY:
1547 gfs2_jdesc_make_dirty(sdp, *(unsigned int *)data);
1548 if (sdp->sd_recoverd_process)
1549 wake_up_process(sdp->sd_recoverd_process);
1552 case LM_CB_DROPLOCKS:
1553 gfs2_gl_hash_clear(sdp, NO_WAIT);
1554 gfs2_quota_scan(sdp);
1558 gfs2_assert_warn(sdp, 0);
1564 * demote_ok - Check to see if it's ok to unlock a glock
1567 * Returns: 1 if it's ok
1570 static int demote_ok(struct gfs2_glock *gl)
1572 const struct gfs2_glock_operations *glops = gl->gl_ops;
1575 if (test_bit(GLF_STICKY, &gl->gl_flags))
1577 else if (glops->go_demote_ok)
1578 demote = glops->go_demote_ok(gl);
1584 * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1589 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
1591 struct gfs2_sbd *sdp = gl->gl_sbd;
1593 spin_lock(&sdp->sd_reclaim_lock);
1594 if (list_empty(&gl->gl_reclaim)) {
1595 gfs2_glock_hold(gl);
1596 list_add(&gl->gl_reclaim, &sdp->sd_reclaim_list);
1597 atomic_inc(&sdp->sd_reclaim_count);
1598 spin_unlock(&sdp->sd_reclaim_lock);
1599 wake_up(&sdp->sd_reclaim_wq);
1601 spin_unlock(&sdp->sd_reclaim_lock);
1605 * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1606 * @sdp: the filesystem
1608 * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1609 * different glock and we notice that there are a lot of glocks in the
1614 void gfs2_reclaim_glock(struct gfs2_sbd *sdp)
1616 struct gfs2_glock *gl;
1618 spin_lock(&sdp->sd_reclaim_lock);
1619 if (list_empty(&sdp->sd_reclaim_list)) {
1620 spin_unlock(&sdp->sd_reclaim_lock);
1623 gl = list_entry(sdp->sd_reclaim_list.next,
1624 struct gfs2_glock, gl_reclaim);
1625 list_del_init(&gl->gl_reclaim);
1626 spin_unlock(&sdp->sd_reclaim_lock);
1628 atomic_dec(&sdp->sd_reclaim_count);
1629 atomic_inc(&sdp->sd_reclaimed);
1631 if (gfs2_glmutex_trylock(gl)) {
1632 if (list_empty(&gl->gl_holders) &&
1633 gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1634 handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1635 gfs2_glmutex_unlock(gl);
1642 * examine_bucket - Call a function for glock in a hash bucket
1643 * @examiner: the function
1644 * @sdp: the filesystem
1645 * @bucket: the bucket
1647 * Returns: 1 if the bucket has entries
1650 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1653 struct gfs2_glock *gl, *prev = NULL;
1654 int has_entries = 0;
1655 struct hlist_head *head = &gl_hash_table[hash].hb_list;
1657 read_lock(gl_lock_addr(hash));
1658 /* Can't use hlist_for_each_entry - don't want prefetch here */
1659 if (hlist_empty(head))
1661 gl = list_entry(head->first, struct gfs2_glock, gl_list);
1663 if (!sdp || gl->gl_sbd == sdp) {
1664 gfs2_glock_hold(gl);
1665 read_unlock(gl_lock_addr(hash));
1667 gfs2_glock_put(prev);
1671 read_lock(gl_lock_addr(hash));
1673 if (gl->gl_list.next == NULL)
1675 gl = list_entry(gl->gl_list.next, struct gfs2_glock, gl_list);
1678 read_unlock(gl_lock_addr(hash));
1680 gfs2_glock_put(prev);
1686 * scan_glock - look at a glock and see if we can reclaim it
1687 * @gl: the glock to look at
1691 static void scan_glock(struct gfs2_glock *gl)
1693 if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object)
1696 if (gfs2_glmutex_trylock(gl)) {
1697 if (list_empty(&gl->gl_holders) &&
1698 gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1700 gfs2_glmutex_unlock(gl);
1705 gfs2_glmutex_unlock(gl);
1706 gfs2_glock_schedule_for_reclaim(gl);
1710 * clear_glock - look at a glock and see if we can free it from glock cache
1711 * @gl: the glock to look at
1715 static void clear_glock(struct gfs2_glock *gl)
1717 struct gfs2_sbd *sdp = gl->gl_sbd;
1720 spin_lock(&sdp->sd_reclaim_lock);
1721 if (!list_empty(&gl->gl_reclaim)) {
1722 list_del_init(&gl->gl_reclaim);
1723 atomic_dec(&sdp->sd_reclaim_count);
1724 spin_unlock(&sdp->sd_reclaim_lock);
1725 released = gfs2_glock_put(gl);
1726 gfs2_assert(sdp, !released);
1728 spin_unlock(&sdp->sd_reclaim_lock);
1731 if (gfs2_glmutex_trylock(gl)) {
1732 if (list_empty(&gl->gl_holders) &&
1733 gl->gl_state != LM_ST_UNLOCKED)
1734 handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1735 gfs2_glmutex_unlock(gl);
1740 * gfs2_gl_hash_clear - Empty out the glock hash table
1741 * @sdp: the filesystem
1742 * @wait: wait until it's all gone
1744 * Called when unmounting the filesystem, or when inter-node lock manager
1745 * requests DROPLOCKS because it is running out of capacity.
