2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2005 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 v.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/kref.h>
19 #include <asm/semaphore.h>
20 #include <asm/uaccess.h>
32 /* Must be kept in sync with the beginning of struct gfs2_glock */
34 struct list_head gl_list;
35 unsigned long gl_flags;
39 struct gfs2_holder gr_gh;
40 struct work_struct gr_work;
43 typedef void (*glock_examiner) (struct gfs2_glock * gl);
46 * relaxed_state_ok - is a requested lock compatible with the current lock mode?
47 * @actual: the current state of the lock
48 * @requested: the lock state that was requested by the caller
49 * @flags: the modifier flags passed in by the caller
51 * Returns: 1 if the locks are compatible, 0 otherwise
54 static inline int relaxed_state_ok(unsigned int actual, unsigned requested,
57 if (actual == requested)
63 if (actual == LM_ST_EXCLUSIVE && requested == LM_ST_SHARED)
66 if (actual != LM_ST_UNLOCKED && (flags & LM_FLAG_ANY))
73 * gl_hash() - Turn glock number into hash bucket number
74 * @lock: The glock number
76 * Returns: The number of the corresponding hash bucket
79 static unsigned int gl_hash(struct lm_lockname *name)
83 h = jhash(&name->ln_number, sizeof(uint64_t), 0);
84 h = jhash(&name->ln_type, sizeof(unsigned int), h);
85 h &= GFS2_GL_HASH_MASK;
91 * glock_free() - Perform a few checks and then release struct gfs2_glock
92 * @gl: The glock to release
94 * Also calls lock module to release its internal structure for this glock.
98 static void glock_free(struct gfs2_glock *gl)
100 struct gfs2_sbd *sdp = gl->gl_sbd;
101 struct inode *aspace = gl->gl_aspace;
103 gfs2_lm_put_lock(sdp, gl->gl_lock);
106 gfs2_aspace_put(aspace);
108 kmem_cache_free(gfs2_glock_cachep, gl);
110 atomic_dec(&sdp->sd_glock_count);
114 * gfs2_glock_hold() - increment reference count on glock
115 * @gl: The glock to hold
119 void gfs2_glock_hold(struct gfs2_glock *gl)
121 kref_get(&gl->gl_ref);
124 /* All work is done after the return from kref_put() so we
125 can release the write_lock before the free. */
127 static void kill_glock(struct kref *kref)
129 struct gfs2_glock *gl = container_of(kref, struct gfs2_glock, gl_ref);
130 struct gfs2_sbd *sdp = gl->gl_sbd;
132 gfs2_assert(sdp, gl->gl_state == LM_ST_UNLOCKED);
133 gfs2_assert(sdp, list_empty(&gl->gl_reclaim));
134 gfs2_assert(sdp, list_empty(&gl->gl_holders));
135 gfs2_assert(sdp, list_empty(&gl->gl_waiters1));
136 gfs2_assert(sdp, list_empty(&gl->gl_waiters2));
137 gfs2_assert(sdp, list_empty(&gl->gl_waiters3));
141 * gfs2_glock_put() - Decrement reference count on glock
142 * @gl: The glock to put
146 int gfs2_glock_put(struct gfs2_glock *gl)
148 struct gfs2_sbd *sdp = gl->gl_sbd;
149 struct gfs2_gl_hash_bucket *bucket = gl->gl_bucket;
152 mutex_lock(&sdp->sd_invalidate_inodes_mutex);
154 write_lock(&bucket->hb_lock);
155 if (kref_put(&gl->gl_ref, kill_glock)) {
156 list_del_init(&gl->gl_list);
157 write_unlock(&bucket->hb_lock);
162 write_unlock(&bucket->hb_lock);
164 mutex_unlock(&sdp->sd_invalidate_inodes_mutex);
169 * queue_empty - check to see if a glock's queue is empty
171 * @head: the head of the queue to check
173 * This function protects the list in the event that a process already
174 * has a holder on the list and is adding a second holder for itself.
175 * The glmutex lock is what generally prevents processes from working
176 * on the same glock at once, but the special case of adding a second
177 * holder for yourself ("recursive" locking) doesn't involve locking
178 * glmutex, making the spin lock necessary.
180 * Returns: 1 if the queue is empty
183 static inline int queue_empty(struct gfs2_glock *gl, struct list_head *head)
186 spin_lock(&gl->gl_spin);
187 empty = list_empty(head);
188 spin_unlock(&gl->gl_spin);
193 * search_bucket() - Find struct gfs2_glock by lock number
194 * @bucket: the bucket to search
195 * @name: The lock name
197 * Returns: NULL, or the struct gfs2_glock with the requested number
200 static struct gfs2_glock *search_bucket(struct gfs2_gl_hash_bucket *bucket,
201 struct lm_lockname *name)
203 struct gfs2_glock *gl;
205 list_for_each_entry(gl, &bucket->hb_list, gl_list) {
206 if (test_bit(GLF_PLUG, &gl->gl_flags))
208 if (!lm_name_equal(&gl->gl_name, name))
211 kref_get(&gl->gl_ref);
220 * gfs2_glock_find() - Find glock by lock number
221 * @sdp: The GFS2 superblock
222 * @name: The lock name
224 * Returns: NULL, or the struct gfs2_glock with the requested number
227 struct gfs2_glock *gfs2_glock_find(struct gfs2_sbd *sdp,
228 struct lm_lockname *name)
230 struct gfs2_gl_hash_bucket *bucket = &sdp->sd_gl_hash[gl_hash(name)];
231 struct gfs2_glock *gl;
233 read_lock(&bucket->hb_lock);
234 gl = search_bucket(bucket, name);
235 read_unlock(&bucket->hb_lock);
241 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
242 * @sdp: The GFS2 superblock
243 * @number: the lock number
244 * @glops: The glock_operations to use
245 * @create: If 0, don't create the glock if it doesn't exist
246 * @glp: the glock is returned here
248 * This does not lock a glock, just finds/creates structures for one.
253 int gfs2_glock_get(struct gfs2_sbd *sdp, uint64_t number,
254 struct gfs2_glock_operations *glops, int create,
255 struct gfs2_glock **glp)
257 struct lm_lockname name;
258 struct gfs2_glock *gl, *tmp;
259 struct gfs2_gl_hash_bucket *bucket;
262 name.ln_number = number;
263 name.ln_type = glops->go_type;
264 bucket = &sdp->sd_gl_hash[gl_hash(&name)];
266 read_lock(&bucket->hb_lock);
267 gl = search_bucket(bucket, &name);
268 read_unlock(&bucket->hb_lock);
275 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
279 memset(gl, 0, sizeof(struct gfs2_glock));
281 INIT_LIST_HEAD(&gl->gl_list);
283 kref_init(&gl->gl_ref);
285 spin_lock_init(&gl->gl_spin);
287 gl->gl_state = LM_ST_UNLOCKED;
288 INIT_LIST_HEAD(&gl->gl_holders);
289 INIT_LIST_HEAD(&gl->gl_waiters1);
290 INIT_LIST_HEAD(&gl->gl_waiters2);
291 INIT_LIST_HEAD(&gl->gl_waiters3);
295 gl->gl_bucket = bucket;
296 INIT_LIST_HEAD(&gl->gl_reclaim);
300 lops_init_le(&gl->gl_le, &gfs2_glock_lops);
301 INIT_LIST_HEAD(&gl->gl_ail_list);
303 /* If this glock protects actual on-disk data or metadata blocks,
304 create a VFS inode to manage the pages/buffers holding them. */
305 if (glops == &gfs2_inode_glops ||
306 glops == &gfs2_rgrp_glops ||
307 glops == &gfs2_meta_glops) {
308 gl->gl_aspace = gfs2_aspace_get(sdp);
309 if (!gl->gl_aspace) {
315 error = gfs2_lm_get_lock(sdp, &name, &gl->gl_lock);
319 atomic_inc(&sdp->sd_glock_count);
321 write_lock(&bucket->hb_lock);
322 tmp = search_bucket(bucket, &name);
324 write_unlock(&bucket->hb_lock);
328 list_add_tail(&gl->gl_list, &bucket->hb_list);
329 write_unlock(&bucket->hb_lock);
338 gfs2_aspace_put(gl->gl_aspace);
341 kmem_cache_free(gfs2_glock_cachep, gl);
347 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
349 * @state: the state we're requesting
350 * @flags: the modifier flags
351 * @gh: the holder structure
355 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, int flags,
356 struct gfs2_holder *gh)
358 INIT_LIST_HEAD(&gh->gh_list);
360 gh->gh_owner = (flags & GL_NEVER_RECURSE) ? NULL : current;
361 gh->gh_state = state;
362 gh->gh_flags = flags;
365 init_completion(&gh->gh_wait);
367 if (gh->gh_state == LM_ST_EXCLUSIVE)
368 gh->gh_flags |= GL_LOCAL_EXCL;
374 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
375 * @state: the state we're requesting
376 * @flags: the modifier flags
377 * @gh: the holder structure
379 * Don't mess with the glock.
