2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
10 #include <linux/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/completion.h>
14 #include <linux/buffer_head.h>
15 #include <linux/delay.h>
16 #include <linux/sort.h>
17 #include <linux/jhash.h>
18 #include <linux/kallsyms.h>
19 #include <linux/gfs2_ondisk.h>
20 #include <linux/list.h>
21 #include <linux/lm_interface.h>
22 #include <asm/uaccess.h>
37 struct gfs2_holder gr_gh;
38 struct work_struct gr_work;
41 struct gfs2_gl_hash_bucket {
42 struct hlist_head hb_list;
45 typedef void (*glock_examiner) (struct gfs2_glock * gl);
47 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp);
48 static int dump_glock(struct gfs2_glock *gl);
49 static int dump_inode(struct gfs2_inode *ip);
51 #define GFS2_GL_HASH_SHIFT 15
52 #define GFS2_GL_HASH_SIZE (1 << GFS2_GL_HASH_SHIFT)
53 #define GFS2_GL_HASH_MASK (GFS2_GL_HASH_SIZE - 1)
55 static struct gfs2_gl_hash_bucket gl_hash_table[GFS2_GL_HASH_SIZE];
58 * Despite what you might think, the numbers below are not arbitrary :-)
59 * They are taken from the ipv4 routing hash code, which is well tested
60 * and thus should be nearly optimal. Later on we might tweek the numbers
61 * but for now this should be fine.
63 * The reason for putting the locks in a separate array from the list heads
64 * is that we can have fewer locks than list heads and save memory. We use
65 * the same hash function for both, but with a different hash mask.
67 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
68 defined(CONFIG_PROVE_LOCKING)
71 # define GL_HASH_LOCK_SZ 256
74 # define GL_HASH_LOCK_SZ 4096
76 # define GL_HASH_LOCK_SZ 2048
78 # define GL_HASH_LOCK_SZ 1024
80 # define GL_HASH_LOCK_SZ 512
82 # define GL_HASH_LOCK_SZ 256
86 /* We never want more locks than chains */
87 #if GFS2_GL_HASH_SIZE < GL_HASH_LOCK_SZ
88 # undef GL_HASH_LOCK_SZ
89 # define GL_HASH_LOCK_SZ GFS2_GL_HASH_SIZE
92 static rwlock_t gl_hash_locks[GL_HASH_LOCK_SZ];
94 static inline rwlock_t *gl_lock_addr(unsigned int x)
96 return &gl_hash_locks[x & (GL_HASH_LOCK_SZ-1)];
98 #else /* not SMP, so no spinlocks required */
99 static inline rwlock_t *gl_lock_addr(x)
106 * relaxed_state_ok - is a requested lock compatible with the current lock mode?
107 * @actual: the current state of the lock
108 * @requested: the lock state that was requested by the caller
109 * @flags: the modifier flags passed in by the caller
111 * Returns: 1 if the locks are compatible, 0 otherwise
114 static inline int relaxed_state_ok(unsigned int actual, unsigned requested,
117 if (actual == requested)
120 if (flags & GL_EXACT)
123 if (actual == LM_ST_EXCLUSIVE && requested == LM_ST_SHARED)
126 if (actual != LM_ST_UNLOCKED && (flags & LM_FLAG_ANY))
133 * gl_hash() - Turn glock number into hash bucket number
134 * @lock: The glock number
136 * Returns: The number of the corresponding hash bucket
139 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
140 const struct lm_lockname *name)
144 h = jhash(&name->ln_number, sizeof(u64), 0);
145 h = jhash(&name->ln_type, sizeof(unsigned int), h);
146 h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
147 h &= GFS2_GL_HASH_MASK;
153 * glock_free() - Perform a few checks and then release struct gfs2_glock
154 * @gl: The glock to release
156 * Also calls lock module to release its internal structure for this glock.
160 static void glock_free(struct gfs2_glock *gl)
162 struct gfs2_sbd *sdp = gl->gl_sbd;
163 struct inode *aspace = gl->gl_aspace;
165 gfs2_lm_put_lock(sdp, gl->gl_lock);
168 gfs2_aspace_put(aspace);
170 kmem_cache_free(gfs2_glock_cachep, gl);
174 * gfs2_glock_hold() - increment reference count on glock
175 * @gl: The glock to hold
179 void gfs2_glock_hold(struct gfs2_glock *gl)
181 atomic_inc(&gl->gl_ref);
185 * gfs2_glock_put() - Decrement reference count on glock
186 * @gl: The glock to put
190 int gfs2_glock_put(struct gfs2_glock *gl)
193 struct gfs2_sbd *sdp = gl->gl_sbd;
195 write_lock(gl_lock_addr(gl->gl_hash));
196 if (atomic_dec_and_test(&gl->gl_ref)) {
197 hlist_del(&gl->gl_list);
198 write_unlock(gl_lock_addr(gl->gl_hash));
199 BUG_ON(spin_is_locked(&gl->gl_spin));
200 gfs2_assert(sdp, gl->gl_state == LM_ST_UNLOCKED);
201 gfs2_assert(sdp, list_empty(&gl->gl_reclaim));
202 gfs2_assert(sdp, list_empty(&gl->gl_holders));
203 gfs2_assert(sdp, list_empty(&gl->gl_waiters1));
204 gfs2_assert(sdp, list_empty(&gl->gl_waiters2));
205 gfs2_assert(sdp, list_empty(&gl->gl_waiters3));
210 write_unlock(gl_lock_addr(gl->gl_hash));
216 * queue_empty - check to see if a glock's queue is empty
218 * @head: the head of the queue to check
220 * This function protects the list in the event that a process already
221 * has a holder on the list and is adding a second holder for itself.
222 * The glmutex lock is what generally prevents processes from working
223 * on the same glock at once, but the special case of adding a second
224 * holder for yourself ("recursive" locking) doesn't involve locking
225 * glmutex, making the spin lock necessary.
227 * Returns: 1 if the queue is empty
230 static inline int queue_empty(struct gfs2_glock *gl, struct list_head *head)
233 spin_lock(&gl->gl_spin);
234 empty = list_empty(head);
235 spin_unlock(&gl->gl_spin);
240 * search_bucket() - Find struct gfs2_glock by lock number
241 * @bucket: the bucket to search
242 * @name: The lock name
244 * Returns: NULL, or the struct gfs2_glock with the requested number
247 static struct gfs2_glock *search_bucket(unsigned int hash,
248 const struct gfs2_sbd *sdp,
249 const struct lm_lockname *name)
251 struct gfs2_glock *gl;
252 struct hlist_node *h;
254 hlist_for_each_entry(gl, h, &gl_hash_table[hash].hb_list, gl_list) {
255 if (!lm_name_equal(&gl->gl_name, name))
257 if (gl->gl_sbd != sdp)
260 atomic_inc(&gl->gl_ref);
269 * gfs2_glock_find() - Find glock by lock number
270 * @sdp: The GFS2 superblock
271 * @name: The lock name
273 * Returns: NULL, or the struct gfs2_glock with the requested number
276 static struct gfs2_glock *gfs2_glock_find(const struct gfs2_sbd *sdp,
277 const struct lm_lockname *name)
279 unsigned int hash = gl_hash(sdp, name);
280 struct gfs2_glock *gl;
282 read_lock(gl_lock_addr(hash));
283 gl = search_bucket(hash, sdp, name);
284 read_unlock(gl_lock_addr(hash));
290 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
291 * @sdp: The GFS2 superblock
292 * @number: the lock number
293 * @glops: The glock_operations to use
294 * @create: If 0, don't create the glock if it doesn't exist
295 * @glp: the glock is returned here
297 * This does not lock a glock, just finds/creates structures for one.
