2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2008 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/buffer_head.h>
14 #include <linux/delay.h>
15 #include <linux/sort.h>
16 #include <linux/jhash.h>
17 #include <linux/kallsyms.h>
18 #include <linux/gfs2_ondisk.h>
19 #include <linux/list.h>
20 #include <linux/wait.h>
21 #include <linux/module.h>
22 #include <asm/uaccess.h>
23 #include <linux/seq_file.h>
24 #include <linux/debugfs.h>
25 #include <linux/kthread.h>
26 #include <linux/freezer.h>
27 #include <linux/workqueue.h>
28 #include <linux/jiffies.h>
41 #define CREATE_TRACE_POINTS
42 #include "trace_gfs2.h"
44 struct gfs2_gl_hash_bucket
{
45 struct hlist_head hb_list
;
48 struct gfs2_glock_iter
{
49 int hash
; /* hash bucket index */
50 struct gfs2_sbd
*sdp
; /* incore superblock */
51 struct gfs2_glock
*gl
; /* current glock struct */
52 char string
[512]; /* scratch space */
55 typedef void (*glock_examiner
) (struct gfs2_glock
* gl
);
57 static int gfs2_dump_lockstate(struct gfs2_sbd
*sdp
);
58 static int __dump_glock(struct seq_file
*seq
, const struct gfs2_glock
*gl
);
59 #define GLOCK_BUG_ON(gl,x) do { if (unlikely(x)) { __dump_glock(NULL, gl); BUG(); } } while(0)
60 static void do_xmote(struct gfs2_glock
*gl
, struct gfs2_holder
*gh
, unsigned int target
);
62 static struct dentry
*gfs2_root
;
63 static struct workqueue_struct
*glock_workqueue
;
64 struct workqueue_struct
*gfs2_delete_workqueue
;
65 static LIST_HEAD(lru_list
);
66 static atomic_t lru_count
= ATOMIC_INIT(0);
67 static DEFINE_SPINLOCK(lru_lock
);
69 #define GFS2_GL_HASH_SHIFT 15
70 #define GFS2_GL_HASH_SIZE (1 << GFS2_GL_HASH_SHIFT)
71 #define GFS2_GL_HASH_MASK (GFS2_GL_HASH_SIZE - 1)
73 static struct gfs2_gl_hash_bucket gl_hash_table
[GFS2_GL_HASH_SIZE
];
74 static struct dentry
*gfs2_root
;
77 * Despite what you might think, the numbers below are not arbitrary :-)
78 * They are taken from the ipv4 routing hash code, which is well tested
79 * and thus should be nearly optimal. Later on we might tweek the numbers
80 * but for now this should be fine.
82 * The reason for putting the locks in a separate array from the list heads
83 * is that we can have fewer locks than list heads and save memory. We use
84 * the same hash function for both, but with a different hash mask.
86 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
87 defined(CONFIG_PROVE_LOCKING)
90 # define GL_HASH_LOCK_SZ 256
93 # define GL_HASH_LOCK_SZ 4096
95 # define GL_HASH_LOCK_SZ 2048
97 # define GL_HASH_LOCK_SZ 1024
99 # define GL_HASH_LOCK_SZ 512
101 # define GL_HASH_LOCK_SZ 256
105 /* We never want more locks than chains */
106 #if GFS2_GL_HASH_SIZE < GL_HASH_LOCK_SZ
107 # undef GL_HASH_LOCK_SZ
108 # define GL_HASH_LOCK_SZ GFS2_GL_HASH_SIZE
111 static rwlock_t gl_hash_locks
[GL_HASH_LOCK_SZ
];
113 static inline rwlock_t
*gl_lock_addr(unsigned int x
)
115 return &gl_hash_locks
[x
& (GL_HASH_LOCK_SZ
-1)];
117 #else /* not SMP, so no spinlocks required */
118 static inline rwlock_t
*gl_lock_addr(unsigned int x
)
125 * gl_hash() - Turn glock number into hash bucket number
126 * @lock: The glock number
128 * Returns: The number of the corresponding hash bucket
131 static unsigned int gl_hash(const struct gfs2_sbd
*sdp
,
132 const struct lm_lockname
*name
)
136 h
= jhash(&name
->ln_number
, sizeof(u64
), 0);
137 h
= jhash(&name
->ln_type
, sizeof(unsigned int), h
);
138 h
= jhash(&sdp
, sizeof(struct gfs2_sbd
*), h
);
139 h
&= GFS2_GL_HASH_MASK
;
145 * glock_free() - Perform a few checks and then release struct gfs2_glock
146 * @gl: The glock to release
148 * Also calls lock module to release its internal structure for this glock.
152 static void glock_free(struct gfs2_glock
*gl
)
154 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
155 struct address_space
*mapping
= gfs2_glock2aspace(gl
);
156 struct kmem_cache
*cachep
= gfs2_glock_cachep
;
158 GLOCK_BUG_ON(gl
, mapping
&& mapping
->nrpages
);
159 trace_gfs2_glock_put(gl
);
161 cachep
= gfs2_glock_aspace_cachep
;
162 sdp
->sd_lockstruct
.ls_ops
->lm_put_lock(cachep
, gl
);
166 * gfs2_glock_hold() - increment reference count on glock
167 * @gl: The glock to hold
171 void gfs2_glock_hold(struct gfs2_glock
*gl
)
173 GLOCK_BUG_ON(gl
, atomic_read(&gl
->gl_ref
) == 0);
174 atomic_inc(&gl
->gl_ref
);
178 * demote_ok - Check to see if it's ok to unlock a glock
181 * Returns: 1 if it's ok
184 static int demote_ok(const struct gfs2_glock
*gl
)
186 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
188 if (gl
->gl_state
== LM_ST_UNLOCKED
)
190 if (!list_empty(&gl
->gl_holders
))
192 if (glops
->go_demote_ok
)
193 return glops
->go_demote_ok(gl
);
198 * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
203 static void gfs2_glock_schedule_for_reclaim(struct gfs2_glock
*gl
)
206 may_reclaim
= (demote_ok(gl
) &&
207 (atomic_read(&gl
->gl_ref
) == 1 ||
208 (gl
->gl_name
.ln_type
== LM_TYPE_INODE
&&
209 atomic_read(&gl
->gl_ref
) <= 2)));
210 spin_lock(&lru_lock
);
211 if (list_empty(&gl
->gl_lru
) && may_reclaim
) {
212 list_add_tail(&gl
->gl_lru
, &lru_list
);
213 atomic_inc(&lru_count
);
215 spin_unlock(&lru_lock
);
219 * gfs2_glock_put_nolock() - Decrement reference count on glock
220 * @gl: The glock to put
222 * This function should only be used if the caller has its own reference
223 * to the glock, in addition to the one it is dropping.
