2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
10 #include <linux/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/completion.h>
14 #include <linux/buffer_head.h>
15 #include <linux/delay.h>
16 #include <linux/sort.h>
17 #include <linux/jhash.h>
18 #include <linux/kallsyms.h>
19 #include <linux/gfs2_ondisk.h>
20 #include <linux/list.h>
21 #include <linux/lm_interface.h>
22 #include <linux/wait.h>
23 #include <linux/module.h>
24 #include <linux/rwsem.h>
25 #include <asm/uaccess.h>
26 #include <linux/seq_file.h>
27 #include <linux/debugfs.h>
28 #include <linux/kthread.h>
29 #include <linux/freezer.h>
30 #include <linux/workqueue.h>
31 #include <linux/jiffies.h>
45 struct gfs2_gl_hash_bucket
{
46 struct hlist_head hb_list
;
50 int hash
; /* hash bucket index */
51 struct gfs2_sbd
*sdp
; /* incore superblock */
52 struct gfs2_glock
*gl
; /* current glock struct */
53 struct seq_file
*seq
; /* sequence file for debugfs */
54 char string
[512]; /* scratch space */
57 typedef void (*glock_examiner
) (struct gfs2_glock
* gl
);
59 static int gfs2_dump_lockstate(struct gfs2_sbd
*sdp
);
60 static int dump_glock(struct glock_iter
*gi
, struct gfs2_glock
*gl
);
61 static void gfs2_glock_xmote_th(struct gfs2_glock
*gl
, struct gfs2_holder
*gh
);
62 static void gfs2_glock_drop_th(struct gfs2_glock
*gl
);
63 static void run_queue(struct gfs2_glock
*gl
);
65 static DECLARE_RWSEM(gfs2_umount_flush_sem
);
66 static struct dentry
*gfs2_root
;
67 static struct task_struct
*scand_process
;
68 static unsigned int scand_secs
= 5;
69 static struct workqueue_struct
*glock_workqueue
;
71 #define GFS2_GL_HASH_SHIFT 15
72 #define GFS2_GL_HASH_SIZE (1 << GFS2_GL_HASH_SHIFT)
73 #define GFS2_GL_HASH_MASK (GFS2_GL_HASH_SIZE - 1)
75 static struct gfs2_gl_hash_bucket gl_hash_table
[GFS2_GL_HASH_SIZE
];
76 static struct dentry
*gfs2_root
;
79 * Despite what you might think, the numbers below are not arbitrary :-)
80 * They are taken from the ipv4 routing hash code, which is well tested
81 * and thus should be nearly optimal. Later on we might tweek the numbers
82 * but for now this should be fine.
84 * The reason for putting the locks in a separate array from the list heads
85 * is that we can have fewer locks than list heads and save memory. We use
86 * the same hash function for both, but with a different hash mask.
88 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
89 defined(CONFIG_PROVE_LOCKING)
92 # define GL_HASH_LOCK_SZ 256
95 # define GL_HASH_LOCK_SZ 4096
97 # define GL_HASH_LOCK_SZ 2048
99 # define GL_HASH_LOCK_SZ 1024
101 # define GL_HASH_LOCK_SZ 512
103 # define GL_HASH_LOCK_SZ 256
107 /* We never want more locks than chains */
108 #if GFS2_GL_HASH_SIZE < GL_HASH_LOCK_SZ
109 # undef GL_HASH_LOCK_SZ
110 # define GL_HASH_LOCK_SZ GFS2_GL_HASH_SIZE
113 static rwlock_t gl_hash_locks
[GL_HASH_LOCK_SZ
];
115 static inline rwlock_t
*gl_lock_addr(unsigned int x
)
117 return &gl_hash_locks
[x
& (GL_HASH_LOCK_SZ
-1)];
119 #else /* not SMP, so no spinlocks required */
120 static inline rwlock_t
*gl_lock_addr(unsigned int x
)
127 * relaxed_state_ok - is a requested lock compatible with the current lock mode?
128 * @actual: the current state of the lock
129 * @requested: the lock state that was requested by the caller
130 * @flags: the modifier flags passed in by the caller
132 * Returns: 1 if the locks are compatible, 0 otherwise
135 static inline int relaxed_state_ok(unsigned int actual
, unsigned requested
,
138 if (actual
== requested
)
141 if (flags
& GL_EXACT
)
144 if (actual
== LM_ST_EXCLUSIVE
&& requested
== LM_ST_SHARED
)
147 if (actual
!= LM_ST_UNLOCKED
&& (flags
& LM_FLAG_ANY
))
154 * gl_hash() - Turn glock number into hash bucket number
155 * @lock: The glock number
157 * Returns: The number of the corresponding hash bucket
160 static unsigned int gl_hash(const struct gfs2_sbd
*sdp
,
161 const struct lm_lockname
*name
)
165 h
= jhash(&name
->ln_number
, sizeof(u64
), 0);
166 h
= jhash(&name
->ln_type
, sizeof(unsigned int), h
);
167 h
= jhash(&sdp
, sizeof(struct gfs2_sbd
*), h
);
168 h
&= GFS2_GL_HASH_MASK
;
174 * glock_free() - Perform a few checks and then release struct gfs2_glock
175 * @gl: The glock to release
177 * Also calls lock module to release its internal structure for this glock.
181 static void glock_free(struct gfs2_glock
*gl
)
183 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
184 struct inode
*aspace
= gl
->gl_aspace
;
186 gfs2_lm_put_lock(sdp
, gl
->gl_lock
);
189 gfs2_aspace_put(aspace
);
191 kmem_cache_free(gfs2_glock_cachep
, gl
);
195 * gfs2_glock_hold() - increment reference count on glock
196 * @gl: The glock to hold
200 void gfs2_glock_hold(struct gfs2_glock
*gl
)
202 atomic_inc(&gl
->gl_ref
);
206 * gfs2_glock_put() - Decrement reference count on glock
207 * @gl: The glock to put
211 int gfs2_glock_put(struct gfs2_glock
*gl
)
214 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
216 write_lock(gl_lock_addr(gl
->gl_hash
));
217 if (atomic_dec_and_test(&gl
->gl_ref
)) {
218 hlist_del(&gl
->gl_list
);
219 write_unlock(gl_lock_addr(gl
->gl_hash
));
220 BUG_ON(spin_is_locked(&gl
->gl_spin
));
221 gfs2_assert(sdp
, gl
->gl_state
== LM_ST_UNLOCKED
);
222 gfs2_assert(sdp
, list_empty(&gl
->gl_reclaim
));
223 gfs2_assert(sdp
, list_empty(&gl
->gl_holders
));
224 gfs2_assert(sdp
, list_empty(&gl
->gl_waiters1
));
225 gfs2_assert(sdp
, list_empty(&gl
->gl_waiters3
));
230 write_unlock(gl_lock_addr(gl
->gl_hash
));
236 * search_bucket() - Find struct gfs2_glock by lock number
237 * @bucket: the bucket to search
238 * @name: The lock name
240 * Returns: NULL, or the struct gfs2_glock with the requested number
243 static struct gfs2_glock
*search_bucket(unsigned int hash
,
244 const struct gfs2_sbd
*sdp
,
245 const struct lm_lockname
*name
)
247 struct gfs2_glock
*gl
;
248 struct hlist_node
*h
;
250 hlist_for_each_entry(gl
, h
, &gl_hash_table
[hash
].hb_list
, gl_list
) {
251 if (!lm_name_equal(&gl
->gl_name
, name
))
253 if (gl
->gl_sbd
!= sdp
)
256 atomic_inc(&gl
->gl_ref
);
265 * gfs2_glock_find() - Find glock by lock number
266 * @sdp: The GFS2 superblock
267 * @name: The lock name
269 * Returns: NULL, or the struct gfs2_glock with the requested number
272 static struct gfs2_glock
*gfs2_glock_find(const struct gfs2_sbd
*sdp
,
273 const struct lm_lockname
*name
)
275 unsigned int hash
= gl_hash(sdp
, name
);
276 struct gfs2_glock
*gl
;
278 read_lock(gl_lock_addr(hash
));
279 gl
= search_bucket(hash
, sdp
, name
);
280 read_unlock(gl_lock_addr(hash
));
285 static void glock_work_func(struct work_struct
*work
)
287 struct gfs2_glock
*gl
= container_of(work
, struct gfs2_glock
, gl_work
.work
);
289 spin_lock(&gl
->gl_spin
);
290 if (test_and_clear_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
))
291 set_bit(GLF_DEMOTE
, &gl
->gl_flags
);
293 spin_unlock(&gl
->gl_spin
);
298 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
299 * @sdp: The GFS2 superblock
300 * @number: the lock number
301 * @glops: The glock_operations to use
302 * @create: If 0, don't create the glock if it doesn't exist
303 * @glp: the glock is returned here
305 * This does not lock a glock, just finds/creates structures for one.