302 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
303 const struct gfs2_glock_operations *glops, int create,
304 struct gfs2_glock **glp)
306 struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
307 struct gfs2_glock *gl, *tmp;
308 unsigned int hash = gl_hash(sdp, &name);
311 read_lock(gl_lock_addr(hash));
312 gl = search_bucket(hash, sdp, &name);
313 read_unlock(gl_lock_addr(hash));
320 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
326 atomic_set(&gl->gl_ref, 1);
327 gl->gl_state = LM_ST_UNLOCKED;
332 gl->gl_req_gh = NULL;
333 gl->gl_req_bh = NULL;
335 gl->gl_stamp = jiffies;
336 gl->gl_object = NULL;
338 gl->gl_aspace = NULL;
339 lops_init_le(&gl->gl_le, &gfs2_glock_lops);
341 /* If this glock protects actual on-disk data or metadata blocks,
342 create a VFS inode to manage the pages/buffers holding them. */
343 if (glops == &gfs2_inode_glops || glops == &gfs2_rgrp_glops) {
344 gl->gl_aspace = gfs2_aspace_get(sdp);
345 if (!gl->gl_aspace) {
351 error = gfs2_lm_get_lock(sdp, &name, &gl->gl_lock);
355 write_lock(gl_lock_addr(hash));
356 tmp = search_bucket(hash, sdp, &name);
358 write_unlock(gl_lock_addr(hash));
362 hlist_add_head(&gl->gl_list, &gl_hash_table[hash].hb_list);
363 write_unlock(gl_lock_addr(hash));
372 gfs2_aspace_put(gl->gl_aspace);
374 kmem_cache_free(gfs2_glock_cachep, gl);
379 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
381 * @state: the state we're requesting
382 * @flags: the modifier flags
383 * @gh: the holder structure
387 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
388 struct gfs2_holder *gh)
390 INIT_LIST_HEAD(&gh->gh_list);
392 gh->gh_ip = (unsigned long)__builtin_return_address(0);
393 gh->gh_owner = current;
394 gh->gh_state = state;
395 gh->gh_flags = flags;
398 init_completion(&gh->gh_wait);
400 if (gh->gh_state == LM_ST_EXCLUSIVE)
401 gh->gh_flags |= GL_LOCAL_EXCL;
407 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
408 * @state: the state we're requesting
409 * @flags: the modifier flags
410 * @gh: the holder structure
412 * Don't mess with the glock.
416 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
418 gh->gh_state = state;
419 gh->gh_flags = flags;
420 if (gh->gh_state == LM_ST_EXCLUSIVE)
421 gh->gh_flags |= GL_LOCAL_EXCL;
423 gh->gh_iflags &= 1 << HIF_ALLOCED;
424 gh->gh_ip = (unsigned long)__builtin_return_address(0);
428 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
429 * @gh: the holder structure
433 void gfs2_holder_uninit(struct gfs2_holder *gh)
435 gfs2_glock_put(gh->gh_gl);
441 * gfs2_holder_get - get a struct gfs2_holder structure
443 * @state: the state we're requesting
444 * @flags: the modifier flags
447 * Figure out how big an impact this function has. Either:
448 * 1) Replace it with a cache of structures hanging off the struct gfs2_sbd
449 * 2) Leave it like it is
451 * Returns: the holder structure, NULL on ENOMEM
454 static struct gfs2_holder *gfs2_holder_get(struct gfs2_glock *gl,
456 int flags, gfp_t gfp_flags)
458 struct gfs2_holder *gh;
460 gh = kmalloc(sizeof(struct gfs2_holder), gfp_flags);
464 gfs2_holder_init(gl, state, flags, gh);
465 set_bit(HIF_ALLOCED, &gh->gh_iflags);
466 gh->gh_ip = (unsigned long)__builtin_return_address(0);
471 * gfs2_holder_put - get rid of a struct gfs2_holder structure
472 * @gh: the holder structure
476 static void gfs2_holder_put(struct gfs2_holder *gh)
478 gfs2_holder_uninit(gh);
483 * rq_mutex - process a mutex request in the queue
484 * @gh: the glock holder
486 * Returns: 1 if the queue is blocked
489 static int rq_mutex(struct gfs2_holder *gh)
491 struct gfs2_glock *gl = gh->gh_gl;
493 list_del_init(&gh->gh_list);
494 /* gh->gh_error never examined. */
495 set_bit(GLF_LOCK, &gl->gl_flags);
496 complete(&gh->gh_wait);
502 * rq_promote - process a promote request in the queue
503 * @gh: the glock holder
505 * Acquire a new inter-node lock, or change a lock state to more restrictive.
507 * Returns: 1 if the queue is blocked
510 static int rq_promote(struct gfs2_holder *gh)
512 struct gfs2_glock *gl = gh->gh_gl;
513 struct gfs2_sbd *sdp = gl->gl_sbd;
514 const struct gfs2_glock_operations *glops = gl->gl_ops;
516 if (!relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
517 if (list_empty(&gl->gl_holders)) {
519 set_bit(GLF_LOCK, &gl->gl_flags);
520 spin_unlock(&gl->gl_spin);
522 if (atomic_read(&sdp->sd_reclaim_count) >
523 gfs2_tune_get(sdp, gt_reclaim_limit) &&
524 !(gh->gh_flags & LM_FLAG_PRIORITY)) {
525 gfs2_reclaim_glock(sdp);
526 gfs2_reclaim_glock(sdp);
529 glops->go_xmote_th(gl, gh->gh_state, gh->gh_flags);
530 spin_lock(&gl->gl_spin);
535 if (list_empty(&gl->gl_holders)) {
536 set_bit(HIF_FIRST, &gh->gh_iflags);
537 set_bit(GLF_LOCK, &gl->gl_flags);
539 struct gfs2_holder *next_gh;
540 if (gh->gh_flags & GL_LOCAL_EXCL)
542 next_gh = list_entry(gl->gl_holders.next, struct gfs2_holder,
544 if (next_gh->gh_flags & GL_LOCAL_EXCL)
548 list_move_tail(&gh->gh_list, &gl->gl_holders);
550 set_bit(HIF_HOLDER, &gh->gh_iflags);
552 complete(&gh->gh_wait);
558 * rq_demote - process a demote request in the queue
559 * @gh: the glock holder
561 * Returns: 1 if the queue is blocked
564 static int rq_demote(struct gfs2_holder *gh)
566 struct gfs2_glock *gl = gh->gh_gl;
567 const struct gfs2_glock_operations *glops = gl->gl_ops;
569 if (!list_empty(&gl->gl_holders))
572 if (gl->gl_state == gh->gh_state || gl->gl_state == LM_ST_UNLOCKED) {
573 list_del_init(&gh->gh_list);
575 spin_unlock(&gl->gl_spin);
576 if (test_bit(HIF_DEALLOC, &gh->gh_iflags))
