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 <linux/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>
37 #include <linux/rhashtable.h>
50 #define CREATE_TRACE_POINTS
51 #include "trace_gfs2.h"
53 struct gfs2_glock_iter
{
54 struct gfs2_sbd
*sdp
; /* incore superblock */
55 struct rhashtable_iter hti
; /* rhashtable iterator */
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 BIT(GFS2_GL_HASH_SHIFT)
74 static struct rhashtable_params ht_parms
= {
75 .nelem_hint
= GFS2_GL_HASH_SIZE
* 3 / 4,
76 .key_len
= offsetofend(struct lm_lockname
, ln_type
),
77 .key_offset
= offsetof(struct gfs2_glock
, gl_name
),
78 .head_offset
= offsetof(struct gfs2_glock
, gl_node
),
81 static struct rhashtable gl_hash_table
;
83 void gfs2_glock_free(struct gfs2_glock
*gl
)
85 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
87 if (gl
->gl_ops
->go_flags
& GLOF_ASPACE
) {
88 kmem_cache_free(gfs2_glock_aspace_cachep
, gl
);
90 kfree(gl
->gl_lksb
.sb_lvbptr
);
91 kmem_cache_free(gfs2_glock_cachep
, gl
);
93 if (atomic_dec_and_test(&sdp
->sd_glock_disposal
))
94 wake_up(&sdp
->sd_glock_wait
);
98 * gfs2_glock_hold() - increment reference count on glock
99 * @gl: The glock to hold
103 static void gfs2_glock_hold(struct gfs2_glock
*gl
)
105 GLOCK_BUG_ON(gl
, __lockref_is_dead(&gl
->gl_lockref
));
106 lockref_get(&gl
->gl_lockref
);
110 * demote_ok - Check to see if it's ok to unlock a glock
113 * Returns: 1 if it's ok
116 static int demote_ok(const struct gfs2_glock
*gl
)
118 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
120 if (gl
->gl_state
== LM_ST_UNLOCKED
)
122 if (!list_empty(&gl
->gl_holders
))
124 if (glops
->go_demote_ok
)
125 return glops
->go_demote_ok(gl
);
130 void gfs2_glock_add_to_lru(struct gfs2_glock
*gl
)
132 spin_lock(&lru_lock
);
134 if (!list_empty(&gl
->gl_lru
))
135 list_del_init(&gl
->gl_lru
);
137 atomic_inc(&lru_count
);
139 list_add_tail(&gl
->gl_lru
, &lru_list
);
140 set_bit(GLF_LRU
, &gl
->gl_flags
);
141 spin_unlock(&lru_lock
);
144 static void gfs2_glock_remove_from_lru(struct gfs2_glock
*gl
)
146 spin_lock(&lru_lock
);
147 if (!list_empty(&gl
->gl_lru
)) {
148 list_del_init(&gl
->gl_lru
);
149 atomic_dec(&lru_count
);
150 clear_bit(GLF_LRU
, &gl
->gl_flags
);
152 spin_unlock(&lru_lock
);
156 * gfs2_glock_put() - Decrement reference count on glock
157 * @gl: The glock to put
161 void gfs2_glock_put(struct gfs2_glock
*gl
)
163 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
164 struct address_space
*mapping
= gfs2_glock2aspace(gl
);
166 if (lockref_put_or_lock(&gl
->gl_lockref
))
169 lockref_mark_dead(&gl
->gl_lockref
);
171 gfs2_glock_remove_from_lru(gl
);
172 spin_unlock(&gl
->gl_lockref
.lock
);
173 rhashtable_remove_fast(&gl_hash_table
, &gl
->gl_node
, ht_parms
);
174 GLOCK_BUG_ON(gl
, !list_empty(&gl
->gl_holders
));
175 GLOCK_BUG_ON(gl
, mapping
&& mapping
->nrpages
);
176 trace_gfs2_glock_put(gl
);
177 sdp
->sd_lockstruct
.ls_ops
->lm_put_lock(gl
);
181 * may_grant - check if its ok to grant a new lock
183 * @gh: The lock request which we wish to grant
185 * Returns: true if its ok to grant the lock
188 static inline int may_grant(const struct gfs2_glock
*gl
, const struct gfs2_holder
*gh
)
190 const struct gfs2_holder
*gh_head
= list_entry(gl
->gl_holders