310 int gfs2_glock_get(struct gfs2_sbd
*sdp
, u64 number
,
311 const struct gfs2_glock_operations
*glops
, int create
,
312 struct gfs2_glock
**glp
)
314 struct lm_lockname name
= { .ln_number
= number
, .ln_type
= glops
->go_type
};
315 struct gfs2_glock
*gl
, *tmp
;
316 unsigned int hash
= gl_hash(sdp
, &name
);
319 read_lock(gl_lock_addr(hash
));
320 gl
= search_bucket(hash
, sdp
, &name
);
321 read_unlock(gl_lock_addr(hash
));
328 gl
= kmem_cache_alloc(gfs2_glock_cachep
, GFP_KERNEL
);
334 atomic_set(&gl
->gl_ref
, 1);
335 gl
->gl_state
= LM_ST_UNLOCKED
;
336 gl
->gl_demote_state
= LM_ST_EXCLUSIVE
;
338 gl
->gl_owner_pid
= 0;
341 gl
->gl_req_gh
= NULL
;
342 gl
->gl_req_bh
= NULL
;
344 gl
->gl_stamp
= jiffies
;
345 gl
->gl_tchange
= jiffies
;
346 gl
->gl_object
= NULL
;
348 gl
->gl_aspace
= NULL
;
349 lops_init_le(&gl
->gl_le
, &gfs2_glock_lops
);
350 INIT_DELAYED_WORK(&gl
->gl_work
, glock_work_func
);
352 /* If this glock protects actual on-disk data or metadata blocks,
353 create a VFS inode to manage the pages/buffers holding them. */
354 if (glops
== &gfs2_inode_glops
|| glops
== &gfs2_rgrp_glops
) {
355 gl
->gl_aspace
= gfs2_aspace_get(sdp
);
356 if (!gl
->gl_aspace
) {
362 error
= gfs2_lm_get_lock(sdp
, &name
, &gl
->gl_lock
);
366 write_lock(gl_lock_addr(hash
));
367 tmp
= search_bucket(hash
, sdp
, &name
);
369 write_unlock(gl_lock_addr(hash
));
373 hlist_add_head(&gl
->gl_list
, &gl_hash_table
[hash
].hb_list
);
374 write_unlock(gl_lock_addr(hash
));
383 gfs2_aspace_put(gl
->gl_aspace
);
385 kmem_cache_free(gfs2_glock_cachep
, gl
);
390 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
392 * @state: the state we're requesting
393 * @flags: the modifier flags
394 * @gh: the holder structure
398 void gfs2_holder_init(struct gfs2_glock
*gl
, unsigned int state
, unsigned flags
,
399 struct gfs2_holder
*gh
)
401 INIT_LIST_HEAD(&gh
->gh_list
);
403 gh
->gh_ip
= (unsigned long)__builtin_return_address(0);
404 gh
->gh_owner_pid
= current
->pid
;
405 gh
->gh_state
= state
;
406 gh
->gh_flags
= flags
;
413 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
414 * @state: the state we're requesting
415 * @flags: the modifier flags
416 * @gh: the holder structure
418 * Don't mess with the glock.
422 void gfs2_holder_reinit(unsigned int state
, unsigned flags
, struct gfs2_holder
*gh
)
424 gh
->gh_state
= state
;
425 gh
->gh_flags
= flags
;
427 gh
->gh_ip
= (unsigned long)__builtin_return_address(0);
431 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
432 * @gh: the holder structure
436 void gfs2_holder_uninit(struct gfs2_holder
*gh
)
438 gfs2_glock_put(gh
->gh_gl
);
443 static void gfs2_holder_wake(struct gfs2_holder
*gh
)
445 clear_bit(HIF_WAIT
, &gh
->gh_iflags
);
446 smp_mb__after_clear_bit();
447 wake_up_bit(&gh
->gh_iflags
, HIF_WAIT
);
450 static int just_schedule(void *word
)
456 static void wait_on_holder(struct gfs2_holder
*gh
)
459 wait_on_bit(&gh
->gh_iflags
, HIF_WAIT
, just_schedule
, TASK_UNINTERRUPTIBLE
);
462 static void gfs2_demote_wake(struct gfs2_glock
*gl
)
464 BUG_ON(!spin_is_locked(&gl
->gl_spin
));
465 gl
->gl_demote_state
= LM_ST_EXCLUSIVE
;
466 clear_bit(GLF_DEMOTE
, &gl
->gl_flags
);
467 smp_mb__after_clear_bit();
468 wake_up_bit(&gl
->gl_flags
, GLF_DEMOTE
);
471 static void wait_on_demote(struct gfs2_glock
*gl
)
474 wait_on_bit(&gl
->gl_flags
, GLF_DEMOTE
, just_schedule
, TASK_UNINTERRUPTIBLE
);
478 * rq_mutex - process a mutex request in the queue
479 * @gh: the glock holder
481 * Returns: 1 if the queue is blocked
484 static int rq_mutex(struct gfs2_holder
*gh
)
486 struct gfs2_glock
*gl
= gh
->gh_gl
;
488 list_del_init(&gh
->gh_list
);
489 /* gh->gh_error never examined. */
490 set_bit(GLF_LOCK
, &gl
->gl_flags
);
491 clear_bit(HIF_WAIT
, &gh
->gh_iflags
);
493 wake_up_bit(&gh
->gh_iflags
, HIF_WAIT
);
499 * rq_promote - process a promote request in the queue
500 * @gh: the glock holder
502 * Acquire a new inter-node lock, or change a lock state to more restrictive.
504 * Returns: 1 if the queue is blocked
507 static int rq_promote(struct gfs2_holder
*gh
)
509 struct gfs2_glock
*gl
= gh
->gh_gl
;
510 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
512 if (!relaxed_state_ok(gl
->gl_state
, gh
->gh_state
, gh
->gh_flags
)) {
513 if (list_empty(&gl
->gl_holders
)) {
515 set_bit(GLF_LOCK
, &gl
->gl_flags
);
516 spin_unlock(&gl
->gl_spin
);
518 if (atomic_read(&sdp
->sd_reclaim_count
) >
519 gfs2_tune_get(sdp
, gt_reclaim_limit
) &&
520 !(gh
->gh_flags
& LM_FLAG_PRIORITY
)) {
521 gfs2_reclaim_glock(sdp
);
522 gfs2_reclaim_glock(sdp
);
525 gfs2_glock_xmote_th(gh
->gh_gl
, gh
);
526 spin_lock(&gl
->gl_spin
);
531 if (list_empty(&gl
->gl_holders
)) {
532 set_bit(HIF_FIRST
, &gh
->gh_iflags
);
533 set_bit(GLF_LOCK
, &gl
->gl_flags
);
535 struct gfs2_holder
*next_gh
;
536 if (gh
->gh_state
== LM_ST_EXCLUSIVE
)
538 next_gh
= list_entry(gl
->gl_holders
.next
, struct gfs2_holder
,
540 if (next_gh
->gh_state
== LM_ST_EXCLUSIVE
)
544 list_move_tail(&gh
->gh_list
, &gl
->gl_holders
);
546 set_bit(HIF_HOLDER
, &gh
->gh_iflags
);
548 gfs2_holder_wake(gh
);
554 * rq_demote - process a demote request in the queue
555 * @gh: the glock holder
557 * Returns: 1 if the queue is blocked
560 static int rq_demote(struct gfs2_glock
*gl
)
562 if (!list_empty(&gl
->gl_holders
))
565 if (gl
->gl_state
== gl
->gl_demote_state
||
566 gl
->gl_state
== LM_ST_UNLOCKED
) {
567 gfs2_demote_wake(gl
);
570 set_bit(GLF_LOCK
, &gl
->gl_flags
);
571 if (gl
->gl_demote_state
== LM_ST_UNLOCKED
||
572 gl
->gl_state
!= LM_ST_EXCLUSIVE
) {
573 spin_unlock(&gl
->gl_spin
);
574 gfs2_glock_drop_th(gl
);
576 spin_unlock(&gl
->gl_spin
);
577 gfs2_glock_xmote_th(gl
, NULL
);
579 spin_lock(&gl
->gl_spin
);
585 * run_queue - process holder structures on a glock
589 static void run_queue(struct gfs2_glock
*gl
)
591 struct gfs2_holder
*gh
;
595 if (test_bit(GLF_LOCK
, &gl
->gl_flags
))
598 if (!list_empty(&gl
->gl_waiters1
)) {
599 gh
= list_entry(gl
->gl_waiters1
.next
,
600 struct gfs2_holder
, gh_list
);
602 if (test_bit(HIF_MUTEX
, &gh
->gh_iflags
))
603 blocked
= rq_mutex(gh
);
605 gfs2_assert_warn(gl
->gl_sbd
, 0);
607 } else if (test_bit(GLF_DEMOTE
, &gl
->gl_flags
)) {
608 blocked
= rq_demote(gl
);
609 } else if (!list_empty(&gl
->gl_waiters3
)) {
610 gh
= list_entry(gl
->gl_waiters3
.next
,
611 struct gfs2_holder
, gh_list
);
613 if (test_bit(HIF_PROMOTE
, &gh
->gh_iflags
))
614 blocked
= rq_promote(gh
);
616 gfs2_assert_warn(gl
->gl_sbd
, 0);
627 * gfs2_glmutex_lock - acquire a local lock on a glock
630 * Gives caller exclusive access to manipulate a glock structure.