226 void gfs2_glock_put_nolock(struct gfs2_glock
*gl
)
228 if (atomic_dec_and_test(&gl
->gl_ref
))
230 gfs2_glock_schedule_for_reclaim(gl
);
234 * gfs2_glock_put() - Decrement reference count on glock
235 * @gl: The glock to put
239 int gfs2_glock_put(struct gfs2_glock
*gl
)
243 write_lock(gl_lock_addr(gl
->gl_hash
));
244 if (atomic_dec_and_lock(&gl
->gl_ref
, &lru_lock
)) {
245 hlist_del(&gl
->gl_list
);
246 if (!list_empty(&gl
->gl_lru
)) {
247 list_del_init(&gl
->gl_lru
);
248 atomic_dec(&lru_count
);
250 spin_unlock(&lru_lock
);
251 write_unlock(gl_lock_addr(gl
->gl_hash
));
252 GLOCK_BUG_ON(gl
, !list_empty(&gl
->gl_holders
));
257 spin_lock(&gl
->gl_spin
);
258 gfs2_glock_schedule_for_reclaim(gl
);
259 spin_unlock(&gl
->gl_spin
);
260 write_unlock(gl_lock_addr(gl
->gl_hash
));
266 * search_bucket() - Find struct gfs2_glock by lock number
267 * @bucket: the bucket to search
268 * @name: The lock name
270 * Returns: NULL, or the struct gfs2_glock with the requested number
273 static struct gfs2_glock
*search_bucket(unsigned int hash
,
274 const struct gfs2_sbd
*sdp
,
275 const struct lm_lockname
*name
)
277 struct gfs2_glock
*gl
;
278 struct hlist_node
*h
;
280 hlist_for_each_entry(gl
, h
, &gl_hash_table
[hash
].hb_list
, gl_list
) {
281 if (!lm_name_equal(&gl
->gl_name
, name
))
283 if (gl
->gl_sbd
!= sdp
)
286 atomic_inc(&gl
->gl_ref
);
295 * may_grant - check if its ok to grant a new lock
297 * @gh: The lock request which we wish to grant
299 * Returns: true if its ok to grant the lock
302 static inline int may_grant(const struct gfs2_glock
*gl
, const struct gfs2_holder
*gh
)
304 const struct gfs2_holder
*gh_head
= list_entry(gl
->gl_holders
.next
, const struct gfs2_holder
, gh_list
);
305 if ((gh
->gh_state
== LM_ST_EXCLUSIVE
||
306 gh_head
->gh_state
== LM_ST_EXCLUSIVE
) && gh
!= gh_head
)
308 if (gl
->gl_state
== gh
->gh_state
)
310 if (gh
->gh_flags
& GL_EXACT
)
312 if (gl
->gl_state
== LM_ST_EXCLUSIVE
) {
313 if (gh
->gh_state
== LM_ST_SHARED
&& gh_head
->gh_state
== LM_ST_SHARED
)
315 if (gh
->gh_state
== LM_ST_DEFERRED
&& gh_head
->gh_state
== LM_ST_DEFERRED
)
318 if (gl
->gl_state
!= LM_ST_UNLOCKED
&& (gh
->gh_flags
& LM_FLAG_ANY
))
323 static void gfs2_holder_wake(struct gfs2_holder
*gh
)
325 clear_bit(HIF_WAIT
, &gh
->gh_iflags
);
326 smp_mb__after_clear_bit();
327 wake_up_bit(&gh
->gh_iflags
, HIF_WAIT
);
331 * do_error - Something unexpected has happened during a lock request
335 static inline void do_error(struct gfs2_glock
*gl
, const int ret
)
337 struct gfs2_holder
*gh
, *tmp
;
339 list_for_each_entry_safe(gh
, tmp
, &gl
->gl_holders
, gh_list
) {
340 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
342 if (ret
& LM_OUT_ERROR
)
344 else if (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
))
345 gh
->gh_error
= GLR_TRYFAILED
;
348 list_del_init(&gh
->gh_list
);
349 trace_gfs2_glock_queue(gh
, 0);
350 gfs2_holder_wake(gh
);
355 * do_promote - promote as many requests as possible on the current queue
358 * Returns: 1 if there is a blocked holder at the head of the list, or 2
359 * if a type specific operation is underway.
362 static int do_promote(struct gfs2_glock
*gl
)
363 __releases(&gl
->gl_spin
)
364 __acquires(&gl
->gl_spin
)
366 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
367 struct gfs2_holder
*gh
, *tmp
;
371 list_for_each_entry_safe(gh
, tmp
, &gl
->gl_holders
, gh_list
) {
372 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
374 if (may_grant(gl
, gh
)) {
375 if (gh
->gh_list
.prev
== &gl
->gl_holders
&&
377 spin_unlock(&gl
->gl_spin
);
378 /* FIXME: eliminate this eventually */
379 ret
= glops
->go_lock(gh
);
380 spin_lock(&gl
->gl_spin
);
385 list_del_init(&gh
->gh_list
);
386 trace_gfs2_glock_queue(gh
, 0);
387 gfs2_holder_wake(gh
);
390 set_bit(HIF_HOLDER
, &gh
->gh_iflags
);
391 trace_gfs2_promote(gh
, 1);
392 gfs2_holder_wake(gh
);
395 set_bit(HIF_HOLDER
, &gh
->gh_iflags
);
396 trace_gfs2_promote(gh
, 0);
397 gfs2_holder_wake(gh
);
400 if (gh
->gh_list
.prev
== &gl
->gl_holders
)
409 * find_first_waiter - find the first gh that's waiting for the glock
413 static inline struct gfs2_holder
*find_first_waiter(const struct gfs2_glock
*gl
)
415 struct gfs2_holder
*gh
;
417 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
) {
418 if (!test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
425 * state_change - record that the glock is now in a different state
427 * @new_state the new state
431 static void state_change(struct gfs2_glock
*gl
, unsigned int new_state
)
435 held1
= (gl
->gl_state
!= LM_ST_UNLOCKED
);
436 held2
= (new_state
!= LM_ST_UNLOCKED
);
438 if (held1
!= held2
) {
442 gfs2_glock_put_nolock(gl
);
444 if (held1
&& held2
&& list_empty(&gl
->gl_holders
))
445 clear_bit(GLF_QUEUED
, &gl
->gl_flags
);
447 gl
->gl_state
= new_state
;
448 gl
->gl_tchange
= jiffies
;
451 static void gfs2_demote_wake(struct gfs2_glock
*gl
)
453 gl
->gl_demote_state
= LM_ST_EXCLUSIVE
;
454 clear_bit(GLF_DEMOTE
, &gl
->gl_flags
);
455 smp_mb__after_clear_bit();
456 wake_up_bit(&gl
->gl_flags
, GLF_DEMOTE
);
460 * finish_xmote - The DLM has replied to one of our lock requests
462 * @ret: The status from the DLM
466 static void finish_xmote(struct gfs2_glock
*gl
, unsigned int ret
)
468 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
469 struct gfs2_holder
*gh
;
470 unsigned state
= ret
& LM_OUT_ST_MASK
;
473 spin_lock(&gl
->gl_spin
);
474 trace_gfs2_glock_state_change(gl
, state
);
475 state_change(gl
, state
);
476 gh
= find_first_waiter(gl
);
478 /* Demote to UN request arrived during demote to SH or DF */
479 if (test_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
) &&
480 state
!= LM_ST_UNLOCKED
&& gl
->gl_demote_state
== LM_ST_UNLOCKED
)
481 gl
->gl_target
= LM_ST_UNLOCKED
;
483 /* Check for state != intended state */
484 if (unlikely(state
!= gl
->gl_target
)) {
485 if (gh
&& !test_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
)) {
486 /* move to back of queue and try next entry */
487 if (ret
& LM_OUT_CANCELED
) {
488 if ((gh
->gh_flags
& LM_FLAG_PRIORITY
) == 0)
489 list_move_tail(&gh
->gh_list
, &gl
->gl_holders
);
490 gh
= find_first_waiter(gl
);
491 gl
->gl_target
= gh
->gh_state
;
494 /* Some error or failed "try lock" - report it */
