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>
29 #include <linux/rcupdate.h>
30 #include <linux/rculist_bl.h>
31 #include <linux/bit_spinlock.h>
32 #include <linux/percpu.h>
33 #include <linux/list_sort.h>
34 #include <linux/lockref.h>
47 #define CREATE_TRACE_POINTS
48 #include "trace_gfs2.h"
50 struct gfs2_glock_iter
{
51 int hash
; /* hash bucket index */
52 unsigned nhash
; /* Index within current bucket */
53 struct gfs2_sbd
*sdp
; /* incore superblock */
54 struct gfs2_glock
*gl
; /* current glock struct */
55 loff_t last_pos
; /* last position */
58 typedef void (*glock_examiner
) (struct gfs2_glock
* gl
);
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 hlist_bl_head gl_hash_table
[GFS2_GL_HASH_SIZE
];
74 static struct dentry
*gfs2_root
;
77 * gl_hash() - Turn glock number into hash bucket number
78 * @lock: The glock number
80 * Returns: The number of the corresponding hash bucket
83 static unsigned int gl_hash(const struct gfs2_sbd
*sdp
,
84 const struct lm_lockname
*name
)
88 h
= jhash(&name
->ln_number
, sizeof(u64
), 0);
89 h
= jhash(&name
->ln_type
, sizeof(unsigned int), h
);
90 h
= jhash(&sdp
, sizeof(struct gfs2_sbd
*), h
);
91 h
&= GFS2_GL_HASH_MASK
;
96 static inline void spin_lock_bucket(unsigned int hash
)
98 hlist_bl_lock(&gl_hash_table
[hash
]);
101 static inline void spin_unlock_bucket(unsigned int hash
)
103 hlist_bl_unlock(&gl_hash_table
[hash
]);
106 static void gfs2_glock_dealloc(struct rcu_head
*rcu
)
108 struct gfs2_glock
*gl
= container_of(rcu
, struct gfs2_glock
, gl_rcu
);
110 if (gl
->gl_ops
->go_flags
& GLOF_ASPACE
) {
111 kmem_cache_free(gfs2_glock_aspace_cachep
, gl
);
113 kfree(gl
->gl_lksb
.sb_lvbptr
);
114 kmem_cache_free(gfs2_glock_cachep
, gl
);
118 void gfs2_glock_free(struct gfs2_glock
*gl
)
120 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
122 call_rcu(&gl
->gl_rcu
, gfs2_glock_dealloc
);
123 if (atomic_dec_and_test(&sdp
->sd_glock_disposal
))
124 wake_up(&sdp
->sd_glock_wait
);
128 * gfs2_glock_hold() - increment reference count on glock
129 * @gl: The glock to hold
133 static void gfs2_glock_hold(struct gfs2_glock
*gl
)
135 GLOCK_BUG_ON(gl
, __lockref_is_dead(&gl
->gl_lockref
));
136 lockref_get(&gl
->gl_lockref
);
140 * demote_ok - Check to see if it's ok to unlock a glock
143 * Returns: 1 if it's ok
146 static int demote_ok(const struct gfs2_glock
*gl
)
148 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
150 if (gl
->gl_state
== LM_ST_UNLOCKED
)
152 if (!list_empty(&gl
->gl_holders
))
154 if (glops
->go_demote_ok
)
155 return glops
->go_demote_ok(gl
);
160 void gfs2_glock_add_to_lru(struct gfs2_glock
*gl
)
162 spin_lock(&lru_lock
);
164 if (!list_empty(&gl
->gl_lru
))
165 list_del_init(&gl
->gl_lru
);
167 atomic_inc(&lru_count
);
169 list_add_tail(&gl
->gl_lru
, &lru_list
);
170 set_bit(GLF_LRU
, &gl
->gl_flags
);
171 spin_unlock(&lru_lock
);
174 static void __gfs2_glock_remove_from_lru(struct gfs2_glock
*gl
)
176 if (!list_empty(&gl
->gl_lru
)) {
177 list_del_init(&gl
->gl_lru
);
178 atomic_dec(&lru_count
);
179 clear_bit(GLF_LRU
, &gl
->gl_flags
);
183 static void gfs2_glock_remove_from_lru(struct gfs2_glock
*gl
)
185 spin_lock(&lru_lock
);
186 __gfs2_glock_remove_from_lru(gl
);
187 spin_unlock(&lru_lock
);
191 * gfs2_glock_put() - Decrement reference count on glock
192 * @gl: The glock to put
196 void gfs2_glock_put(struct gfs2_glock
*gl
)
198 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
199 struct address_space
*mapping
= gfs2_glock2aspace(gl
);
201 if (lockref_put_or_lock(&gl
->gl_lockref
))
204 lockref_mark_dead(&gl
->gl_lockref
);
206 spin_lock(&lru_lock
);
207 __gfs2_glock_remove_from_lru(gl
);
208 spin_unlock(&lru_lock
);
209 spin_unlock(&gl
->gl_lockref
.lock
);
210 spin_lock_bucket(gl
->gl_hash
);
211 hlist_bl_del_rcu(&gl
->gl_list
);
212 spin_unlock_bucket(gl
->gl_hash
);
213 GLOCK_BUG_ON(gl
, !list_empty(&gl
->gl_holders
));
214 GLOCK_BUG_ON(gl
, mapping
&& mapping
->nrpages
);
215 trace_gfs2_glock_put(gl
);
216 sdp
->sd_lockstruct
.ls_ops
->lm_put_lock(gl
);
220 * search_bucket() - Find struct gfs2_glock by lock number
221 * @bucket: the bucket to search
222 * @name: The lock name
224 * Returns: NULL, or the struct gfs2_glock with the requested number
227 static struct gfs2_glock
*search_bucket(unsigned int hash
,
228 const struct gfs2_sbd
*sdp
,
229 const struct lm_lockname
*name
)
231 struct gfs2_glock
*gl
;
232 struct hlist_bl_node
*h
;
234 hlist_bl_for_each_entry_rcu(gl
, h
, &gl_hash_table
[hash
], gl_list
) {
235 if (!lm_name_equal(&gl
->gl_name
, name
))
237 if (gl
->gl_sbd
!= sdp
)
239 if (lockref_get_not_dead(&gl
->gl_lockref
))
247 * may_grant - check if its ok to grant a new lock
249 * @gh: The lock request which we wish to grant
251 * Returns: true if its ok to grant the lock
254 static inline int may_grant(const struct gfs2_glock
*gl
, const struct gfs2_holder
*gh
)
256 const struct gfs2_holder
*gh_head
= list_entry(gl
->gl_holders
.next
, const struct gfs2_holder
, gh_list
);
257 if ((gh
->gh_state
== LM_ST_EXCLUSIVE
||
258 gh_head
->gh_state
== LM_ST_EXCLUSIVE
) && gh
!= gh_head
)
260 if (gl
->gl_state
== gh
->gh_state
)
262 if (gh
->gh_flags
& GL_EXACT
)
264 if (gl
->gl_state
== LM_ST_EXCLUSIVE
) {
265 if (gh
->gh_state
== LM_ST_SHARED
&& gh_head
->gh_state
== LM_ST_SHARED
)
267 if (gh
->gh_state
== LM_ST_DEFERRED
&& gh_head
->gh_state
== LM_ST_DEFERRED
)
270 if (gl
->gl_state
!= LM_ST_UNLOCKED
&& (gh
->gh_flags
& LM_FLAG_ANY
))
275 static void gfs2_holder_wake(struct gfs2_holder
*gh
)
277 clear_bit(HIF_WAIT
, &gh
->gh_iflags
);
278 smp_mb__after_clear_bit();
279 wake_up_bit(&gh
->gh_iflags
, HIF_WAIT
);
283 * do_error - Something unexpected has happened during a lock request
287 static inline void do_error(struct gfs2_glock
*gl
, const int ret
)
289 struct gfs2_holder
*gh
, *tmp
;
291 list_for_each_entry_safe(gh
, tmp
, &gl
->gl_holders
, gh_list
) {
292 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
294 if (ret
& LM_OUT_ERROR
)
296 else if (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
))
297 gh
->gh_error
= GLR_TRYFAILED
;
300 list_del_init(&gh
->gh_list
);
301 trace_gfs2_glock_queue(gh
, 0);
302 gfs2_holder_wake(gh
);
307 * do_promote - promote as many requests as possible on the current queue
310 * Returns: 1 if there is a blocked holder at the head of the list, or 2
311 * if a type specific operation is underway.