633 static void gfs2_glmutex_lock(struct gfs2_glock
*gl
)
635 struct gfs2_holder gh
;
637 gfs2_holder_init(gl
, 0, 0, &gh
);
638 set_bit(HIF_MUTEX
, &gh
.gh_iflags
);
639 if (test_and_set_bit(HIF_WAIT
, &gh
.gh_iflags
))
642 spin_lock(&gl
->gl_spin
);
643 if (test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
)) {
644 list_add_tail(&gh
.gh_list
, &gl
->gl_waiters1
);
646 gl
->gl_owner_pid
= current
->pid
;
647 gl
->gl_ip
= (unsigned long)__builtin_return_address(0);
648 clear_bit(HIF_WAIT
, &gh
.gh_iflags
);
650 wake_up_bit(&gh
.gh_iflags
, HIF_WAIT
);
652 spin_unlock(&gl
->gl_spin
);
655 gfs2_holder_uninit(&gh
);
659 * gfs2_glmutex_trylock - try to acquire a local lock on a glock
662 * Returns: 1 if the glock is acquired
665 static int gfs2_glmutex_trylock(struct gfs2_glock
*gl
)
669 spin_lock(&gl
->gl_spin
);
670 if (test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
)) {
673 gl
->gl_owner_pid
= current
->pid
;
674 gl
->gl_ip
= (unsigned long)__builtin_return_address(0);
676 spin_unlock(&gl
->gl_spin
);
682 * gfs2_glmutex_unlock - release a local lock on a glock
687 static void gfs2_glmutex_unlock(struct gfs2_glock
*gl
)
689 spin_lock(&gl
->gl_spin
);
690 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
691 gl
->gl_owner_pid
= 0;
694 BUG_ON(!spin_is_locked(&gl
->gl_spin
));
695 spin_unlock(&gl
->gl_spin
);
699 * handle_callback - process a demote request
701 * @state: the state the caller wants us to change to
703 * There are only two requests that we are going to see in actual
704 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
707 static void handle_callback(struct gfs2_glock
*gl
, unsigned int state
,
708 int remote
, unsigned long delay
)
710 int bit
= delay
? GLF_PENDING_DEMOTE
: GLF_DEMOTE
;
712 spin_lock(&gl
->gl_spin
);
713 set_bit(bit
, &gl
->gl_flags
);
714 if (gl
->gl_demote_state
== LM_ST_EXCLUSIVE
) {
715 gl
->gl_demote_state
= state
;
716 gl
->gl_demote_time
= jiffies
;
717 if (remote
&& gl
->gl_ops
->go_type
== LM_TYPE_IOPEN
&&
719 struct inode
*inode
= igrab(gl
->gl_object
);
720 spin_unlock(&gl
->gl_spin
);
722 d_prune_aliases(inode
);
727 } else if (gl
->gl_demote_state
!= LM_ST_UNLOCKED
&&
728 gl
->gl_demote_state
!= state
) {
729 gl
->gl_demote_state
= LM_ST_UNLOCKED
;
731 spin_unlock(&gl
->gl_spin
);
735 * state_change - record that the glock is now in a different state
737 * @new_state the new state
741 static void state_change(struct gfs2_glock
*gl
, unsigned int new_state
)
745 held1
= (gl
->gl_state
!= LM_ST_UNLOCKED
);
746 held2
= (new_state
!= LM_ST_UNLOCKED
);
748 if (held1
!= held2
) {
755 gl
->gl_state
= new_state
;
756 gl
->gl_tchange
= jiffies
;
760 * xmote_bh - Called after the lock module is done acquiring a lock
761 * @gl: The glock in question
762 * @ret: the int returned from the lock module
766 static void xmote_bh(struct gfs2_glock
*gl
, unsigned int ret
)
768 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
769 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
770 struct gfs2_holder
*gh
= gl
->gl_req_gh
;
771 int prev_state
= gl
->gl_state
;
774 gfs2_assert_warn(sdp
, test_bit(GLF_LOCK
, &gl
->gl_flags
));
775 gfs2_assert_warn(sdp
, list_empty(&gl
->gl_holders
));
776 gfs2_assert_warn(sdp
, !(ret
& LM_OUT_ASYNC
));
778 state_change(gl
, ret
& LM_OUT_ST_MASK
);
780 if (prev_state
!= LM_ST_UNLOCKED
&& !(ret
& LM_OUT_CACHEABLE
)) {
782 glops
->go_inval(gl
, DIO_METADATA
);
783 } else if (gl
->gl_state
== LM_ST_DEFERRED
) {
784 /* We might not want to do this here.
