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
;
511 if (!relaxed_state_ok(gl
->gl_state
, gh
->gh_state
, gh
->gh_flags
)) {
512 if (list_empty(&gl
->gl_holders
)) {
514 set_bit(GLF_LOCK
, &gl
->gl_flags
);
515 spin_unlock(&gl
->gl_spin
);
516 gfs2_glock_xmote_th(gh
->gh_gl
, gh
);
517 spin_lock(&gl
->gl_spin
);
522 if (list_empty(&gl
->gl_holders
)) {
523 set_bit(HIF_FIRST
, &gh
->gh_iflags
);
524 set_bit(GLF_LOCK
, &gl
->gl_flags
);
526 struct gfs2_holder
*next_gh
;
527 if (gh
->gh_state
== LM_ST_EXCLUSIVE
)
529 next_gh
= list_entry(gl
->gl_holders
.next
, struct gfs2_holder
,
531 if (next_gh
->gh_state
== LM_ST_EXCLUSIVE
)
535 list_move_tail(&gh
->gh_list
, &gl
->gl_holders
);
537 set_bit(HIF_HOLDER
, &gh
->gh_iflags
);
539 gfs2_holder_wake(gh
);
545 * rq_demote - process a demote request in the queue
546 * @gh: the glock holder
548 * Returns: 1 if the queue is blocked
551 static int rq_demote(struct gfs2_glock
*gl
)
553 if (!list_empty(&gl
->gl_holders
))
556 if (gl
->gl_state
== gl
->gl_demote_state
||
557 gl
->gl_state
== LM_ST_UNLOCKED
) {
558 gfs2_demote_wake(gl
);
562 set_bit(GLF_LOCK
, &gl
->gl_flags
);
563 set_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
);
565 if (gl
->gl_demote_state
== LM_ST_UNLOCKED
||
566 gl
->gl_state
!= LM_ST_EXCLUSIVE
) {
567 spin_unlock(&gl
->gl_spin
);
568 gfs2_glock_drop_th(gl
);
570 spin_unlock(&gl
->gl_spin
);
571 gfs2_glock_xmote_th(gl
, NULL
);
574 spin_lock(&gl
->gl_spin
);
575 clear_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
);
581 * run_queue - process holder structures on a glock
585 static void run_queue(struct gfs2_glock
*gl
)
587 struct gfs2_holder
*gh
;
591 if (test_bit(GLF_LOCK
, &gl
->gl_flags
))
594 if (!list_empty(&gl
->gl_waiters1
)) {
595 gh
= list_entry(gl
->gl_waiters1
.next
,
596 struct gfs2_holder
, gh_list
);
598 if (test_bit(HIF_MUTEX
, &gh
->gh_iflags
))
599 blocked
= rq_mutex(gh
);
601 gfs2_assert_warn(gl
->gl_sbd
, 0);
603 } else if (test_bit(GLF_DEMOTE
, &gl
->gl_flags
)) {
604 blocked
= rq_demote(gl
);
605 if (gl
->gl_waiters2
&& !blocked
) {
606 set_bit(GLF_DEMOTE
, &gl
->gl_flags
);
607 gl
->gl_demote_state
= LM_ST_UNLOCKED
;
610 } else if (!list_empty(&gl
->gl_waiters3
)) {
611 gh
= list_entry(gl
->gl_waiters3
.next
,
612 struct gfs2_holder
, gh_list
);
614 if (test_bit(HIF_PROMOTE
, &gh
->gh_iflags
))
615 blocked
= rq_promote(gh
);
617 gfs2_assert_warn(gl
->gl_sbd
, 0);
628 * gfs2_glmutex_lock - acquire a local lock on a glock
631 * Gives caller exclusive access to manipulate a glock structure.
