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>
45 #define CREATE_TRACE_POINTS
46 #include "trace_gfs2.h"
48 struct gfs2_glock_iter
{
49 int hash
; /* hash bucket index */
50 unsigned nhash
; /* Index within current bucket */
51 struct gfs2_sbd
*sdp
; /* incore superblock */
52 struct gfs2_glock
*gl
; /* current glock struct */
53 loff_t last_pos
; /* last position */
56 typedef void (*glock_examiner
) (struct gfs2_glock
* gl
);
58 static int __dump_glock(struct seq_file
*seq
, const struct gfs2_glock
*gl
);
59 #define GLOCK_BUG_ON(gl,x) do { if (unlikely(x)) { __dump_glock(NULL, gl); BUG(); } } while(0)
60 static void do_xmote(struct gfs2_glock
*gl
, struct gfs2_holder
*gh
, unsigned int target
);
62 static struct dentry
*gfs2_root
;
63 static struct workqueue_struct
*glock_workqueue
;
64 struct workqueue_struct
*gfs2_delete_workqueue
;
65 static LIST_HEAD(lru_list
);
66 static atomic_t lru_count
= ATOMIC_INIT(0);
67 static DEFINE_SPINLOCK(lru_lock
);
69 #define GFS2_GL_HASH_SHIFT 15
70 #define GFS2_GL_HASH_SIZE (1 << GFS2_GL_HASH_SHIFT)
71 #define GFS2_GL_HASH_MASK (GFS2_GL_HASH_SIZE - 1)
73 static struct 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 kmem_cache_free(gfs2_glock_cachep
, gl
);
116 void gfs2_glock_free(struct gfs2_glock
*gl
)
118 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
120 call_rcu(&gl
->gl_rcu
, gfs2_glock_dealloc
);
121 if (atomic_dec_and_test(&sdp
->sd_glock_disposal
))
122 wake_up(&sdp
->sd_glock_wait
);
126 * gfs2_glock_hold() - increment reference count on glock
127 * @gl: The glock to hold
131 void gfs2_glock_hold(struct gfs2_glock
*gl
)
133 GLOCK_BUG_ON(gl
, atomic_read(&gl
->gl_ref
) == 0);
134 atomic_inc(&gl
->gl_ref
);
138 * demote_ok - Check to see if it's ok to unlock a glock
141 * Returns: 1 if it's ok
144 static int demote_ok(const struct gfs2_glock
*gl
)
146 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
148 if (gl
->gl_state
== LM_ST_UNLOCKED
)
150 if (!list_empty(&gl
->gl_holders
))
152 if (glops
->go_demote_ok
)
153 return glops
->go_demote_ok(gl
);
158 void gfs2_glock_add_to_lru(struct gfs2_glock
*gl
)
160 spin_lock(&lru_lock
);
162 if (!list_empty(&gl
->gl_lru
))
163 list_del_init(&gl
->gl_lru
);
165 atomic_inc(&lru_count
);
167 list_add_tail(&gl
->gl_lru
, &lru_list
);
168 set_bit(GLF_LRU
, &gl
->gl_flags
);
169 spin_unlock(&lru_lock
);
172 static void __gfs2_glock_remove_from_lru(struct gfs2_glock
*gl
)
174 if (!list_empty(&gl
->gl_lru
)) {
175 list_del_init(&gl
->gl_lru
);
176 atomic_dec(&lru_count
);
177 clear_bit(GLF_LRU
, &gl
->gl_flags
);
181 static void gfs2_glock_remove_from_lru(struct gfs2_glock
*gl
)
183 spin_lock(&lru_lock
);
184 __gfs2_glock_remove_from_lru(gl
);
185 spin_unlock(&lru_lock
);
189 * gfs2_glock_put_nolock() - Decrement reference count on glock
190 * @gl: The glock to put
192 * This function should only be used if the caller has its own reference
193 * to the glock, in addition to the one it is dropping.
196 void gfs2_glock_put_nolock(struct gfs2_glock
*gl
)
198 if (atomic_dec_and_test(&gl
->gl_ref
))
203 * gfs2_glock_put() - Decrement reference count on glock
204 * @gl: The glock to put
208 void gfs2_glock_put(struct gfs2_glock
*gl
)
210 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
211 struct address_space
*mapping
= gfs2_glock2aspace(gl
);
213 if (atomic_dec_and_lock(&gl
->gl_ref
, &lru_lock
)) {
214 __gfs2_glock_remove_from_lru(gl
);
215 spin_unlock(&lru_lock
);
216 spin_lock_bucket(gl
->gl_hash
);
217 hlist_bl_del_rcu(&gl
->gl_list
);
218 spin_unlock_bucket(gl
->gl_hash
);
219 GLOCK_BUG_ON(gl
, !list_empty(&gl
->gl_holders
));
220 GLOCK_BUG_ON(gl
, mapping
&& mapping
->nrpages
);
221 trace_gfs2_glock_put(gl
);
222 sdp
->sd_lockstruct
.ls_ops
->lm_put_lock(gl
);
227 * search_bucket() - Find struct gfs2_glock by lock number
228 * @bucket: the bucket to search
229 * @name: The lock name
231 * Returns: NULL, or the struct gfs2_glock with the requested number
234 static struct gfs2_glock
*search_bucket(unsigned int hash
,
235 const struct gfs2_sbd
*sdp
,
236 const struct lm_lockname
*name
)
238 struct gfs2_glock
*gl
;
239 struct hlist_bl_node
*h
;
241 hlist_bl_for_each_entry_rcu(gl
, h
, &gl_hash_table
[hash
], gl_list
) {
242 if (!lm_name_equal(&gl
->gl_name
, name
))
244 if (gl
->gl_sbd
!= sdp
)
246 if (atomic_inc_not_zero(&gl
->gl_ref
))
254 * may_grant - check if its ok to grant a new lock
256 * @gh: The lock request which we wish to grant
258 * Returns: true if its ok to grant the lock
261 static inline int may_grant(const struct gfs2_glock
*gl
, const struct gfs2_holder
*gh
)
263 const struct gfs2_holder
*gh_head
= list_entry(gl
->gl_holders
.next
, const struct gfs2_holder
, gh_list
);
264 if ((gh
->gh_state
== LM_ST_EXCLUSIVE
||
265 gh_head
->gh_state
== LM_ST_EXCLUSIVE
) && gh
!= gh_head
)
267 if (gl
->gl_state
== gh
->gh_state
)
269 if (gh
->gh_flags
& GL_EXACT
)
271 if (gl
->gl_state
== LM_ST_EXCLUSIVE
) {
272 if (gh
->gh_state
== LM_ST_SHARED
&& gh_head
->gh_state
== LM_ST_SHARED
)
274 if (gh
->gh_state
== LM_ST_DEFERRED
&& gh_head
->gh_state
== LM_ST_DEFERRED
)
277 if (gl
->gl_state
!= LM_ST_UNLOCKED
&& (gh
->gh_flags
& LM_FLAG_ANY
))
282 static void gfs2_holder_wake(struct gfs2_holder
*gh
)
284 clear_bit(HIF_WAIT
, &gh
->gh_iflags
);
285 smp_mb__after_clear_bit();
286 wake_up_bit(&gh
->gh_iflags
, HIF_WAIT
);
290 * do_error - Something unexpected has happened during a lock request
294 static inline void do_error(struct gfs2_glock
*gl
, const int ret
)
296 struct gfs2_holder
*gh
, *tmp
;
298 list_for_each_entry_safe(gh
, tmp
, &gl
->gl_holders
, gh_list
) {
299 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
301 if (ret
& LM_OUT_ERROR
)
303 else if (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
))
304 gh
->gh_error
= GLR_TRYFAILED
;
307 list_del_init(&gh
->gh_list
);
308 trace_gfs2_glock_queue(gh
, 0);
309 gfs2_holder_wake(gh
);
314 * do_promote - promote as many requests as possible on the current queue
317 * Returns: 1 if there is a blocked holder at the head of the list, or 2
318 * if a type specific operation is underway.