314 static int do_promote(struct gfs2_glock
*gl
)
315 __releases(&gl
->gl_spin
)
316 __acquires(&gl
->gl_spin
)
318 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
319 struct gfs2_holder
*gh
, *tmp
;
323 list_for_each_entry_safe(gh
, tmp
, &gl
->gl_holders
, gh_list
) {
324 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
326 if (may_grant(gl
, gh
)) {
327 if (gh
->gh_list
.prev
== &gl
->gl_holders
&&
329 spin_unlock(&gl
->gl_spin
);
330 /* FIXME: eliminate this eventually */
331 ret
= glops
->go_lock(gh
);
332 spin_lock(&gl
->gl_spin
);
337 list_del_init(&gh
->gh_list
);
338 trace_gfs2_glock_queue(gh
, 0);
339 gfs2_holder_wake(gh
);
342 set_bit(HIF_HOLDER
, &gh
->gh_iflags
);
343 trace_gfs2_promote(gh
, 1);
344 gfs2_holder_wake(gh
);
347 set_bit(HIF_HOLDER
, &gh
->gh_iflags
);
348 trace_gfs2_promote(gh
, 0);
349 gfs2_holder_wake(gh
);
352 if (gh
->gh_list
.prev
== &gl
->gl_holders
)
361 * find_first_waiter - find the first gh that's waiting for the glock
365 static inline struct gfs2_holder
*find_first_waiter(const struct gfs2_glock
*gl
)
367 struct gfs2_holder
*gh
;
369 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
) {
370 if (!test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
377 * state_change - record that the glock is now in a different state
379 * @new_state the new state
383 static void state_change(struct gfs2_glock
*gl
, unsigned int new_state
)
387 held1
= (gl
->gl_state
!= LM_ST_UNLOCKED
);
388 held2
= (new_state
!= LM_ST_UNLOCKED
);
390 if (held1
!= held2
) {
391 GLOCK_BUG_ON(gl
, __lockref_is_dead(&gl
->gl_lockref
));
393 gl
->gl_lockref
.count
++;
395 gl
->gl_lockref
.count
--;
397 if (held1
&& held2
&& list_empty(&gl
->gl_holders
))
398 clear_bit(GLF_QUEUED
, &gl
->gl_flags
);
400 if (new_state
!= gl
->gl_target
)
401 /* shorten our minimum hold time */
402 gl
->gl_hold_time
= max(gl
->gl_hold_time
- GL_GLOCK_HOLD_DECR
,
404 gl
->gl_state
= new_state
;
405 gl
->gl_tchange
= jiffies
;
408 static void gfs2_demote_wake(struct gfs2_glock
*gl
)
410 gl
->gl_demote_state
= LM_ST_EXCLUSIVE
;
411 clear_bit(GLF_DEMOTE
, &gl
->gl_flags
);
412 smp_mb__after_clear_bit();
413 wake_up_bit(&gl
->gl_flags
, GLF_DEMOTE
);
417 * finish_xmote - The DLM has replied to one of our lock requests
419 * @ret: The status from the DLM
423 static void finish_xmote(struct gfs2_glock
*gl
, unsigned int ret
)
425 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
426 struct gfs2_holder
*gh
;
427 unsigned state
= ret
& LM_OUT_ST_MASK
;
430 spin_lock(&gl
->gl_spin
);
431 trace_gfs2_glock_state_change(gl
, state
);
432 state_change(gl
, state
);
433 gh
= find_first_waiter(gl
);
435 /* Demote to UN request arrived during demote to SH or DF */
436 if (test_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
) &&
437 state
!= LM_ST_UNLOCKED
&& gl
->gl_demote_state
== LM_ST_UNLOCKED
)
438 gl
->gl_target
= LM_ST_UNLOCKED
;
440 /* Check for state != intended state */
441 if (unlikely(state
!= gl
->gl_target
)) {
442 if (gh
&& !test_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
)) {
443 /* move to back of queue and try next entry */
444 if (ret
& LM_OUT_CANCELED
) {
445 if ((gh
->gh_flags
& LM_FLAG_PRIORITY
) == 0)
446 list_move_tail(&gh
->gh_list
, &gl
->gl_holders
);
447 gh
= find_first_waiter(gl
);
448 gl
->gl_target
= gh
->gh_state
;
451 /* Some error or failed "try lock" - report it */
452 if ((ret
& LM_OUT_ERROR
) ||
453 (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
))) {
454 gl
->gl_target
= gl
->gl_state
;
460 /* Unlocked due to conversion deadlock, try again */
463 do_xmote(gl
, gh
, gl
->gl_target
);
465 /* Conversion fails, unlock and try again */
468 do_xmote(gl
, gh
, LM_ST_UNLOCKED
);
470 default: /* Everything else */
471 printk(KERN_ERR
"GFS2: wanted %u got %u\n", gl
->gl_target
, state
);
474 spin_unlock(&gl
->gl_spin
);
478 /* Fast path - we got what we asked for */
479 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
))
480 gfs2_demote_wake(gl
);
481 if (state
!= LM_ST_UNLOCKED
) {
482 if (glops
->go_xmote_bh
) {
483 spin_unlock(&gl
->gl_spin
);
484 rv
= glops
->go_xmote_bh(gl
, gh
);
485 spin_lock(&gl
->gl_spin
);
496 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
498 spin_unlock(&gl
->gl_spin
);
502 * do_xmote - Calls the DLM to change the state of a lock
503 * @gl: The lock state
504 * @gh: The holder (only for promotes)
505 * @target: The target lock state
509 static void do_xmote(struct gfs2_glock
*gl
, struct gfs2_holder
*gh
, unsigned int target
)
510 __releases(&gl
->gl_spin
)
511 __acquires(&gl
->gl_spin
)
513 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
514 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
515 unsigned int lck_flags
= gh
? gh
->gh_flags
: 0;
518 lck_flags
&= (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
| LM_FLAG_NOEXP
|
520 GLOCK_BUG_ON(gl
, gl
->gl_state
== target
);
521 GLOCK_BUG_ON(gl
, gl
->gl_state
== gl
->gl_target
);
522 if ((target
== LM_ST_UNLOCKED
|| target
== LM_ST_DEFERRED
) &&
524 set_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
);
525 do_error(gl
, 0); /* Fail queued try locks */
528 set_bit(GLF_BLOCKING
, &gl
->gl_flags
);
529 if ((gl
->gl_req
== LM_ST_UNLOCKED
) ||
530 (gl
->gl_state
== LM_ST_EXCLUSIVE
) ||
531 (lck_flags
& (LM_FLAG_TRY
|LM_FLAG_TRY_1CB
)))
532 clear_bit(GLF_BLOCKING
, &gl
->gl_flags
);
533 spin_unlock(&gl
->gl_spin
);
536 if (test_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
))
537 glops
->go_inval(gl
, target
== LM_ST_DEFERRED
? 0 : DIO_METADATA
);
538 clear_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
);
541 if (sdp
->sd_lockstruct
.ls_ops
->lm_lock
) {
543 ret
= sdp
->sd_lockstruct
.ls_ops
->lm_lock(gl
, target
, lck_flags
);
545 printk(KERN_ERR
"GFS2: lm_lock ret %d\n", ret
);
548 } else { /* lock_nolock */
549 finish_xmote(gl
, target
);
550 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
554 spin_lock(&gl
->gl_spin
);
558 * find_first_holder - find the first "holder" gh
562 static inline struct gfs2_holder
*find_first_holder(const struct gfs2_glock
*gl
)
564 struct gfs2_holder
*gh
;
566 if (!list_empty(&gl
->gl_holders
)) {
567 gh
= list_entry(gl
->gl_holders
.next
