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 <linux/kallsyms.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <asm/semaphore.h>
22 #include <asm/uaccess.h>
25 #include "lm_interface.h"
37 /* Must be kept in sync with the beginning of struct gfs2_glock */
39 struct list_head gl_list
;
40 unsigned long gl_flags
;
44 struct gfs2_holder gr_gh
;
45 struct work_struct gr_work
;
48 typedef void (*glock_examiner
) (struct gfs2_glock
* gl
);
51 * relaxed_state_ok - is a requested lock compatible with the current lock mode?
52 * @actual: the current state of the lock
53 * @requested: the lock state that was requested by the caller
54 * @flags: the modifier flags passed in by the caller
56 * Returns: 1 if the locks are compatible, 0 otherwise
59 static inline int relaxed_state_ok(unsigned int actual
, unsigned requested
,
62 if (actual
== requested
)
68 if (actual
== LM_ST_EXCLUSIVE
&& requested
== LM_ST_SHARED
)
71 if (actual
!= LM_ST_UNLOCKED
&& (flags
& LM_FLAG_ANY
))
78 * gl_hash() - Turn glock number into hash bucket number
79 * @lock: The glock number
81 * Returns: The number of the corresponding hash bucket
84 static unsigned int gl_hash(struct lm_lockname
*name
)
88 h
= jhash(&name
->ln_number
, sizeof(uint64_t), 0);
89 h
= jhash(&name
->ln_type
, sizeof(unsigned int), h
);
90 h
&= GFS2_GL_HASH_MASK
;
96 * glock_free() - Perform a few checks and then release struct gfs2_glock
97 * @gl: The glock to release
99 * Also calls lock module to release its internal structure for this glock.
103 static void glock_free(struct gfs2_glock
*gl
)
105 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
106 struct inode
*aspace
= gl
->gl_aspace
;
108 gfs2_lm_put_lock(sdp
, gl
->gl_lock
);
111 gfs2_aspace_put(aspace
);
113 kmem_cache_free(gfs2_glock_cachep
, gl
);
117 * gfs2_glock_hold() - increment reference count on glock
118 * @gl: The glock to hold
122 void gfs2_glock_hold(struct gfs2_glock
*gl
)
124 kref_get(&gl
->gl_ref
);
127 /* All work is done after the return from kref_put() so we
128 can release the write_lock before the free. */
130 static void kill_glock(struct kref
*kref
)
132 struct gfs2_glock
*gl
= container_of(kref
, struct gfs2_glock
, gl_ref
);
133 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
135 gfs2_assert(sdp
, gl
->gl_state
== LM_ST_UNLOCKED
);
136 gfs2_assert(sdp
, list_empty(&gl
->gl_reclaim
));
137 gfs2_assert(sdp
, list_empty(&gl
->gl_holders
));
138 gfs2_assert(sdp
, list_empty(&gl
->gl_waiters1
));
139 gfs2_assert(sdp
, list_empty(&gl
->gl_waiters2
));
140 gfs2_assert(sdp
, list_empty(&gl
->gl_waiters3
));
144 * gfs2_glock_put() - Decrement reference count on glock
145 * @gl: The glock to put
149 int gfs2_glock_put(struct gfs2_glock
*gl
)
151 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
152 struct gfs2_gl_hash_bucket
*bucket
= gl
->gl_bucket
;
155 mutex_lock(&sdp
->sd_invalidate_inodes_mutex
);
157 write_lock(&bucket
->hb_lock
);
158 if (kref_put(&gl
->gl_ref
, kill_glock
)) {
159 list_del_init(&gl
->gl_list
);
160 write_unlock(&bucket
->hb_lock
);
161 BUG_ON(spin_is_locked(&gl
->gl_spin
));
166 write_unlock(&bucket
->hb_lock
);
168 mutex_unlock(&sdp
->sd_invalidate_inodes_mutex
);
173 * queue_empty - check to see if a glock's queue is empty
175 * @head: the head of the queue to check
177 * This function protects the list in the event that a process already
178 * has a holder on the list and is adding a second holder for itself.
179 * The glmutex lock is what generally prevents processes from working
180 * on the same glock at once, but the special case of adding a second
181 * holder for yourself ("recursive" locking) doesn't involve locking
182 * glmutex, making the spin lock necessary.
184 * Returns: 1 if the queue is empty
187 static inline int queue_empty(struct gfs2_glock
*gl
, struct list_head
*head
)
190 spin_lock(&gl
->gl_spin
);
191 empty
= list_empty(head
);
192 spin_unlock(&gl
->gl_spin
);
197 * search_bucket() - Find struct gfs2_glock by lock number
198 * @bucket: the bucket to search
199 * @name: The lock name
201 * Returns: NULL, or the struct gfs2_glock with the requested number
204 static struct gfs2_glock
*search_bucket(struct gfs2_gl_hash_bucket
*bucket
,
205 struct lm_lockname
*name
)
207 struct gfs2_glock
*gl
;
209 list_for_each_entry(gl
, &bucket
->hb_list
, gl_list
) {
210 if (test_bit(GLF_PLUG
, &gl
->gl_flags
))
212 if (!lm_name_equal(&gl
->gl_name
, name
))
215 kref_get(&gl
->gl_ref
);
224 * gfs2_glock_find() - Find glock by lock number
225 * @sdp: The GFS2 superblock
226 * @name: The lock name
228 * Returns: NULL, or the struct gfs2_glock with the requested number
231 struct gfs2_glock
*gfs2_glock_find(struct gfs2_sbd
*sdp
,
232 struct lm_lockname
*name
)
234 struct gfs2_gl_hash_bucket
*bucket
= &sdp
->sd_gl_hash
[gl_hash(name
)];
235 struct gfs2_glock
*gl
;
237 read_lock(&bucket
->hb_lock
);
238 gl
= search_bucket(bucket
, name
);
239 read_unlock(&bucket
->hb_lock
);
245 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
246 * @sdp: The GFS2 superblock
247 * @number: the lock number
248 * @glops: The glock_operations to use
249 * @create: If 0, don't create the glock if it doesn't exist
250 * @glp: the glock is returned here
252 * This does not lock a glock, just finds/creates structures for one.
257 int gfs2_glock_get(struct gfs2_sbd
*sdp
, uint64_t number
,
258 struct gfs2_glock_operations
*glops
, int create
,
259 struct gfs2_glock
**glp
)
261 struct lm_lockname name
;
262 struct gfs2_glock
*gl
, *tmp
;
263 struct gfs2_gl_hash_bucket
*bucket
;
266 name
.ln_number
= number
;
267 name
.ln_type
= glops
->go_type
;
268 bucket
= &sdp
->sd_gl_hash
[gl_hash(&name
)];
270 read_lock(&bucket
->hb_lock
);
271 gl
= search_bucket(bucket
, &name
);
272 read_unlock(&bucket
->hb_lock
);
279 gl
= kmem_cache_alloc(gfs2_glock_cachep
, GFP_KERNEL
);
283 memset(gl
, 0, sizeof(struct gfs2_glock
));
285 INIT_LIST_HEAD(&gl
->gl_list
);
287 kref_init(&gl
->gl_ref
);
289 spin_lock_init(&gl
->gl_spin
);
291 gl
->gl_state
= LM_ST_UNLOCKED
;
292 INIT_LIST_HEAD(&gl
->gl_holders
);
293 INIT_LIST_HEAD(&gl
->gl_waiters1
);
294 INIT_LIST_HEAD(&gl
->gl_waiters2
);
295 INIT_LIST_HEAD(&gl
->gl_waiters3
);
299 gl
->gl_bucket
= bucket
;
300 INIT_LIST_HEAD(&gl
->gl_reclaim
);
304 lops_init_le(&gl
->gl_le
, &gfs2_glock_lops
);
305 INIT_LIST_HEAD(&gl
->gl_ail_list
);
307 /* If this glock protects actual on-disk data or metadata blocks,
308 create a VFS inode to manage the pages/buffers holding them. */
309 if (glops
== &gfs2_inode_glops
||
310 glops
== &gfs2_rgrp_glops
||
311 glops
== &gfs2_meta_glops
) {
312 gl
->gl_aspace
= gfs2_aspace_get(sdp
);
313 if (!gl
->gl_aspace
) {
319 error
= gfs2_lm_get_lock(sdp
, &name
, &gl
->gl_lock
);
323 write_lock(&bucket
->hb_lock
);
324 tmp
= search_bucket(bucket
, &name
);
326 write_unlock(&bucket
->hb_lock
);
330 list_add_tail(&gl
->gl_list
, &bucket
->hb_list
);
331 write_unlock(&bucket
->hb_lock
);
340 gfs2_aspace_put(gl
->gl_aspace
);
343 kmem_cache_free(gfs2_glock_cachep
, gl
);
349 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
351 * @state: the state we're requesting
352 * @flags: the modifier flags
353 * @gh: the holder structure
357 void gfs2_holder_init(struct gfs2_glock
*gl
, unsigned int state
, unsigned flags
,
358 struct gfs2_holder
*gh
)
360 INIT_LIST_HEAD(&gh
->gh_list
);
362 gh
->gh_ip
= (unsigned long)__builtin_return_address(0);
363 gh
->gh_owner
= current
;
364 gh
->gh_state
= state
;
365 gh
->gh_flags
= flags
;
368 init_completion(&gh
->gh_wait
);
370 if (gh
->gh_state
== LM_ST_EXCLUSIVE
)
371 gh
->gh_flags
|= GL_LOCAL_EXCL
;
377 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
378 * @state: the state we're requesting
379 * @flags: the modifier flags
380 * @gh: the holder structure
382 * Don't mess with the glock.