321 static int do_promote(struct gfs2_glock
*gl
)
322 __releases(&gl
->gl_spin
)
323 __acquires(&gl
->gl_spin
)
325 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
326 struct gfs2_holder
*gh
, *tmp
;
330 list_for_each_entry_safe(gh
, tmp
, &gl
->gl_holders
, gh_list
) {
331 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
333 if (may_grant(gl
, gh
)) {
334 if (gh
->gh_list
.prev
== &gl
->gl_holders
&&
336 spin_unlock(&gl
->gl_spin
);
337 /* FIXME: eliminate this eventually */
338 ret
= glops
->go_lock(gh
);
339 spin_lock(&gl
->gl_spin
);
344 list_del_init(&gh
->gh_list
);
345 trace_gfs2_glock_queue(gh
, 0);
346 gfs2_holder_wake(gh
);
349 set_bit(HIF_HOLDER
, &gh
->gh_iflags
);
350 trace_gfs2_promote(gh
, 1);
351 gfs2_holder_wake(gh
);
354 set_bit(HIF_HOLDER
, &gh
->gh_iflags
);
355 trace_gfs2_promote(gh
, 0);
356 gfs2_holder_wake(gh
);
359 if (gh
->gh_list
.prev
== &gl
->gl_holders
)
368 * find_first_waiter - find the first gh that's waiting for the glock
372 static inline struct gfs2_holder
*find_first_waiter(const struct gfs2_glock
*gl
)
374 struct gfs2_holder
*gh
;
376 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
) {
377 if (!test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
384 * state_change - record that the glock is now in a different state
386 * @new_state the new state
390 static void state_change(struct gfs2_glock
*gl
, unsigned int new_state
)
394 held1
= (gl
->gl_state
!= LM_ST_UNLOCKED
);
395 held2
= (new_state
!= LM_ST_UNLOCKED
);
397 if (held1
!= held2
) {
401 gfs2_glock_put_nolock(gl
);
403 if (held1
&& held2
&& list_empty(&gl
->gl_holders
))
404 clear_bit(GLF_QUEUED
, &gl
->gl_flags
);
406 if (new_state
!= gl
->gl_target
)
407 /* shorten our minimum hold time */
408 gl
->gl_hold_time
= max(gl
->gl_hold_time
- GL_GLOCK_HOLD_DECR
,
410 gl
->gl_state
= new_state
;
411 gl
->gl_tchange
= jiffies
;
414 static void gfs2_demote_wake(struct gfs2_glock
*gl
)
416 gl
->gl_demote_state
= LM_ST_EXCLUSIVE
;
417 clear_bit(GLF_DEMOTE
, &gl
->gl_flags
);
418 smp_mb__after_clear_bit();
419 wake_up_bit(&gl
->gl_flags
, GLF_DEMOTE
);
423 * finish_xmote - The DLM has replied to one of our lock requests
425 * @ret: The status from the DLM
429 static void finish_xmote(struct gfs2_glock
*gl
, unsigned int ret
)
431 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
432 struct gfs2_holder
*gh
;
433 unsigned state
= ret
& LM_OUT_ST_MASK
;
436 spin_lock(&gl
->gl_spin
);
437 trace_gfs2_glock_state_change(gl
, state
);
438 state_change(gl
, state
);
439 gh
= find_first_waiter(gl
);
441 /* Demote to UN request arrived during demote to SH or DF */
442 if (test_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
) &&
443 state
!= LM_ST_UNLOCKED
&& gl
->gl_demote_state
== LM_ST_UNLOCKED
)
444 gl
->gl_target
= LM_ST_UNLOCKED
;
446 /* Check for state != intended state */
447 if (unlikely(state
!= gl
->gl_target
)) {
448 if (gh
&& !test_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
)) {
449 /* move to back of queue and try next entry */
450 if (ret
& LM_OUT_CANCELED
) {
451 if ((gh
->gh_flags
& LM_FLAG_PRIORITY
) == 0)
452 list_move_tail(&gh
->gh_list
, &gl
->gl_holders
);
453 gh
= find_first_waiter(gl
);
454 gl
->gl_target
= gh
->gh_state
;
457 /* Some error or failed "try lock" - report it */
458 if ((ret
& LM_OUT_ERROR
) ||
459 (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
))) {
460 gl
->gl_target
= gl
->gl_state
;
466 /* Unlocked due to conversion deadlock, try again */
469 do_xmote(gl
, gh
, gl
->gl_target
);
471 /* Conversion fails, unlock and try again */
474 do_xmote(gl
, gh
, LM_ST_UNLOCKED
);
476 default: /* Everything else */
477 printk(KERN_ERR
"GFS2: wanted %u got %u\n", gl
->gl_target
, state
);
480 spin_unlock(&gl
->gl_spin
);
484 /* Fast path - we got what we asked for */
485 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
))
486 gfs2_demote_wake(gl
);
487 if (state
!= LM_ST_UNLOCKED
) {
488 if (glops
->go_xmote_bh
) {
489 spin_unlock(&gl
->gl_spin
);
490 rv
= glops
->go_xmote_bh(gl
, gh
);
491 spin_lock(&gl
->gl_spin
);
502 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
504 spin_unlock(&gl
->gl_spin
);
508 * do_xmote - Calls the DLM to change the state of a lock
509 * @gl: The lock state
510 * @gh: The holder (only for promotes)
511 * @target: The target lock state
515 static void do_xmote(struct gfs2_glock
*gl
, struct gfs2_holder
*gh
, unsigned int target
)
516 __releases(&gl
->gl_spin
)
517 __acquires(&gl
->gl_spin
)
519 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
520 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
521 unsigned int lck_flags
= gh
? gh
->gh_flags
: 0;
524 lck_flags
&= (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
| LM_FLAG_NOEXP
|
526 GLOCK_BUG_ON(gl
, gl
->gl_state
== target
);
527 GLOCK_BUG_ON(gl
, gl
->gl_state
== gl
->gl_target
);
528 if ((target
== LM_ST_UNLOCKED
|| target
== LM_ST_DEFERRED
) &&
530 set_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
);
531 do_error(gl
, 0); /* Fail queued try locks */
534 set_bit(GLF_BLOCKING
, &gl
->gl_flags
);
535 if ((gl
->gl_req
== LM_ST_UNLOCKED
) ||
536 (gl
->gl_state
== LM_ST_EXCLUSIVE
) ||
537 (lck_flags
& (LM_FLAG_TRY
|LM_FLAG_TRY_1CB
)))
538 clear_bit(GLF_BLOCKING
, &gl
->gl_flags
);
539 spin_unlock(&gl
->gl_spin
);
540 if (glops
->go_xmote_th
)
541 glops
->go_xmote_th(gl
);
542 if (test_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
))
543 glops
->go_inval(gl
, target
== LM_ST_DEFERRED
? 0 : DIO_METADATA
);
544 clear_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
);
547 if (sdp
->sd_lockstruct
.ls_ops
->lm_lock
) {
549 ret
= sdp
->sd_lockstruct
.ls_ops
->lm_lock(gl
, target
, lck_flags
);
550 GLOCK_BUG_ON(gl
, ret
);
551 } else { /* lock_nolock */
552 finish_xmote(gl
, target
);
553 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
557 spin_lock(&gl
->gl_spin
);
561 * find_first_holder - find the first "holder" gh
565 static inline struct gfs2_holder
*find_first_holder(const struct gfs2_glock
*gl
)
567 struct gfs2_holder
*gh
;
569 if (!list_empty(&gl
->gl_holders
)) {