634 static void gfs2_glmutex_lock(struct gfs2_glock
*gl
)
636 struct gfs2_holder gh
;
638 gfs2_holder_init(gl
, 0, 0, &gh
);
639 set_bit(HIF_MUTEX
, &gh
.gh_iflags
);
640 if (test_and_set_bit(HIF_WAIT
, &gh
.gh_iflags
))
643 spin_lock(&gl
->gl_spin
);
644 if (test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
)) {
645 list_add_tail(&gh
.gh_list
, &gl
->gl_waiters1
);
647 gl
->gl_owner_pid
= current
->pid
;
648 gl
->gl_ip
= (unsigned long)__builtin_return_address(0);
649 clear_bit(HIF_WAIT
, &gh
.gh_iflags
);
651 wake_up_bit(&gh
.gh_iflags
, HIF_WAIT
);
653 spin_unlock(&gl
->gl_spin
);
656 gfs2_holder_uninit(&gh
);
660 * gfs2_glmutex_trylock - try to acquire a local lock on a glock
663 * Returns: 1 if the glock is acquired
666 static int gfs2_glmutex_trylock(struct gfs2_glock
*gl
)
670 spin_lock(&gl
->gl_spin
);
671 if (test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
)) {
674 gl
->gl_owner_pid
= current
->pid
;
675 gl
->gl_ip
= (unsigned long)__builtin_return_address(0);
677 spin_unlock(&gl
->gl_spin
);
683 * gfs2_glmutex_unlock - release a local lock on a glock
688 static void gfs2_glmutex_unlock(struct gfs2_glock
*gl
)
690 spin_lock(&gl
->gl_spin
);
691 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
692 gl
->gl_owner_pid
= 0;
695 BUG_ON(!spin_is_locked(&gl
->gl_spin
));
696 spin_unlock(&gl
->gl_spin
);
700 * handle_callback - process a demote request
702 * @state: the state the caller wants us to change to
704 * There are only two requests that we are going to see in actual
705 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
708 static void handle_callback(struct gfs2_glock
*gl
, unsigned int state
,
709 int remote
, unsigned long delay
)
711 int bit
= delay
? GLF_PENDING_DEMOTE
: GLF_DEMOTE
;
713 spin_lock(&gl
->gl_spin
);
714 set_bit(bit
, &gl
->gl_flags
);
715 if (gl
->gl_demote_state
== LM_ST_EXCLUSIVE
) {
716 gl
->gl_demote_state
= state
;
717 gl
->gl_demote_time
= jiffies
;
718 if (remote
&& gl
->gl_ops
->go_type
== LM_TYPE_IOPEN
&&
720 gfs2_glock_schedule_for_reclaim(gl
);
721 spin_unlock(&gl
->gl_spin
);
724 } else if (gl
->gl_demote_state
!= LM_ST_UNLOCKED
&&
725 gl
->gl_demote_state
!= state
) {
726 if (test_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
))
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 if (!(gh
->gh_flags
& GL_FLOCK
)) {
1110 existing
= find_holder_by_owner(&gl
->gl_holders
,
1113 print_symbol(KERN_WARNING
"original: %s\n",
1115 printk(KERN_INFO
"pid : %d\n", existing
->gh_owner_pid
);
1116 printk(KERN_INFO
"lock type : %d lock state : %d\n",
1117 existing
->gh_gl
->gl_name
.ln_type
,
1118 existing
->gh_gl
->gl_state
);
1119 print_symbol(KERN_WARNING
"new: %s\n", gh
->gh_ip
);
1120 printk(KERN_INFO
"pid : %d\n", gh
->gh_owner_pid
);
1121 printk(KERN_INFO
"lock type : %d lock state : %d\n",
1122 gl
->gl_name
.ln_type
, gl
->gl_state
);
1126 existing
= find_holder_by_owner(&gl
->gl_waiters3
,
1129 print_symbol(KERN_WARNING
"original: %s\n",
1131 print_symbol(KERN_WARNING
"new: %s\n", gh
->gh_ip
);
1136 if (gh
->gh_flags
& LM_FLAG_PRIORITY
)
1137 list_add(&gh
->gh_list
, &gl
->gl_waiters3
);
1139 list_add_tail(&gh
->gh_list
, &gl
->gl_waiters3
);
1143 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1144 * @gh: the holder structure
1146 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1148 * Returns: 0, GLR_TRYFAILED, or errno on failure
1151 int gfs2_glock_nq(struct gfs2_holder
*gh
)
1153 struct gfs2_glock
*gl
= gh
->gh_gl
;
1154 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
1158 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
))) {
1159 set_bit(HIF_ABORTED
, &gh
->gh_iflags
);
1163 set_bit(HIF_PROMOTE
, &gh
->gh_iflags
);
1165 spin_lock(&gl
->gl_spin
);
1168 spin_unlock(&gl
->gl_spin
);
1170 if (!(gh
->gh_flags
& GL_ASYNC
)) {
1171 error
= glock_wait_internal(gh
);
1172 if (error
== GLR_CANCELED
) {
1182 * gfs2_glock_poll - poll to see if an async request has been completed