495 if ((ret
& LM_OUT_ERROR
) ||
496 (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
))) {
497 gl
->gl_target
= gl
->gl_state
;
503 /* Unlocked due to conversion deadlock, try again */
506 do_xmote(gl
, gh
, gl
->gl_target
);
508 /* Conversion fails, unlock and try again */
511 do_xmote(gl
, gh
, LM_ST_UNLOCKED
);
513 default: /* Everything else */
514 printk(KERN_ERR
"GFS2: wanted %u got %u\n", gl
->gl_target
, state
);
517 spin_unlock(&gl
->gl_spin
);
521 /* Fast path - we got what we asked for */
522 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
))
523 gfs2_demote_wake(gl
);
524 if (state
!= LM_ST_UNLOCKED
) {
525 if (glops
->go_xmote_bh
) {
526 spin_unlock(&gl
->gl_spin
);
527 rv
= glops
->go_xmote_bh(gl
, gh
);
528 spin_lock(&gl
->gl_spin
);
539 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
541 spin_unlock(&gl
->gl_spin
);
544 static unsigned int gfs2_lm_lock(struct gfs2_sbd
*sdp
, void *lock
,
545 unsigned int req_state
,
548 int ret
= LM_OUT_ERROR
;
550 if (!sdp
->sd_lockstruct
.ls_ops
->lm_lock
)
551 return req_state
== LM_ST_UNLOCKED
? 0 : req_state
;
553 if (likely(!test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)))
554 ret
= sdp
->sd_lockstruct
.ls_ops
->lm_lock(lock
,
560 * do_xmote - Calls the DLM to change the state of a lock
561 * @gl: The lock state
562 * @gh: The holder (only for promotes)
563 * @target: The target lock state
567 static void do_xmote(struct gfs2_glock
*gl
, struct gfs2_holder
*gh
, unsigned int target
)
568 __releases(&gl
->gl_spin
)
569 __acquires(&gl
->gl_spin
)
571 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
572 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
573 unsigned int lck_flags
= gh
? gh
->gh_flags
: 0;
576 lck_flags
&= (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
| LM_FLAG_NOEXP
|
578 BUG_ON(gl
->gl_state
== target
);
579 BUG_ON(gl
->gl_state
== gl
->gl_target
);
580 if ((target
== LM_ST_UNLOCKED
|| target
== LM_ST_DEFERRED
) &&
582 set_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
);
583 do_error(gl
, 0); /* Fail queued try locks */
585 spin_unlock(&gl
->gl_spin
);
586 if (glops
->go_xmote_th
)
587 glops
->go_xmote_th(gl
);
588 if (test_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
))
589 glops
->go_inval(gl
, target
== LM_ST_DEFERRED
? 0 : DIO_METADATA
);
590 clear_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
);
593 if (target
!= LM_ST_UNLOCKED
&& (gl
->gl_state
== LM_ST_SHARED
||
594 gl
->gl_state
== LM_ST_DEFERRED
) &&
595 !(lck_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
)))
596 lck_flags
|= LM_FLAG_TRY_1CB
;
597 ret
= gfs2_lm_lock(sdp
, gl
, target
, lck_flags
);
599 if (!(ret
& LM_OUT_ASYNC
)) {
600 finish_xmote(gl
, ret
);
601 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
604 GLOCK_BUG_ON(gl
, ret
!= LM_OUT_ASYNC
);
606 spin_lock(&gl
->gl_spin
);
610 * find_first_holder - find the first "holder" gh
614 static inline struct gfs2_holder
*find_first_holder(const struct gfs2_glock
*gl
)
616 struct gfs2_holder
*gh
;
618 if (!list_empty(&gl
->gl_holders
)) {
619 gh
= list_entry(gl
->gl_holders
.next
, struct gfs2_holder
, gh_list
);
620 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
627 * run_queue - do all outstanding tasks related to a glock
628 * @gl: The glock in question
629 * @nonblock: True if we must not block in run_queue
633 static void run_queue(struct gfs2_glock
*gl
, const int nonblock
)
634 __releases(&gl
->gl_spin
)
635 __acquires(&gl
->gl_spin
)
637 struct gfs2_holder
*gh
= NULL
;
640 if (test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
))
643 GLOCK_BUG_ON(gl
, test_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
));
645 if (test_bit(GLF_DEMOTE
, &gl
->gl_flags
) &&
646 gl
->gl_demote_state
!= gl
->gl_state
) {
647 if (find_first_holder(gl
))
651 set_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
);
652 GLOCK_BUG_ON(gl
, gl
->gl_demote_state
== LM_ST_EXCLUSIVE
);
653 gl
->gl_target
= gl
->gl_demote_state
;
655 if (test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
656 gfs2_demote_wake(gl
);
657 ret
= do_promote(gl
);
662 gh
= find_first_waiter(gl
);
663 gl
->gl_target
= gh
->gh_state
;
664 if (!(gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
)))
665 do_error(gl
, 0); /* Fail queued try locks */
667 do_xmote(gl
, gh
, gl
->gl_target
);
672 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
673 smp_mb__after_clear_bit();
675 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
676 gfs2_glock_put_nolock(gl
);
680 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
681 smp_mb__after_clear_bit();
685 static void delete_work_func(struct work_struct
*work
)
687 struct gfs2_glock
*gl
= container_of(work
, struct gfs2_glock
, gl_delete
);
688 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
689 struct gfs2_inode
*ip
= NULL
;
693 spin_lock(&gl
->gl_spin
);
694 ip
= (struct gfs2_inode
*)gl
->gl_object
;
696 no_addr
= ip
->i_no_addr
;
697 spin_unlock(&gl
->gl_spin
);
699 inode
= gfs2_ilookup(sdp
->sd_vfs
, no_addr
);
701 d_prune_aliases(inode
);
708 static void glock_work_func(struct work_struct
*work
)
710 unsigned long delay
= 0;
711 struct gfs2_glock
*gl
= container_of(work
, struct gfs2_glock
, gl_work
.work
);
714 if (test_and_clear_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
)) {
715 finish_xmote(gl
, gl
->gl_reply
);
718 spin_lock(&gl
->gl_spin
);
719 if (test_and_clear_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
) &&
720 gl
->gl_state
!= LM_ST_UNLOCKED
&&
721 gl
->gl_demote_state
!= LM_ST_EXCLUSIVE
) {
722 unsigned long holdtime
, now
= jiffies
;
723 holdtime
= gl
->gl_tchange
+ gl
->gl_ops
->go_min_hold_time
;
724 if (time_before(now
, holdtime
))
725 delay
= holdtime
- now
;
726 set_bit(delay
? GLF_PENDING_DEMOTE
: GLF_DEMOTE
, &gl
->gl_flags
);
729 spin_unlock(&gl
->gl_spin
);
731 queue_delayed_work(glock_workqueue
, &gl
->gl_work
, delay
) == 0)
738 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
739 * @sdp: The GFS2 superblock
740 * @number: the lock number
741 * @glops: The glock_operations to use
742 * @create: If 0, don't create the glock if it doesn't exist
743 * @glp: the glock is returned here
745 * This does not lock a glock, just finds/creates structures for one.
