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/kref.h>
19 #include <linux/kallsyms.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <asm/uaccess.h>
24 #include "lm_interface.h"
36 /* Must be kept in sync with the beginning of struct gfs2_glock */
38 struct list_head gl_list;
39 unsigned long gl_flags;
43 struct gfs2_holder gr_gh;
44 struct work_struct gr_work;
47 typedef void (*glock_examiner) (struct gfs2_glock * gl);
49 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp);
50 static int dump_glock(struct gfs2_glock *gl);
52 static struct gfs2_gl_hash_bucket gl_hash_table[GFS2_GL_HASH_SIZE];
55 * relaxed_state_ok - is a requested lock compatible with the current lock mode?
56 * @actual: the current state of the lock
57 * @requested: the lock state that was requested by the caller
58 * @flags: the modifier flags passed in by the caller
60 * Returns: 1 if the locks are compatible, 0 otherwise
63 static inline int relaxed_state_ok(unsigned int actual, unsigned requested,
66 if (actual == requested)
72 if (actual == LM_ST_EXCLUSIVE && requested == LM_ST_SHARED)
75 if (actual != LM_ST_UNLOCKED && (flags & LM_FLAG_ANY))
82 * gl_hash() - Turn glock number into hash bucket number
83 * @lock: The glock number
85 * Returns: The number of the corresponding hash bucket
88 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
89 const struct lm_lockname *name)
93 h = jhash(&name->ln_number, sizeof(u64), 0);
94 h = jhash(&name->ln_type, sizeof(unsigned int), h);
95 h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
96 h &= GFS2_GL_HASH_MASK;
102 * glock_free() - Perform a few checks and then release struct gfs2_glock
103 * @gl: The glock to release
105 * Also calls lock module to release its internal structure for this glock.
109 static void glock_free(struct gfs2_glock *gl)
111 struct gfs2_sbd *sdp = gl->gl_sbd;
112 struct inode *aspace = gl->gl_aspace;
114 gfs2_lm_put_lock(sdp, gl->gl_lock);
117 gfs2_aspace_put(aspace);
119 kmem_cache_free(gfs2_glock_cachep, gl);
123 * gfs2_glock_hold() - increment reference count on glock
124 * @gl: The glock to hold
128 void gfs2_glock_hold(struct gfs2_glock *gl)
130 kref_get(&gl->gl_ref);
133 /* All work is done after the return from kref_put() so we
134 can release the write_lock before the free. */
136 static void kill_glock(struct kref *kref)
138 struct gfs2_glock *gl = container_of(kref, struct gfs2_glock, gl_ref);
139 struct gfs2_sbd *sdp = gl->gl_sbd;
141 gfs2_assert(sdp, gl->gl_state == LM_ST_UNLOCKED);
142 gfs2_assert(sdp, list_empty(&gl->gl_reclaim));
143 gfs2_assert(sdp, list_empty(&gl->gl_holders));
144 gfs2_assert(sdp, list_empty(&gl->gl_waiters1));
145 gfs2_assert(sdp, list_empty(&gl->gl_waiters2));
146 gfs2_assert(sdp, list_empty(&gl->gl_waiters3));
150 * gfs2_glock_put() - Decrement reference count on glock
151 * @gl: The glock to put
155 int gfs2_glock_put(struct gfs2_glock *gl)
157 struct gfs2_gl_hash_bucket *bucket = gl->gl_bucket;
160 write_lock(&bucket->hb_lock);
161 if (kref_put(&gl->gl_ref, kill_glock)) {
162 list_del_init(&gl->gl_list);
163 write_unlock(&bucket->hb_lock);
164 BUG_ON(spin_is_locked(&gl->gl_spin));
169 write_unlock(&bucket->hb_lock);
175 * queue_empty - check to see if a glock's queue is empty
177 * @head: the head of the queue to check
179 * This function protects the list in the event that a process already
180 * has a holder on the list and is adding a second holder for itself.
181 * The glmutex lock is what generally prevents processes from working
182 * on the same glock at once, but the special case of adding a second
183 * holder for yourself ("recursive" locking) doesn't involve locking
184 * glmutex, making the spin lock necessary.
186 * Returns: 1 if the queue is empty
189 static inline int queue_empty(struct gfs2_glock *gl, struct list_head *head)
192 spin_lock(&gl->gl_spin);
193 empty = list_empty(head);
194 spin_unlock(&gl->gl_spin);
199 * search_bucket() - Find struct gfs2_glock by lock number
200 * @bucket: the bucket to search
201 * @name: The lock name
203 * Returns: NULL, or the struct gfs2_glock with the requested number
206 static struct gfs2_glock *search_bucket(struct gfs2_gl_hash_bucket *bucket,
207 const struct gfs2_sbd *sdp,
208 const struct lm_lockname *name)
210 struct gfs2_glock *gl;
212 list_for_each_entry(gl, &bucket->hb_list, gl_list) {
213 if (test_bit(GLF_PLUG, &gl->gl_flags))
215 if (!lm_name_equal(&gl->gl_name, name))
217 if (gl->gl_sbd != sdp)
220 kref_get(&gl->gl_ref);
229 * gfs2_glock_find() - Find glock by lock number
230 * @sdp: The GFS2 superblock
231 * @name: The lock name
233 * Returns: NULL, or the struct gfs2_glock with the requested number
236 static struct gfs2_glock *gfs2_glock_find(const struct gfs2_sbd *sdp,
237 const struct lm_lockname *name)
239 struct gfs2_gl_hash_bucket *bucket = &gl_hash_table[gl_hash(sdp, name)];
240 struct gfs2_glock *gl;
242 read_lock(&bucket->hb_lock);
243 gl = search_bucket(bucket, sdp, name);
244 read_unlock(&bucket->hb_lock);
250 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
251 * @sdp: The GFS2 superblock
252 * @number: the lock number
253 * @glops: The glock_operations to use
254 * @create: If 0, don't create the glock if it doesn't exist
255 * @glp: the glock is returned here
257 * This does not lock a glock, just finds/creates structures for one.