1185 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1188 int gfs2_glock_poll(struct gfs2_holder
*gh
)
1190 struct gfs2_glock
*gl
= gh
->gh_gl
;
1193 spin_lock(&gl
->gl_spin
);
1195 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
1197 else if (list_empty(&gh
->gh_list
)) {
1198 if (gh
->gh_error
== GLR_CANCELED
) {
1199 spin_unlock(&gl
->gl_spin
);
1201 if (gfs2_glock_nq(gh
))
1208 spin_unlock(&gl
->gl_spin
);
1214 * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
1215 * @gh: the holder structure
1217 * Returns: 0, GLR_TRYFAILED, or errno on failure
1220 int gfs2_glock_wait(struct gfs2_holder
*gh
)
1224 error
= glock_wait_internal(gh
);
1225 if (error
== GLR_CANCELED
) {
1227 gh
->gh_flags
&= ~GL_ASYNC
;
1228 error
= gfs2_glock_nq(gh
);
1235 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1236 * @gh: the glock holder
1240 void gfs2_glock_dq(struct gfs2_holder
*gh
)
1242 struct gfs2_glock
*gl
= gh
->gh_gl
;
1243 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1246 if (gh
->gh_flags
& GL_NOCACHE
)
1247 handle_callback(gl
, LM_ST_UNLOCKED
, 0, 0);
1249 gfs2_glmutex_lock(gl
);
1251 spin_lock(&gl
->gl_spin
);
1252 list_del_init(&gh
->gh_list
);
1254 if (list_empty(&gl
->gl_holders
)) {
1255 spin_unlock(&gl
->gl_spin
);
1257 if (glops
->go_unlock
)
1258 glops
->go_unlock(gh
);
1260 spin_lock(&gl
->gl_spin
);
1261 gl
->gl_stamp
= jiffies
;
1264 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
1265 spin_unlock(&gl
->gl_spin
);
1267 gfs2_glock_hold(gl
);
1268 if (test_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
) &&
1269 !test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
1270 delay
= gl
->gl_ops
->go_min_hold_time
;
1271 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, delay
) == 0)
1275 void gfs2_glock_dq_wait(struct gfs2_holder
*gh
)
1277 struct gfs2_glock
*gl
= gh
->gh_gl
;
1283 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1284 * @gh: the holder structure
1288 void gfs2_glock_dq_uninit(struct gfs2_holder
*gh
)
1291 gfs2_holder_uninit(gh
);
1295 * gfs2_glock_nq_num - acquire a glock based on lock number
1296 * @sdp: the filesystem
1297 * @number: the lock number
1298 * @glops: the glock operations for the type of glock
1299 * @state: the state to acquire the glock in
1300 * @flags: modifier flags for the aquisition
1301 * @gh: the struct gfs2_holder
1306 int gfs2_glock_nq_num(struct gfs2_sbd
*sdp
, u64 number
,
1307 const struct gfs2_glock_operations
*glops
,
1308 unsigned int state
, int flags
, struct gfs2_holder
*gh
)
1310 struct gfs2_glock
*gl
;
1313 error
= gfs2_glock_get(sdp
, number
, glops
, CREATE
, &gl
);
1315 error
= gfs2_glock_nq_init(gl
, state
, flags
, gh
);
1323 * glock_compare - Compare two struct gfs2_glock structures for sorting
1324 * @arg_a: the first structure
1325 * @arg_b: the second structure
1329 static int glock_compare(const void *arg_a
, const void *arg_b
)
1331 const struct gfs2_holder
*gh_a
= *(const struct gfs2_holder
**)arg_a
;
1332 const struct gfs2_holder
*gh_b
= *(const struct gfs2_holder
**)arg_b
;
1333 const struct lm_lockname
*a
= &gh_a
->gh_gl
->gl_name
;
1334 const struct lm_lockname
*b
= &gh_b
->gh_gl
->gl_name
;
1336 if (a
->ln_number
> b
->ln_number
)
1338 if (a
->ln_number
< b
->ln_number
)
1340 BUG_ON(gh_a
->gh_gl
->gl_ops
->go_type
== gh_b
->gh_gl
->gl_ops
->go_type
);
1345 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1346 * @num_gh: the number of structures
1347 * @ghs: an array of struct gfs2_holder structures
1349 * Returns: 0 on success (all glocks acquired),
1350 * errno on failure (no glocks acquired)
1353 static int nq_m_sync(unsigned int num_gh
, struct gfs2_holder
*ghs
,
1354 struct gfs2_holder
**p
)
1359 for (x
= 0; x
< num_gh
; x
++)
1362 sort(p
, num_gh
, sizeof(struct gfs2_holder
*), glock_compare
, NULL
);
1364 for (x
= 0; x
< num_gh
; x
++) {
1365 p
[x
]->gh_flags
&= ~(LM_FLAG_TRY
| GL_ASYNC
);
1367 error
= gfs2_glock_nq(p
[x
]);
1370 gfs2_glock_dq(p
[x
]);
1379 * gfs2_glock_nq_m - acquire multiple glocks