.next
, const struct gfs2_holder
, gh_list
);
191 if ((gh
->gh_state
== LM_ST_EXCLUSIVE
||
192 gh_head
->gh_state
== LM_ST_EXCLUSIVE
) && gh
!= gh_head
)
194 if (gl
->gl_state
== gh
->gh_state
)
196 if (gh
->gh_flags
& GL_EXACT
)
198 if (gl
->gl_state
== LM_ST_EXCLUSIVE
) {
199 if (gh
->gh_state
== LM_ST_SHARED
&& gh_head
->gh_state
== LM_ST_SHARED
)
201 if (gh
->gh_state
== LM_ST_DEFERRED
&& gh_head
->gh_state
== LM_ST_DEFERRED
)
204 if (gl
->gl_state
!= LM_ST_UNLOCKED
&& (gh
->gh_flags
& LM_FLAG_ANY
))
209 static void gfs2_holder_wake(struct gfs2_holder
*gh
)
211 clear_bit(HIF_WAIT
, &gh
->gh_iflags
);
212 smp_mb__after_atomic();
213 wake_up_bit(&gh
->gh_iflags
, HIF_WAIT
);
217 * do_error - Something unexpected has happened during a lock request
221 static void do_error(struct gfs2_glock
*gl
, const int ret
)
223 struct gfs2_holder
*gh
, *tmp
;
225 list_for_each_entry_safe(gh
, tmp
, &gl
->gl_holders
, gh_list
) {
226 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
228 if (ret
& LM_OUT_ERROR
)
230 else if (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
))
231 gh
->gh_error
= GLR_TRYFAILED
;
234 list_del_init(&gh
->gh_list
);
235 trace_gfs2_glock_queue(gh
, 0);
236 gfs2_holder_wake(gh
);
241 * do_promote - promote as many requests as possible on the current queue
244 * Returns: 1 if there is a blocked holder at the head of the list, or 2
245 * if a type specific operation is underway.
248 static int do_promote(struct gfs2_glock
*gl
)
249 __releases(&gl
->gl_lockref
.lock
)
250 __acquires(&gl
->gl_lockref
.lock
)
252 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
253 struct gfs2_holder
*gh
, *tmp
;
257 list_for_each_entry_safe(gh
, tmp
, &gl
->gl_holders
, gh_list
) {
258 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
260 if (may_grant(gl
, gh
)) {
261 if (gh
->gh_list
.prev
== &gl
->gl_holders
&&
263 spin_unlock(&gl
->gl_lockref
.lock
);
264 /* FIXME: eliminate this eventually */
265 ret
= glops
->go_lock(gh
);
266 spin_lock(&gl
->gl_lockref
.lock
);
271 list_del_init(&gh
->gh_list
);
272 trace_gfs2_glock_queue(gh
, 0);
273 gfs2_holder_wake(gh
);
276 set_bit(HIF_HOLDER
, &gh
->gh_iflags
);
277 trace_gfs2_promote(gh
, 1);
278 gfs2_holder_wake(gh
);
281 set_bit(HIF_HOLDER
, &gh
->gh_iflags
);
282 trace_gfs2_promote(gh
, 0);
283 gfs2_holder_wake(gh
);
286 if (gh
->gh_list
.prev
== &gl
->gl_holders
)
295 * find_first_waiter - find the first gh that's waiting for the glock
299 static inline struct gfs2_holder
*find_first_waiter(const struct gfs2_glock
*gl
)
301 struct gfs2_holder
*gh
;
303 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
) {
304 if (!test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
311 * state_change - record that the glock is now in a different state
313 * @new_state the new state
317 static void state_change(struct gfs2_glock
*gl
, unsigned int new_state
)
321 held1
= (gl
->gl_state
!= LM_ST_UNLOCKED
);
322 held2
= (new_state
!= LM_ST_UNLOCKED
);
324 if (held1
!= held2
) {
325 GLOCK_BUG_ON(gl
, __lockref_is_dead(&gl
->gl_lockref
));
327 gl
->gl_lockref
.count
++;
329 gl
->gl_lockref
.count
--;
331 if (held1
&& held2
&& list_empty(&gl
->gl_holders
))
332 clear_bit(GLF_QUEUED
, &gl
->gl_flags
);
334 if (new_state
!= gl
->gl_target
)