262 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
263 const struct gfs2_glock_operations *glops, int create,
264 struct gfs2_glock **glp)
266 struct lm_lockname name;
267 struct gfs2_glock *gl, *tmp;
268 struct gfs2_gl_hash_bucket *bucket;
271 name.ln_number = number;
272 name.ln_type = glops->go_type;
273 bucket = &gl_hash_table[gl_hash(sdp, &name)];
275 read_lock(&bucket->hb_lock);
276 gl = search_bucket(bucket, sdp, &name);
277 read_unlock(&bucket->hb_lock);
284 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
290 kref_init(&gl->gl_ref);
291 gl->gl_state = LM_ST_UNLOCKED;
295 gl->gl_req_gh = NULL;
296 gl->gl_req_bh = NULL;
298 gl->gl_stamp = jiffies;
299 gl->gl_object = NULL;
300 gl->gl_bucket = bucket;
302 gl->gl_aspace = NULL;
303 lops_init_le(&gl->gl_le, &gfs2_glock_lops);
305 /* If this glock protects actual on-disk data or metadata blocks,
306 create a VFS inode to manage the pages/buffers holding them. */
307 if (glops == &gfs2_inode_glops || glops == &gfs2_rgrp_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 write_lock(&bucket->hb_lock);
320 tmp = search_bucket(bucket, sdp, &name);
322 write_unlock(&bucket->hb_lock);
326 list_add_tail(&gl->gl_list, &bucket->hb_list);
327 write_unlock(&bucket->hb_lock);
336 gfs2_aspace_put(gl->gl_aspace);
338 kmem_cache_free(gfs2_glock_cachep, gl);
343 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
345 * @state: the state we're requesting
346 * @flags: the modifier flags
347 * @gh: the holder structure
351 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
352 struct gfs2_holder *gh)
354 INIT_LIST_HEAD(&gh->gh_list);
356 gh->gh_ip = (unsigned long)__builtin_return_address(0);
357 gh->gh_owner = current;
358 gh->gh_state = state;
359 gh->gh_flags = flags;
362 init_completion(&gh->gh_wait);
364 if (gh->gh_state == LM_ST_EXCLUSIVE)
365 gh->gh_flags |= GL_LOCAL_EXCL;
371 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
372 * @state: the state we're requesting
373 * @flags: the modifier flags
374 * @gh: the holder structure
376 * Don't mess with the glock.
380 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
382 gh->gh_state = state;
383 gh->gh_flags = flags;
384 if (gh->gh_state == LM_ST_EXCLUSIVE)
385 gh->gh_flags |= GL_LOCAL_EXCL;
387 gh->gh_iflags &= 1 << HIF_ALLOCED;
388 gh->gh_ip = (unsigned long)__builtin_return_address(0);
392 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
393 * @gh: the holder structure
397 void gfs2_holder_uninit(struct gfs2_holder *gh)
399 gfs2_glock_put(gh->gh_gl);
405 * gfs2_holder_get - get a struct gfs2_holder structure
407 * @state: the state we're requesting
408 * @flags: the modifier flags
411 * Figure out how big an impact this function has. Either:
412 * 1) Replace it with a cache of structures hanging off the struct gfs2_sbd
413 * 2) Leave it like it is
415 * Returns: the holder structure, NULL on ENOMEM
418 static struct gfs2_holder *gfs2_holder_get(struct gfs2_glock *gl,
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);
430 gh->gh_ip = (unsigned long)__builtin_return_address(0);
435 * gfs2_holder_put - get rid of a struct gfs2_holder structure
436 * @gh: the holder structure
440 static void gfs2_holder_put(struct gfs2_holder *gh)
442 gfs2_holder_uninit(gh);
447 * rq_mutex - process a mutex request in the queue
448 * @gh: the glock holder
450 * Returns: 1 if the queue is blocked
453 static int rq_mutex(struct gfs2_holder *gh)
455 struct gfs2_glock *gl = gh->gh_gl;
457 list_del_init(&gh->gh_list);
458 /* gh->gh_error never examined. */
459 set_bit(GLF_LOCK, &gl->gl_flags);
460 complete(&gh->gh_wait);
466 * rq_promote - process a promote request in the queue
467 * @gh: the glock holder
469 * Acquire a new inter-node lock, or change a lock state to more restrictive.
471 * Returns: 1 if the queue is blocked
474 static int rq_promote(struct gfs2_holder *gh)
476 struct gfs2_glock *gl = gh->gh_gl;
477 struct gfs2_sbd *sdp = gl->gl_sbd;
478 const struct gfs2_glock_operations *glops = gl->gl_ops;
480 if (!relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
481 if (list_empty(&gl->gl_holders)) {
483 set_bit(GLF_LOCK, &gl->gl_flags);
484 spin_unlock(&gl->gl_spin);
486 if (atomic_read(&sdp->sd_reclaim_count) >
487 gfs2_tune_get(sdp, gt_reclaim_limit) &&
488 !(gh->gh_flags & LM_FLAG_PRIORITY)) {
489 gfs2_reclaim_glock(sdp);
490 gfs2_reclaim_glock(sdp);
493 glops->go_xmote_th(gl, gh->gh_state, gh->gh_flags);
494 spin_lock(&gl->gl_spin);
499 if (list_empty(&gl->gl_holders)) {
500 set_bit(HIF_FIRST, &gh->gh_iflags);
501 set_bit(GLF_LOCK, &gl->gl_flags);
503 struct gfs2_holder *next_gh;
504 if (gh->gh_flags & GL_LOCAL_EXCL)
506 next_gh = list_entry(gl->gl_holders.next, struct gfs2_holder,
508 if (next_gh->gh_flags & GL_LOCAL_EXCL)
512 list_move_tail(&gh->gh_list, &gl->gl_holders);
514 set_bit(HIF_HOLDER, &gh->gh_iflags);
516 complete(&gh->gh_wait);
522 * rq_demote - process a demote request in the queue
523 * @gh: the glock holder
525 * Returns: 1 if the queue is blocked
528 static int rq_demote(struct gfs2_holder *gh)
530 struct gfs2_glock *gl = gh->gh_gl;
531 const struct gfs2_glock_operations *glops = gl->gl_ops;
533 if (!list_empty(&gl->gl_holders))
536 if (gl->gl_state == gh->gh_state || gl->gl_state == LM_ST_UNLOCKED) {
537 list_del_init(&gh->gh_list);
539 spin_unlock(&gl->gl_spin);
540 if (test_bit(HIF_DEALLOC, &gh->gh_iflags))
543 complete(&gh->gh_wait);
544 spin_lock(&gl->gl_spin);
547 set_bit(GLF_LOCK, &gl->gl_flags);
548 spin_unlock(&gl->gl_spin);
550 if (gh->gh_state == LM_ST_UNLOCKED ||
551 gl->gl_state != LM_ST_EXCLUSIVE)
552 glops->go_drop_th(gl);
554 glops->go_xmote_th(gl, gh->gh_state, gh->gh_flags);
556 spin_lock(&gl->gl_spin);
563 * rq_greedy - process a queued request to drop greedy status
564 * @gh: the glock holder
566 * Returns: 1 if the queue is blocked
569 static int rq_greedy(struct gfs2_holder *gh)
571 struct gfs2_glock *gl = gh->gh_gl;
573 list_del_init(&gh->gh_list);
574 /* gh->gh_error never examined. */
575 clear_bit(GLF_GREEDY, &gl->gl_flags);
576 spin_unlock(&gl->gl_spin);
578 gfs2_holder_uninit(gh);
579 kfree(container_of(gh, struct greedy, gr_gh));
581 spin_lock(&gl->gl_spin);
587 * run_queue - process holder structures on a glock
591 static void run_queue(struct gfs2_glock *gl)
593 struct gfs2_holder *gh;
597 if (test_bit(GLF_LOCK, &gl->gl_flags))
600 if (!list_empty(&gl->gl_waiters1)) {
601 gh = list_entry(gl->gl_waiters1.next,
602 struct gfs2_holder, gh_list);
604 if (test_bit(HIF_MUTEX, &gh->gh_iflags))
605 blocked = rq_mutex(gh);
607 gfs2_assert_warn(gl->gl_sbd, 0);
609 } else if (!list_empty(&gl->gl_waiters2) &&
610 !test_bit(GLF_SKIP_WAITERS2, &gl->gl_flags)) {
611 gh = list_entry(gl->gl_waiters2.next,
612 struct gfs2_holder, gh_list);
614 if (test_bit(HIF_DEMOTE, &gh->gh_iflags))
615 blocked = rq_demote(gh);
616 else if (test_bit(HIF_GREEDY, &gh->gh_iflags))
617 blocked = rq_greedy(gh);
619 gfs2_assert_warn(gl->gl_sbd, 0);
621 } else if (!list_empty(&gl->gl_waiters3)) {
622 gh = list_entry(gl->gl_waiters3.next,
623 struct gfs2_holder, gh_list);
625 if (test_bit(HIF_PROMOTE, &gh->gh_iflags))
626 blocked = rq_promote(gh);
628 gfs2_assert_warn(gl->gl_sbd, 0);
639 * gfs2_glmutex_lock - acquire a local lock on a glock
642 * Gives caller exclusive access to manipulate a glock structure.