386 void gfs2_holder_reinit(unsigned int state
, unsigned flags
, struct gfs2_holder
*gh
)
388 gh
->gh_state
= state
;
389 gh
->gh_flags
= flags
;
390 if (gh
->gh_state
== LM_ST_EXCLUSIVE
)
391 gh
->gh_flags
|= GL_LOCAL_EXCL
;
393 gh
->gh_iflags
&= 1 << HIF_ALLOCED
;
394 gh
->gh_ip
= (unsigned long)__builtin_return_address(0);
398 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
399 * @gh: the holder structure
403 void gfs2_holder_uninit(struct gfs2_holder
*gh
)
405 gfs2_glock_put(gh
->gh_gl
);
411 * gfs2_holder_get - get a struct gfs2_holder structure
413 * @state: the state we're requesting
414 * @flags: the modifier flags
415 * @gfp_flags: __GFP_NOFAIL
417 * Figure out how big an impact this function has. Either:
418 * 1) Replace it with a cache of structures hanging off the struct gfs2_sbd
419 * 2) Leave it like it is
421 * Returns: the holder structure, NULL on ENOMEM
424 struct gfs2_holder
*gfs2_holder_get(struct gfs2_glock
*gl
, unsigned int state
,
425 int flags
, gfp_t gfp_flags
)
427 struct gfs2_holder
*gh
;
429 gh
= kmalloc(sizeof(struct gfs2_holder
), gfp_flags
);
433 gfs2_holder_init(gl
, state
, flags
, gh
);
434 set_bit(HIF_ALLOCED
, &gh
->gh_iflags
);
435 gh
->gh_ip
= (unsigned long)__builtin_return_address(0);
440 * gfs2_holder_put - get rid of a struct gfs2_holder structure
441 * @gh: the holder structure
445 void gfs2_holder_put(struct gfs2_holder
*gh
)
447 gfs2_holder_uninit(gh
);
452 * rq_mutex - process a mutex request in the queue
453 * @gh: the glock holder
455 * Returns: 1 if the queue is blocked
458 static int rq_mutex(struct gfs2_holder
*gh
)
460 struct gfs2_glock
*gl
= gh
->gh_gl
;
462 list_del_init(&gh
->gh_list
);
463 /* gh->gh_error never examined. */
464 set_bit(GLF_LOCK
, &gl
->gl_flags
);
465 complete(&gh
->gh_wait
);
471 * rq_promote - process a promote request in the queue
472 * @gh: the glock holder
474 * Acquire a new inter-node lock, or change a lock state to more restrictive.
476 * Returns: 1 if the queue is blocked
479 static int rq_promote(struct gfs2_holder
*gh
)
481 struct gfs2_glock
*gl
= gh
->gh_gl
;
482 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
483 struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
485 if (!relaxed_state_ok(gl
->gl_state
, gh
->gh_state
, gh
->gh_flags
)) {
486 if (list_empty(&gl
->gl_holders
)) {
488 set_bit(GLF_LOCK
, &gl
->gl_flags
);
489 spin_unlock(&gl
->gl_spin
);
491 if (atomic_read(&sdp
->sd_reclaim_count
) >
492 gfs2_tune_get(sdp
, gt_reclaim_limit
) &&
493 !(gh
->gh_flags
& LM_FLAG_PRIORITY
)) {
494 gfs2_reclaim_glock(sdp
);
495 gfs2_reclaim_glock(sdp
);
498 glops
->go_xmote_th(gl
, gh
->gh_state
,
501 spin_lock(&gl
->gl_spin
);
506 if (list_empty(&gl
->gl_holders
)) {
507 set_bit(HIF_FIRST
, &gh
->gh_iflags
);
508 set_bit(GLF_LOCK
, &gl
->gl_flags
);
510 struct gfs2_holder
*next_gh
;
511 if (gh
->gh_flags
& GL_LOCAL_EXCL
)
513 next_gh
= list_entry(gl
->gl_holders
.next
, struct gfs2_holder
,
515 if (next_gh
->gh_flags
& GL_LOCAL_EXCL
)
519 list_move_tail(&gh
->gh_list
, &gl
->gl_holders
);
521 set_bit(HIF_HOLDER
, &gh
->gh_iflags
);
523 complete(&gh
->gh_wait
);
529 * rq_demote - process a demote request in the queue
530 * @gh: the glock holder
532 * Returns: 1 if the queue is blocked
535 static int rq_demote(struct gfs2_holder
*gh
)
537 struct gfs2_glock
*gl
= gh
->gh_gl
;
538 struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
540 if (!list_empty(&gl
->gl_holders
))
543 if (gl
->gl_state
== gh
->gh_state
|| gl
->gl_state
== LM_ST_UNLOCKED
) {
544 list_del_init(&gh
->gh_list
);
546 spin_unlock(&gl
->gl_spin
);
547 if (test_bit(HIF_DEALLOC
, &gh
->gh_iflags
))
550 complete(&gh
->gh_wait
);
551 spin_lock(&gl
->gl_spin
);
554 set_bit(GLF_LOCK
, &gl
->gl_flags
);
555 spin_unlock(&gl
->gl_spin
);
557 if (gh
->gh_state
== LM_ST_UNLOCKED
||
558 gl
->gl_state
!= LM_ST_EXCLUSIVE
)
559 glops
->go_drop_th(gl
);
561 glops
->go_xmote_th(gl
, gh
->gh_state
, gh
->gh_flags
);
563 spin_lock(&gl
->gl_spin
);
570 * rq_greedy - process a queued request to drop greedy status
571 * @gh: the glock holder
573 * Returns: 1 if the queue is blocked
576 static int rq_greedy(struct gfs2_holder
*gh
)
578 struct gfs2_glock
*gl
= gh
->gh_gl
;
580 list_del_init(&gh
->gh_list
);
581 /* gh->gh_error never examined. */
582 clear_bit(GLF_GREEDY
, &gl
->gl_flags
);
583 spin_unlock(&gl
->gl_spin
);
585 gfs2_holder_uninit(gh
);
586 kfree(container_of(gh
, struct greedy
, gr_gh
));
588 spin_lock(&gl
->gl_spin
);
594 * run_queue - process holder structures on a glock
598 static void run_queue(struct gfs2_glock
*gl
)
600 struct gfs2_holder
*gh
;
604 if (test_bit(GLF_LOCK
, &gl
->gl_flags
))
607 if (!list_empty(&gl
->gl_waiters1
)) {
608 gh
= list_entry(gl
->gl_waiters1
.next
,
609 struct gfs2_holder
, gh_list
);
611 if (test_bit(HIF_MUTEX
, &gh
->gh_iflags
))
612 blocked
= rq_mutex(gh
);
614 gfs2_assert_warn(gl
->gl_sbd
, 0);
616 } else if (!list_empty(&gl
->gl_waiters2
) &&
617 !test_bit(GLF_SKIP_WAITERS2
, &gl
->gl_flags
)) {
618 gh
= list_entry(gl
->gl_waiters2
.next
,
619 struct gfs2_holder
, gh_list
);
621 if (test_bit(HIF_DEMOTE
, &gh
->gh_iflags
))
622 blocked
= rq_demote(gh
);
623 else if (test_bit(HIF_GREEDY
, &gh
->gh_iflags
))
624 blocked
= rq_greedy(gh
);
626 gfs2_assert_warn(gl
->gl_sbd
, 0);
628 } else if (!list_empty(&gl
->gl_waiters3
)) {
629 gh
= list_entry(gl
->gl_waiters3
.next
,
630 struct gfs2_holder
, gh_list
);
632 if (test_bit(HIF_PROMOTE
, &gh
->gh_iflags
))
633 blocked
= rq_promote(gh
);
635 gfs2_assert_warn(gl
->gl_sbd
, 0);
646 * gfs2_glmutex_lock - acquire a local lock on a glock
649 * Gives caller exclusive access to manipulate a glock structure.