785 Look at moving to the inode glops. */
787 glops
->go_inval(gl
, 0);
790 /* Deal with each possible exit condition */
793 gl
->gl_stamp
= jiffies
;
794 if (ret
& LM_OUT_CANCELED
) {
797 spin_lock(&gl
->gl_spin
);
798 if (gl
->gl_state
!= gl
->gl_demote_state
) {
799 gl
->gl_req_bh
= NULL
;
800 spin_unlock(&gl
->gl_spin
);
801 gfs2_glock_drop_th(gl
);
805 gfs2_demote_wake(gl
);
806 spin_unlock(&gl
->gl_spin
);
809 spin_lock(&gl
->gl_spin
);
810 list_del_init(&gh
->gh_list
);
812 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)))
814 gh
->gh_error
= GLR_CANCELED
;
815 if (ret
& LM_OUT_CANCELED
)
817 if (relaxed_state_ok(gl
->gl_state
, gh
->gh_state
, gh
->gh_flags
)) {
818 list_add_tail(&gh
->gh_list
, &gl
->gl_holders
);
820 set_bit(HIF_HOLDER
, &gh
->gh_iflags
);
821 set_bit(HIF_FIRST
, &gh
->gh_iflags
);
825 gh
->gh_error
= GLR_TRYFAILED
;
826 if (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
))
828 gh
->gh_error
= -EINVAL
;
829 if (gfs2_assert_withdraw(sdp
, 0) == -1)
830 fs_err(sdp
, "ret = 0x%.8X\n", ret
);
832 spin_unlock(&gl
->gl_spin
);
835 if (glops
->go_xmote_bh
)
836 glops
->go_xmote_bh(gl
);
839 spin_lock(&gl
->gl_spin
);
840 gl
->gl_req_gh
= NULL
;
841 gl
->gl_req_bh
= NULL
;
842 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
843 spin_unlock(&gl
->gl_spin
);
849 gfs2_holder_wake(gh
);
853 * gfs2_glock_xmote_th - Call into the lock module to acquire or change a glock
854 * @gl: The glock in question
855 * @state: the requested state
856 * @flags: modifier flags to the lock call
860 static void gfs2_glock_xmote_th(struct gfs2_glock
*gl
, struct gfs2_holder
*gh
)
862 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
863 int flags
= gh
? gh
->gh_flags
: 0;
864 unsigned state
= gh
? gh
->gh_state
: gl
->gl_demote_state
;
865 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
866 int lck_flags
= flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
|
867 LM_FLAG_NOEXP
| LM_FLAG_ANY
|
869 unsigned int lck_ret
;
871 if (glops
->go_xmote_th
)
872 glops
->go_xmote_th(gl
);
874 gfs2_assert_warn(sdp
, test_bit(GLF_LOCK
, &gl
->gl_flags
));
875 gfs2_assert_warn(sdp
, list_empty(&gl
->gl_holders
));
876 gfs2_assert_warn(sdp
, state
!= LM_ST_UNLOCKED
);
877 gfs2_assert_warn(sdp
, state
!= gl
->gl_state
);
880 gl
->gl_req_bh
= xmote_bh
;
882 lck_ret
= gfs2_lm_lock(sdp
, gl
->gl_lock
, gl
->gl_state
, state
, lck_flags
);
884 if (gfs2_assert_withdraw(sdp
, !(lck_ret
& LM_OUT_ERROR
)))
887 if (lck_ret
& LM_OUT_ASYNC
)
888 gfs2_assert_warn(sdp
, lck_ret
== LM_OUT_ASYNC
);
890 xmote_bh(gl
, lck_ret
);
894 * drop_bh - Called after a lock module unlock completes
896 * @ret: the return status
898 * Doesn't wake up the process waiting on the struct gfs2_holder (if any)
899 * Doesn't drop the reference on the glock the top half took out
903 static void drop_bh(struct gfs2_glock
*gl
, unsigned int ret
)
905 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
906 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
907 struct gfs2_holder
*gh
= gl
->gl_req_gh
;
909 gfs2_assert_warn(sdp
, test_bit(GLF_LOCK
, &gl
->gl_flags
));
910 gfs2_assert_warn(sdp
, list_empty(&gl
->gl_holders
));
911 gfs2_assert_warn(sdp
, !ret
);
913 state_change(gl
, LM_ST_UNLOCKED
);
916 glops
->go_inval(gl
, DIO_METADATA
);
919 spin_lock(&gl
->gl_spin
);
920 list_del_init(&gh
->gh_list
);
922 spin_unlock(&gl
->gl_spin
);
925 spin_lock(&gl
->gl_spin
);
926 gfs2_demote_wake(gl
);
927 gl
->gl_req_gh
= NULL
;
928 gl
->gl_req_bh
= NULL
;
929 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
930 spin_unlock(&gl
->gl_spin
);
935 gfs2_holder_wake(gh
);
939 * gfs2_glock_drop_th - call into the lock module to unlock a lock
944 static void gfs2_glock_drop_th(struct gfs2_glock
*gl
)
946 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
947 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
950 if (glops
->go_drop_th
)
951 glops
->go_drop_th(gl
);
953 gfs2_assert_warn(sdp
, test_bit(GLF_LOCK
, &gl
->gl_flags
));
954 gfs2_assert_warn(sdp
, list_empty(&gl
->gl_holders
));
955 gfs2_assert_warn(sdp
, gl
->gl_state
!= LM_ST_UNLOCKED
);
958 gl
->gl_req_bh
= drop_bh
;
960 ret
= gfs2_lm_unlock(sdp
, gl
->gl_lock
, gl
->gl_state
);
962 if (gfs2_assert_withdraw(sdp
, !(ret
& LM_OUT_ERROR
)))
968 gfs2_assert_warn(sdp
, ret
== LM_OUT_ASYNC
);
972 * do_cancels - cancel requests for locks stuck waiting on an expire flag
973 * @gh: the LM_FLAG_PRIORITY holder waiting to acquire the lock
975 * Don't cancel GL_NOCANCEL requests.
978 static void do_cancels(struct gfs2_holder
*gh
)
980 struct gfs2_glock
*gl
= gh
->gh_gl
;
982 spin_lock(&gl
->gl_spin
);
984 while (gl
->gl_req_gh
!= gh
&&
985 !test_bit(HIF_HOLDER
, &gh
->gh_iflags
) &&
986 !list_empty(&gh
->gh_list
)) {
987 if (gl
->gl_req_bh
&& !(gl
->gl_req_gh
&&
988 (gl
->gl_req_gh
->gh_flags
& GL_NOCANCEL
))) {
989 spin_unlock(&gl
->gl_spin
);
990 gfs2_lm_cancel(gl
->gl_sbd
, gl
->gl_lock
);
992 spin_lock(&gl
->gl_spin
);
994 spin_unlock(&gl
->gl_spin
);
996 spin_lock(&gl
->gl_spin
);
1000 spin_unlock(&gl
->gl_spin
);
1004 * glock_wait_internal - wait on a glock acquisition
1005 * @gh: the glock holder
1007 * Returns: 0 on success
1010 static int glock_wait_internal(struct gfs2_holder
*gh
)
1012 struct gfs2_glock
*gl
= gh
->gh_gl
;
1013 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
1014 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1016 if (test_bit(HIF_ABORTED
, &gh
->gh_iflags
))
1019 if (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
)) {
1020 spin_lock(&gl
->gl_spin
);
1021 if (gl
->gl_req_gh
!= gh
&&
1022 !test_bit(HIF_HOLDER
, &gh
->gh_iflags
) &&
1023 !list_empty(&gh
->gh_list
)) {
1024 list_del_init(&gh
->gh_list
);
1025 gh
->gh_error
= GLR_TRYFAILED
;
1027 spin_unlock(&gl
->gl_spin
);
1028 return gh
->gh_error
;
1030 spin_unlock(&gl
->gl_spin
);
1033 if (gh
->gh_flags
& LM_FLAG_PRIORITY
)
1038 return gh
->gh_error
;
1040 gfs2_assert_withdraw(sdp
, test_bit(HIF_HOLDER
, &gh
->gh_iflags
));
1041 gfs2_assert_withdraw(sdp
, relaxed_state_ok(gl
->gl_state
, gh
->gh_state
,
1044 if (test_bit(HIF_FIRST
, &gh
->gh_iflags
)) {
1045 gfs2_assert_warn(sdp
, test_bit(GLF_LOCK
, &gl
->gl_flags
));
1047 if (glops
->go_lock
) {
1048 gh
->gh_error
= glops
->go_lock(gh
);
1050 spin_lock(&gl
->gl_spin
);
1051 list_del_init(&gh
->gh_list
);
1052 spin_unlock(&gl
->gl_spin
);
1056 spin_lock(&gl
->gl_spin
);
1057 gl
->gl_req_gh
= NULL
;
1058 gl
->gl_req_bh
= NULL
;
1059 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
1061 spin_unlock(&gl
->gl_spin
);
1064 return gh
->gh_error
;
1067 static inline struct gfs2_holder
*
1068 find_holder_by_owner(struct list_head
*head
, pid_t pid
)
1070 struct gfs2_holder
*gh
;
1072 list_for_each_entry(gh
, head
, gh_list
) {
1073 if (gh
->gh_owner_pid
== pid
)
1080 static void print_dbg(struct glock_iter
*gi
, const char *fmt
, ...)