335 /* shorten our minimum hold time */
336 gl
->gl_hold_time
= max(gl
->gl_hold_time
- GL_GLOCK_HOLD_DECR
,
338 gl
->gl_state
= new_state
;
339 gl
->gl_tchange
= jiffies
;
342 static void gfs2_demote_wake(struct gfs2_glock
*gl
)
344 gl
->gl_demote_state
= LM_ST_EXCLUSIVE
;
345 clear_bit(GLF_DEMOTE
, &gl
->gl_flags
);
346 smp_mb__after_atomic();
347 wake_up_bit(&gl
->gl_flags
, GLF_DEMOTE
);
351 * finish_xmote - The DLM has replied to one of our lock requests
353 * @ret: The status from the DLM
357 static void finish_xmote(struct gfs2_glock
*gl
, unsigned int ret
)
359 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
360 struct gfs2_holder
*gh
;
361 unsigned state
= ret
& LM_OUT_ST_MASK
;
364 spin_lock(&gl
->gl_lockref
.lock
);
365 trace_gfs2_glock_state_change(gl
, state
);
366 state_change(gl
, state
);
367 gh
= find_first_waiter(gl
);
369 /* Demote to UN request arrived during demote to SH or DF */
370 if (test_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
) &&
371 state
!= LM_ST_UNLOCKED
&& gl
->gl_demote_state
== LM_ST_UNLOCKED
)
372 gl
->gl_target
= LM_ST_UNLOCKED
;
374 /* Check for state != intended state */
375 if (unlikely(state
!= gl
->gl_target
)) {
376 if (gh
&& !test_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
)) {
377 /* move to back of queue and try next entry */
378 if (ret
& LM_OUT_CANCELED
) {
379 if ((gh
->gh_flags
& LM_FLAG_PRIORITY
) == 0)
380 list_move_tail(&gh
->gh_list
, &gl
->gl_holders
);
381 gh
= find_first_waiter(gl
);
382 gl
->gl_target
= gh
->gh_state
;
385 /* Some error or failed "try lock" - report it */
386 if ((ret
& LM_OUT_ERROR
) ||
387 (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
))) {
388 gl
->gl_target
= gl
->gl_state
;
394 /* Unlocked due to conversion deadlock, try again */
397 do_xmote(gl
, gh
, gl
->gl_target
);
399 /* Conversion fails, unlock and try again */
402 do_xmote(gl
, gh
, LM_ST_UNLOCKED
);
404 default: /* Everything else */
405 pr_err("wanted %u got %u\n", gl
->gl_target
, state
);
408 spin_unlock(&gl
->gl_lockref
.lock
);
412 /* Fast path - we got what we asked for */
413 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
))
414 gfs2_demote_wake(gl
);
415 if (state
!= LM_ST_UNLOCKED
) {
416 if (glops
->go_xmote_bh
) {
417 spin_unlock(&gl
->gl_lockref
.lock
);
418 rv
= glops
->go_xmote_bh(gl
, gh
);
419 spin_lock(&gl
->gl_lockref
.lock
);
430 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
432 spin_unlock(&gl
->gl_lockref
.lock
);
436 * do_xmote - Calls the DLM to change the state of a lock
437 * @gl: The lock state
438 * @gh: The holder (only for promotes)
439 * @target: The target lock state
443 static void do_xmote(struct gfs2_glock
*gl
, struct gfs2_holder
*gh
, unsigned int target
)
444 __releases(&gl
->gl_lockref
.lock
)
445 __acquires(&gl
->gl_lockref
.lock
)
447 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
448 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
449 unsigned int lck_flags
= (unsigned int)(gh
? gh
->gh_flags
: 0);
452 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)) &&
453 target
!= LM_ST_UNLOCKED
)
455 lck_flags
&= (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
| LM_FLAG_NOEXP
|
457 GLOCK_BUG_ON(gl
, gl
->gl_state
== target
);
458 GLOCK_BUG_ON(gl
, gl