570 gh
= list_entry(gl
->gl_holders
.next
, struct gfs2_holder
, gh_list
);
571 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
578 * run_queue - do all outstanding tasks related to a glock
579 * @gl: The glock in question
580 * @nonblock: True if we must not block in run_queue
584 static void run_queue(struct gfs2_glock
*gl
, const int nonblock
)
585 __releases(&gl
->gl_spin
)
586 __acquires(&gl
->gl_spin
)
588 struct gfs2_holder
*gh
= NULL
;
591 if (test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
))
594 GLOCK_BUG_ON(gl
, test_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
));
596 if (test_bit(GLF_DEMOTE
, &gl
->gl_flags
) &&
597 gl
->gl_demote_state
!= gl
->gl_state
) {
598 if (find_first_holder(gl
))
602 set_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
);
603 GLOCK_BUG_ON(gl
, gl
->gl_demote_state
== LM_ST_EXCLUSIVE
);
604 gl
->gl_target
= gl
->gl_demote_state
;
606 if (test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
607 gfs2_demote_wake(gl
);
608 ret
= do_promote(gl
);
613 gh
= find_first_waiter(gl
);
614 gl
->gl_target
= gh
->gh_state
;
615 if (!(gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
)))
616 do_error(gl
, 0); /* Fail queued try locks */
618 do_xmote(gl
, gh
, gl
->gl_target
);
623 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
624 smp_mb__after_clear_bit();
626 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
627 gfs2_glock_put_nolock(gl
);
631 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
632 smp_mb__after_clear_bit();
636 static void delete_work_func(struct work_struct
*work
)
638 struct gfs2_glock
*gl
= container_of(work
, struct gfs2_glock
, gl_delete
);
639 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
640 struct gfs2_inode
*ip
;
642 u64 no_addr
= gl
->gl_name
.ln_number
;
645 /* Note: Unsafe to dereference ip as we don't hold right refs/locks */
648 inode
= gfs2_ilookup(sdp
->sd_vfs
, no_addr
, 1);
650 inode
= gfs2_lookup_by_inum(sdp
, no_addr
, NULL
, GFS2_BLKST_UNLINKED
);
651 if (inode
&& !IS_ERR(inode
)) {
652 d_prune_aliases(inode
);
658 static void glock_work_func(struct work_struct
*work
)
660 unsigned long delay
= 0;
661 struct gfs2_glock
*gl
= container_of(work
, struct gfs2_glock
, gl_work
.work
);
664 if (test_and_clear_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
)) {
665 finish_xmote(gl
, gl
->gl_reply
);
668 spin_lock(&gl
->gl_spin
);
669 if (test_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
) &&
670 gl
->gl_state
!= LM_ST_UNLOCKED
&&
671 gl
->gl_demote_state
!= LM_ST_EXCLUSIVE
) {
672 unsigned long holdtime
, now
= jiffies
;
674 holdtime
= gl
->gl_tchange
+ gl
->gl_hold_time
;
675 if (time_before(now
, holdtime
))
676 delay
= holdtime
- now
;
679 clear_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
);
680 set_bit(GLF_DEMOTE
, &gl
->gl_flags
);
684 spin_unlock(&gl
->gl_spin
);
688 if (gl
->gl_name
.ln_type
!= LM_TYPE_INODE
)
690 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, delay
) == 0)
698 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
699 * @sdp: The GFS2 superblock
700 * @number: the lock number
701 * @glops: The glock_operations to use
702 * @create: If 0, don't create the glock if it doesn't exist
703 * @glp: the glock is returned here
705 * This does not lock a glock, just finds/creates structures for one.
710 int gfs2_glock_get(struct gfs2_sbd
*sdp
, u64 number
,
711 const struct gfs2_glock_operations
*glops
, int create
,
712 struct gfs2_glock
**glp
)
714 struct super_block
*s
= sdp
->sd_vfs
;
715 struct lm_lockname name
= { .ln_number
= number
, .ln_type
= glops
->go_type
};
716 struct gfs2_glock
*gl
, *tmp
;
717 unsigned int hash
= gl_hash(sdp
, &name
);
718 struct address_space
*mapping
;
719 struct kmem_cache
*cachep
;
722 gl
= search_bucket(hash
, sdp
, &name
);
731 if (glops
->go_flags
& GLOF_ASPACE
)
732 cachep
= gfs2_glock_aspace_cachep
;
734 cachep
= gfs2_glock_cachep
;
735 gl
= kmem_cache_alloc(cachep
, GFP_KERNEL
);
739 atomic_inc(&sdp
->sd_glock_disposal
);
743 atomic_set(&gl
->gl_ref
, 1);
744 gl
->gl_state
= LM_ST_UNLOCKED
;
745 gl
->gl_target
= LM_ST_UNLOCKED
;
746 gl
->gl_demote_state
= LM_ST_EXCLUSIVE
;
749 gl
->gl_dstamp
= ktime_set(0, 0);
751 /* We use the global stats to estimate the initial per-glock stats */
752 gl
->gl_stats
= this_cpu_ptr(sdp
->sd_lkstats
)->lkstats
[glops
->go_type
];
754 gl
->gl_stats
.stats
[GFS2_LKS_DCOUNT
] = 0;
755 gl
->gl_stats
.stats
[GFS2_LKS_QCOUNT
] = 0;
756 memset(&gl
->gl_lksb
, 0, sizeof(struct dlm_lksb
));
757 memset(gl
->gl_lvb
, 0, 32 * sizeof(char));
758 gl
->gl_lksb
.sb_lvbptr
= gl
->gl_lvb
;
759 gl
->gl_tchange
= jiffies
;
760 gl
->gl_object
= NULL
;
761 gl
->gl_hold_time
= GL_GLOCK_DFT_HOLD
;
762 INIT_DELAYED_WORK(&gl
->gl_work
, glock_work_func
);
763 INIT_WORK(&gl
->gl_delete
, delete_work_func
);
765 mapping
= gfs2_glock2aspace(gl
);
767 mapping
->a_ops
= &gfs2_meta_aops
;
768 mapping
->host
= s
->s_bdev
->bd_inode
;
770 mapping_set_gfp_mask(mapping
, GFP_NOFS
);
771 mapping
->private_data
= NULL
;
772 mapping
->backing_dev_info
= s
->s_bdi
;
773 mapping
->writeback_index
= 0;
776 spin_lock_bucket(hash
);
777 tmp
= search_bucket(hash
, sdp
, &name
);
779 spin_unlock_bucket(hash
);
780 kmem_cache_free(cachep
, gl
);
781 atomic_dec(&sdp
->sd_glock_disposal
);
784 hlist_bl_add_head_rcu(&gl
->gl_list
, &gl_hash_table
[hash
]);
785 spin_unlock_bucket(hash
);
794 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
796 * @state: the state we're requesting
797 * @flags: the modifier flags
798 * @gh: the holder structure
802 void gfs2_holder_init(struct gfs2_glock
*gl
, unsigned int state
, unsigned flags
,
803 struct gfs2_holder
*gh
)
805 INIT_LIST_HEAD(&gh
->gh_list
);
807 gh
->gh_ip
= (unsigned long)__builtin_return_address(0);
808 gh
->gh_owner_pid
= get_pid(task_pid(current
));
809 gh
->gh_state
= state
;
810 gh
->gh_flags
= flags
;
817 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
818 * @state: the state we're requesting
819 * @flags: the modifier flags
820 * @gh: the holder structure
822 * Don't mess with the glock.