1084 va_start(args
, fmt
);
1086 vsprintf(gi
->string
, fmt
, args
);
1087 seq_printf(gi
->seq
, gi
->string
);
1095 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1096 * @gh: the holder structure to add
1100 static void add_to_queue(struct gfs2_holder
*gh
)
1102 struct gfs2_glock
*gl
= gh
->gh_gl
;
1103 struct gfs2_holder
*existing
;
1105 BUG_ON(!gh
->gh_owner_pid
);
1106 if (test_and_set_bit(HIF_WAIT
, &gh
->gh_iflags
))
1109 existing
= find_holder_by_owner(&gl
->gl_holders
, gh
->gh_owner_pid
);
1111 print_symbol(KERN_WARNING
"original: %s\n", existing
->gh_ip
);
1112 printk(KERN_INFO
"pid : %d\n", existing
->gh_owner_pid
);
1113 printk(KERN_INFO
"lock type : %d lock state : %d\n",
1114 existing
->gh_gl
->gl_name
.ln_type
, existing
->gh_gl
->gl_state
);
1115 print_symbol(KERN_WARNING
"new: %s\n", gh
->gh_ip
);
1116 printk(KERN_INFO
"pid : %d\n", gh
->gh_owner_pid
);
1117 printk(KERN_INFO
"lock type : %d lock state : %d\n",
1118 gl
->gl_name
.ln_type
, gl
->gl_state
);
1122 existing
= find_holder_by_owner(&gl
->gl_waiters3
, gh
->gh_owner_pid
);
1124 print_symbol(KERN_WARNING
"original: %s\n", existing
->gh_ip
);
1125 print_symbol(KERN_WARNING
"new: %s\n", gh
->gh_ip
);
1129 if (gh
->gh_flags
& LM_FLAG_PRIORITY
)
1130 list_add(&gh
->gh_list
, &gl
->gl_waiters3
);
1132 list_add_tail(&gh
->gh_list
, &gl
->gl_waiters3
);
1136 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1137 * @gh: the holder structure
1139 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1141 * Returns: 0, GLR_TRYFAILED, or errno on failure
1144 int gfs2_glock_nq(struct gfs2_holder
*gh
)
1146 struct gfs2_glock
*gl
= gh
->gh_gl
;
1147 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
1151 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
))) {
1152 set_bit(HIF_ABORTED
, &gh
->gh_iflags
);
1156 set_bit(HIF_PROMOTE
, &gh
->gh_iflags
);
1158 spin_lock(&gl
->gl_spin
);
1161 spin_unlock(&gl
->gl_spin
);
1163 if (!(gh
->gh_flags
& GL_ASYNC
)) {
1164 error
= glock_wait_internal(gh
);
1165 if (error
== GLR_CANCELED
) {
1175 * gfs2_glock_poll - poll to see if an async request has been completed
1178 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1181 int gfs2_glock_poll(struct gfs2_holder
*gh
)
1183 struct gfs2_glock
*gl
= gh
->gh_gl
;
1186 spin_lock(&gl
->gl_spin
);
1188 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
1190 else if (list_empty(&gh
->gh_list
)) {
1191 if (gh
->gh_error
== GLR_CANCELED
) {
1192 spin_unlock(&gl
->gl_spin
);
1194 if (gfs2_glock_nq(gh
))
1201 spin_unlock(&gl
->gl_spin
);
1207 * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
1208 * @gh: the holder structure
1210 * Returns: 0, GLR_TRYFAILED, or errno on failure
1213 int gfs2_glock_wait(struct gfs2_holder
*gh
)
1217 error
= glock_wait_internal(gh
);
1218 if (error
== GLR_CANCELED
) {
1220 gh
->gh_flags
&= ~GL_ASYNC
;
1221 error
= gfs2_glock_nq(gh
);
1228 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1229 * @gh: the glock holder
1233 void gfs2_glock_dq(struct gfs2_holder
*gh
)
1235 struct gfs2_glock
*gl
= gh
->gh_gl
;
1236 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1239 if (gh
->gh_flags
& GL_NOCACHE
)
1240 handle_callback(gl
, LM_ST_UNLOCKED
, 0, 0);
1242 gfs2_glmutex_lock(gl
);
1244 spin_lock(&gl
->gl_spin
);
1245 list_del_init(&gh
->gh_list
);
1247 if (list_empty(&gl
->gl_holders
)) {
1248 spin_unlock(&gl
->gl_spin
);
1250 if (glops
->go_unlock
)
1251 glops
->go_unlock(gh
);
1253 spin_lock(&gl
->gl_spin
);
1254 gl
->gl_stamp
= jiffies
;
1257 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
1258 spin_unlock(&gl
->gl_spin
);
1260 gfs2_glock_hold(gl
);
1261 if (test_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
) &&
1262 !test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
1263 delay
= gl
->gl_ops
->go_min_hold_time
;
1264 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, delay
) == 0)
1268 void gfs2_glock_dq_wait(struct gfs2_holder
*gh
)
1270 struct gfs2_glock
*gl
= gh
->gh_gl
;
1276 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1277 * @gh: the holder structure
1281 void gfs2_glock_dq_uninit(struct gfs2_holder
*gh
)
1284 gfs2_holder_uninit(gh
);
1288 * gfs2_glock_nq_num - acquire a glock based on lock number
1289 * @sdp: the filesystem
1290 * @number: the lock number
1291 * @glops: the glock operations for the type of glock
1292 * @state: the state to acquire the glock in
1293 * @flags: modifier flags for the aquisition
1294 * @gh: the struct gfs2_holder
1299 int gfs2_glock_nq_num(struct gfs2_sbd
*sdp
, u64 number
,
1300 const struct gfs2_glock_operations
*glops
,
1301 unsigned int state
, int flags
, struct gfs2_holder
*gh
)
1303 struct gfs2_glock
*gl
;
1306 error
= gfs2_glock_get(sdp
, number
, glops
, CREATE
, &gl
);
1308 error
= gfs2_glock_nq_init(gl
, state
, flags
, gh
);
1316 * glock_compare - Compare two struct gfs2_glock structures for sorting
1317 * @arg_a: the first structure
1318 * @arg_b: the second structure
1322 static int glock_compare(const void *arg_a
, const void *arg_b
)
1324 const struct gfs2_holder
*gh_a
= *(const struct gfs2_holder
**)arg_a
;
1325 const struct gfs2_holder
*gh_b
= *(const struct gfs2_holder
**)arg_b
;
1326 const struct lm_lockname
*a
= &gh_a
->gh_gl
->gl_name
;
1327 const struct lm_lockname
*b
= &gh_b
->gh_gl
->gl_name
;
1329 if (a
->ln_number
> b
->ln_number
)
1331 if (a
->ln_number
< b
->ln_number
)
1333 BUG_ON(gh_a
->gh_gl
->gl_ops
->go_type
== gh_b
->gh_gl
->gl_ops
->go_type
);
1338 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1339 * @num_gh: the number of structures
1340 * @ghs: an array of struct gfs2_holder structures
1342 * Returns: 0 on success (all glocks acquired),
1343 * errno on failure (no glocks acquired)
1346 static int nq_m_sync(unsigned int num_gh
, struct gfs2_holder
*ghs
,
1347 struct gfs2_holder
**p
)
1352 for (x
= 0; x
< num_gh
; x
++)
1355 sort(p
, num_gh
, sizeof(struct gfs2_holder
*), glock_compare
, NULL
);
1357 for (x
= 0; x
< num_gh
; x
++) {
1358 p
[x
]->gh_flags
&= ~(LM_FLAG_TRY
| GL_ASYNC
);
1360 error
= gfs2_glock_nq(p
[x
]);
1363 gfs2_glock_dq(p
[x
]);
1372 * gfs2_glock_nq_m - acquire multiple glocks
1373 * @num_gh: the number of structures
1374 * @ghs: an array of struct gfs2_holder structures
1377 * Returns: 0 on success (all glocks acquired),
1378 * errno on failure (no glocks acquired)
1381 int gfs2_glock_nq_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1383 struct gfs2_holder
*tmp
[4];
1384 struct gfs2_holder
**pph
= tmp
;
1391 ghs
->gh_flags
&= ~(LM_FLAG_TRY
| GL_ASYNC
);
1392 return gfs2_glock_nq(ghs
);
1396 pph
= kmalloc(num_gh
* sizeof(struct gfs2_holder
*), GFP_NOFS
);
1401 error
= nq_m_sync(num_gh
, ghs
, pph
);
1410 * gfs2_glock_dq_m - release multiple glocks
1411 * @num_gh: the number of structures
1412 * @ghs: an array of struct gfs2_holder structures
1416 void gfs2_glock_dq_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1420 for (x
= 0; x
< num_gh
; x
++)
1421 gfs2_glock_dq(&ghs
[x
]);
1425 * gfs2_glock_dq_uninit_m - release multiple glocks
1426 * @num_gh: the number of structures
1427 * @ghs: an array of struct gfs2_holder structures
1431 void gfs2_glock_dq_uninit_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1435 for (x
= 0; x
< num_gh
; x
++)
1436 gfs2_glock_dq_uninit(&ghs
[x
]);
1440 * gfs2_lvb_hold - attach a LVB from a glock
1441 * @gl: The glock in question
1445 int gfs2_lvb_hold(struct gfs2_glock
*gl
)
1449 gfs2_glmutex_lock(gl
);
1451 if (!atomic_read(&gl
->gl_lvb_count
)) {
1452 error
= gfs2_lm_hold_lvb(gl
->gl_sbd
, gl
->gl_lock
, &gl
->gl_lvb
);
1454 gfs2_glmutex_unlock(gl
);
1457 gfs2_glock_hold(gl
);
1459 atomic_inc(&gl
->gl_lvb_count
);
1461 gfs2_glmutex_unlock(gl
);
1467 * gfs2_lvb_unhold - detach a LVB from a glock
1468 * @gl: The glock in question
1472 void gfs2_lvb_unhold(struct gfs2_glock
*gl
)
1474 gfs2_glock_hold(gl
);
1475 gfs2_glmutex_lock(gl
);
1477 gfs2_assert(gl
->gl_sbd
, atomic_read(&gl
->gl_lvb_count
) > 0);
1478 if (atomic_dec_and_test(&gl
->gl_lvb_count
)) {
1479 gfs2_lm_unhold_lvb(gl
->gl_sbd
, gl
->gl_lock
, gl
->gl_lvb
);
1484 gfs2_glmutex_unlock(gl
);
1488 static void blocking_cb(struct gfs2_sbd
*sdp
, struct lm_lockname
*name
,
1491 struct gfs2_glock
*gl
;
1492 unsigned long delay
= 0;
1493 unsigned long holdtime
;
1494 unsigned long now
= jiffies
;
1496 gl
= gfs2_glock_find(sdp
, name
);
1500 holdtime
= gl
->gl_tchange
+ gl
->gl_ops
->go_min_hold_time
;
1501 if (time_before(now
, holdtime
))
1502 delay
= holdtime
- now
;
1504 handle_callback(gl
, state
, 1, delay
);
1505 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, delay
) == 0)
1510 * gfs2_glock_cb - Callback used by locking module
1511 * @sdp: Pointer to the superblock
1512 * @type: Type of callback
1513 * @data: Type dependent data pointer
1515 * Called by the locking module when it wants to tell us something.