652 void gfs2_glmutex_lock(struct gfs2_glock
*gl
)
654 struct gfs2_holder gh
;
656 gfs2_holder_init(gl
, 0, 0, &gh
);
657 set_bit(HIF_MUTEX
, &gh
.gh_iflags
);
659 spin_lock(&gl
->gl_spin
);
660 if (test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
))
661 list_add_tail(&gh
.gh_list
, &gl
->gl_waiters1
);
663 complete(&gh
.gh_wait
);
664 spin_unlock(&gl
->gl_spin
);
666 wait_for_completion(&gh
.gh_wait
);
667 gfs2_holder_uninit(&gh
);
671 * gfs2_glmutex_trylock - try to acquire a local lock on a glock
674 * Returns: 1 if the glock is acquired
677 int gfs2_glmutex_trylock(struct gfs2_glock
*gl
)
681 spin_lock(&gl
->gl_spin
);
682 if (test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
))
684 spin_unlock(&gl
->gl_spin
);
690 * gfs2_glmutex_unlock - release a local lock on a glock
695 void gfs2_glmutex_unlock(struct gfs2_glock
*gl
)
697 spin_lock(&gl
->gl_spin
);
698 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
700 BUG_ON(!spin_is_locked(&gl
->gl_spin
));
701 spin_unlock(&gl
->gl_spin
);
705 * handle_callback - add a demote request to a lock's queue
707 * @state: the state the caller wants us to change to
711 static void handle_callback(struct gfs2_glock
*gl
, unsigned int state
)
713 struct gfs2_holder
*gh
, *new_gh
= NULL
;
716 spin_lock(&gl
->gl_spin
);
718 list_for_each_entry(gh
, &gl
->gl_waiters2
, gh_list
) {
719 if (test_bit(HIF_DEMOTE
, &gh
->gh_iflags
) &&
720 gl
->gl_req_gh
!= gh
) {
721 if (gh
->gh_state
!= state
)
722 gh
->gh_state
= LM_ST_UNLOCKED
;
728 list_add_tail(&new_gh
->gh_list
, &gl
->gl_waiters2
);
731 spin_unlock(&gl
->gl_spin
);
733 new_gh
= gfs2_holder_get(gl
, state
, LM_FLAG_TRY
,
734 GFP_KERNEL
| __GFP_NOFAIL
),
735 set_bit(HIF_DEMOTE
, &new_gh
->gh_iflags
);
736 set_bit(HIF_DEALLOC
, &new_gh
->gh_iflags
);
742 spin_unlock(&gl
->gl_spin
);
745 gfs2_holder_put(new_gh
);
749 * state_change - record that the glock is now in a different state
751 * @new_state the new state
755 static void state_change(struct gfs2_glock
*gl
, unsigned int new_state
)
759 held1
= (gl
->gl_state
!= LM_ST_UNLOCKED
);
760 held2
= (new_state
!= LM_ST_UNLOCKED
);
762 if (held1
!= held2
) {
769 gl
->gl_state
= new_state
;
773 * xmote_bh - Called after the lock module is done acquiring a lock
774 * @gl: The glock in question
775 * @ret: the int returned from the lock module
779 static void xmote_bh(struct gfs2_glock
*gl
, unsigned int ret
)
781 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
782 struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
783 struct gfs2_holder
*gh
= gl
->gl_req_gh
;
784 int prev_state
= gl
->gl_state
;
787 gfs2_assert_warn(sdp
, test_bit(GLF_LOCK
, &gl
->gl_flags
));
788 gfs2_assert_warn(sdp
, queue_empty(gl
, &gl
->gl_holders
));
789 gfs2_assert_warn(sdp
, !(ret
& LM_OUT_ASYNC
));
791 state_change(gl
, ret
& LM_OUT_ST_MASK
);
793 if (prev_state
!= LM_ST_UNLOCKED
&& !(ret
& LM_OUT_CACHEABLE
)) {
795 glops
->go_inval(gl
, DIO_METADATA
| DIO_DATA
);
796 } else if (gl
->gl_state
== LM_ST_DEFERRED
) {
797 /* We might not want to do this here.
798 Look at moving to the inode glops. */
800 glops
->go_inval(gl
, DIO_DATA
);
803 /* Deal with each possible exit condition */
806 gl
->gl_stamp
= jiffies
;
808 else if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
))) {
809 spin_lock(&gl
->gl_spin
);
810 list_del_init(&gh
->gh_list
);
812 spin_unlock(&gl
->gl_spin
);
814 } else if (test_bit(HIF_DEMOTE
, &gh
->gh_iflags
)) {
815 spin_lock(&gl
->gl_spin
);
816 list_del_init(&gh
->gh_list
);
817 if (gl
->gl_state
== gh
->gh_state
||
818 gl
->gl_state
== LM_ST_UNLOCKED
)
821 if (gfs2_assert_warn(sdp
, gh
->gh_flags
&
822 (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
)) == -1)
823 fs_warn(sdp
, "ret = 0x%.8X\n", ret
);
824 gh
->gh_error
= GLR_TRYFAILED
;
826 spin_unlock(&gl
->gl_spin
);
828 if (ret
& LM_OUT_CANCELED
)
829 handle_callback(gl
, LM_ST_UNLOCKED
); /* Lame */
831 } else if (ret
& LM_OUT_CANCELED
) {
832 spin_lock(&gl
->gl_spin
);
833 list_del_init(&gh
->gh_list
);
834 gh
->gh_error
= GLR_CANCELED
;
835 spin_unlock(&gl
->gl_spin
);
837 } else if (relaxed_state_ok(gl
->gl_state
, gh
->gh_state
, gh
->gh_flags
)) {
838 spin_lock(&gl
->gl_spin
);
839 list_move_tail(&gh
->gh_list
, &gl
->gl_holders
);
841 set_bit(HIF_HOLDER
, &gh
->gh_iflags
);
842 spin_unlock(&gl
->gl_spin
);
844 set_bit(HIF_FIRST
, &gh
->gh_iflags
);
848 } else if (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
)) {
849 spin_lock(&gl
->gl_spin
);
850 list_del_init(&gh
->gh_list
);
851 gh
->gh_error
= GLR_TRYFAILED
;
852 spin_unlock(&gl
->gl_spin
);
855 if (gfs2_assert_withdraw(sdp
, 0) == -1)
856 fs_err(sdp
, "ret = 0x%.8X\n", ret
);
859 if (glops
->go_xmote_bh
)
860 glops
->go_xmote_bh(gl
);
863 spin_lock(&gl
->gl_spin
);
864 gl
->gl_req_gh
= NULL
;
865 gl
->gl_req_bh
= NULL
;
866 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
868 spin_unlock(&gl
->gl_spin
);
874 if (test_bit(HIF_DEALLOC
, &gh
->gh_iflags
))
877 complete(&gh
->gh_wait
);
882 * gfs2_glock_xmote_th - Call into the lock module to acquire or change a glock
883 * @gl: The glock in question
884 * @state: the requested state
885 * @flags: modifier flags to the lock call
889 void gfs2_glock_xmote_th(struct gfs2_glock
*gl
, unsigned int state
, int flags
)
891 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
892 struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
893 int lck_flags
= flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
|
894 LM_FLAG_NOEXP
| LM_FLAG_ANY
|
896 unsigned int lck_ret
;
898 gfs2_assert_warn(sdp
, test_bit(GLF_LOCK
, &gl
->gl_flags
));
899 gfs2_assert_warn(sdp
, queue_empty(gl
, &gl
->gl_holders
));
900 gfs2_assert_warn(sdp
, state
!= LM_ST_UNLOCKED
