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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 #include <linux/sched.h>
13 #include <linux/slab.h>
14 #include <linux/spinlock.h>
15 #include <linux/buffer_head.h>
16 #include <linux/delay.h>
17 #include <linux/sort.h>
18 #include <linux/jhash.h>
19 #include <linux/kallsyms.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <linux/list.h>
22 #include <linux/wait.h>
23 #include <linux/module.h>
24 #include <asm/uaccess.h>
25 #include <linux/seq_file.h>
26 #include <linux/debugfs.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/workqueue.h>
30 #include <linux/jiffies.h>
31 #include <linux/rcupdate.h>
32 #include <linux/rculist_bl.h>
33 #include <linux/bit_spinlock.h>
34 #include <linux/percpu.h>
35 #include <linux/list_sort.h>
36 #include <linux/lockref.h>
49 #define CREATE_TRACE_POINTS
50 #include "trace_gfs2.h"
52 struct gfs2_glock_iter
{
53 int hash
; /* hash bucket index */
54 unsigned nhash
; /* Index within current bucket */
55 struct gfs2_sbd
*sdp
; /* incore superblock */
56 struct gfs2_glock
*gl
; /* current glock struct */
57 loff_t last_pos
; /* last position */
60 typedef void (*glock_examiner
) (struct gfs2_glock
* gl
);
62 static void do_xmote(struct gfs2_glock
*gl
, struct gfs2_holder
*gh
, unsigned int target
);
64 static struct dentry
*gfs2_root
;
65 static struct workqueue_struct
*glock_workqueue
;
66 struct workqueue_struct
*gfs2_delete_workqueue
;
67 static LIST_HEAD(lru_list
);
68 static atomic_t lru_count
= ATOMIC_INIT(0);
69 static DEFINE_SPINLOCK(lru_lock
);
71 #define GFS2_GL_HASH_SHIFT 15
72 #define GFS2_GL_HASH_SIZE (1 << GFS2_GL_HASH_SHIFT)
73 #define GFS2_GL_HASH_MASK (GFS2_GL_HASH_SIZE - 1)
75 static struct hlist_bl_head gl_hash_table
[GFS2_GL_HASH_SIZE
];
76 static struct dentry
*gfs2_root
;
79 * gl_hash() - Turn glock number into hash bucket number
80 * @lock: The glock number
82 * Returns: The number of the corresponding hash bucket
85 static unsigned int gl_hash(const struct gfs2_sbd
*sdp
,
86 const struct lm_lockname
*name
)
90 h
= jhash(&name
->ln_number
, sizeof(u64
), 0);
91 h
= jhash(&name
->ln_type
, sizeof(unsigned int), h
);
92 h
= jhash(&sdp
, sizeof(struct gfs2_sbd
*), h
);
93 h
&= GFS2_GL_HASH_MASK
;
98 static inline void spin_lock_bucket(unsigned int hash
)
100 hlist_bl_lock(&gl_hash_table
[hash
]);
103 static inline void spin_unlock_bucket(unsigned int hash
)
105 hlist_bl_unlock(&gl_hash_table
[hash
]);
108 static void gfs2_glock_dealloc(struct rcu_head
*rcu
)
110 struct gfs2_glock
*gl
= container_of(rcu
, struct gfs2_glock
, gl_rcu
);
112 if (gl
->gl_ops
->go_flags
& GLOF_ASPACE
) {
113 kmem_cache_free(gfs2_glock_aspace_cachep
, gl
);
115 kfree(gl
->gl_lksb
.sb_lvbptr
);
116 kmem_cache_free(gfs2_glock_cachep
, gl
);
120 void gfs2_glock_free(struct gfs2_glock
*gl
)
122 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
124 call_rcu(&gl
->gl_rcu
, gfs2_glock_dealloc
);
125 if (atomic_dec_and_test(&sdp
->sd_glock_disposal
))
126 wake_up(&sdp
->sd_glock_wait
);
130 * gfs2_glock_hold() - increment reference count on glock
131 * @gl: The glock to hold
135 static void gfs2_glock_hold(struct gfs2_glock
*gl
)
137 GLOCK_BUG_ON(gl
, __lockref_is_dead(&gl
->gl_lockref
));
138 lockref_get(&gl
->gl_lockref
);
142 * demote_ok - Check to see if it's ok to unlock a glock
145 * Returns: 1 if it's ok
148 static int demote_ok(const struct gfs2_glock
*gl
)
150 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
152 if (gl
->gl_state
== LM_ST_UNLOCKED
)
154 if (!list_empty(&gl
->gl_holders
))
156 if (glops
->go_demote_ok
)
157 return glops
->go_demote_ok(gl
);
162 void gfs2_glock_add_to_lru(struct gfs2_glock
*gl
)
164 spin_lock(&lru_lock
);
166 if (!list_empty(&gl
->gl_lru
))
167 list_del_init(&gl
->gl_lru
);
169 atomic_inc(&lru_count
);
171 list_add_tail(&gl
->gl_lru
, &lru_list
);
172 set_bit(GLF_LRU
, &gl
->gl_flags
);
173 spin_unlock(&lru_lock
);
176 static void __gfs2_glock_remove_from_lru(struct gfs2_glock
*gl
)
178 if (!list_empty(&gl
->gl_lru
)) {
179 list_del_init(&gl
->gl_lru
);
180 atomic_dec(&lru_count
);
181 clear_bit(GLF_LRU
, &gl
->gl_flags
);
185 static void gfs2_glock_remove_from_lru(struct gfs2_glock
*gl
)
187 spin_lock(&lru_lock
);
188 __gfs2_glock_remove_from_lru(gl
);
189 spin_unlock(&lru_lock
);
193 * gfs2_glock_put() - Decrement reference count on glock
194 * @gl: The glock to put
198 void gfs2_glock_put(struct gfs2_glock
*gl
)
200 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
201 struct address_space
*mapping
= gfs2_glock2aspace(gl
);
203 if (lockref_put_or_lock(&gl
->gl_lockref
))
206 lockref_mark_dead(&gl
->gl_lockref
);
208 spin_lock(&lru_lock
);
209 __gfs2_glock_remove_from_lru(gl
);
210 spin_unlock(&lru_lock
);
211 spin_unlock(&gl
->gl_lockref
.lock
);
212 spin_lock_bucket(gl
->gl_hash
);
213 hlist_bl_del_rcu(&gl
->gl_list
);
214 spin_unlock_bucket(gl
->gl_hash
);
215 GLOCK_BUG_ON(gl
, !list_empty(&gl
->gl_holders
));
216 GLOCK_BUG_ON(gl
, mapping
&& mapping
->nrpages
);
217 trace_gfs2_glock_put(gl
);
218 sdp
->sd_lockstruct
.ls_ops
->lm_put_lock(gl
);
222 * search_bucket() - Find struct gfs2_glock by lock number
223 * @bucket: the bucket to search
224 * @name: The lock name
226 * Returns: NULL, or the struct gfs2_glock with the requested number
229 static struct gfs2_glock
*search_bucket(unsigned int hash
,
230 const struct gfs2_sbd
*sdp
,
231 const struct lm_lockname
*name
)
233 struct gfs2_glock
*gl
;
234 struct hlist_bl_node
*h
;
236 hlist_bl_for_each_entry_rcu(gl
, h
, &gl_hash_table
[hash
], gl_list
) {
237 if (!lm_name_equal(&gl
->gl_name
, name
))
239 if (gl
->gl_sbd
!= sdp
)
241 if (lockref_get_not_dead(&gl
->gl_lockref
))
249 * may_grant - check if its ok to grant a new lock
251 * @gh: The lock request which we wish to grant
253 * Returns: true if its ok to grant the lock
256 static inline int may_grant(const struct gfs2_glock
*gl
, const struct gfs2_holder
*gh
)
258 const struct gfs2_holder
*gh_head
= list_entry(gl
->gl_holders
.next
, const struct gfs2_holder
, gh_list
);
259 if ((gh
->gh_state
== LM_ST_EXCLUSIVE
||
260 gh_head
->gh_state
== LM_ST_EXCLUSIVE
) && gh
!= gh_head
)
262 if (gl
->gl_state
== gh
->gh_state
)
264 if (gh
->gh_flags
& GL_EXACT
)
266 if (gl
->gl_state
== LM_ST_EXCLUSIVE
) {
267 if (gh
->gh_state
== LM_ST_SHARED
&& gh_head
->gh_state
== LM_ST_SHARED
)
269 if (gh
->gh_state
== LM_ST_DEFERRED
&& gh_head
->gh_state
== LM_ST_DEFERRED
)
272 if (gl
->gl_state
!= LM_ST_UNLOCKED
&& (gh
->gh_flags
& LM_FLAG_ANY
))
277 static void gfs2_holder_wake(struct gfs2_holder
*gh
)
279 clear_bit(HIF_WAIT
, &gh
->gh_iflags
);
280 smp_mb__after_atomic();
281 wake_up_bit(&gh
->gh_iflags
, HIF_WAIT
);
285 * do_error - Something unexpected has happened during a lock request
289 static inline void do_error(struct gfs2_glock
*gl
, const int ret
)
291 struct gfs2_holder
*gh
, *tmp
;
293 list_for_each_entry_safe(gh
, tmp
, &gl
->gl_holders
, gh_list
) {
294 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
296 if (ret
& LM_OUT_ERROR
)
298 else if (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
))
299 gh
->gh_error
= GLR_TRYFAILED
;
302 list_del_init(&gh
->gh_list
);
303 trace_gfs2_glock_queue(gh
, 0);
304 gfs2_holder_wake(gh
);
309 * do_promote - promote as many requests as possible on the current queue
312 * Returns: 1 if there is a blocked holder at the head of the list, or 2
313 * if a type specific operation is underway.