1380 * @num_gh: the number of structures
1381 * @ghs: an array of struct gfs2_holder structures
1384 * Returns: 0 on success (all glocks acquired),
1385 * errno on failure (no glocks acquired)
1388 int gfs2_glock_nq_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1390 struct gfs2_holder
*tmp
[4];
1391 struct gfs2_holder
**pph
= tmp
;
1398 ghs
->gh_flags
&= ~(LM_FLAG_TRY
| GL_ASYNC
);
1399 return gfs2_glock_nq(ghs
);
1403 pph
= kmalloc(num_gh
* sizeof(struct gfs2_holder
*), GFP_NOFS
);
1408 error
= nq_m_sync(num_gh
, ghs
, pph
);
1417 * gfs2_glock_dq_m - release multiple glocks
1418 * @num_gh: the number of structures
1419 * @ghs: an array of struct gfs2_holder structures
1423 void gfs2_glock_dq_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1427 for (x
= 0; x
< num_gh
; x
++)
1428 gfs2_glock_dq(&ghs
[x
]);
1432 * gfs2_glock_dq_uninit_m - release multiple glocks
1433 * @num_gh: the number of structures
1434 * @ghs: an array of struct gfs2_holder structures
1438 void gfs2_glock_dq_uninit_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1442 for (x
= 0; x
< num_gh
; x
++)
1443 gfs2_glock_dq_uninit(&ghs
[x
]);
1447 * gfs2_lvb_hold - attach a LVB from a glock
1448 * @gl: The glock in question
1452 int gfs2_lvb_hold(struct gfs2_glock
*gl
)
1456 gfs2_glmutex_lock(gl
);
1458 if (!atomic_read(&gl
->gl_lvb_count
)) {
1459 error
= gfs2_lm_hold_lvb(gl
->gl_sbd
, gl
->gl_lock
, &gl
->gl_lvb
);
1461 gfs2_glmutex_unlock(gl
);
1464 gfs2_glock_hold(gl
);
1466 atomic_inc(&gl
->gl_lvb_count
);
1468 gfs2_glmutex_unlock(gl
);
1474 * gfs2_lvb_unhold - detach a LVB from a glock
1475 * @gl: The glock in question
1479 void gfs2_lvb_unhold(struct gfs2_glock
*gl
)
1481 gfs2_glock_hold(gl
);
1482 gfs2_glmutex_lock(gl
);
1484 gfs2_assert(gl
->gl_sbd
, atomic_read(&gl
->gl_lvb_count
) > 0);
1485 if (atomic_dec_and_test(&gl
->gl_lvb_count
)) {
1486 gfs2_lm_unhold_lvb(gl
->gl_sbd
, gl
->gl_lock
, gl
->gl_lvb
);
1491 gfs2_glmutex_unlock(gl
);
1495 static void blocking_cb(struct gfs2_sbd
*sdp
, struct lm_lockname
*name
,
1498 struct gfs2_glock
*gl
;
1499 unsigned long delay
= 0;
1500 unsigned long holdtime
;
1501 unsigned long now
= jiffies
;
1503 gl
= gfs2_glock_find(sdp
, name
);
1507 holdtime
= gl
->gl_tchange
+ gl
->gl_ops
->go_min_hold_time
;
1508 if (time_before(now
, holdtime
))
1509 delay
= holdtime
- now
;
1511 handle_callback(gl
, state
, 1, delay
);
1512 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, delay
) == 0)
1517 * gfs2_glock_cb - Callback used by locking module
1518 * @sdp: Pointer to the superblock
1519 * @type: Type of callback
1520 * @data: Type dependent data pointer
1522 * Called by the locking module when it wants to tell us something.
1523 * Either we need to drop a lock, one of our ASYNC requests completed, or
1524 * a journal from another client needs to be recovered.
1527 void gfs2_glock_cb(void *cb_data
, unsigned int type
, void *data
)
1529 struct gfs2_sbd
*sdp
= cb_data
;
1533 blocking_cb(sdp
, data
, LM_ST_UNLOCKED
);
1537 blocking_cb(sdp
, data
, LM_ST_DEFERRED
);
1541 blocking_cb(sdp
, data
, LM_ST_SHARED
);
1545 struct lm_async_cb
*async
= data
;
1546 struct gfs2_glock
*gl
;
1548 down_read(&gfs2_umount_flush_sem
);
1549 gl
= gfs2_glock_find(sdp
, &async
->lc_name
);
1550 if (gfs2_assert_warn(sdp
, gl
))
1552 if (!gfs2_assert_warn(sdp
, gl
->gl_req_bh
))
1553 gl
->gl_req_bh(gl
, async
->lc_ret
);
1554 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
1556 up_read(&gfs2_umount_flush_sem
);
1560 case LM_CB_NEED_RECOVERY
:
1561 gfs2_jdesc_make_dirty(sdp
, *(unsigned int *)data
);
1562 if (sdp
->sd_recoverd_process
)
1563 wake_up_process(sdp
->sd_recoverd_process
);
1566 case LM_CB_DROPLOCKS
:
1567 gfs2_gl_hash_clear(sdp
, NO_WAIT
);
1568 gfs2_quota_scan(sdp
);
1572 gfs2_assert_warn(sdp
, 0);
1578 * demote_ok - Check to see if it's ok to unlock a glock
1581 * Returns: 1 if it's ok
1584 static int demote_ok(struct gfs2_glock
*gl
)