579 complete(&gh->gh_wait);
580 spin_lock(&gl->gl_spin);
583 set_bit(GLF_LOCK, &gl->gl_flags);
584 spin_unlock(&gl->gl_spin);
586 if (gh->gh_state == LM_ST_UNLOCKED ||
587 gl->gl_state != LM_ST_EXCLUSIVE)
588 glops->go_drop_th(gl);
590 glops->go_xmote_th(gl, gh->gh_state, gh->gh_flags);
592 spin_lock(&gl->gl_spin);
599 * rq_greedy - process a queued request to drop greedy status
600 * @gh: the glock holder
602 * Returns: 1 if the queue is blocked
605 static int rq_greedy(struct gfs2_holder *gh)
607 struct gfs2_glock *gl = gh->gh_gl;
609 list_del_init(&gh->gh_list);
610 /* gh->gh_error never examined. */
611 clear_bit(GLF_GREEDY, &gl->gl_flags);
612 spin_unlock(&gl->gl_spin);
614 gfs2_holder_uninit(gh);
615 kfree(container_of(gh, struct greedy, gr_gh));
617 spin_lock(&gl->gl_spin);
623 * run_queue - process holder structures on a glock
627 static void run_queue(struct gfs2_glock *gl)
629 struct gfs2_holder *gh;
633 if (test_bit(GLF_LOCK, &gl->gl_flags))
636 if (!list_empty(&gl->gl_waiters1)) {
637 gh = list_entry(gl->gl_waiters1.next,
638 struct gfs2_holder, gh_list);
640 if (test_bit(HIF_MUTEX, &gh->gh_iflags))
641 blocked = rq_mutex(gh);
643 gfs2_assert_warn(gl->gl_sbd, 0);
645 } else if (!list_empty(&gl->gl_waiters2) &&
646 !test_bit(GLF_SKIP_WAITERS2, &gl->gl_flags)) {
647 gh = list_entry(gl->gl_waiters2.next,
648 struct gfs2_holder, gh_list);
650 if (test_bit(HIF_DEMOTE, &gh->gh_iflags))
651 blocked = rq_demote(gh);
652 else if (test_bit(HIF_GREEDY, &gh->gh_iflags))
653 blocked = rq_greedy(gh);
655 gfs2_assert_warn(gl->gl_sbd, 0);
657 } else if (!list_empty(&gl->gl_waiters3)) {
658 gh = list_entry(gl->gl_waiters3.next,
659 struct gfs2_holder, gh_list);
661 if (test_bit(HIF_PROMOTE, &gh->gh_iflags))
662 blocked = rq_promote(gh);
664 gfs2_assert_warn(gl->gl_sbd, 0);
675 * gfs2_glmutex_lock - acquire a local lock on a glock
678 * Gives caller exclusive access to manipulate a glock structure.
681 static void gfs2_glmutex_lock(struct gfs2_glock *gl)
683 struct gfs2_holder gh;
685 gfs2_holder_init(gl, 0, 0, &gh);
686 set_bit(HIF_MUTEX, &gh.gh_iflags);
688 spin_lock(&gl->gl_spin);
689 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
690 list_add_tail(&gh.gh_list, &gl->gl_waiters1);
692 gl->gl_owner = current;
693 gl->gl_ip = (unsigned long)__builtin_return_address(0);
694 complete(&gh.gh_wait);
696 spin_unlock(&gl->gl_spin);
698 wait_for_completion(&gh.gh_wait);
699 gfs2_holder_uninit(&gh);
703 * gfs2_glmutex_trylock - try to acquire a local lock on a glock
706 * Returns: 1 if the glock is acquired
709 static int gfs2_glmutex_trylock(struct gfs2_glock *gl)
713 spin_lock(&gl->gl_spin);
714 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
717 gl->gl_owner = current;
718 gl->gl_ip = (unsigned long)__builtin_return_address(0);
720 spin_unlock(&gl->gl_spin);
726 * gfs2_glmutex_unlock - release a local lock on a glock
731 static void gfs2_glmutex_unlock(struct gfs2_glock *gl)
733 spin_lock(&gl->gl_spin);
734 clear_bit(GLF_LOCK, &gl->gl_flags);
738 BUG_ON(!spin_is_locked(&gl->gl_spin));
739 spin_unlock(&gl->gl_spin);
743 * handle_callback - add a demote request to a lock's queue
745 * @state: the state the caller wants us to change to
747 * Note: This may fail sliently if we are out of memory.
750 static void handle_callback(struct gfs2_glock *gl, unsigned int state)
752 struct gfs2_holder *gh, *new_gh = NULL;
755 spin_lock(&gl->gl_spin);
757 list_for_each_entry(gh, &gl->gl_waiters2, gh_list) {
758 if (test_bit(HIF_DEMOTE, &gh->gh_iflags) &&
759 gl->gl_req_gh != gh) {
760 if (gh->gh_state != state)
761 gh->gh_state = LM_ST_UNLOCKED;
767 list_add_tail(&new_gh->gh_list, &gl->gl_waiters2);
770 spin_unlock(&gl->gl_spin);
772 new_gh = gfs2_holder_get(gl, state, LM_FLAG_TRY, GFP_KERNEL);
775 set_bit(HIF_DEMOTE, &new_gh->gh_iflags);
776 set_bit(HIF_DEALLOC, &new_gh->gh_iflags);
782 spin_unlock(&gl->gl_spin);
785 gfs2_holder_put(new_gh);
788 void gfs2_glock_inode_squish(struct inode *inode)
790 struct gfs2_holder gh;
791 struct gfs2_glock *gl = GFS2_I(inode)->i_gl;
792 gfs2_holder_init(gl, LM_ST_UNLOCKED, 0, &gh);
793 set_bit(HIF_DEMOTE, &gh.gh_iflags);
794 spin_lock(&gl->gl_spin);
795 gfs2_assert(inode->i_sb->s_fs_info, list_empty(&gl->gl_holders));
796 list_add_tail(&gh.gh_list, &gl->gl_waiters2);
798 spin_unlock(&gl->gl_spin);
799 wait_for_completion(&gh.gh_wait);
800 gfs2_holder_uninit(&gh);
804 * state_change - record that the glock is now in a different state
806 * @new_state the new state
810 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
814 held1 = (gl->gl_state != LM_ST_UNLOCKED);
815 held2 = (new_state != LM_ST_UNLOCKED);
817 if (held1 != held2) {
824 gl->gl_state = new_state;
828 * xmote_bh - Called after the lock module is done acquiring a lock
829 * @gl: The glock in question
830 * @ret: the int returned from the lock module
834 static void xmote_bh(struct gfs2_glock *gl, unsigned int ret)
836 struct gfs2_sbd *sdp = gl->gl_sbd;
837 const struct gfs2_glock_operations *glops = gl->gl_ops;
838 struct gfs2_holder *gh = gl->gl_req_gh;
839 int prev_state = gl->gl_state;
842 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
843 gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
844 gfs2_assert_warn(sdp, !(ret & LM_OUT_ASYNC));
846 state_change(gl, ret & LM_OUT_ST_MASK);
848 if (prev_state != LM_ST_UNLOCKED && !(ret & LM_OUT_CACHEABLE)) {
850 glops->go_inval(gl, DIO_METADATA | DIO_DATA);
851 } else if (gl->gl_state == LM_ST_DEFERRED) {
852 /* We might not want to do this here.