316 static int do_promote(struct gfs2_glock
*gl
)
317 __releases(&gl
->gl_spin
)
318 __acquires(&gl
->gl_spin
)
320 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
321 struct gfs2_holder
*gh
, *tmp
;
325 list_for_each_entry_safe(gh
, tmp
, &gl
->gl_holders
, gh_list
) {
326 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
328 if (may_grant(gl
, gh
)) {
329 if (gh
->gh_list
.prev
== &gl
->gl_holders
&&
331 spin_unlock(&gl
->gl_spin
);
332 /* FIXME: eliminate this eventually */
333 ret
= glops
->go_lock(gh
);
334 spin_lock(&gl
->gl_spin
);
339 list_del_init(&gh
->gh_list
);
340 trace_gfs2_glock_queue(gh
, 0);
341 gfs2_holder_wake(gh
);
344 set_bit(HIF_HOLDER
, &gh
->gh_iflags
);
345 trace_gfs2_promote(gh
, 1);
346 gfs2_holder_wake(gh
);
349 set_bit(HIF_HOLDER
, &gh
->gh_iflags
);
350 trace_gfs2_promote(gh
, 0);
351 gfs2_holder_wake(gh
);
354 if (gh
->gh_list
.prev
== &gl
->gl_holders
)
363 * find_first_waiter - find the first gh that's waiting for the glock
367 static inline struct gfs2_holder
*find_first_waiter(const struct gfs2_glock
*gl
)
369 struct gfs2_holder
*gh
;
371 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
) {
372 if (!test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
379 * state_change - record that the glock is now in a different state
381 * @new_state the new state
385 static void state_change(struct gfs2_glock
*gl
, unsigned int new_state
)
389 held1
= (gl
->gl_state
!= LM_ST_UNLOCKED
);
390 held2
= (new_state
!= LM_ST_UNLOCKED
);
392 if (held1
!= held2
) {
393 GLOCK_BUG_ON(gl
, __lockref_is_dead(&gl
->gl_lockref
));
395 gl
->gl_lockref
.count
++;
397 gl
->gl_lockref
.count
--;
399 if (held1
&& held2
&& list_empty(&gl
->gl_holders
))
400 clear_bit(GLF_QUEUED
, &gl
->gl_flags
);
402 if (new_state
!= gl
->gl_target
)
403 /* shorten our minimum hold time */
404 gl
->gl_hold_time
= max(gl
->gl_hold_time
- GL_GLOCK_HOLD_DECR
,
406 gl
->gl_state
= new_state
;
407 gl
->gl_tchange
= jiffies
;
410 static void gfs2_demote_wake(struct gfs2_glock
*gl
)
412 gl
->gl_demote_state
= LM_ST_EXCLUSIVE
;
413 clear_bit(GLF_DEMOTE
, &gl
->gl_flags
);
414 smp_mb__after_atomic();
415 wake_up_bit(&gl
->gl_flags
, GLF_DEMOTE
);
419 * finish_xmote - The DLM has replied to one of our lock requests
421 * @ret: The status from the DLM
425 static void finish_xmote(struct gfs2_glock
*gl
, unsigned int ret
)
427 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
428 struct gfs2_holder
*gh
;
429 unsigned state
= ret
& LM_OUT_ST_MASK
;
432 spin_lock(&gl
->gl_spin
);
433 trace_gfs2_glock_state_change(gl
, state
);
434 state_change(gl
, state
);
435 gh
= find_first_waiter(gl
);
437 /* Demote to UN request arrived during demote to SH or DF */
438 if (test_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
) &&
439 state
!= LM_ST_UNLOCKED
&& gl
->gl_demote_state
== LM_ST_UNLOCKED
)
440 gl
->gl_target
= LM_ST_UNLOCKED
;
442 /* Check for state != intended state */
443 if (unlikely(state
!= gl
->gl_target
)) {
444 if (gh
&& !test_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
)) {
445 /* move to back of queue and try next entry */
446 if (ret
& LM_OUT_CANCELED
) {
447 if ((gh
->gh_flags
& LM_FLAG_PRIORITY
) == 0)
448 list_move_tail(&gh
->gh_list
, &gl
->gl_holders
);
449 gh
= find_first_waiter(gl
);
450 gl
->gl_target
= gh
->gh_state
;
453 /* Some error or failed "try lock" - report it */
454 if ((ret
& LM_OUT_ERROR
) ||
455 (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
))) {
456 gl
->gl_target
= gl
->gl_state
;
462 /* Unlocked due to conversion deadlock, try again */
465 do_xmote(gl
, gh
, gl
->gl_target
);
467 /* Conversion fails, unlock and try again */
470 do_xmote(gl
, gh
, LM_ST_UNLOCKED
);
472 default: /* Everything else */
473 pr_err("wanted %u got %u\n", gl
->gl_target
, state
);
476 spin_unlock(&gl
->gl_spin
);
480 /* Fast path - we got what we asked for */
481 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
))
482 gfs2_demote_wake(gl
);
483 if (state
!= LM_ST_UNLOCKED
) {
484 if (glops
->go_xmote_bh
) {
485 spin_unlock(&gl
->gl_spin
);
486 rv
= glops
->go_xmote_bh(gl
, gh
);
487 spin_lock(&gl
->gl_spin
);
498 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
500 spin_unlock(&gl
->gl_spin
);
504 * do_xmote - Calls the DLM to change the state of a lock
505 * @gl: The lock state
506 * @gh: The holder (only for promotes)
507 * @target: The target lock state
511 static void do_xmote(struct gfs2_glock
*gl
, struct gfs2_holder
*gh
, unsigned int target
)
512 __releases(&gl
->gl_spin
)
513 __acquires(&gl
->gl_spin
)
515 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
516 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
517 unsigned int lck_flags
= gh
? gh
->gh_flags
: 0;
520 lck_flags
&= (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
| LM_FLAG_NOEXP
|
522 GLOCK_BUG_ON(gl
, gl
->gl_state
== target
);
523 GLOCK_BUG_ON(gl
, gl
->gl_state
== gl
->gl_target
);
524 if ((target
== LM_ST_UNLOCKED
|| target
== LM_ST_DEFERRED
) &&
526 set_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
);
527 do_error(gl
, 0); /* Fail queued try locks */
530 set_bit(GLF_BLOCKING
, &gl
->gl_flags
);
531 if ((gl
->gl_req
== LM_ST_UNLOCKED
) ||
532 (gl
->gl_state
== LM_ST_EXCLUSIVE
) ||
533 (lck_flags
& (LM_FLAG_TRY
|LM_FLAG_TRY_1CB
)))
534 clear_bit(GLF_BLOCKING
, &gl
->gl_flags
);
535 spin_unlock(&gl
->gl_spin
);
538 if (test_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
))
539 glops
->go_inval(gl
, target
== LM_ST_DEFERRED
? 0 : DIO_METADATA
);
540 clear_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
);
543 if (sdp
->sd_lockstruct
.ls_ops
->lm_lock
) {
545 ret
= sdp
->sd_lockstruct
.ls_ops
->lm_lock(gl
, target
, lck_flags
);
547 pr_err("lm_lock ret %d\n", ret
);
550 } else { /* lock_nolock */
551 finish_xmote(gl
, target
);
552 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
556 spin_lock(&gl
->gl_spin
);
560 * find_first_holder - find the first "holder" gh
564 static inline struct gfs2_holder
*find_first_holder(const struct gfs2_glock
*gl
)
566 struct gfs2_holder
*gh
;
568 if (!list_empty(&gl
->gl_holders
)) {
569 gh
= list_entry(gl
->gl_holders
.next
, struct gfs2_holder