1586 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1589 if (test_bit(GLF_STICKY
, &gl
->gl_flags
))
1591 else if (glops
->go_demote_ok
)
1592 demote
= glops
->go_demote_ok(gl
);
1598 * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1603 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock
*gl
)
1605 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
1607 spin_lock(&sdp
->sd_reclaim_lock
);
1608 if (list_empty(&gl
->gl_reclaim
)) {
1609 gfs2_glock_hold(gl
);
1610 list_add(&gl
->gl_reclaim
, &sdp
->sd_reclaim_list
);
1611 atomic_inc(&sdp
->sd_reclaim_count
);
1613 spin_unlock(&sdp
->sd_reclaim_lock
);
1615 wake_up(&sdp
->sd_reclaim_wq
);
1619 * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1620 * @sdp: the filesystem
1622 * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1623 * different glock and we notice that there are a lot of glocks in the
1628 void gfs2_reclaim_glock(struct gfs2_sbd
*sdp
)
1630 struct gfs2_glock
*gl
;
1632 spin_lock(&sdp
->sd_reclaim_lock
);
1633 if (list_empty(&sdp
->sd_reclaim_list
)) {
1634 spin_unlock(&sdp
->sd_reclaim_lock
);
1637 gl
= list_entry(sdp
->sd_reclaim_list
.next
,
1638 struct gfs2_glock
, gl_reclaim
);
1639 list_del_init(&gl
->gl_reclaim
);
1640 spin_unlock(&sdp
->sd_reclaim_lock
);
1642 atomic_dec(&sdp
->sd_reclaim_count
);
1643 atomic_inc(&sdp
->sd_reclaimed
);
1645 if (gfs2_glmutex_trylock(gl
)) {
1646 if (list_empty(&gl
->gl_holders
) &&
1647 gl
->gl_state
!= LM_ST_UNLOCKED
&& demote_ok(gl
))
1648 handle_callback(gl
, LM_ST_UNLOCKED
, 0, 0);
1649 gfs2_glmutex_unlock(gl
);
1656 * examine_bucket - Call a function for glock in a hash bucket
1657 * @examiner: the function
1658 * @sdp: the filesystem
1659 * @bucket: the bucket
1661 * Returns: 1 if the bucket has entries
1664 static int examine_bucket(glock_examiner examiner
, struct gfs2_sbd
*sdp
,
1667 struct gfs2_glock
*gl
, *prev
= NULL
;
1668 int has_entries
= 0;
1669 struct hlist_head
*head
= &gl_hash_table
[hash
].hb_list
;
1671 read_lock(gl_lock_addr(hash
));
1672 /* Can't use hlist_for_each_entry - don't want prefetch here */
1673 if (hlist_empty(head
))
1675 gl
= list_entry(head
->first
, struct gfs2_glock
, gl_list
);
1677 if (!sdp
|| gl
->gl_sbd
== sdp
) {
1678 gfs2_glock_hold(gl
);
1679 read_unlock(gl_lock_addr(hash
));
1681 gfs2_glock_put(prev
);
1685 read_lock(gl_lock_addr(hash
));
1687 if (gl
->gl_list
.next
== NULL
)
1689 gl
= list_entry(gl
->gl_list
.next
, struct gfs2_glock
, gl_list
);
1692 read_unlock(gl_lock_addr(hash
));
1694 gfs2_glock_put(prev
);
1700 * scan_glock - look at a glock and see if we can reclaim it
1701 * @gl: the glock to look at
1705 static void scan_glock(struct gfs2_glock
*gl
)
1707 if (gl
->gl_ops
== &gfs2_inode_glops
&& gl
->gl_object
)
1710 if (gfs2_glmutex_trylock(gl
)) {
1711 if (list_empty(&gl
->gl_holders
) &&
1712 gl
->gl_state
!= LM_ST_UNLOCKED
&& demote_ok(gl
))
1714 gfs2_glmutex_unlock(gl
);
1719 gfs2_glmutex_unlock(gl
);
1720 gfs2_glock_schedule_for_reclaim(gl
);
1724 * clear_glock - look at a glock and see if we can free it from glock cache
1725 * @gl: the glock to look at
1729 static void clear_glock(struct gfs2_glock
*gl
)
1731 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
1734 spin_lock(&sdp
->sd_reclaim_lock
);
1735 if (!list_empty(&gl
->gl_reclaim
)) {
1736 list_del_init(&gl
->gl_reclaim
);
1737 atomic_dec(&sdp
->sd_reclaim_count
);
1738 spin_unlock(&sdp
->sd_reclaim_lock
);
1739 released
= gfs2_glock_put(gl
);
1740 gfs2_assert(sdp
, !released
);
1742 spin_unlock(&sdp
->sd_reclaim_lock
);
1745 if (gfs2_glmutex_trylock(gl
)) {
1746 if (list_empty(&gl
->gl_holders
) &&
1747 gl
->gl_state
!= LM_ST_UNLOCKED
)
1748 handle_callback(gl
, LM_ST_UNLOCKED
, 0, 0);
1749 gfs2_glmutex_unlock(gl
);
1754 * gfs2_gl_hash_clear - Empty out the glock hash table
1755 * @sdp: the filesystem
1756 * @wait: wait until it's all gone
1758 * Called when unmounting the filesystem, or when inter-node lock manager
1759 * requests DROPLOCKS because it is running out of capacity.