);
901 gfs2_assert_warn(sdp
, state
!= gl
->gl_state
);
903 if (gl
->gl_state
== LM_ST_EXCLUSIVE
) {
906 DIO_METADATA
| DIO_DATA
| DIO_RELEASE
);
910 gl
->gl_req_bh
= xmote_bh
;
912 lck_ret
= gfs2_lm_lock(sdp
, gl
->gl_lock
, gl
->gl_state
, state
,
915 if (gfs2_assert_withdraw(sdp
, !(lck_ret
& LM_OUT_ERROR
)))
918 if (lck_ret
& LM_OUT_ASYNC
)
919 gfs2_assert_warn(sdp
, lck_ret
== LM_OUT_ASYNC
);
921 xmote_bh(gl
, lck_ret
);
925 * drop_bh - Called after a lock module unlock completes
927 * @ret: the return status
929 * Doesn't wake up the process waiting on the struct gfs2_holder (if any)
930 * Doesn't drop the reference on the glock the top half took out
934 static void drop_bh(struct gfs2_glock
*gl
, unsigned int ret
)
936 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
937 struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
938 struct gfs2_holder
*gh
= gl
->gl_req_gh
;
940 clear_bit(GLF_PREFETCH
, &gl
->gl_flags
);
942 gfs2_assert_warn(sdp
, test_bit(GLF_LOCK
, &gl
->gl_flags
));
943 gfs2_assert_warn(sdp
, queue_empty(gl
, &gl
->gl_holders
));
944 gfs2_assert_warn(sdp
, !ret
);
946 state_change(gl
, LM_ST_UNLOCKED
);
949 glops
->go_inval(gl
, DIO_METADATA
| DIO_DATA
);
952 spin_lock(&gl
->gl_spin
);
953 list_del_init(&gh
->gh_list
);
955 spin_unlock(&gl
->gl_spin
);
958 if (glops
->go_drop_bh
)
959 glops
->go_drop_bh(gl
);
961 spin_lock(&gl
->gl_spin
);
962 gl
->gl_req_gh
= NULL
;
963 gl
->gl_req_bh
= NULL
;
964 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
966 spin_unlock(&gl
->gl_spin
);
971 if (test_bit(HIF_DEALLOC
, &gh
->gh_iflags
))
974 complete(&gh
->gh_wait
);
979 * gfs2_glock_drop_th - call into the lock module to unlock a lock
984 void gfs2_glock_drop_th(struct gfs2_glock
*gl
)
986 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
987 struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
990 gfs2_assert_warn(sdp
, test_bit(GLF_LOCK
, &gl
->gl_flags
));
991 gfs2_assert_warn(sdp
, queue_empty(gl
, &gl
->gl_holders
));
992 gfs2_assert_warn(sdp
, gl
->gl_state
!= LM_ST_UNLOCKED
);
994 if (gl
->gl_state
== LM_ST_EXCLUSIVE
) {
997 DIO_METADATA
| DIO_DATA
| DIO_RELEASE
);
1000 gfs2_glock_hold(gl
);
1001 gl
->gl_req_bh
= drop_bh
;
1003 ret
= gfs2_lm_unlock(sdp
, gl
->gl_lock
, gl
->gl_state
);
1005 if (gfs2_assert_withdraw(sdp
, !(ret
& LM_OUT_ERROR
)))
1011 gfs2_assert_warn(sdp
, ret
== LM_OUT_ASYNC
);
1015 * do_cancels - cancel requests for locks stuck waiting on an expire flag
1016 * @gh: the LM_FLAG_PRIORITY holder waiting to acquire the lock
1018 * Don't cancel GL_NOCANCEL requests.
1021 static void do_cancels(struct gfs2_holder
*gh
)
1023 struct gfs2_glock
*gl
= gh
->gh_gl
;
1025 spin_lock(&gl
->gl_spin
);
1027 while (gl
->gl_req_gh
!= gh
&&
1028 !test_bit(HIF_HOLDER
, &gh
->gh_iflags
) &&
1029 !list_empty(&gh
->gh_list
)) {
1030 if (gl
->gl_req_bh
&&
1032 (gl
->gl_req_gh
->gh_flags
& GL_NOCANCEL
))) {
1033 spin_unlock(&gl
->gl_spin
);
1034 gfs2_lm_cancel(gl
->gl_sbd
, gl
->gl_lock
);
1036 spin_lock(&gl
->gl_spin
);
1038 spin_unlock(&gl
->gl_spin
);
1040 spin_lock(&gl
->gl_spin
);
1044 spin_unlock(&gl
->gl_spin
);
1048 * glock_wait_internal - wait on a glock acquisition
1049 * @gh: the glock holder
1051 * Returns: 0 on success
1054 static int glock_wait_internal(struct gfs2_holder
*gh
)
1056 struct gfs2_glock
*gl
= gh
->gh_gl
;
1057 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
1058 struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1060 if (test_bit(HIF_ABORTED
, &gh
->gh_iflags
))
1063 if (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
)) {
1064 spin_lock(&gl
->gl_spin
);
1065 if (gl
->gl_req_gh
!= gh
&&
1066 !test_bit(HIF_HOLDER
, &gh
->gh_iflags
) &&
1067 !list_empty(&gh
->gh_list
)) {
1068 list_del_init(&gh
->gh_list
);
1069 gh
->gh_error
= GLR_TRYFAILED
;
1071 spin_unlock(&gl
->gl_spin
);
1072 return gh
->gh_error
;
1074 spin_unlock(&gl
->gl_spin
);
1077 if (gh
->gh_flags
& LM_FLAG_PRIORITY
)
1080 wait_for_completion(&gh
->gh_wait
);
1083 return gh
->gh_error
;
1085 gfs2_assert_withdraw(sdp
, test_bit(HIF_HOLDER
, &gh
->gh_iflags
));
1086 gfs2_assert_withdraw(sdp
, relaxed_state_ok(gl
->gl_state
,
1090 if (test_bit(HIF_FIRST
, &gh
->gh_iflags
)) {
1091 gfs2_assert_warn(sdp
, test_bit(GLF_LOCK
, &gl
->gl_flags
));
1093 if (glops
->go_lock
) {
1094 gh
->gh_error
= glops
->go_lock(gh
);
1096 spin_lock(&gl
->gl_spin
);
1097 list_del_init(&gh
->gh_list
);
1098 spin_unlock(&gl
->gl_spin
);
1102 spin_lock(&gl
->gl_spin
);
1103 gl
->gl_req_gh
= NULL
;
1104 gl
->gl_req_bh
= NULL
;
1105 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
1107 spin_unlock(&gl
->gl_spin
);
1110 return gh
->gh_error
;
1113 static inline struct gfs2_holder
*
1114 find_holder_by_owner(struct list_head
*head
, struct task_struct
*owner
)
1116 struct gfs2_holder
*gh
;
1118 list_for_each_entry(gh
, head
, gh_list
) {
1119 if (gh
->gh_owner
== owner
)
1127 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1128 * @gh: the holder structure to add
1132 static void add_to_queue(struct gfs2_holder
*gh
)
1134 struct gfs2_glock
*gl
= gh
->gh_gl
;
1135 struct gfs2_holder
*existing
;
1137 BUG_ON(!gh
->gh_owner
);
1139 existing
= find_holder_by_owner(&gl
->gl_holders
, gh
->gh_owner
);
1141 print_symbol(KERN_WARNING
"original: %s\n", existing
->gh_ip
);
1142 print_symbol(KERN_WARNING
"new: %s\n", gh
->gh_ip
);
1146 existing
= find_holder_by_owner(&gl
->gl_waiters3
, gh
->gh_owner
);
1148 print_symbol(KERN_WARNING
"original: %s\n", existing
->gh_ip
);
1149 print_symbol(KERN_WARNING
"new: %s\n", gh
->gh_ip
);
1153 if (gh
->gh_flags
& LM_FLAG_PRIORITY
)
1154 list_add(&gh
->gh_list
, &gl
->gl_waiters3
);
1156 list_add_tail(&gh
->gh_list
, &gl
->gl_waiters3
);