, gh_list
);
570 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
577 * run_queue - do all outstanding tasks related to a glock
578 * @gl: The glock in question
579 * @nonblock: True if we must not block in run_queue
583 static void run_queue(struct gfs2_glock
*gl
, const int nonblock
)
584 __releases(&gl
->gl_spin
)
585 __acquires(&gl
->gl_spin
)
587 struct gfs2_holder
*gh
= NULL
;
590 if (test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
))
593 GLOCK_BUG_ON(gl
, test_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
));
595 if (test_bit(GLF_DEMOTE
, &gl
->gl_flags
) &&
596 gl
->gl_demote_state
!= gl
->gl_state
) {
597 if (find_first_holder(gl
))
601 set_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
);
602 GLOCK_BUG_ON(gl
, gl
->gl_demote_state
== LM_ST_EXCLUSIVE
);
603 gl
->gl_target
= gl
->gl_demote_state
;
605 if (test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
606 gfs2_demote_wake(gl
);
607 ret
= do_promote(gl
);
612 gh
= find_first_waiter(gl
);
613 gl
->gl_target
= gh
->gh_state
;
614 if (!(gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
)))
615 do_error(gl
, 0); /* Fail queued try locks */
617 do_xmote(gl
, gh
, gl
->gl_target
);
622 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
623 smp_mb__after_atomic();
624 gl
->gl_lockref
.count
++;
625 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
626 gl
->gl_lockref
.count
--;
630 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
631 smp_mb__after_atomic();
635 static void delete_work_func(struct work_struct
*work
)
637 struct gfs2_glock
*gl
= container_of(work
, struct gfs2_glock
, gl_delete
);
638 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
639 struct gfs2_inode
*ip
;
641 u64 no_addr
= gl
->gl_name
.ln_number
;
644 /* Note: Unsafe to dereference ip as we don't hold right refs/locks */
647 inode
= gfs2_ilookup(sdp
->sd_vfs
, no_addr
, 1);
649 inode
= gfs2_lookup_by_inum(sdp
, no_addr
, NULL
, GFS2_BLKST_UNLINKED
);
650 if (inode
&& !IS_ERR(inode
)) {
651 d_prune_aliases(inode
);
657 static void glock_work_func(struct work_struct
*work
)
659 unsigned long delay
= 0;
660 struct gfs2_glock
*gl
= container_of(work
, struct gfs2_glock
, gl_work
.work
);
663 if (test_and_clear_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
)) {
664 finish_xmote(gl
, gl
->gl_reply
);
667 spin_lock(&gl
->gl_spin
);
668 if (test_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
) &&
669 gl
->gl_state
!= LM_ST_UNLOCKED
&&
670 gl
->gl_demote_state
!= LM_ST_EXCLUSIVE
) {
671 unsigned long holdtime
, now
= jiffies
;
673 holdtime
= gl
->gl_tchange
+ gl
->gl_hold_time
;
674 if (time_before(now
, holdtime
))
675 delay
= holdtime
- now
;
678 clear_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
);
679 set_bit(GLF_DEMOTE
, &gl
->gl_flags
);
683 spin_unlock(&gl
->gl_spin
);
687 if (gl
->gl_name
.ln_type
!= LM_TYPE_INODE
)
689 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, delay
) == 0)
697 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
698 * @sdp: The GFS2 superblock
699 * @number: the lock number
700 * @glops: The glock_operations to use
701 * @create: If 0, don't create the glock if it doesn't exist
702 * @glp: the glock is returned here
704 * This does not lock a glock, just finds/creates structures for one.
709 int gfs2_glock_get(struct gfs2_sbd
*sdp
, u64 number
,
710 const struct gfs2_glock_operations
*glops
, int create
,
711 struct gfs2_glock
**glp
)
713 struct super_block
*s
= sdp
->sd_vfs
;
714 struct lm_lockname name
= { .ln_number
= number
, .ln_type
= glops
->go_type
};
715 struct gfs2_glock
*gl
, *tmp
;
716 unsigned int hash
= gl_hash(sdp
, &name
);
717 struct address_space
*mapping
;
718 struct kmem_cache
*cachep
;
721 gl
= search_bucket(hash
, sdp
, &name
);
730 if (glops
->go_flags
& GLOF_ASPACE
)
731 cachep
= gfs2_glock_aspace_cachep
;
733 cachep
= gfs2_glock_cachep
;
734 gl
= kmem_cache_alloc(cachep
, GFP_NOFS
);
738 memset(&gl
->gl_lksb
, 0, sizeof(struct dlm_lksb
));
740 if (glops
->go_flags
& GLOF_LVB
) {
741 gl
->gl_lksb
.sb_lvbptr
= kzalloc(GFS2_MIN_LVB_SIZE
, GFP_NOFS
);
742 if (!gl
->gl_lksb
.sb_lvbptr
) {
743 kmem_cache_free(cachep
, gl
);
748 atomic_inc(&sdp
->sd_glock_disposal
);
752 gl
->gl_lockref
.count
= 1;
753 gl
->gl_state
= LM_ST_UNLOCKED
;
754 gl
->gl_target
= LM_ST_UNLOCKED
;
755 gl
->gl_demote_state
= LM_ST_EXCLUSIVE
;
758 gl
->gl_dstamp
= ktime_set(0, 0);
760 /* We use the global stats to estimate the initial per-glock stats */
761 gl
->gl_stats
= this_cpu_ptr(sdp
->sd_lkstats
)->lkstats
[glops
->go_type
];
763 gl
->gl_stats
.stats
[GFS2_LKS_DCOUNT
] = 0;
764 gl
->gl_stats
.stats
[GFS2_LKS_QCOUNT
] = 0;
765 gl
->gl_tchange
= jiffies
;
766 gl
->gl_object
= NULL
;
767 gl
->gl_hold_time
= GL_GLOCK_DFT_HOLD
;
768 INIT_DELAYED_WORK(&gl
->gl_work
, glock_work_func
);
769 INIT_WORK(&gl
->gl_delete
, delete_work_func
);
771 mapping
= gfs2_glock2aspace(gl
);
773 mapping
->a_ops
= &gfs2_meta_aops
;
774 mapping
->host
= s
->s_bdev
->bd_inode
;
776 mapping_set_gfp_mask(mapping
, GFP_NOFS
);
777 mapping
->private_data
= NULL
;
778 mapping
->backing_dev_info
= s
->s_bdi
;
779 mapping
->writeback_index
= 0;
782 spin_lock_bucket(hash
);
783 tmp
= search_bucket(hash
, sdp
, &name
);
785 spin_unlock_bucket(hash
);
786 kfree(gl
->gl_lksb
.sb_lvbptr
);
787 kmem_cache_free(cachep
, gl
);
788 atomic_dec(&sdp
->sd_glock_disposal
);
791 hlist_bl_add_head_rcu(&gl
->gl_list
, &gl_hash_table
[hash
]);
792 spin_unlock_bucket(hash
);
801 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
803 * @state: the state we're requesting
804 * @flags: the modifier flags
805 * @gh: the holder structure
809 void gfs2_holder_init(struct gfs2_glock
*gl
, unsigned int state
, unsigned flags
,
810 struct gfs2_holder
*gh
)
812 INIT_LIST_HEAD(&gh
->gh_list
);
814 gh
->gh_ip
= (unsigned long)__builtin_return_address(0);
815 gh
->gh_owner_pid
= get_pid(task_pid(current
));
816 gh
->gh_state
= state
;
817 gh
->gh_flags
= flags
;
824 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
825 * @state: the state we're requesting
826 * @flags: the modifier flags
827 * @gh: the holder structure
829 * Don't mess with the glock.