1762 void gfs2_gl_hash_clear(struct gfs2_sbd
*sdp
, int wait
)
1772 for (x
= 0; x
< GFS2_GL_HASH_SIZE
; x
++) {
1773 if (examine_bucket(clear_glock
, sdp
, x
))
1780 if (time_after_eq(jiffies
,
1781 t
+ gfs2_tune_get(sdp
, gt_stall_secs
) * HZ
)) {
1782 fs_warn(sdp
, "Unmount seems to be stalled. "
1783 "Dumping lock state...\n");
1784 gfs2_dump_lockstate(sdp
);
1788 down_write(&gfs2_umount_flush_sem
);
1789 invalidate_inodes(sdp
->sd_vfs
);
1790 up_write(&gfs2_umount_flush_sem
);
1796 * Diagnostic routines to help debug distributed deadlock
1799 static void gfs2_print_symbol(struct glock_iter
*gi
, const char *fmt
,
1800 unsigned long address
)
1802 char buffer
[KSYM_SYMBOL_LEN
];
1804 sprint_symbol(buffer
, address
);
1805 print_dbg(gi
, fmt
, buffer
);
1809 * dump_holder - print information about a glock holder
1810 * @str: a string naming the type of holder
1811 * @gh: the glock holder
1813 * Returns: 0 on success, -ENOBUFS when we run out of space
1816 static int dump_holder(struct glock_iter
*gi
, char *str
,
1817 struct gfs2_holder
*gh
)
1820 struct task_struct
*gh_owner
;
1822 print_dbg(gi
, " %s\n", str
);
1823 if (gh
->gh_owner_pid
) {
1824 print_dbg(gi
, " owner = %ld ", (long)gh
->gh_owner_pid
);
1825 gh_owner
= find_task_by_pid(gh
->gh_owner_pid
);
1827 print_dbg(gi
, "(%s)\n", gh_owner
->comm
);
1829 print_dbg(gi
, "(ended)\n");
1831 print_dbg(gi
, " owner = -1\n");
1832 print_dbg(gi
, " gh_state = %u\n", gh
->gh_state
);
1833 print_dbg(gi
, " gh_flags =");
1834 for (x
= 0; x
< 32; x
++)
1835 if (gh
->gh_flags
& (1 << x
))
1836 print_dbg(gi
, " %u", x
);
1837 print_dbg(gi
, " \n");
1838 print_dbg(gi
, " error = %d\n", gh
->gh_error
);
1839 print_dbg(gi
, " gh_iflags =");
1840 for (x
= 0; x
< 32; x
++)
1841 if (test_bit(x
, &gh
->gh_iflags
))
1842 print_dbg(gi
, " %u", x
);
1843 print_dbg(gi
, " \n");
1844 gfs2_print_symbol(gi
, " initialized at: %s\n", gh
->gh_ip
);
1850 * dump_inode - print information about an inode
1853 * Returns: 0 on success, -ENOBUFS when we run out of space
1856 static int dump_inode(struct glock_iter
*gi
, struct gfs2_inode
*ip
)
1860 print_dbg(gi
, " Inode:\n");
1861 print_dbg(gi
, " num = %llu/%llu\n",
1862 (unsigned long long)ip
->i_no_formal_ino
,
1863 (unsigned long long)ip
->i_no_addr
);
1864 print_dbg(gi
, " type = %u\n", IF2DT(ip
->i_inode
.i_mode
));
1865 print_dbg(gi
, " i_flags =");
1866 for (x
= 0; x
< 32; x
++)
1867 if (test_bit(x
, &ip
->i_flags
))
1868 print_dbg(gi
, " %u", x
);
1869 print_dbg(gi
, " \n");
1874 * dump_glock - print information about a glock
1876 * @count: where we are in the buffer
1878 * Returns: 0 on success, -ENOBUFS when we run out of space
1881 static int dump_glock(struct glock_iter
*gi
, struct gfs2_glock
*gl
)
1883 struct gfs2_holder
*gh
;
1885 int error
= -ENOBUFS
;
1886 struct task_struct
*gl_owner
;
1888 spin_lock(&gl
->gl_spin
);
1890 print_dbg(gi
, "Glock 0x%p (%u, 0x%llx)\n", gl
, gl
->gl_name
.ln_type
,
1891 (unsigned long long)gl
->gl_name
.ln_number
);
1892 print_dbg(gi
, " gl_flags =");
1893 for (x
