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 gfs2_assert(sdp
, gl
->gl_state
== LM_ST_UNLOCKED
);
221 gfs2_assert(sdp
, list_empty(&gl
->gl_reclaim
));
222 gfs2_assert(sdp
, list_empty(&gl
->gl_holders
));
223 gfs2_assert(sdp
, list_empty(&gl
->gl_waiters1
));
224 gfs2_assert(sdp
, list_empty(&gl
->gl_waiters3
));
229 write_unlock(gl_lock_addr(gl
->gl_hash
));
235 * search_bucket() - Find struct gfs2_glock by lock number
236 * @bucket: the bucket to search
237 * @name: The lock name
239 * Returns: NULL, or the struct gfs2_glock with the requested number
242 static struct gfs2_glock
*search_bucket(unsigned int hash
,
243 const struct gfs2_sbd
*sdp
,
244 const struct lm_lockname
*name
)
246 struct gfs2_glock
*gl
;
247 struct hlist_node
*h
;
249 hlist_for_each_entry(gl
, h
, &gl_hash_table
[hash
].hb_list
, gl_list
) {
250 if (!lm_name_equal(&gl
->gl_name
, name
))
252 if (gl
->gl_sbd
!= sdp
)
255 atomic_inc(&gl
->gl_ref
);
264 * gfs2_glock_find() - Find glock by lock number
265 * @sdp: The GFS2 superblock
266 * @name: The lock name
268 * Returns: NULL, or the struct gfs2_glock with the requested number
271 static struct gfs2_glock
*gfs2_glock_find(const struct gfs2_sbd
*sdp
,
272 const struct lm_lockname
*name
)
274 unsigned int hash
= gl_hash(sdp
, name
);
275 struct gfs2_glock
*gl
;
277 read_lock(gl_lock_addr(hash
));
278 gl
= search_bucket(hash
, sdp
, name
);
279 read_unlock(gl_lock_addr(hash
));
284 static void glock_work_func(struct work_struct
*work
)
286 struct gfs2_glock
*gl
= container_of(work
, struct gfs2_glock
, gl_work
.work
);
288 spin_lock(&gl
->gl_spin
);
289 if (test_and_clear_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
))
290 set_bit(GLF_DEMOTE
, &gl
->gl_flags
);
292 spin_unlock(&gl
->gl_spin
);
297 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
298 * @sdp: The GFS2 superblock
299 * @number: the lock number
300 * @glops: The glock_operations to use
301 * @create: If 0, don't create the glock if it doesn't exist
302 * @glp: the glock is returned here
304 * This does not lock a glock, just finds/creates structures for one.
309 int gfs2_glock_get(struct gfs2_sbd
*sdp
, u64 number
,
310 const struct gfs2_glock_operations
*glops
, int create
,
311 struct gfs2_glock
**glp
)
313 struct lm_lockname name
= { .ln_number
= number
, .ln_type
= glops
->go_type
};
314 struct gfs2_glock
*gl
, *tmp
;
315 unsigned int hash
= gl_hash(sdp
, &name
);
318 read_lock(gl_lock_addr(hash
));
319 gl
= search_bucket(hash
, sdp
, &name
);
320 read_unlock(gl_lock_addr(hash
));
327 gl
= kmem_cache_alloc(gfs2_glock_cachep
, GFP_KERNEL
);
333 atomic_set(&gl
->gl_ref
, 1);
334 gl
->gl_state
= LM_ST_UNLOCKED
;
335 gl
->gl_demote_state
= LM_ST_EXCLUSIVE
;
337 gl
->gl_owner_pid
= 0;
340 gl
->gl_req_gh
= NULL
;
341 gl
->gl_req_bh
= NULL
;
343 gl
->gl_stamp
= jiffies
;
344 gl
->gl_tchange
= jiffies
;
345 gl
->gl_object
= NULL
;
347 gl
->gl_aspace
= NULL
;
348 INIT_DELAYED_WORK(&gl
->gl_work
, glock_work_func
);
350 /* If this glock protects actual on-disk data or metadata blocks,
351 create a VFS inode to manage the pages/buffers holding them. */
352 if (glops
== &gfs2_inode_glops
|| glops
== &gfs2_rgrp_glops
) {
353 gl
->gl_aspace
= gfs2_aspace_get(sdp
);
354 if (!gl
->gl_aspace
) {
360 error
= gfs2_lm_get_lock(sdp
, &name
, &gl
->gl_lock
);
364 write_lock(gl_lock_addr(hash
));
365 tmp
= search_bucket(hash
, sdp
, &name
);
367 write_unlock(gl_lock_addr(hash
));
371 hlist_add_head(&gl
->gl_list
, &gl_hash_table
[hash
].hb_list
);
372 write_unlock(gl_lock_addr(hash
));
381 gfs2_aspace_put(gl
->gl_aspace
);
383 kmem_cache_free(gfs2_glock_cachep
, gl
);
388 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
390 * @state: the state we're requesting
391 * @flags: the modifier flags
392 * @gh: the holder structure
396 void gfs2_holder_init(struct gfs2_glock
*gl
, unsigned int state
, unsigned flags
,
397 struct gfs2_holder
*gh
)
399 INIT_LIST_HEAD(&gh
->gh_list
);
401 gh
->gh_ip
= (unsigned long)__builtin_return_address(0);
402 gh
->gh_owner_pid
= current
->pid
;
403 gh
->gh_state
= state
;
404 gh
->gh_flags
= flags
;
411 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
412 * @state: the state we're requesting
413 * @flags: the modifier flags
414 * @gh: the holder structure
416 * Don't mess with the glock.
420 void gfs2_holder_reinit(unsigned int state
, unsigned flags
, struct gfs2_holder
*gh
)
422 gh
->gh_state
= state
;
423 gh
->gh_flags
= flags
;
425 gh
->gh_ip
= (unsigned long)__builtin_return_address(0);
429 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
430 * @gh: the holder structure
434 void gfs2_holder_uninit(struct gfs2_holder
*gh
)
436 gfs2_glock_put(gh
->gh_gl
);
441 static void gfs2_holder_wake(struct gfs2_holder
*gh
)
443 clear_bit(HIF_WAIT
, &gh
->gh_iflags
);
444 smp_mb__after_clear_bit();
445 wake_up_bit(&gh
->gh_iflags
, HIF_WAIT
);
448 static int just_schedule(void *word
)
454 static void wait_on_holder(struct gfs2_holder
*gh
)
457 wait_on_bit(&gh
->gh_iflags
, HIF_WAIT
, just_schedule
, TASK_UNINTERRUPTIBLE
);
460 static void gfs2_demote_wake(struct gfs2_glock
*gl
)
462 gl
->gl_demote_state
= LM_ST_EXCLUSIVE
;
463 clear_bit(GLF_DEMOTE
, &gl
->gl_flags
);
464 smp_mb__after_clear_bit();
465 wake_up_bit(&gl
->gl_flags
, GLF_DEMOTE
);
468 static void wait_on_demote(struct gfs2_glock
*gl
)
471 wait_on_bit(&gl
->gl_flags
, GLF_DEMOTE
, just_schedule
, TASK_UNINTERRUPTIBLE
);
475 * rq_mutex - process a mutex request in the queue
476 * @gh: the glock holder
478 * Returns: 1 if the queue is blocked
481 static int rq_mutex(struct gfs2_holder
*gh
)
483 struct gfs2_glock
*gl
= gh
->gh_gl
;
485 list_del_init(&gh
->gh_list
);
486 /* gh->gh_error never examined. */
487 set_bit(GLF_LOCK
, &gl
->gl_flags
);
488 clear_bit(HIF_WAIT
, &gh
->gh_iflags
);
490 wake_up_bit(&gh
->gh_iflags
, HIF_WAIT
);
496 * rq_promote - process a promote request in the queue
497 * @gh: the glock holder
499 * Acquire a new inter-node lock, or change a lock state to more restrictive.