833 void gfs2_holder_reinit(unsigned int state
, unsigned flags
, struct gfs2_holder
*gh
)
835 gh
->gh_state
= state
;
836 gh
->gh_flags
= flags
;
838 gh
->gh_ip
= (unsigned long)__builtin_return_address(0);
839 if (gh
->gh_owner_pid
)
840 put_pid(gh
->gh_owner_pid
);
841 gh
->gh_owner_pid
= get_pid(task_pid(current
));
845 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
846 * @gh: the holder structure
850 void gfs2_holder_uninit(struct gfs2_holder
*gh
)
852 put_pid(gh
->gh_owner_pid
);
853 gfs2_glock_put(gh
->gh_gl
);
859 * gfs2_glock_wait - wait on a glock acquisition
860 * @gh: the glock holder
862 * Returns: 0 on success
865 int gfs2_glock_wait(struct gfs2_holder
*gh
)
867 unsigned long time1
= jiffies
;
870 wait_on_bit(&gh
->gh_iflags
, HIF_WAIT
, TASK_UNINTERRUPTIBLE
);
871 if (time_after(jiffies
, time1
+ HZ
)) /* have we waited > a second? */
872 /* Lengthen the minimum hold time. */
873 gh
->gh_gl
->gl_hold_time
= min(gh
->gh_gl
->gl_hold_time
+
880 * handle_callback - process a demote request
882 * @state: the state the caller wants us to change to
884 * There are only two requests that we are going to see in actual
885 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
888 static void handle_callback(struct gfs2_glock
*gl
, unsigned int state
,
889 unsigned long delay
, bool remote
)
891 int bit
= delay
? GLF_PENDING_DEMOTE
: GLF_DEMOTE
;
893 set_bit(bit
, &gl
->gl_flags
);
894 if (gl
->gl_demote_state
== LM_ST_EXCLUSIVE
) {
895 gl
->gl_demote_state
= state
;
896 gl
->gl_demote_time
= jiffies
;
897 } else if (gl
->gl_demote_state
!= LM_ST_UNLOCKED
&&
898 gl
->gl_demote_state
!= state
) {
899 gl
->gl_demote_state
= LM_ST_UNLOCKED
;
901 if (gl
->gl_ops
->go_callback
)
902 gl
->gl_ops
->go_callback(gl
, remote
);
903 trace_gfs2_demote_rq(gl
, remote
);
906 void gfs2_print_dbg(struct seq_file
*seq
, const char *fmt
, ...)
908 struct va_format vaf
;
914 seq_vprintf(seq
, fmt
, args
);
926 * add_to_queue - Add a holder to the wait queue (but look for recursion)
927 * @gh: the holder structure to add
929 * Eventually we should move the recursive locking trap to a
930 * debugging option or something like that. This is the fast
931 * path and needs to have the minimum number of distractions.
935 static inline void add_to_queue(struct gfs2_holder
*gh
)
936 __releases(&gl
->gl_spin
)
937 __acquires(&gl
->gl_spin
)
939 struct gfs2_glock
*gl
= gh
->gh_gl
;
940 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
941 struct list_head
*insert_pt
= NULL
;
942 struct gfs2_holder
*gh2
;
945 BUG_ON(gh
->gh_owner_pid
== NULL
);
946 if (test_and_set_bit(HIF_WAIT
, &gh
->gh_iflags
))
949 if (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
)) {
950 if (test_bit(GLF_LOCK
, &gl
->gl_flags
))
951 try_futile
= !may_grant(gl
, gh
);
952 if (test_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
))
956 list_for_each_entry(gh2
, &gl
->gl_holders
, gh_list
) {
957 if (unlikely(gh2
->gh_owner_pid
== gh
->gh_owner_pid
&&
958 (gh
->gh_gl
->gl_ops
->go_type
!= LM_TYPE_FLOCK
)))
961 !(gh2
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
))) {
963 gh
->gh_error
= GLR_TRYFAILED
;
964 gfs2_holder_wake(gh
);
967 if (test_bit(HIF_HOLDER
, &gh2
->gh_iflags
))
969 if (unlikely((gh
->gh_flags
& LM_FLAG_PRIORITY
) && !insert_pt
))
970 insert_pt
= &gh2
->gh_list
;
972 set_bit(GLF_QUEUED
, &gl
->gl_flags
);
973 trace_gfs2_glock_queue(gh
, 1);
974 gfs2_glstats_inc(gl
, GFS2_LKS_QCOUNT
);
975 gfs2_sbstats_inc(gl
, GFS2_LKS_QCOUNT
);
976 if (likely(insert_pt
== NULL
)) {
977 list_add_tail(&gh
->gh_list
, &gl
->gl_holders
);
978 if (unlikely(gh
->gh_flags
& LM_FLAG_PRIORITY
))
982 list_add_tail(&gh
->gh_list
, insert_pt
);
984 gh
= list_entry(gl
->gl_holders
.next
, struct gfs2_holder
, gh_list
);
985 if (!(gh
->gh_flags
& LM_FLAG_PRIORITY
)) {
986 spin_unlock(&gl
->gl_spin
);
987 if (sdp
->sd_lockstruct
.ls_ops
->lm_cancel
)
988 sdp
->sd_lockstruct
.ls_ops
->lm_cancel(gl
);
989 spin_lock(&gl
->gl_spin
);
994 pr_err("original: %pSR\n", (void *)gh2
->gh_ip
);
995 pr_err("pid: %d\n", pid_nr(gh2
->gh_owner_pid
));
996 pr_err("lock type: %d req lock state : %d\n",
997 gh2
->gh_gl
->gl_name
.ln_type
, gh2
->gh_state
);
998 pr_err("new: %pSR\n", (void *)gh
->gh_ip
);
999 pr_err("pid: %d\n", pid_nr(gh
->gh_owner_pid
));
1000 pr_err("lock type: %d req lock state : %d\n",
1001 gh
->gh_gl
->gl_name
.ln_type
, gh
->gh_state
);
1002 gfs2_dump_glock(NULL
, gl
);
1007 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1008 * @gh: the holder structure
1010 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1012 * Returns: 0, GLR_TRYFAILED, or errno on failure
1015 int gfs2_glock_nq(struct gfs2_holder
*gh
)
1017 struct gfs2_glock
*gl
= gh
->gh_gl
;
1018 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
1021 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)))
1024 if (test_bit(GLF_LRU
, &gl
->gl_flags
))
1025 gfs2_glock_remove_from_lru(gl
);
1027 spin_lock(&gl
->gl_spin
);
1029 if (unlikely((LM_FLAG_NOEXP
& gh
->gh_flags
) &&
1030 test_and_clear_bit(GLF_FROZEN
, &gl
->gl_flags
))) {
1031 set_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
);
1032 gl
->gl_lockref
.count
++;
1033 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
1034 gl
->gl_lockref
.count
--;
1037 spin_unlock(&gl
->gl_spin
);
1039 if (!(gh
->gh_flags
& GL_ASYNC
))
1040 error
= gfs2_glock_wait(gh
);
1046 * gfs2_glock_poll - poll to see if an async request has been completed
1049 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1052 int gfs2_glock_poll(struct gfs2_holder
*gh
)
1054 return test_bit(HIF_WAIT
, &gh
->gh_iflags
) ? 0 : 1;
1058 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1059 * @gh: the glock holder
1063 void gfs2_glock_dq(struct gfs2_holder
*gh
)
1065 struct gfs2_glock
*gl
= gh
->gh_gl
;
1066 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1070 spin_lock(&gl
->gl_spin
);
1071 if (gh
->gh_flags
& GL_NOCACHE
)
1072 handle_callback(gl
, LM_ST_UNLOCKED
, 0, false);
1074 list_del_init(&gh
->gh_list
);
1075 if (find_first_holder(gl
) == NULL
) {
1076 if (glops
->go_unlock
) {
1077 GLOCK_BUG_ON(gl
, test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
));
1078 spin_unlock(&gl
->gl_spin
);
1079 glops
->go_unlock(gh
);
1080 spin_lock(&gl
->gl_spin
);
1081 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
1083 if (list_empty(&gl
->gl_holders
) &&
1084 !test_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
) &&
1085 !test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