750 int gfs2_glock_get(struct gfs2_sbd
*sdp
, u64 number
,
751 const struct gfs2_glock_operations
*glops
, int create
,
752 struct gfs2_glock
**glp
)
754 struct super_block
*s
= sdp
->sd_vfs
;
755 struct lm_lockname name
= { .ln_number
= number
, .ln_type
= glops
->go_type
};
756 struct gfs2_glock
*gl
, *tmp
;
757 unsigned int hash
= gl_hash(sdp
, &name
);
758 struct address_space
*mapping
;
760 read_lock(gl_lock_addr(hash
));
761 gl
= search_bucket(hash
, sdp
, &name
);
762 read_unlock(gl_lock_addr(hash
));
770 if (glops
->go_flags
& GLOF_ASPACE
)
771 gl
= kmem_cache_alloc(gfs2_glock_aspace_cachep
, GFP_KERNEL
);
773 gl
= kmem_cache_alloc(gfs2_glock_cachep
, GFP_KERNEL
);
777 atomic_inc(&sdp
->sd_glock_disposal
);
780 atomic_set(&gl
->gl_ref
, 1);
781 gl
->gl_state
= LM_ST_UNLOCKED
;
782 gl
->gl_target
= LM_ST_UNLOCKED
;
783 gl
->gl_demote_state
= LM_ST_EXCLUSIVE
;
786 snprintf(gl
->gl_strname
, GDLM_STRNAME_BYTES
, "%8x%16llx", name
.ln_type
, (unsigned long long)number
);
787 memset(&gl
->gl_lksb
, 0, sizeof(struct dlm_lksb
));
788 gl
->gl_lksb
.sb_lvbptr
= gl
->gl_lvb
;
789 gl
->gl_tchange
= jiffies
;
790 gl
->gl_object
= NULL
;
792 INIT_DELAYED_WORK(&gl
->gl_work
, glock_work_func
);
793 INIT_WORK(&gl
->gl_delete
, delete_work_func
);
795 mapping
= gfs2_glock2aspace(gl
);
797 mapping
->a_ops
= &gfs2_meta_aops
;
798 mapping
->host
= s
->s_bdev
->bd_inode
;
800 mapping_set_gfp_mask(mapping
, GFP_NOFS
);
801 mapping
->assoc_mapping
= NULL
;
802 mapping
->backing_dev_info
= s
->s_bdi
;
803 mapping
->writeback_index
= 0;
806 write_lock(gl_lock_addr(hash
));
807 tmp
= search_bucket(hash
, sdp
, &name
);
809 write_unlock(gl_lock_addr(hash
));
813 hlist_add_head(&gl
->gl_list
, &gl_hash_table
[hash
].hb_list
);
814 write_unlock(gl_lock_addr(hash
));
823 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
825 * @state: the state we're requesting
826 * @flags: the modifier flags
827 * @gh: the holder structure
831 void gfs2_holder_init(struct gfs2_glock
*gl
, unsigned int state
, unsigned flags
,
832 struct gfs2_holder
*gh
)
834 INIT_LIST_HEAD(&gh
->gh_list
);
836 gh
->gh_ip
= (unsigned long)__builtin_return_address(0);
837 gh
->gh_owner_pid
= get_pid(task_pid(current
));
838 gh
->gh_state
= state
;
839 gh
->gh_flags
= flags
;
846 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
847 * @state: the state we're requesting
848 * @flags: the modifier flags
849 * @gh: the holder structure
851 * Don't mess with the glock.
855 void gfs2_holder_reinit(unsigned int state
, unsigned flags
, struct gfs2_holder
*gh
)
857 gh
->gh_state
= state
;
858 gh
->gh_flags
= flags
;
860 gh
->gh_ip
= (unsigned long)__builtin_return_address(0);
861 if (gh
->gh_owner_pid
)
862 put_pid(gh
->gh_owner_pid
);
863 gh
->gh_owner_pid
= get_pid(task_pid(current
));
867 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
868 * @gh: the holder structure
872 void gfs2_holder_uninit(struct gfs2_holder
*gh
)
874 put_pid(gh
->gh_owner_pid
);
875 gfs2_glock_put(gh
->gh_gl
);
881 * gfs2_glock_holder_wait
884 * This function and gfs2_glock_demote_wait both show up in the WCHAN
885 * field. Thus I've separated these otherwise identical functions in
886 * order to be more informative to the user.
889 static int gfs2_glock_holder_wait(void *word
)
895 static int gfs2_glock_demote_wait(void *word
)
901 static void wait_on_holder(struct gfs2_holder
*gh
)
904 wait_on_bit(&gh
->gh_iflags
, HIF_WAIT
, gfs2_glock_holder_wait
, TASK_UNINTERRUPTIBLE
);
907 static void wait_on_demote(struct gfs2_glock
*gl
)
910 wait_on_bit(&gl
->gl_flags
, GLF_DEMOTE
, gfs2_glock_demote_wait
, TASK_UNINTERRUPTIBLE
);
914 * handle_callback - process a demote request
916 * @state: the state the caller wants us to change to
918 * There are only two requests that we are going to see in actual
919 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
922 static void handle_callback(struct gfs2_glock
*gl
, unsigned int state
,
925 int bit
= delay
? GLF_PENDING_DEMOTE
: GLF_DEMOTE
;
927 set_bit(bit
, &gl
->gl_flags
);
928 if (gl
->gl_demote_state
== LM_ST_EXCLUSIVE
) {
929 gl
->gl_demote_state
= state
;
930 gl
->gl_demote_time
= jiffies
;
931 } else if (gl
->gl_demote_state
!= LM_ST_UNLOCKED
&&
932 gl
->gl_demote_state
!= state
) {
933 gl
->gl_demote_state
= LM_ST_UNLOCKED
;
935 if (gl
->gl_ops
->go_callback
)
936 gl
->gl_ops
->go_callback(gl
);
937 trace_gfs2_demote_rq(gl
);
941 * gfs2_glock_wait - wait on a glock acquisition
942 * @gh: the glock holder
944 * Returns: 0 on success
947 int gfs2_glock_wait(struct gfs2_holder
*gh
)
953 void gfs2_print_dbg(struct seq_file
*seq
, const char *fmt
, ...)
959 struct gfs2_glock_iter
*gi
= seq
->private;
960 vsprintf(gi
->string
, fmt
, args
);
961 seq_printf(seq
, gi
->string
);
963 printk(KERN_ERR
" ");
970 * add_to_queue - Add a holder to the wait queue (but look for recursion)
971 * @gh: the holder structure to add
973 * Eventually we should move the recursive locking trap to a
974 * debugging option or something like that. This is the fast
975 * path and needs to have the minimum number of distractions.