501 * Returns: 1 if the queue is blocked
504 static int rq_promote(struct gfs2_holder
*gh
)
506 struct gfs2_glock
*gl
= gh
->gh_gl
;
508 if (!relaxed_state_ok(gl
->gl_state
, gh
->gh_state
, gh
->gh_flags
)) {
509 if (list_empty(&gl
->gl_holders
)) {
511 set_bit(GLF_LOCK
, &gl
->gl_flags
);
512 spin_unlock(&gl
->gl_spin
);
513 gfs2_glock_xmote_th(gh
->gh_gl
, gh
);
514 spin_lock(&gl
->gl_spin
);
519 if (list_empty(&gl
->gl_holders
)) {
520 set_bit(HIF_FIRST
, &gh
->gh_iflags
);
521 set_bit(GLF_LOCK
, &gl
->gl_flags
);
523 struct gfs2_holder
*next_gh
;
524 if (gh
->gh_state
== LM_ST_EXCLUSIVE
)
526 next_gh
= list_entry(gl
->gl_holders
.next
, struct gfs2_holder
,
528 if (next_gh
->gh_state
== LM_ST_EXCLUSIVE
)
532 list_move_tail(&gh
->gh_list
, &gl
->gl_holders
);
534 set_bit(HIF_HOLDER
, &gh
->gh_iflags
);
536 gfs2_holder_wake(gh
);
542 * rq_demote - process a demote request in the queue
543 * @gh: the glock holder
545 * Returns: 1 if the queue is blocked
548 static int rq_demote(struct gfs2_glock
*gl
)
550 if (!list_empty(&gl
->gl_holders
))
553 if (gl
->gl_state
== gl
->gl_demote_state
||
554 gl
->gl_state
== LM_ST_UNLOCKED
) {
555 gfs2_demote_wake(gl
);
559 set_bit(GLF_LOCK
, &gl
->gl_flags
);
560 set_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
);
562 if (gl
->gl_demote_state
== LM_ST_UNLOCKED
||
563 gl
->gl_state
!= LM_ST_EXCLUSIVE
) {
564 spin_unlock(&gl
->gl_spin
);
565 gfs2_glock_drop_th(gl
);
567 spin_unlock(&gl
->gl_spin
);
568 gfs2_glock_xmote_th(gl
, NULL
);
571 spin_lock(&gl
->gl_spin
);
572 clear_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
);
578 * run_queue - process holder structures on a glock
582 static void run_queue(struct gfs2_glock
*gl
)
584 struct gfs2_holder
*gh
;
588 if (test_bit(GLF_LOCK
, &gl
->gl_flags
))
591 if (!list_empty(&gl
->gl_waiters1
)) {
592 gh
= list_entry(gl
->gl_waiters1
.next
,
593 struct gfs2_holder
, gh_list
);
594 blocked
= rq_mutex(gh
);
595 } else if (test_bit(GLF_DEMOTE
, &gl
->gl_flags
)) {
596 blocked
= rq_demote(gl
);
597 if (gl
->gl_waiters2
&& !blocked
) {
598 set_bit(GLF_DEMOTE
, &gl
->gl_flags
);
599 gl
->gl_demote_state
= LM_ST_UNLOCKED
;
602 } else if (!list_empty(&gl
->gl_waiters3
)) {
603 gh
= list_entry(gl
->gl_waiters3
.next
,
604 struct gfs2_holder
, gh_list
);
605 blocked
= rq_promote(gh
);
615 * gfs2_glmutex_lock - acquire a local lock on a glock
618 * Gives caller exclusive access to manipulate a glock structure.
621 static void gfs2_glmutex_lock(struct gfs2_glock
*gl
)
623 struct gfs2_holder gh
;
625 gfs2_holder_init(gl
, 0, 0, &gh
);
626 if (test_and_set_bit(HIF_WAIT
, &gh
.gh_iflags
))
629 spin_lock(&gl
->gl_spin
);
630 if (test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
)) {
631 list_add_tail(&gh
.gh_list
, &gl
->gl_waiters1
);
633 gl
->gl_owner_pid
= current
->pid
;
634 gl
->gl_ip
= (unsigned long)__builtin_return_address(0);
635 clear_bit(HIF_WAIT
, &gh
.gh_iflags
);
637 wake_up_bit(&gh
.gh_iflags
, HIF_WAIT
);
639 spin_unlock(&gl
->gl_spin
);
642 gfs2_holder_uninit(&gh
);
646 * gfs2_glmutex_trylock - try to acquire a local lock on a glock
649 * Returns: 1 if the glock is acquired
652 static int gfs2_glmutex_trylock(struct gfs2_glock
*gl
)
656 spin_lock(&gl
->gl_spin
);
657 if (test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
)) {
660 gl
->gl_owner_pid
= current
->pid
;
661 gl
->gl_ip
= (unsigned long)__builtin_return_address(0);
663 spin_unlock(&gl
->gl_spin
);
669 * gfs2_glmutex_unlock - release a local lock on a glock
674 static void gfs2_glmutex_unlock(struct gfs2_glock
*gl
)
676 spin_lock(&gl
->gl_spin
);
677 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
678 gl
->gl_owner_pid
= 0;
681 spin_unlock(&gl
->gl_spin
);
685 * handle_callback - process a demote request
687 * @state: the state the caller wants us to change to
689 * There are only two requests that we are going to see in actual
690 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
693 static void handle_callback(struct gfs2_glock
*gl
, unsigned int state
,
694 int remote
, unsigned long delay
)
696 int bit
= delay
? GLF_PENDING_DEMOTE
: GLF_DEMOTE
;
698 spin_lock(&gl
->gl_spin
);
699 set_bit(bit
, &gl
->gl_flags
);
700 if (gl
->gl_demote_state
== LM_ST_EXCLUSIVE
) {
701 gl
->gl_demote_state
= state
;
702 gl
->gl_demote_time
= jiffies
;
703 if (remote
&& gl
->gl_ops
->go_type
== LM_TYPE_IOPEN
&&
705 gfs2_glock_schedule_for_reclaim(gl
);
706 spin_unlock(&gl
->gl_spin
);
709 } else if (gl
->gl_demote_state
!= LM_ST_UNLOCKED
&&
710 gl
->gl_demote_state
!= state
) {
711 if (test_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
))
714 gl
->gl_demote_state
= LM_ST_UNLOCKED
;
716 spin_unlock(&gl
->gl_spin
);
720 * state_change - record that the glock is now in a different state
722 * @new_state the new state
726 static void state_change(struct gfs2_glock
*gl
, unsigned int new_state
)
730 held1
= (gl
->gl_state
!= LM_ST_UNLOCKED
);
731 held2
= (new_state
!= LM_ST_UNLOCKED
);
733 if (held1
!= held2
) {
740 gl
->gl_state
= new_state
;
741 gl
->gl_tchange
= jiffies
;
745 * xmote_bh - Called after the lock module is done acquiring a lock
746 * @gl: The glock in question
747 * @ret: the int returned from the lock module
751 static void xmote_bh(struct gfs2_glock
*gl
, unsigned int ret
)
753 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
754 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
755 struct gfs2_holder
*gh
= gl
->gl_req_gh
;
756 int prev_state
= gl
->gl_state
;
759 gfs2_assert_warn(sdp
, test_bit(GLF_LOCK
, &gl
->gl_flags
));
760 gfs2_assert_warn(sdp
, list_empty(&gl
->gl_holders
));
761 gfs2_assert_warn(sdp
, !(ret
& LM_OUT_ASYNC
));
763 state_change(gl
, ret
& LM_OUT_ST_MASK
);
765 if (prev_state
!= LM_ST_UNLOCKED
&& !(ret
& LM_OUT_CACHEABLE
)) {
767 glops
->go_inval(gl
, DIO_METADATA
);
768 } else if (gl
->gl_state
== LM_ST_DEFERRED
) {
769 /* We might not want to do this here.