1088 if (!test_bit(GLF_LFLUSH
, &gl
->gl_flags
) && demote_ok(gl
))
1089 gfs2_glock_add_to_lru(gl
);
1091 trace_gfs2_glock_queue(gh
, 0);
1092 spin_unlock(&gl
->gl_spin
);
1093 if (likely(fast_path
))
1096 gfs2_glock_hold(gl
);
1097 if (test_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
) &&
1098 !test_bit(GLF_DEMOTE
, &gl
->gl_flags
) &&
1099 gl
->gl_name
.ln_type
== LM_TYPE_INODE
)
1100 delay
= gl
->gl_hold_time
;
1101 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, delay
) == 0)
1105 void gfs2_glock_dq_wait(struct gfs2_holder
*gh
)
1107 struct gfs2_glock
*gl
= gh
->gh_gl
;
1110 wait_on_bit(&gl
->gl_flags
, GLF_DEMOTE
, TASK_UNINTERRUPTIBLE
);
1114 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1115 * @gh: the holder structure
1119 void gfs2_glock_dq_uninit(struct gfs2_holder
*gh
)
1122 gfs2_holder_uninit(gh
);
1126 * gfs2_glock_nq_num - acquire a glock based on lock number
1127 * @sdp: the filesystem
1128 * @number: the lock number
1129 * @glops: the glock operations for the type of glock
1130 * @state: the state to acquire the glock in
1131 * @flags: modifier flags for the acquisition
1132 * @gh: the struct gfs2_holder
1137 int gfs2_glock_nq_num(struct gfs2_sbd
*sdp
, u64 number
,
1138 const struct gfs2_glock_operations
*glops
,
1139 unsigned int state
, int flags
, struct gfs2_holder
*gh
)
1141 struct gfs2_glock
*gl
;
1144 error
= gfs2_glock_get(sdp
, number
, glops
, CREATE
, &gl
);
1146 error
= gfs2_glock_nq_init(gl
, state
, flags
, gh
);
1154 * glock_compare - Compare two struct gfs2_glock structures for sorting
1155 * @arg_a: the first structure
1156 * @arg_b: the second structure
1160 static int glock_compare(const void *arg_a
, const void *arg_b
)
1162 const struct gfs2_holder
*gh_a
= *(const struct gfs2_holder
**)arg_a
;
1163 const struct gfs2_holder
*gh_b
= *(const struct gfs2_holder
**)arg_b
;
1164 const struct lm_lockname
*a
= &gh_a
->gh_gl
->gl_name
;
1165 const struct lm_lockname
*b
= &gh_b
->gh_gl
->gl_name
;
1167 if (a
->ln_number
> b
->ln_number
)
1169 if (a
->ln_number
< b
->ln_number
)
1171 BUG_ON(gh_a
->gh_gl
->gl_ops
->go_type
== gh_b
->gh_gl
->gl_ops
->go_type
);
1176 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1177 * @num_gh: the number of structures
1178 * @ghs: an array of struct gfs2_holder structures
1180 * Returns: 0 on success (all glocks acquired),
1181 * errno on failure (no glocks acquired)
1184 static int nq_m_sync(unsigned int num_gh
, struct gfs2_holder
*ghs
,
1185 struct gfs2_holder
**p
)
1190 for (x
= 0; x
< num_gh
; x
++)
1193 sort(p
, num_gh
, sizeof(struct gfs2_holder
*), glock_compare
, NULL
);
1195 for (x
= 0; x
< num_gh
; x
++) {
1196 p
[x
]->gh_flags
&= ~(LM_FLAG_TRY
| GL_ASYNC
);
1198 error
= gfs2_glock_nq(p
[x
]);
1201 gfs2_glock_dq(p
[x
]);
1210 * gfs2_glock_nq_m - acquire multiple glocks
1211 * @num_gh: the number of structures
1212 * @ghs: an array of struct gfs2_holder structures
1215 * Returns: 0 on success (all glocks acquired),
1216 * errno on failure (no glocks acquired)
1219 int gfs2_glock_nq_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1221 struct gfs2_holder
*tmp
[4];
1222 struct gfs2_holder
**pph
= tmp
;
1229 ghs
->gh_flags
&= ~(LM_FLAG_TRY
| GL_ASYNC
);
1230 return gfs2_glock_nq(ghs
);
1234 pph
= kmalloc(num_gh
* sizeof(struct gfs2_holder
*), GFP_NOFS
);
1239 error
= nq_m_sync(num_gh
, ghs
, pph
);
1248 * gfs2_glock_dq_m - release multiple glocks
1249 * @num_gh: the number of structures
1250 * @ghs: an array of struct gfs2_holder structures
1254 void gfs2_glock_dq_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1257 gfs2_glock_dq(&ghs
[num_gh
]);
1260 void gfs2_glock_cb(struct gfs2_glock
*gl
, unsigned int state
)
1262 unsigned long delay
= 0;
1263 unsigned long holdtime
;
1264 unsigned long now
= jiffies
;
1266 gfs2_glock_hold(gl
);
1267 holdtime
= gl
->gl_tchange
+ gl
->gl_hold_time
;
1268 if (test_bit(GLF_QUEUED
, &gl
->gl_flags
) &&
1269 gl
->gl_name
.ln_type
== LM_TYPE_INODE
) {
1270 if (time_before(now
, holdtime
))
1271 delay
= holdtime
- now
;
1272 if (test_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
))
1273 delay
= gl
->gl_hold_time
;
1276 spin_lock(&gl
->gl_spin
);
1277 handle_callback(gl
, state
, delay
, true);
1278 spin_unlock(&gl
->gl_spin
);
1279 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, delay
) == 0)
1284 * gfs2_should_freeze - Figure out if glock should be frozen
1285 * @gl: The glock in question
1287 * Glocks are not frozen if (a) the result of the dlm operation is
1288 * an error, (b) the locking operation was an unlock operation or
1289 * (c) if there is a "noexp" flagged request anywhere in the queue
1291 * Returns: 1 if freezing should occur, 0 otherwise
1294 static int gfs2_should_freeze(const struct gfs2_glock
*gl
)
1296 const struct gfs2_holder
*gh
;
1298 if (gl
->gl_reply
& ~LM_OUT_ST_MASK
)
1300 if (gl
->gl_target
== LM_ST_UNLOCKED
)
1303 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
) {
1304 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
1306 if (LM_FLAG_NOEXP
& gh
->gh_flags
)
1314 * gfs2_glock_complete - Callback used by locking
1315 * @gl: Pointer to the glock
1316 * @ret: The return value from the dlm
1318 * The gl_reply field is under the gl_spin lock so that it is ok
1319 * to use a bitfield shared with other glock state fields.
1322 void gfs2_glock_complete(struct gfs2_glock
*gl
, int ret
)
1324 struct lm_lockstruct
*ls
= &gl
->gl_sbd
->sd_lockstruct
;
1326 spin_lock(&gl
->gl_spin
);
1329 if (unlikely(test_bit(DFL_BLOCK_LOCKS
, &ls
->ls_recover_flags
))) {
1330 if (gfs2_should_freeze(gl
)) {
1331 set_bit(GLF_FROZEN
, &gl
->gl_flags
);
1332 spin_unlock(&gl
->gl_spin
);
1337 gl
->gl_lockref
.count
++;
1338 set_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
);
1339 spin_unlock(&gl
->gl_spin
);
1341 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
1345 static int glock_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
1347 struct gfs2_glock
*gla
, *glb
;
1349 gla
= list_entry(a
, struct gfs2_glock
, gl_lru
);
1350 glb
= list_entry(b
, struct gfs2_glock
, gl_lru
);
1352 if (gla
->gl_name
.ln_number
> glb
->gl_name
.ln_number
)
1354 if (gla
->gl_name
.ln_number
< glb
->gl_name
.ln_number
)
1361 * gfs2_dispose_glock_lru - Demote a list of glocks
1362 * @list: The list to dispose of
1364 * Disposing of glocks may involve disk accesses, so that here we sort
1365 * the glocks by number (i.e. disk location of the inodes) so that if
1366 * there are any such accesses, they'll be sent in order (mostly).