, struct gfs2_holder
, gh_list
);
568 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
575 * run_queue - do all outstanding tasks related to a glock
576 * @gl: The glock in question
577 * @nonblock: True if we must not block in run_queue
581 static void run_queue(struct gfs2_glock
*gl
, const int nonblock
)
582 __releases(&gl
->gl_spin
)
583 __acquires(&gl
->gl_spin
)
585 struct gfs2_holder
*gh
= NULL
;
588 if (test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
))
591 GLOCK_BUG_ON(gl
, test_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
));
593 if (test_bit(GLF_DEMOTE
, &gl
->gl_flags
) &&
594 gl
->gl_demote_state
!= gl
->gl_state
) {
595 if (find_first_holder(gl
))
599 set_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
);
600 GLOCK_BUG_ON(gl
, gl
->gl_demote_state
== LM_ST_EXCLUSIVE
);
601 gl
->gl_target
= gl
->gl_demote_state
;
603 if (test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
604 gfs2_demote_wake(gl
);
605 ret
= do_promote(gl
);
610 gh
= find_first_waiter(gl
);
611 gl
->gl_target
= gh
->gh_state
;
612 if (!(gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
)))
613 do_error(gl
, 0); /* Fail queued try locks */
615 do_xmote(gl
, gh
, gl
->gl_target
);
620 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
621 smp_mb__after_clear_bit();
622 gl
->gl_lockref
.count
++;
623 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
624 gl
->gl_lockref
.count
--;
628 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
629 smp_mb__after_clear_bit();
633 static void delete_work_func(struct work_struct
*work
)
635 struct gfs2_glock
*gl
= container_of(work
, struct gfs2_glock
, gl_delete
);
636 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
637 struct gfs2_inode
*ip
;
639 u64 no_addr
= gl
->gl_name
.ln_number
;
642 /* Note: Unsafe to dereference ip as we don't hold right refs/locks */
645 inode
= gfs2_ilookup(sdp
->sd_vfs
, no_addr
, 1);
647 inode
= gfs2_lookup_by_inum(sdp
, no_addr
, NULL
, GFS2_BLKST_UNLINKED
);
648 if (inode
&& !IS_ERR(inode
)) {
649 d_prune_aliases(inode
);
655 static void glock_work_func(struct work_struct
*work
)
657 unsigned long delay
= 0;
658 struct gfs2_glock
*gl
= container_of(work
, struct gfs2_glock
, gl_work
.work
);
661 if (test_and_clear_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
)) {
662 finish_xmote(gl
, gl
->gl_reply
);
665 spin_lock(&gl
->gl_spin
);
666 if (test_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
) &&
667 gl
->gl_state
!= LM_ST_UNLOCKED
&&
668 gl
->gl_demote_state
!= LM_ST_EXCLUSIVE
) {
669 unsigned long holdtime
, now
= jiffies
;
671 holdtime
= gl
->gl_tchange
+ gl
->gl_hold_time
;
672 if (time_before(now
, holdtime
))
673 delay
= holdtime
- now
;
676 clear_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
);
677 set_bit(GLF_DEMOTE
, &gl
->gl_flags
);
681 spin_unlock(&gl
->gl_spin
);
685 if (gl
->gl_name
.ln_type
!= LM_TYPE_INODE
)
687 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, delay
) == 0)
695 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
696 * @sdp: The GFS2 superblock
697 * @number: the lock number
698 * @glops: The glock_operations to use
699 * @create: If 0, don't create the glock if it doesn't exist
700 * @glp: the glock is returned here
702 * This does not lock a glock, just finds/creates structures for one.
707 int gfs2_glock_get(struct gfs2_sbd
*sdp
, u64 number
,
708 const struct gfs2_glock_operations
*glops
, int create
,
709 struct gfs2_glock
**glp
)
711 struct super_block
*s
= sdp
->sd_vfs
;
712 struct lm_lockname name
= { .ln_number
= number
, .ln_type
= glops
->go_type
};
713 struct gfs2_glock
*gl
, *tmp
;
714 unsigned int hash
= gl_hash(sdp
, &name
);
715 struct address_space
*mapping
;
716 struct kmem_cache
*cachep
;
719 gl
= search_bucket(hash
, sdp
, &name
);
728 if (glops
->go_flags
& GLOF_ASPACE
)
729 cachep
= gfs2_glock_aspace_cachep
;
731 cachep
= gfs2_glock_cachep
;
732 gl
= kmem_cache_alloc(cachep
, GFP_KERNEL
);
736 memset(&gl
->gl_lksb
, 0, sizeof(struct dlm_lksb
));
738 if (glops
->go_flags
& GLOF_LVB
) {
739 gl
->gl_lksb
.sb_lvbptr
= kzalloc(GFS2_MIN_LVB_SIZE
, GFP_KERNEL
);
740 if (!gl
->gl_lksb
.sb_lvbptr
) {
741 kmem_cache_free(cachep
, gl
);
746 atomic_inc(&sdp
->sd_glock_disposal
);
750 gl
->gl_lockref
.count
= 1;
751 gl
->gl_state
= LM_ST_UNLOCKED
;
752 gl
->gl_target
= LM_ST_UNLOCKED
;
753 gl
->gl_demote_state
= LM_ST_EXCLUSIVE
;
756 gl
->gl_dstamp
= ktime_set(0, 0);
758 /* We use the global stats to estimate the initial per-glock stats */
759 gl
->gl_stats
= this_cpu_ptr(sdp
->sd_lkstats
)->lkstats
[glops
->go_type
];
761 gl
->gl_stats
.stats
[GFS2_LKS_DCOUNT
] = 0;
762 gl
->gl_stats
.stats
[GFS2_LKS_QCOUNT
] = 0;
763 gl
->gl_tchange
= jiffies
;
764 gl
->gl_object
= NULL
;
765 gl
->gl_hold_time
= GL_GLOCK_DFT_HOLD
;
766 INIT_DELAYED_WORK(&gl
->gl_work
, glock_work_func
);
767 INIT_WORK(&gl
->gl_delete
, delete_work_func
);
769 mapping
= gfs2_glock2aspace(gl
);
771 mapping
->a_ops
= &gfs2_meta_aops
;
772 mapping
->host
= s
->s_bdev
->bd_inode
;
774 mapping_set_gfp_mask(mapping
, GFP_NOFS
);
775 mapping
->private_data
= NULL
;
776 mapping
->backing_dev_info
= s
->s_bdi
;
777 mapping
->writeback_index
= 0;
780 spin_lock_bucket(hash
);
781 tmp
= search_bucket(hash
, sdp
, &name
);
783 spin_unlock_bucket(hash
);
784 kfree(gl
->gl_lksb
.sb_lvbptr
);
785 kmem_cache_free(cachep
, gl
);
786 atomic_dec(&sdp
->sd_glock_disposal
);
789 hlist_bl_add_head_rcu(&gl
->gl_list
, &gl_hash_table
[hash
]);
790 spin_unlock_bucket(hash
);
799 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
801 * @state: the state we're requesting
802 * @flags: the modifier flags
803 * @gh: the holder structure
807 void gfs2_holder_init(struct gfs2_glock
*gl
, unsigned int state
, unsigned flags
,
808 struct gfs2_holder
*gh
)
810 INIT_LIST_HEAD(&gh
->gh_list
);
812 gh
->gh_ip
= (unsigned long)__builtin_return_address(0);
813 gh
->gh_owner_pid
= get_pid(task_pid(current
));
814 gh
->gh_state
= state
;
815 gh
->gh_flags
= flags
;
822 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
823 * @state: the state we're requesting
824 * @flags: the modifier flags
825 * @gh: the holder structure
827 * Don't mess with the glock.