979 static inline void add_to_queue(struct gfs2_holder
*gh
)
980 __releases(&gl
->gl_spin
)
981 __acquires(&gl
->gl_spin
)
983 struct gfs2_glock
*gl
= gh
->gh_gl
;
984 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
985 struct list_head
*insert_pt
= NULL
;
986 struct gfs2_holder
*gh2
;
989 BUG_ON(gh
->gh_owner_pid
== NULL
);
990 if (test_and_set_bit(HIF_WAIT
, &gh
->gh_iflags
))
993 if (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
)) {
994 if (test_bit(GLF_LOCK
, &gl
->gl_flags
))
996 if (test_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
))
1000 list_for_each_entry(gh2
, &gl
->gl_holders
, gh_list
) {
1001 if (unlikely(gh2
->gh_owner_pid
== gh
->gh_owner_pid
&&
1002 (gh
->gh_gl
->gl_ops
->go_type
!= LM_TYPE_FLOCK
)))
1003 goto trap_recursive
;
1005 !(gh2
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
)) &&
1006 !may_grant(gl
, gh
)) {
1008 gh
->gh_error
= GLR_TRYFAILED
;
1009 gfs2_holder_wake(gh
);
1012 if (test_bit(HIF_HOLDER
, &gh2
->gh_iflags
))
1014 if (unlikely((gh
->gh_flags
& LM_FLAG_PRIORITY
) && !insert_pt
))
1015 insert_pt
= &gh2
->gh_list
;
1017 set_bit(GLF_QUEUED
, &gl
->gl_flags
);
1018 if (likely(insert_pt
== NULL
)) {
1019 list_add_tail(&gh
->gh_list
, &gl
->gl_holders
);
1020 if (unlikely(gh
->gh_flags
& LM_FLAG_PRIORITY
))
1024 trace_gfs2_glock_queue(gh
, 1);
1025 list_add_tail(&gh
->gh_list
, insert_pt
);
1027 gh
= list_entry(gl
->gl_holders
.next
, struct gfs2_holder
, gh_list
);
1028 if (!(gh
->gh_flags
& LM_FLAG_PRIORITY
)) {
1029 spin_unlock(&gl
->gl_spin
);
1030 if (sdp
->sd_lockstruct
.ls_ops
->lm_cancel
)
1031 sdp
->sd_lockstruct
.ls_ops
->lm_cancel(gl
);
1032 spin_lock(&gl
->gl_spin
);
1037 print_symbol(KERN_ERR
"original: %s\n", gh2
->gh_ip
);
1038 printk(KERN_ERR
"pid: %d\n", pid_nr(gh2
->gh_owner_pid
));
1039 printk(KERN_ERR
"lock type: %d req lock state : %d\n",
1040 gh2
->gh_gl
->gl_name
.ln_type
, gh2
->gh_state
);
1041 print_symbol(KERN_ERR
"new: %s\n", gh
->gh_ip
);
1042 printk(KERN_ERR
"pid: %d\n", pid_nr(gh
->gh_owner_pid
));
1043 printk(KERN_ERR
"lock type: %d req lock state : %d\n",
1044 gh
->gh_gl
->gl_name
.ln_type
, gh
->gh_state
);
1045 __dump_glock(NULL
, gl
);
1050 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1051 * @gh: the holder structure
1053 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1055 * Returns: 0, GLR_TRYFAILED, or errno on failure
1058 int gfs2_glock_nq(struct gfs2_holder
*gh
)
1060 struct gfs2_glock
*gl
= gh
->gh_gl
;
1061 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
1064 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)))
1067 spin_lock(&gl
->gl_spin
);
1069 if ((LM_FLAG_NOEXP
& gh
->gh_flags
) &&
1070 test_and_clear_bit(GLF_FROZEN
, &gl
->gl_flags
))
1071 set_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
);
1073 spin_unlock(&gl
->gl_spin
);
1075 if (!(gh
->gh_flags
& GL_ASYNC
))
1076 error
= gfs2_glock_wait(gh
);
1082 * gfs2_glock_poll - poll to see if an async request has been completed
1085 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1088 int gfs2_glock_poll(struct gfs2_holder
*gh
)
1090 return test_bit(HIF_WAIT
, &gh
->gh_iflags
) ? 0 : 1;
1094 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1095 * @gh: the glock holder
1099 void gfs2_glock_dq(struct gfs2_holder
*gh
)
1101 struct gfs2_glock
*gl
= gh
->gh_gl
;
1102 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1106 spin_lock(&gl
->gl_spin
);
1107 if (gh
->gh_flags
& GL_NOCACHE
)
1108 handle_callback(gl
, LM_ST_UNLOCKED
, 0);
1110 list_del_init(&gh
->gh_list
);
1111 if (find_first_holder(gl
) == NULL
) {
1112 if (glops
->go_unlock
) {
1113 GLOCK_BUG_ON(gl
, test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
));
1114 spin_unlock(&gl
->gl_spin
);
1115 glops
->go_unlock(gh
);
1116 spin_lock(&gl
->gl_spin
);
1117 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
1119 if (list_empty(&gl
->gl_holders
) &&
1120 !test_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
) &&
1121 !test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
1124 trace_gfs2_glock_queue(gh
, 0);
1125 spin_unlock(&gl
->gl_spin
);
1126 if (likely(fast_path
))
1129 gfs2_glock_hold(gl
);
1130 if (test_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
) &&
1131 !test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
1132 delay
= gl
->gl_ops
->go_min_hold_time
;
1133 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, delay
) == 0)
1137 void gfs2_glock_dq_wait(struct gfs2_holder
*gh
)
1139 struct gfs2_glock
*gl
= gh
->gh_gl
;
1145 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1146 * @gh: the holder structure
1150 void gfs2_glock_dq_uninit(struct gfs2_holder
*gh
)
1153 gfs2_holder_uninit(gh
);
1157 * gfs2_glock_nq_num - acquire a glock based on lock number
1158 * @sdp: the filesystem
1159 * @number: the lock number
1160 * @glops: the glock operations for the type of glock
1161 * @state: the state to acquire the glock in
1162 * @flags: modifier flags for the aquisition
1163 * @gh: the struct gfs2_holder
1168 int gfs2_glock_nq_num(struct gfs2_sbd
*sdp
, u64 number
,
1169 const struct gfs2_glock_operations
*glops
,
1170 unsigned int state
, int flags
, struct gfs2_holder
*gh
)
1172 struct gfs2_glock
*gl
;
1175 error
= gfs2_glock_get(sdp
, number
, glops
, CREATE
, &gl
);
1177 error
= gfs2_glock_nq_init(gl
, state
, flags
, gh
);
1185 * glock_compare - Compare two struct gfs2_glock structures for sorting
1186 * @arg_a: the first structure
1187 * @arg_b: the second structure
1191 static int glock_compare(const void *arg_a
, const void *arg_b
)
1193 const struct gfs2_holder
*gh_a
= *(const struct gfs2_holder
**)arg_a
;
1194 const struct gfs2_holder
*gh_b
= *(const struct gfs2_holder
**)arg_b
;
1195 const struct lm_lockname
*a
= &gh_a
->gh_gl
->gl_name
;
1196 const struct lm_lockname
*b
= &gh_b
->gh_gl
->gl_name
;
1198 if (a
->ln_number
> b
->ln_number
)
1200 if (a
->ln_number
< b
->ln_number
)
1202 BUG_ON(gh_a
->gh_gl
->gl_ops
->go_type
== gh_b
->gh_gl
->gl_ops
->go_type
);
1207 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1208 * @num_gh: the number of structures
1209 * @ghs: an array of struct gfs2_holder structures