770 Look at moving to the inode glops. */
772 glops
->go_inval(gl
, 0);
775 /* Deal with each possible exit condition */
778 gl
->gl_stamp
= jiffies
;
779 if (ret
& LM_OUT_CANCELED
) {
782 spin_lock(&gl
->gl_spin
);
783 if (gl
->gl_state
!= gl
->gl_demote_state
) {
784 gl
->gl_req_bh
= NULL
;
785 spin_unlock(&gl
->gl_spin
);
786 gfs2_glock_drop_th(gl
);
790 gfs2_demote_wake(gl
);
791 spin_unlock(&gl
->gl_spin
);
794 spin_lock(&gl
->gl_spin
);
795 list_del_init(&gh
->gh_list
);
797 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)))
799 gh
->gh_error
= GLR_CANCELED
;
800 if (ret
& LM_OUT_CANCELED
)
802 if (relaxed_state_ok(gl
->gl_state
, gh
->gh_state
, gh
->gh_flags
)) {
803 list_add_tail(&gh
->gh_list
, &gl
->gl_holders
);
805 set_bit(HIF_HOLDER
, &gh
->gh_iflags
);
806 set_bit(HIF_FIRST
, &gh
->gh_iflags
);
810 gh
->gh_error
= GLR_TRYFAILED
;
811 if (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
))
813 gh
->gh_error
= -EINVAL
;
814 if (gfs2_assert_withdraw(sdp
, 0) == -1)
815 fs_err(sdp
, "ret = 0x%.8X\n", ret
);
817 spin_unlock(&gl
->gl_spin
);
820 if (glops
->go_xmote_bh
)
821 glops
->go_xmote_bh(gl
);
824 spin_lock(&gl
->gl_spin
);
825 gl
->gl_req_gh
= NULL
;
826 gl
->gl_req_bh
= NULL
;
827 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
828 spin_unlock(&gl
->gl_spin
);
834 gfs2_holder_wake(gh
);
838 * gfs2_glock_xmote_th - Call into the lock module to acquire or change a glock
839 * @gl: The glock in question
840 * @state: the requested state
841 * @flags: modifier flags to the lock call
845 static void gfs2_glock_xmote_th(struct gfs2_glock
*gl
, struct gfs2_holder
*gh
)
847 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
848 int flags
= gh
? gh
->gh_flags
: 0;
849 unsigned state
= gh
? gh
->gh_state
: gl
->gl_demote_state
;
850 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
851 int lck_flags
= flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
|
852 LM_FLAG_NOEXP
| LM_FLAG_ANY
|
854 unsigned int lck_ret
;
856 if (glops
->go_xmote_th
)
857 glops
->go_xmote_th(gl
);
859 gfs2_assert_warn(sdp
, test_bit(GLF_LOCK
, &gl
->gl_flags
));
860 gfs2_assert_warn(sdp
, list_empty(&gl
->gl_holders
));
861 gfs2_assert_warn(sdp
, state
!= LM_ST_UNLOCKED
);
862 gfs2_assert_warn(sdp
, state
!= gl
->gl_state
);
865 gl
->gl_req_bh
= xmote_bh
;
867 lck_ret
= gfs2_lm_lock(sdp
, gl
->gl_lock
, gl
->gl_state
, state
, lck_flags
);
869 if (gfs2_assert_withdraw(sdp
, !(lck_ret
& LM_OUT_ERROR
)))
872 if (lck_ret
& LM_OUT_ASYNC
)
873 gfs2_assert_warn(sdp
, lck_ret
== LM_OUT_ASYNC
);
875 xmote_bh(gl
, lck_ret
);
879 * drop_bh - Called after a lock module unlock completes
881 * @ret: the return status
883 * Doesn't wake up the process waiting on the struct gfs2_holder (if any)
884 * Doesn't drop the reference on the glock the top half took out
888 static void drop_bh(struct gfs2_glock
*gl
, unsigned int ret
)
890 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
891 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
892 struct gfs2_holder
*gh
= gl
->gl_req_gh
;
894 gfs2_assert_warn(sdp
, test_bit(GLF_LOCK
, &gl
->gl_flags
));
895 gfs2_assert_warn(sdp
, list_empty(&gl
->gl_holders
));
896 gfs2_assert_warn(sdp
, !ret
);
898 state_change(gl
, LM_ST_UNLOCKED
);
901 glops
->go_inval(gl
, DIO_METADATA
);
904 spin_lock(&gl
->gl_spin
);
905 list_del_init(&gh
->gh_list
);
907 spin_unlock(&gl
->gl_spin
);
910 spin_lock(&gl
->gl_spin
);
911 gfs2_demote_wake(gl
);
912 gl
->gl_req_gh
= NULL
;
913 gl
->gl_req_bh
= NULL
;
914 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
915 spin_unlock(&gl
->gl_spin
);
920 gfs2_holder_wake(gh
);
924 * gfs2_glock_drop_th - call into the lock module to unlock a lock
929 static void gfs2_glock_drop_th(struct gfs2_glock
*gl
)
931 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
932 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
935 if (glops
->go_xmote_th
)
936 glops
->go_xmote_th(gl
);
938 gfs2_assert_warn(sdp
, test_bit(GLF_LOCK
, &gl
->gl_flags
));
939 gfs2_assert_warn(sdp
, list_empty(&gl
->gl_holders
));
940 gfs2_assert_warn(sdp
, gl
->gl_state
!= LM_ST_UNLOCKED
);
943 gl
->gl_req_bh
= drop_bh
;
945 ret
= gfs2_lm_unlock(sdp
, gl
->gl_lock
, gl
->gl_state
);
947 if (gfs2_assert_withdraw(sdp
, !(ret
& LM_OUT_ERROR
)))
953 gfs2_assert_warn(sdp
, ret
== LM_OUT_ASYNC
);
957 * do_cancels - cancel requests for locks stuck waiting on an expire flag
958 * @gh: the LM_FLAG_PRIORITY holder waiting to acquire the lock
960 * Don't cancel GL_NOCANCEL requests.