383 void gfs2_holder_reinit(unsigned int state, int flags, struct gfs2_holder *gh)
385 gh->gh_state = state;
386 gh->gh_flags = flags;
387 if (gh->gh_state == LM_ST_EXCLUSIVE)
388 gh->gh_flags |= GL_LOCAL_EXCL;
390 gh->gh_iflags &= 1 << HIF_ALLOCED;
394 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
395 * @gh: the holder structure
399 void gfs2_holder_uninit(struct gfs2_holder *gh)
401 gfs2_glock_put(gh->gh_gl);
406 * gfs2_holder_get - get a struct gfs2_holder structure
408 * @state: the state we're requesting
409 * @flags: the modifier flags
410 * @gfp_flags: __GFP_NOFAIL
412 * Figure out how big an impact this function has. Either:
413 * 1) Replace it with a cache of structures hanging off the struct gfs2_sbd
414 * 2) Leave it like it is
416 * Returns: the holder structure, NULL on ENOMEM
419 struct gfs2_holder *gfs2_holder_get(struct gfs2_glock *gl, unsigned int state,
420 int flags, gfp_t gfp_flags)
422 struct gfs2_holder *gh;
424 gh = kmalloc(sizeof(struct gfs2_holder), gfp_flags);
428 gfs2_holder_init(gl, state, flags, gh);
429 set_bit(HIF_ALLOCED, &gh->gh_iflags);
435 * gfs2_holder_put - get rid of a struct gfs2_holder structure
436 * @gh: the holder structure
440 void gfs2_holder_put(struct gfs2_holder *gh)
442 gfs2_holder_uninit(gh);
447 * handle_recurse - put other holder structures (marked recursive)
448 * into the holders list
449 * @gh: the holder structure
453 static void handle_recurse(struct gfs2_holder *gh)
455 struct gfs2_glock *gl = gh->gh_gl;
456 struct gfs2_sbd *sdp = gl->gl_sbd;
457 struct gfs2_holder *tmp_gh, *safe;
460 if (gfs2_assert_warn(sdp, gh->gh_owner))
463 list_for_each_entry_safe(tmp_gh, safe, &gl->gl_waiters3, gh_list) {
464 if (tmp_gh->gh_owner != gh->gh_owner)
467 gfs2_assert_warn(sdp,
468 test_bit(HIF_RECURSE, &tmp_gh->gh_iflags));
470 list_move_tail(&tmp_gh->gh_list, &gl->gl_holders);
471 tmp_gh->gh_error = 0;
472 set_bit(HIF_HOLDER, &tmp_gh->gh_iflags);
474 complete(&tmp_gh->gh_wait);
479 gfs2_assert_warn(sdp, found);
483 * do_unrecurse - a recursive holder was just dropped of the waiters3 list
486 * If there is only one other recursive holder, clear its HIF_RECURSE bit.
487 * If there is more than one, leave them alone.
491 static void do_unrecurse(struct gfs2_holder *gh)
493 struct gfs2_glock *gl = gh->gh_gl;
494 struct gfs2_sbd *sdp = gl->gl_sbd;
495 struct gfs2_holder *tmp_gh, *last_gh = NULL;
498 if (gfs2_assert_warn(sdp, gh->gh_owner))
501 list_for_each_entry(tmp_gh, &gl->gl_waiters3, gh_list) {
502 if (tmp_gh->gh_owner != gh->gh_owner)
505 gfs2_assert_warn(sdp,
506 test_bit(HIF_RECURSE, &tmp_gh->gh_iflags));
515 if (!gfs2_assert_warn(sdp, found))
516 clear_bit(HIF_RECURSE, &last_gh->gh_iflags);
520 * rq_mutex - process a mutex request in the queue
521 * @gh: the glock holder
523 * Returns: 1 if the queue is blocked
526 static int rq_mutex(struct gfs2_holder *gh)
528 struct gfs2_glock *gl = gh->gh_gl;
530 list_del_init(&gh->gh_list);
531 /* gh->gh_error never examined. */
532 set_bit(GLF_LOCK, &gl->gl_flags);
533 complete(&gh->gh_wait);
539 * rq_promote - process a promote request in the queue
540 * @gh: the glock holder
542 * Acquire a new inter-node lock, or change a lock state to more restrictive.
544 * Returns: 1 if the queue is blocked
547 static int rq_promote(struct gfs2_holder *gh)
549 struct gfs2_glock *gl = gh->gh_gl;
550 struct gfs2_sbd *sdp = gl->gl_sbd;
551 struct gfs2_glock_operations *glops = gl->gl_ops;
554 if (!relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
555 if (list_empty(&gl->gl_holders)) {
557 set_bit(GLF_LOCK, &gl->gl_flags);
558 spin_unlock(&gl->gl_spin);
560 if (atomic_read(&sdp->sd_reclaim_count) >
561 gfs2_tune_get(sdp, gt_reclaim_limit) &&
562 !(gh->gh_flags & LM_FLAG_PRIORITY)) {
563 gfs2_reclaim_glock(sdp);
564 gfs2_reclaim_glock(sdp);
567 glops->go_xmote_th(gl, gh->gh_state,
570 spin_lock(&gl->gl_spin);
575 if (list_empty(&gl->gl_holders)) {
576 set_bit(HIF_FIRST, &gh->gh_iflags);
577 set_bit(GLF_LOCK, &gl->gl_flags);
580 struct gfs2_holder *next_gh;
581 if (gh->gh_flags & GL_LOCAL_EXCL)
583 next_gh = list_entry(gl->gl_holders.next, struct gfs2_holder,
585 if (next_gh->gh_flags & GL_LOCAL_EXCL)
587 recurse = test_bit(HIF_RECURSE, &gh->gh_iflags);
590 list_move_tail(&gh->gh_list, &gl->gl_holders);
592 set_bit(HIF_HOLDER, &gh->gh_iflags);
597 complete(&gh->gh_wait);
603 * rq_demote - process a demote request in the queue
604 * @gh: the glock holder
606 * Returns: 1 if the queue is blocked
609 static int rq_demote(struct gfs2_holder *gh)
611 struct gfs2_glock *gl = gh->gh_gl;
612 struct gfs2_glock_operations *glops = gl->gl_ops;
614 if (!list_empty(&gl->gl_holders))
617 if (gl->gl_state == gh->gh_state || gl->gl_state == LM_ST_UNLOCKED) {
618 list_del_init(&gh->gh_list);
620 spin_unlock(&gl->gl_spin);
621 if (test_bit(HIF_DEALLOC, &gh->gh_iflags))
624 complete(&gh->gh_wait);
625 spin_lock(&gl->gl_spin);
628 set_bit(GLF_LOCK, &gl->gl_flags);
629 spin_unlock(&gl->gl_spin);
631 if (gh->gh_state == LM_ST_UNLOCKED ||
632 gl->gl_state != LM_ST_EXCLUSIVE)
633 glops->go_drop_th(gl);
635 glops->go_xmote_th(gl, gh->gh_state, gh->gh_flags);
637 spin_lock(&gl->gl_spin);
644 * rq_greedy - process a queued request to drop greedy status
645 * @gh: the glock holder
647 * Returns: 1 if the queue is blocked
650 static int rq_greedy(struct gfs2_holder *gh)
652 struct gfs2_glock *gl = gh->gh_gl;
654 list_del_init(&gh->gh_list);
655 /* gh->gh_error never examined. */
656 clear_bit(GLF_GREEDY, &gl->gl_flags);
657 spin_unlock(&gl->gl_spin);
659 gfs2_holder_uninit(gh);
660 kfree(container_of(gh, struct greedy, gr_gh));
662 spin_lock(&gl->gl_spin);
668 * run_queue - process holder structures on a glock
673 static void run_queue(struct gfs2_glock *gl)
675 struct gfs2_holder *gh;
679 if (test_bit(GLF_LOCK, &gl->gl_flags))
682 if (!list_empty(&gl->gl_waiters1)) {
683 gh = list_entry(gl->gl_waiters1.next,
684 struct gfs2_holder, gh_list);
686 if (test_bit(HIF_MUTEX, &gh->gh_iflags))
687 blocked = rq_mutex(gh);
689 gfs2_assert_warn(gl->gl_sbd, 0);
691 } else if (!list_empty(&gl->gl_waiters2) &&
692 !test_bit(GLF_SKIP_WAITERS2, &gl->gl_flags)) {
693 gh = list_entry(gl->gl_waiters2.next,
694 struct gfs2_holder, gh_list);
696 if (test_bit(HIF_DEMOTE, &gh->gh_iflags))
697 blocked = rq_demote(gh);
698 else if (test_bit(HIF_GREEDY, &gh->gh_iflags))
699 blocked = rq_greedy(gh);
701 gfs2_assert_warn(gl->gl_sbd, 0);
703 } else if (!list_empty(&gl->gl_waiters3)) {
704 gh = list_entry(gl->gl_waiters3.next,
705 struct gfs2_holder, gh_list);
707 if (test_bit(HIF_PROMOTE, &gh->gh_iflags))
708 blocked = rq_promote(gh);
710 gfs2_assert_warn(gl->gl_sbd, 0);
721 * gfs2_glmutex_lock - acquire a local lock on a glock
724 * Gives caller exclusive access to manipulate a glock structure.