645 static void gfs2_glmutex_lock(struct gfs2_glock *gl)
647 struct gfs2_holder gh;
649 gfs2_holder_init(gl, 0, 0, &gh);
650 set_bit(HIF_MUTEX, &gh.gh_iflags);
652 spin_lock(&gl->gl_spin);
653 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
654 list_add_tail(&gh.gh_list, &gl->gl_waiters1);
656 gl->gl_owner = current;
657 gl->gl_ip = (unsigned long)__builtin_return_address(0);
658 complete(&gh.gh_wait);
660 spin_unlock(&gl->gl_spin);
662 wait_for_completion(&gh.gh_wait);
663 gfs2_holder_uninit(&gh);
667 * gfs2_glmutex_trylock - try to acquire a local lock on a glock
670 * Returns: 1 if the glock is acquired
673 static int gfs2_glmutex_trylock(struct gfs2_glock *gl)
677 spin_lock(&gl->gl_spin);
678 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
681 gl->gl_owner = current;
682 gl->gl_ip = (unsigned long)__builtin_return_address(0);
684 spin_unlock(&gl->gl_spin);
690 * gfs2_glmutex_unlock - release a local lock on a glock
695 static void gfs2_glmutex_unlock(struct gfs2_glock *gl)
697 spin_lock(&gl->gl_spin);
698 clear_bit(GLF_LOCK, &gl->gl_flags);
702 BUG_ON(!spin_is_locked(&gl->gl_spin));
703 spin_unlock(&gl->gl_spin);
707 * handle_callback - add a demote request to a lock's queue
709 * @state: the state the caller wants us to change to
711 * Note: This may fail sliently if we are out of memory.
714 static void handle_callback(struct gfs2_glock *gl, unsigned int state)
716 struct gfs2_holder *gh, *new_gh = NULL;
719 spin_lock(&gl->gl_spin);
721 list_for_each_entry(gh, &gl->gl_waiters2, gh_list) {
722 if (test_bit(HIF_DEMOTE, &gh->gh_iflags) &&
723 gl->gl_req_gh != gh) {
724 if (gh->gh_state != state)
725 gh->gh_state = LM_ST_UNLOCKED;
731 list_add_tail(&new_gh->gh_list, &gl->gl_waiters2);
734 spin_unlock(&gl->gl_spin);
736 new_gh = gfs2_holder_get(gl, state, LM_FLAG_TRY, GFP_KERNEL);
739 set_bit(HIF_DEMOTE, &new_gh->gh_iflags);
740 set_bit(HIF_DEALLOC, &new_gh->gh_iflags);
746 spin_unlock(&gl->gl_spin);
749 gfs2_holder_put(new_gh);
752 void gfs2_glock_inode_squish(struct inode *inode)
754 struct gfs2_holder gh;
755 struct gfs2_glock *gl = GFS2_I(inode)->i_gl;
756 gfs2_holder_init(gl, LM_ST_UNLOCKED, 0, &gh);
757 set_bit(HIF_DEMOTE, &gh.gh_iflags);
758 spin_lock(&gl->gl_spin);
759 gfs2_assert(inode->i_sb->s_fs_info, list_empty(&gl->gl_holders));
760 list_add_tail(&gh.gh_list, &gl->gl_waiters2);
762 spin_unlock(&gl->gl_spin);
763 wait_for_completion(&gh.gh_wait);
764 gfs2_holder_uninit(&gh);
768 * state_change - record that the glock is now in a different state
770 * @new_state the new state
774 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
778 held1 = (gl->gl_state != LM_ST_UNLOCKED);
779 held2 = (new_state != LM_ST_UNLOCKED);
781 if (held1 != held2) {
788 gl->gl_state = new_state;
792 * xmote_bh - Called after the lock module is done acquiring a lock
793 * @gl: The glock in question
794 * @ret: the int returned from the lock module
798 static void xmote_bh(struct gfs2_glock *gl, unsigned int ret)
800 struct gfs2_sbd *sdp = gl->gl_sbd;
801 const struct gfs2_glock_operations *glops = gl->gl_ops;
802 struct gfs2_holder *gh = gl->gl_req_gh;
803 int prev_state = gl->gl_state;
806 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
807 gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
808 gfs2_assert_warn(sdp, !(ret & LM_OUT_ASYNC));
810 state_change(gl, ret & LM_OUT_ST_MASK);
812 if (prev_state != LM_ST_UNLOCKED && !(ret & LM_OUT_CACHEABLE)) {
814 glops->go_inval(gl, DIO_METADATA | DIO_DATA);
815 } else if (gl->gl_state == LM_ST_DEFERRED) {
816 /* We might not want to do this here.