963 static void do_cancels(struct gfs2_holder
*gh
)
965 struct gfs2_glock
*gl
= gh
->gh_gl
;
967 spin_lock(&gl
->gl_spin
);
969 while (gl
->gl_req_gh
!= gh
&&
970 !test_bit(HIF_HOLDER
, &gh
->gh_iflags
) &&
971 !list_empty(&gh
->gh_list
)) {
972 if (gl
->gl_req_bh
&& !(gl
->gl_req_gh
&&
973 (gl
->gl_req_gh
->gh_flags
& GL_NOCANCEL
))) {
974 spin_unlock(&gl
->gl_spin
);
975 gfs2_lm_cancel(gl
->gl_sbd
, gl
->gl_lock
);
977 spin_lock(&gl
->gl_spin
);
979 spin_unlock(&gl
->gl_spin
);
981 spin_lock(&gl
->gl_spin
);
985 spin_unlock(&gl
->gl_spin
);
989 * glock_wait_internal - wait on a glock acquisition
990 * @gh: the glock holder
992 * Returns: 0 on success
995 static int glock_wait_internal(struct gfs2_holder
*gh
)
997 struct gfs2_glock
*gl
= gh
->gh_gl
;
998 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
999 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1001 if (test_bit(HIF_ABORTED
, &gh
->gh_iflags
))
1004 if (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
)) {
1005 spin_lock(&gl
->gl_spin
);
1006 if (gl
->gl_req_gh
!= gh
&&
1007 !test_bit(HIF_HOLDER
, &gh
->gh_iflags
) &&
1008 !list_empty(&gh
->gh_list
)) {
1009 list_del_init(&gh
->gh_list
);
1010 gh
->gh_error
= GLR_TRYFAILED
;
1012 spin_unlock(&gl
->gl_spin
);
1013 return gh
->gh_error
;
1015 spin_unlock(&gl
->gl_spin
);
1018 if (gh
->gh_flags
& LM_FLAG_PRIORITY
)
1023 return gh
->gh_error
;
1025 gfs2_assert_withdraw(sdp
, test_bit(HIF_HOLDER
, &gh
->gh_iflags
));
1026 gfs2_assert_withdraw(sdp
, relaxed_state_ok(gl
->gl_state
, gh
->gh_state
,
1029 if (test_bit(HIF_FIRST
, &gh
->gh_iflags
)) {
1030 gfs2_assert_warn(sdp
, test_bit(GLF_LOCK
, &gl
->gl_flags
));
1032 if (glops
->go_lock
) {
1033 gh
->gh_error
= glops
->go_lock(gh
);
1035 spin_lock(&gl
->gl_spin
);
1036 list_del_init(&gh
->gh_list
);
1037 spin_unlock(&gl
->gl_spin
);
1041 spin_lock(&gl
->gl_spin
);
1042 gl
->gl_req_gh
= NULL
;
1043 gl
->gl_req_bh
= NULL
;
1044 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
1046 spin_unlock(&gl
->gl_spin
);
1049 return gh
->gh_error
;
1052 static inline struct gfs2_holder
*
1053 find_holder_by_owner(struct list_head
*head
, pid_t pid
)
1055 struct gfs2_holder
*gh
;
1057 list_for_each_entry(gh
, head
, gh_list
) {
1058 if (gh
->gh_owner_pid
== pid
)
1065 static void print_dbg(struct glock_iter
*gi
, const char *fmt
, ...)
1069 va_start(args
, fmt
);
1071 vsprintf(gi
->string
, fmt
, args
);
1072 seq_printf(gi
->seq
, gi
->string
);
1080 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1081 * @gh: the holder structure to add
1085 static void add_to_queue(struct gfs2_holder
*gh
)
1087 struct gfs2_glock
*gl
= gh
->gh_gl
;
1088 struct gfs2_holder
*existing
;
1090 BUG_ON(!gh
->gh_owner_pid
);
1091 if (test_and_set_bit(HIF_WAIT
, &gh
->gh_iflags
))
1094 if (!(gh
->gh_flags
& GL_FLOCK
)) {
1095 existing
= find_holder_by_owner(&gl
->gl_holders
,
1098 print_symbol(KERN_WARNING
"original: %s\n",
1100 printk(KERN_INFO
"pid : %d\n", existing
->gh_owner_pid
);
1101 printk(KERN_INFO
"lock type : %d lock state : %d\n",
1102 existing
->gh_gl
->gl_name
.ln_type
,
1103 existing
->gh_gl
->gl_state
);
1104 print_symbol(KERN_WARNING
"new: %s\n", gh
->gh_ip
);
1105 printk(KERN_INFO
"pid : %d\n", gh
->gh_owner_pid
);
1106 printk(KERN_INFO
"lock type : %d lock state : %d\n",
1107 gl
->gl_name
.ln_type
, gl
->gl_state
);
1111 existing
= find_holder_by_owner(&gl
->gl_waiters3
,
1114 print_symbol(KERN_WARNING
"original: %s\n",
1116 print_symbol(KERN_WARNING
"new: %s\n", gh
->gh_ip
);
1121 if (gh
->gh_flags
& LM_FLAG_PRIORITY
)
1122 list_add(&gh
->gh_list
, &gl
->gl_waiters3
);
1124 list_add_tail(&gh
->gh_list
, &gl
->gl_waiters3
);
1128 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1129 * @gh: the holder structure
1131 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1133 * Returns: 0, GLR_TRYFAILED, or errno on failure
1136 int gfs2_glock_nq(struct gfs2_holder
*gh
)
1138 struct gfs2_glock
*gl
= gh
->gh_gl
;
1139 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
1143 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
))) {
1144 set_bit(HIF_ABORTED
, &gh
->gh_iflags
);
1148 spin_lock(&gl
->gl_spin
);
1151 spin_unlock(&gl
->gl_spin
);
1153 if (!(gh
->gh_flags
& GL_ASYNC
)) {
1154 error
= glock_wait_internal(gh
);
1155 if (error
== GLR_CANCELED
) {
1165 * gfs2_glock_poll - poll to see if an async request has been completed
1168 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1171 int gfs2_glock_poll(struct gfs2_holder
*gh
)
1173 struct gfs2_glock
*gl
= gh
->gh_gl
;
1176 spin_lock(&gl
->gl_spin
);
1178 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
1180 else if (list_empty(&gh
->gh_list
)) {
1181 if (gh
->gh_error
== GLR_CANCELED
) {
1182 spin_unlock(&gl
->gl_spin
);
1184 if (gfs2_glock_nq(gh
))
1191 spin_unlock(&gl
->gl_spin
);
1197 * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
1198 * @gh: the holder structure
1200 * Returns: 0, GLR_TRYFAILED, or errno on failure
1203 int gfs2_glock_wait(struct gfs2_holder
*gh
)
1207 error
= glock_wait_internal(gh
);
1208 if (error
== GLR_CANCELED
) {
1210 gh
->gh_flags
&= ~GL_ASYNC
;
1211 error
= gfs2_glock_nq(gh
);
1218 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1219 * @gh: the glock holder
1223 void gfs2_glock_dq(struct gfs2_holder
*gh
)
1225 struct gfs2_glock
*gl
= gh
->gh_gl
;
1226 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1229 if (gh
->gh_flags
& GL_NOCACHE
)
1230 handle_callback(gl
, LM_ST_UNLOCKED
, 0, 0);
1232 gfs2_glmutex_lock(gl
);
1234 spin_lock(&gl
->gl_spin
);
1235 list_del_init(&gh
->gh_list
);
1237 if (list_empty(&gl
->gl_holders
)) {
1238 if (glops