727 void gfs2_glmutex_lock(struct gfs2_glock *gl)
729 struct gfs2_holder gh;
731 gfs2_holder_init(gl, 0, 0, &gh);
732 set_bit(HIF_MUTEX, &gh.gh_iflags);
734 spin_lock(&gl->gl_spin);
735 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
736 list_add_tail(&gh.gh_list, &gl->gl_waiters1);
738 complete(&gh.gh_wait);
739 spin_unlock(&gl->gl_spin);
741 wait_for_completion(&gh.gh_wait);
742 gfs2_holder_uninit(&gh);
746 * gfs2_glmutex_trylock - try to acquire a local lock on a glock
749 * Returns: 1 if the glock is acquired
752 int gfs2_glmutex_trylock(struct gfs2_glock *gl)
756 spin_lock(&gl->gl_spin);
757 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
759 spin_unlock(&gl->gl_spin);
765 * gfs2_glmutex_unlock - release a local lock on a glock
770 void gfs2_glmutex_unlock(struct gfs2_glock *gl)
772 spin_lock(&gl->gl_spin);
773 clear_bit(GLF_LOCK, &gl->gl_flags);
775 spin_unlock(&gl->gl_spin);
779 * handle_callback - add a demote request to a lock's queue
781 * @state: the state the caller wants us to change to
785 static void handle_callback(struct gfs2_glock *gl, unsigned int state)
787 struct gfs2_holder *gh, *new_gh = NULL;
790 spin_lock(&gl->gl_spin);
792 list_for_each_entry(gh, &gl->gl_waiters2, gh_list) {
793 if (test_bit(HIF_DEMOTE, &gh->gh_iflags) &&
794 gl->gl_req_gh != gh) {
795 if (gh->gh_state != state)
796 gh->gh_state = LM_ST_UNLOCKED;
802 list_add_tail(&new_gh->gh_list, &gl->gl_waiters2);
805 spin_unlock(&gl->gl_spin);
807 new_gh = gfs2_holder_get(gl, state,
808 LM_FLAG_TRY | GL_NEVER_RECURSE,
809 GFP_KERNEL | __GFP_NOFAIL),
810 set_bit(HIF_DEMOTE, &new_gh->gh_iflags);
811 set_bit(HIF_DEALLOC, &new_gh->gh_iflags);
817 spin_unlock(&gl->gl_spin);
820 gfs2_holder_put(new_gh);
824 * state_change - record that the glock is now in a different state
826 * @new_state the new state
830 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
832 struct gfs2_sbd *sdp = gl->gl_sbd;
835 held1 = (gl->gl_state != LM_ST_UNLOCKED);
836 held2 = (new_state != LM_ST_UNLOCKED);
838 if (held1 != held2) {
840 atomic_inc(&sdp->sd_glock_held_count);
843 atomic_dec(&sdp->sd_glock_held_count);
848 gl->gl_state = new_state;
852 * xmote_bh - Called after the lock module is done acquiring a lock
853 * @gl: The glock in question
854 * @ret: the int returned from the lock module
858 static void xmote_bh(struct gfs2_glock *gl, unsigned int ret)
860 struct gfs2_sbd *sdp = gl->gl_sbd;
861 struct gfs2_glock_operations *glops = gl->gl_ops;
862 struct gfs2_holder *gh = gl->gl_req_gh;
863 int prev_state = gl->gl_state;
866 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
867 gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
868 gfs2_assert_warn(sdp, !(ret & LM_OUT_ASYNC));
870 state_change(gl, ret & LM_OUT_ST_MASK);
872 if (prev_state != LM_ST_UNLOCKED && !(ret & LM_OUT_CACHEABLE)) {
874 glops->go_inval(gl, DIO_METADATA | DIO_DATA);
875 } else if (gl->gl_state == LM_ST_DEFERRED) {
876 /* We might not want to do this here.