1748 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp, int wait)
1758 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1759 if (examine_bucket(clear_glock, sdp, x))
1766 if (time_after_eq(jiffies,
1767 t + gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
1768 fs_warn(sdp, "Unmount seems to be stalled. "
1769 "Dumping lock state...\n");
1770 gfs2_dump_lockstate(sdp);
1774 down_write(&gfs2_umount_flush_sem);
1775 invalidate_inodes(sdp->sd_vfs);
1776 up_write(&gfs2_umount_flush_sem);
1782 * Diagnostic routines to help debug distributed deadlock
1785 static void gfs2_print_symbol(struct glock_iter *gi, const char *fmt,
1786 unsigned long address)
1788 char buffer[KSYM_SYMBOL_LEN];
1790 sprint_symbol(buffer, address);
1791 print_dbg(gi, fmt, buffer);
1795 * dump_holder - print information about a glock holder
1796 * @str: a string naming the type of holder
1797 * @gh: the glock holder
1799 * Returns: 0 on success, -ENOBUFS when we run out of space
1802 static int dump_holder(struct glock_iter *gi, char *str,
1803 struct gfs2_holder *gh)
1806 struct task_struct *gh_owner;
1808 print_dbg(gi, " %s\n", str);
1809 if (gh->gh_owner_pid) {
1810 print_dbg(gi, " owner = %ld ",
1811 (long)pid_nr(gh->gh_owner_pid));
1812 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1814 print_dbg(gi, "(%s)\n", gh_owner->comm);
1816 print_dbg(gi, "(ended)\n");
1818 print_dbg(gi, " owner = -1\n");
1819 print_dbg(gi, " gh_state = %u\n", gh->gh_state);
1820 print_dbg(gi, " gh_flags =");
1821 for (x = 0; x < 32; x++)
1822 if (gh->gh_flags & (1 << x))
1823 print_dbg(gi, " %u", x);
1824 print_dbg(gi, " \n");
1825 print_dbg(gi, " error = %d\n", gh->gh_error);
1826 print_dbg(gi, " gh_iflags =");
1827 for (x = 0; x < 32; x++)
1828 if (test_bit(x, &gh->gh_iflags))
1829 print_dbg(gi, " %u", x);
1830 print_dbg(gi, " \n");
1831 gfs2_print_symbol(gi, " initialized at: %s\n", gh->gh_ip);
1837 * dump_inode - print information about an inode
1840 * Returns: 0 on success, -ENOBUFS when we run out of space
1843 static int dump_inode(struct glock_iter *gi, struct gfs2_inode *ip)
1847 print_dbg(gi, " Inode:\n");
1848 print_dbg(gi, " num = %llu/%llu\n",
1849 (unsigned long long)ip->i_no_formal_ino,
1850 (unsigned long long)ip->i_no_addr);
1851 print_dbg(gi, " type = %u\n", IF2DT(ip->i_inode.i_mode));
1852 print_dbg(gi, " i_flags =");
1853 for (x = 0; x < 32; x++)
1854 if (test_bit(x, &ip->i_flags))
1855 print_dbg(gi, " %u", x);
1856 print_dbg(gi, " \n");
1861 * dump_glock - print information about a glock
1863 * @count: where we are in the buffer
1865 * Returns: 0 on success, -ENOBUFS when we run out of space
1868 static int dump_glock(struct glock_iter *gi, struct gfs2_glock *gl)
1870 struct gfs2_holder *gh;
1872 int error = -ENOBUFS;
1873 struct task_struct *gl_owner;
1875 spin_lock(&gl->gl_spin);
1877 print_dbg(gi, "Glock 0x%p (%u, 0x%llx)\n", gl, gl->gl_name.ln_type,
1878 (unsigned long long)gl->gl_name.ln_number);