831 void gfs2_holder_reinit(unsigned int state
, unsigned flags
, struct gfs2_holder
*gh
)
833 gh
->gh_state
= state
;
834 gh
->gh_flags
= flags
;
836 gh
->gh_ip
= (unsigned long)__builtin_return_address(0);
837 if (gh
->gh_owner_pid
)
838 put_pid(gh
->gh_owner_pid
);
839 gh
->gh_owner_pid
= get_pid(task_pid(current
));
843 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
844 * @gh: the holder structure
848 void gfs2_holder_uninit(struct gfs2_holder
*gh
)
850 put_pid(gh
->gh_owner_pid
);
851 gfs2_glock_put(gh
->gh_gl
);
857 * gfs2_glock_holder_wait
860 * This function and gfs2_glock_demote_wait both show up in the WCHAN
861 * field. Thus I've separated these otherwise identical functions in
862 * order to be more informative to the user.
865 static int gfs2_glock_holder_wait(void *word
)
871 static int gfs2_glock_demote_wait(void *word
)
878 * gfs2_glock_wait - wait on a glock acquisition
879 * @gh: the glock holder
881 * Returns: 0 on success
884 int gfs2_glock_wait(struct gfs2_holder
*gh
)
886 unsigned long time1
= jiffies
;
889 wait_on_bit(&gh
->gh_iflags
, HIF_WAIT
, gfs2_glock_holder_wait
, TASK_UNINTERRUPTIBLE
);
890 if (time_after(jiffies
, time1
+ HZ
)) /* have we waited > a second? */
891 /* Lengthen the minimum hold time. */
892 gh
->gh_gl
->gl_hold_time
= min(gh
->gh_gl
->gl_hold_time
+
899 * handle_callback - process a demote request
901 * @state: the state the caller wants us to change to
903 * There are only two requests that we are going to see in actual
904 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
907 static void handle_callback(struct gfs2_glock
*gl
, unsigned int state
,
908 unsigned long delay
, bool remote
)
910 int bit
= delay
? GLF_PENDING_DEMOTE
: GLF_DEMOTE
;
912 set_bit(bit
, &gl
->gl_flags
);
913 if (gl
->gl_demote_state
== LM_ST_EXCLUSIVE
) {
914 gl
->gl_demote_state
= state
;
915 gl
->gl_demote_time
= jiffies
;
916 } else if (gl
->gl_demote_state
!= LM_ST_UNLOCKED
&&
917 gl
->gl_demote_state
!= state
) {
918 gl
->gl_demote_state
= LM_ST_UNLOCKED
;
920 if (gl
->gl_ops
->go_callback
)
921 gl
->gl_ops
->go_callback(gl
, remote
);
922 trace_gfs2_demote_rq(gl
, remote
);
925 void gfs2_print_dbg(struct seq_file
*seq
, const char *fmt
, ...)
927 struct va_format vaf
;
933 seq_vprintf(seq
, fmt
, args
);
938 printk(KERN_ERR
" %pV", &vaf
);
945 * add_to_queue - Add a holder to the wait queue (but look for recursion)
946 * @gh: the holder structure to add
948 * Eventually we should move the recursive locking trap to a
949 * debugging option or something like that. This is the fast
950 * path and needs to have the minimum number of distractions.
954 static inline void add_to_queue(struct gfs2_holder
*gh
)
955 __releases(&gl
->gl_spin
)
956 __acquires(&gl
->gl_spin
)
958 struct gfs2_glock
*gl
= gh
->gh_gl
;
959 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
960 struct list_head
*insert_pt
= NULL
;
961 struct gfs2_holder
*gh2
;
964 BUG_ON(gh
->gh_owner_pid
== NULL
);
965 if (test_and_set_bit(HIF_WAIT
, &gh
->gh_iflags
))
968 if (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
)) {
969 if (test_bit(GLF_LOCK
, &gl
->gl_flags
))
970 try_futile
= !may_grant(gl
, gh
);
971 if (test_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
))
975 list_for_each_entry(gh2
, &gl
->gl_holders
, gh_list
) {
976 if (unlikely(gh2
->gh_owner_pid
== gh
->gh_owner_pid
&&
977 (gh
->gh_gl
->gl_ops
->go_type
!= LM_TYPE_FLOCK
)))
980 !(gh2
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
))) {
982 gh
->gh_error
= GLR_TRYFAILED
;
983 gfs2_holder_wake(gh
);
986 if (test_bit(HIF_HOLDER
, &gh2
->gh_iflags
))
988 if (unlikely((gh
->gh_flags
& LM_FLAG_PRIORITY
) && !insert_pt
))
989 insert_pt
= &gh2
->gh_list
;
991 set_bit(GLF_QUEUED
, &gl
->gl_flags
);
992 trace_gfs2_glock_queue(gh
, 1);
993 gfs2_glstats_inc(gl
, GFS2_LKS_QCOUNT
);
994 gfs2_sbstats_inc(gl
, GFS2_LKS_QCOUNT
);
995 if (likely(insert_pt
== NULL
)) {
996 list_add_tail(&gh
->gh_list
, &gl
->gl_holders
);
997 if (unlikely(gh
->gh_flags
& LM_FLAG_PRIORITY
))
1001 list_add_tail(&gh
->gh_list
, insert_pt
);
1003 gh
= list_entry(gl
->gl_holders
.next
, struct gfs2_holder
, gh_list
);
1004 if (!(gh
->gh_flags
& LM_FLAG_PRIORITY
)) {
1005 spin_unlock(&gl
->gl_spin
);
1006 if (sdp
->sd_lockstruct
.ls_ops
->lm_cancel
)
1007 sdp
->sd_lockstruct
.ls_ops
->lm_cancel(gl
);
1008 spin_lock(&gl
->gl_spin
);
1013 printk(KERN_ERR
"original: %pSR\n", (void *)gh2
->gh_ip
);
1014 printk(KERN_ERR
"pid: %d\n", pid_nr(gh2
->gh_owner_pid
));
1015 printk(KERN_ERR
"lock type: %d req lock state : %d\n",
1016 gh2
->gh_gl
->gl_name
.ln_type
, gh2
->gh_state
);
1017 printk(KERN_ERR
"new: %pSR\n", (void *)gh
->gh_ip
);
1018 printk(KERN_ERR
"pid: %d\n", pid_nr(gh
->gh_owner_pid
));
1019 printk(KERN_ERR
"lock type: %d req lock state : %d\n",
1020 gh
->gh_gl
->gl_name
.ln_type
, gh
->gh_state
);
1021 gfs2_dump_glock(NULL
, gl
);
1026 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1027 * @gh: the holder structure
1029 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1031 * Returns: 0, GLR_TRYFAILED, or errno on failure
1034 int gfs2_glock_nq(struct gfs2_holder
*gh
)
1036 struct gfs2_glock
*gl
= gh
->gh_gl
;
1037 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
1040 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)))
1043 if (test_bit(GLF_LRU
, &gl
->gl_flags
))
1044 gfs2_glock_remove_from_lru(gl
);
1046 spin_lock(&gl
->gl_spin
);
1048 if ((LM_FLAG_NOEXP
& gh
->gh_flags
) &&
1049 test_and_clear_bit(GLF_FROZEN
, &gl
->gl_flags
))
1050 set_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
);
1052 spin_unlock(&gl
->gl_spin
);
1054 if (!(gh
->gh_flags
& GL_ASYNC
))
1055 error
= gfs2_glock_wait(gh
);
1061 * gfs2_glock_poll - poll to see if an async request has been completed
1064 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1067 int gfs2_glock_poll(struct gfs2_holder
*gh
)
1069 return test_bit(HIF_WAIT
, &gh
->gh_iflags
) ? 0 : 1;
1073 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1074 * @gh: the glock holder
1078 void gfs2_glock_dq(struct gfs2_holder
*gh
)
1080 struct gfs2_glock
*gl
= gh
->gh_gl
;
1081 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1085 spin_lock(&gl
->gl_spin
);
1086 if (gh
->gh_flags
& GL_NOCACHE
)
1087 handle_callback(gl
, LM_ST_UNLOCKED
, 0, false);
1089 list_del_init(&gh
->gh_list
);
1090 if (find_first_holder(gl
) == NULL
) {
1091 if (glops
->go_unlock
) {
1092 GLOCK_BUG_ON(gl
, test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
));
1093 spin_unlock(&gl
->gl_spin
);
1094 glops
->go_unlock(gh
);
1095 spin_lock(&gl
->gl_spin
);
1096 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
1098 if (list_empty(&gl
->gl_holders
) &&