853 Look at moving to the inode glops. */
855 glops->go_inval(gl, DIO_DATA);
858 /* Deal with each possible exit condition */
861 gl->gl_stamp = jiffies;
862 else if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
863 spin_lock(&gl->gl_spin);
864 list_del_init(&gh->gh_list);
866 spin_unlock(&gl->gl_spin);
867 } else if (test_bit(HIF_DEMOTE, &gh->gh_iflags)) {
868 spin_lock(&gl->gl_spin);
869 list_del_init(&gh->gh_list);
870 if (gl->gl_state == gh->gh_state ||
871 gl->gl_state == LM_ST_UNLOCKED) {
874 if (gfs2_assert_warn(sdp, gh->gh_flags &
875 (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) == -1)
876 fs_warn(sdp, "ret = 0x%.8X\n", ret);
877 gh->gh_error = GLR_TRYFAILED;
879 spin_unlock(&gl->gl_spin);
881 if (ret & LM_OUT_CANCELED)
882 handle_callback(gl, LM_ST_UNLOCKED);
884 } else if (ret & LM_OUT_CANCELED) {
885 spin_lock(&gl->gl_spin);
886 list_del_init(&gh->gh_list);
887 gh->gh_error = GLR_CANCELED;
888 spin_unlock(&gl->gl_spin);
890 } else if (relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
891 spin_lock(&gl->gl_spin);
892 list_move_tail(&gh->gh_list, &gl->gl_holders);
894 set_bit(HIF_HOLDER, &gh->gh_iflags);
895 spin_unlock(&gl->gl_spin);
897 set_bit(HIF_FIRST, &gh->gh_iflags);
901 } else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
902 spin_lock(&gl->gl_spin);
903 list_del_init(&gh->gh_list);
904 gh->gh_error = GLR_TRYFAILED;
905 spin_unlock(&gl->gl_spin);
908 if (gfs2_assert_withdraw(sdp, 0) == -1)
909 fs_err(sdp, "ret = 0x%.8X\n", ret);
912 if (glops->go_xmote_bh)
913 glops->go_xmote_bh(gl);
916 spin_lock(&gl->gl_spin);
917 gl->gl_req_gh = NULL;
918 gl->gl_req_bh = NULL;
919 clear_bit(GLF_LOCK, &gl->gl_flags);
921 spin_unlock(&gl->gl_spin);
927 if (test_bit(HIF_DEALLOC, &gh->gh_iflags))
930 complete(&gh->gh_wait);
935 * gfs2_glock_xmote_th - Call into the lock module to acquire or change a glock
936 * @gl: The glock in question
937 * @state: the requested state
938 * @flags: modifier flags to the lock call
942 void gfs2_glock_xmote_th(struct gfs2_glock *gl, unsigned int state, int flags)
944 struct gfs2_sbd *sdp = gl->gl_sbd;
945 const struct gfs2_glock_operations *glops = gl->gl_ops;
946 int lck_flags = flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB |
947 LM_FLAG_NOEXP | LM_FLAG_ANY |
949 unsigned int lck_ret;
951 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
952 gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
953 gfs2_assert_warn(sdp, state != LM_ST_UNLOCKED);
954 gfs2_assert_warn(sdp, state != gl->gl_state);
956 if (gl->gl_state == LM_ST_EXCLUSIVE && glops->go_sync)
957 glops->go_sync(gl, DIO_METADATA | DIO_DATA | DIO_RELEASE);
960 gl->gl_req_bh = xmote_bh;
962 lck_ret = gfs2_lm_lock(sdp, gl->gl_lock, gl->gl_state, state, lck_flags);
964 if (gfs2_assert_withdraw(sdp, !(lck_ret & LM_OUT_ERROR)))
967 if (lck_ret & LM_OUT_ASYNC)
968 gfs2_assert_warn(sdp, lck_ret == LM_OUT_ASYNC);
970 xmote_bh(gl, lck_ret);
974 * drop_bh - Called after a lock module unlock completes
976 * @ret: the return status
978 * Doesn't wake up the process waiting on the struct gfs2_holder (if any)
979 * Doesn't drop the reference on the glock the top half took out
983 static void drop_bh(struct gfs2_glock *gl, unsigned int ret)
985 struct gfs2_sbd *sdp = gl->gl_sbd;
986 const struct gfs2_glock_operations *glops = gl->gl_ops;
987 struct gfs2_holder *gh = gl->gl_req_gh;
989 clear_bit(GLF_PREFETCH, &gl->gl_flags);
991 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
992 gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
993 gfs2_assert_warn(sdp, !ret);
995 state_change(gl, LM_ST_UNLOCKED);
998 glops->go_inval(gl, DIO_METADATA | DIO_DATA);
1001 spin_lock(&gl->gl_spin);
1002 list_del_init(&gh->gh_list);
1004 spin_unlock(&gl->gl_spin);
1007 if (glops->go_drop_bh)
1008 glops->go_drop_bh(gl);
1010 spin_lock(&gl->gl_spin);
1011 gl->gl_req_gh = NULL;
1012 gl->gl_req_bh = NULL;
1013 clear_bit(GLF_LOCK, &gl->gl_flags);
1015 spin_unlock(&gl->gl_spin);
1020 if (test_bit(HIF_DEALLOC, &gh->gh_iflags))
1021 gfs2_holder_put(gh);
1023 complete(&gh->gh_wait);
1028 * gfs2_glock_drop_th - call into the lock module to unlock a lock
1033 void gfs2_glock_drop_th(struct gfs2_glock *gl)
1035 struct gfs2_sbd *sdp = gl->gl_sbd;
1036 const struct gfs2_glock_operations *glops = gl->gl_ops;
1039 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1040 gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
1041 gfs2_assert_warn(sdp, gl->gl_state != LM_ST_UNLOCKED);
1043 if (gl->gl_state == LM_ST_EXCLUSIVE && glops->go_sync)
1044 glops->go_sync(gl, DIO_METADATA | DIO_DATA | DIO_RELEASE);
1046 gfs2_glock_hold(gl);
1047 gl->gl_req_bh = drop_bh;
1049 ret = gfs2_lm_unlock(sdp, gl->gl_lock, gl->gl_state);
1051 if (gfs2_assert_withdraw(sdp, !(ret & LM_OUT_ERROR)))
1057 gfs2_assert_warn(sdp, ret == LM_OUT_ASYNC);
1061 * do_cancels - cancel requests for locks stuck waiting on an expire flag
1062 * @gh: the LM_FLAG_PRIORITY holder waiting to acquire the lock
1064 * Don't cancel GL_NOCANCEL requests.