1879 print_dbg(gi, " gl_flags =");
1880 for (x = 0; x < 32; x++) {
1881 if (test_bit(x, &gl->gl_flags))
1882 print_dbg(gi, " %u", x);
1884 if (!test_bit(GLF_LOCK, &gl->gl_flags))
1885 print_dbg(gi, " (unlocked)");
1886 print_dbg(gi, " \n");
1887 print_dbg(gi, " gl_ref = %d\n", atomic_read(&gl->gl_ref));
1888 print_dbg(gi, " gl_state = %u\n", gl->gl_state);
1889 if (gl->gl_owner_pid) {
1890 gl_owner = pid_task(gl->gl_owner_pid, PIDTYPE_PID);
1892 print_dbg(gi, " gl_owner = pid %d (%s)\n",
1893 pid_nr(gl->gl_owner_pid), gl_owner->comm);
1895 print_dbg(gi, " gl_owner = %d (ended)\n",
1896 pid_nr(gl->gl_owner_pid));
1898 print_dbg(gi, " gl_owner = -1\n");
1899 print_dbg(gi, " gl_ip = %lu\n", gl->gl_ip);
1900 print_dbg(gi, " req_gh = %s\n", (gl->gl_req_gh) ? "yes" : "no");
1901 print_dbg(gi, " req_bh = %s\n", (gl->gl_req_bh) ? "yes" : "no");
1902 print_dbg(gi, " lvb_count = %d\n", atomic_read(&gl->gl_lvb_count));
1903 print_dbg(gi, " object = %s\n", (gl->gl_object) ? "yes" : "no");
1904 print_dbg(gi, " reclaim = %s\n",
1905 (list_empty(&gl->gl_reclaim)) ? "no" : "yes");
1907 print_dbg(gi, " aspace = 0x%p nrpages = %lu\n", gl->gl_aspace,
1908 gl->gl_aspace->i_mapping->nrpages);
1910 print_dbg(gi, " aspace = no\n");
1911 print_dbg(gi, " ail = %d\n", atomic_read(&gl->gl_ail_count));
1912 if (gl->gl_req_gh) {
1913 error = dump_holder(gi, "Request", gl->gl_req_gh);
1917 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1918 error = dump_holder(gi, "Holder", gh);
1922 list_for_each_entry(gh, &gl->gl_waiters1, gh_list) {
1923 error = dump_holder(gi, "Waiter1", gh);
1927 list_for_each_entry(gh, &gl->gl_waiters3, gh_list) {
1928 error = dump_holder(gi, "Waiter3", gh);
1932 if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
1933 print_dbg(gi, " Demotion req to state %u (%llu uS ago)\n",
1934 gl->gl_demote_state, (unsigned long long)
1935 (jiffies - gl->gl_demote_time)*(1000000/HZ));
1937 if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object) {
1938 if (!test_bit(GLF_LOCK, &gl->gl_flags) &&
1939 list_empty(&gl->gl_holders)) {
1940 error = dump_inode(gi, gl->gl_object);
1945 print_dbg(gi, " Inode: busy\n");
1952 spin_unlock(&gl->gl_spin);
1957 * gfs2_dump_lockstate - print out the current lockstate
1958 * @sdp: the filesystem
1959 * @ub: the buffer to copy the information into
1961 * If @ub is NULL, dump the lockstate to the console.
1965 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
1967 struct gfs2_glock *gl;
1968 struct hlist_node *h;
1972 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1974 read_lock(gl_lock_addr(x));
1976 hlist_for_each_entry(gl, h, &gl_hash_table[x].hb_list, gl_list) {
1977 if (gl->gl_sbd != sdp)
1980 error = dump_glock(NULL, gl);
1985 read_unlock(gl_lock_addr(x));
1996 * gfs2_scand - Look for cached glocks and inodes to toss from memory
1997 * @sdp: Pointer to GFS2 superblock
1999 * One of these daemons runs, finding candidates to add to sd_reclaim_list.