->go_unlock
) {
1239 spin_unlock(&gl
->gl_spin
);
1240 glops
->go_unlock(gh
);
1241 spin_lock(&gl
->gl_spin
);
1243 gl
->gl_stamp
= jiffies
;
1246 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
1247 spin_unlock(&gl
->gl_spin
);
1249 gfs2_glock_hold(gl
);
1250 if (test_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
) &&
1251 !test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
1252 delay
= gl
->gl_ops
->go_min_hold_time
;
1253 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, delay
) == 0)
1257 void gfs2_glock_dq_wait(struct gfs2_holder
*gh
)
1259 struct gfs2_glock
*gl
= gh
->gh_gl
;
1265 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1266 * @gh: the holder structure
1270 void gfs2_glock_dq_uninit(struct gfs2_holder
*gh
)
1273 gfs2_holder_uninit(gh
);
1277 * gfs2_glock_nq_num - acquire a glock based on lock number
1278 * @sdp: the filesystem
1279 * @number: the lock number
1280 * @glops: the glock operations for the type of glock
1281 * @state: the state to acquire the glock in
1282 * @flags: modifier flags for the aquisition
1283 * @gh: the struct gfs2_holder
1288 int gfs2_glock_nq_num(struct gfs2_sbd
*sdp
, u64 number
,
1289 const struct gfs2_glock_operations
*glops
,
1290 unsigned int state
, int flags
, struct gfs2_holder
*gh
)
1292 struct gfs2_glock
*gl
;
1295 error
= gfs2_glock_get(sdp
, number
, glops
, CREATE
, &gl
);
1297 error
= gfs2_glock_nq_init(gl
, state
, flags
, gh
);
1305 * glock_compare - Compare two struct gfs2_glock structures for sorting
1306 * @arg_a: the first structure
1307 * @arg_b: the second structure
1311 static int glock_compare(const void *arg_a
, const void *arg_b
)
1313 const struct gfs2_holder
*gh_a
= *(const struct gfs2_holder
**)arg_a
;
1314 const struct gfs2_holder
*gh_b
= *(const struct gfs2_holder
**)arg_b
;
1315 const struct lm_lockname
*a
= &gh_a
->gh_gl
->gl_name
;
1316 const struct lm_lockname
*b
= &gh_b
->gh_gl
->gl_name
;
1318 if (a
->ln_number
> b
->ln_number
)
1320 if (a
->ln_number
< b
->ln_number
)
1322 BUG_ON(gh_a
->gh_gl
->gl_ops
->go_type
== gh_b
->gh_gl
->gl_ops
->go_type
);
1327 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1328 * @num_gh: the number of structures
1329 * @ghs: an array of struct gfs2_holder structures
1331 * Returns: 0 on success (all glocks acquired),
1332 * errno on failure (no glocks acquired)
1335 static int nq_m_sync(unsigned int num_gh
, struct gfs2_holder
*ghs
,
1336 struct gfs2_holder
**p
)
1341 for (x
= 0; x
< num_gh
; x
++)
1344 sort(p
, num_gh
, sizeof(struct gfs2_holder
*), glock_compare
, NULL
);
1346 for (x
= 0; x
< num_gh
; x
++) {
1347 p
[x
]->gh_flags
&= ~(LM_FLAG_TRY
| GL_ASYNC
);
1349 error
= gfs2_glock_nq(p
[x
]);
1352 gfs2_glock_dq(p
[x
]);
1361 * gfs2_glock_nq_m - acquire multiple glocks
1362 * @num_gh: the number of structures
1363 * @ghs: an array of struct gfs2_holder structures
1366 * Returns: 0 on success (all glocks acquired),
1367 * errno on failure (no glocks acquired)
1370 int gfs2_glock_nq_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1372 struct gfs2_holder
*tmp
[4];
1373 struct gfs2_holder
**pph
= tmp
;
1380 ghs
->gh_flags
&= ~(LM_FLAG_TRY
| GL_ASYNC
);
1381 return gfs2_glock_nq(ghs
);
1385 pph
= kmalloc(num_gh
* sizeof(struct gfs2_holder
*), GFP_NOFS
);
1390 error
= nq_m_sync(num_gh
, ghs
, pph
);
1399 * gfs2_glock_dq_m - release multiple glocks
1400 * @num_gh: the number of structures
1401 * @ghs: an array of struct gfs2_holder structures
1405 void gfs2_glock_dq_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1409 for (x
= 0; x
< num_gh
; x
++)
1410 gfs2_glock_dq(&ghs
[x
]);
1414 * gfs2_glock_dq_uninit_m - release multiple glocks
1415 * @num_gh: the number of structures
1416 * @ghs: an array of struct gfs2_holder structures
1420 void gfs2_glock_dq_uninit_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1424 for (x
= 0; x
< num_gh
; x
++)
1425 gfs2_glock_dq_uninit(&ghs
[x
]);
1429 * gfs2_lvb_hold - attach a LVB from a glock
1430 * @gl: The glock in question
1434 int gfs2_lvb_hold(struct gfs2_glock
*gl
)
1438 gfs2_glmutex_lock(gl
);
1440 if (!atomic_read(&gl
->gl_lvb_count
)) {
1441 error
= gfs2_lm_hold_lvb(gl
->gl_sbd
, gl
->gl_lock
, &gl
->gl_lvb
);
1443 gfs2_glmutex_unlock(gl
);
1446 gfs2_glock_hold(gl
);
1448 atomic_inc(&gl
->gl_lvb_count
);
1450 gfs2_glmutex_unlock(gl
);
1456 * gfs2_lvb_unhold - detach a LVB from a glock
1457 * @gl: The glock in question
1461 void gfs2_lvb_unhold(struct gfs2_glock
*gl
)
1463 gfs2_glock_hold(gl
);
1464 gfs2_glmutex_lock(gl
);
1466 gfs2_assert(gl
->gl_sbd
, atomic_read(&gl
->gl_lvb_count
) > 0);
1467 if (atomic_dec_and_test(&gl
->gl_lvb_count
)) {
1468 gfs2_lm_unhold_lvb(gl
->gl_sbd
, gl
->gl_lock
, gl
->gl_lvb
);
1473 gfs2_glmutex_unlock(gl
);
1477 static void blocking_cb(struct gfs2_sbd
*sdp
, struct lm_lockname
*name
,
1480 struct gfs2_glock
*gl
;
1481 unsigned long delay
= 0;
1482 unsigned long holdtime
;
1483 unsigned long now
= jiffies
;
1485 gl
= gfs2_glock_find(sdp
, name
);
1489 holdtime
= gl
->gl_tchange
+ gl
->gl_ops
->go_min_hold_time
;
1490 if (time_before(now
, holdtime
))
1491 delay
= holdtime
- now
;
1493 handle_callback(gl
, state
, 1, delay
);
1494 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, delay
) == 0)
1499 * gfs2_glock_cb - Callback used by locking module
1500 * @sdp: Pointer to the superblock
1501 * @type: Type of callback
1502 * @data: Type dependent data pointer
1504 * Called by the locking module when it wants to tell us something.
1505 * Either we need to drop a lock, one of our ASYNC requests completed, or
1506 * a journal from another client needs to be recovered.