877 Look at moving to the inode glops. */
879 glops->go_inval(gl, DIO_DATA);
882 /* Deal with each possible exit condition */
885 gl->gl_stamp = jiffies;
887 else if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
888 spin_lock(&gl->gl_spin);
889 list_del_init(&gh->gh_list);
891 if (test_bit(HIF_RECURSE, &gh->gh_iflags))
893 spin_unlock(&gl->gl_spin);
895 } else if (test_bit(HIF_DEMOTE, &gh->gh_iflags)) {
896 spin_lock(&gl->gl_spin);
897 list_del_init(&gh->gh_list);
898 if (gl->gl_state == gh->gh_state ||
899 gl->gl_state == LM_ST_UNLOCKED)
902 if (gfs2_assert_warn(sdp, gh->gh_flags &
903 (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) == -1)
904 fs_warn(sdp, "ret = 0x%.8X\n", ret);
905 gh->gh_error = GLR_TRYFAILED;
907 spin_unlock(&gl->gl_spin);
909 if (ret & LM_OUT_CANCELED)
910 handle_callback(gl, LM_ST_UNLOCKED); /* Lame */
912 } else if (ret & LM_OUT_CANCELED) {
913 spin_lock(&gl->gl_spin);
914 list_del_init(&gh->gh_list);
915 gh->gh_error = GLR_CANCELED;
916 if (test_bit(HIF_RECURSE, &gh->gh_iflags))
918 spin_unlock(&gl->gl_spin);
920 } else if (relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
921 spin_lock(&gl->gl_spin);
922 list_move_tail(&gh->gh_list, &gl->gl_holders);
924 set_bit(HIF_HOLDER, &gh->gh_iflags);
925 spin_unlock(&gl->gl_spin);
927 set_bit(HIF_FIRST, &gh->gh_iflags);
931 } else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
932 spin_lock(&gl->gl_spin);
933 list_del_init(&gh->gh_list);
934 gh->gh_error = GLR_TRYFAILED;
935 if (test_bit(HIF_RECURSE, &gh->gh_iflags))
937 spin_unlock(&gl->gl_spin);
940 if (gfs2_assert_withdraw(sdp, 0) == -1)
941 fs_err(sdp, "ret = 0x%.8X\n", ret);
944 if (glops->go_xmote_bh)
945 glops->go_xmote_bh(gl);
948 spin_lock(&gl->gl_spin);
949 gl->gl_req_gh = NULL;
950 gl->gl_req_bh = NULL;
951 clear_bit(GLF_LOCK, &gl->gl_flags);
953 spin_unlock(&gl->gl_spin);
959 if (test_bit(HIF_DEALLOC, &gh->gh_iflags))
962 complete(&gh->gh_wait);
967 * gfs2_glock_xmote_th - Call into the lock module to acquire or change a glock
968 * @gl: The glock in question
969 * @state: the requested state
970 * @flags: modifier flags to the lock call
974 void gfs2_glock_xmote_th(struct gfs2_glock *gl, unsigned int state, int flags)
976 struct gfs2_sbd *sdp = gl->gl_sbd;
977 struct gfs2_glock_operations *glops = gl->gl_ops;
978 int lck_flags = flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB |
979 LM_FLAG_NOEXP | LM_FLAG_ANY |
981 unsigned int lck_ret;
983 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
984 gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
985 gfs2_assert_warn(sdp, state != LM_ST_UNLOCKED);
986 gfs2_assert_warn(sdp, state != gl->gl_state);
988 if (gl->gl_state == LM_ST_EXCLUSIVE) {
991 DIO_METADATA | DIO_DATA | DIO_RELEASE);
995 gl->gl_req_bh = xmote_bh;
997 atomic_inc(&sdp->sd_lm_lock_calls);
999 lck_ret = gfs2_lm_lock(sdp, gl->gl_lock, gl->gl_state, state,
1002 if (gfs2_assert_withdraw(sdp, !(lck_ret & LM_OUT_ERROR)))
1005 if (lck_ret & LM_OUT_ASYNC)
1006 gfs2_assert_warn(sdp, lck_ret == LM_OUT_ASYNC);
1008 xmote_bh(gl, lck_ret);
1012 * drop_bh - Called after a lock module unlock completes
1014 * @ret: the return status
1016 * Doesn't wake up the process waiting on the struct gfs2_holder (if any)
1017 * Doesn't drop the reference on the glock the top half took out
1021 static void drop_bh(struct gfs2_glock *gl, unsigned int ret)
1023 struct gfs2_sbd *sdp = gl->gl_sbd;
1024 struct gfs2_glock_operations *glops = gl->gl_ops;
1025 struct gfs2_holder *gh = gl->gl_req_gh;
1027 clear_bit(GLF_PREFETCH, &gl->gl_flags);
1029 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1030 gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
1031 gfs2_assert_warn(sdp, !ret);
1033 state_change(gl, LM_ST_UNLOCKED);
1035 if (glops->go_inval)
1036 glops->go_inval(gl, DIO_METADATA | DIO_DATA);
1039 spin_lock(&gl->gl_spin);
1040 list_del_init(&gh->gh_list);
1042 spin_unlock(&gl->gl_spin);
1045 if (glops->go_drop_bh)
1046 glops->go_drop_bh(gl);
1048 spin_lock(&gl->gl_spin);
1049 gl->gl_req_gh = NULL;
1050 gl->gl_req_bh = NULL;
1051 clear_bit(GLF_LOCK, &gl->gl_flags);
1053 spin_unlock(&gl->gl_spin);
1058 if (test_bit(HIF_DEALLOC, &gh->gh_iflags))
1059 gfs2_holder_put(gh);
1061 complete(&gh->gh_wait);
1066 * gfs2_glock_drop_th - call into the lock module to unlock a lock
1071 void gfs2_glock_drop_th(struct gfs2_glock *gl)
1073 struct gfs2_sbd *sdp = gl->gl_sbd;
1074 struct gfs2_glock_operations *glops = gl->gl_ops;
1077 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1078 gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
1079 gfs2_assert_warn(sdp, gl->gl_state != LM_ST_UNLOCKED);
1081 if (gl->gl_state == LM_ST_EXCLUSIVE) {
1084 DIO_METADATA | DIO_DATA | DIO_RELEASE);
1087 gfs2_glock_hold(gl);
1088 gl->gl_req_bh = drop_bh;
1090 atomic_inc(&sdp->sd_lm_unlock_calls);
1092 ret = gfs2_lm_unlock(sdp, gl->gl_lock, gl->gl_state);
1094 if (gfs2_assert_withdraw(sdp, !(ret & LM_OUT_ERROR)))
1100 gfs2_assert_warn(sdp, ret == LM_OUT_ASYNC);
1104 * do_cancels - cancel requests for locks stuck waiting on an expire flag
1105 * @gh: the LM_FLAG_PRIORITY holder waiting to acquire the lock
1107 * Don't cancel GL_NOCANCEL requests.
1110 static void do_cancels(struct gfs2_holder *gh)
1112 struct gfs2_glock *gl = gh->gh_gl;
1114 spin_lock(&gl->gl_spin);
1116 while (gl->gl_req_gh != gh &&
1117 !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1118 !list_empty(&gh->gh_list)) {
1119 if (gl->gl_req_bh &&
1121 (gl->gl_req_gh->gh_flags & GL_NOCANCEL))) {
1122 spin_unlock(&gl->gl_spin);
1123 gfs2_lm_cancel(gl->gl_sbd, gl->gl_lock);
1125 spin_lock(&gl->gl_spin);
1127 spin_unlock(&gl->gl_spin);
1129 spin_lock(&gl->gl_spin);
1133 spin_unlock(&gl->gl_spin);
1137 * glock_wait_internal - wait on a glock acquisition
1138 * @gh: the glock holder
1140 * Returns: 0 on success
1143 static int glock_wait_internal(struct gfs2_holder *gh)
1145 struct gfs2_glock *gl = gh->gh_gl;
1146 struct gfs2_sbd *sdp = gl->gl_sbd;
1147 struct gfs2_glock_operations *glops = gl->gl_ops;
1149 if (test_bit(HIF_ABORTED, &gh->gh_iflags))
1152 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1153 spin_lock(&gl->gl_spin);
1154 if (gl->gl_req_gh != gh &&
1155 !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1156 !list_empty(&gh->gh_list)) {
1157 list_del_init(&gh->gh_list);
1158 gh->gh_error = GLR_TRYFAILED;
1159 if (test_bit(HIF_RECURSE, &gh->gh_iflags))
1162 spin_unlock(&gl->gl_spin);
1163 return gh->gh_error;
1165 spin_unlock(&gl->gl_spin);
1168 if (gh->gh_flags & LM_FLAG_PRIORITY)
1171 wait_for_completion(&gh->gh_wait);
1174 return gh->gh_error;
1176 gfs2_assert_withdraw(sdp, test_bit(HIF_HOLDER, &gh->gh_iflags));
1177 gfs2_assert_withdraw(sdp, relaxed_state_ok(gl->gl_state,
1181 if (test_bit(HIF_FIRST, &gh->gh_iflags)) {
1182 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1184 if (glops->go_lock) {
1185 gh->gh_error = glops->go_lock(gh);
1187 spin_lock(&gl->gl_spin);
1188 list_del_init(&gh->gh_list);
1189 if (test_and_clear_bit(HIF_RECURSE,
1192 spin_unlock(&gl->gl_spin);
1196 spin_lock(&gl->gl_spin);
1197 gl->gl_req_gh = NULL;
1198 gl->gl_req_bh = NULL;
1199 clear_bit(GLF_LOCK, &gl->gl_flags);
1200 if (test_bit(HIF_RECURSE, &gh->gh_iflags))
1203 spin_unlock(&gl->gl_spin);
1206 return gh->gh_error;
1209 static inline struct gfs2_holder *
1210 find_holder_by_owner(struct list_head *head, struct task_struct *owner)
1212 struct gfs2_holder *gh;
1214 list_for_each_entry(gh, head, gh_list) {
1215 if (gh->gh_owner == owner)
1225 * Make sure the new holder is compatible with the pre-existing one.