1067 static void do_cancels(struct gfs2_holder *gh)
1069 struct gfs2_glock *gl = gh->gh_gl;
1071 spin_lock(&gl->gl_spin);
1073 while (gl->gl_req_gh != gh &&
1074 !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1075 !list_empty(&gh->gh_list)) {
1076 if (gl->gl_req_bh && !(gl->gl_req_gh &&
1077 (gl->gl_req_gh->gh_flags & GL_NOCANCEL))) {
1078 spin_unlock(&gl->gl_spin);
1079 gfs2_lm_cancel(gl->gl_sbd, gl->gl_lock);
1081 spin_lock(&gl->gl_spin);
1083 spin_unlock(&gl->gl_spin);
1085 spin_lock(&gl->gl_spin);
1089 spin_unlock(&gl->gl_spin);
1093 * glock_wait_internal - wait on a glock acquisition
1094 * @gh: the glock holder
1096 * Returns: 0 on success
1099 static int glock_wait_internal(struct gfs2_holder *gh)
1101 struct gfs2_glock *gl = gh->gh_gl;
1102 struct gfs2_sbd *sdp = gl->gl_sbd;
1103 const struct gfs2_glock_operations *glops = gl->gl_ops;
1105 if (test_bit(HIF_ABORTED, &gh->gh_iflags))
1108 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1109 spin_lock(&gl->gl_spin);
1110 if (gl->gl_req_gh != gh &&
1111 !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1112 !list_empty(&gh->gh_list)) {
1113 list_del_init(&gh->gh_list);
1114 gh->gh_error = GLR_TRYFAILED;
1116 spin_unlock(&gl->gl_spin);
1117 return gh->gh_error;
1119 spin_unlock(&gl->gl_spin);
1122 if (gh->gh_flags & LM_FLAG_PRIORITY)
1125 wait_for_completion(&gh->gh_wait);
1128 return gh->gh_error;
1130 gfs2_assert_withdraw(sdp, test_bit(HIF_HOLDER, &gh->gh_iflags));
1131 gfs2_assert_withdraw(sdp, relaxed_state_ok(gl->gl_state, gh->gh_state,
1134 if (test_bit(HIF_FIRST, &gh->gh_iflags)) {
1135 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1137 if (glops->go_lock) {
1138 gh->gh_error = glops->go_lock(gh);
1140 spin_lock(&gl->gl_spin);
1141 list_del_init(&gh->gh_list);
1142 spin_unlock(&gl->gl_spin);
1146 spin_lock(&gl->gl_spin);
1147 gl->gl_req_gh = NULL;
1148 gl->gl_req_bh = NULL;
1149 clear_bit(GLF_LOCK, &gl->gl_flags);
1151 spin_unlock(&gl->gl_spin);
1154 return gh->gh_error;
1157 static inline struct gfs2_holder *
1158 find_holder_by_owner(struct list_head *head, struct task_struct *owner)
1160 struct gfs2_holder *gh;
1162 list_for_each_entry(gh, head, gh_list) {
1163 if (gh->gh_owner == owner)
1171 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1172 * @gh: the holder structure to add
1176 static void add_to_queue(struct gfs2_holder *gh)
1178 struct gfs2_glock *gl = gh->gh_gl;
1179 struct gfs2_holder *existing;
1181 BUG_ON(!gh->gh_owner);
1183 existing = find_holder_by_owner(&gl->gl_holders, gh->gh_owner);
1185 print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
1186 printk(KERN_INFO "pid : %d\n", existing->gh_owner->pid);
1187 printk(KERN_INFO "lock type : %d lock state : %d\n",
1188 existing->gh_gl->gl_name.ln_type, existing->gh_gl->gl_state);
1189 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1190 printk(KERN_INFO "pid : %d\n", gh->gh_owner->pid);
1191 printk(KERN_INFO "lock type : %d lock state : %d\n",
1192 gl->gl_name.ln_type, gl->gl_state);
1196 existing = find_holder_by_owner(&gl->gl_waiters3, gh->gh_owner);
1198 print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
1199 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1203 if (gh->gh_flags & LM_FLAG_PRIORITY)
1204 list_add(&gh->gh_list, &gl->gl_waiters3);
1206 list_add_tail(&gh->gh_list, &gl->gl_waiters3);
1210 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1211 * @gh: the holder structure
1213 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1215 * Returns: 0, GLR_TRYFAILED, or errno on failure
1218 int gfs2_glock_nq(struct gfs2_holder *gh)
1220 struct gfs2_glock *gl = gh->gh_gl;
1221 struct gfs2_sbd *sdp = gl->gl_sbd;
1225 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1226 set_bit(HIF_ABORTED, &gh->gh_iflags);
1230 set_bit(HIF_PROMOTE, &gh->gh_iflags);
1232 spin_lock(&gl->gl_spin);
1235 spin_unlock(&gl->gl_spin);
1237 if (!(gh->gh_flags & GL_ASYNC)) {
1238 error = glock_wait_internal(gh);
1239 if (error == GLR_CANCELED) {
1245 clear_bit(GLF_PREFETCH, &gl->gl_flags);
1247 if (error == GLR_TRYFAILED && (gh->gh_flags & GL_DUMP))
1254 * gfs2_glock_poll - poll to see if an async request has been completed
1257 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1260 int gfs2_glock_poll(struct gfs2_holder *gh)
1262 struct gfs2_glock *gl = gh->gh_gl;
1265 spin_lock(&gl->gl_spin);
1267 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1269 else if (list_empty(&gh->gh_list)) {
1270 if (gh->gh_error == GLR_CANCELED) {
1271 spin_unlock(&gl->gl_spin);
1273 if (gfs2_glock_nq(gh))
1280 spin_unlock(&gl->gl_spin);
1286 * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
1287 * @gh: the holder structure
1289 * Returns: 0, GLR_TRYFAILED, or errno on failure
1292 int gfs2_glock_wait(struct gfs2_holder *gh)
1296 error = glock_wait_internal(gh);
1297 if (error == GLR_CANCELED) {
1299 gh->gh_flags &= ~GL_ASYNC;
1300 error = gfs2_glock_nq(gh);
1307 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1308 * @gh: the glock holder
1312 void gfs2_glock_dq(struct gfs2_holder *gh)
1314 struct gfs2_glock *gl = gh->gh_gl;
1315 const struct gfs2_glock_operations *glops = gl->gl_ops;
1317 if (gh->gh_flags & GL_NOCACHE)
1318 handle_callback(gl, LM_ST_UNLOCKED);
1320 gfs2_glmutex_lock(gl);
1322 spin_lock(&gl->gl_spin);
1323 list_del_init(&gh->gh_list);
1325 if (list_empty(&gl->gl_holders)) {
1326 spin_unlock(&gl->gl_spin);
1328 if (glops->go_unlock)
1329 glops->go_unlock(gh);
1331 gl->gl_stamp = jiffies;
1333 spin_lock(&gl->gl_spin);
1336 clear_bit(GLF_LOCK, &gl->gl_flags);
1338 spin_unlock(&gl->gl_spin);
1342 * gfs2_glock_prefetch - Try to prefetch a glock
1344 * @state: the state to prefetch in
1345 * @flags: flags passed to go_xmote_th()
1349 static void gfs2_glock_prefetch(struct gfs2_glock *gl, unsigned int state,
1352 const struct gfs2_glock_operations *glops = gl->gl_ops;
1354 spin_lock(&gl->gl_spin);
1356 if (test_bit(GLF_LOCK, &gl->gl_flags) || !list_empty(&gl->gl_holders) ||
1357 !list_empty(&gl->gl_waiters1) || !list_empty(&gl->gl_waiters2) ||
1358 !list_empty(&gl->gl_waiters3) ||
1359 relaxed_state_ok(gl->gl_state, state, flags)) {
1360 spin_unlock(&gl->gl_spin);
1364 set_bit(GLF_PREFETCH, &gl->gl_flags);
1365 set_bit(GLF_LOCK, &gl->gl_flags);
1366 spin_unlock(&gl->gl_spin);
1368 glops->go_xmote_th(gl, state, flags);
1371 static void greedy_work(void *data)
1373 struct greedy *gr = data;
1374 struct gfs2_holder *gh = &gr->gr_gh;
1375 struct gfs2_glock *gl = gh->gh_gl;
1376 const struct gfs2_glock_operations *glops = gl->gl_ops;
1378 clear_bit(GLF_SKIP_WAITERS2, &gl->gl_flags);
1380 if (glops->go_greedy)
1381 glops->go_greedy(gl);
1383 spin_lock(&gl->gl_spin);