1160 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1161 * @gh: the holder structure
1163 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1165 * Returns: 0, GLR_TRYFAILED, or errno on failure
1168 int gfs2_glock_nq(struct gfs2_holder
*gh
)
1170 struct gfs2_glock
*gl
= gh
->gh_gl
;
1171 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
1175 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
))) {
1176 set_bit(HIF_ABORTED
, &gh
->gh_iflags
);
1180 set_bit(HIF_PROMOTE
, &gh
->gh_iflags
);
1182 spin_lock(&gl
->gl_spin
);
1185 spin_unlock(&gl
->gl_spin
);
1187 if (!(gh
->gh_flags
& GL_ASYNC
)) {
1188 error
= glock_wait_internal(gh
);
1189 if (error
== GLR_CANCELED
) {
1195 clear_bit(GLF_PREFETCH
, &gl
->gl_flags
);
1201 * gfs2_glock_poll - poll to see if an async request has been completed
1204 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1207 int gfs2_glock_poll(struct gfs2_holder
*gh
)
1209 struct gfs2_glock
*gl
= gh
->gh_gl
;
1212 spin_lock(&gl
->gl_spin
);
1214 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
1216 else if (list_empty(&gh
->gh_list
)) {
1217 if (gh
->gh_error
== GLR_CANCELED
) {
1218 spin_unlock(&gl
->gl_spin
);
1220 if (gfs2_glock_nq(gh
))
1227 spin_unlock(&gl
->gl_spin
);
1233 * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
1234 * @gh: the holder structure
1236 * Returns: 0, GLR_TRYFAILED, or errno on failure
1239 int gfs2_glock_wait(struct gfs2_holder
*gh
)
1243 error
= glock_wait_internal(gh
);
1244 if (error
== GLR_CANCELED
) {
1246 gh
->gh_flags
&= ~GL_ASYNC
;
1247 error
= gfs2_glock_nq(gh
);
1254 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1255 * @gh: the glock holder
1259 void gfs2_glock_dq(struct gfs2_holder
*gh
)
1261 struct gfs2_glock
*gl
= gh
->gh_gl
;
1262 struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1264 if (gh
->gh_flags
& GL_SYNC
)
1265 set_bit(GLF_SYNC
, &gl
->gl_flags
);
1267 if (gh
->gh_flags
& GL_NOCACHE
)
1268 handle_callback(gl
, LM_ST_UNLOCKED
);
1270 gfs2_glmutex_lock(gl
);
1272 spin_lock(&gl
->gl_spin
);
1273 list_del_init(&gh
->gh_list
);
1275 if (list_empty(&gl
->gl_holders
)) {
1276 spin_unlock(&gl
->gl_spin
);
1278 if (glops
->go_unlock
)
1279 glops
->go_unlock(gh
);
1281 if (test_bit(GLF_SYNC
, &gl
->gl_flags
)) {
1283 glops
->go_sync(gl
, DIO_METADATA
| DIO_DATA
);
1286 gl
->gl_stamp
= jiffies
;
1288 spin_lock(&gl
->gl_spin
);
1291 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
1293 spin_unlock(&gl
->gl_spin
);
1297 * gfs2_glock_prefetch - Try to prefetch a glock
1299 * @state: the state to prefetch in
1300 * @flags: flags passed to go_xmote_th()
1304 void gfs2_glock_prefetch(struct gfs2_glock
*gl
, unsigned int state
, int flags
)
1306 struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1308 spin_lock(&gl
->gl_spin
);
1310 if (test_bit(GLF_LOCK
, &gl
->gl_flags
) ||
1311 !list_empty(&gl
->gl_holders
) ||
1312 !list_empty(&gl
->gl_waiters1
) ||
1313 !list_empty(&gl
->gl_waiters2
) ||
1314 !list_empty(&gl
->gl_waiters3
) ||
1315 relaxed_state_ok(gl
->gl_state
, state
, flags
)) {
1316 spin_unlock(&gl
->gl_spin
);
1320 set_bit(GLF_PREFETCH
, &gl
->gl_flags
);
1321 set_bit(GLF_LOCK
, &gl
->gl_flags
);
1322 spin_unlock(&gl
->gl_spin
);
1324 glops
->go_xmote_th(gl
, state
, flags
);
1328 * gfs2_glock_force_drop - Force a glock to be uncached
1333 void gfs2_glock_force_drop(struct gfs2_glock
*gl
)
1335 struct gfs2_holder gh
;
1337 gfs2_holder_init(gl
, LM_ST_UNLOCKED
, 0, &gh
);
1338 set_bit(HIF_DEMOTE
, &gh
.gh_iflags
);
1340 spin_lock(&gl
->gl_spin
);
1341 list_add_tail(&gh
.gh_list
, &gl
->gl_waiters2
);
1343 spin_unlock(&gl
->gl_spin
);
1345 wait_for_completion(&gh
.gh_wait
);
1346 gfs2_holder_uninit(&gh
);
1349 static void greedy_work(void *data
)
1351 struct greedy
*gr
= data
;
1352 struct gfs2_holder
*gh
= &gr
->gr_gh
;
1353 struct gfs2_glock
*gl
= gh
->gh_gl
;
1354 struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1356 clear_bit(GLF_SKIP_WAITERS2
, &gl
->gl_flags
);
1358 if (glops
->go_greedy
)
1359 glops
->go_greedy(gl
);
1361 spin_lock(&gl
->gl_spin
);
1363 if (list_empty(&gl
->gl_waiters2
)) {
1364 clear_bit(GLF_GREEDY
, &gl
->gl_flags
);
1365 spin_unlock(&gl
->gl_spin
);
1366 gfs2_holder_uninit(gh
);
1369 gfs2_glock_hold(gl
);
1370 list_add_tail(&gh
->gh_list
, &gl
->gl_waiters2
);
1372 spin_unlock(&gl
->gl_spin
);
1378 * gfs2_glock_be_greedy -
1382 * Returns: 0 if go_greedy will be called, 1 otherwise
1385 int gfs2_glock_be_greedy(struct gfs2_glock
*gl
, unsigned int time
)
1388 struct gfs2_holder
*gh
;
1391 gl
->gl_sbd
->sd_args
.ar_localcaching
||
1392 test_and_set_bit(GLF_GREEDY
, &gl
->gl_flags
))
1395 gr
= kmalloc(sizeof(struct greedy
), GFP_KERNEL
);
1397 clear_bit(GLF_GREEDY
, &gl
->gl_flags
);
1402 gfs2_holder_init(gl
, 0, 0, gh
);
1403 set_bit(HIF_GREEDY
, &gh
->gh_iflags
);
1404 INIT_WORK(&gr
->gr_work
, greedy_work
, gr
);
1406 set_bit(GLF_SKIP_WAITERS2
, &gl
->gl_flags
);
1407 schedule_delayed_work(&gr
->gr_work
, time
);
1413 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1414 * @gh: the holder structure
1418 void gfs2_glock_dq_uninit(struct gfs2_holder
*gh
)
1421 gfs2_holder_uninit(gh
);
1425 * gfs2_glock_nq_num - acquire a glock based on lock number
1426 * @sdp: the filesystem
1427 * @number: the lock number
1428 * @glops: the glock operations for the type of glock
1429 * @state: the state to acquire the glock in
1430 * @flags: modifier flags for the aquisition
1431 * @gh: the struct gfs2_holder
1436 int gfs2_glock_nq_num(struct gfs2_sbd
*sdp
, uint64_t number
,
1437 struct gfs2_glock_operations
*glops
, unsigned int state
,
1438 int flags
, struct gfs2_holder
*gh
)
1440 struct gfs2_glock
*gl
;
1443 error
= gfs2_glock_get(sdp
, number