->gl_state
== gl
->gl_target
);
459 if ((target
== LM_ST_UNLOCKED
|| target
== LM_ST_DEFERRED
) &&
461 set_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
);
462 do_error(gl
, 0); /* Fail queued try locks */
465 set_bit(GLF_BLOCKING
, &gl
->gl_flags
);
466 if ((gl
->gl_req
== LM_ST_UNLOCKED
) ||
467 (gl
->gl_state
== LM_ST_EXCLUSIVE
) ||
468 (lck_flags
& (LM_FLAG_TRY
|LM_FLAG_TRY_1CB
)))
469 clear_bit(GLF_BLOCKING
, &gl
->gl_flags
);
470 spin_unlock(&gl
->gl_lockref
.lock
);
473 if (test_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
))
474 glops
->go_inval(gl
, target
== LM_ST_DEFERRED
? 0 : DIO_METADATA
);
475 clear_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
);
478 if (sdp
->sd_lockstruct
.ls_ops
->lm_lock
) {
480 ret
= sdp
->sd_lockstruct
.ls_ops
->lm_lock(gl
, target
, lck_flags
);
481 if (ret
== -EINVAL
&& gl
->gl_target
== LM_ST_UNLOCKED
&&
482 target
== LM_ST_UNLOCKED
&&
483 test_bit(SDF_SKIP_DLM_UNLOCK
, &sdp
->sd_flags
)) {
484 finish_xmote(gl
, target
);
485 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
489 pr_err("lm_lock ret %d\n", ret
);
490 GLOCK_BUG_ON(gl
, !test_bit(SDF_SHUTDOWN
,
493 } else { /* lock_nolock */
494 finish_xmote(gl
, target
);
495 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
499 spin_lock(&gl
->gl_lockref
.lock
);
503 * find_first_holder - find the first "holder" gh
507 static inline struct gfs2_holder
*find_first_holder(const struct gfs2_glock
*gl
)
509 struct gfs2_holder
*gh
;
511 if (!list_empty(&gl
->gl_holders
)) {
512 gh
= list_entry(gl
->gl_holders
.next
, struct gfs2_holder
, gh_list
);
513 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
520 * run_queue - do all outstanding tasks related to a glock
521 * @gl: The glock in question
522 * @nonblock: True if we must not block in run_queue
526 static void run_queue(struct gfs2_glock
*gl
, const int nonblock
)
527 __releases(&gl
->gl_lockref
.lock
)
528 __acquires(&gl
->gl_lockref
.lock
)
530 struct gfs2_holder
*gh
= NULL
;
533 if (test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
))
536 GLOCK_BUG_ON(gl
, test_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
));
538 if (test_bit(GLF_DEMOTE
, &gl
->gl_flags
) &&
539 gl
->gl_demote_state
!= gl
->gl_state
) {
540 if (find_first_holder(gl
))
544 set_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
);
545 GLOCK_BUG_ON(gl
, gl
->gl_demote_state
== LM_ST_EXCLUSIVE
);
546 gl
->gl_target
= gl
->gl_demote_state
;
548 if (test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
549 gfs2_demote_wake(gl
);
550 ret
= do_promote(gl
);
555 gh
= find_first_waiter(gl
);
556 gl
->gl_target
= gh
->gh_state
;
557 if (!(gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
)))
558 do_error(gl
, 0); /* Fail queued try locks */
560 do_xmote(gl
, gh
, gl
->gl_target
);
565 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
566 smp_mb__after_atomic();
567 gl
->gl_lockref
.count
++;
568 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
569 gl
->gl_lockref
.count
--;
573 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
574 smp_mb__after_atomic();
578 static void delete_work_func(struct work_struct
*work
)
580 struct gfs2_glock
*gl
= container_of(work
, struct gfs2_glock
, gl_delete
);
581 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