1516 * Either we need to drop a lock, one of our ASYNC requests completed, or
1517 * a journal from another client needs to be recovered.
1520 void gfs2_glock_cb(void *cb_data
, unsigned int type
, void *data
)
1522 struct gfs2_sbd
*sdp
= cb_data
;
1526 blocking_cb(sdp
, data
, LM_ST_UNLOCKED
);
1530 blocking_cb(sdp
, data
, LM_ST_DEFERRED
);
1534 blocking_cb(sdp
, data
, LM_ST_SHARED
);
1538 struct lm_async_cb
*async
= data
;
1539 struct gfs2_glock
*gl
;
1541 down_read(&gfs2_umount_flush_sem
);
1542 gl
= gfs2_glock_find(sdp
, &async
->lc_name
);
1543 if (gfs2_assert_warn(sdp
, gl
))
1545 if (!gfs2_assert_warn(sdp
, gl
->gl_req_bh
))
1546 gl
->gl_req_bh(gl
, async
->lc_ret
);
1547 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
1549 up_read(&gfs2_umount_flush_sem
);
1553 case LM_CB_NEED_RECOVERY
:
1554 gfs2_jdesc_make_dirty(sdp
, *(unsigned int *)data
);
1555 if (sdp
->sd_recoverd_process
)
1556 wake_up_process(sdp
->sd_recoverd_process
);
1559 case LM_CB_DROPLOCKS
:
1560 gfs2_gl_hash_clear(sdp
, NO_WAIT
);
1561 gfs2_quota_scan(sdp
);
1565 gfs2_assert_warn(sdp
, 0);
1571 * demote_ok - Check to see if it's ok to unlock a glock
1574 * Returns: 1 if it's ok
1577 static int demote_ok(struct gfs2_glock
*gl
)
1579 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1582 if (test_bit(GLF_STICKY
, &gl
->gl_flags
))
1584 else if (glops
->go_demote_ok
)
1585 demote
= glops
->go_demote_ok(gl
);
1591 * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1596 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock
*gl
)
1598 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
1600 spin_lock(&sdp
->sd_reclaim_lock
);
1601 if (list_empty(&gl
->gl_reclaim
)) {
1602 gfs2_glock_hold(gl
);
1603 list_add(&gl
->gl_reclaim
, &sdp
->sd_reclaim_list
);
1604 atomic_inc(&sdp
->sd_reclaim_count
);
1606 spin_unlock(&sdp
->sd_reclaim_lock
);
1608 wake_up(&sdp
->sd_reclaim_wq
);
1612 * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1613 * @sdp: the filesystem
1615 * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1616 * different glock and we notice that there are a lot of glocks in the
1621 void gfs2_reclaim_glock(struct gfs2_sbd
*sdp
)
1623 struct gfs2_glock
*gl
;
1625 spin_lock(&sdp
->sd_reclaim_lock
);
1626 if (list_empty(&sdp
->sd_reclaim_list
)) {
1627 spin_unlock(&sdp
->sd_reclaim_lock
);
1630 gl
= list_entry(sdp
->sd_reclaim_list
.next
,
1631 struct gfs2_glock
, gl_reclaim
);
1632 list_del_init(&gl
->gl_reclaim
);
1633 spin_unlock(&sdp
->sd_reclaim_lock
);
1635 atomic_dec(&sdp
->sd_reclaim_count
);
1636 atomic_inc(&sdp
->sd_reclaimed
);
1638 if (gfs2_glmutex_trylock(gl
)) {
1639 if (list_empty(&gl
->gl_holders
) &&
1640 gl
->gl_state
!= LM_ST_UNLOCKED
&& demote_ok(gl
))
1641 handle_callback(gl
, LM_ST_UNLOCKED
, 0, 0);
1642 gfs2_glmutex_unlock(gl
);
1649 * examine_bucket - Call a function for glock in a hash bucket
1650 * @examiner: the function
1651 * @sdp: the filesystem
1652 * @bucket: the bucket
1654 * Returns: 1 if the bucket has entries
1657 static int examine_bucket(glock_examiner examiner
, struct gfs2_sbd
*sdp
,
1660 struct gfs2_glock
*gl
, *prev
= NULL
;
1661 int has_entries
= 0;
1662 struct hlist_head
*head
= &gl_hash_table
[hash
].hb_list
;
1664 read_lock(gl_lock_addr(hash
));
1665 /* Can't use hlist_for_each_entry - don't want prefetch here */
1666 if (hlist_empty(head
))
1668 gl
= list_entry(head
->first
, struct gfs2_glock
, gl_list
);
1670 if (!sdp
|| gl
->gl_sbd
== sdp
) {
1671 gfs2_glock_hold(gl
);
1672 read_unlock(gl_lock_addr(hash
));
1674 gfs2_glock_put(prev
);
1678 read_lock(gl_lock_addr(hash
));
1680 if (gl
->gl_list
.next
== NULL
)
1682 gl
= list_entry(gl
->gl_list
.next
, struct gfs2_glock
, gl_list
);
1685 read_unlock(gl_lock_addr(hash
));
1687 gfs2_glock_put(prev
);
1693 * scan_glock - look at a glock and see if we can reclaim it
1694 * @gl: the glock to look at
1698 static void scan_glock(struct gfs2_glock
*gl
)
1700 if (gl
->gl_ops
== &gfs2_inode_glops
&& gl
->gl_object
)
1703 if (gfs2_glmutex_trylock(gl
)) {
1704 if (list_empty(&gl
->gl_holders
) &&
1705 gl
->gl_state
!= LM_ST_UNLOCKED
&& demote_ok(gl
))
1707 gfs2_glmutex_unlock(gl
);
1712 gfs2_glmutex_unlock(gl
);
1713 gfs2_glock_schedule_for_reclaim(gl
);
1717 * clear_glock - look at a glock and see if we can free it from glock cache
1718 * @gl: the glock to look at
1722 static void clear_glock(struct gfs2_glock
*gl
)
1724 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
1727 spin_lock(&sdp
->sd_reclaim_lock
);
1728 if (!list_empty(&gl
->gl_reclaim
)) {
1729 list_del_init(&gl
->gl_reclaim
);
1730 atomic_dec(&sdp
->sd_reclaim_count
);
1731 spin_unlock(&sdp
->sd_reclaim_lock
);
1732 released
= gfs2_glock_put(gl
);
1733 gfs2_assert(sdp
, !released
);
1735 spin_unlock(&sdp
->sd_reclaim_lock
);
1738 if (gfs2_glmutex_trylock(gl
)) {
1739 if (list_empty(&gl
->gl_holders
) &&
1740 gl
->gl_state
!= LM_ST_UNLOCKED
)
1741 handle_callback(gl
, LM_ST_UNLOCKED
, 0, 0);
1742 gfs2_glmutex_unlock(gl
);
1747 * gfs2_gl_hash_clear - Empty out the glock hash table
1748 * @sdp: the filesystem
1749 * @wait: wait until it's all gone
1751 * Called when unmounting the filesystem, or when inter-node lock manager
1752 * requests DROPLOCKS because it is running out of capacity.