, glops
, CREATE
, &gl
);
1445 error
= gfs2_glock_nq_init(gl
, state
, flags
, gh
);
1453 * glock_compare - Compare two struct gfs2_glock structures for sorting
1454 * @arg_a: the first structure
1455 * @arg_b: the second structure
1459 static int glock_compare(const void *arg_a
, const void *arg_b
)
1461 struct gfs2_holder
*gh_a
= *(struct gfs2_holder
**)arg_a
;
1462 struct gfs2_holder
*gh_b
= *(struct gfs2_holder
**)arg_b
;
1463 struct lm_lockname
*a
= &gh_a
->gh_gl
->gl_name
;
1464 struct lm_lockname
*b
= &gh_b
->gh_gl
->gl_name
;
1467 if (a
->ln_number
> b
->ln_number
)
1469 else if (a
->ln_number
< b
->ln_number
)
1472 if (gh_a
->gh_state
== LM_ST_SHARED
&&
1473 gh_b
->gh_state
== LM_ST_EXCLUSIVE
)
1475 else if (!(gh_a
->gh_flags
& GL_LOCAL_EXCL
) &&
1476 (gh_b
->gh_flags
& GL_LOCAL_EXCL
))
1484 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1485 * @num_gh: the number of structures
1486 * @ghs: an array of struct gfs2_holder structures
1488 * Returns: 0 on success (all glocks acquired),
1489 * errno on failure (no glocks acquired)
1492 static int nq_m_sync(unsigned int num_gh
, struct gfs2_holder
*ghs
,
1493 struct gfs2_holder
**p
)
1498 for (x
= 0; x
< num_gh
; x
++)
1501 sort(p
, num_gh
, sizeof(struct gfs2_holder
*), glock_compare
, NULL
);
1503 for (x
= 0; x
< num_gh
; x
++) {
1504 p
[x
]->gh_flags
&= ~(LM_FLAG_TRY
| GL_ASYNC
);
1506 error
= gfs2_glock_nq(p
[x
]);
1509 gfs2_glock_dq(p
[x
]);
1518 * gfs2_glock_nq_m - acquire multiple glocks
1519 * @num_gh: the number of structures
1520 * @ghs: an array of struct gfs2_holder structures
1522 * Figure out how big an impact this function has. Either:
1523 * 1) Replace this code with code that calls gfs2_glock_prefetch()
1524 * 2) Forget async stuff and just call nq_m_sync()
1525 * 3) Leave it like it is
1527 * Returns: 0 on success (all glocks acquired),
1528 * errno on failure (no glocks acquired)
1531 int gfs2_glock_nq_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1535 int borked
= 0, serious
= 0;
1542 ghs
->gh_flags
&= ~(LM_FLAG_TRY
| GL_ASYNC
);
1543 return gfs2_glock_nq(ghs
);
1546 e
= kcalloc(num_gh
, sizeof(struct gfs2_holder
*), GFP_KERNEL
);
1550 for (x
= 0; x
< num_gh
; x
++) {
1551 ghs
[x
].gh_flags
|= LM_FLAG_TRY
| GL_ASYNC
;
1552 error
= gfs2_glock_nq(&ghs
[x
]);
1561 for (x
= 0; x
< num_gh
; x
++) {
1562 error
= e
[x
] = glock_wait_internal(&ghs
[x
]);
1565 if (error
!= GLR_TRYFAILED
&& error
!= GLR_CANCELED
)
1575 for (x
= 0; x
< num_gh
; x
++)
1577 gfs2_glock_dq(&ghs
[x
]);
1582 for (x
= 0; x
< num_gh
; x
++)
1583 gfs2_holder_reinit(ghs
[x
].gh_state
, ghs
[x
].gh_flags
,
1585 error
= nq_m_sync(num_gh
, ghs
, (struct gfs2_holder
**)e
);
1594 * gfs2_glock_dq_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_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1604 for (x
= 0; x
< num_gh
; x
++)
1605 gfs2_glock_dq(&ghs
[x
]);
1609 * gfs2_glock_dq_uninit_m - release multiple glocks
1610 * @num_gh: the number of structures
1611 * @ghs: an array of struct gfs2_holder structures
1615 void gfs2_glock_dq_uninit_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1619 for (x
= 0; x
< num_gh
; x
++)
1620 gfs2_glock_dq_uninit(&ghs
[x
]);
1624 * gfs2_glock_prefetch_num - prefetch a glock based on lock number
1625 * @sdp: the filesystem
1626 * @number: the lock number
1627 * @glops: the glock operations for the type of glock
1628 * @state: the state to acquire the glock in
1629 * @flags: modifier flags for the aquisition
1634 void gfs2_glock_prefetch_num(struct gfs2_sbd
*sdp
, uint64_t number
,
1635 struct gfs2_glock_operations
*glops
,
1636 unsigned int state
, int flags
)
1638 struct gfs2_glock
*gl
;
1641 if (atomic_read(&sdp
->sd_reclaim_count
) <
1642 gfs2_tune_get(sdp
, gt_reclaim_limit
)) {
1643 error
= gfs2_glock_get(sdp
, number
, glops
, CREATE
, &gl
);
1645 gfs2_glock_prefetch(gl
, state
, flags
);
1652 * gfs2_lvb_hold - attach a LVB from a glock
1653 * @gl: The glock in question
1657 int gfs2_lvb_hold(struct gfs2_glock
*gl
)
1661 gfs2_glmutex_lock(gl
);
1663 if (!atomic_read(&gl
->gl_lvb_count
)) {
1664 error
= gfs2_lm_hold_lvb(gl
->gl_sbd
, gl
->gl_lock
, &gl
->gl_lvb
);
1666 gfs2_glmutex_unlock(gl
);
1669 gfs2_glock_hold(gl
);
1671 atomic_inc(&gl
->gl_lvb_count
);
1673 gfs2_glmutex_unlock(gl
);
1679 * gfs2_lvb_unhold - detach a LVB from a glock
1680 * @gl: The glock in question
1684 void gfs2_lvb_unhold(struct gfs2_glock
*gl
)
1686 gfs2_glock_hold(gl
);
1687 gfs2_glmutex_lock(gl
);
1689 gfs2_assert(gl
->gl_sbd
, atomic_read(&gl
->gl_lvb_count
) > 0);
1690 if (atomic_dec_and_test(&gl
->gl_lvb_count
)) {
1691 gfs2_lm_unhold_lvb(gl
->gl_sbd
, gl
->gl_lock
, gl
->gl_lvb
);
1696 gfs2_glmutex_unlock(gl
);
1700 void gfs2_lvb_sync(struct gfs2_glock
*gl
)
1702 gfs2_glmutex_lock(gl
);
1704 gfs2_assert(gl
->gl_sbd
, atomic_read(&gl
->gl_lvb_count
));
1705 if (!gfs2_assert_warn(gl
->gl_sbd
, gfs2_glock_is_held_excl(gl
)))
1706 gfs2_lm_sync_lvb(gl
->gl_sbd
, gl
->gl_lock
, gl
->gl_lvb
);
1708 gfs2_glmutex_unlock(gl
);
1711 static void blocking_cb(struct gfs2_sbd
*sdp
, struct lm_lockname
*name
,
1714 struct gfs2_glock
*gl
;
1716 gl
= gfs2_glock_find(sdp
, name
);
1720 if (gl
->gl_ops
->go_callback
)
1721 gl
->gl_ops
->go_callback(gl
, state
);
1722 handle_callback(gl
, state
);
1724 spin_lock(&gl
->gl_spin
);
1726 spin_unlock(&gl
->gl_spin
);
1732 * gfs2_glock_cb - Callback used by locking module
1733 * @fsdata: Pointer to the superblock
1734 * @type: Type of callback
1735 * @data: Type dependent data pointer
1737 * Called by the locking module when it wants to tell us something.
1738 * Either we need to drop a lock, one of our ASYNC requests completed, or
1739 * a journal from another client needs to be recovered.