1229 static int recurse_check(struct gfs2_holder *existing, struct gfs2_holder *new,
1232 struct gfs2_sbd *sdp = existing->gh_gl->gl_sbd;
1234 if (gfs2_assert_warn(sdp, (new->gh_flags & LM_FLAG_ANY) ||
1235 !(existing->gh_flags & LM_FLAG_ANY)))
1238 if (gfs2_assert_warn(sdp, (existing->gh_flags & GL_LOCAL_EXCL) ||
1239 !(new->gh_flags & GL_LOCAL_EXCL)))
1242 if (gfs2_assert_warn(sdp, relaxed_state_ok(state, new->gh_state,
1249 set_bit(HIF_ABORTED, &new->gh_iflags);
1254 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1255 * @gh: the holder structure to add
1259 static void add_to_queue(struct gfs2_holder *gh)
1261 struct gfs2_glock *gl = gh->gh_gl;
1262 struct gfs2_holder *existing;
1267 existing = find_holder_by_owner(&gl->gl_holders, gh->gh_owner);
1269 if (recurse_check(existing, gh, gl->gl_state))
1272 list_add_tail(&gh->gh_list, &gl->gl_holders);
1273 set_bit(HIF_HOLDER, &gh->gh_iflags);
1276 complete(&gh->gh_wait);
1281 existing = find_holder_by_owner(&gl->gl_waiters3, gh->gh_owner);
1283 if (recurse_check(existing, gh, existing->gh_state))
1286 set_bit(HIF_RECURSE, &gh->gh_iflags);
1287 set_bit(HIF_RECURSE, &existing->gh_iflags);
1289 list_add_tail(&gh->gh_list, &gl->gl_waiters3);
1295 if (gh->gh_flags & LM_FLAG_PRIORITY)
1296 list_add(&gh->gh_list, &gl->gl_waiters3);
1298 list_add_tail(&gh->gh_list, &gl->gl_waiters3);
1302 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1303 * @gh: the holder structure
1305 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1307 * Returns: 0, GLR_TRYFAILED, or errno on failure
1310 int gfs2_glock_nq(struct gfs2_holder *gh)
1312 struct gfs2_glock *gl = gh->gh_gl;
1313 struct gfs2_sbd *sdp = gl->gl_sbd;
1316 atomic_inc(&sdp->sd_glock_nq_calls);
1319 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1320 set_bit(HIF_ABORTED, &gh->gh_iflags);
1324 set_bit(HIF_PROMOTE, &gh->gh_iflags);
1326 spin_lock(&gl->gl_spin);
1329 spin_unlock(&gl->gl_spin);
1331 if (!(gh->gh_flags & GL_ASYNC)) {
1332 error = glock_wait_internal(gh);
1333 if (error == GLR_CANCELED) {
1339 clear_bit(GLF_PREFETCH, &gl->gl_flags);
1345 * gfs2_glock_poll - poll to see if an async request has been completed
1348 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1351 int gfs2_glock_poll(struct gfs2_holder *gh)
1353 struct gfs2_glock *gl = gh->gh_gl;
1356 spin_lock(&gl->gl_spin);
1358 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1360 else if (list_empty(&gh->gh_list)) {
1361 if (gh->gh_error == GLR_CANCELED) {
1362 spin_unlock(&gl->gl_spin);
1364 if (gfs2_glock_nq(gh))
1371 spin_unlock(&gl->gl_spin);
1377 * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
1378 * @gh: the holder structure
1380 * Returns: 0, GLR_TRYFAILED, or errno on failure
1383 int gfs2_glock_wait(struct gfs2_holder *gh)
1387 error = glock_wait_internal(gh);
1388 if (error == GLR_CANCELED) {
1390 gh->gh_flags &= ~GL_ASYNC;
1391 error = gfs2_glock_nq(gh);
1398 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1399 * @gh: the glock holder
1403 void gfs2_glock_dq(struct gfs2_holder *gh)
1405 struct gfs2_glock *gl = gh->gh_gl;
1406 struct gfs2_sbd *sdp = gl->gl_sbd;
1407 struct gfs2_glock_operations *glops = gl->gl_ops;
1409 atomic_inc(&sdp->sd_glock_dq_calls);
1411 if (gh->gh_flags & GL_SYNC)
1412 set_bit(GLF_SYNC, &gl->gl_flags);
1414 if (gh->gh_flags & GL_NOCACHE)
1415 handle_callback(gl, LM_ST_UNLOCKED);
1417 gfs2_glmutex_lock(gl);
1419 spin_lock(&gl->gl_spin);
1420 list_del_init(&gh->gh_list);
1422 if (list_empty(&gl->gl_holders)) {
1423 spin_unlock(&gl->gl_spin);
1425 if (glops->go_unlock)
1426 glops->go_unlock(gh);
1428 if (test_bit(GLF_SYNC, &gl->gl_flags)) {
1430 glops->go_sync(gl, DIO_METADATA | DIO_DATA);
1433 gl->gl_stamp = jiffies;
1435 spin_lock(&gl->gl_spin);
1438 clear_bit(GLF_LOCK, &gl->gl_flags);
1440 spin_unlock(&gl->gl_spin);
1444 * gfs2_glock_prefetch - Try to prefetch a glock
1446 * @state: the state to prefetch in
1447 * @flags: flags passed to go_xmote_th()
1451 void gfs2_glock_prefetch(struct gfs2_glock *gl, unsigned int state, int flags)
1453 struct gfs2_glock_operations *glops = gl->gl_ops;
1455 spin_lock(&gl->gl_spin);
1457 if (test_bit(GLF_LOCK, &gl->gl_flags) ||
1458 !list_empty(&gl->gl_holders) ||
1459 !list_empty(&gl->gl_waiters1) ||
1460 !list_empty(&gl->gl_waiters2) ||
1461 !list_empty(&gl->gl_waiters3) ||
1462 relaxed_state_ok(gl->gl_state, state, flags)) {
1463 spin_unlock(&gl->gl_spin);
1467 set_bit(GLF_PREFETCH, &gl->gl_flags);
1468 set_bit(GLF_LOCK, &gl->gl_flags);
1469 spin_unlock(&gl->gl_spin);
1471 glops->go_xmote_th(gl, state, flags);
1473 atomic_inc(&gl->gl_sbd->sd_glock_prefetch_calls);
1477 * gfs2_glock_force_drop - Force a glock to be uncached
1482 void gfs2_glock_force_drop(struct gfs2_glock *gl)
1484 struct gfs2_holder gh;
1486 gfs2_holder_init(gl, LM_ST_UNLOCKED, GL_NEVER_RECURSE, &gh);
1487 set_bit(HIF_DEMOTE, &gh.gh_iflags);
1489 spin_lock(&gl->gl_spin);
1490 list_add_tail(&gh.gh_list, &gl->gl_waiters2);
1492 spin_unlock(&gl->gl_spin);
1494 wait_for_completion(&gh.gh_wait);
1495 gfs2_holder_uninit(&gh);
1498 static void greedy_work(void *data)
1500 struct greedy *gr = (struct greedy *)data;
1501 struct gfs2_holder *gh = &gr->gr_gh;
1502 struct gfs2_glock *gl = gh->gh_gl;
1503 struct gfs2_glock_operations *glops = gl->gl_ops;
1505 clear_bit(GLF_SKIP_WAITERS2, &gl->gl_flags);
1507 if (glops->go_greedy)
1508 glops->go_greedy(gl);
1510 spin_lock(&gl->gl_spin);
1512 if (list_empty(&gl->gl_waiters2)) {
1513 clear_bit(GLF_GREEDY, &gl->gl_flags);
1514 spin_unlock(&gl->gl_spin);
1515 gfs2_holder_uninit(gh);
1518 gfs2_glock_hold(gl);
1519 list_add_tail(&gh->gh_list, &gl->gl_waiters2);
1521 spin_unlock(&gl->gl_spin);
1527 * gfs2_glock_be_greedy -
1531 * Returns: 0 if go_greedy will be called, 1 otherwise
1534 int gfs2_glock_be_greedy(struct gfs2_glock *gl, unsigned int time)
1537 struct gfs2_holder *gh;
1540 gl->gl_sbd->sd_args.ar_localcaching ||
1541 test_and_set_bit(GLF_GREEDY, &gl->gl_flags))
1544 gr = kmalloc(sizeof(struct greedy), GFP_KERNEL);
1546 clear_bit(GLF_GREEDY, &gl->gl_flags);
1551 gfs2_holder_init(gl, 0, GL_NEVER_RECURSE, gh);
1552 set_bit(HIF_GREEDY, &gh->gh_iflags);
1553 INIT_WORK(&gr->gr_work, greedy_work, gr);
1555 set_bit(GLF_SKIP_WAITERS2, &gl->gl_flags);
1556 schedule_delayed_work(&gr->gr_work, time);
1562 * gfs2_glock_nq_init - intialize a holder and enqueue it on a glock
1564 * @state: the state we're requesting
1565 * @flags: the modifier flags