826 void gfs2_holder_reinit(unsigned int state
, unsigned flags
, struct gfs2_holder
*gh
)
828 gh
->gh_state
= state
;
829 gh
->gh_flags
= flags
;
831 gh
->gh_ip
= (unsigned long)__builtin_return_address(0);
832 if (gh
->gh_owner_pid
)
833 put_pid(gh
->gh_owner_pid
);
834 gh
->gh_owner_pid
= get_pid(task_pid(current
));
838 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
839 * @gh: the holder structure
843 void gfs2_holder_uninit(struct gfs2_holder
*gh
)
845 put_pid(gh
->gh_owner_pid
);
846 gfs2_glock_put(gh
->gh_gl
);
852 * gfs2_glock_holder_wait
855 * This function and gfs2_glock_demote_wait both show up in the WCHAN
856 * field. Thus I've separated these otherwise identical functions in
857 * order to be more informative to the user.
860 static int gfs2_glock_holder_wait(void *word
)
866 static int gfs2_glock_demote_wait(void *word
)
873 * gfs2_glock_wait - wait on a glock acquisition
874 * @gh: the glock holder
876 * Returns: 0 on success
879 int gfs2_glock_wait(struct gfs2_holder
*gh
)
881 unsigned long time1
= jiffies
;
884 wait_on_bit(&gh
->gh_iflags
, HIF_WAIT
, gfs2_glock_holder_wait
, TASK_UNINTERRUPTIBLE
);
885 if (time_after(jiffies
, time1
+ HZ
)) /* have we waited > a second? */
886 /* Lengthen the minimum hold time. */
887 gh
->gh_gl
->gl_hold_time
= min(gh
->gh_gl
->gl_hold_time
+
894 * handle_callback - process a demote request
896 * @state: the state the caller wants us to change to
898 * There are only two requests that we are going to see in actual
899 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
902 static void handle_callback(struct gfs2_glock
*gl
, unsigned int state
,
905 int bit
= delay
? GLF_PENDING_DEMOTE
: GLF_DEMOTE
;
907 set_bit(bit
, &gl
->gl_flags
);
908 if (gl
->gl_demote_state
== LM_ST_EXCLUSIVE
) {
909 gl
->gl_demote_state
= state
;
910 gl
->gl_demote_time
= jiffies
;
911 } else if (gl
->gl_demote_state
!= LM_ST_UNLOCKED
&&
912 gl
->gl_demote_state
!= state
) {
913 gl
->gl_demote_state
= LM_ST_UNLOCKED
;
915 if (gl
->gl_ops
->go_callback
)
916 gl
->gl_ops
->go_callback(gl
);
917 trace_gfs2_demote_rq(gl
);
920 void gfs2_print_dbg(struct seq_file
*seq
, const char *fmt
, ...)
922 struct va_format vaf
;
928 seq_vprintf(seq
, fmt
, args
);
933 printk(KERN_ERR
" %pV", &vaf
);
940 * add_to_queue - Add a holder to the wait queue (but look for recursion)
941 * @gh: the holder structure to add
943 * Eventually we should move the recursive locking trap to a
944 * debugging option or something like that. This is the fast
945 * path and needs to have the minimum number of distractions.
949 static inline void add_to_queue(struct gfs2_holder
*gh
)
950 __releases(&gl
->gl_spin
)
951 __acquires(&gl
->gl_spin
)
953 struct gfs2_glock
*gl
= gh
->gh_gl
;
954 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
955 struct list_head
*insert_pt
= NULL
;
956 struct gfs2_holder
*gh2
;
959 BUG_ON(gh
->gh_owner_pid
== NULL
);
960 if (test_and_set_bit(HIF_WAIT
, &gh
->gh_iflags
))
963 if (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
)) {
964 if (test_bit(GLF_LOCK
, &gl
->gl_flags
))
965 try_futile
= !may_grant(gl
, gh
);
966 if (test_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
))
970 list_for_each_entry(gh2
, &gl
->gl_holders
, gh_list
) {
971 if (unlikely(gh2
->gh_owner_pid
== gh
->gh_owner_pid
&&
972 (gh
->gh_gl
->gl_ops
->go_type
!= LM_TYPE_FLOCK
)))
975 !(gh2
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
))) {
977 gh
->gh_error
= GLR_TRYFAILED
;
978 gfs2_holder_wake(gh
);
981 if (test_bit(HIF_HOLDER
, &gh2
->gh_iflags
))
983 if (unlikely((gh
->gh_flags
& LM_FLAG_PRIORITY
) && !insert_pt
))
984 insert_pt
= &gh2
->gh_list
;
986 set_bit(GLF_QUEUED
, &gl
->gl_flags
);
987 trace_gfs2_glock_queue(gh
, 1);
988 gfs2_glstats_inc(gl
, GFS2_LKS_QCOUNT
);
989 gfs2_sbstats_inc(gl
, GFS2_LKS_QCOUNT
);
990 if (likely(insert_pt
== NULL
)) {
991 list_add_tail(&gh
->gh_list
, &gl
->gl_holders
);
992 if (unlikely(gh
->gh_flags
& LM_FLAG_PRIORITY
))
996 list_add_tail(&gh
->gh_list
, insert_pt
);
998 gh
= list_entry(gl
->gl_holders
.next
, struct gfs2_holder
, gh_list
);
999 if (!(gh
->gh_flags
& LM_FLAG_PRIORITY
)) {
1000 spin_unlock(&gl
->gl_spin
);
1001 if (sdp
->sd_lockstruct
.ls_ops
->lm_cancel
)
1002 sdp
->sd_lockstruct
.ls_ops
->lm_cancel(gl
);
1003 spin_lock(&gl
->gl_spin
);
1008 print_symbol(KERN_ERR
"original: %s\n", gh2
->gh_ip
);
1009 printk(KERN_ERR
"pid: %d\n", pid_nr(gh2
->gh_owner_pid
));
1010 printk(KERN_ERR
"lock type: %d req lock state : %d\n",
1011 gh2
->gh_gl
->gl_name
.ln_type
, gh2
->gh_state
);
1012 print_symbol(KERN_ERR
"new: %s\n", gh
->gh_ip
);
1013 printk(KERN_ERR
"pid: %d\n", pid_nr(gh
->gh_owner_pid
));
1014 printk(KERN_ERR
"lock type: %d req lock state : %d\n",
1015 gh
->gh_gl
->gl_name
.ln_type
, gh
->gh_state
);
1016 __dump_glock(NULL
, gl
);
1021 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1022 * @gh: the holder structure
1024 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1026 * Returns: 0, GLR_TRYFAILED, or errno on failure
1029 int gfs2_glock_nq(struct gfs2_holder
*gh
)
1031 struct gfs2_glock
*gl
= gh
->gh_gl
;
1032 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
1035 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)))
1038 if (test_bit(GLF_LRU
, &gl
->gl_flags
))
1039 gfs2_glock_remove_from_lru(gl
);
1041 spin_lock(&gl
->gl_spin
);
1043 if ((LM_FLAG_NOEXP
& gh
->gh_flags
) &&
1044 test_and_clear_bit(GLF_FROZEN
, &gl
->gl_flags
))
1045 set_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
);
1047 spin_unlock(&gl
->gl_spin
);
1049 if (!(gh
->gh_flags
& GL_ASYNC
))
1050 error
= gfs2_glock_wait(gh
);
1056 * gfs2_glock_poll - poll to see if an async request has been completed
1059 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1062 int gfs2_glock_poll(struct gfs2_holder
*gh
)
1064 return test_bit(HIF_WAIT
, &gh
->gh_iflags
) ? 0 : 1;
1068 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1069 * @gh: the glock holder
1073 void gfs2_glock_dq(struct gfs2_holder
*gh
)
1075 struct gfs2_glock
*gl
= gh
->gh_gl
;
1076 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1080 spin_lock(&gl
->gl_spin
);
1081 if (gh
->gh_flags
& GL_NOCACHE
)