1742 void gfs2_glock_cb(lm_fsdata_t
*fsdata
, unsigned int type
, void *data
)
1744 struct gfs2_sbd
*sdp
= (struct gfs2_sbd
*)fsdata
;
1748 blocking_cb(sdp
, data
, LM_ST_UNLOCKED
);
1752 blocking_cb(sdp
, data
, LM_ST_DEFERRED
);
1756 blocking_cb(sdp
, data
, LM_ST_SHARED
);
1760 struct lm_async_cb
*async
= data
;
1761 struct gfs2_glock
*gl
;
1763 gl
= gfs2_glock_find(sdp
, &async
->lc_name
);
1764 if (gfs2_assert_warn(sdp
, gl
))
1766 if (!gfs2_assert_warn(sdp
, gl
->gl_req_bh
))
1767 gl
->gl_req_bh(gl
, async
->lc_ret
);
1772 case LM_CB_NEED_RECOVERY
:
1773 gfs2_jdesc_make_dirty(sdp
, *(unsigned int *)data
);
1774 if (sdp
->sd_recoverd_process
)
1775 wake_up_process(sdp
->sd_recoverd_process
);
1778 case LM_CB_DROPLOCKS
:
1779 gfs2_gl_hash_clear(sdp
, NO_WAIT
);
1780 gfs2_quota_scan(sdp
);
1784 gfs2_assert_warn(sdp
, 0);
1790 * gfs2_try_toss_inode - try to remove a particular inode struct from cache
1791 * sdp: the filesystem
1792 * inum: the inode number
1796 void gfs2_try_toss_inode(struct gfs2_sbd
*sdp
, struct gfs2_inum
*inum
)
1798 struct gfs2_glock
*gl
;
1799 struct gfs2_inode
*ip
;
1802 error
= gfs2_glock_get(sdp
, inum
->no_addr
, &gfs2_inode_glops
,
1807 if (!gfs2_glmutex_trylock(gl
))
1814 if (atomic_read(&ip
->i_count
))
1817 gfs2_inode_destroy(ip
);
1820 gfs2_glmutex_unlock(gl
);
1827 * gfs2_iopen_go_callback - Try to kick the inode/vnode associated with an
1828 * iopen glock from memory
1829 * @io_gl: the iopen glock
1830 * @state: the state into which the glock should be put
1834 void gfs2_iopen_go_callback(struct gfs2_glock
*io_gl
, unsigned int state
)
1836 struct gfs2_glock
*i_gl
;
1838 if (state
!= LM_ST_UNLOCKED
)
1841 spin_lock(&io_gl
->gl_spin
);
1842 i_gl
= io_gl
->gl_object
;
1844 gfs2_glock_hold(i_gl
);
1845 spin_unlock(&io_gl
->gl_spin
);
1847 spin_unlock(&io_gl
->gl_spin
);
1851 if (gfs2_glmutex_trylock(i_gl
)) {
1852 struct gfs2_inode
*ip
= i_gl
->gl_object
;
1854 gfs2_try_toss_vnode(ip
);
1855 gfs2_glmutex_unlock(i_gl
);
1856 gfs2_glock_schedule_for_reclaim(i_gl
);
1859 gfs2_glmutex_unlock(i_gl
);
1863 gfs2_glock_put(i_gl
);
1867 * demote_ok - Check to see if it's ok to unlock a glock
1870 * Returns: 1 if it's ok
1873 static int demote_ok(struct gfs2_glock
*gl
)
1875 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
1876 struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1879 if (test_bit(GLF_STICKY
, &gl
->gl_flags
))
1881 else if (test_bit(GLF_PREFETCH
, &gl
->gl_flags
))
1882 demote
= time_after_eq(jiffies
,
1884 gfs2_tune_get(sdp
, gt_prefetch_secs
) * HZ
);
1885 else if (glops
->go_demote_ok
)
1886 demote
= glops
->go_demote_ok(gl
);
1892 * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1897 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock
*gl
)
1899 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
1901 spin_lock(&sdp
->sd_reclaim_lock
);
1902 if (list_empty(&gl
->gl_reclaim
)) {
1903 gfs2_glock_hold(gl
);
1904 list_add(&gl
->gl_reclaim
, &sdp
->sd_reclaim_list
);
1905 atomic_inc(&sdp
->sd_reclaim_count
);
1907 spin_unlock(&sdp
->sd_reclaim_lock
);
1909 wake_up(&sdp
->sd_reclaim_wq
);
1913 * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1914 * @sdp: the filesystem
1916 * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1917 * different glock and we notice that there are a lot of glocks in the
1922 void gfs2_reclaim_glock(struct gfs2_sbd
*sdp
)
1924 struct gfs2_glock
*gl
;
1926 spin_lock(&sdp
->sd_reclaim_lock
);
1927 if (list_empty(&sdp
->sd_reclaim_list
)) {
1928 spin_unlock(&sdp
->sd_reclaim_lock
);
1931 gl
= list_entry(sdp
->sd_reclaim_list
.next
,
1932 struct gfs2_glock
, gl_reclaim
);
1933 list_del_init(&gl
->gl_reclaim
);
1934 spin_unlock(&sdp
->sd_reclaim_lock
);
1936 atomic_dec(&sdp
->sd_reclaim_count
);
1937 atomic_inc(&sdp
->sd_reclaimed
);
1939 if (gfs2_glmutex_trylock(gl
)) {
1940 if (gl
->gl_ops
== &gfs2_inode_glops
) {
1941 struct gfs2_inode
*ip
= gl
->gl_object
;
1942 if (ip
&& !atomic_read(&ip
->i_count
))
1943 gfs2_inode_destroy(ip
);
1945 if (queue_empty(gl
, &gl
->gl_holders
) &&
1946 gl
->gl_state
!= LM_ST_UNLOCKED
&&
1948 handle_callback(gl
, LM_ST_UNLOCKED
);
1949 gfs2_glmutex_unlock(gl
);
1956 * examine_bucket - Call a function for glock in a hash bucket
1957 * @examiner: the function
1958 * @sdp: the filesystem
1959 * @bucket: the bucket
1961 * Returns: 1 if the bucket has entries
1964 static int examine_bucket(glock_examiner examiner
, struct gfs2_sbd
*sdp
,
1965 struct gfs2_gl_hash_bucket
*bucket
)
1967 struct glock_plug plug
;
1968 struct list_head
*tmp
;
1969 struct gfs2_glock
*gl
;
1972 /* Add "plug" to end of bucket list, work back up list from there */
1973 memset(&plug
.gl_flags
, 0, sizeof(unsigned long));
1974 set_bit(GLF_PLUG
, &plug
.gl_flags
);
1976 write_lock(&bucket
->hb_lock
);
1977 list_add(&plug
.gl_list
, &bucket
->hb_list
);
1978 write_unlock(&bucket
->hb_lock
);
1981 write_lock(&bucket
->hb_lock
);
1984 tmp
= plug
.gl_list
.next
;
1986 if (tmp
== &bucket
->hb_list
) {
1987 list_del(&plug
.gl_list
);
1988 entries
= !list_empty(&bucket
->hb_list
);
1989 write_unlock(&bucket
->hb_lock
);
1992 gl
= list_entry(tmp
, struct gfs2_glock
, gl_list
);
1994 /* Move plug up list */
1995 list_move(&plug
.gl_list
, &gl
->gl_list
);
1997 if (test_bit(GLF_PLUG
, &gl
->gl_flags
))
2000 /* examiner() must glock_put() */
2001 gfs2_glock_hold(gl
);
2006 write_unlock(&bucket
->hb_lock
);
2013 * scan_glock - look at a glock and see if we can reclaim it
2014 * @gl: the glock to look at
2018 static void scan_glock(struct gfs2_glock
*gl
)
2020 if (gfs2_glmutex_trylock(gl
)) {
2021 if (gl
->gl_ops
== &gfs2_inode_glops
) {
2022 struct gfs2_inode
*ip
= gl
->gl_object
;
2023 if (ip
&& !atomic_read(&ip
->i_count
))
2026 if (queue_empty(gl
, &gl
->gl_holders
) &&
2027 gl
->gl_state
!= LM_ST_UNLOCKED
&&
2031 gfs2_glmutex_unlock(gl
);
2039 gfs2_glmutex_unlock(gl
);
2040 gfs2_glock_schedule_for_reclaim(gl
);
2045 * gfs2_scand_internal - Look for glocks and inodes to toss from memory
2046 * @sdp: the filesystem
2050 void gfs2_scand_internal(struct gfs2_sbd
*sdp
)
2054 for (x
= 0; x
< GFS2_GL_HASH_SIZE
; x
++) {
2055 examine_bucket(scan_glock
, sdp
, &sdp
->sd_gl_hash
[x
]);
2061 * clear_glock - look at a glock and see if we can free it from glock cache
2062 * @gl: the glock to look at
2066 static void clear_glock(struct gfs2_glock
*gl
)
2068 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
2071 spin_lock(&sdp
->sd_reclaim_lock
);
2072 if (!list_empty(&gl
->gl_reclaim
)) {
2073 list_del_init(&gl
->gl_reclaim
);
2074 atomic_dec(&sdp
->sd_reclaim_count
);
2075 spin_unlock(&sdp
->sd_reclaim_lock
);
2076 released
= gfs2_glock_put(gl
);
2077 gfs2_assert(sdp
, !released
);
2079 spin_unlock(&sdp
->sd_reclaim_lock
);
2082 if (gfs2_glmutex_trylock(gl
)) {
2083 if (gl
->gl_ops
== &gfs2_inode_glops
) {
2084 struct gfs2_inode
*ip
= gl
->gl_object
;
2085 if (ip
&& !atomic_read(&ip
->i_count
))
2086 gfs2_inode_destroy(ip
);
2088 if (queue_empty(gl
, &gl
->gl_holders
) &&
2089 gl
->gl_state
!= LM_ST_UNLOCKED
)
2090 handle_callback(gl
, LM_ST_UNLOCKED
);
2092 gfs2_glmutex_unlock(gl
);
2099 * gfs2_gl_hash_clear - Empty out the glock hash table
2100 * @sdp: the filesystem
2101 * @wait: wait until it's all gone
2103 * Called when unmounting the filesystem, or when inter-node lock manager
2104 * requests DROPLOCKS because it is running out of capacity.