1385 if (list_empty(&gl->gl_waiters2)) {
1386 clear_bit(GLF_GREEDY, &gl->gl_flags);
1387 spin_unlock(&gl->gl_spin);
1388 gfs2_holder_uninit(gh);
1391 gfs2_glock_hold(gl);
1392 list_add_tail(&gh->gh_list, &gl->gl_waiters2);
1394 spin_unlock(&gl->gl_spin);
1400 * gfs2_glock_be_greedy -
1404 * Returns: 0 if go_greedy will be called, 1 otherwise
1407 int gfs2_glock_be_greedy(struct gfs2_glock *gl, unsigned int time)
1410 struct gfs2_holder *gh;
1412 if (!time || gl->gl_sbd->sd_args.ar_localcaching ||
1413 test_and_set_bit(GLF_GREEDY, &gl->gl_flags))
1416 gr = kmalloc(sizeof(struct greedy), GFP_KERNEL);
1418 clear_bit(GLF_GREEDY, &gl->gl_flags);
1423 gfs2_holder_init(gl, 0, 0, gh);
1424 set_bit(HIF_GREEDY, &gh->gh_iflags);
1425 INIT_WORK(&gr->gr_work, greedy_work, gr);
1427 set_bit(GLF_SKIP_WAITERS2, &gl->gl_flags);
1428 schedule_delayed_work(&gr->gr_work, time);
1434 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1435 * @gh: the holder structure
1439 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1442 gfs2_holder_uninit(gh);
1446 * gfs2_glock_nq_num - acquire a glock based on lock number
1447 * @sdp: the filesystem
1448 * @number: the lock number
1449 * @glops: the glock operations for the type of glock
1450 * @state: the state to acquire the glock in
1451 * @flags: modifier flags for the aquisition
1452 * @gh: the struct gfs2_holder
1457 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1458 const struct gfs2_glock_operations *glops,
1459 unsigned int state, int flags, struct gfs2_holder *gh)
1461 struct gfs2_glock *gl;
1464 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1466 error = gfs2_glock_nq_init(gl, state, flags, gh);
1474 * glock_compare - Compare two struct gfs2_glock structures for sorting
1475 * @arg_a: the first structure
1476 * @arg_b: the second structure
1480 static int glock_compare(const void *arg_a, const void *arg_b)
1482 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1483 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1484 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1485 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1487 if (a->ln_number > b->ln_number)
1489 if (a->ln_number < b->ln_number)
1491 if (gh_a->gh_state == LM_ST_SHARED && gh_b->gh_state == LM_ST_EXCLUSIVE)
1493 if (!(gh_a->gh_flags & GL_LOCAL_EXCL) && (gh_b->gh_flags & GL_LOCAL_EXCL))
1499 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1500 * @num_gh: the number of structures
1501 * @ghs: an array of struct gfs2_holder structures
1503 * Returns: 0 on success (all glocks acquired),
1504 * errno on failure (no glocks acquired)
1507 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1508 struct gfs2_holder **p)
1513 for (x = 0; x < num_gh; x++)
1516 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1518 for (x = 0; x < num_gh; x++) {
1519 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1521 error = gfs2_glock_nq(p[x]);
1524 gfs2_glock_dq(p[x]);
1533 * gfs2_glock_nq_m - acquire multiple glocks
1534 * @num_gh: the number of structures
1535 * @ghs: an array of struct gfs2_holder structures
1537 * Figure out how big an impact this function has. Either:
1538 * 1) Replace this code with code that calls gfs2_glock_prefetch()
1539 * 2) Forget async stuff and just call nq_m_sync()
1540 * 3) Leave it like it is
1542 * Returns: 0 on success (all glocks acquired),
1543 * errno on failure (no glocks acquired)
1546 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1550 int borked = 0, serious = 0;
1557 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1558 return gfs2_glock_nq(ghs);
1561 e = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL);
1565 for (x = 0; x < num_gh; x++) {
1566 ghs[x].gh_flags |= LM_FLAG_TRY | GL_ASYNC;
1567 error = gfs2_glock_nq(&ghs[x]);
1576 for (x = 0; x < num_gh; x++) {
1577 error = e[x] = glock_wait_internal(&ghs[x]);
1580 if (error != GLR_TRYFAILED && error != GLR_CANCELED)
1590 for (x = 0; x < num_gh; x++)
1592 gfs2_glock_dq(&ghs[x]);
1597 for (x = 0; x < num_gh; x++)
1598 gfs2_holder_reinit(ghs[x].gh_state, ghs[x].gh_flags,
1600 error = nq_m_sync(num_gh, ghs, (struct gfs2_holder **)e);
1609 * gfs2_glock_dq_m - release multiple glocks
1610 * @num_gh: the number of structures
1611 * @ghs: an array of struct gfs2_holder structures
1615 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1619 for (x = 0; x < num_gh; x++)
1620 gfs2_glock_dq(&ghs[x]);
1624 * gfs2_glock_dq_uninit_m - release multiple glocks
1625 * @num_gh: the number of structures
1626 * @ghs: an array of struct gfs2_holder structures
1630 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1634 for (x = 0; x < num_gh; x++)
1635 gfs2_glock_dq_uninit(&ghs[x]);
1639 * gfs2_glock_prefetch_num - prefetch a glock based on lock number
1640 * @sdp: the filesystem
1641 * @number: the lock number
1642 * @glops: the glock operations for the type of glock
1643 * @state: the state to acquire the glock in
1644 * @flags: modifier flags for the aquisition
1649 void gfs2_glock_prefetch_num(struct gfs2_sbd *sdp, u64 number,
1650 const struct gfs2_glock_operations *glops,
1651 unsigned int state, int flags)
1653 struct gfs2_glock *gl;
1656 if (atomic_read(&sdp->sd_reclaim_count) <
1657 gfs2_tune_get(sdp, gt_reclaim_limit)) {
1658 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1660 gfs2_glock_prefetch(gl, state, flags);
1667 * gfs2_lvb_hold - attach a LVB from a glock
1668 * @gl: The glock in question
1672 int gfs2_lvb_hold(struct gfs2_glock *gl)
1676 gfs2_glmutex_lock(gl);
1678 if (!atomic_read(&gl->gl_lvb_count)) {
1679 error = gfs2_lm_hold_lvb(gl->gl_sbd, gl->gl_lock, &gl->gl_lvb);
1681 gfs2_glmutex_unlock(gl);
1684 gfs2_glock_hold(gl);
1686 atomic_inc(&gl->gl_lvb_count);
1688 gfs2_glmutex_unlock(gl);
1694 * gfs2_lvb_unhold - detach a LVB from a glock
1695 * @gl: The glock in question
1699 void gfs2_lvb_unhold(struct gfs2_glock *gl)
1701 gfs2_glock_hold(gl);
1702 gfs2_glmutex_lock(gl);
1704 gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count) > 0);
1705 if (atomic_dec_and_test(&gl->gl_lvb_count)) {
1706 gfs2_lm_unhold_lvb(gl->gl_sbd, gl->gl_lock, gl->gl_lvb);
1711 gfs2_glmutex_unlock(gl);
1715 static void blocking_cb(struct gfs2_sbd *sdp, struct lm_lockname *name,
1718 struct gfs2_glock *gl;
1720 gl = gfs2_glock_find(sdp, name);
1724 if (gl->gl_ops->go_callback)
1725 gl->gl_ops->go_callback(gl, state);
1726 handle_callback(gl, state);
1728 spin_lock(&gl->gl_spin);
1730 spin_unlock(&gl->gl_spin);
1736 * gfs2_glock_cb - Callback used by locking module
1737 * @sdp: Pointer to the superblock
1738 * @type: Type of callback
1739 * @data: Type dependent data pointer
1741 * Called by the locking module when it wants to tell us something.