1755 void gfs2_gl_hash_clear(struct gfs2_sbd
*sdp
, int wait
)
1765 for (x
= 0; x
< GFS2_GL_HASH_SIZE
; x
++) {
1766 if (examine_bucket(clear_glock
, sdp
, x
))
1773 if (time_after_eq(jiffies
,
1774 t
+ gfs2_tune_get(sdp
, gt_stall_secs
) * HZ
)) {
1775 fs_warn(sdp
, "Unmount seems to be stalled. "
1776 "Dumping lock state...\n");
1777 gfs2_dump_lockstate(sdp
);
1781 down_write(&gfs2_umount_flush_sem
);
1782 invalidate_inodes(sdp
->sd_vfs
);
1783 up_write(&gfs2_umount_flush_sem
);
1789 * Diagnostic routines to help debug distributed deadlock
1792 static void gfs2_print_symbol(struct glock_iter
*gi
, const char *fmt
,
1793 unsigned long address
)
1795 char buffer
[KSYM_SYMBOL_LEN
];
1797 sprint_symbol(buffer
, address
);
1798 print_dbg(gi
, fmt
, buffer
);
1802 * dump_holder - print information about a glock holder
1803 * @str: a string naming the type of holder
1804 * @gh: the glock holder
1806 * Returns: 0 on success, -ENOBUFS when we run out of space
1809 static int dump_holder(struct glock_iter
*gi
, char *str
,
1810 struct gfs2_holder
*gh
)
1813 struct task_struct
*gh_owner
;
1815 print_dbg(gi
, " %s\n", str
);
1816 if (gh
->gh_owner_pid
) {
1817 print_dbg(gi
, " owner = %ld ", (long)gh
->gh_owner_pid
);
1818 gh_owner
= find_task_by_pid(gh
->gh_owner_pid
);
1820 print_dbg(gi
, "(%s)\n", gh_owner
->comm
);
1822 print_dbg(gi
, "(ended)\n");
1824 print_dbg(gi
, " owner = -1\n");
1825 print_dbg(gi
, " gh_state = %u\n", gh
->gh_state
);
1826 print_dbg(gi
, " gh_flags =");
1827 for (x
= 0; x
< 32; x
++)
1828 if (gh
->gh_flags
& (1 << x
))
1829 print_dbg(gi
, " %u", x
);
1830 print_dbg(gi
, " \n");
1831 print_dbg(gi
, " error = %d\n", gh
->gh_error
);
1832 print_dbg(gi
, " gh_iflags =");
1833 for (x
= 0; x
< 32; x
++)
1834 if (test_bit(x
, &gh
->gh_iflags
))
1835 print_dbg(gi
, " %u", x
);
1836 print_dbg(gi
, " \n");
1837 gfs2_print_symbol(gi
, " initialized at: %s\n", gh
->gh_ip
);
1843 * dump_inode - print information about an inode
1846 * Returns: 0 on success, -ENOBUFS when we run out of space
1849 static int dump_inode(struct glock_iter
*gi
, struct gfs2_inode
*ip
)
1853 print_dbg(gi
, " Inode:\n");
1854 print_dbg(gi
, " num = %llu/%llu\n",
1855 (unsigned long long)ip
->i_no_formal_ino
,
1856 (unsigned long long)ip
->i_no_addr
);
1857 print_dbg(gi
, " type = %u\n", IF2DT(ip
->i_inode
.i_mode
));
1858 print_dbg(gi
, " i_flags =");
1859 for (x
= 0; x
< 32; x
++)
1860 if (test_bit(x
, &ip
->i_flags
))
1861 print_dbg(gi
, " %u", x
);
1862 print_dbg(gi
, " \n");
1867 * dump_glock - print information about a glock
1869 * @count: where we are in the buffer
1871 * Returns: 0 on success, -ENOBUFS when we run out of space
1874 static int dump_glock(struct glock_iter
*gi
, struct gfs2_glock
*gl
)
1876 struct gfs2_holder
*gh
;
1878 int error
= -ENOBUFS
;
1879 struct task_struct
*gl_owner
;
1881 spin_lock(&gl
->gl_spin
);
1883 print_dbg(gi
, "Glock 0x%p (%u, 0x%llx)\n", gl
, gl
->gl_name
.ln_type
,
1884 (unsigned long long)gl
->gl_name
.ln_number
);
1885 print_dbg(gi
, " gl_flags =");
1886 for (x
= 0; x
< 32; x
++) {
1887 if (test_bit(x
, &gl
->gl_flags
))
1888 print_dbg(gi
, " %u", x
);
1890 if (!test_bit(GLF_LOCK
, &gl
->gl_flags
))
1891 print_dbg(gi
, " (unlocked)");
1892 print_dbg(gi
, " \n");
1893 print_dbg(gi
, " gl_ref = %d\n", atomic_read(&gl
->gl_ref
));
1894 print_dbg(gi
, " gl_state = %u\n", gl
->gl_state
);
1895 if (gl
->gl_owner_pid
) {
1896 gl_owner
= find_task_by_pid(gl
->gl_owner_pid
);
1898 print_dbg(gi
, " gl_owner = pid %d (%s)\n",
1899 gl
->gl_owner_pid
, gl_owner
->comm
);
1901 print_dbg(gi
, " gl_owner = %d (ended)\n",
1904 print_dbg(gi
, " gl_owner = -1\n");
1905 print_dbg(gi
, " gl_ip = %lu\n", gl
->gl_ip
);
1906 print_dbg(gi
, " req_gh = %s\n", (gl
->gl_req_gh
) ? "yes" : "no");
1907 print_dbg(gi
, " req_bh = %s\n", (gl
->gl_req_bh
) ? "yes" : "no");
1908 print_dbg(gi
, " lvb_count = %d\n", atomic_read(&gl
->gl_lvb_count
));
1909 print_dbg(gi
, " object = %s\n", (gl
->gl_object
) ? "yes" : "no");
1910 print_dbg(gi
, " le = %s\n",
1911 (list_empty(&gl
->gl_le
.le_list
)) ? "no" : "yes");
1912 print_dbg(gi
, " reclaim = %s\n",
1913 (list_empty(&gl
->gl_reclaim
)) ? "no" : "yes");
1915 print_dbg(gi
, " aspace = 0x%p nrpages = %lu\n", gl
->gl_aspace
,
1916 gl
->gl_aspace
->i_mapping
->nrpages
);
1918 print_dbg(gi
, " aspace = no\n");
1919 print_dbg(gi
, " ail = %d\n", atomic_read(&gl
->gl_ail_count
));
1920 if (gl
->gl_req_gh
) {
1921 error
= dump_holder(gi
, "Request", gl
->gl_req_gh
);
1925 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
) {
1926 error
= dump_holder(gi
, "Holder", gh
);
1930 list_for_each_entry(gh
, &gl
->gl_waiters1
, gh_list
) {
1931 error
= dump_holder(gi
, "Waiter1", gh
);
1935 list_for_each_entry(gh
, &gl
->gl_waiters3
, gh_list
) {
1936 error
= dump_holder(gi
, "Waiter3", gh
);
1940 if (test_bit(GLF_DEMOTE
, &gl
->gl_flags
)) {
1941 print_dbg(gi
, " Demotion req to state %u (%llu uS ago)\n",
1942 gl
->gl_demote_state
, (unsigned long long)
1943 (jiffies
- gl
->gl_demote_time
)*(1000000/HZ
));
1945 if (gl
->gl_ops
== &gfs2_inode_glops
&& gl
->gl_object
) {
1946 if (!test_bit(GLF_LOCK
, &gl
->gl_flags
) &&
1947 list_empty(&gl
->gl_holders
)) {
1948 error
= dump_inode(gi
, gl
->gl_object
);
1953 print_dbg(gi
, " Inode: busy\n");
1960 spin_unlock(&gl
->gl_spin
);
1965 * gfs2_dump_lockstate - print out the current lockstate
1966 * @sdp: the filesystem
1967 * @ub: the buffer to copy the information into
1969 * If @ub is NULL, dump the lockstate to the console.