817 Look at moving to the inode glops. */
819 glops->go_inval(gl, DIO_DATA);
822 /* Deal with each possible exit condition */
825 gl->gl_stamp = jiffies;
826 else if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
827 spin_lock(&gl->gl_spin);
828 list_del_init(&gh->gh_list);
830 spin_unlock(&gl->gl_spin);
831 } else if (test_bit(HIF_DEMOTE, &gh->gh_iflags)) {
832 spin_lock(&gl->gl_spin);
833 list_del_init(&gh->gh_list);
834 if (gl->gl_state == gh->gh_state ||
835 gl->gl_state == LM_ST_UNLOCKED) {
838 if (gfs2_assert_warn(sdp, gh->gh_flags &
839 (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) == -1)
840 fs_warn(sdp, "ret = 0x%.8X\n", ret);
841 gh->gh_error = GLR_TRYFAILED;
843 spin_unlock(&gl->gl_spin);
845 if (ret & LM_OUT_CANCELED)
846 handle_callback(gl, LM_ST_UNLOCKED);
848 } else if (ret & LM_OUT_CANCELED) {
849 spin_lock(&gl->gl_spin);
850 list_del_init(&gh->gh_list);
851 gh->gh_error = GLR_CANCELED;
852 spin_unlock(&gl->gl_spin);
854 } else if (relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
855 spin_lock(&gl->gl_spin);
856 list_move_tail(&gh->gh_list, &gl->gl_holders);
858 set_bit(HIF_HOLDER, &gh->gh_iflags);
859 spin_unlock(&gl->gl_spin);
861 set_bit(HIF_FIRST, &gh->gh_iflags);
865 } else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
866 spin_lock(&gl->gl_spin);
867 list_del_init(&gh->gh_list);
868 gh->gh_error = GLR_TRYFAILED;
869 spin_unlock(&gl->gl_spin);
872 if (gfs2_assert_withdraw(sdp, 0) == -1)
873 fs_err(sdp, "ret = 0x%.8X\n", ret);
876 if (glops->go_xmote_bh)
877 glops->go_xmote_bh(gl);
880 spin_lock(&gl->gl_spin);
881 gl->gl_req_gh = NULL;
882 gl->gl_req_bh = NULL;
883 clear_bit(GLF_LOCK, &gl->gl_flags);
885 spin_unlock(&gl->gl_spin);
891 if (test_bit(HIF_DEALLOC, &gh->gh_iflags))
894 complete(&gh->gh_wait);
899 * gfs2_glock_xmote_th - Call into the lock module to acquire or change a glock
900 * @gl: The glock in question
901 * @state: the requested state
902 * @flags: modifier flags to the lock call
906 void gfs2_glock_xmote_th(struct gfs2_glock *gl, unsigned int state, int flags)
908 struct gfs2_sbd *sdp = gl->gl_sbd;
909 const struct gfs2_glock_operations *glops = gl->gl_ops;
910 int lck_flags = flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB |
911 LM_FLAG_NOEXP | LM_FLAG_ANY |
913 unsigned int lck_ret;
915 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
916 gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
917 gfs2_assert_warn(sdp, state != LM_ST_UNLOCKED);
918 gfs2_assert_warn(sdp, state != gl->gl_state);
920 if (gl->gl_state == LM_ST_EXCLUSIVE && glops->go_sync)
921 glops->go_sync(gl, DIO_METADATA | DIO_DATA | DIO_RELEASE);
924 gl->gl_req_bh = xmote_bh;
926 lck_ret = gfs2_lm_lock(sdp, gl->gl_lock, gl->gl_state, state, lck_flags);
928 if (gfs2_assert_withdraw(sdp, !(lck_ret & LM_OUT_ERROR)))
931 if (lck_ret & LM_OUT_ASYNC)
932 gfs2_assert_warn(sdp, lck_ret == LM_OUT_ASYNC);
934 xmote_bh(gl, lck_ret);
938 * drop_bh - Called after a lock module unlock completes
940 * @ret: the return status
942 * Doesn't wake up the process waiting on the struct gfs2_holder (if any)
943 * Doesn't drop the reference on the glock the top half took out
947 static void drop_bh(struct gfs2_glock *gl, unsigned int ret)
949 struct gfs2_sbd *sdp = gl->gl_sbd;
950 const struct gfs2_glock_operations *glops = gl->gl_ops;
951 struct gfs2_holder *gh = gl->gl_req_gh;
953 clear_bit(GLF_PREFETCH, &gl->gl_flags);
955 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
956 gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
957 gfs2_assert_warn(sdp, !ret);
959 state_change(gl, LM_ST_UNLOCKED);
962 glops->go_inval(gl, DIO_METADATA | DIO_DATA);
965 spin_lock(&gl->gl_spin);
966 list_del_init(&gh->gh_list);
968 spin_unlock(&gl->gl_spin);
971 if (glops->go_drop_bh)
972 glops->go_drop_bh(gl);
974 spin_lock(&gl->gl_spin);
975 gl->gl_req_gh = NULL;
976 gl->gl_req_bh = NULL;
977 clear_bit(GLF_LOCK, &gl->gl_flags);
979 spin_unlock(&gl->gl_spin);
984 if (test_bit(HIF_DEALLOC, &gh->gh_iflags))
987 complete(&gh->gh_wait);
992 * gfs2_glock_drop_th - call into the lock module to unlock a lock
997 void gfs2_glock_drop_th(struct gfs2_glock *gl)
999 struct gfs2_sbd *sdp = gl->gl_sbd;
1000 const struct gfs2_glock_operations *glops = gl->gl_ops;
1003 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1004 gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
1005 gfs2_assert_warn(sdp, gl->gl_state != LM_ST_UNLOCKED);
1007 if (gl->gl_state == LM_ST_EXCLUSIVE && glops->go_sync)
1008 glops->go_sync(gl, DIO_METADATA | DIO_DATA | DIO_RELEASE);
1010 gfs2_glock_hold(gl);
1011 gl->gl_req_bh = drop_bh;
1013 ret = gfs2_lm_unlock(sdp, gl->gl_lock, gl->gl_state);
1015 if (gfs2_assert_withdraw(sdp, !(ret & LM_OUT_ERROR)))
1021 gfs2_assert_warn(sdp, ret == LM_OUT_ASYNC);
1025 * do_cancels - cancel requests for locks stuck waiting on an expire flag
1026 * @gh: the LM_FLAG_PRIORITY holder waiting to acquire the lock
1028 * Don't cancel GL_NOCANCEL requests.
1031 static void do_cancels(struct gfs2_holder *gh)
1033 struct gfs2_glock *gl = gh->gh_gl;
1035 spin_lock(&gl->gl_spin);
1037 while (gl->gl_req_gh != gh &&
1038 !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1039 !list_empty(&gh->gh_list)) {
1040 if (gl->gl_req_bh && !(gl->gl_req_gh &&
1041 (gl->gl_req_gh->gh_flags & GL_NOCANCEL))) {
1042 spin_unlock(&gl->gl_spin);
1043 gfs2_lm_cancel(gl->gl_sbd, gl->gl_lock);
1045 spin_lock(&gl->gl_spin);
1047 spin_unlock(&gl->gl_spin);
1049 spin_lock(&gl->gl_spin);
1053 spin_unlock(&gl->gl_spin);
1057 * glock_wait_internal - wait on a glock acquisition
1058 * @gh: the glock holder
1060 * Returns: 0 on success
1063 static int glock_wait_internal(struct gfs2_holder *gh)
1065 struct gfs2_glock *gl = gh->gh_gl;
1066 struct gfs2_sbd *sdp = gl->gl_sbd;
1067 const struct gfs2_glock_operations *glops = gl->gl_ops;
1069 if (test_bit(HIF_ABORTED, &gh->gh_iflags))
1072 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1073 spin_lock(&gl->gl_spin);
1074 if (gl->gl_req_gh != gh &&
1075 !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1076 !list_empty(&gh->gh_list)) {
1077 list_del_init(&gh->gh_list);