= 0; x
< 32; x
++) {
1894 if (test_bit(x
, &gl
->gl_flags
))
1895 print_dbg(gi
, " %u", x
);
1897 if (!test_bit(GLF_LOCK
, &gl
->gl_flags
))
1898 print_dbg(gi
, " (unlocked)");
1899 print_dbg(gi
, " \n");
1900 print_dbg(gi
, " gl_ref = %d\n", atomic_read(&gl
->gl_ref
));
1901 print_dbg(gi
, " gl_state = %u\n", gl
->gl_state
);
1902 if (gl
->gl_owner_pid
) {
1903 gl_owner
= find_task_by_pid(gl
->gl_owner_pid
);
1905 print_dbg(gi
, " gl_owner = pid %d (%s)\n",
1906 gl
->gl_owner_pid
, gl_owner
->comm
);
1908 print_dbg(gi
, " gl_owner = %d (ended)\n",
1911 print_dbg(gi
, " gl_owner = -1\n");
1912 print_dbg(gi
, " gl_ip = %lu\n", gl
->gl_ip
);
1913 print_dbg(gi
, " req_gh = %s\n", (gl
->gl_req_gh
) ? "yes" : "no");
1914 print_dbg(gi
, " req_bh = %s\n", (gl
->gl_req_bh
) ? "yes" : "no");
1915 print_dbg(gi
, " lvb_count = %d\n", atomic_read(&gl
->gl_lvb_count
));
1916 print_dbg(gi
, " object = %s\n", (gl
->gl_object
) ? "yes" : "no");
1917 print_dbg(gi
, " le = %s\n",
1918 (list_empty(&gl
->gl_le
.le_list
)) ? "no" : "yes");
1919 print_dbg(gi
, " reclaim = %s\n",
1920 (list_empty(&gl
->gl_reclaim
)) ? "no" : "yes");
1922 print_dbg(gi
, " aspace = 0x%p nrpages = %lu\n", gl
->gl_aspace
,
1923 gl
->gl_aspace
->i_mapping
->nrpages
);
1925 print_dbg(gi
, " aspace = no\n");
1926 print_dbg(gi
, " ail = %d\n", atomic_read(&gl
->gl_ail_count
));
1927 if (gl
->gl_req_gh
) {
1928 error
= dump_holder(gi
, "Request", gl
->gl_req_gh
);
1932 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
) {
1933 error
= dump_holder(gi
, "Holder", gh
);
1937 list_for_each_entry(gh
, &gl
->gl_waiters1
, gh_list
) {
1938 error
= dump_holder(gi
, "Waiter1", gh
);
1942 list_for_each_entry(gh
, &gl
->gl_waiters3
, gh_list
) {
1943 error
= dump_holder(gi
, "Waiter3", gh
);
1947 if (test_bit(GLF_DEMOTE
, &gl
->gl_flags
)) {
1948 print_dbg(gi
, " Demotion req to state %u (%llu uS ago)\n",
1949 gl
->gl_demote_state
, (unsigned long long)
1950 (jiffies
- gl
->gl_demote_time
)*(1000000/HZ
));
1952 if (gl
->gl_ops
== &gfs2_inode_glops
&& gl
->gl_object
) {
1953 if (!test_bit(GLF_LOCK
, &gl
->gl_flags
) &&
1954 list_empty(&gl
->gl_holders
)) {
1955 error
= dump_inode(gi
, gl
->gl_object
);
1960 print_dbg(gi
, " Inode: busy\n");
1967 spin_unlock(&gl
->gl_spin
);
1972 * gfs2_dump_lockstate - print out the current lockstate
1973 * @sdp: the filesystem
1974 * @ub: the buffer to copy the information into
1976 * If @ub is NULL, dump the lockstate to the console.
1980 static int gfs2_dump_lockstate(struct gfs2_sbd
*sdp
)
1982 struct gfs2_glock
*gl
;
1983 struct hlist_node
*h
;
1987 for (x
= 0; x
< GFS2_GL_HASH_SIZE
; x
++) {
1989 read_lock(gl_lock_addr(x
));
1991 hlist_for_each_entry(gl
, h
, &gl_hash_table
[x
].hb_list
, gl_list
) {
1992 if (gl
->gl_sbd
!= sdp
)
1995 error
= dump_glock(NULL
, gl
);
2000 read_unlock(gl_lock_addr(x
));
2011 * gfs2_scand - Look for cached glocks and inodes to toss from memory
2012 * @sdp: Pointer to GFS2 superblock
2014 * One of these daemons runs, finding candidates to add to sd_reclaim_list.