1211 * Returns: 0 on success (all glocks acquired),
1212 * errno on failure (no glocks acquired)
1215 static int nq_m_sync(unsigned int num_gh
, struct gfs2_holder
*ghs
,
1216 struct gfs2_holder
**p
)
1221 for (x
= 0; x
< num_gh
; x
++)
1224 sort(p
, num_gh
, sizeof(struct gfs2_holder
*), glock_compare
, NULL
);
1226 for (x
= 0; x
< num_gh
; x
++) {
1227 p
[x
]->gh_flags
&= ~(LM_FLAG_TRY
| GL_ASYNC
);
1229 error
= gfs2_glock_nq(p
[x
]);
1232 gfs2_glock_dq(p
[x
]);
1241 * gfs2_glock_nq_m - acquire multiple glocks
1242 * @num_gh: the number of structures
1243 * @ghs: an array of struct gfs2_holder structures
1246 * Returns: 0 on success (all glocks acquired),
1247 * errno on failure (no glocks acquired)
1250 int gfs2_glock_nq_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1252 struct gfs2_holder
*tmp
[4];
1253 struct gfs2_holder
**pph
= tmp
;
1260 ghs
->gh_flags
&= ~(LM_FLAG_TRY
| GL_ASYNC
);
1261 return gfs2_glock_nq(ghs
);
1265 pph
= kmalloc(num_gh
* sizeof(struct gfs2_holder
*), GFP_NOFS
);
1270 error
= nq_m_sync(num_gh
, ghs
, pph
);
1279 * gfs2_glock_dq_m - release multiple glocks
1280 * @num_gh: the number of structures
1281 * @ghs: an array of struct gfs2_holder structures
1285 void gfs2_glock_dq_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1289 for (x
= 0; x
< num_gh
; x
++)
1290 gfs2_glock_dq(&ghs
[x
]);
1294 * gfs2_glock_dq_uninit_m - release multiple glocks
1295 * @num_gh: the number of structures
1296 * @ghs: an array of struct gfs2_holder structures
1300 void gfs2_glock_dq_uninit_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1304 for (x
= 0; x
< num_gh
; x
++)
1305 gfs2_glock_dq_uninit(&ghs
[x
]);
1308 void gfs2_glock_cb(struct gfs2_glock
*gl
, unsigned int state
)
1310 unsigned long delay
= 0;
1311 unsigned long holdtime
;
1312 unsigned long now
= jiffies
;
1314 gfs2_glock_hold(gl
);
1315 holdtime
= gl
->gl_tchange
+ gl
->gl_ops
->go_min_hold_time
;
1316 if (test_bit(GLF_QUEUED
, &gl
->gl_flags
)) {
1317 if (time_before(now
, holdtime
))
1318 delay
= holdtime
- now
;
1319 if (test_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
))
1320 delay
= gl
->gl_ops
->go_min_hold_time
;
1323 spin_lock(&gl
->gl_spin
);
1324 handle_callback(gl
, state
, delay
);
1325 spin_unlock(&gl
->gl_spin
);
1326 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, delay
) == 0)
1331 * gfs2_should_freeze - Figure out if glock should be frozen
1332 * @gl: The glock in question
1334 * Glocks are not frozen if (a) the result of the dlm operation is
1335 * an error, (b) the locking operation was an unlock operation or
1336 * (c) if there is a "noexp" flagged request anywhere in the queue
1338 * Returns: 1 if freezing should occur, 0 otherwise
1341 static int gfs2_should_freeze(const struct gfs2_glock
*gl
)
1343 const struct gfs2_holder
*gh
;
1345 if (gl
->gl_reply
& ~LM_OUT_ST_MASK
)
1347 if (gl
->gl_target
== LM_ST_UNLOCKED
)
1350 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
) {
1351 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
1353 if (LM_FLAG_NOEXP
& gh
->gh_flags
)
1361 * gfs2_glock_complete - Callback used by locking
1362 * @gl: Pointer to the glock
1363 * @ret: The return value from the dlm
1367 void gfs2_glock_complete(struct gfs2_glock
*gl
, int ret
)
1369 struct lm_lockstruct
*ls
= &gl
->gl_sbd
->sd_lockstruct
;
1373 if (unlikely(test_bit(DFL_BLOCK_LOCKS
, &ls
->ls_flags
))) {
1374 spin_lock(&gl
->gl_spin
);
1375 if (gfs2_should_freeze(gl
)) {
1376 set_bit(GLF_FROZEN
, &gl
->gl_flags
);
1377 spin_unlock(&gl
->gl_spin
);
1380 spin_unlock(&gl
->gl_spin
);
1382 set_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
);
1383 gfs2_glock_hold(gl
);
1384 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
1389 static int gfs2_shrink_glock_memory(struct shrinker
*shrink
, int nr
, gfp_t gfp_mask
)
1391 struct gfs2_glock
*gl
;
1399 if (!(gfp_mask
& __GFP_FS
))
1402 spin_lock(&lru_lock
);
1403 while(nr
&& !list_empty(&lru_list
)) {
1404 gl
= list_entry(lru_list
.next
, struct gfs2_glock
, gl_lru
);
1405 list_del_init(&gl
->gl_lru
);
1406 atomic_dec(&lru_count
);
1408 /* Test for being demotable */
1409 if (!test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
)) {
1410 gfs2_glock_hold(gl
);
1411 spin_unlock(&lru_lock
);
1412 spin_lock(&gl
->gl_spin
);
1413 may_demote
= demote_ok(gl
);
1415 handle_callback(gl
, LM_ST_UNLOCKED
, 0);
1418 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
1419 smp_mb__after_clear_bit();
1420 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
1421 gfs2_glock_put_nolock(gl
);
1422 spin_unlock(&gl
->gl_spin
);
1423 spin_lock(&lru_lock
);
1427 list_add(&gl
->gl_lru
, &skipped
);
1429 list_splice(&skipped
, &lru_list
);
1430 atomic_add(nr_skipped
, &lru_count
);
1431 spin_unlock(&lru_lock
);
1433 return (atomic_read(&lru_count
) / 100) * sysctl_vfs_cache_pressure
;
1436 static struct shrinker glock_shrinker
= {
1437 .shrink
= gfs2_shrink_glock_memory
,
1438 .seeks
= DEFAULT_SEEKS
,
1442 * examine_bucket - Call a function for glock in a hash bucket
1443 * @examiner: the function
1444 * @sdp: the filesystem
1445 * @bucket: the bucket
1447 * Returns: 1 if the bucket has entries
1450 static int examine_bucket(glock_examiner examiner
, struct gfs2_sbd
*sdp
,
1453 struct gfs2_glock
*gl
, *prev
= NULL
;
1454 int has_entries
= 0;
1455 struct hlist_head
*head
= &gl_hash_table
[hash
].hb_list
;
1457 read_lock(gl_lock_addr(hash
));
1458 /* Can't use hlist_for_each_entry - don't want prefetch here */
1459 if (hlist_empty(head
))
1461 gl
= list_entry(head
->first
, struct gfs2_glock
, gl_list
);
1463 if (!sdp
|| gl
->gl_sbd
== sdp
) {
1464 gfs2_glock_hold(gl
);
1465 read_unlock(gl_lock_addr(hash
));
1467 gfs2_glock_put(prev
);
1471 read_lock(gl_lock_addr(hash
));
1473 if (gl
->gl_list
.next
== NULL
)
1475 gl
= list_entry(gl
->gl_list
.next
, struct gfs2_glock
, gl_list
);
1478 read_unlock(gl_lock_addr(hash
));
1480 gfs2_glock_put(prev
);
1487 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1488 * @gl: The glock to thaw
1490 * N.B. When we freeze a glock, we leave a ref to the glock outstanding,
1491 * so this has to result in the ref count being dropped by one.