1099 !test_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
) &&
1100 !test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
1103 if (!test_bit(GLF_LFLUSH
, &gl
->gl_flags
) && demote_ok(gl
))
1104 gfs2_glock_add_to_lru(gl
);
1106 trace_gfs2_glock_queue(gh
, 0);
1107 spin_unlock(&gl
->gl_spin
);
1108 if (likely(fast_path
))
1111 gfs2_glock_hold(gl
);
1112 if (test_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
) &&
1113 !test_bit(GLF_DEMOTE
, &gl
->gl_flags
) &&
1114 gl
->gl_name
.ln_type
== LM_TYPE_INODE
)
1115 delay
= gl
->gl_hold_time
;
1116 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, delay
) == 0)
1120 void gfs2_glock_dq_wait(struct gfs2_holder
*gh
)
1122 struct gfs2_glock
*gl
= gh
->gh_gl
;
1125 wait_on_bit(&gl
->gl_flags
, GLF_DEMOTE
, gfs2_glock_demote_wait
, TASK_UNINTERRUPTIBLE
);
1129 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1130 * @gh: the holder structure
1134 void gfs2_glock_dq_uninit(struct gfs2_holder
*gh
)
1137 gfs2_holder_uninit(gh
);
1141 * gfs2_glock_nq_num - acquire a glock based on lock number
1142 * @sdp: the filesystem
1143 * @number: the lock number
1144 * @glops: the glock operations for the type of glock
1145 * @state: the state to acquire the glock in
1146 * @flags: modifier flags for the acquisition
1147 * @gh: the struct gfs2_holder
1152 int gfs2_glock_nq_num(struct gfs2_sbd
*sdp
, u64 number
,
1153 const struct gfs2_glock_operations
*glops
,
1154 unsigned int state
, int flags
, struct gfs2_holder
*gh
)
1156 struct gfs2_glock
*gl
;
1159 error
= gfs2_glock_get(sdp
, number
, glops
, CREATE
, &gl
);
1161 error
= gfs2_glock_nq_init(gl
, state
, flags
, gh
);
1169 * glock_compare - Compare two struct gfs2_glock structures for sorting
1170 * @arg_a: the first structure
1171 * @arg_b: the second structure
1175 static int glock_compare(const void *arg_a
, const void *arg_b
)
1177 const struct gfs2_holder
*gh_a
= *(const struct gfs2_holder
**)arg_a
;
1178 const struct gfs2_holder
*gh_b
= *(const struct gfs2_holder
**)arg_b
;
1179 const struct lm_lockname
*a
= &gh_a
->gh_gl
->gl_name
;
1180 const struct lm_lockname
*b
= &gh_b
->gh_gl
->gl_name
;
1182 if (a
->ln_number
> b
->ln_number
)
1184 if (a
->ln_number
< b
->ln_number
)
1186 BUG_ON(gh_a
->gh_gl
->gl_ops
->go_type
== gh_b
->gh_gl
->gl_ops
->go_type
);
1191 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1192 * @num_gh: the number of structures
1193 * @ghs: an array of struct gfs2_holder structures
1195 * Returns: 0 on success (all glocks acquired),
1196 * errno on failure (no glocks acquired)
1199 static int nq_m_sync(unsigned int num_gh
, struct gfs2_holder
*ghs
,
1200 struct gfs2_holder
**p
)
1205 for (x
= 0; x
< num_gh
; x
++)
1208 sort(p
, num_gh
, sizeof(struct gfs2_holder
*), glock_compare
, NULL
);
1210 for (x
= 0; x
< num_gh
; x
++) {
1211 p
[x
]->gh_flags
&= ~(LM_FLAG_TRY
| GL_ASYNC
);
1213 error
= gfs2_glock_nq(p
[x
]);
1216 gfs2_glock_dq(p
[x
]);
1225 * gfs2_glock_nq_m - acquire multiple glocks
1226 * @num_gh: the number of structures
1227 * @ghs: an array of struct gfs2_holder structures
1230 * Returns: 0 on success (all glocks acquired),
1231 * errno on failure (no glocks acquired)
1234 int gfs2_glock_nq_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1236 struct gfs2_holder
*tmp
[4];
1237 struct gfs2_holder
**pph
= tmp
;
1244 ghs
->gh_flags
&= ~(LM_FLAG_TRY
| GL_ASYNC
);
1245 return gfs2_glock_nq(ghs
);
1249 pph
= kmalloc(num_gh
* sizeof(struct gfs2_holder
*), GFP_NOFS
);
1254 error
= nq_m_sync(num_gh
, ghs
, pph
);
1263 * gfs2_glock_dq_m - release multiple glocks
1264 * @num_gh: the number of structures
1265 * @ghs: an array of struct gfs2_holder structures
1269 void gfs2_glock_dq_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1272 gfs2_glock_dq(&ghs
[num_gh
]);
1275 void gfs2_glock_cb(struct gfs2_glock
*gl
, unsigned int state
)
1277 unsigned long delay
= 0;
1278 unsigned long holdtime
;
1279 unsigned long now
= jiffies
;
1281 gfs2_glock_hold(gl
);
1282 holdtime
= gl
->gl_tchange
+ gl
->gl_hold_time
;
1283 if (test_bit(GLF_QUEUED
, &gl
->gl_flags
) &&
1284 gl
->gl_name
.ln_type
== LM_TYPE_INODE
) {
1285 if (time_before(now
, holdtime
))
1286 delay
= holdtime
- now
;
1287 if (test_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
))
1288 delay
= gl
->gl_hold_time
;
1291 spin_lock(&gl
->gl_spin
);
1292 handle_callback(gl
, state
, delay
, true);
1293 spin_unlock(&gl
->gl_spin
);
1294 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, delay
) == 0)
1299 * gfs2_should_freeze - Figure out if glock should be frozen
1300 * @gl: The glock in question
1302 * Glocks are not frozen if (a) the result of the dlm operation is
1303 * an error, (b) the locking operation was an unlock operation or
1304 * (c) if there is a "noexp" flagged request anywhere in the queue
1306 * Returns: 1 if freezing should occur, 0 otherwise
1309 static int gfs2_should_freeze(const struct gfs2_glock
*gl
)
1311 const struct gfs2_holder
*gh
;
1313 if (gl
->gl_reply
& ~LM_OUT_ST_MASK
)
1315 if (gl
->gl_target
== LM_ST_UNLOCKED
)
1318 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
) {
1319 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
1321 if (LM_FLAG_NOEXP
& gh
->gh_flags
)
1329 * gfs2_glock_complete - Callback used by locking
1330 * @gl: Pointer to the glock
1331 * @ret: The return value from the dlm
1333 * The gl_reply field is under the gl_spin lock so that it is ok
1334 * to use a bitfield shared with other glock state fields.
1337 void gfs2_glock_complete(struct gfs2_glock
*gl
, int ret
)
1339 struct lm_lockstruct
*ls
= &gl
->gl_sbd
->sd_lockstruct
;
1341 spin_lock(&gl
->gl_spin
);
1344 if (unlikely(test_bit(DFL_BLOCK_LOCKS
, &ls
->ls_recover_flags
))) {
1345 if (gfs2_should_freeze(gl
)) {
1346 set_bit(GLF_FROZEN
, &gl
->gl_flags
);
1347 spin_unlock(&gl
->gl_spin
);
1352 gl
->gl_lockref
.count
++;
1353 set_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
);
1354 spin_unlock(&gl
->gl_spin
);
1356 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
1360 static int glock_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
1362 struct gfs2_glock
*gla
, *glb
;
1364 gla
= list_entry(a
, struct gfs2_glock
, gl_lru
);
1365 glb
= list_entry(b
, struct gfs2_glock
, gl_lru
);
1367 if (gla
->gl_name
.ln_number
> glb
->gl_name
.ln_number
)
1369 if (gla
->gl_name
.ln_number
< glb
->gl_name
.ln_number
)
1376 * gfs2_dispose_glock_lru - Demote a list of glocks
1377 * @list: The list to dispose of
1379 * Disposing of glocks may involve disk accesses, so that here we sort
1380 * the glocks by number (i.e. disk location of the inodes) so that if
1381 * there are any such accesses, they'll be sent in order (mostly).