1509 void gfs2_glock_cb(void *cb_data
, unsigned int type
, void *data
)
1511 struct gfs2_sbd
*sdp
= cb_data
;
1515 blocking_cb(sdp
, data
, LM_ST_UNLOCKED
);
1519 blocking_cb(sdp
, data
, LM_ST_DEFERRED
);
1523 blocking_cb(sdp
, data
, LM_ST_SHARED
);
1527 struct lm_async_cb
*async
= data
;
1528 struct gfs2_glock
*gl
;
1530 down_read(&gfs2_umount_flush_sem
);
1531 gl
= gfs2_glock_find(sdp
, &async
->lc_name
);
1532 if (gfs2_assert_warn(sdp
, gl
))
1534 if (!gfs2_assert_warn(sdp
, gl
->gl_req_bh
))
1535 gl
->gl_req_bh(gl
, async
->lc_ret
);
1536 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
1538 up_read(&gfs2_umount_flush_sem
);
1542 case LM_CB_NEED_RECOVERY
:
1543 gfs2_jdesc_make_dirty(sdp
, *(unsigned int *)data
);
1544 if (sdp
->sd_recoverd_process
)
1545 wake_up_process(sdp
->sd_recoverd_process
);
1548 case LM_CB_DROPLOCKS
:
1549 gfs2_gl_hash_clear(sdp
, NO_WAIT
);
1550 gfs2_quota_scan(sdp
);
1554 gfs2_assert_warn(sdp
, 0);
1560 * demote_ok - Check to see if it's ok to unlock a glock
1563 * Returns: 1 if it's ok
1566 static int demote_ok(struct gfs2_glock
*gl
)
1568 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1571 if (test_bit(GLF_STICKY
, &gl
->gl_flags
))
1573 else if (glops
->go_demote_ok
)
1574 demote
= glops
->go_demote_ok(gl
);
1580 * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1585 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock
*gl
)
1587 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
1589 spin_lock(&sdp
->sd_reclaim_lock
);
1590 if (list_empty(&gl
->gl_reclaim
)) {
1591 gfs2_glock_hold(gl
);
1592 list_add(&gl
->gl_reclaim
, &sdp
->sd_reclaim_list
);
1593 atomic_inc(&sdp
->sd_reclaim_count
);
1595 spin_unlock(&sdp
->sd_reclaim_lock
);
1597 wake_up(&sdp
->sd_reclaim_wq
);
1601 * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1602 * @sdp: the filesystem
1604 * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1605 * different glock and we notice that there are a lot of glocks in the
1610 void gfs2_reclaim_glock(struct gfs2_sbd
*sdp
)
1612 struct gfs2_glock
*gl
;
1614 spin_lock(&sdp
->sd_reclaim_lock
);
1615 if (list_empty(&sdp
->sd_reclaim_list
)) {
1616 spin_unlock(&sdp
->sd_reclaim_lock
);
1619 gl
= list_entry(sdp
->sd_reclaim_list
.next
,
1620 struct gfs2_glock
, gl_reclaim
);
1621 list_del_init(&gl
->gl_reclaim
);
1622 spin_unlock(&sdp
->sd_reclaim_lock
);
1624 atomic_dec(&sdp
->sd_reclaim_count
);
1625 atomic_inc(&sdp
->sd_reclaimed
);
1627 if (gfs2_glmutex_trylock(gl
)) {
1628 if (list_empty(&gl
->gl_holders
) &&
1629 gl
->gl_state
!= LM_ST_UNLOCKED
&& demote_ok(gl
))
1630 handle_callback(gl
, LM_ST_UNLOCKED
, 0, 0);
1631 gfs2_glmutex_unlock(gl
);
1638 * examine_bucket - Call a function for glock in a hash bucket
1639 * @examiner: the function
1640 * @sdp: the filesystem
1641 * @bucket: the bucket
1643 * Returns: 1 if the bucket has entries
1646 static int examine_bucket(glock_examiner examiner
, struct gfs2_sbd
*sdp
,
1649 struct gfs2_glock
*gl
, *prev
= NULL
;
1650 int has_entries
= 0;
1651 struct hlist_head
*head
= &gl_hash_table
[hash
].hb_list
;
1653 read_lock(gl_lock_addr(hash
));
1654 /* Can't use hlist_for_each_entry - don't want prefetch here */
1655 if (hlist_empty(head
))
1657 gl
= list_entry(head
->first
, struct gfs2_glock
, gl_list
);
1659 if (!sdp
|| gl
->gl_sbd
== sdp
) {
1660 gfs2_glock_hold(gl
);
1661 read_unlock(gl_lock_addr(hash
));
1663 gfs2_glock_put(prev
);
1667 read_lock(gl_lock_addr(hash
));
1669 if (gl
->gl_list
.next
== NULL
)
1671 gl
= list_entry(gl
->gl_list
.next
, struct gfs2_glock
, gl_list
);
1674 read_unlock(gl_lock_addr(hash
));
1676 gfs2_glock_put(prev
);
1682 * scan_glock - look at a glock and see if we can reclaim it
1683 * @gl: the glock to look at
1687 static void scan_glock(struct gfs2_glock
*gl
)
1689 if (gl
->gl_ops
== &gfs2_inode_glops
&& gl
->gl_object
)
1692 if (gfs2_glmutex_trylock(gl
)) {
1693 if (list_empty(&gl
->gl_holders
) &&
1694 gl
->gl_state
!= LM_ST_UNLOCKED
&& demote_ok(gl
))
1696 gfs2_glmutex_unlock(gl
);
1701 gfs2_glmutex_unlock(gl
);
1702 gfs2_glock_schedule_for_reclaim(gl
);
1706 * clear_glock - look at a glock and see if we can free it from glock cache
1707 * @gl: the glock to look at
1711 static void clear_glock(struct gfs2_glock
*gl
)
1713 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
1716 spin_lock(&sdp
->sd_reclaim_lock
);
1717 if (!list_empty(&gl
->gl_reclaim
)) {
1718 list_del_init(&gl
->gl_reclaim
);
1719 atomic_dec(&sdp
->sd_reclaim_count
);
1720 spin_unlock(&sdp
->sd_reclaim_lock
);
1721 released
= gfs2_glock_put(gl
);
1722 gfs2_assert(sdp
, !released
);
1724 spin_unlock(&sdp
->sd_reclaim_lock
);
1727 if (gfs2_glmutex_trylock(gl
)) {
1728 if (list_empty(&gl
->gl_holders
) &&
1729 gl
->gl_state
!= LM_ST_UNLOCKED
)
1730 handle_callback(gl
, LM_ST_UNLOCKED
, 0, 0);
1731 gfs2_glmutex_unlock(gl
);
1736 * gfs2_gl_hash_clear - Empty out the glock hash table
1737 * @sdp: the filesystem
1738 * @wait: wait until it's all gone
1740 * Called when unmounting the filesystem, or when inter-node lock manager
1741 * requests DROPLOCKS because it is running out of capacity.