1494 static void thaw_glock(struct gfs2_glock
*gl
)
1496 if (!test_and_clear_bit(GLF_FROZEN
, &gl
->gl_flags
))
1498 set_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
);
1499 gfs2_glock_hold(gl
);
1500 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
1505 * clear_glock - look at a glock and see if we can free it from glock cache
1506 * @gl: the glock to look at
1510 static void clear_glock(struct gfs2_glock
*gl
)
1512 spin_lock(&lru_lock
);
1513 if (!list_empty(&gl
->gl_lru
)) {
1514 list_del_init(&gl
->gl_lru
);
1515 atomic_dec(&lru_count
);
1517 spin_unlock(&lru_lock
);
1519 spin_lock(&gl
->gl_spin
);
1520 if (gl
->gl_state
!= LM_ST_UNLOCKED
)
1521 handle_callback(gl
, LM_ST_UNLOCKED
, 0);
1522 spin_unlock(&gl
->gl_spin
);
1523 gfs2_glock_hold(gl
);
1524 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
1529 * gfs2_glock_thaw - Thaw any frozen glocks
1530 * @sdp: The super block
1534 void gfs2_glock_thaw(struct gfs2_sbd
*sdp
)
1538 for (x
= 0; x
< GFS2_GL_HASH_SIZE
; x
++)
1539 examine_bucket(thaw_glock
, sdp
, x
);
1543 * gfs2_gl_hash_clear - Empty out the glock hash table
1544 * @sdp: the filesystem
1545 * @wait: wait until it's all gone
1547 * Called when unmounting the filesystem.
1550 void gfs2_gl_hash_clear(struct gfs2_sbd
*sdp
)
1554 for (x
= 0; x
< GFS2_GL_HASH_SIZE
; x
++)
1555 examine_bucket(clear_glock
, sdp
, x
);
1556 flush_workqueue(glock_workqueue
);
1557 wait_event(sdp
->sd_glock_wait
, atomic_read(&sdp
->sd_glock_disposal
) == 0);
1558 gfs2_dump_lockstate(sdp
);
1561 void gfs2_glock_finish_truncate(struct gfs2_inode
*ip
)
1563 struct gfs2_glock
*gl
= ip
->i_gl
;
1566 ret
= gfs2_truncatei_resume(ip
);
1567 gfs2_assert_withdraw(gl
->gl_sbd
, ret
== 0);
1569 spin_lock(&gl
->gl_spin
);
1570 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
1572 spin_unlock(&gl
->gl_spin
);
1575 static const char *state2str(unsigned state
)
1578 case LM_ST_UNLOCKED
:
1582 case LM_ST_DEFERRED
:
1584 case LM_ST_EXCLUSIVE
:
1590 static const char *hflags2str(char *buf
, unsigned flags
, unsigned long iflags
)
1593 if (flags
& LM_FLAG_TRY
)
1595 if (flags
& LM_FLAG_TRY_1CB
)
1597 if (flags
& LM_FLAG_NOEXP
)
1599 if (flags
& LM_FLAG_ANY
)
1601 if (flags
& LM_FLAG_PRIORITY
)
1603 if (flags
& GL_ASYNC
)
1605 if (flags
& GL_EXACT
)
1607 if (flags
& GL_NOCACHE
)
1609 if (test_bit(HIF_HOLDER
, &iflags
))
1611 if (test_bit(HIF_WAIT
, &iflags
))
1613 if (test_bit(HIF_FIRST
, &iflags
))
1620 * dump_holder - print information about a glock holder
1621 * @seq: the seq_file struct
1622 * @gh: the glock holder
1624 * Returns: 0 on success, -ENOBUFS when we run out of space
1627 static int dump_holder(struct seq_file
*seq
, const struct gfs2_holder
*gh
)
1629 struct task_struct
*gh_owner
= NULL
;
1630 char buffer
[KSYM_SYMBOL_LEN
];
1633 sprint_symbol(buffer
, gh
->gh_ip
);
1634 if (gh
->gh_owner_pid
)
1635 gh_owner
= pid_task(gh
->gh_owner_pid
, PIDTYPE_PID
);
1636 gfs2_print_dbg(seq
, " H: s:%s f:%s e:%d p:%ld [%s] %s\n",
1637 state2str(gh
->gh_state
),
1638 hflags2str(flags_buf
, gh
->gh_flags
, gh
->gh_iflags
),
1640 gh
->gh_owner_pid
? (long)pid_nr(gh
->gh_owner_pid
) : -1,
1641 gh_owner
? gh_owner
->comm
: "(ended)", buffer
);
1645 static const char *gflags2str(char *buf
, const unsigned long *gflags
)
1648 if (test_bit(GLF_LOCK
, gflags
))
1650 if (test_bit(GLF_DEMOTE
, gflags
))
1652 if (test_bit(GLF_PENDING_DEMOTE
, gflags
))
1654 if (test_bit(GLF_DEMOTE_IN_PROGRESS
, gflags
))
1656 if (test_bit(GLF_DIRTY
, gflags
))
1658 if (test_bit(GLF_LFLUSH
, gflags
))
1660 if (test_bit(GLF_INVALIDATE_IN_PROGRESS
, gflags
))
1662 if (test_bit(GLF_REPLY_PENDING
, gflags
))
1664 if (test_bit(GLF_INITIAL
, gflags
))
1666 if (test_bit(GLF_FROZEN
, gflags
))
1668 if (test_bit(GLF_QUEUED
, gflags
))
1675 * __dump_glock - print information about a glock
1676 * @seq: The seq_file struct
1679 * The file format is as follows:
1680 * One line per object, capital letters are used to indicate objects
1681 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1682 * other objects are indented by a single space and follow the glock to
1683 * which they are related. Fields are indicated by lower case letters
1684 * followed by a colon and the field value, except for strings which are in
1685 * [] so that its possible to see if they are composed of spaces for
1686 * example. The field's are n = number (id of the object), f = flags,
1687 * t = type, s = state, r = refcount, e = error, p = pid.