1383 * Must be called under the lru_lock, but may drop and retake this
1384 * lock. While the lru_lock is dropped, entries may vanish from the
1385 * list, but no new entries will appear on the list (since it is
1389 static void gfs2_dispose_glock_lru(struct list_head
*list
)
1390 __releases(&lru_lock
)
1391 __acquires(&lru_lock
)
1393 struct gfs2_glock
*gl
;
1395 list_sort(NULL
, list
, glock_cmp
);
1397 while(!list_empty(list
)) {
1398 gl
= list_entry(list
->next
, struct gfs2_glock
, gl_lru
);
1399 list_del_init(&gl
->gl_lru
);
1400 if (!spin_trylock(&gl
->gl_spin
)) {
1401 list_add(&gl
->gl_lru
, &lru_list
);
1402 atomic_inc(&lru_count
);
1405 clear_bit(GLF_LRU
, &gl
->gl_flags
);
1406 spin_unlock(&lru_lock
);
1407 gl
->gl_lockref
.count
++;
1409 handle_callback(gl
, LM_ST_UNLOCKED
, 0, false);
1410 WARN_ON(!test_and_clear_bit(GLF_LOCK
, &gl
->gl_flags
));
1411 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
1412 gl
->gl_lockref
.count
--;
1413 spin_unlock(&gl
->gl_spin
);
1414 spin_lock(&lru_lock
);
1419 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1420 * @nr: The number of entries to scan
1422 * This function selects the entries on the LRU which are able to
1423 * be demoted, and then kicks off the process by calling
1424 * gfs2_dispose_glock_lru() above.
1427 static long gfs2_scan_glock_lru(int nr
)
1429 struct gfs2_glock
*gl
;
1434 spin_lock(&lru_lock
);
1435 while ((nr
-- >= 0) && !list_empty(&lru_list
)) {
1436 gl
= list_entry(lru_list
.next
, struct gfs2_glock
, gl_lru
);
1438 /* Test for being demotable */
1439 if (!test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
)) {
1440 list_move(&gl
->gl_lru
, &dispose
);
1441 atomic_dec(&lru_count
);
1446 list_move(&gl
->gl_lru
, &skipped
);
1448 list_splice(&skipped
, &lru_list
);
1449 if (!list_empty(&dispose
))
1450 gfs2_dispose_glock_lru(&dispose
);
1451 spin_unlock(&lru_lock
);
1456 static unsigned long gfs2_glock_shrink_scan(struct shrinker
*shrink
,
1457 struct shrink_control
*sc
)
1459 if (!(sc
->gfp_mask
& __GFP_FS
))
1461 return gfs2_scan_glock_lru(sc
->nr_to_scan
);
1464 static unsigned long gfs2_glock_shrink_count(struct shrinker
*shrink
,
1465 struct shrink_control
*sc
)
1467 return vfs_pressure_ratio(atomic_read(&lru_count
));
1470 static struct shrinker glock_shrinker
= {
1471 .seeks
= DEFAULT_SEEKS
,
1472 .count_objects
= gfs2_glock_shrink_count
,
1473 .scan_objects
= gfs2_glock_shrink_scan
,
1477 * examine_bucket - Call a function for glock in a hash bucket
1478 * @examiner: the function
1479 * @sdp: the filesystem
1480 * @bucket: the bucket
1484 static void examine_bucket(glock_examiner examiner
, const struct gfs2_sbd
*sdp
,
1487 struct gfs2_glock
*gl
;
1488 struct hlist_bl_head
*head
= &gl_hash_table
[hash
];
1489 struct hlist_bl_node
*pos
;
1492 hlist_bl_for_each_entry_rcu(gl
, pos
, head
, gl_list
) {
1493 if ((gl
->gl_sbd
== sdp
) && lockref_get_not_dead(&gl
->gl_lockref
))
1500 static void glock_hash_walk(glock_examiner examiner
, const struct gfs2_sbd
*sdp
)
1504 for (x
= 0; x
< GFS2_GL_HASH_SIZE
; x
++)
1505 examine_bucket(examiner
, sdp
, x
);
1510 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1511 * @gl: The glock to thaw
1515 static void thaw_glock(struct gfs2_glock
*gl
)
1517 if (!test_and_clear_bit(GLF_FROZEN
, &gl
->gl_flags
))
1519 set_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
);
1520 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0) {
1527 * clear_glock - look at a glock and see if we can free it from glock cache
1528 * @gl: the glock to look at
1532 static void clear_glock(struct gfs2_glock
*gl
)
1534 gfs2_glock_remove_from_lru(gl
);
1536 spin_lock(&gl
->gl_spin
);
1537 if (gl
->gl_state
!= LM_ST_UNLOCKED
)
1538 handle_callback(gl
, LM_ST_UNLOCKED
, 0, false);
1539 spin_unlock(&gl
->gl_spin
);
1540 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
1545 * gfs2_glock_thaw - Thaw any frozen glocks
1546 * @sdp: The super block
1550 void gfs2_glock_thaw(struct gfs2_sbd
*sdp
)
1552 glock_hash_walk(thaw_glock
, sdp
);
1555 static void dump_glock(struct seq_file
*seq
, struct gfs2_glock
*gl
)
1557 spin_lock(&gl
->gl_spin
);
1558 gfs2_dump_glock(seq
, gl
);
1559 spin_unlock(&gl
->gl_spin
);
1562 static void dump_glock_func(struct gfs2_glock
*gl
)
1564 dump_glock(NULL
, gl
);
1568 * gfs2_gl_hash_clear - Empty out the glock hash table
1569 * @sdp: the filesystem
1570 * @wait: wait until it's all gone
1572 * Called when unmounting the filesystem.