1082 handle_callback(gl
, LM_ST_UNLOCKED
, 0);
1084 list_del_init(&gh
->gh_list
);
1085 if (find_first_holder(gl
) == NULL
) {
1086 if (glops
->go_unlock
) {
1087 GLOCK_BUG_ON(gl
, test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
));
1088 spin_unlock(&gl
->gl_spin
);
1089 glops
->go_unlock(gh
);
1090 spin_lock(&gl
->gl_spin
);
1091 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
1093 if (list_empty(&gl
->gl_holders
) &&
1094 !test_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
) &&
1095 !test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
1098 if (!test_bit(GLF_LFLUSH
, &gl
->gl_flags
) && demote_ok(gl
))
1099 gfs2_glock_add_to_lru(gl
);
1101 trace_gfs2_glock_queue(gh
, 0);
1102 spin_unlock(&gl
->gl_spin
);
1103 if (likely(fast_path
))
1106 gfs2_glock_hold(gl
);
1107 if (test_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
) &&
1108 !test_bit(GLF_DEMOTE
, &gl
->gl_flags
) &&
1109 gl
->gl_name
.ln_type
== LM_TYPE_INODE
)
1110 delay
= gl
->gl_hold_time
;
1111 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, delay
) == 0)
1115 void gfs2_glock_dq_wait(struct gfs2_holder
*gh
)
1117 struct gfs2_glock
*gl
= gh
->gh_gl
;
1120 wait_on_bit(&gl
->gl_flags
, GLF_DEMOTE
, gfs2_glock_demote_wait
, TASK_UNINTERRUPTIBLE
);
1124 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1125 * @gh: the holder structure
1129 void gfs2_glock_dq_uninit(struct gfs2_holder
*gh
)
1132 gfs2_holder_uninit(gh
);
1136 * gfs2_glock_nq_num - acquire a glock based on lock number
1137 * @sdp: the filesystem
1138 * @number: the lock number
1139 * @glops: the glock operations for the type of glock
1140 * @state: the state to acquire the glock in
1141 * @flags: modifier flags for the acquisition
1142 * @gh: the struct gfs2_holder
1147 int gfs2_glock_nq_num(struct gfs2_sbd
*sdp
, u64 number
,
1148 const struct gfs2_glock_operations
*glops
,
1149 unsigned int state
, int flags
, struct gfs2_holder
*gh
)
1151 struct gfs2_glock
*gl
;
1154 error
= gfs2_glock_get(sdp
, number
, glops
, CREATE
, &gl
);
1156 error
= gfs2_glock_nq_init(gl
, state
, flags
, gh
);
1164 * glock_compare - Compare two struct gfs2_glock structures for sorting
1165 * @arg_a: the first structure
1166 * @arg_b: the second structure
1170 static int glock_compare(const void *arg_a
, const void *arg_b
)
1172 const struct gfs2_holder
*gh_a
= *(const struct gfs2_holder
**)arg_a
;
1173 const struct gfs2_holder
*gh_b
= *(const struct gfs2_holder
**)arg_b
;
1174 const struct lm_lockname
*a
= &gh_a
->gh_gl
->gl_name
;
1175 const struct lm_lockname
*b
= &gh_b
->gh_gl
->gl_name
;
1177 if (a
->ln_number
> b
->ln_number
)
1179 if (a
->ln_number
< b
->ln_number
)
1181 BUG_ON(gh_a
->gh_gl
->gl_ops
->go_type
== gh_b
->gh_gl
->gl_ops
->go_type
);
1186 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1187 * @num_gh: the number of structures
1188 * @ghs: an array of struct gfs2_holder structures
1190 * Returns: 0 on success (all glocks acquired),
1191 * errno on failure (no glocks acquired)
1194 static int nq_m_sync(unsigned int num_gh
, struct gfs2_holder
*ghs
,
1195 struct gfs2_holder
**p
)
1200 for (x
= 0; x
< num_gh
; x
++)
1203 sort(p
, num_gh
, sizeof(struct gfs2_holder
*), glock_compare
, NULL
);
1205 for (x
= 0; x
< num_gh
; x
++) {
1206 p
[x
]->gh_flags
&= ~(LM_FLAG_TRY
| GL_ASYNC
);
1208 error
= gfs2_glock_nq(p
[x
]);
1211 gfs2_glock_dq(p
[x
]);
1220 * gfs2_glock_nq_m - acquire multiple glocks
1221 * @num_gh: the number of structures
1222 * @ghs: an array of struct gfs2_holder structures
1225 * Returns: 0 on success (all glocks acquired),
1226 * errno on failure (no glocks acquired)
1229 int gfs2_glock_nq_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1231 struct gfs2_holder
*tmp
[4];
1232 struct gfs2_holder
**pph
= tmp
;
1239 ghs
->gh_flags
&= ~(LM_FLAG_TRY
| GL_ASYNC
);
1240 return gfs2_glock_nq(ghs
);
1244 pph
= kmalloc(num_gh
* sizeof(struct gfs2_holder
*), GFP_NOFS
);
1249 error
= nq_m_sync(num_gh
, ghs
, pph
);
1258 * gfs2_glock_dq_m - release multiple glocks
1259 * @num_gh: the number of structures
1260 * @ghs: an array of struct gfs2_holder structures
1264 void gfs2_glock_dq_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1267 gfs2_glock_dq(&ghs
[num_gh
]);
1271 * gfs2_glock_dq_uninit_m - release multiple glocks
1272 * @num_gh: the number of structures
1273 * @ghs: an array of struct gfs2_holder structures
1277 void gfs2_glock_dq_uninit_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1280 gfs2_glock_dq_uninit(&ghs
[num_gh
]);
1283 void gfs2_glock_cb(struct gfs2_glock
*gl
, unsigned int state
)
1285 unsigned long delay
= 0;
1286 unsigned long holdtime
;
1287 unsigned long now
= jiffies
;
1289 gfs2_glock_hold(gl
);
1290 holdtime
= gl
->gl_tchange
+ gl
->gl_hold_time
;
1291 if (test_bit(GLF_QUEUED
, &gl
->gl_flags
) &&
1292 gl
->gl_name
.ln_type
== LM_TYPE_INODE
) {
1293 if (time_before(now
, holdtime
))
1294 delay
= holdtime
- now
;
1295 if (test_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
))
1296 delay
= gl
->gl_hold_time
;
1299 spin_lock(&gl
->gl_spin
);
1300 handle_callback(gl
, state
, delay
);
1301 spin_unlock(&gl
->gl_spin
);
1302 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, delay
) == 0)
1307 * gfs2_should_freeze - Figure out if glock should be frozen
1308 * @gl: The glock in question
1310 * Glocks are not frozen if (a) the result of the dlm operation is
1311 * an error, (b) the locking operation was an unlock operation or
1312 * (c) if there is a "noexp" flagged request anywhere in the queue
1314 * Returns: 1 if freezing should occur, 0 otherwise
1317 static int gfs2_should_freeze(const struct gfs2_glock
*gl
)
1319 const struct gfs2_holder
*gh
;
1321 if (gl
->gl_reply
& ~LM_OUT_ST_MASK
)
1323 if (gl
->gl_target
== LM_ST_UNLOCKED
)
1326 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
) {
1327 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
1329 if (LM_FLAG_NOEXP
& gh
->gh_flags
)
1337 * gfs2_glock_complete - Callback used by locking
1338 * @gl: Pointer to the glock
1339 * @ret: The return value from the dlm
1341 * The gl_reply field is under the gl_spin lock so that it is ok
1342 * to use a bitfield shared with other glock state fields.