1742 * Either we need to drop a lock, one of our ASYNC requests completed, or
1743 * a journal from another client needs to be recovered.
1746 void gfs2_glock_cb(void *cb_data, unsigned int type, void *data)
1748 struct gfs2_sbd *sdp = cb_data;
1752 blocking_cb(sdp, data, LM_ST_UNLOCKED);
1756 blocking_cb(sdp, data, LM_ST_DEFERRED);
1760 blocking_cb(sdp, data, LM_ST_SHARED);
1764 struct lm_async_cb *async = data;
1765 struct gfs2_glock *gl;
1767 gl = gfs2_glock_find(sdp, &async->lc_name);
1768 if (gfs2_assert_warn(sdp, gl))
1770 if (!gfs2_assert_warn(sdp, gl->gl_req_bh))
1771 gl->gl_req_bh(gl, async->lc_ret);
1776 case LM_CB_NEED_RECOVERY:
1777 gfs2_jdesc_make_dirty(sdp, *(unsigned int *)data);
1778 if (sdp->sd_recoverd_process)
1779 wake_up_process(sdp->sd_recoverd_process);
1782 case LM_CB_DROPLOCKS:
1783 gfs2_gl_hash_clear(sdp, NO_WAIT);
1784 gfs2_quota_scan(sdp);
1788 gfs2_assert_warn(sdp, 0);
1794 * demote_ok - Check to see if it's ok to unlock a glock
1797 * Returns: 1 if it's ok
1800 static int demote_ok(struct gfs2_glock *gl)
1802 struct gfs2_sbd *sdp = gl->gl_sbd;
1803 const struct gfs2_glock_operations *glops = gl->gl_ops;
1806 if (test_bit(GLF_STICKY, &gl->gl_flags))
1808 else if (test_bit(GLF_PREFETCH, &gl->gl_flags))
1809 demote = time_after_eq(jiffies, gl->gl_stamp +
1810 gfs2_tune_get(sdp, gt_prefetch_secs) * HZ);
1811 else if (glops->go_demote_ok)
1812 demote = glops->go_demote_ok(gl);
1818 * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1823 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
1825 struct gfs2_sbd *sdp = gl->gl_sbd;
1827 spin_lock(&sdp->sd_reclaim_lock);
1828 if (list_empty(&gl->gl_reclaim)) {
1829 gfs2_glock_hold(gl);
1830 list_add(&gl->gl_reclaim, &sdp->sd_reclaim_list);
1831 atomic_inc(&sdp->sd_reclaim_count);
1833 spin_unlock(&sdp->sd_reclaim_lock);
1835 wake_up(&sdp->sd_reclaim_wq);
1839 * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1840 * @sdp: the filesystem
1842 * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1843 * different glock and we notice that there are a lot of glocks in the
1848 void gfs2_reclaim_glock(struct gfs2_sbd *sdp)
1850 struct gfs2_glock *gl;
1852 spin_lock(&sdp->sd_reclaim_lock);
1853 if (list_empty(&sdp->sd_reclaim_list)) {
1854 spin_unlock(&sdp->sd_reclaim_lock);
1857 gl = list_entry(sdp->sd_reclaim_list.next,
1858 struct gfs2_glock, gl_reclaim);
1859 list_del_init(&gl->gl_reclaim);
1860 spin_unlock(&sdp->sd_reclaim_lock);
1862 atomic_dec(&sdp->sd_reclaim_count);
1863 atomic_inc(&sdp->sd_reclaimed);
1865 if (gfs2_glmutex_trylock(gl)) {
1866 if (queue_empty(gl, &gl->gl_holders) &&
1867 gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1868 handle_callback(gl, LM_ST_UNLOCKED);
1869 gfs2_glmutex_unlock(gl);
1876 * examine_bucket - Call a function for glock in a hash bucket
1877 * @examiner: the function
1878 * @sdp: the filesystem
1879 * @bucket: the bucket
1881 * Returns: 1 if the bucket has entries
1884 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1887 struct gfs2_glock *gl, *prev = NULL;
1888 int has_entries = 0;
1889 struct hlist_head *head = &gl_hash_table[hash].hb_list;
1891 read_lock(gl_lock_addr(hash));
1892 /* Can't use hlist_for_each_entry - don't want prefetch here */
1893 if (hlist_empty(head))
1895 gl = list_entry(head->first, struct gfs2_glock, gl_list);
1897 if (gl->gl_sbd == sdp) {
1898 gfs2_glock_hold(gl);
1899 read_unlock(gl_lock_addr(hash));
1901 gfs2_glock_put(prev);
1905 read_lock(gl_lock_addr(hash));
1907 if (gl->gl_list.next == NULL)
1909 gl = list_entry(gl->gl_list.next, struct gfs2_glock, gl_list);
1912 read_unlock(gl_lock_addr(hash));
1914 gfs2_glock_put(prev);
1919 * scan_glock - look at a glock and see if we can reclaim it
1920 * @gl: the glock to look at
1924 static void scan_glock(struct gfs2_glock *gl)
1926 if (gl->gl_ops == &gfs2_inode_glops)
1929 if (gfs2_glmutex_trylock(gl)) {
1930 if (queue_empty(gl, &gl->gl_holders) &&
1931 gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1933 gfs2_glmutex_unlock(gl);
1938 gfs2_glmutex_unlock(gl);
1939 gfs2_glock_schedule_for_reclaim(gl);
1943 * gfs2_scand_internal - Look for glocks and inodes to toss from memory
1944 * @sdp: the filesystem
1948 void gfs2_scand_internal(struct gfs2_sbd *sdp)
1952 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1953 examine_bucket(scan_glock, sdp, x);
1957 * clear_glock - look at a glock and see if we can free it from glock cache
1958 * @gl: the glock to look at
1962 static void clear_glock(struct gfs2_glock *gl)
1964 struct gfs2_sbd *sdp = gl->gl_sbd;
1967 spin_lock(&sdp->sd_reclaim_lock);
1968 if (!list_empty(&gl->gl_reclaim)) {
1969 list_del_init(&gl->gl_reclaim);
1970 atomic_dec(&sdp->sd_reclaim_count);
1971 spin_unlock(&sdp->sd_reclaim_lock);
1972 released = gfs2_glock_put(gl);
1973 gfs2_assert(sdp, !released);
1975 spin_unlock(&sdp->sd_reclaim_lock);
1978 if (gfs2_glmutex_trylock(gl)) {
1979 if (queue_empty(gl, &gl->gl_holders) &&
1980 gl->gl_state != LM_ST_UNLOCKED)
1981 handle_callback(gl, LM_ST_UNLOCKED);
1982 gfs2_glmutex_unlock(gl);
1987 * gfs2_gl_hash_clear - Empty out the glock hash table
1988 * @sdp: the filesystem
1989 * @wait: wait until it's all gone
1991 * Called when unmounting the filesystem, or when inter-node lock manager
1992 * requests DROPLOCKS because it is running out of capacity.