1575 void gfs2_gl_hash_clear(struct gfs2_sbd
*sdp
)
1577 set_bit(SDF_SKIP_DLM_UNLOCK
, &sdp
->sd_flags
);
1578 flush_workqueue(glock_workqueue
);
1579 glock_hash_walk(clear_glock
, sdp
);
1580 flush_workqueue(glock_workqueue
);
1581 wait_event(sdp
->sd_glock_wait
, atomic_read(&sdp
->sd_glock_disposal
) == 0);
1582 glock_hash_walk(dump_glock_func
, sdp
);
1585 void gfs2_glock_finish_truncate(struct gfs2_inode
*ip
)
1587 struct gfs2_glock
*gl
= ip
->i_gl
;
1590 ret
= gfs2_truncatei_resume(ip
);
1591 gfs2_assert_withdraw(gl
->gl_sbd
, ret
== 0);
1593 spin_lock(&gl
->gl_spin
);
1594 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
1596 spin_unlock(&gl
->gl_spin
);
1599 static const char *state2str(unsigned state
)
1602 case LM_ST_UNLOCKED
:
1606 case LM_ST_DEFERRED
:
1608 case LM_ST_EXCLUSIVE
:
1614 static const char *hflags2str(char *buf
, unsigned flags
, unsigned long iflags
)
1617 if (flags
& LM_FLAG_TRY
)
1619 if (flags
& LM_FLAG_TRY_1CB
)
1621 if (flags
& LM_FLAG_NOEXP
)
1623 if (flags
& LM_FLAG_ANY
)
1625 if (flags
& LM_FLAG_PRIORITY
)
1627 if (flags
& GL_ASYNC
)
1629 if (flags
& GL_EXACT
)
1631 if (flags
& GL_NOCACHE
)
1633 if (test_bit(HIF_HOLDER
, &iflags
))
1635 if (test_bit(HIF_WAIT
, &iflags
))
1637 if (test_bit(HIF_FIRST
, &iflags
))
1644 * dump_holder - print information about a glock holder
1645 * @seq: the seq_file struct
1646 * @gh: the glock holder
1650 static void dump_holder(struct seq_file
*seq
, const struct gfs2_holder
*gh
)
1652 struct task_struct
*gh_owner
= NULL
;
1656 if (gh
->gh_owner_pid
)
1657 gh_owner
= pid_task(gh
->gh_owner_pid
, PIDTYPE_PID
);
1658 gfs2_print_dbg(seq
, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1659 state2str(gh
->gh_state
),
1660 hflags2str(flags_buf
, gh
->gh_flags
, gh
->gh_iflags
),
1662 gh
->gh_owner_pid
? (long)pid_nr(gh
->gh_owner_pid
) : -1,
1663 gh_owner
? gh_owner
->comm
: "(ended)",
1668 static const char *gflags2str(char *buf
, const struct gfs2_glock
*gl
)
1670 const unsigned long *gflags
= &gl
->gl_flags
;
1673 if (test_bit(GLF_LOCK
, gflags
))
1675 if (test_bit(GLF_DEMOTE
, gflags
))
1677 if (test_bit(GLF_PENDING_DEMOTE
, gflags
))
1679 if (test_bit(GLF_DEMOTE_IN_PROGRESS
, gflags
))
1681 if (test_bit(GLF_DIRTY
, gflags
))
1683 if (test_bit(GLF_LFLUSH
, gflags
))
1685 if (test_bit(GLF_INVALIDATE_IN_PROGRESS
, gflags
))
1687 if (test_bit(GLF_REPLY_PENDING
, gflags
))
1689 if (test_bit(GLF_INITIAL
, gflags
))
1691 if (test_bit(GLF_FROZEN
, gflags
))
1693 if (test_bit(GLF_QUEUED
, gflags
))
1695 if (test_bit(GLF_LRU
, gflags
))
1699 if (test_bit(GLF_BLOCKING
, gflags
))
1706 * gfs2_dump_glock - print information about a glock
1707 * @seq: The seq_file struct
1710 * The file format is as follows:
1711 * One line per object, capital letters are used to indicate objects
1712 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1713 * other objects are indented by a single space and follow the glock to
1714 * which they are related. Fields are indicated by lower case letters
1715 * followed by a colon and the field value, except for strings which are in
1716 * [] so that its possible to see if they are composed of spaces for
1717 * example. The field's are n = number (id of the object), f = flags,
1718 * t = type, s = state, r = refcount, e = error, p = pid.
1722 void gfs2_dump_glock(struct seq_file
*seq
, const struct gfs2_glock
*gl
)
1724 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1725 unsigned long long dtime
;
1726 const struct gfs2_holder
*gh
;
1727 char gflags_buf
[32];
1729 dtime
= jiffies
- gl
->gl_demote_time
;
1730 dtime
*= 1000000/HZ
; /* demote time in uSec */
1731 if (!test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
1733 gfs2_print_dbg(seq
, "G: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d v:%d r:%d m:%ld\n",
1734 state2str(gl
->gl_state
),
1735 gl
->gl_name
.ln_type
,
1736 (unsigned long long)gl
->gl_name
.ln_number
,
1737 gflags2str(gflags_buf
, gl
),
1738 state2str(gl
->gl_target
),
1739 state2str(gl
->gl_demote_state
), dtime
,
1740 atomic_read(&gl
->gl_ail_count
),
1741 atomic_read(&gl
->gl_revokes
),
1742 (int)gl
->gl_lockref
.count
, gl
->gl_hold_time
);
1744 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
)
1745 dump_holder(seq
, gh
);
1747 if (gl
->gl_state
!= LM_ST_UNLOCKED
&& glops
->go_dump
)
1748 glops
->go_dump(seq
, gl
);
1751 static int gfs2_glstats_seq_show(struct seq_file
*seq
, void *iter_ptr
)
1753 struct gfs2_glock
*gl
= iter_ptr
;
1755 seq_printf(seq
, "G: n:%u/%llx rtt:%lld/%lld rttb:%lld/%lld irt:%lld/%lld dcnt: %lld qcnt: %lld\n",
1756 gl
->gl_name
.ln_type
,
1757 (unsigned long long)gl
->gl_name
.ln_number
,
1758 (long long)gl
->gl_stats
.stats
[GFS2_LKS_SRTT
],
1759 (long long)gl
->gl_stats
.stats
[GFS2_LKS_SRTTVAR
],
1760 (long long)gl
->gl_stats
.stats
[GFS2_LKS_SRTTB
],
1761 (long long)gl
->gl_stats
.stats
[GFS2_LKS_SRTTVARB
],
1762 (long long)gl
->gl_stats
.stats
[GFS2_LKS_SIRT
],
1763 (long long)gl
->gl_stats
.stats
[GFS2_LKS_SIRTVAR
],
1764 (long long)gl
->gl_stats
.stats
[GFS2_LKS_DCOUNT
],
1765 (long long)gl
->gl_stats
.stats
[GFS2_LKS_QCOUNT
]);
1769 static const char *gfs2_gltype
[] = {
1783 static const char *gfs2_stype
[] = {
1784 [GFS2_LKS_SRTT
] = "srtt",
1785 [GFS2_LKS_SRTTVAR
] = "srttvar",
1786 [GFS2_LKS_SRTTB
] = "srttb",
1787 [GFS2_LKS_SRTTVARB
] = "srttvarb",
1788 [GFS2_LKS_SIRT
] = "sirt",
1789 [GFS2_LKS_SIRTVAR
] = "sirtvar",
1790 [GFS2_LKS_DCOUNT
] = "dlm",
1791 [GFS2_LKS_QCOUNT
] = "queue",
1794 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1796 static int gfs2_sbstats_seq_show(struct seq_file
*seq
, void *iter_ptr
)
1798 struct gfs2_glock_iter
*gi
= seq
->private;
1799 struct gfs2_sbd
*sdp
= gi
->sdp
;
1800 unsigned index
= gi
->hash
>> 3;
1801 unsigned subindex
= gi
->hash
& 0x07;
1805 if (index
== 0 && subindex
!= 0)
1808 seq_printf(seq
, "%-10s %8s:", gfs2_gltype
[index
],
1809 (index
== 0) ? "cpu": gfs2_stype
[subindex
]);
1811 for_each_possible_cpu(i
) {
1812 const struct gfs2_pcpu_lkstats
*lkstats
= per_cpu_ptr(sdp
->sd_lkstats
, i
);
1816 value
= lkstats
->lkstats
[index
- 1].stats
[subindex
];
1818 seq_printf(seq
, " %15lld", (long long)value
);
1820 seq_putc(seq
, '\n');
1824 int __init
gfs2_glock_init(void)
1827 for(i
= 0; i
< GFS2_GL_HASH_SIZE
; i
++) {
1828 INIT_HLIST_BL_HEAD(&gl_hash_table
[i
]);
1831 glock_workqueue
= alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM
|
1832 WQ_HIGHPRI
| WQ_FREEZABLE
, 0);
1833 if (!glock_workqueue
)
1835 gfs2_delete_workqueue
= alloc_workqueue("delete_workqueue",
1836 WQ_MEM_RECLAIM
| WQ_FREEZABLE
,
1838 if (!gfs2_delete_workqueue
) {
1839 destroy_workqueue(glock_workqueue
);
1843 register_shrinker(&glock_shrinker
);
1848 void gfs2_glock_exit(void)
1850 unregister_shrinker(&glock_shrinker
);
1851 destroy_workqueue(glock_workqueue
);
1852 destroy_workqueue(gfs2_delete_workqueue
);
1855 static inline struct gfs2_glock
*glock_hash_chain(unsigned hash
)
1857 return hlist_bl_entry(hlist_bl_first_rcu(&gl_hash_table
[hash
]),
1858 struct gfs2_glock
, gl_list
);
1861 static inline struct gfs2_glock
*glock_hash_next(struct gfs2_glock
*gl
)
1863 return hlist_bl_entry(rcu_dereference(gl
->gl_list
.next
),
1864 struct gfs2_glock
, gl_list
);
1867 static int gfs2_glock_iter_next(struct gfs2_glock_iter
*gi
)
1869 struct gfs2_glock
*gl
;
1874 gi
->gl
= glock_hash_next(gl
);
1877 if (gi
->hash
>= GFS2_GL_HASH_SIZE
) {
1881 gi
->gl
= glock_hash_chain(gi
->hash
);
1884 while (gi
->gl
== NULL
) {
1886 if (gi
->hash
>= GFS2_GL_HASH_SIZE
) {
1890 gi
->gl
= glock_hash_chain(gi
->hash
);
1893 /* Skip entries for other sb and dead entries */
1894 } while (gi
->sdp
!= gi
->gl
->gl_sbd
||
1895 __lockref_is_dead(&gi
->gl
->gl_lockref
));
1900 static void *gfs2_glock_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1902 struct gfs2_glock_iter
*gi
= seq
->private;
1905 if (gi
->last_pos
<= *pos
)
1906 n
= gi
->nhash
+ (*pos
- gi
->last_pos
);
1914 if (gfs2_glock_iter_next(gi
))
1918 gi
->last_pos
= *pos
;
1922 static void *gfs2_glock_seq_next(struct seq_file
*seq
, void *iter_ptr
,
1925 struct gfs2_glock_iter
*gi
= seq
->private;
1928 gi
->last_pos
= *pos
;
1929 if (gfs2_glock_iter_next(gi
))
1935 static void gfs2_glock_seq_stop(struct seq_file
*seq
, void *iter_ptr
)
1937 struct gfs2_glock_iter
*gi
= seq
->private;
1944 static int gfs2_glock_seq_show(struct seq_file
*seq
, void *iter_ptr
)
1946 dump_glock(seq
, iter_ptr
);
1950 static void *gfs2_sbstats_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1952 struct gfs2_glock_iter
*gi
= seq
->private;
1955 if (*pos
>= GFS2_NR_SBSTATS
)
1958 return SEQ_START_TOKEN
;
1961 static void *gfs2_sbstats_seq_next(struct seq_file
*seq
, void *iter_ptr
,
1964 struct gfs2_glock_iter
*gi
= seq
->private;
1967 if (gi
->hash
>= GFS2_NR_SBSTATS
) {
1971 return SEQ_START_TOKEN
;
1974 static void gfs2_sbstats_seq_stop(struct seq_file
*seq
, void *iter_ptr
)
1979 static const struct seq_operations gfs2_glock_seq_ops
= {
1980 .start
= gfs2_glock_seq_start
,
1981 .next
= gfs2_glock_seq_next
,
1982 .stop
= gfs2_glock_seq_stop
,
1983 .show
= gfs2_glock_seq_show
,
1986 static const struct seq_operations gfs2_glstats_seq_ops
= {
1987 .start
= gfs2_glock_seq_start
,
1988 .next
= gfs2_glock_seq_next
,
1989 .stop
= gfs2_glock_seq_stop
,
1990 .show
= gfs2_glstats_seq_show
,
1993 static const struct seq_operations gfs2_sbstats_seq_ops
= {
1994 .start
= gfs2_sbstats_seq_start
,
1995 .next
= gfs2_sbstats_seq_next
,
1996 .stop
= gfs2_sbstats_seq_stop
,
1997 .show
= gfs2_sbstats_seq_show
,
2000 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2002 static int gfs2_glocks_open(struct inode
*inode
, struct file
*file
)
2004 int ret
= seq_open_private(file
, &gfs2_glock_seq_ops
,
2005 sizeof(struct gfs2_glock_iter
));
2007 struct seq_file
*seq
= file
->private_data
;
2008 struct gfs2_glock_iter
*gi
= seq
->private;
2009 gi
->sdp
= inode
->i_private
;
2010 seq
->buf
= kmalloc(GFS2_SEQ_GOODSIZE
, GFP_KERNEL
| __GFP_NOWARN
);
2012 seq
->size
= GFS2_SEQ_GOODSIZE
;
2017 static int gfs2_glstats_open(struct inode
*inode
, struct file
*file
)
2019 int ret
= seq_open_private(file
, &gfs2_glstats_seq_ops
,
2020 sizeof(struct gfs2_glock_iter
));
2022 struct seq_file
*seq
= file
->private_data
;
2023 struct gfs2_glock_iter
*gi
= seq
->private;
2024 gi
->sdp
= inode
->i_private
;
2025 seq
->buf
= kmalloc(GFS2_SEQ_GOODSIZE
, GFP_KERNEL
| __GFP_NOWARN
);
2027 seq
->size
= GFS2_SEQ_GOODSIZE
;
2032 static int gfs2_sbstats_open(struct inode
*inode
, struct file
*file
)
2034 int ret
= seq_open_private(file
, &gfs2_sbstats_seq_ops
,
2035 sizeof(struct gfs2_glock_iter
));
2037 struct seq_file
*seq
= file
->private_data
;
2038 struct gfs2_glock_iter
*gi
= seq
->private;
2039 gi
->sdp
= inode
->i_private
;
2044 static const struct file_operations gfs2_glocks_fops
= {
2045 .owner
= THIS_MODULE
,
2046 .open
= gfs2_glocks_open
,
2048 .llseek
= seq_lseek
,
2049 .release
= seq_release_private
,
2052 static const struct file_operations gfs2_glstats_fops
= {
2053 .owner
= THIS_MODULE
,
2054 .open
= gfs2_glstats_open
,
2056 .llseek
= seq_lseek
,
2057 .release
= seq_release_private
,
2060 static const struct file_operations gfs2_sbstats_fops
= {
2061 .owner
= THIS_MODULE
,
2062 .open
= gfs2_sbstats_open
,
2064 .llseek
= seq_lseek
,
2065 .release
= seq_release_private
,
2068 int gfs2_create_debugfs_file(struct gfs2_sbd
*sdp
)
2070 sdp
->debugfs_dir
= debugfs_create_dir(sdp
->sd_table_name
, gfs2_root
);
2071 if (!sdp
->debugfs_dir
)
2073 sdp
->debugfs_dentry_glocks
= debugfs_create_file("glocks",
2075 sdp
->debugfs_dir
, sdp
,
2077 if (!sdp
->debugfs_dentry_glocks
)
2080 sdp
->debugfs_dentry_glstats
= debugfs_create_file("glstats",
2082 sdp
->debugfs_dir
, sdp
,
2083 &gfs2_glstats_fops
);
2084 if (!sdp
->debugfs_dentry_glstats
)
2087 sdp
->debugfs_dentry_sbstats
= debugfs_create_file("sbstats",
2089 sdp
->debugfs_dir
, sdp
,
2090 &gfs2_sbstats_fops
);
2091 if (!sdp
->debugfs_dentry_sbstats
)
2096 gfs2_delete_debugfs_file(sdp
);
2100 void gfs2_delete_debugfs_file(struct gfs2_sbd
*sdp
)
2102 if (sdp
->debugfs_dir
) {
2103 if (sdp
->debugfs_dentry_glocks
) {
2104 debugfs_remove(sdp
->debugfs_dentry_glocks
);
2105 sdp
->debugfs_dentry_glocks
= NULL
;
2107 if (sdp
->debugfs_dentry_glstats
) {
2108 debugfs_remove(sdp
->debugfs_dentry_glstats
);
2109 sdp
->debugfs_dentry_glstats
= NULL
;
2111 if (sdp
->debugfs_dentry_sbstats
) {
2112 debugfs_remove(sdp
->debugfs_dentry_sbstats
);
2113 sdp
->debugfs_dentry_sbstats
= NULL
;
2115 debugfs_remove(sdp
->debugfs_dir
);
2116 sdp
->debugfs_dir
= NULL
;
2120 int gfs2_register_debugfs(void)
2122 gfs2_root
= debugfs_create_dir("gfs2", NULL
);
2123 return gfs2_root
? 0 : -ENOMEM
;
2126 void gfs2_unregister_debugfs(void)
2128 debugfs_remove(gfs2_root
);