1078 gh->gh_error = GLR_TRYFAILED;
1080 spin_unlock(&gl->gl_spin);
1081 return gh->gh_error;
1083 spin_unlock(&gl->gl_spin);
1086 if (gh->gh_flags & LM_FLAG_PRIORITY)
1089 wait_for_completion(&gh->gh_wait);
1092 return gh->gh_error;
1094 gfs2_assert_withdraw(sdp, test_bit(HIF_HOLDER, &gh->gh_iflags));
1095 gfs2_assert_withdraw(sdp, relaxed_state_ok(gl->gl_state, gh->gh_state,
1098 if (test_bit(HIF_FIRST, &gh->gh_iflags)) {
1099 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1101 if (glops->go_lock) {
1102 gh->gh_error = glops->go_lock(gh);
1104 spin_lock(&gl->gl_spin);
1105 list_del_init(&gh->gh_list);
1106 spin_unlock(&gl->gl_spin);
1110 spin_lock(&gl->gl_spin);
1111 gl->gl_req_gh = NULL;
1112 gl->gl_req_bh = NULL;
1113 clear_bit(GLF_LOCK, &gl->gl_flags);
1115 spin_unlock(&gl->gl_spin);
1118 return gh->gh_error;
1121 static inline struct gfs2_holder *
1122 find_holder_by_owner(struct list_head *head, struct task_struct *owner)
1124 struct gfs2_holder *gh;
1126 list_for_each_entry(gh, head, gh_list) {
1127 if (gh->gh_owner == owner)
1135 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1136 * @gh: the holder structure to add
1140 static void add_to_queue(struct gfs2_holder *gh)
1142 struct gfs2_glock *gl = gh->gh_gl;
1143 struct gfs2_holder *existing;
1145 BUG_ON(!gh->gh_owner);
1147 existing = find_holder_by_owner(&gl->gl_holders, gh->gh_owner);
1149 print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
1150 printk(KERN_INFO "pid : %d\n", existing->gh_owner->pid);
1151 printk(KERN_INFO "lock type : %d lock state : %d\n",
1152 existing->gh_gl->gl_name.ln_type, existing->gh_gl->gl_state);
1153 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1154 printk(KERN_INFO "pid : %d\n", gh->gh_owner->pid);
1155 printk(KERN_INFO "lock type : %d lock state : %d\n",
1156 gl->gl_name.ln_type, gl->gl_state);
1160 existing = find_holder_by_owner(&gl->gl_waiters3, gh->gh_owner);
1162 print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
1163 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1167 if (gh->gh_flags & LM_FLAG_PRIORITY)
1168 list_add(&gh->gh_list, &gl->gl_waiters3);
1170 list_add_tail(&gh->gh_list, &gl->gl_waiters3);
1174 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1175 * @gh: the holder structure
1177 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1179 * Returns: 0, GLR_TRYFAILED, or errno on failure
1182 int gfs2_glock_nq(struct gfs2_holder *gh)
1184 struct gfs2_glock *gl = gh->gh_gl;
1185 struct gfs2_sbd *sdp = gl->gl_sbd;
1189 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1190 set_bit(HIF_ABORTED, &gh->gh_iflags);
1194 set_bit(HIF_PROMOTE, &gh->gh_iflags);
1196 spin_lock(&gl->gl_spin);
1199 spin_unlock(&gl->gl_spin);
1201 if (!(gh->gh_flags & GL_ASYNC)) {
1202 error = glock_wait_internal(gh);
1203 if (error == GLR_CANCELED) {
1209 clear_bit(GLF_PREFETCH, &gl->gl_flags);
1211 if (error == GLR_TRYFAILED && (gh->gh_flags & GL_DUMP))
1218 * gfs2_glock_poll - poll to see if an async request has been completed
1221 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1224 int gfs2_glock_poll(struct gfs2_holder *gh)
1226 struct gfs2_glock *gl = gh->gh_gl;
1229 spin_lock(&gl->gl_spin);
1231 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1233 else if (list_empty(&gh->gh_list)) {
1234 if (gh->gh_error == GLR_CANCELED) {
1235 spin_unlock(&gl->gl_spin);
1237 if (gfs2_glock_nq(gh))
1244 spin_unlock(&gl->gl_spin);
1250 * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
1251 * @gh: the holder structure
1253 * Returns: 0, GLR_TRYFAILED, or errno on failure
1256 int gfs2_glock_wait(struct gfs2_holder *gh)
1260 error = glock_wait_internal(gh);
1261 if (error == GLR_CANCELED) {
1263 gh->gh_flags &= ~GL_ASYNC;
1264 error = gfs2_glock_nq(gh);
1271 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1272 * @gh: the glock holder
1276 void gfs2_glock_dq(struct gfs2_holder *gh)
1278 struct gfs2_glock *gl = gh->gh_gl;
1279 const struct gfs2_glock_operations *glops = gl->gl_ops;
1281 if (gh->gh_flags & GL_NOCACHE)
1282 handle_callback(gl, LM_ST_UNLOCKED);
1284 gfs2_glmutex_lock(gl);
1286 spin_lock(&gl->gl_spin);
1287 list_del_init(&gh->gh_list);
1289 if (list_empty(&gl->gl_holders)) {
1290 spin_unlock(&gl->gl_spin);
1292 if (glops->go_unlock)
1293 glops->go_unlock(gh);
1295 gl->gl_stamp = jiffies;
1297 spin_lock(&gl->gl_spin);
1300 clear_bit(GLF_LOCK, &gl->gl_flags);
1302 spin_unlock(&gl->gl_spin);
1306 * gfs2_glock_prefetch - Try to prefetch a glock
1308 * @state: the state to prefetch in
1309 * @flags: flags passed to go_xmote_th()
1313 static void gfs2_glock_prefetch(struct gfs2_glock *gl, unsigned int state,
1316 const struct gfs2_glock_operations *glops = gl->gl_ops;
1318 spin_lock(&gl->gl_spin);
1320 if (test_bit(GLF_LOCK, &gl->gl_flags) || !list_empty(&gl->gl_holders) ||
1321 !list_empty(&gl->gl_waiters1) || !list_empty(&gl->gl_waiters2) ||
1322 !list_empty(&gl->gl_waiters3) ||
1323 relaxed_state_ok(gl->gl_state, state, flags)) {
1324 spin_unlock(&gl->gl_spin);
1328 set_bit(GLF_PREFETCH, &gl->gl_flags);
1329 set_bit(GLF_LOCK, &gl->gl_flags);
1330 spin_unlock(&gl->gl_spin);
1332 glops->go_xmote_th(gl, state, flags);
1335 static void greedy_work(void *data)
1337 struct greedy *gr = data;
1338 struct gfs2_holder *gh = &gr->gr_gh;
1339 struct gfs2_glock *gl = gh->gh_gl;
1340 const struct gfs2_glock_operations *glops = gl->gl_ops;
1342 clear_bit(GLF_SKIP_WAITERS2, &gl->gl_flags);
1344 if (glops->go_greedy)
1345 glops->go_greedy(gl);
1347 spin_lock(&gl->gl_spin);
1349 if (list_empty(&gl->gl_waiters2)) {
1350 clear_bit(GLF_GREEDY, &gl->gl_flags);
1351 spin_unlock(&gl->gl_spin);
1352 gfs2_holder_uninit(gh);
1355 gfs2_glock_hold(gl);
1356 list_add_tail(&gh->gh_list, &gl->gl_waiters2);
1358 spin_unlock(&gl->gl_spin);
1364 * gfs2_glock_be_greedy -
1368 * Returns: 0 if go_greedy will be called, 1 otherwise
1371 int gfs2_glock_be_greedy(struct gfs2_glock *gl, unsigned int time)
1374 struct gfs2_holder *gh;
1376 if (!time || gl->gl_sbd->sd_args.ar_localcaching ||
1377 test_and_set_bit(GLF_GREEDY, &gl->gl_flags))
1380 gr = kmalloc(sizeof(struct greedy), GFP_KERNEL);