2018 static int gfs2_scand(void *data
)
2023 while (!kthread_should_stop()) {
2024 for (x
= 0; x
< GFS2_GL_HASH_SIZE
; x
++)
2025 examine_bucket(scan_glock
, NULL
, x
);
2026 if (freezing(current
))
2031 schedule_timeout_interruptible(delay
* HZ
);
2039 int __init
gfs2_glock_init(void)
2042 for(i
= 0; i
< GFS2_GL_HASH_SIZE
; i
++) {
2043 INIT_HLIST_HEAD(&gl_hash_table
[i
].hb_list
);
2045 #ifdef GL_HASH_LOCK_SZ
2046 for(i
= 0; i
< GL_HASH_LOCK_SZ
; i
++) {
2047 rwlock_init(&gl_hash_locks
[i
]);
2051 scand_process
= kthread_run(gfs2_scand
, NULL
, "gfs2_scand");
2052 if (IS_ERR(scand_process
))
2053 return PTR_ERR(scand_process
);
2055 glock_workqueue
= create_workqueue("glock_workqueue");
2056 if (IS_ERR(glock_workqueue
)) {
2057 kthread_stop(scand_process
);
2058 return PTR_ERR(glock_workqueue
);
2064 void gfs2_glock_exit(void)
2066 destroy_workqueue(glock_workqueue
);
2067 kthread_stop(scand_process
);
2070 module_param(scand_secs
, uint
, S_IRUGO
|S_IWUSR
);
2071 MODULE_PARM_DESC(scand_secs
, "The number of seconds between scand runs");
2073 static int gfs2_glock_iter_next(struct glock_iter
*gi
)
2075 struct gfs2_glock
*gl
;
2078 read_lock(gl_lock_addr(gi
->hash
));
2081 gi
->gl
= hlist_entry(gl
->gl_list
.next
,
2082 struct gfs2_glock
, gl_list
);
2084 gfs2_glock_hold(gi
->gl
);
2086 read_unlock(gl_lock_addr(gi
->hash
));
2089 if (gl
&& gi
->gl
== NULL
)
2091 while(gi
->gl
== NULL
) {
2092 if (gi
->hash
>= GFS2_GL_HASH_SIZE
)
2094 read_lock(gl_lock_addr(gi
->hash
));
2095 gi
->gl
= hlist_entry(gl_hash_table
[gi
->hash
].hb_list
.first
,
2096 struct gfs2_glock
, gl_list
);
2098 gfs2_glock_hold(gi
->gl
);
2099 read_unlock(gl_lock_addr(gi
->hash
));
2103 if (gi
->sdp
!= gi
->gl
->gl_sbd
)
2109 static void gfs2_glock_iter_free(struct glock_iter
*gi
)
2112 gfs2_glock_put(gi
->gl
);
2116 static struct glock_iter
*gfs2_glock_iter_init(struct gfs2_sbd
*sdp
)
2118 struct glock_iter
*gi
;
2120 gi
= kmalloc(sizeof (*gi
), GFP_KERNEL
);
2128 memset(gi
->string
, 0, sizeof(gi
->string
));
2130 if (gfs2_glock_iter_next(gi
)) {
2131 gfs2_glock_iter_free(gi
);
2138 static void *gfs2_glock_seq_start(struct seq_file
*file
, loff_t
*pos
)
2140 struct glock_iter
*gi
;
2143 gi
= gfs2_glock_iter_init(file
->private);
2148 if (gfs2_glock_iter_next(gi
)) {
2149 gfs2_glock_iter_free(gi
);
2157 static void *gfs2_glock_seq_next(struct seq_file
*file
, void *iter_ptr
,
2160 struct glock_iter
*gi
= iter_ptr
;
2164 if (gfs2_glock_iter_next(gi
)) {
2165 gfs2_glock_iter_free(gi
);
2172 static void gfs2_glock_seq_stop(struct seq_file
*file
, void *iter_ptr
)
2174 struct glock_iter
*gi
= iter_ptr
;
2176 gfs2_glock_iter_free(gi
);
2179 static int gfs2_glock_seq_show(struct seq_file
*file
, void *iter_ptr
)
2181 struct glock_iter
*gi
= iter_ptr
;
2184 dump_glock(gi
, gi
->gl
);
2189 static const struct seq_operations gfs2_glock_seq_ops
= {
2190 .start
= gfs2_glock_seq_start
,
2191 .next
= gfs2_glock_seq_next
,
2192 .stop
= gfs2_glock_seq_stop
,
2193 .show
= gfs2_glock_seq_show
,
2196 static int gfs2_debugfs_open(struct inode
*inode
, struct file
*file
)
2198 struct seq_file
*seq
;
2201 ret
= seq_open(file
, &gfs2_glock_seq_ops
);
2205 seq
= file
->private_data
;
2206 seq
->private = inode
->i_private
;
2211 static const struct file_operations gfs2_debug_fops
= {
2212 .owner
= THIS_MODULE
,
2213 .open
= gfs2_debugfs_open
,
2215 .llseek
= seq_lseek
,
2216 .release
= seq_release
2219 int gfs2_create_debugfs_file(struct gfs2_sbd
*sdp
)
2221 sdp
->debugfs_dir
= debugfs_create_dir(sdp
->sd_table_name
, gfs2_root
);
2222 if (!sdp
->debugfs_dir
)
2224 sdp
->debugfs_dentry_glocks
= debugfs_create_file("glocks",
2226 sdp
->debugfs_dir
, sdp
,
2228 if (!sdp
->debugfs_dentry_glocks
)
2234 void gfs2_delete_debugfs_file(struct gfs2_sbd
*sdp
)
2236 if (sdp
&& sdp
->debugfs_dir
) {
2237 if (sdp
->debugfs_dentry_glocks
) {
2238 debugfs_remove(sdp
->debugfs_dentry_glocks
);
2239 sdp
->debugfs_dentry_glocks
= NULL
;
2241 debugfs_remove(sdp
->debugfs_dir
);
2242 sdp
->debugfs_dir
= NULL
;
2246 int gfs2_register_debugfs(void)
2248 gfs2_root
= debugfs_create_dir("gfs2", NULL
);
2249 return gfs2_root
? 0 : -ENOMEM
;
2252 void gfs2_unregister_debugfs(void)
2254 debugfs_remove(gfs2_root
);