2107 void gfs2_gl_hash_clear(struct gfs2_sbd
*sdp
, int wait
)
2118 for (x
= 0; x
< GFS2_GL_HASH_SIZE
; x
++)
2119 if (examine_bucket(clear_glock
, sdp
,
2120 &sdp
->sd_gl_hash
[x
]))
2126 if (time_after_eq(jiffies
,
2127 t
+ gfs2_tune_get(sdp
, gt_stall_secs
) * HZ
)) {
2128 fs_warn(sdp
, "Unmount seems to be stalled. "
2129 "Dumping lock state...\n");
2130 gfs2_dump_lockstate(sdp
);
2134 /* invalidate_inodes() requires that the sb inodes list
2135 not change, but an async completion callback for an
2136 unlock can occur which does glock_put() which
2137 can call iput() which will change the sb inodes list.
2138 invalidate_inodes_mutex prevents glock_put()'s during
2139 an invalidate_inodes() */
2141 mutex_lock(&sdp
->sd_invalidate_inodes_mutex
);
2142 invalidate_inodes(sdp
->sd_vfs
);
2143 mutex_unlock(&sdp
->sd_invalidate_inodes_mutex
);
2149 * Diagnostic routines to help debug distributed deadlock
2153 * dump_holder - print information about a glock holder
2154 * @str: a string naming the type of holder
2155 * @gh: the glock holder
2157 * Returns: 0 on success, -ENOBUFS when we run out of space
2160 static int dump_holder(char *str
, struct gfs2_holder
*gh
)
2163 int error
= -ENOBUFS
;
2165 printk(KERN_INFO
" %s\n", str
);
2166 printk(KERN_INFO
" owner = %ld\n",
2167 (gh
->gh_owner
) ? (long)gh
->gh_owner
->pid
: -1);
2168 printk(KERN_INFO
" gh_state = %u\n", gh
->gh_state
);
2169 printk(KERN_INFO
" gh_flags =");
2170 for (x
= 0; x
< 32; x
++)
2171 if (gh
->gh_flags
& (1 << x
))
2174 printk(KERN_INFO
" error = %d\n", gh
->gh_error
);
2175 printk(KERN_INFO
" gh_iflags =");
2176 for (x
= 0; x
< 32; x
++)
2177 if (test_bit(x
, &gh
->gh_iflags
))
2180 print_symbol(KERN_INFO
" initialized at: %s\n", gh
->gh_ip
);
2188 * dump_inode - print information about an inode
2191 * Returns: 0 on success, -ENOBUFS when we run out of space
2194 static int dump_inode(struct gfs2_inode
*ip
)
2197 int error
= -ENOBUFS
;
2199 printk(KERN_INFO
" Inode:\n");
2200 printk(KERN_INFO
" num = %llu %llu\n",
2201 ip
->i_num
.no_formal_ino
, ip
->i_num
.no_addr
);
2202 printk(KERN_INFO
" type = %u\n", IF2DT(ip
->i_di
.di_mode
));
2203 printk(KERN_INFO
" i_count = %d\n", atomic_read(&ip
->i_count
));
2204 printk(KERN_INFO
" i_flags =");
2205 for (x
= 0; x
< 32; x
++)
2206 if (test_bit(x
, &ip
->i_flags
))
2209 printk(KERN_INFO
" vnode = %s\n", (ip
->i_vnode
) ? "yes" : "no");
2217 * dump_glock - print information about a glock
2219 * @count: where we are in the buffer
2221 * Returns: 0 on success, -ENOBUFS when we run out of space
2224 static int dump_glock(struct gfs2_glock
*gl
)
2226 struct gfs2_holder
*gh
;
2228 int error
= -ENOBUFS
;
2230 spin_lock(&gl
->gl_spin
);
2232 printk(KERN_INFO
"Glock (%u, %llu)\n",
2233 gl
->gl_name
.ln_type
,
2234 gl
->gl_name
.ln_number
);
2235 printk(KERN_INFO
" gl_flags =");
2236 for (x
= 0; x
< 32; x
++)
2237 if (test_bit(x
, &gl
->gl_flags
))
2240 printk(KERN_INFO
" gl_ref = %d\n", atomic_read(&gl
->gl_ref
.refcount
));
2241 printk(KERN_INFO
" gl_state = %u\n", gl
->gl_state
);
2242 printk(KERN_INFO
" req_gh = %s\n", (gl
->gl_req_gh
) ? "yes" : "no");
2243 printk(KERN_INFO
" req_bh = %s\n", (gl
->gl_req_bh
) ? "yes" : "no");
2244 printk(KERN_INFO
" lvb_count = %d\n", atomic_read(&gl
->gl_lvb_count
));
2245 printk(KERN_INFO
" object = %s\n", (gl
->gl_object
) ? "yes" : "no");
2246 printk(KERN_INFO
" le = %s\n",
2247 (list_empty(&gl
->gl_le
.le_list
)) ? "no" : "yes");
2248 printk(KERN_INFO
" reclaim = %s\n",
2249 (list_empty(&gl
->gl_reclaim
)) ? "no" : "yes");
2251 printk(KERN_INFO
" aspace = %lu\n",
2252 gl
->gl_aspace
->i_mapping
->nrpages
);
2254 printk(KERN_INFO
" aspace = no\n");
2255 printk(KERN_INFO
" ail = %d\n", atomic_read(&gl
->gl_ail_count
));
2256 if (gl
->gl_req_gh
) {
2257 error
= dump_holder("Request", gl
->gl_req_gh
);
2261 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
) {
2262 error
= dump_holder("Holder", gh
);
2266 list_for_each_entry(gh
, &gl
->gl_waiters1
, gh_list
) {
2267 error
= dump_holder("Waiter1", gh
);
2271 list_for_each_entry(gh
, &gl
->gl_waiters2
, gh_list
) {
2272 error
= dump_holder("Waiter2", gh
);
2276 list_for_each_entry(gh
, &gl
->gl_waiters3
, gh_list
) {
2277 error
= dump_holder("Waiter3", gh
);
2281 if (gl
->gl_ops
== &gfs2_inode_glops
&& gl
->gl_object
) {
2282 if (!test_bit(GLF_LOCK
, &gl
->gl_flags
) &&
2283 list_empty(&gl
->gl_holders
)) {
2284 error
= dump_inode(gl
->gl_object
);
2289 printk(KERN_INFO
" Inode: busy\n");
2296 spin_unlock(&gl
->gl_spin
);
2302 * gfs2_dump_lockstate - print out the current lockstate
2303 * @sdp: the filesystem
2304 * @ub: the buffer to copy the information into
2306 * If @ub is NULL, dump the lockstate to the console.
2310 int gfs2_dump_lockstate(struct gfs2_sbd
*sdp
)
2312 struct gfs2_gl_hash_bucket
*bucket
;
2313 struct gfs2_glock
*gl
;
2317 for (x
= 0; x
< GFS2_GL_HASH_SIZE
; x
++) {
2318 bucket
= &sdp
->sd_gl_hash
[x
];
2320 read_lock(&bucket
->hb_lock
);
2322 list_for_each_entry(gl
, &bucket
->hb_list
, gl_list
) {
2323 if (test_bit(GLF_PLUG
, &gl
->gl_flags
))
2326 error
= dump_glock(gl
);
2331 read_unlock(&bucket
->hb_lock
);