1566 * @gh: the holder structure
1568 * Returns: 0, GLR_*, or errno
1571 int gfs2_glock_nq_init(struct gfs2_glock *gl, unsigned int state, int flags,
1572 struct gfs2_holder *gh)
1576 gfs2_holder_init(gl, state, flags, gh);
1578 error = gfs2_glock_nq(gh);
1580 gfs2_holder_uninit(gh);
1586 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1587 * @gh: the holder structure
1591 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1594 gfs2_holder_uninit(gh);
1598 * gfs2_glock_nq_num - acquire a glock based on lock number
1599 * @sdp: the filesystem
1600 * @number: the lock number
1601 * @glops: the glock operations for the type of glock
1602 * @state: the state to acquire the glock in
1603 * @flags: modifier flags for the aquisition
1604 * @gh: the struct gfs2_holder
1609 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, uint64_t number,
1610 struct gfs2_glock_operations *glops, unsigned int state,
1611 int flags, struct gfs2_holder *gh)
1613 struct gfs2_glock *gl;
1616 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1618 error = gfs2_glock_nq_init(gl, state, flags, gh);
1626 * glock_compare - Compare two struct gfs2_glock structures for sorting
1627 * @arg_a: the first structure
1628 * @arg_b: the second structure
1632 static int glock_compare(const void *arg_a, const void *arg_b)
1634 struct gfs2_holder *gh_a = *(struct gfs2_holder **)arg_a;
1635 struct gfs2_holder *gh_b = *(struct gfs2_holder **)arg_b;
1636 struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1637 struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1640 if (a->ln_number > b->ln_number)
1642 else if (a->ln_number < b->ln_number)
1645 if (gh_a->gh_state == LM_ST_SHARED &&
1646 gh_b->gh_state == LM_ST_EXCLUSIVE)
1648 else if (!(gh_a->gh_flags & GL_LOCAL_EXCL) &&
1649 (gh_b->gh_flags & GL_LOCAL_EXCL))
1657 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1658 * @num_gh: the number of structures
1659 * @ghs: an array of struct gfs2_holder structures
1661 * Returns: 0 on success (all glocks acquired),
1662 * errno on failure (no glocks acquired)
1665 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1666 struct gfs2_holder **p)
1671 for (x = 0; x < num_gh; x++)
1674 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1676 for (x = 0; x < num_gh; x++) {
1677 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1679 error = gfs2_glock_nq(p[x]);
1682 gfs2_glock_dq(p[x]);
1691 * gfs2_glock_nq_m - acquire multiple glocks
1692 * @num_gh: the number of structures
1693 * @ghs: an array of struct gfs2_holder structures
1695 * Figure out how big an impact this function has. Either:
1696 * 1) Replace this code with code that calls gfs2_glock_prefetch()
1697 * 2) Forget async stuff and just call nq_m_sync()
1698 * 3) Leave it like it is
1700 * Returns: 0 on success (all glocks acquired),
1701 * errno on failure (no glocks acquired)
1704 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1708 int borked = 0, serious = 0;
1715 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1716 return gfs2_glock_nq(ghs);
1719 e = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL);
1723 for (x = 0; x < num_gh; x++) {
1724 ghs[x].gh_flags |= LM_FLAG_TRY | GL_ASYNC;
1725 error = gfs2_glock_nq(&ghs[x]);
1734 for (x = 0; x < num_gh; x++) {
1735 error = e[x] = glock_wait_internal(&ghs[x]);
1738 if (error != GLR_TRYFAILED && error != GLR_CANCELED)
1748 for (x = 0; x < num_gh; x++)
1750 gfs2_glock_dq(&ghs[x]);
1755 for (x = 0; x < num_gh; x++)
1756 gfs2_holder_reinit(ghs[x].gh_state, ghs[x].gh_flags,
1758 error = nq_m_sync(num_gh, ghs, (struct gfs2_holder **)e);
1767 * gfs2_glock_dq_m - release multiple glocks
1768 * @num_gh: the number of structures
1769 * @ghs: an array of struct gfs2_holder structures
1773 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1777 for (x = 0; x < num_gh; x++)
1778 gfs2_glock_dq(&ghs[x]);
1782 * gfs2_glock_dq_uninit_m - release multiple glocks
1783 * @num_gh: the number of structures
1784 * @ghs: an array of struct gfs2_holder structures
1788 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1792 for (x = 0; x < num_gh; x++)
1793 gfs2_glock_dq_uninit(&ghs[x]);
1797 * gfs2_glock_prefetch_num - prefetch a glock based on lock number
1798 * @sdp: the filesystem
1799 * @number: the lock number
1800 * @glops: the glock operations for the type of glock
1801 * @state: the state to acquire the glock in
1802 * @flags: modifier flags for the aquisition
1807 void gfs2_glock_prefetch_num(struct gfs2_sbd *sdp, uint64_t number,
1808 struct gfs2_glock_operations *glops,
1809 unsigned int state, int flags)
1811 struct gfs2_glock *gl;
1814 if (atomic_read(&sdp->sd_reclaim_count) <
1815 gfs2_tune_get(sdp, gt_reclaim_limit)) {
1816 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1818 gfs2_glock_prefetch(gl, state, flags);
1825 * gfs2_lvb_hold - attach a LVB from a glock
1826 * @gl: The glock in question
1830 int gfs2_lvb_hold(struct gfs2_glock *gl)
1834 gfs2_glmutex_lock(gl);
1836 if (!atomic_read(&gl->gl_lvb_count)) {
1837 error = gfs2_lm_hold_lvb(gl->gl_sbd, gl->gl_lock, &gl->gl_lvb);
1839 gfs2_glmutex_unlock(gl);
1842 gfs2_glock_hold(gl);
1844 atomic_inc(&gl->gl_lvb_count);
1846 gfs2_glmutex_unlock(gl);
1852 * gfs2_lvb_unhold - detach a LVB from a glock
1853 * @gl: The glock in question
1857 void gfs2_lvb_unhold(struct gfs2_glock *gl)
1859 gfs2_glock_hold(gl);
1860 gfs2_glmutex_lock(gl);
1862 gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count) > 0);
1863 if (atomic_dec_and_test(&gl->gl_lvb_count)) {
1864 gfs2_lm_unhold_lvb(gl->gl_sbd, gl->gl_lock, gl->gl_lvb);
1869 gfs2_glmutex_unlock(gl);
1873 void gfs2_lvb_sync(struct gfs2_glock *gl)
1875 gfs2_glmutex_lock(gl);
1877 gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count));
1878 if (!gfs2_assert_warn(gl->gl_sbd, gfs2_glock_is_held_excl(gl)))
1879 gfs2_lm_sync_lvb(gl->gl_sbd, gl->gl_lock, gl->gl_lvb);
1881 gfs2_glmutex_unlock(gl);
1884 static void blocking_cb(struct gfs2_sbd *sdp, struct lm_lockname *name,
1887 struct gfs2_glock *gl;
1889 gl = gfs2_glock_find(sdp, name);
1893 if (gl->gl_ops->go_callback)
1894 gl->gl_ops->go_callback(gl, state);
1895 handle_callback(gl, state);
1897 spin_lock(&gl->gl_spin);
1899 spin_unlock(&gl->gl_spin);
1905 * gfs2_glock_cb - Callback used by locking module
1906 * @fsdata: Pointer to the superblock
1907 * @type: Type of callback
1908 * @data: Type dependent data pointer
1910 * Called by the locking module when it wants to tell us something.