1345 void gfs2_glock_complete(struct gfs2_glock
*gl
, int ret
)
1347 struct lm_lockstruct
*ls
= &gl
->gl_sbd
->sd_lockstruct
;
1349 spin_lock(&gl
->gl_spin
);
1352 if (unlikely(test_bit(DFL_BLOCK_LOCKS
, &ls
->ls_recover_flags
))) {
1353 if (gfs2_should_freeze(gl
)) {
1354 set_bit(GLF_FROZEN
, &gl
->gl_flags
);
1355 spin_unlock(&gl
->gl_spin
);
1360 spin_unlock(&gl
->gl_spin
);
1361 set_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
);
1363 gfs2_glock_hold(gl
);
1364 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
1369 static int gfs2_shrink_glock_memory(struct shrinker
*shrink
,
1370 struct shrink_control
*sc
)
1372 struct gfs2_glock
*gl
;
1375 int nr
= sc
->nr_to_scan
;
1376 gfp_t gfp_mask
= sc
->gfp_mask
;
1382 if (!(gfp_mask
& __GFP_FS
))
1385 spin_lock(&lru_lock
);
1386 while(nr
&& !list_empty(&lru_list
)) {
1387 gl
= list_entry(lru_list
.next
, struct gfs2_glock
, gl_lru
);
1388 list_del_init(&gl
->gl_lru
);
1389 clear_bit(GLF_LRU
, &gl
->gl_flags
);
1390 atomic_dec(&lru_count
);
1392 /* Test for being demotable */
1393 if (!test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
)) {
1394 gfs2_glock_hold(gl
);
1395 spin_unlock(&lru_lock
);
1396 spin_lock(&gl
->gl_spin
);
1397 may_demote
= demote_ok(gl
);
1399 handle_callback(gl
, LM_ST_UNLOCKED
, 0);
1402 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
1403 smp_mb__after_clear_bit();
1404 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
1405 gfs2_glock_put_nolock(gl
);
1406 spin_unlock(&gl
->gl_spin
);
1407 spin_lock(&lru_lock
);
1411 list_add(&gl
->gl_lru
, &skipped
);
1412 set_bit(GLF_LRU
, &gl
->gl_flags
);
1414 list_splice(&skipped
, &lru_list
);
1415 atomic_add(nr_skipped
, &lru_count
);
1416 spin_unlock(&lru_lock
);
1418 return (atomic_read(&lru_count
) / 100) * sysctl_vfs_cache_pressure
;
1421 static struct shrinker glock_shrinker
= {
1422 .shrink
= gfs2_shrink_glock_memory
,
1423 .seeks
= DEFAULT_SEEKS
,
1427 * examine_bucket - Call a function for glock in a hash bucket
1428 * @examiner: the function
1429 * @sdp: the filesystem
1430 * @bucket: the bucket
1434 static void examine_bucket(glock_examiner examiner
, const struct gfs2_sbd
*sdp
,
1437 struct gfs2_glock
*gl
;
1438 struct hlist_bl_head
*head
= &gl_hash_table
[hash
];
1439 struct hlist_bl_node
*pos
;
1442 hlist_bl_for_each_entry_rcu(gl
, pos
, head
, gl_list
) {
1443 if ((gl
->gl_sbd
== sdp
) && atomic_read(&gl
->gl_ref
))
1450 static void glock_hash_walk(glock_examiner examiner
, const struct gfs2_sbd
*sdp
)
1454 for (x
= 0; x
< GFS2_GL_HASH_SIZE
; x
++)
1455 examine_bucket(examiner
, sdp
, x
);
1460 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1461 * @gl: The glock to thaw
1463 * N.B. When we freeze a glock, we leave a ref to the glock outstanding,
1464 * so this has to result in the ref count being dropped by one.
1467 static void thaw_glock(struct gfs2_glock
*gl
)
1469 if (!test_and_clear_bit(GLF_FROZEN
, &gl
->gl_flags
))
1471 set_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
);
1472 gfs2_glock_hold(gl
);
1473 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
1478 * clear_glock - look at a glock and see if we can free it from glock cache
1479 * @gl: the glock to look at
1483 static void clear_glock(struct gfs2_glock
*gl
)
1485 gfs2_glock_remove_from_lru(gl
);
1487 spin_lock(&gl
->gl_spin
);
1488 if (gl
->gl_state
!= LM_ST_UNLOCKED
)
1489 handle_callback(gl
, LM_ST_UNLOCKED
, 0);
1490 spin_unlock(&gl
->gl_spin
);
1491 gfs2_glock_hold(gl
);
1492 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
1497 * gfs2_glock_thaw - Thaw any frozen glocks
1498 * @sdp: The super block
1502 void gfs2_glock_thaw(struct gfs2_sbd
*sdp
)
1504 glock_hash_walk(thaw_glock
, sdp
);
1507 static int dump_glock(struct seq_file
*seq
, struct gfs2_glock
*gl
)
1510 spin_lock(&gl
->gl_spin
);
1511 ret
= __dump_glock(seq
, gl
);
1512 spin_unlock(&gl
->gl_spin
);
1516 static void dump_glock_func(struct gfs2_glock
*gl
)
1518 dump_glock(NULL
, gl
);
1522 * gfs2_gl_hash_clear - Empty out the glock hash table
1523 * @sdp: the filesystem
1524 * @wait: wait until it's all gone
1526 * Called when unmounting the filesystem.
1529 void gfs2_gl_hash_clear(struct gfs2_sbd
*sdp
)
1531 glock_hash_walk(clear_glock
, sdp
);
1532 flush_workqueue(glock_workqueue
);
1533 wait_event(sdp
->sd_glock_wait
, atomic_read(&sdp
->sd_glock_disposal
) == 0);
1534 glock_hash_walk(dump_glock_func
, sdp
);
1537 void gfs2_glock_finish_truncate(struct gfs2_inode
*ip
)
1539 struct gfs2_glock
*gl
= ip
->i_gl
;
1542 ret
= gfs2_truncatei_resume(ip
);
1543 gfs2_assert_withdraw(gl
->gl_sbd
, ret
== 0);
1545 spin_lock(&gl
->gl_spin
);
1546 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
1548 spin_unlock(&gl
->gl_spin
);
1551 static const char *state2str(unsigned state
)
1554 case LM_ST_UNLOCKED
:
1558 case LM_ST_DEFERRED
:
1560 case LM_ST_EXCLUSIVE
:
1566 static const char *hflags2str(char *buf
, unsigned flags
, unsigned long iflags
)
1569 if (flags
& LM_FLAG_TRY
)
1571 if (flags
& LM_FLAG_TRY_1CB
)
1573 if (flags
& LM_FLAG_NOEXP
)
1575 if (flags
& LM_FLAG_ANY
)
1577 if (flags
& LM_FLAG_PRIORITY
)
1579 if (flags
& GL_ASYNC
)
1581 if (flags
& GL_EXACT
)
1583 if (flags
& GL_NOCACHE
)
1585 if (test_bit(HIF_HOLDER
, &iflags
))
1587 if (test_bit(HIF_WAIT
, &iflags
))
1589 if (test_bit(HIF_FIRST
, &iflags
))
1596 * dump_holder - print information about a glock holder
1597 * @seq: the seq_file struct
1598 * @gh: the glock holder
1600 * Returns: 0 on success, -ENOBUFS when we run out of space
1603 static int dump_holder(struct seq_file
*seq
, const struct gfs2_holder
*gh
)
1605 struct task_struct
*gh_owner
= NULL
;
1608 if (gh
->gh_owner_pid
)
1609 gh_owner
= pid_task(gh
->gh_owner_pid
, PIDTYPE_PID
);
1610 gfs2_print_dbg(seq
, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1611 state2str(gh
->gh_state
),
1612 hflags2str(flags_buf
, gh
->gh_flags
, gh
->gh_iflags
),
1614 gh
->gh_owner_pid
? (long)pid_nr(gh
->gh_owner_pid
) : -1,
1615 gh_owner
? gh_owner
->comm
: "(ended)",
1620 static const char *gflags2str(char *buf
, const struct gfs2_glock
*gl
)
1622 const unsigned long *gflags
= &gl
->gl_flags
;
1625 if (test_bit(GLF_LOCK
, gflags
))
1627 if (test_bit(GLF_DEMOTE
, gflags
))
1629 if (test_bit(GLF_PENDING_DEMOTE
, gflags
))
1631 if (test_bit(GLF_DEMOTE_IN_PROGRESS
, gflags
))
1633 if (test_bit(GLF_DIRTY
, gflags
))
1635 if (test_bit(GLF_LFLUSH
, gflags
))
1637 if (test_bit(GLF_INVALIDATE_IN_PROGRESS
, gflags
))
1639 if (test_bit(GLF_REPLY_PENDING
, gflags
))
1641 if (test_bit(GLF_INITIAL
, gflags
))
1643 if (test_bit(GLF_FROZEN
, gflags
))
1645 if (test_bit(GLF_QUEUED
, gflags
))
1647 if (test_bit(GLF_LRU
, gflags
))
1651 if (test_bit(GLF_BLOCKING
, gflags
))
1658 * __dump_glock - print information about a glock
1659 * @seq: The seq_file struct
1662 * The file format is as follows:
1663 * One line per object, capital letters are used to indicate objects
1664 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1665 * other objects are indented by a single space and follow the glock to
1666 * which they are related. Fields are indicated by lower case letters
1667 * followed by a colon and the field value, except for strings which are in
1668 * [] so that its possible to see if they are composed of spaces for
1669 * example. The field's are n = number (id of the object), f = flags,
1670 * t = type, s = state, r = refcount, e = error, p = pid.