2003 static int gfs2_scand(void *data)
2008 while (!kthread_should_stop()) {
2009 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
2010 examine_bucket(scan_glock, NULL, x);
2011 if (freezing(current))
2016 schedule_timeout_interruptible(delay * HZ);
2024 int __init gfs2_glock_init(void)
2027 for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
2028 INIT_HLIST_HEAD(&gl_hash_table[i].hb_list);
2030 #ifdef GL_HASH_LOCK_SZ
2031 for(i = 0; i < GL_HASH_LOCK_SZ; i++) {
2032 rwlock_init(&gl_hash_locks[i]);
2036 scand_process = kthread_run(gfs2_scand, NULL, "gfs2_scand");
2037 if (IS_ERR(scand_process))
2038 return PTR_ERR(scand_process);
2040 glock_workqueue = create_workqueue("glock_workqueue");
2041 if (IS_ERR(glock_workqueue)) {
2042 kthread_stop(scand_process);
2043 return PTR_ERR(glock_workqueue);
2049 void gfs2_glock_exit(void)
2051 destroy_workqueue(glock_workqueue);
2052 kthread_stop(scand_process);
2055 module_param(scand_secs, uint, S_IRUGO|S_IWUSR);
2056 MODULE_PARM_DESC(scand_secs, "The number of seconds between scand runs");
2058 static int gfs2_glock_iter_next(struct glock_iter *gi)
2060 struct gfs2_glock *gl;
2063 read_lock(gl_lock_addr(gi->hash));
2066 gi->gl = hlist_entry(gl->gl_list.next,
2067 struct gfs2_glock, gl_list);
2069 gfs2_glock_hold(gi->gl);
2071 read_unlock(gl_lock_addr(gi->hash));
2074 if (gl && gi->gl == NULL)
2076 while(gi->gl == NULL) {
2077 if (gi->hash >= GFS2_GL_HASH_SIZE)
2079 read_lock(gl_lock_addr(gi->hash));
2080 gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
2081 struct gfs2_glock, gl_list);
2083 gfs2_glock_hold(gi->gl);
2084 read_unlock(gl_lock_addr(gi->hash));
2088 if (gi->sdp != gi->gl->gl_sbd)
2094 static void gfs2_glock_iter_free(struct glock_iter *gi)
2097 gfs2_glock_put(gi->gl);
2101 static struct glock_iter *gfs2_glock_iter_init(struct gfs2_sbd *sdp)
2103 struct glock_iter *gi;
2105 gi = kmalloc(sizeof (*gi), GFP_KERNEL);
2113 memset(gi->string, 0, sizeof(gi->string));
2115 if (gfs2_glock_iter_next(gi)) {
2116 gfs2_glock_iter_free(gi);
2123 static void *gfs2_glock_seq_start(struct seq_file *file, loff_t *pos)
2125 struct glock_iter *gi;
2128 gi = gfs2_glock_iter_init(file->private);
2133 if (gfs2_glock_iter_next(gi)) {
2134 gfs2_glock_iter_free(gi);
2142 static void *gfs2_glock_seq_next(struct seq_file *file, void *iter_ptr,
2145 struct glock_iter *gi = iter_ptr;
2149 if (gfs2_glock_iter_next(gi)) {
2150 gfs2_glock_iter_free(gi);
2157 static void gfs2_glock_seq_stop(struct seq_file *file, void *iter_ptr)
2159 struct glock_iter *gi = iter_ptr;
2161 gfs2_glock_iter_free(gi);
2164 static int gfs2_glock_seq_show(struct seq_file *file, void *iter_ptr)
2166 struct glock_iter *gi = iter_ptr;
2169 dump_glock(gi, gi->gl);
2174 static const struct seq_operations gfs2_glock_seq_ops = {
2175 .start = gfs2_glock_seq_start,
2176 .next = gfs2_glock_seq_next,
2177 .stop = gfs2_glock_seq_stop,
2178 .show = gfs2_glock_seq_show,
2181 static int gfs2_debugfs_open(struct inode *inode, struct file *file)
2183 struct seq_file *seq;
2186 ret = seq_open(file, &gfs2_glock_seq_ops);
2190 seq = file->private_data;
2191 seq->private = inode->i_private;
2196 static const struct file_operations gfs2_debug_fops = {
2197 .owner = THIS_MODULE,
2198 .open = gfs2_debugfs_open,
2200 .llseek = seq_lseek,
2201 .release = seq_release
2204 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2206 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2207 if (!sdp->debugfs_dir)
2209 sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
2211 sdp->debugfs_dir, sdp,
2213 if (!sdp->debugfs_dentry_glocks)
2219 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2221 if (sdp && sdp->debugfs_dir) {
2222 if (sdp->debugfs_dentry_glocks) {
2223 debugfs_remove(sdp->debugfs_dentry_glocks);
2224 sdp->debugfs_dentry_glocks = NULL;
2226 debugfs_remove(sdp->debugfs_dir);
2227 sdp->debugfs_dir = NULL;
2231 int gfs2_register_debugfs(void)
2233 gfs2_root = debugfs_create_dir("gfs2", NULL);
2234 return gfs2_root ? 0 : -ENOMEM;
2237 void gfs2_unregister_debugfs(void)
2239 debugfs_remove(gfs2_root);