1911 * Either we need to drop a lock, one of our ASYNC requests completed, or
1912 * a journal from another client needs to be recovered.
1915 void gfs2_glock_cb(lm_fsdata_t *fsdata, unsigned int type, void *data)
1917 struct gfs2_sbd *sdp = (struct gfs2_sbd *)fsdata;
1919 atomic_inc(&sdp->sd_lm_callbacks);
1923 blocking_cb(sdp, (struct lm_lockname *)data, LM_ST_UNLOCKED);
1927 blocking_cb(sdp, (struct lm_lockname *)data, LM_ST_DEFERRED);
1931 blocking_cb(sdp, (struct lm_lockname *)data, LM_ST_SHARED);
1935 struct lm_async_cb *async = (struct lm_async_cb *)data;
1936 struct gfs2_glock *gl;
1938 gl = gfs2_glock_find(sdp, &async->lc_name);
1939 if (gfs2_assert_warn(sdp, gl))
1941 if (!gfs2_assert_warn(sdp, gl->gl_req_bh))
1942 gl->gl_req_bh(gl, async->lc_ret);
1948 case LM_CB_NEED_RECOVERY:
1949 gfs2_jdesc_make_dirty(sdp, *(unsigned int *)data);
1950 if (sdp->sd_recoverd_process)
1951 wake_up_process(sdp->sd_recoverd_process);
1954 case LM_CB_DROPLOCKS:
1955 gfs2_gl_hash_clear(sdp, NO_WAIT);
1956 gfs2_quota_scan(sdp);
1960 gfs2_assert_warn(sdp, 0);
1966 * gfs2_try_toss_inode - try to remove a particular inode struct from cache
1967 * sdp: the filesystem
1968 * inum: the inode number
1972 void gfs2_try_toss_inode(struct gfs2_sbd *sdp, struct gfs2_inum *inum)
1974 struct gfs2_glock *gl;
1975 struct gfs2_inode *ip;
1978 error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_inode_glops,
1983 if (!gfs2_glmutex_trylock(gl))
1990 if (atomic_read(&ip->i_count))
1993 gfs2_inode_destroy(ip);
1996 gfs2_glmutex_unlock(gl);
2003 * gfs2_iopen_go_callback - Try to kick the inode/vnode associated with an
2004 * iopen glock from memory
2005 * @io_gl: the iopen glock
2006 * @state: the state into which the glock should be put
2010 void gfs2_iopen_go_callback(struct gfs2_glock *io_gl, unsigned int state)
2012 struct gfs2_glock *i_gl;
2014 if (state != LM_ST_UNLOCKED)
2017 spin_lock(&io_gl->gl_spin);
2018 i_gl = get_gl2gl(io_gl);
2020 gfs2_glock_hold(i_gl);
2021 spin_unlock(&io_gl->gl_spin);
2023 spin_unlock(&io_gl->gl_spin);
2027 if (gfs2_glmutex_trylock(i_gl)) {
2028 struct gfs2_inode *ip = get_gl2ip(i_gl);
2030 gfs2_try_toss_vnode(ip);
2031 gfs2_glmutex_unlock(i_gl);
2032 gfs2_glock_schedule_for_reclaim(i_gl);
2035 gfs2_glmutex_unlock(i_gl);
2039 gfs2_glock_put(i_gl);
2043 * demote_ok - Check to see if it's ok to unlock a glock
2046 * Returns: 1 if it's ok
2049 static int demote_ok(struct gfs2_glock *gl)
2051 struct gfs2_sbd *sdp = gl->gl_sbd;
2052 struct gfs2_glock_operations *glops = gl->gl_ops;
2055 if (test_bit(GLF_STICKY, &gl->gl_flags))
2057 else if (test_bit(GLF_PREFETCH, &gl->gl_flags))
2058 demote = time_after_eq(jiffies,
2060 gfs2_tune_get(sdp, gt_prefetch_secs) * HZ);
2061 else if (glops->go_demote_ok)
2062 demote = glops->go_demote_ok(gl);
2068 * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
2073 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
2075 struct gfs2_sbd *sdp = gl->gl_sbd;
2077 spin_lock(&sdp->sd_reclaim_lock);
2078 if (list_empty(&gl->gl_reclaim)) {
2079 gfs2_glock_hold(gl);
2080 list_add(&gl->gl_reclaim, &sdp->sd_reclaim_list);
2081 atomic_inc(&sdp->sd_reclaim_count);
2083 spin_unlock(&sdp->sd_reclaim_lock);
2085 wake_up(&sdp->sd_reclaim_wq);
2089 * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
2090 * @sdp: the filesystem
2092 * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
2093 * different glock and we notice that there are a lot of glocks in the
2098 void gfs2_reclaim_glock(struct gfs2_sbd *sdp)
2100 struct gfs2_glock *gl;
2102 spin_lock(&sdp->sd_reclaim_lock);
2103 if (list_empty(&sdp->sd_reclaim_list)) {
2104 spin_unlock(&sdp->sd_reclaim_lock);
2107 gl = list_entry(sdp->sd_reclaim_list.next,
2108 struct gfs2_glock, gl_reclaim);
2109 list_del_init(&gl->gl_reclaim);
2110 spin_unlock(&sdp->sd_reclaim_lock);
2112 atomic_dec(&sdp->sd_reclaim_count);
2113 atomic_inc(&sdp->sd_reclaimed);
2115 if (gfs2_glmutex_trylock(gl)) {
2116 if (gl->gl_ops == &gfs2_inode_glops) {
2117 struct gfs2_inode *ip = get_gl2ip(gl);
2118 if (ip && !atomic_read(&ip->i_count))
2119 gfs2_inode_destroy(ip);
2121 if (queue_empty(gl, &gl->gl_holders) &&
2122 gl->gl_state != LM_ST_UNLOCKED &&
2124 handle_callback(gl, LM_ST_UNLOCKED);
2125 gfs2_glmutex_unlock(gl);
2132 * examine_bucket - Call a function for glock in a hash bucket
2133 * @examiner: the function
2134 * @sdp: the filesystem
2135 * @bucket: the bucket
2137 * Returns: 1 if the bucket has entries
2140 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
2141 struct gfs2_gl_hash_bucket *bucket)
2143 struct glock_plug plug;
2144 struct list_head *tmp;
2145 struct gfs2_glock *gl;
2148 /* Add "plug" to end of bucket list, work back up list from there */
2149 memset(&plug.gl_flags, 0, sizeof(unsigned long));
2150 set_bit(GLF_PLUG, &plug.gl_flags);
2152 write_lock(&bucket->hb_lock);
2153 list_add(&plug.gl_list, &bucket->hb_list);
2154 write_unlock(&bucket->hb_lock);
2157 write_lock(&bucket->hb_lock);
2160 tmp = plug.gl_list.next;
2162 if (tmp == &bucket->hb_list) {
2163 list_del(&plug.gl_list);
2164 entries = !list_empty(&bucket->hb_list);
2165 write_unlock(&bucket->hb_lock);
2168 gl = list_entry(tmp, struct gfs2_glock, gl_list);
2170 /* Move plug up list */
2171 list_move(&plug.gl_list, &gl->gl_list);
2173 if (test_bit(GLF_PLUG, &gl->gl_flags))
2176 /* examiner() must glock_put() */
2177 gfs2_glock_hold(gl);
2182 write_unlock(&bucket->hb_lock);
2189 * scan_glock - look at a glock and see if we can reclaim it
2190 * @gl: the glock to look at
2194 static void scan_glock(struct gfs2_glock *gl)
2196 if (gfs2_glmutex_trylock(gl)) {
2197 if (gl->gl_ops == &gfs2_inode_glops) {
2198 struct gfs2_inode *ip = get_gl2ip(gl);
2199 if (ip && !atomic_read(&ip->i_count))
2202 if (queue_empty(gl, &gl->gl_holders) &&
2203 gl->gl_state != LM_ST_UNLOCKED &&
2207 gfs2_glmutex_unlock(gl);
2215 gfs2_glmutex_unlock(gl);
2216 gfs2_glock_schedule_for_reclaim(gl);
2221 * gfs2_scand_internal - Look for glocks and inodes to toss from memory
2222 * @sdp: the filesystem
2226 void gfs2_scand_internal(struct gfs2_sbd *sdp)
2230 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
2231 examine_bucket(scan_glock, sdp, &sdp->sd_gl_hash[x]);
2237 * clear_glock - look at a glock and see if we can free it from glock cache
2238 * @gl: the glock to look at
2242 static void clear_glock(struct gfs2_glock *gl)
2244 struct gfs2_sbd *sdp = gl->gl_sbd;
2247 spin_lock(&sdp->sd_reclaim_lock);
2248 if (!list_empty(&gl->gl_reclaim)) {
2249 list_del_init(&gl->gl_reclaim);
2250 atomic_dec(&sdp->sd_reclaim_count);
2251 released = gfs2_glock_put(gl);
2252 gfs2_assert(sdp, !released);
2254 spin_unlock(&sdp->sd_reclaim_lock);
2256 if (gfs2_glmutex_trylock(gl)) {
2257 if (gl->gl_ops == &gfs2_inode_glops) {
2258 struct gfs2_inode *ip = get_gl2ip(gl);
2259 if (ip && !atomic_read(&ip->i_count))
2260 gfs2_inode_destroy(ip);
2262 if (queue_empty(gl, &gl->gl_holders) &&
2263 gl->gl_state != LM_ST_UNLOCKED)
2264 handle_callback(gl, LM_ST_UNLOCKED);
2266 gfs2_glmutex_unlock(gl);
2273 * gfs2_gl_hash_clear - Empty out the glock hash table
2274 * @sdp: the filesystem
2275 * @wait: wait until it's all gone
2277 * Called when unmounting the filesystem, or when inter-node lock manager
2278 * requests DROPLOCKS because it is running out of capacity.