1672 * Returns: 0 on success, -ENOBUFS when we run out of space
1675 static int __dump_glock(struct seq_file
*seq
, const struct gfs2_glock
*gl
)
1677 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1678 unsigned long long dtime
;
1679 const struct gfs2_holder
*gh
;
1680 char gflags_buf
[32];
1683 dtime
= jiffies
- gl
->gl_demote_time
;
1684 dtime
*= 1000000/HZ
; /* demote time in uSec */
1685 if (!test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
1687 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",
1688 state2str(gl
->gl_state
),
1689 gl
->gl_name
.ln_type
,
1690 (unsigned long long)gl
->gl_name
.ln_number
,
1691 gflags2str(gflags_buf
, gl
),
1692 state2str(gl
->gl_target
),
1693 state2str(gl
->gl_demote_state
), dtime
,
1694 atomic_read(&gl
->gl_ail_count
),
1695 atomic_read(&gl
->gl_revokes
),
1696 atomic_read(&gl
->gl_ref
), gl
->gl_hold_time
);
1698 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
) {
1699 error
= dump_holder(seq
, gh
);
1703 if (gl
->gl_state
!= LM_ST_UNLOCKED
&& glops
->go_dump
)
1704 error
= glops
->go_dump(seq
, gl
);
1709 static int gfs2_glstats_seq_show(struct seq_file
*seq
, void *iter_ptr
)
1711 struct gfs2_glock
*gl
= iter_ptr
;
1713 seq_printf(seq
, "G: n:%u/%llx rtt:%lld/%lld rttb:%lld/%lld irt:%lld/%lld dcnt: %lld qcnt: %lld\n",
1714 gl
->gl_name
.ln_type
,
1715 (unsigned long long)gl
->gl_name
.ln_number
,
1716 (long long)gl
->gl_stats
.stats
[GFS2_LKS_SRTT
],
1717 (long long)gl
->gl_stats
.stats
[GFS2_LKS_SRTTVAR
],
1718 (long long)gl
->gl_stats
.stats
[GFS2_LKS_SRTTB
],
1719 (long long)gl
->gl_stats
.stats
[GFS2_LKS_SRTTVARB
],
1720 (long long)gl
->gl_stats
.stats
[GFS2_LKS_SIRT
],
1721 (long long)gl
->gl_stats
.stats
[GFS2_LKS_SIRTVAR
],
1722 (long long)gl
->gl_stats
.stats
[GFS2_LKS_DCOUNT
],
1723 (long long)gl
->gl_stats
.stats
[GFS2_LKS_QCOUNT
]);
1727 static const char *gfs2_gltype
[] = {
1741 static const char *gfs2_stype
[] = {
1742 [GFS2_LKS_SRTT
] = "srtt",
1743 [GFS2_LKS_SRTTVAR
] = "srttvar",
1744 [GFS2_LKS_SRTTB
] = "srttb",
1745 [GFS2_LKS_SRTTVARB
] = "srttvarb",
1746 [GFS2_LKS_SIRT
] = "sirt",
1747 [GFS2_LKS_SIRTVAR
] = "sirtvar",
1748 [GFS2_LKS_DCOUNT
] = "dlm",
1749 [GFS2_LKS_QCOUNT
] = "queue",
1752 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1754 static int gfs2_sbstats_seq_show(struct seq_file
*seq
, void *iter_ptr
)
1756 struct gfs2_glock_iter
*gi
= seq
->private;
1757 struct gfs2_sbd
*sdp
= gi
->sdp
;
1758 unsigned index
= gi
->hash
>> 3;
1759 unsigned subindex
= gi
->hash
& 0x07;
1763 if (index
== 0 && subindex
!= 0)
1766 seq_printf(seq
, "%-10s %8s:", gfs2_gltype
[index
],
1767 (index
== 0) ? "cpu": gfs2_stype
[subindex
]);
1769 for_each_possible_cpu(i
) {
1770 const struct gfs2_pcpu_lkstats
*lkstats
= per_cpu_ptr(sdp
->sd_lkstats
, i
);
1774 value
= lkstats
->lkstats
[index
- 1].stats
[subindex
];
1776 seq_printf(seq
, " %15lld", (long long)value
);
1778 seq_putc(seq
, '\n');
1782 int __init
gfs2_glock_init(void)
1785 for(i
= 0; i
< GFS2_GL_HASH_SIZE
; i
++) {
1786 INIT_HLIST_BL_HEAD(&gl_hash_table
[i
]);
1789 glock_workqueue
= alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM
|
1790 WQ_HIGHPRI
| WQ_FREEZABLE
, 0);
1791 if (IS_ERR(glock_workqueue
))
1792 return PTR_ERR(glock_workqueue
);
1793 gfs2_delete_workqueue
= alloc_workqueue("delete_workqueue",
1794 WQ_MEM_RECLAIM
| WQ_FREEZABLE
,
1796 if (IS_ERR(gfs2_delete_workqueue
)) {
1797 destroy_workqueue(glock_workqueue
);
1798 return PTR_ERR(gfs2_delete_workqueue
);
1801 register_shrinker(&glock_shrinker
);
1806 void gfs2_glock_exit(void)
1808 unregister_shrinker(&glock_shrinker
);
1809 destroy_workqueue(glock_workqueue
);
1810 destroy_workqueue(gfs2_delete_workqueue
);
1813 static inline struct gfs2_glock
*glock_hash_chain(unsigned hash
)
1815 return hlist_bl_entry(hlist_bl_first_rcu(&gl_hash_table
[hash
]),
1816 struct gfs2_glock
, gl_list
);
1819 static inline struct gfs2_glock
*glock_hash_next(struct gfs2_glock
*gl
)
1821 return hlist_bl_entry(rcu_dereference(gl
->gl_list
.next
),
1822 struct gfs2_glock
, gl_list
);
1825 static int gfs2_glock_iter_next(struct gfs2_glock_iter
*gi
)
1827 struct gfs2_glock
*gl
;
1832 gi
->gl
= glock_hash_next(gl
);
1835 if (gi
->hash
>= GFS2_GL_HASH_SIZE
) {
1839 gi
->gl
= glock_hash_chain(gi
->hash
);
1842 while (gi
->gl
== NULL
) {
1844 if (gi
->hash
>= GFS2_GL_HASH_SIZE
) {
1848 gi
->gl
= glock_hash_chain(gi
->hash
);
1851 /* Skip entries for other sb and dead entries */
1852 } while (gi
->sdp
!= gi
->gl
->gl_sbd
|| atomic_read(&gi
->gl
->gl_ref
) == 0);
1857 static void *gfs2_glock_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1859 struct gfs2_glock_iter
*gi
= seq
->private;
1862 if (gi
->last_pos
<= *pos
)
1863 n
= gi
->nhash
+ (*pos
- gi
->last_pos
);
1871 if (gfs2_glock_iter_next(gi
))
1875 gi