1744 void gfs2_gl_hash_clear(struct gfs2_sbd
*sdp
, int wait
)
1754 for (x
= 0; x
< GFS2_GL_HASH_SIZE
; x
++) {
1755 if (examine_bucket(clear_glock
, sdp
, x
))
1762 if (time_after_eq(jiffies
,
1763 t
+ gfs2_tune_get(sdp
, gt_stall_secs
) * HZ
)) {
1764 fs_warn(sdp
, "Unmount seems to be stalled. "
1765 "Dumping lock state...\n");
1766 gfs2_dump_lockstate(sdp
);
1770 down_write(&gfs2_umount_flush_sem
);
1771 invalidate_inodes(sdp
->sd_vfs
);
1772 up_write(&gfs2_umount_flush_sem
);
1778 * Diagnostic routines to help debug distributed deadlock
1781 static void gfs2_print_symbol(struct glock_iter
*gi
, const char *fmt
,
1782 unsigned long address
)
1784 char buffer
[KSYM_SYMBOL_LEN
];
1786 sprint_symbol(buffer
, address
);
1787 print_dbg(gi
, fmt
, buffer
);
1791 * dump_holder - print information about a glock holder
1792 * @str: a string naming the type of holder
1793 * @gh: the glock holder
1795 * Returns: 0 on success, -ENOBUFS when we run out of space
1798 static int dump_holder(struct glock_iter
*gi
, char *str
,
1799 struct gfs2_holder
*gh
)
1802 struct task_struct
*gh_owner
;
1804 print_dbg(gi
, " %s\n", str
);
1805 if (gh
->gh_owner_pid
) {
1806 print_dbg(gi
, " owner = %ld ", (long)gh
->gh_owner_pid
);
1807 gh_owner
= find_task_by_pid(gh
->gh_owner_pid
);
1809 print_dbg(gi
, "(%s)\n", gh_owner
->comm
);
1811 print_dbg(gi
, "(ended)\n");
1813 print_dbg(gi
, " owner = -1\n");
1814 print_dbg(gi
, " gh_state = %u\n", gh
->gh_state
);
1815 print_dbg(gi
, " gh_flags =");
1816 for (x
= 0; x
< 32; x
++)
1817 if (gh
->gh_flags
& (1 << x
))
1818 print_dbg(gi
, " %u", x
);
1819 print_dbg(gi
, " \n");
1820 print_dbg(gi
, " error = %d\n", gh
->gh_error
);
1821 print_dbg(gi
, " gh_iflags =");
1822 for (x
= 0; x
< 32; x
++)
1823 if (test_bit(x
, &gh
->gh_iflags
))
1824 print_dbg(gi
, " %u", x
);
1825 print_dbg(gi
, " \n");
1826 gfs2_print_symbol(gi
, " initialized at: %s\n", gh
->gh_ip
);
1832 * dump_inode - print information about an inode
1835 * Returns: 0 on success, -ENOBUFS when we run out of space
1838 static int dump_inode(struct glock_iter
*gi
, struct gfs2_inode
*ip
)
1842 print_dbg(gi
, " Inode:\n");
1843 print_dbg(gi
, " num = %llu/%llu\n",
1844 (unsigned long long)ip
->i_no_formal_ino
,
1845 (unsigned long long)ip
->i_no_addr
);
1846 print_dbg(gi
, " type = %u\n", IF2DT(ip
->i_inode
.i_mode
));
1847 print_dbg(gi
, " i_flags =");
1848 for (x
= 0; x
< 32; x
++)
1849 if (test_bit(x
, &ip
->i_flags
))
1850 print_dbg(gi
, " %u", x
);
1851 print_dbg(gi
, " \n");
1856 * dump_glock - print information about a glock
1858 * @count: where we are in the buffer
1860 * Returns: 0 on success, -ENOBUFS when we run out of space
1863 static int dump_glock(struct glock_iter
*gi
, struct gfs2_glock
*gl
)
1865 struct gfs2_holder
*gh
;
1867 int error
= -ENOBUFS
;
1868 struct task_struct
*gl_owner
;
1870 spin_lock(&gl
->gl_spin
);
1872 print_dbg(gi
, "Glock 0x%p (%u, 0x%llx)\n", gl
, gl
->gl_name
.ln_type
,
1873 (unsigned long long)gl
->gl_name
.ln_number
);
1874 print_dbg(gi
, " gl_flags =");
1875 for (x
= 0; x
< 32; x
++) {
1876 if (test_bit(x
, &gl
->gl_flags
))
1877 print_dbg(gi
, " %u", x
);
1879 if (!test_bit(GLF_LOCK
, &gl
->gl_flags
))
1880 print_dbg(gi
, " (unlocked)");
1881 print_dbg(gi
, " \n");
1882 print_dbg(gi
, " gl_ref = %d\n", atomic_read(&gl
->gl_ref
));
1883 print_dbg(gi
, " gl_state = %u\n", gl
->gl_state
);
1884 if (gl
->gl_owner_pid
) {
1885 gl_owner
= find_task_by_pid(gl
->gl_owner_pid
);
1887 print_dbg(gi
, " gl_owner = pid %d (%s)\n",
1888 gl
->gl_owner_pid
, gl_owner
->comm
);
1890 print_dbg(gi
, " gl_owner = %d (ended)\n",
1893 print_dbg(gi
, " gl_owner = -1\n");
1894 print_dbg(gi
, " gl_ip = %lu\n", gl
->gl_ip
);
1895 print_dbg(gi
, " req_gh = %s\n", (gl
->gl_req_gh
) ? "yes" : "no");
1896 print_dbg(gi
, " req_bh = %s\n", (gl
->gl_req_bh
) ? "yes" : "no");
1897 print_dbg(gi
, " lvb_count = %d\n", atomic_read(&gl
->gl_lvb_count
));
1898 print_dbg(gi
, " object = %s\n", (gl
->gl_object
) ? "yes" : "no");
1899 print_dbg(gi
, " reclaim = %s\n",
1900 (list_empty(&gl
->gl_reclaim
)) ? "no" : "yes");
1902 print_dbg(gi
, " aspace = 0x%p nrpages = %lu\n", gl
->gl_aspace
,
1903 gl
->gl_aspace
->i_mapping
->nrpages
);
1905 print_dbg(gi
, " aspace = no\n");
1906 print_dbg(gi
, " ail = %d\n", atomic_read(&gl
->gl_ail_count
));
1907 if (gl
->gl_req_gh
) {
1908 error
= dump_holder(gi
, "Request", gl
->gl_req_gh
);
1912 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
) {
1913 error
= dump_holder(gi
, "Holder", gh
);
1917 list_for_each_entry(gh
, &gl
->gl_waiters1
, gh_list
) {
1918 error
= dump_holder(gi
, "Waiter1", gh
);
1922 list_for_each_entry(gh
, &gl
->gl_waiters3
, gh_list
) {
1923 error
= dump_holder(gi
, "Waiter3", gh
);
1927 if (test_bit(GLF_DEMOTE
, &gl
->gl_flags
)) {
1928 print_dbg(gi
, " Demotion req to state %u (%llu uS ago)\n",
1929 gl
->gl_demote_state
, (unsigned long long)
1930 (jiffies
- gl
->gl_demote_time
)*(1000000/HZ
));
1932 if (gl
->gl_ops
== &gfs2_inode_glops
&& gl
->gl_object
) {
1933 if (!test_bit(GLF_LOCK
, &gl
->gl_flags
) &&
1934 list_empty(&gl
->gl_holders
)) {
1935 error
= dump_inode(gi
, gl
->gl_object
);
1940 print_dbg(gi
, " Inode: busy\n");
1947 spin_unlock(&gl
->gl_spin
);
1952 * gfs2_dump_lockstate - print out the current lockstate
1953 * @sdp: the filesystem
1954 * @ub: the buffer to copy the information into
1956 * If @ub is NULL, dump the lockstate to the console.