1368 * Must be called under the lru_lock, but may drop and retake this
1369 * lock. While the lru_lock is dropped, entries may vanish from the
1370 * list, but no new entries will appear on the list (since it is
1374 static void gfs2_dispose_glock_lru(struct list_head
*list
)
1375 __releases(&lru_lock
)
1376 __acquires(&lru_lock
)
1378 struct gfs2_glock
*gl
;
1380 list_sort(NULL
, list
, glock_cmp
);
1382 while(!list_empty(list
)) {
1383 gl
= list_entry(list
->next
, struct gfs2_glock
, gl_lru
);
1384 list_del_init(&gl
->gl_lru
);
1385 if (!spin_trylock(&gl
->gl_spin
)) {
1387 list_add(&gl
->gl_lru
, &lru_list
);
1388 atomic_inc(&lru_count
);
1391 if (test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
)) {
1392 spin_unlock(&gl
->gl_spin
);
1393 goto add_back_to_lru
;
1395 clear_bit(GLF_LRU
, &gl
->gl_flags
);
1396 gl
->gl_lockref
.count
++;
1398 handle_callback(gl
, LM_ST_UNLOCKED
, 0, false);
1399 WARN_ON(!test_and_clear_bit(GLF_LOCK
, &gl
->gl_flags
));
1400 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
1401 gl
->gl_lockref
.count
--;
1402 spin_unlock(&gl
->gl_spin
);
1403 cond_resched_lock(&lru_lock
);
1408 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1409 * @nr: The number of entries to scan
1411 * This function selects the entries on the LRU which are able to
1412 * be demoted, and then kicks off the process by calling
1413 * gfs2_dispose_glock_lru() above.
1416 static long gfs2_scan_glock_lru(int nr
)
1418 struct gfs2_glock
*gl
;
1423 spin_lock(&lru_lock
);
1424 while ((nr
-- >= 0) && !list_empty(&lru_list
)) {
1425 gl
= list_entry(lru_list
.next
, struct gfs2_glock
, gl_lru
);
1427 /* Test for being demotable */
1428 if (!test_bit(GLF_LOCK
, &gl
->gl_flags
)) {
1429 list_move(&gl
->gl_lru
, &dispose
);
1430 atomic_dec(&lru_count
);
1435 list_move(&gl
->gl_lru
, &skipped
);
1437 list_splice(&skipped
, &lru_list
);
1438 if (!list_empty(&dispose
))
1439 gfs2_dispose_glock_lru(&dispose
);
1440 spin_unlock(&lru_lock
);
1445 static unsigned long gfs2_glock_shrink_scan(struct shrinker
*shrink
,
1446 struct shrink_control
*sc
)
1448 if (!(sc
->gfp_mask
& __GFP_FS
))
1450 return gfs2_scan_glock_lru(sc
->nr_to_scan
);
1453 static unsigned long gfs2_glock_shrink_count(struct shrinker
*shrink
,
1454 struct shrink_control
*sc
)
1456 return vfs_pressure_ratio(atomic_read(&lru_count
));
1459 static struct shrinker glock_shrinker
= {
1460 .seeks
= DEFAULT_SEEKS
,
1461 .count_objects
= gfs2_glock_shrink_count
,
1462 .scan_objects
= gfs2_glock_shrink_scan
,
1466 * examine_bucket - Call a function for glock in a hash bucket
1467 * @examiner: the function
1468 * @sdp: the filesystem
1469 * @bucket: the bucket
1473 static void examine_bucket(glock_examiner examiner
, const struct gfs2_sbd
*sdp
,
1476 struct gfs2_glock
*gl
;
1477 struct hlist_bl_head
*head
= &gl_hash_table
[hash
];
1478 struct hlist_bl_node
*pos
;
1481 hlist_bl_for_each_entry_rcu(gl
, pos
, head
, gl_list
) {
1482 if ((gl
->gl_sbd
== sdp
) && lockref_get_not_dead(&gl
->gl_lockref
))
1489 static void glock_hash_walk(glock_examiner examiner
, const struct gfs2_sbd
*sdp
)
1493 for (x
= 0; x
< GFS2_GL_HASH_SIZE
; x
++)
1494 examine_bucket(examiner
, sdp
, x
);
1499 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1500 * @gl: The glock to thaw
1504 static void thaw_glock(struct gfs2_glock
*gl
)
1506 if (!test_and_clear_bit(GLF_FROZEN
, &gl
->gl_flags
))
1508 set_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
);
1509 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0) {
1516 * clear_glock - look at a glock and see if we can free it from glock cache
1517 * @gl: the glock to look at
1521 static void clear_glock(struct gfs2_glock
*gl
)
1523 gfs2_glock_remove_from_lru(gl
);
1525 spin_lock(&gl
->gl_spin
);
1526 if (gl
->gl_state
!= LM_ST_UNLOCKED
)
1527 handle_callback(gl
, LM_ST_UNLOCKED
, 0, false);
1528 spin_unlock(&gl
->gl_spin
);
1529 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
1534 * gfs2_glock_thaw - Thaw any frozen glocks
1535 * @sdp: The super block
1539 void gfs2_glock_thaw(struct gfs2_sbd
*sdp
)
1541 glock_hash_walk(thaw_glock
, sdp
);
1544 static void dump_glock(struct seq_file
*seq
, struct gfs2_glock
*gl
)
1546 spin_lock(&gl
->gl_spin
);
1547 gfs2_dump_glock(seq
, gl
);
1548 spin_unlock(&gl
->gl_spin
);
1551 static void dump_glock_func(struct gfs2_glock
*gl
)
1553 dump_glock(NULL
, gl
);
1557 * gfs2_gl_hash_clear - Empty out the glock hash table
1558 * @sdp: the filesystem
1559 * @wait: wait until it's all gone
1561 * Called when unmounting the filesystem.