2281 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp, int wait)
2292 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
2293 if (examine_bucket(clear_glock, sdp,
2294 &sdp->sd_gl_hash[x]))
2300 if (time_after_eq(jiffies,
2301 t + gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
2302 fs_warn(sdp, "Unmount seems to be stalled. "
2303 "Dumping lock state...\n");
2304 gfs2_dump_lockstate(sdp);
2308 /* invalidate_inodes() requires that the sb inodes list
2309 not change, but an async completion callback for an
2310 unlock can occur which does glock_put() which
2311 can call iput() which will change the sb inodes list.
2312 invalidate_inodes_mutex prevents glock_put()'s during
2313 an invalidate_inodes() */
2315 mutex_lock(&sdp->sd_invalidate_inodes_mutex);
2316 invalidate_inodes(sdp->sd_vfs);
2317 mutex_unlock(&sdp->sd_invalidate_inodes_mutex);
2323 * Diagnostic routines to help debug distributed deadlock
2327 * dump_holder - print information about a glock holder
2328 * @str: a string naming the type of holder
2329 * @gh: the glock holder
2331 * Returns: 0 on success, -ENOBUFS when we run out of space
2334 static int dump_holder(char *str, struct gfs2_holder *gh)
2337 int error = -ENOBUFS;
2339 printk(" %s\n", str);
2340 printk(" owner = %ld\n",
2341 (gh->gh_owner) ? (long)gh->gh_owner->pid : -1);
2342 printk(" gh_state = %u\n", gh->gh_state);
2343 printk(" gh_flags =");
2344 for (x = 0; x < 32; x++)
2345 if (gh->gh_flags & (1 << x))
2348 printk(" error = %d\n", gh->gh_error);
2349 printk(" gh_iflags =");
2350 for (x = 0; x < 32; x++)
2351 if (test_bit(x, &gh->gh_iflags))
2361 * dump_inode - print information about an inode
2364 * Returns: 0 on success, -ENOBUFS when we run out of space
2367 static int dump_inode(struct gfs2_inode *ip)
2370 int error = -ENOBUFS;
2372 printk(" Inode:\n");
2373 printk(" num = %llu %llu\n",
2374 ip->i_num.no_formal_ino, ip->i_num.no_addr);
2375 printk(" type = %u\n", IF2DT(ip->i_di.di_mode));
2376 printk(" i_count = %d\n", atomic_read(&ip->i_count));
2377 printk(" i_flags =");
2378 for (x = 0; x < 32; x++)
2379 if (test_bit(x, &ip->i_flags))
2382 printk(" vnode = %s\n", (ip->i_vnode) ? "yes" : "no");
2390 * dump_glock - print information about a glock
2392 * @count: where we are in the buffer
2394 * Returns: 0 on success, -ENOBUFS when we run out of space
2397 static int dump_glock(struct gfs2_glock *gl)
2399 struct gfs2_holder *gh;
2401 int error = -ENOBUFS;
2403 spin_lock(&gl->gl_spin);
2405 printk("Glock (%u, %llu)\n",
2406 gl->gl_name.ln_type,
2407 gl->gl_name.ln_number);
2408 printk(" gl_flags =");
2409 for (x = 0; x < 32; x++)
2410 if (test_bit(x, &gl->gl_flags))
2413 printk(" gl_ref = %d\n", atomic_read(&gl->gl_ref.refcount));
2414 printk(" gl_state = %u\n", gl->gl_state);
2415 printk(" req_gh = %s\n", (gl->gl_req_gh) ? "yes" : "no");
2416 printk(" req_bh = %s\n", (gl->gl_req_bh) ? "yes" : "no");
2417 printk(" lvb_count = %d\n", atomic_read(&gl->gl_lvb_count));
2418 printk(" object = %s\n", (gl->gl_object) ? "yes" : "no");
2419 printk(" le = %s\n",
2420 (list_empty(&gl->gl_le.le_list)) ? "no" : "yes");
2421 printk(" reclaim = %s\n",
2422 (list_empty(&gl->gl_reclaim)) ? "no" : "yes");
2424 printk(" aspace = %lu\n",
2425 gl->gl_aspace->i_mapping->nrpages);
2427 printk(" aspace = no\n");
2428 printk(" ail = %d\n", atomic_read(&gl->gl_ail_count));
2429 if (gl->gl_req_gh) {
2430 error = dump_holder("Request", gl->gl_req_gh);
2434 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
2435 error = dump_holder("Holder", gh);
2439 list_for_each_entry(gh, &gl->gl_waiters1, gh_list) {
2440 error = dump_holder("Waiter1", gh);
2444 list_for_each_entry(gh, &gl->gl_waiters2, gh_list) {
2445 error = dump_holder("Waiter2", gh);
2449 list_for_each_entry(gh, &gl->gl_waiters3, gh_list) {
2450 error = dump_holder("Waiter3", gh);
2454 if (gl->gl_ops == &gfs2_inode_glops && get_gl2ip(gl)) {
2455 if (!test_bit(GLF_LOCK, &gl->gl_flags) &&
2456 list_empty(&gl->gl_holders)) {
2457 error = dump_inode(get_gl2ip(gl));
2462 printk(" Inode: busy\n");
2469 spin_unlock(&gl->gl_spin);
2475 * gfs2_dump_lockstate - print out the current lockstate
2476 * @sdp: the filesystem
2477 * @ub: the buffer to copy the information into
2479 * If @ub is NULL, dump the lockstate to the console.
2483 int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
2485 struct gfs2_gl_hash_bucket *bucket;
2486 struct gfs2_glock *gl;
2490 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
2491 bucket = &sdp->sd_gl_hash[x];
2493 read_lock(&bucket->hb_lock);
2495 list_for_each_entry(gl, &bucket->hb_list, gl_list) {
2496 if (test_bit(GLF_PLUG, &gl->gl_flags))
2499 error = dump_glock(gl);
2504 read_unlock(&bucket->hb_lock);