1960 static int gfs2_dump_lockstate(struct gfs2_sbd
*sdp
)
1962 struct gfs2_glock
*gl
;
1963 struct hlist_node
*h
;
1967 for (x
= 0; x
< GFS2_GL_HASH_SIZE
; x
++) {
1969 read_lock(gl_lock_addr(x
));
1971 hlist_for_each_entry(gl
, h
, &gl_hash_table
[x
].hb_list
, gl_list
) {
1972 if (gl
->gl_sbd
!= sdp
)
1975 error
= dump_glock(NULL
, gl
);
1980 read_unlock(gl_lock_addr(x
));
1991 * gfs2_scand - Look for cached glocks and inodes to toss from memory
1992 * @sdp: Pointer to GFS2 superblock
1994 * One of these daemons runs, finding candidates to add to sd_reclaim_list.
1998 static int gfs2_scand(void *data
)
2003 while (!kthread_should_stop()) {
2004 for (x
= 0; x
< GFS2_GL_HASH_SIZE
; x
++)
2005 examine_bucket(scan_glock
, NULL
, x
);
2006 if (freezing(current
))
2011 schedule_timeout_interruptible(delay
* HZ
);
2019 int __init
gfs2_glock_init(void)
2022 for(i
= 0; i
< GFS2_GL_HASH_SIZE
; i
++) {
2023 INIT_HLIST_HEAD(&gl_hash_table
[i
].hb_list
);
2025 #ifdef GL_HASH_LOCK_SZ
2026 for(i
= 0; i
< GL_HASH_LOCK_SZ
; i
++) {
2027 rwlock_init(&gl_hash_locks
[i
]);
2031 scand_process
= kthread_run(gfs2_scand
, NULL
, "gfs2_scand");
2032 if (IS_ERR(scand_process
))
2033 return PTR_ERR(scand_process
);
2035 glock_workqueue
= create_workqueue("glock_workqueue");
2036 if (IS_ERR(glock_workqueue
)) {
2037 kthread_stop(scand_process
);
2038 return PTR_ERR(glock_workqueue
);
2044 void gfs2_glock_exit(void)
2046 destroy_workqueue(glock_workqueue
);
2047 kthread_stop(scand_process
);
2050 module_param(scand_secs
, uint
, S_IRUGO
|S_IWUSR
);
2051 MODULE_PARM_DESC(scand_secs
, "The number of seconds between scand runs");
2053 static int gfs2_glock_iter_next(struct glock_iter
*gi
)
2055 struct gfs2_glock
*gl
;
2058 read_lock(gl_lock_addr(gi
->hash
));
2061 gi
->gl
= hlist_entry(gl
->gl_list
.next
,
2062 struct gfs2_glock
, gl_list
);
2064 gfs2_glock_hold(gi
->gl
);
2066 read_unlock(gl_lock_addr(gi
->hash
));
2069 if (gl
&& gi
->gl
== NULL
)
2071 while(gi
->gl
== NULL
) {
2072 if (gi
->hash
>= GFS2_GL_HASH_SIZE
)
2074 read_lock(gl_lock_addr(gi
->hash
));
2075 gi
->gl
= hlist_entry(gl_hash_table
[gi
->hash
].hb_list
.first
,
2076 struct gfs2_glock
, gl_list
);
2078 gfs2_glock_hold(gi
->gl
);
2079 read_unlock(gl_lock_addr(gi
->hash
));
2083 if (gi
->sdp
!= gi
->gl
->gl_sbd
)
2089 static void gfs2_glock_iter_free(struct glock_iter
*gi
)
2092 gfs2_glock_put(gi
->gl
);
2096 static struct glock_iter
*gfs2_glock_iter_init(struct gfs2_sbd
*sdp
)
2098 struct glock_iter
*gi
;
2100 gi
= kmalloc(sizeof (*gi
), GFP_KERNEL
);
2108 memset(gi
->string
, 0, sizeof(gi
->string
));
2110 if (gfs2_glock_iter_next(gi
)) {
2111 gfs2_glock_iter_free(gi
);
2118 static void *gfs2_glock_seq_start(struct seq_file
*file
, loff_t
*pos
)
2120 struct glock_iter
*gi
;
2123 gi
= gfs2_glock_iter_init(file
->private);
2128 if (gfs2_glock_iter_next(gi
)) {
2129 gfs2_glock_iter_free(gi
);
2137 static void *gfs2_glock_seq_next(struct seq_file
*file
, void *iter_ptr
,
2140 struct glock_iter
*gi
= iter_ptr
;
2144 if (gfs2_glock_iter_next(gi
)) {
2145 gfs2_glock_iter_free(gi
);
2152 static void gfs2_glock_seq_stop(struct seq_file
*file
, void *iter_ptr
)
2154 struct glock_iter
*gi
= iter_ptr
;
2156 gfs2_glock_iter_free(gi
);
2159 static int gfs2_glock_seq_show(struct seq_file
*file
, void *iter_ptr
)
2161 struct glock_iter
*gi
= iter_ptr
;
2164 dump_glock(gi
, gi
->gl
);
2169 static const struct seq_operations gfs2_glock_seq_ops
= {
2170 .start
= gfs2_glock_seq_start
,
2171 .next
= gfs2_glock_seq_next
,
2172 .stop
= gfs2_glock_seq_stop
,
2173 .show
= gfs2_glock_seq_show
,
2176 static int gfs2_debugfs_open(struct inode
*inode
, struct file
*file
)
2178 struct seq_file
*seq
;
2181 ret
= seq_open(file
, &gfs2_glock_seq_ops
);
2185 seq
= file
->private_data
;
2186 seq
->private = inode
->i_private
;
2191 static const struct file_operations gfs2_debug_fops
= {
2192 .owner
= THIS_MODULE
,
2193 .open
= gfs2_debugfs_open
,
2195 .llseek
= seq_lseek
,
2196 .release
= seq_release
2199 int gfs2_create_debugfs_file(struct gfs2_sbd
*sdp
)
2201 sdp
->debugfs_dir
= debugfs_create_dir(sdp
->sd_table_name
, gfs2_root
);
2202 if (!sdp
->debugfs_dir
)
2204 sdp
->debugfs_dentry_glocks
= debugfs_create_file("glocks",
2206 sdp
->debugfs_dir
, sdp
,
2208 if (!sdp
->debugfs_dentry_glocks
)
2214 void gfs2_delete_debugfs_file(struct gfs2_sbd
*sdp
)
2216 if (sdp
&& sdp
->debugfs_dir
) {
2217 if (sdp
->debugfs_dentry_glocks
) {
2218 debugfs_remove(sdp
->debugfs_dentry_glocks
);
2219 sdp
->debugfs_dentry_glocks
= NULL
;
2221 debugfs_remove(sdp
->debugfs_dir
);
2222 sdp
->debugfs_dir
= NULL
;
2226 int gfs2_register_debugfs(void)
2228 gfs2_root
= debugfs_create_dir("gfs2", NULL
);
2229 return gfs2_root
? 0 : -ENOMEM
;
2232 void gfs2_unregister_debugfs(void)
2234 debugfs_remove(gfs2_root
);