1973 static int gfs2_dump_lockstate(struct gfs2_sbd
*sdp
)
1975 struct gfs2_glock
*gl
;
1976 struct hlist_node
*h
;
1980 for (x
= 0; x
< GFS2_GL_HASH_SIZE
; x
++) {
1982 read_lock(gl_lock_addr(x
));
1984 hlist_for_each_entry(gl
, h
, &gl_hash_table
[x
].hb_list
, gl_list
) {
1985 if (gl
->gl_sbd
!= sdp
)
1988 error
= dump_glock(NULL
, gl
);
1993 read_unlock(gl_lock_addr(x
));
2004 * gfs2_scand - Look for cached glocks and inodes to toss from memory
2005 * @sdp: Pointer to GFS2 superblock
2007 * One of these daemons runs, finding candidates to add to sd_reclaim_list.
2011 static int gfs2_scand(void *data
)
2016 while (!kthread_should_stop()) {
2017 for (x
= 0; x
< GFS2_GL_HASH_SIZE
; x
++)
2018 examine_bucket(scan_glock
, NULL
, x
);
2019 if (freezing(current
))
2024 schedule_timeout_interruptible(delay
* HZ
);
2032 int __init
gfs2_glock_init(void)
2035 for(i
= 0; i
< GFS2_GL_HASH_SIZE
; i
++) {
2036 INIT_HLIST_HEAD(&gl_hash_table
[i
].hb_list
);
2038 #ifdef GL_HASH_LOCK_SZ
2039 for(i
= 0; i
< GL_HASH_LOCK_SZ
; i
++) {
2040 rwlock_init(&gl_hash_locks
[i
]);
2044 scand_process
= kthread_run(gfs2_scand
, NULL
, "gfs2_scand");
2045 if (IS_ERR(scand_process
))
2046 return PTR_ERR(scand_process
);
2048 glock_workqueue
= create_workqueue("glock_workqueue");
2049 if (IS_ERR(glock_workqueue
)) {
2050 kthread_stop(scand_process
);
2051 return PTR_ERR(glock_workqueue
);
2057 void gfs2_glock_exit(void)
2059 destroy_workqueue(glock_workqueue
);
2060 kthread_stop(scand_process
);
2063 module_param(scand_secs
, uint
, S_IRUGO
|S_IWUSR
);
2064 MODULE_PARM_DESC(scand_secs
, "The number of seconds between scand runs");
2066 static int gfs2_glock_iter_next(struct glock_iter
*gi
)
2068 struct gfs2_glock
*gl
;
2071 read_lock(gl_lock_addr(gi
->hash
));
2074 gi
->gl
= hlist_entry(gl
->gl_list
.next
,
2075 struct gfs2_glock
, gl_list
);
2077 gfs2_glock_hold(gi
->gl
);
2079 read_unlock(gl_lock_addr(gi
->hash
));
2082 if (gl
&& gi
->gl
== NULL
)
2084 while(gi
->gl
== NULL
) {
2085 if (gi
->hash
>= GFS2_GL_HASH_SIZE
)
2087 read_lock(gl_lock_addr(gi
->hash
));
2088 gi
->gl
= hlist_entry(gl_hash_table
[gi
->hash
].hb_list
.first
,
2089 struct gfs2_glock
, gl_list
);
2091 gfs2_glock_hold(gi
->gl
);
2092 read_unlock(gl_lock_addr(gi
->hash
));
2096 if (gi
->sdp
!= gi
->gl
->gl_sbd
)
2102 static void gfs2_glock_iter_free(struct glock_iter
*gi
)
2105 gfs2_glock_put(gi
->gl
);
2109 static struct glock_iter
*gfs2_glock_iter_init(struct gfs2_sbd
*sdp
)
2111 struct glock_iter
*gi
;
2113 gi
= kmalloc(sizeof (*gi
), GFP_KERNEL
);
2121 memset(gi
->string
, 0, sizeof(gi
->string
));
2123 if (gfs2_glock_iter_next(gi
)) {
2124 gfs2_glock_iter_free(gi
);
2131 static void *gfs2_glock_seq_start(struct seq_file
*file
, loff_t
*pos
)
2133 struct glock_iter
*gi
;
2136 gi
= gfs2_glock_iter_init(file
->private);
2141 if (gfs2_glock_iter_next(gi
)) {
2142 gfs2_glock_iter_free(gi
);
2150 static void *gfs2_glock_seq_next(struct seq_file
*file
, void *iter_ptr
,
2153 struct glock_iter
*gi
= iter_ptr
;
2157 if (gfs2_glock_iter_next(gi
)) {
2158 gfs2_glock_iter_free(gi
);
2165 static void gfs2_glock_seq_stop(struct seq_file
*file
, void *iter_ptr
)
2167 struct glock_iter
*gi
= iter_ptr
;
2169 gfs2_glock_iter_free(gi
);
2172 static int gfs2_glock_seq_show(struct seq_file
*file
, void *iter_ptr
)
2174 struct glock_iter
*gi
= iter_ptr
;
2177 dump_glock(gi
, gi
->gl
);
2182 static const struct seq_operations gfs2_glock_seq_ops
= {
2183 .start
= gfs2_glock_seq_start
,
2184 .next
= gfs2_glock_seq_next
,
2185 .stop
= gfs2_glock_seq_stop
,
2186 .show
= gfs2_glock_seq_show
,
2189 static int gfs2_debugfs_open(struct inode
*inode
, struct file
*file
)
2191 struct seq_file
*seq
;
2194 ret
= seq_open(file
, &gfs2_glock_seq_ops
);
2198 seq
= file
->private_data
;
2199 seq
->private = inode
->i_private
;
2204 static const struct file_operations gfs2_debug_fops
= {
2205 .owner
= THIS_MODULE
,
2206 .open
= gfs2_debugfs_open
,
2208 .llseek
= seq_lseek
,
2209 .release
= seq_release
2212 int gfs2_create_debugfs_file(struct gfs2_sbd
*sdp
)
2214 sdp
->debugfs_dir
= debugfs_create_dir(sdp
->sd_table_name
, gfs2_root
);
2215 if (!sdp
->debugfs_dir
)
2217 sdp
->debugfs_dentry_glocks
= debugfs_create_file("glocks",
2219 sdp
->debugfs_dir
, sdp
,
2221 if (!sdp
->debugfs_dentry_glocks
)
2227 void gfs2_delete_debugfs_file(struct gfs2_sbd
*sdp
)
2229 if (sdp
&& sdp
->debugfs_dir
) {
2230 if (sdp
->debugfs_dentry_glocks
) {
2231 debugfs_remove(sdp
->debugfs_dentry_glocks
);
2232 sdp
->debugfs_dentry_glocks
= NULL
;
2234 debugfs_remove(sdp
->debugfs_dir
);
2235 sdp
->debugfs_dir
= NULL
;
2239 int gfs2_register_debugfs(void)
2241 gfs2_root
= debugfs_create_dir("gfs2", NULL
);
2242 return gfs2_root
? 0 : -ENOMEM
;
2245 void gfs2_unregister_debugfs(void)
2247 debugfs_remove(gfs2_root
);