->last_pos
= *pos
;
1879 static void *gfs2_glock_seq_next(struct seq_file
*seq
, void *iter_ptr
,
1882 struct gfs2_glock_iter
*gi
= seq
->private;
1885 gi
->last_pos
= *pos
;
1886 if (gfs2_glock_iter_next(gi
))
1892 static void gfs2_glock_seq_stop(struct seq_file
*seq
, void *iter_ptr
)
1894 struct gfs2_glock_iter
*gi
= seq
->private;
1901 static int gfs2_glock_seq_show(struct seq_file
*seq
, void *iter_ptr
)
1903 return dump_glock(seq
, iter_ptr
);
1906 static void *gfs2_sbstats_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1908 struct gfs2_glock_iter
*gi
= seq
->private;
1911 if (*pos
>= GFS2_NR_SBSTATS
)
1914 return SEQ_START_TOKEN
;
1917 static void *gfs2_sbstats_seq_next(struct seq_file
*seq
, void *iter_ptr
,
1920 struct gfs2_glock_iter
*gi
= seq
->private;
1923 if (gi
->hash
>= GFS2_NR_SBSTATS
) {
1927 return SEQ_START_TOKEN
;
1930 static void gfs2_sbstats_seq_stop(struct seq_file
*seq
, void *iter_ptr
)
1935 static const struct seq_operations gfs2_glock_seq_ops
= {
1936 .start
= gfs2_glock_seq_start
,
1937 .next
= gfs2_glock_seq_next
,
1938 .stop
= gfs2_glock_seq_stop
,
1939 .show
= gfs2_glock_seq_show
,
1942 static const struct seq_operations gfs2_glstats_seq_ops
= {
1943 .start
= gfs2_glock_seq_start
,
1944 .next
= gfs2_glock_seq_next
,
1945 .stop
= gfs2_glock_seq_stop
,
1946 .show
= gfs2_glstats_seq_show
,
1949 static const struct seq_operations gfs2_sbstats_seq_ops
= {
1950 .start
= gfs2_sbstats_seq_start
,
1951 .next
= gfs2_sbstats_seq_next
,
1952 .stop
= gfs2_sbstats_seq_stop
,
1953 .show
= gfs2_sbstats_seq_show
,
1956 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
1958 static int gfs2_glocks_open(struct inode
*inode
, struct file
*file
)
1960 int ret
= seq_open_private(file
, &gfs2_glock_seq_ops
,
1961 sizeof(struct gfs2_glock_iter
));
1963 struct seq_file
*seq
= file
->private_data
;
1964 struct gfs2_glock_iter
*gi
= seq
->private;
1965 gi
->sdp
= inode
->i_private
;
1966 seq
->buf
= kmalloc(GFS2_SEQ_GOODSIZE
, GFP_KERNEL
| __GFP_NOWARN
);
1968 seq
->size
= GFS2_SEQ_GOODSIZE
;
1973 static int gfs2_glstats_open(struct inode
*inode
, struct file
*file
)
1975 int ret
= seq_open_private(file
, &gfs2_glstats_seq_ops
,
1976 sizeof(struct gfs2_glock_iter
));
1978 struct seq_file
*seq
= file
->private_data
;
1979 struct gfs2_glock_iter
*gi
= seq
->private;
1980 gi
->sdp
= inode
->i_private
;
1981 seq
->buf
= kmalloc(GFS2_SEQ_GOODSIZE
, GFP_KERNEL
| __GFP_NOWARN
);
1983 seq
->size
= GFS2_SEQ_GOODSIZE
;
1988 static int gfs2_sbstats_open(struct inode
*inode
, struct file
*file
)
1990 int ret
= seq_open_private(file
, &gfs2_sbstats_seq_ops
,
1991 sizeof(struct gfs2_glock_iter
));
1993 struct seq_file
*seq
= file
->private_data
;
1994 struct gfs2_glock_iter
*gi
= seq
->private;
1995 gi
->sdp
= inode
->i_private
;
2000 static const struct file_operations gfs2_glocks_fops
= {
2001 .owner
= THIS_MODULE
,
2002 .open
= gfs2_glocks_open
,
2004 .llseek
= seq_lseek
,
2005 .release
= seq_release_private
,
2008 static const struct file_operations gfs2_glstats_fops
= {
2009 .owner
= THIS_MODULE
,
2010 .open
= gfs2_glstats_open
,
2012 .llseek
= seq_lseek
,
2013 .release
= seq_release_private
,
2016 static const struct file_operations gfs2_sbstats_fops
= {
2017 .owner
= THIS_MODULE
,
2018 .open
= gfs2_sbstats_open
,
2020 .llseek
= seq_lseek
,
2021 .release
= seq_release_private
,
2024 int gfs2_create_debugfs_file(struct gfs2_sbd
*sdp
)
2026 sdp
->debugfs_dir
= debugfs_create_dir(sdp
->sd_table_name
, gfs2_root
);
2027 if (!sdp
->debugfs_dir
)
2029 sdp
->debugfs_dentry_glocks
= debugfs_create_file("glocks",
2031 sdp
->debugfs_dir
, sdp
,
2033 if (!sdp
->debugfs_dentry_glocks
)
2036 sdp
->debugfs_dentry_glstats
= debugfs_create_file("glstats",
2038 sdp
->debugfs_dir
, sdp
,
2039 &gfs2_glstats_fops
);
2040 if (!sdp
->debugfs_dentry_glstats
)
2043 sdp
->debugfs_dentry_sbstats
= debugfs_create_file("sbstats",
2045 sdp
->debugfs_dir
, sdp
,
2046 &gfs2_sbstats_fops
);
2047 if (!sdp
->debugfs_dentry_sbstats
)
2052 gfs2_delete_debugfs_file(sdp
);
2056 void gfs2_delete_debugfs_file(struct gfs2_sbd
*sdp
)
2058 if (sdp
->debugfs_dir
) {
2059 if (sdp
->debugfs_dentry_glocks
) {
2060 debugfs_remove(sdp
->debugfs_dentry_glocks
);
2061 sdp
->debugfs_dentry_glocks
= NULL
;
2063 if (sdp
->debugfs_dentry_glstats
) {
2064 debugfs_remove(sdp
->debugfs_dentry_glstats
);
2065 sdp
->debugfs_dentry_glstats
= NULL
;
2067 if (sdp
->debugfs_dentry_sbstats
) {
2068 debugfs_remove(sdp
->debugfs_dentry_sbstats
);
2069 sdp
->debugfs_dentry_sbstats
= NULL
;
2071 debugfs_remove(sdp
->debugfs_dir
);
2072 sdp
->debugfs_dir
= NULL
;
2076 int gfs2_register_debugfs(void)
2078 gfs2_root
= debugfs_create_dir("gfs2", NULL
);
2079 return gfs2_root
? 0 : -ENOMEM
;
2082 void gfs2_unregister_debugfs(void)
2084 debugfs_remove(gfs2_root
);