1995 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp, int wait)
2005 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
2006 if (examine_bucket(clear_glock, sdp, x))
2013 if (time_after_eq(jiffies,
2014 t + gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
2015 fs_warn(sdp, "Unmount seems to be stalled. "
2016 "Dumping lock state...\n");
2017 gfs2_dump_lockstate(sdp);
2021 invalidate_inodes(sdp->sd_vfs);
2027 * Diagnostic routines to help debug distributed deadlock
2031 * dump_holder - print information about a glock holder
2032 * @str: a string naming the type of holder
2033 * @gh: the glock holder
2035 * Returns: 0 on success, -ENOBUFS when we run out of space
2038 static int dump_holder(char *str, struct gfs2_holder *gh)
2041 int error = -ENOBUFS;
2043 printk(KERN_INFO " %s\n", str);
2044 printk(KERN_INFO " owner = %ld\n",
2045 (gh->gh_owner) ? (long)gh->gh_owner->pid : -1);
2046 printk(KERN_INFO " gh_state = %u\n", gh->gh_state);
2047 printk(KERN_INFO " gh_flags =");
2048 for (x = 0; x < 32; x++)
2049 if (gh->gh_flags & (1 << x))
2052 printk(KERN_INFO " error = %d\n", gh->gh_error);
2053 printk(KERN_INFO " gh_iflags =");
2054 for (x = 0; x < 32; x++)
2055 if (test_bit(x, &gh->gh_iflags))
2058 print_symbol(KERN_INFO " initialized at: %s\n", gh->gh_ip);
2066 * dump_inode - print information about an inode
2069 * Returns: 0 on success, -ENOBUFS when we run out of space
2072 static int dump_inode(struct gfs2_inode *ip)
2075 int error = -ENOBUFS;
2077 printk(KERN_INFO " Inode:\n");
2078 printk(KERN_INFO " num = %llu %llu\n",
2079 (unsigned long long)ip->i_num.no_formal_ino,
2080 (unsigned long long)ip->i_num.no_addr);
2081 printk(KERN_INFO " type = %u\n", IF2DT(ip->i_inode.i_mode));
2082 printk(KERN_INFO " i_flags =");
2083 for (x = 0; x < 32; x++)
2084 if (test_bit(x, &ip->i_flags))
2094 * dump_glock - print information about a glock
2096 * @count: where we are in the buffer
2098 * Returns: 0 on success, -ENOBUFS when we run out of space
2101 static int dump_glock(struct gfs2_glock *gl)
2103 struct gfs2_holder *gh;
2105 int error = -ENOBUFS;
2107 spin_lock(&gl->gl_spin);
2109 printk(KERN_INFO "Glock 0x%p (%u, %llu)\n", gl, gl->gl_name.ln_type,
2110 (unsigned long long)gl->gl_name.ln_number);
2111 printk(KERN_INFO " gl_flags =");
2112 for (x = 0; x < 32; x++) {
2113 if (test_bit(x, &gl->gl_flags))
2117 printk(KERN_INFO " gl_ref = %d\n", atomic_read(&gl->gl_ref));
2118 printk(KERN_INFO " gl_state = %u\n", gl->gl_state);
2119 printk(KERN_INFO " gl_owner = %s\n", gl->gl_owner->comm);
2120 print_symbol(KERN_INFO " gl_ip = %s\n", gl->gl_ip);
2121 printk(KERN_INFO " req_gh = %s\n", (gl->gl_req_gh) ? "yes" : "no");
2122 printk(KERN_INFO " req_bh = %s\n", (gl->gl_req_bh) ? "yes" : "no");
2123 printk(KERN_INFO " lvb_count = %d\n", atomic_read(&gl->gl_lvb_count));
2124 printk(KERN_INFO " object = %s\n", (gl->gl_object) ? "yes" : "no");
2125 printk(KERN_INFO " le = %s\n",
2126 (list_empty(&gl->gl_le.le_list)) ? "no" : "yes");
2127 printk(KERN_INFO " reclaim = %s\n",
2128 (list_empty(&gl->gl_reclaim)) ? "no" : "yes");
2130 printk(KERN_INFO " aspace = 0x%p nrpages = %lu\n", gl->gl_aspace,
2131 gl->gl_aspace->i_mapping->nrpages);
2133 printk(KERN_INFO " aspace = no\n");
2134 printk(KERN_INFO " ail = %d\n", atomic_read(&gl->gl_ail_count));
2135 if (gl->gl_req_gh) {
2136 error = dump_holder("Request", gl->gl_req_gh);
2140 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
2141 error = dump_holder("Holder", gh);
2145 list_for_each_entry(gh, &gl->gl_waiters1, gh_list) {
2146 error = dump_holder("Waiter1", gh);
2150 list_for_each_entry(gh, &gl->gl_waiters2, gh_list) {
2151 error = dump_holder("Waiter2", gh);
2155 list_for_each_entry(gh, &gl->gl_waiters3, gh_list) {
2156 error = dump_holder("Waiter3", gh);
2160 if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object) {
2161 if (!test_bit(GLF_LOCK, &gl->gl_flags) &&
2162 list_empty(&gl->gl_holders)) {
2163 error = dump_inode(gl->gl_object);
2168 printk(KERN_INFO " Inode: busy\n");
2175 spin_unlock(&gl->gl_spin);
2180 * gfs2_dump_lockstate - print out the current lockstate
2181 * @sdp: the filesystem
2182 * @ub: the buffer to copy the information into
2184 * If @ub is NULL, dump the lockstate to the console.
2188 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
2190 struct gfs2_glock *gl;
2191 struct hlist_node *h;
2195 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
2197 read_lock(gl_lock_addr(x));
2199 hlist_for_each_entry(gl, h, &gl_hash_table[x].hb_list, gl_list) {
2200 if (gl->gl_sbd != sdp)
2203 error = dump_glock(gl);
2208 read_unlock(gl_lock_addr(x));
2218 int __init gfs2_glock_init(void)
2221 for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
2222 INIT_HLIST_HEAD(&gl_hash_table[i].hb_list);
2224 #ifdef GL_HASH_LOCK_SZ
2225 for(i = 0; i < GL_HASH_LOCK_SZ; i++) {
2226 rwlock_init(&gl_hash_locks[i]);