1382 clear_bit(GLF_GREEDY, &gl->gl_flags);
1387 gfs2_holder_init(gl, 0, 0, gh);
1388 set_bit(HIF_GREEDY, &gh->gh_iflags);
1389 INIT_WORK(&gr->gr_work, greedy_work, gr);
1391 set_bit(GLF_SKIP_WAITERS2, &gl->gl_flags);
1392 schedule_delayed_work(&gr->gr_work, time);
1398 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1399 * @gh: the holder structure
1403 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1406 gfs2_holder_uninit(gh);
1410 * gfs2_glock_nq_num - acquire a glock based on lock number
1411 * @sdp: the filesystem
1412 * @number: the lock number
1413 * @glops: the glock operations for the type of glock
1414 * @state: the state to acquire the glock in
1415 * @flags: modifier flags for the aquisition
1416 * @gh: the struct gfs2_holder
1421 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1422 const struct gfs2_glock_operations *glops,
1423 unsigned int state, int flags, struct gfs2_holder *gh)
1425 struct gfs2_glock *gl;
1428 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1430 error = gfs2_glock_nq_init(gl, state, flags, gh);
1438 * glock_compare - Compare two struct gfs2_glock structures for sorting
1439 * @arg_a: the first structure
1440 * @arg_b: the second structure
1444 static int glock_compare(const void *arg_a, const void *arg_b)
1446 struct gfs2_holder *gh_a = *(struct gfs2_holder **)arg_a;
1447 struct gfs2_holder *gh_b = *(struct gfs2_holder **)arg_b;
1448 struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1449 struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1452 if (a->ln_number > b->ln_number)
1454 else if (a->ln_number < b->ln_number)
1457 if (gh_a->gh_state == LM_ST_SHARED &&
1458 gh_b->gh_state == LM_ST_EXCLUSIVE)
1460 else if (!(gh_a->gh_flags & GL_LOCAL_EXCL) &&
1461 (gh_b->gh_flags & GL_LOCAL_EXCL))
1469 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1470 * @num_gh: the number of structures
1471 * @ghs: an array of struct gfs2_holder structures
1473 * Returns: 0 on success (all glocks acquired),
1474 * errno on failure (no glocks acquired)
1477 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1478 struct gfs2_holder **p)
1483 for (x = 0; x < num_gh; x++)
1486 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1488 for (x = 0; x < num_gh; x++) {
1489 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1491 error = gfs2_glock_nq(p[x]);
1494 gfs2_glock_dq(p[x]);
1503 * gfs2_glock_nq_m - acquire multiple glocks
1504 * @num_gh: the number of structures
1505 * @ghs: an array of struct gfs2_holder structures
1507 * Figure out how big an impact this function has. Either:
1508 * 1) Replace this code with code that calls gfs2_glock_prefetch()
1509 * 2) Forget async stuff and just call nq_m_sync()
1510 * 3) Leave it like it is
1512 * Returns: 0 on success (all glocks acquired),
1513 * errno on failure (no glocks acquired)
1516 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1520 int borked = 0, serious = 0;
1527 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1528 return gfs2_glock_nq(ghs);
1531 e = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL);
1535 for (x = 0; x < num_gh; x++) {
1536 ghs[x].gh_flags |= LM_FLAG_TRY | GL_ASYNC;
1537 error = gfs2_glock_nq(&ghs[x]);
1546 for (x = 0; x < num_gh; x++) {
1547 error = e[x] = glock_wait_internal(&ghs[x]);
1550 if (error != GLR_TRYFAILED && error != GLR_CANCELED)
1560 for (x = 0; x < num_gh; x++)
1562 gfs2_glock_dq(&ghs[x]);
1567 for (x = 0; x < num_gh; x++)
1568 gfs2_holder_reinit(ghs[x].gh_state, ghs[x].gh_flags,
1570 error = nq_m_sync(num_gh, ghs, (struct gfs2_holder **)e);
1579 * gfs2_glock_dq_m - release multiple glocks
1580 * @num_gh: the number of structures
1581 * @ghs: an array of struct gfs2_holder structures
1585 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1589 for (x = 0; x < num_gh; x++)
1590 gfs2_glock_dq(&ghs[x]);
1594 * gfs2_glock_dq_uninit_m - release multiple glocks
1595 * @num_gh: the number of structures
1596 * @ghs: an array of struct gfs2_holder structures
1600 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1604 for (x = 0; x < num_gh; x++)
1605 gfs2_glock_dq_uninit(&ghs[x]);
1609 * gfs2_glock_prefetch_num - prefetch a glock based on lock number
1610 * @sdp: the filesystem
1611 * @number: the lock number
1612 * @glops: the glock operations for the type of glock
1613 * @state: the state to acquire the glock in
1614 * @flags: modifier flags for the aquisition
1619 void gfs2_glock_prefetch_num(struct gfs2_sbd *sdp, u64 number,
1620 const struct gfs2_glock_operations *glops,
1621 unsigned int state, int flags)
1623 struct gfs2_glock *gl;
1626 if (atomic_read(&sdp->sd_reclaim_count) <
1627 gfs2_tune_get(sdp, gt_reclaim_limit)) {
1628 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1630 gfs2_glock_prefetch(gl, state, flags);
1637 * gfs2_lvb_hold - attach a LVB from a glock
1638 * @gl: The glock in question
1642 int gfs2_lvb_hold(struct gfs2_glock *gl)
1646 gfs2_glmutex_lock(gl);
1648 if (!atomic_read(&gl->gl_lvb_count)) {
1649 error = gfs2_lm_hold_lvb(gl->gl_sbd, gl->gl_lock, &gl->gl_lvb);
1651 gfs2_glmutex_unlock(gl);
1654 gfs2_glock_hold(gl);
1656 atomic_inc(&gl->gl_lvb_count);
1658 gfs2_glmutex_unlock(gl);
1664 * gfs2_lvb_unhold - detach a LVB from a glock
1665 * @gl: The glock in question
1669 void gfs2_lvb_unhold(struct gfs2_glock *gl)
1671 gfs2_glock_hold(gl);
1672 gfs2_glmutex_lock(gl);
1674 gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count) > 0);
1675 if (atomic_dec_and_test(&gl->gl_lvb_count)) {
1676 gfs2_lm_unhold_lvb(gl->gl_sbd, gl->gl_lock, gl->gl_lvb);
1681 gfs2_glmutex_unlock(gl);
1685 static void blocking_cb(struct gfs2_sbd *sdp, struct lm_lockname *name,
1688 struct gfs2_glock *gl;
1690 gl = gfs2_glock_find(sdp, name);
1694 if (gl->gl_ops->go_callback)
1695 gl->gl_ops->go_callback(gl, state);
1696 handle_callback(gl, state);
1698 spin_lock(&gl->gl_spin);
1700 spin_unlock(&gl->gl_spin);
1706 * gfs2_glock_cb - Callback used by locking module
1707 * @fsdata: Pointer to the superblock
1708 * @type: Type of callback
1709 * @data: Type dependent data pointer
1711 * Called by the locking module when it wants to tell us something.