1564 void gfs2_gl_hash_clear(struct gfs2_sbd
*sdp
)
1566 set_bit(SDF_SKIP_DLM_UNLOCK
, &sdp
->sd_flags
);
1567 flush_workqueue(glock_workqueue
);
1568 glock_hash_walk(clear_glock
, sdp
);
1569 flush_workqueue(glock_workqueue
);
1570 wait_event(sdp
->sd_glock_wait
, atomic_read(&sdp
->sd_glock_disposal
) == 0);
1571 glock_hash_walk(dump_glock_func
, sdp
);
1574 void gfs2_glock_finish_truncate(struct gfs2_inode
*ip
)
1576 struct gfs2_glock
*gl
= ip
->i_gl
;
1579 ret
= gfs2_truncatei_resume(ip
);
1580 gfs2_assert_withdraw(gl
->gl_sbd
, ret
== 0);
1582 spin_lock(&gl
->gl_spin
);
1583 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
1585 spin_unlock(&gl
->gl_spin
);
1588 static const char *state2str(unsigned state
)
1591 case LM_ST_UNLOCKED
:
1595 case LM_ST_DEFERRED
:
1597 case LM_ST_EXCLUSIVE
:
1603 static const char *hflags2str(char *buf
, unsigned flags
, unsigned long iflags
)
1606 if (flags
& LM_FLAG_TRY
)
1608 if (flags
& LM_FLAG_TRY_1CB
)
1610 if (flags
& LM_FLAG_NOEXP
)
1612 if (flags
& LM_FLAG_ANY
)
1614 if (flags
& LM_FLAG_PRIORITY
)
1616 if (flags
& GL_ASYNC
)
1618 if (flags
& GL_EXACT
)
1620 if (flags
& GL_NOCACHE
)
1622 if (test_bit(HIF_HOLDER
, &iflags
))
1624 if (test_bit(HIF_WAIT
, &iflags
))
1626 if (test_bit(HIF_FIRST
, &iflags
))
1633 * dump_holder - print information about a glock holder
1634 * @seq: the seq_file struct
1635 * @gh: the glock holder
1639 static void dump_holder(struct seq_file
*seq
, const struct gfs2_holder
*gh
)
1641 struct task_struct
*gh_owner
= NULL
;
1645 if (gh
->gh_owner_pid
)
1646 gh_owner
= pid_task(gh
->gh_owner_pid
, PIDTYPE_PID
);
1647 gfs2_print_dbg(seq
, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1648 state2str(gh
->gh_state
),
1649 hflags2str(flags_buf
, gh
->gh_flags
, gh
->gh_iflags
),
1651 gh
->gh_owner_pid
? (long)pid_nr(gh
->gh_owner_pid
) : -1,
1652 gh_owner
? gh_owner
->comm
: "(ended)",
1657 static const char *gflags2str(char *buf
, const struct gfs2_glock
*gl
)
1659 const unsigned long *gflags
= &gl
->gl_flags
;
1662 if (test_bit(GLF_LOCK
, gflags
))
1664 if (test_bit(GLF_DEMOTE
, gflags
))
1666 if (test_bit(GLF_PENDING_DEMOTE
, gflags
))
1668 if (test_bit(GLF_DEMOTE_IN_PROGRESS
, gflags
))
1670 if (test_bit(GLF_DIRTY
, gflags
))
1672 if (test_bit(GLF_LFLUSH
, gflags
))
1674 if (test_bit(GLF_INVALIDATE_IN_PROGRESS
, gflags
))
1676 if (test_bit(GLF_REPLY_PENDING
, gflags
))
1678 if (test_bit(GLF_INITIAL
, gflags
))
1680 if (test_bit(GLF_FROZEN
, gflags
))
1682 if (test_bit(GLF_QUEUED
, gflags
))
1684 if (test_bit(GLF_LRU
, gflags
))
1688 if (test_bit(GLF_BLOCKING
, gflags
))
1695 * gfs2_dump_glock - print information about a glock
1696 * @seq: The seq_file struct
1699 * The file format is as follows:
1700 * One line per object, capital letters are used to indicate objects
1701 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1702 * other objects are indented by a single space and follow the glock to
1703 * which they are related. Fields are indicated by lower case letters
1704 * followed by a colon and the field value, except for strings which are in
1705 * [] so that its possible to see if they are composed of spaces for
1706 * example. The field's are n = number (id of the object), f = flags,
1707 * t = type, s = state, r = refcount, e = error, p = pid.
1711 void gfs2_dump_glock(struct seq_file
*seq
, const struct gfs2_glock
*gl
)
1713 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1714 unsigned long long dtime
;
1715 const struct gfs2_holder
*gh
;
1716 char gflags_buf
[32];
1718 dtime
= jiffies
- gl
->gl_demote_time
;
1719 dtime
*= 1000000/HZ
; /* demote time in uSec */
1720 if (!test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
1722 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",
1723 state2str(gl
->gl_state
),
1724 gl
->gl_name
.ln_type
,
1725 (unsigned long long)gl
->gl_name
.ln_number
,
1726 gflags2str(gflags_buf
, gl
),
1727 state2str(gl
->gl_target
),
1728 state2str(gl
->gl_demote_state
), dtime
,
1729 atomic_read(&gl
->gl_ail_count
),
1730 atomic_read(&gl
->gl_revokes
),
1731 (int)gl
->gl_lockref
.count
, gl
->gl_hold_time
);
1733 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
)
1734 dump_holder(seq
, gh
);
1736 if (gl
->gl_state
!= LM_ST_UNLOCKED
&& glops
->go_dump
)
1737 glops
->go_dump(seq
, gl
);
1740 static int gfs2_glstats_seq_show(struct seq_file
*seq
, void *iter_ptr
)
1742 struct gfs2_glock
*gl
= iter_ptr
;
1744 seq_printf(seq
, "G: n:%u/%llx rtt:%lld/%lld rttb:%lld/%lld irt:%lld/%lld dcnt: %lld qcnt: %lld\n",
1745 gl
->gl_name
.ln_type
,
1746 (unsigned long long)gl
->gl_name
.ln_number
,
1747 (long long)gl
->gl_stats
.stats
[GFS2_LKS_SRTT
],
1748 (long long)gl
->gl_stats
.stats
[GFS2_LKS_SRTTVAR
],
1749 (long long)gl
->gl_stats
.stats
[GFS2_LKS_SRTTB
],
1750 (long long)gl
->gl_stats
.stats
[GFS2_LKS_SRTTVARB
],
1751 (long long)gl
->gl_stats
.stats
[GFS2_LKS_SIRT
],
1752 (long long)gl
->gl_stats
.stats
[GFS2_LKS_SIRTVAR
],
1753 (long long)gl
->gl_stats
.stats
[GFS2_LKS_DCOUNT
],
1754 (long long)gl
->gl_stats
.stats
[GFS2_LKS_QCOUNT
]);
1758 static const char *gfs2_gltype
[] = {
1772 static const char *gfs2_stype
[] = {
1773 [GFS2_LKS_SRTT
] = "srtt",
1774 [GFS2_LKS_SRTTVAR
] = "srttvar",
1775 [GFS2_LKS_SRTTB
] = "srttb",
1776 [GFS2_LKS_SRTTVARB
] = "srttvarb",
1777 [GFS2_LKS_SIRT
] = "sirt",
1778 [GFS2_LKS_SIRTVAR
] = "sirtvar",
1779 [GFS2_LKS_DCOUNT
] = "dlm",
1780 [GFS2_LKS_QCOUNT
] = "queue",
1783 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1785 static int gfs2_sbstats_seq_show(struct seq_file
*seq
, void *iter_ptr
)
1787 struct gfs2_glock_iter
*gi
= seq
->private;
1788 struct gfs2_sbd
*sdp
= gi
->sdp
;
1789 unsigned index
= gi
->hash
>> 3;
1790 unsigned subindex
= gi
->hash
& 0x07;
1794 if (index
== 0 && subindex
!= 0)
1797 seq_printf(seq
, "%-10s %8s:", gfs2_gltype
[index
],
1798 (index
== 0) ? "cpu": gfs2_stype
[subindex
]);
1800 for_each_possible_cpu(i
) {
1801 const struct gfs2_pcpu_lkstats
*lkstats
= per_cpu_ptr(sdp
->sd_lkstats
, i
);
1805 value
= lkstats
->lkstats
[index
- 1].stats
[subindex
];
1807 seq_printf(seq
, " %15lld", (long long)value
);
1809 seq_putc(seq
, '\n');
1813 int __init
gfs2_glock_init(void)
1816 for(i
= 0; i
< GFS2_GL_HASH_SIZE
; i
++) {
1817 INIT_HLIST_BL_HEAD(&gl_hash_table
[i
]);
1820 glock_workqueue
= alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM
|
1821 WQ_HIGHPRI
| WQ_FREEZABLE
, 0);
1822 if (!glock_workqueue
)
1824 gfs2_delete_workqueue
= alloc_workqueue("delete_workqueue",
1825 WQ_MEM_RECLAIM
| WQ_FREEZABLE
,
1827 if (!gfs2_delete_workqueue
) {
1828 destroy_workqueue(glock_workqueue
);
1832 register_shrinker(&glock_shrinker
);
1837 void gfs2_glock_exit(void)
1839 unregister_shrinker(&glock_shrinker
);
1840 destroy_workqueue(glock_workqueue
);
1841 destroy_workqueue(gfs2_delete_workqueue
);
1844 static inline struct gfs2_glock
*glock_hash_chain(unsigned hash
)
1846 return hlist_bl_entry(hlist_bl_first_rcu(&gl_hash_table
[hash
]),
1847 struct gfs2_glock
, gl_list
);
1850 static inline struct gfs2_glock
*glock_hash_next(struct gfs2_glock
*gl
)
1852 return hlist_bl_entry(rcu_dereference(gl
->gl_list
.next
),
1853 struct gfs2_glock
, gl_list
);
1856 static int gfs2_glock_iter_next(struct gfs2_glock_iter
*gi
)
1858 struct gfs2_glock
*gl
;
1863 gi
->gl
= glock_hash_next(gl
);
1866 if (gi
->hash
>= GFS2_GL_HASH_SIZE
) {
1870 gi
->gl
= glock_hash_chain(gi
->hash
);
1873 while (gi
->gl
== NULL
) {
1875 if (gi
->hash
>= GFS2_GL_HASH_SIZE
) {
1879 gi
->gl
= glock_hash_chain(gi
->hash
);
1882 /* Skip entries for other sb and dead entries */
1883 } while (gi
->sdp
!= gi
->gl
->gl_sbd
||
1884 __lockref_is_dead(&gi
->gl
->gl_lockref
));
1889 static void *gfs2_glock_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1891 struct gfs2_glock_iter
*gi
= seq
->private;
1894 if (gi
->last_pos
<= *pos
)
1895 n
= gi
->nhash
+ (*pos
- gi
->last_pos
);
1903 if (gfs2_glock_iter_next(gi
))
1907 gi
->last_pos
= *pos
;
1911 static void *gfs2_glock_seq_next(struct seq_file
*seq
, void *iter_ptr
,
1914 struct gfs2_glock_iter
*gi
= seq
->private;
1917 gi
->last_pos
= *pos
;
1918 if (gfs2_glock_iter_next(gi
))
1924 static void gfs2_glock_seq_stop(struct seq_file
*seq
, void *iter_ptr
)
1926 struct gfs2_glock_iter
*gi
= seq
->private;
1933 static int gfs2_glock_seq_show(struct seq_file
*seq
, void *iter_ptr
)
1935 dump_glock(seq
, iter_ptr
);
1939 static void *gfs2_sbstats_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1941 struct gfs2_glock_iter
*gi
= seq
->private;
1944 if (*pos
>= GFS2_NR_SBSTATS
)
1947 return SEQ_START_TOKEN
;
1950 static void *gfs2_sbstats_seq_next(struct seq_file
*seq
, void *iter_ptr
,
1953 struct gfs2_glock_iter
*gi
= seq
->private;
1956 if (gi
->hash
>= GFS2_NR_SBSTATS
) {
1960 return SEQ_START_TOKEN
;
1963 static void gfs2_sbstats_seq_stop(struct seq_file
*seq
, void *iter_ptr
)
1968 static const struct seq_operations gfs2_glock_seq_ops
= {
1969 .start
= gfs2_glock_seq_start
,
1970 .next
= gfs2_glock_seq_next
,
1971 .stop
= gfs2_glock_seq_stop
,
1972 .show
= gfs2_glock_seq_show
,
1975 static const struct seq_operations gfs2_glstats_seq_ops
= {
1976 .start
= gfs2_glock_seq_start
,
1977 .next
= gfs2_glock_seq_next
,
1978 .stop
= gfs2_glock_seq_stop
,
1979 .show
= gfs2_glstats_seq_show
,
1982 static const struct seq_operations gfs2_sbstats_seq_ops
= {
1983 .start
= gfs2_sbstats_seq_start
,
1984 .next
= gfs2_sbstats_seq_next
,
1985 .stop
= gfs2_sbstats_seq_stop
,
1986 .show
= gfs2_sbstats_seq_show
,
1989 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
1991 static int gfs2_glocks_open(struct inode
*inode
, struct file
*file
)
1993 int ret
= seq_open_private(file
, &gfs2_glock_seq_ops
,
1994 sizeof(struct gfs2_glock_iter
));
1996 struct seq_file
*seq
= file
->private_data
;
1997 struct gfs2_glock_iter
*gi
= seq
->private;
1998 gi
->sdp
= inode
->i_private
;
1999 seq
->buf
= kmalloc(GFS2_SEQ_GOODSIZE
, GFP_KERNEL
| __GFP_NOWARN
);
2001 seq
->size
= GFS2_SEQ_GOODSIZE
;
2006 static int gfs2_glstats_open(struct inode
*inode
, struct file
*file
)
2008 int ret
= seq_open_private(file
, &gfs2_glstats_seq_ops
,
2009 sizeof(struct gfs2_glock_iter
));
2011 struct seq_file
*seq
= file
->private_data
;
2012 struct gfs2_glock_iter
*gi
= seq
->private;
2013 gi
->sdp
= inode
->i_private
;
2014 seq
->buf
= kmalloc(GFS2_SEQ_GOODSIZE
, GFP_KERNEL
| __GFP_NOWARN
);
2016 seq
->size
= GFS2_SEQ_GOODSIZE
;
2021 static int gfs2_sbstats_open(struct inode
*inode
, struct file
*file
)
2023 int ret
= seq_open_private(file
, &gfs2_sbstats_seq_ops
,
2024 sizeof(struct gfs2_glock_iter
));
2026 struct seq_file
*seq
= file
->private_data
;
2027 struct gfs2_glock_iter
*gi
= seq
->private;
2028 gi
->sdp
= inode
->i_private
;
2033 static const struct file_operations gfs2_glocks_fops
= {
2034 .owner
= THIS_MODULE
,
2035 .open
= gfs2_glocks_open
,
2037 .llseek
= seq_lseek
,
2038 .release
= seq_release_private
,
2041 static const struct file_operations gfs2_glstats_fops
= {
2042 .owner
= THIS_MODULE
,
2043 .open
= gfs2_glstats_open
,
2045 .llseek
= seq_lseek
,
2046 .release
= seq_release_private
,
2049 static const struct file_operations gfs2_sbstats_fops
= {
2050 .owner
= THIS_MODULE
,
2051 .open
= gfs2_sbstats_open
,
2053 .llseek
= seq_lseek
,
2054 .release
= seq_release_private
,
2057 int gfs2_create_debugfs_file(struct gfs2_sbd
*sdp
)
2059 sdp
->debugfs_dir
= debugfs_create_dir(sdp
->sd_table_name
, gfs2_root
);
2060 if (!sdp
->debugfs_dir
)
2062 sdp
->debugfs_dentry_glocks
= debugfs_create_file("glocks",
2064 sdp
->debugfs_dir
, sdp
,
2066 if (!sdp
->debugfs_dentry_glocks
)
2069 sdp
->debugfs_dentry_glstats
= debugfs_create_file("glstats",
2071 sdp
->debugfs_dir
, sdp
,
2072 &gfs2_glstats_fops
);
2073 if (!sdp
->debugfs_dentry_glstats
)
2076 sdp
->debugfs_dentry_sbstats
= debugfs_create_file("sbstats",
2078 sdp
->debugfs_dir
, sdp
,
2079 &gfs2_sbstats_fops
);
2080 if (!sdp
->debugfs_dentry_sbstats
)
2085 gfs2_delete_debugfs_file(sdp
);
2089 void gfs2_delete_debugfs_file(struct gfs2_sbd
*sdp
)
2091 if (sdp
->debugfs_dir
) {
2092 if (sdp
->debugfs_dentry_glocks
) {
2093 debugfs_remove(sdp
->debugfs_dentry_glocks
);
2094 sdp
->debugfs_dentry_glocks
= NULL
;
2096 if (sdp
->debugfs_dentry_glstats
) {
2097 debugfs_remove(sdp
->debugfs_dentry_glstats
);
2098 sdp
->debugfs_dentry_glstats
= NULL
;
2100 if (sdp
->debugfs_dentry_sbstats
) {
2101 debugfs_remove(sdp
->debugfs_dentry_sbstats
);
2102 sdp
->debugfs_dentry_sbstats
= NULL
;
2104 debugfs_remove(sdp
->debugfs_dir
);
2105 sdp
->debugfs_dir
= NULL
;
2109 int gfs2_register_debugfs(void)
2111 gfs2_root
= debugfs_create_dir("gfs2", NULL
);
2112 return gfs2_root
? 0 : -ENOMEM
;
2115 void gfs2_unregister_debugfs(void)
2117 debugfs_remove(gfs2_root
);