1689 * Returns: 0 on success, -ENOBUFS when we run out of space
1692 static int __dump_glock(struct seq_file
*seq
, const struct gfs2_glock
*gl
)
1694 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1695 unsigned long long dtime
;
1696 const struct gfs2_holder
*gh
;
1697 char gflags_buf
[32];
1700 dtime
= jiffies
- gl
->gl_demote_time
;
1701 dtime
*= 1000000/HZ
; /* demote time in uSec */
1702 if (!test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
1704 gfs2_print_dbg(seq
, "G: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d r:%d\n",
1705 state2str(gl
->gl_state
),
1706 gl
->gl_name
.ln_type
,
1707 (unsigned long long)gl
->gl_name
.ln_number
,
1708 gflags2str(gflags_buf
, &gl
->gl_flags
),
1709 state2str(gl
->gl_target
),
1710 state2str(gl
->gl_demote_state
), dtime
,
1711 atomic_read(&gl
->gl_ail_count
),
1712 atomic_read(&gl
->gl_ref
));
1714 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
) {
1715 error
= dump_holder(seq
, gh
);
1719 if (gl
->gl_state
!= LM_ST_UNLOCKED
&& glops
->go_dump
)
1720 error
= glops
->go_dump(seq
, gl
);
1725 static int dump_glock(struct seq_file
*seq
, struct gfs2_glock
*gl
)
1728 spin_lock(&gl
->gl_spin
);
1729 ret
= __dump_glock(seq
, gl
);
1730 spin_unlock(&gl
->gl_spin
);
1735 * gfs2_dump_lockstate - print out the current lockstate
1736 * @sdp: the filesystem
1737 * @ub: the buffer to copy the information into
1739 * If @ub is NULL, dump the lockstate to the console.
1743 static int gfs2_dump_lockstate(struct gfs2_sbd
*sdp
)
1745 struct gfs2_glock
*gl
;
1746 struct hlist_node
*h
;
1750 for (x
= 0; x
< GFS2_GL_HASH_SIZE
; x
++) {
1752 read_lock(gl_lock_addr(x
));
1754 hlist_for_each_entry(gl
, h
, &gl_hash_table
[x
].hb_list
, gl_list
) {
1755 if (gl
->gl_sbd
!= sdp
)
1758 error
= dump_glock(NULL
, gl
);
1763 read_unlock(gl_lock_addr(x
));
1774 int __init
gfs2_glock_init(void)
1777 for(i
= 0; i
< GFS2_GL_HASH_SIZE
; i
++) {
1778 INIT_HLIST_HEAD(&gl_hash_table
[i
].hb_list
);
1780 #ifdef GL_HASH_LOCK_SZ
1781 for(i
= 0; i
< GL_HASH_LOCK_SZ
; i
++) {
1782 rwlock_init(&gl_hash_locks
[i
]);
1786 glock_workqueue
= alloc_workqueue("glock_workqueue", WQ_RESCUER
|
1787 WQ_HIGHPRI
| WQ_FREEZEABLE
, 0);
1788 if (IS_ERR(glock_workqueue
))
1789 return PTR_ERR(glock_workqueue
);
1790 gfs2_delete_workqueue
= alloc_workqueue("delete_workqueue", WQ_RESCUER
|
1792 if (IS_ERR(gfs2_delete_workqueue
)) {
1793 destroy_workqueue(glock_workqueue
);
1794 return PTR_ERR(gfs2_delete_workqueue
);
1797 register_shrinker(&glock_shrinker
);
1802 void gfs2_glock_exit(void)
1804 unregister_shrinker(&glock_shrinker
);
1805 destroy_workqueue(glock_workqueue
);
1806 destroy_workqueue(gfs2_delete_workqueue
);
1809 static int gfs2_glock_iter_next(struct gfs2_glock_iter
*gi
)
1811 struct gfs2_glock
*gl
;
1814 read_lock(gl_lock_addr(gi
->hash
));
1817 gi
->gl
= hlist_entry(gl
->gl_list
.next
,
1818 struct gfs2_glock
, gl_list
);
1820 gi
->gl
= hlist_entry(gl_hash_table
[gi
->hash
].hb_list
.first
,
1821 struct gfs2_glock
, gl_list
);
1824 gfs2_glock_hold(gi
->gl
);
1825 read_unlock(gl_lock_addr(gi
->hash
));
1828 while (gi
->gl
== NULL
) {
1830 if (gi
->hash
>= GFS2_GL_HASH_SIZE
)
1832 read_lock(gl_lock_addr(gi
->hash
));
1833 gi
->gl
= hlist_entry(gl_hash_table
[gi
->hash
].hb_list
.first
,
1834 struct gfs2_glock
, gl_list
);
1836 gfs2_glock_hold(gi
->gl
);
1837 read_unlock(gl_lock_addr(gi
->hash
));
1840 if (gi
->sdp
!= gi
->gl
->gl_sbd
)
1846 static void gfs2_glock_iter_free(struct gfs2_glock_iter
*gi
)
1849 gfs2_glock_put(gi
->gl
);
1853 static void *gfs2_glock_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1855 struct gfs2_glock_iter
*gi
= seq
->private;
1861 if (gfs2_glock_iter_next(gi
)) {
1862 gfs2_glock_iter_free(gi
);
1870 static void *gfs2_glock_seq_next(struct seq_file
*seq
, void *iter_ptr
,
1873 struct gfs2_glock_iter
*gi
= seq
->private;
1877 if (gfs2_glock_iter_next(gi
)) {
1878 gfs2_glock_iter_free(gi
);
1885 static void gfs2_glock_seq_stop(struct seq_file
*seq
, void *iter_ptr
)
1887 struct gfs2_glock_iter
*gi
= seq
->private;
1888 gfs2_glock_iter_free(gi
);
1891 static int gfs2_glock_seq_show(struct seq_file
*seq
, void *iter_ptr
)
1893 return dump_glock(seq
, iter_ptr
);
1896 static const struct seq_operations gfs2_glock_seq_ops
= {
1897 .start
= gfs2_glock_seq_start
,
1898 .next
= gfs2_glock_seq_next
,
1899 .stop
= gfs2_glock_seq_stop
,
1900 .show
= gfs2_glock_seq_show
,
1903 static int gfs2_debugfs_open(struct inode
*inode
, struct file
*file
)
1905 int ret
= seq_open_private(file
, &gfs2_glock_seq_ops
,
1906 sizeof(struct gfs2_glock_iter
));
1908 struct seq_file
*seq
= file
->private_data
;
1909 struct gfs2_glock_iter
*gi
= seq
->private;
1910 gi
->sdp
= inode
->i_private
;
1915 static const struct file_operations gfs2_debug_fops
= {
1916 .owner
= THIS_MODULE
,
1917 .open
= gfs2_debugfs_open
,
1919 .llseek
= seq_lseek
,
1920 .release
= seq_release_private
,
1923 int gfs2_create_debugfs_file(struct gfs2_sbd
*sdp
)
1925 sdp
->debugfs_dir
= debugfs_create_dir(sdp
->sd_table_name
, gfs2_root
);
1926 if (!sdp
->debugfs_dir
)
1928 sdp
->debugfs_dentry_glocks
= debugfs_create_file("glocks",
1930 sdp
->debugfs_dir
, sdp
,
1932 if (!sdp
->debugfs_dentry_glocks
)
1938 void gfs2_delete_debugfs_file(struct gfs2_sbd
*sdp
)
1940 if (sdp
&& sdp
->debugfs_dir
) {
1941 if (sdp
->debugfs_dentry_glocks
) {
1942 debugfs_remove(sdp
->debugfs_dentry_glocks
);
1943 sdp
->debugfs_dentry_glocks
= NULL
;
1945 debugfs_remove(sdp
->debugfs_dir
);
1946 sdp
->debugfs_dir
= NULL
;
1950 int gfs2_register_debugfs(void)
1952 gfs2_root
= debugfs_create_dir("gfs2", NULL
);
1953 return gfs2_root
? 0 : -ENOMEM
;
1956 void gfs2_unregister_debugfs(void)
1958 debugfs_remove(gfs2_root
);