1712 * Either we need to drop a lock, one of our ASYNC requests completed, or
1713 * a journal from another client needs to be recovered.
1716 void gfs2_glock_cb(lm_fsdata_t *fsdata, unsigned int type, void *data)
1718 struct gfs2_sbd *sdp = (struct gfs2_sbd *)fsdata;
1722 blocking_cb(sdp, data, LM_ST_UNLOCKED);
1726 blocking_cb(sdp, data, LM_ST_DEFERRED);
1730 blocking_cb(sdp, data, LM_ST_SHARED);
1734 struct lm_async_cb *async = data;
1735 struct gfs2_glock *gl;
1737 gl = gfs2_glock_find(sdp, &async->lc_name);
1738 if (gfs2_assert_warn(sdp, gl))
1740 if (!gfs2_assert_warn(sdp, gl->gl_req_bh))
1741 gl->gl_req_bh(gl, async->lc_ret);
1746 case LM_CB_NEED_RECOVERY:
1747 gfs2_jdesc_make_dirty(sdp, *(unsigned int *)data);
1748 if (sdp->sd_recoverd_process)
1749 wake_up_process(sdp->sd_recoverd_process);
1752 case LM_CB_DROPLOCKS:
1753 gfs2_gl_hash_clear(sdp, NO_WAIT);
1754 gfs2_quota_scan(sdp);
1758 gfs2_assert_warn(sdp, 0);
1764 * gfs2_iopen_go_callback - Try to kick the inode/vnode associated with an
1765 * iopen glock from memory
1766 * @io_gl: the iopen glock
1767 * @state: the state into which the glock should be put
1771 void gfs2_iopen_go_callback(struct gfs2_glock *io_gl, unsigned int state)
1774 if (state != LM_ST_UNLOCKED)
1776 /* FIXME: remove this? */
1780 * demote_ok - Check to see if it's ok to unlock a glock
1783 * Returns: 1 if it's ok
1786 static int demote_ok(struct gfs2_glock *gl)
1788 struct gfs2_sbd *sdp = gl->gl_sbd;
1789 const struct gfs2_glock_operations *glops = gl->gl_ops;
1792 if (test_bit(GLF_STICKY, &gl->gl_flags))
1794 else if (test_bit(GLF_PREFETCH, &gl->gl_flags))
1795 demote = time_after_eq(jiffies, gl->gl_stamp +
1796 gfs2_tune_get(sdp, gt_prefetch_secs) * HZ);
1797 else if (glops->go_demote_ok)
1798 demote = glops->go_demote_ok(gl);
1804 * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1809 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
1811 struct gfs2_sbd *sdp = gl->gl_sbd;
1813 spin_lock(&sdp->sd_reclaim_lock);
1814 if (list_empty(&gl->gl_reclaim)) {
1815 gfs2_glock_hold(gl);
1816 list_add(&gl->gl_reclaim, &sdp->sd_reclaim_list);
1817 atomic_inc(&sdp->sd_reclaim_count);
1819 spin_unlock(&sdp->sd_reclaim_lock);
1821 wake_up(&sdp->sd_reclaim_wq);
1825 * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1826 * @sdp: the filesystem
1828 * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1829 * different glock and we notice that there are a lot of glocks in the
1834 void gfs2_reclaim_glock(struct gfs2_sbd *sdp)
1836 struct gfs2_glock *gl;
1838 spin_lock(&sdp->sd_reclaim_lock);
1839 if (list_empty(&sdp->sd_reclaim_list)) {
1840 spin_unlock(&sdp->sd_reclaim_lock);
1843 gl = list_entry(sdp->sd_reclaim_list.next,
1844 struct gfs2_glock, gl_reclaim);
1845 list_del_init(&gl->gl_reclaim);
1846 spin_unlock(&sdp->sd_reclaim_lock);
1848 atomic_dec(&sdp->sd_reclaim_count);
1849 atomic_inc(&sdp->sd_reclaimed);
1851 if (gfs2_glmutex_trylock(gl)) {
1852 if (queue_empty(gl, &gl->gl_holders) &&
1853 gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1854 handle_callback(gl, LM_ST_UNLOCKED);
1855 gfs2_glmutex_unlock(gl);
1862 * examine_bucket - Call a function for glock in a hash bucket
1863 * @examiner: the function
1864 * @sdp: the filesystem
1865 * @bucket: the bucket
1867 * Returns: 1 if the bucket has entries
1870 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1871 struct gfs2_gl_hash_bucket *bucket)
1873 struct glock_plug plug;
1874 struct list_head *tmp;
1875 struct gfs2_glock *gl;
1878 /* Add "plug" to end of bucket list, work back up list from there */
1879 memset(&plug.gl_flags, 0, sizeof(unsigned long));
1880 set_bit(GLF_PLUG, &plug.gl_flags);
1882 write_lock(&bucket->hb_lock);
1883 list_add(&plug.gl_list, &bucket->hb_list);
1884 write_unlock(&bucket->hb_lock);
1887 write_lock(&bucket->hb_lock);
1890 tmp = plug.gl_list.next;
1892 if (tmp == &bucket->hb_list) {
1893 list_del(&plug.gl_list);
1894 entries = !list_empty(&bucket->hb_list);
1895 write_unlock(&bucket->hb_lock);
1898 gl = list_entry(tmp, struct gfs2_glock, gl_list);
1900 /* Move plug up list */
1901 list_move(&plug.gl_list, &gl->gl_list);
1903 if (test_bit(GLF_PLUG, &gl->gl_flags))
1905 if (gl->gl_sbd != sdp)
1908 /* examiner() must glock_put() */
1909 gfs2_glock_hold(gl);
1914 write_unlock(&bucket->hb_lock);
1921 * scan_glock - look at a glock and see if we can reclaim it
1922 * @gl: the glock to look at
1926 static void scan_glock(struct gfs2_glock *gl)
1928 if (gl->gl_ops == &gfs2_inode_glops)
1931 if (gfs2_glmutex_trylock(gl)) {
1932 if (queue_empty(gl, &gl->gl_holders) &&
1933 gl->gl_state != LM_ST_UNLOCKED &&
1936 gfs2_glmutex_unlock(gl);
1943 gfs2_glmutex_unlock(gl);
1944 gfs2_glock_schedule_for_reclaim(gl);
1949 * gfs2_scand_internal - Look for glocks and inodes to toss from memory
1950 * @sdp: the filesystem
1954 void gfs2_scand_internal(struct gfs2_sbd *sdp)
1958 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1959 examine_bucket(scan_glock, sdp, &gl_hash_table[x]);
1965 * clear_glock - look at a glock and see if we can free it from glock cache
1966 * @gl: the glock to look at
1970 static void clear_glock(struct gfs2_glock *gl)
1972 struct gfs2_sbd *sdp = gl->gl_sbd;
1975 spin_lock(&sdp->sd_reclaim_lock);
1976 if (!list_empty(&gl->gl_reclaim)) {
1977 list_del_init(&gl->gl_reclaim);
1978 atomic_dec(&sdp->sd_reclaim_count);
1979 spin_unlock(&sdp->sd_reclaim_lock);
1980 released = gfs2_glock_put(gl);
1981 gfs2_assert(sdp, !released);
1983 spin_unlock(&sdp->sd_reclaim_lock);
1986 if (gfs2_glmutex_trylock(gl)) {
1987 if (queue_empty(gl, &gl->gl_holders) &&
1988 gl->gl_state != LM_ST_UNLOCKED)
1989 handle_callback(gl, LM_ST_UNLOCKED);
1991 gfs2_glmutex_unlock(gl);
1998 * gfs2_gl_hash_clear - Empty out the glock hash table
1999 * @sdp: the filesystem
2000 * @wait: wait until it's all gone
2002 * Called when unmounting the filesystem, or when inter-node lock manager
2003 * requests DROPLOCKS because it is running out of capacity.
2006 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp, int wait)
2017 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
2018 if (examine_bucket(clear_glock, sdp, &gl_hash_table[x]))
2024 if (time_after_eq(jiffies,
2025 t + gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
2026 fs_warn(sdp, "Unmount seems to be stalled. "
2027 "Dumping lock state...\n");
2028 gfs2_dump_lockstate(sdp);
2032 invalidate_inodes(sdp->sd_vfs);
2038 * Diagnostic routines to help debug distributed deadlock
2042 * dump_holder - print information about a glock holder
2043 * @str: a string naming the type of holder
2044 * @gh: the glock holder
2046 * Returns: 0 on success, -ENOBUFS when we run out of space
2049 static int dump_holder(char *str, struct gfs2_holder *gh)
2052 int error = -ENOBUFS;
2054 printk(KERN_INFO " %s\n", str);
2055 printk(KERN_INFO " owner = %ld\n",
2056 (gh->gh_owner) ? (long)gh->gh_owner->pid : -1);
2057 printk(KERN_INFO " gh_state = %u\n", gh->gh_state);
2058 printk(KERN_INFO " gh_flags =");
2059 for (x = 0; x < 32; x++)
2060 if (gh->gh_flags & (1 << x))
2063 printk(KERN_INFO " error = %d\n", gh->gh_error);
2064 printk(KERN_INFO " gh_iflags =");
2065 for (x = 0; x < 32; x++)
2066 if (test_bit(x, &gh->gh_iflags))
2069 print_symbol(KERN_INFO " initialized at: %s\n", gh->gh_ip);
2077 * dump_inode - print information about an inode
2080 * Returns: 0 on success, -ENOBUFS when we run out of space
2083 static int dump_inode(struct gfs2_inode *ip)
2086 int error = -ENOBUFS;
2088 printk(KERN_INFO " Inode:\n");
2089 printk(KERN_INFO " num = %llu %llu\n",
2090 (unsigned long long)ip->i_num.no_formal_ino,
2091 (unsigned long long)ip->i_num.no_addr);
2092 printk(KERN_INFO " type = %u\n", IF2DT(ip->i_di.di_mode));
2093 printk(KERN_INFO " i_flags =");
2094 for (x = 0; x < 32; x++)
2095 if (test_bit(x, &ip->i_flags))
2105 * dump_glock - print information about a glock
2107 * @count: where we are in the buffer
2109 * Returns: 0 on success, -ENOBUFS when we run out of space
2112 static int dump_glock(struct gfs2_glock *gl)
2114 struct gfs2_holder *gh;
2116 int error = -ENOBUFS;
2118 spin_lock(&gl->gl_spin);
2120 printk(KERN_INFO "Glock 0x%p (%u, %llu)\n", gl, gl->gl_name.ln_type,
2121 (unsigned long long)gl->gl_name.ln_number);
2122 printk(KERN_INFO " gl_flags =");
2123 for (x = 0; x < 32; x++) {
2124 if (test_bit(x, &gl->gl_flags))
2128 printk(KERN_INFO " gl_ref = %d\n", atomic_read(&gl->gl_ref.refcount));
2129 printk(KERN_INFO " gl_state = %u\n", gl->gl_state);
2130 printk(KERN_INFO " gl_owner = %s\n", gl->gl_owner->comm);
2131 print_symbol(KERN_INFO " gl_ip = %s\n", gl->gl_ip);
2132 printk(KERN_INFO " req_gh = %s\n", (gl->gl_req_gh) ? "yes" : "no");
2133 printk(KERN_INFO " req_bh = %s\n", (gl->gl_req_bh) ? "yes" : "no");
2134 printk(KERN_INFO " lvb_count = %d\n", atomic_read(&gl->gl_lvb_count));
2135 printk(KERN_INFO " object = %s\n", (gl->gl_object) ? "yes" : "no");
2136 printk(KERN_INFO " le = %s\n",
2137 (list_empty(&gl->gl_le.le_list)) ? "no" : "yes");
2138 printk(KERN_INFO " reclaim = %s\n",
2139 (list_empty(&gl->gl_reclaim)) ? "no" : "yes");
2141 printk(KERN_INFO " aspace = 0x%p nrpages = %lu\n", gl->gl_aspace,
2142 gl->gl_aspace->i_mapping->nrpages);
2144 printk(KERN_INFO " aspace = no\n");
2145 printk(KERN_INFO " ail = %d\n", atomic_read(&gl->gl_ail_count));
2146 if (gl->gl_req_gh) {
2147 error = dump_holder("Request", gl->gl_req_gh);
2151 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
2152 error = dump_holder("Holder", gh);
2156 list_for_each_entry(gh, &gl->gl_waiters1, gh_list) {
2157 error = dump_holder("Waiter1", gh);
2161 list_for_each_entry(gh, &gl->gl_waiters2, gh_list) {
2162 error = dump_holder("Waiter2", gh);
2166 list_for_each_entry(gh, &gl->gl_waiters3, gh_list) {
2167 error = dump_holder("Waiter3", gh);
2171 if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object) {
2172 if (!test_bit(GLF_LOCK, &gl->gl_flags) &&
2173 list_empty(&gl->gl_holders)) {
2174 error = dump_inode(gl->gl_object);
2179 printk(KERN_INFO " Inode: busy\n");
2186 spin_unlock(&gl->gl_spin);
2191 * gfs2_dump_lockstate - print out the current lockstate
2192 * @sdp: the filesystem
2193 * @ub: the buffer to copy the information into
2195 * If @ub is NULL, dump the lockstate to the console.
2199 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
2201 struct gfs2_gl_hash_bucket *bucket;
2202 struct gfs2_glock *gl;
2206 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
2207 bucket = &gl_hash_table[x];
2209 read_lock(&bucket->hb_lock);
2211 list_for_each_entry(gl, &bucket->hb_list, gl_list) {
2212 if (test_bit(GLF_PLUG, &gl->gl_flags))
2214 if (gl->gl_sbd != sdp)
2217 error = dump_glock(gl);
2222 read_unlock(&bucket->hb_lock);
2232 int __init gfs2_glock_init(void)
2235 for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
2236 struct gfs2_gl_hash_bucket *hb = &gl_hash_table[i];
2237 rwlock_init(&hb->hb_lock);
2238 INIT_LIST_HEAD(&hb->hb_list);