583 u64 no_addr
= gl
->gl_name
.ln_number
;
585 /* If someone's using this glock to create a new dinode, the block must
586 have been freed by another node, then re-used, in which case our
587 iopen callback is too late after the fact. Ignore it. */
588 if (test_bit(GLF_INODE_CREATING
, &gl
->gl_flags
))
591 inode
= gfs2_lookup_by_inum(sdp
, no_addr
, NULL
, GFS2_BLKST_UNLINKED
);
592 if (inode
&& !IS_ERR(inode
)) {
593 d_prune_aliases(inode
);
600 static void glock_work_func(struct work_struct
*work
)
602 unsigned long delay
= 0;
603 struct gfs2_glock
*gl
= container_of(work
, struct gfs2_glock
, gl_work
.work
);
606 if (test_and_clear_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
)) {
607 finish_xmote(gl
, gl
->gl_reply
);
610 spin_lock(&gl
->gl_lockref
.lock
);
611 if (test_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
) &&
612 gl
->gl_state
!= LM_ST_UNLOCKED
&&
613 gl
->gl_demote_state
!= LM_ST_EXCLUSIVE
) {
614 unsigned long holdtime
, now
= jiffies
;
616 holdtime
= gl
->gl_tchange
+ gl
->gl_hold_time
;
617 if (time_before(now
, holdtime
))
618 delay
= holdtime
- now
;
621 clear_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
);
622 set_bit(GLF_DEMOTE
, &gl
->gl_flags
);
626 spin_unlock(&gl
->gl_lockref
.lock
);
630 if (gl
->gl_name
.ln_type
!= LM_TYPE_INODE
)
632 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, delay
) == 0)
640 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
641 * @sdp: The GFS2 superblock
642 * @number: the lock number
643 * @glops: The glock_operations to use
644 * @create: If 0, don't create the glock if it doesn't exist
645 * @glp: the glock is returned here
647 * This does not lock a glock, just finds/creates structures for one.
652 int gfs2_glock_get(struct gfs2_sbd
*sdp
, u64 number
,
653 const struct gfs2_glock_operations
*glops
, int create
,
654 struct gfs2_glock
**glp
)
656 struct super_block
*s
= sdp
->sd_vfs
;
657 struct lm_lockname name
= { .ln_number
= number
,
658 .ln_type
= glops
->go_type
,
660 struct gfs2_glock
*gl
, *tmp
;
661 struct address_space
*mapping
;
662 struct kmem_cache
*cachep
;
666 gl
= rhashtable_lookup_fast(&gl_hash_table
, &name
, ht_parms
);
667 if (gl
&& !lockref_get_not_dead(&gl
->gl_lockref
))
677 if (glops
->go_flags
& GLOF_ASPACE
)
678 cachep
= gfs2_glock_aspace_cachep
;
680 cachep
= gfs2_glock_cachep
;
681 gl
= kmem_cache_alloc(cachep
, GFP_NOFS
);
685 memset(&gl
->gl_lksb
, 0, sizeof(struct dlm_lksb
));
687 if (glops
->go_flags
& GLOF_LVB
) {
688 gl
->gl_lksb
.sb_lvbptr
= kzalloc(GFS2_MIN_LVB_SIZE
, GFP_NOFS
);
689 if (!gl
->gl_lksb
.sb_lvbptr
) {
690 kmem_cache_free(cachep
, gl
);
695 atomic_inc(&sdp
->sd_glock_disposal
);
696 gl
->gl_node
.next
= NULL
;
699 gl
->gl_lockref
.count
= 1;
700 gl
->gl_state
= LM_ST_UNLOCKED
;
701 gl
->gl_target
= LM_ST_UNLOCKED
;
702 gl
->gl_demote_state
= LM_ST_EXCLUSIVE
;
706 /* We use the global stats to estimate the initial per-glock stats */
707 gl
->gl_stats
= this_cpu_ptr(sdp
->sd_lkstats
)->lkstats
[glops
->go_type
];
709 gl
->gl_stats
.stats
[GFS2_LKS_DCOUNT
] = 0;
710 gl
->gl_stats
.stats
[GFS2_LKS_QCOUNT
] = 0;
711 gl
->gl_tchange
= jiffies
;
712 gl
->gl_object
= NULL
;
713 gl
->gl_hold_time
= GL_GLOCK_DFT_HOLD
;
714 INIT_DELAYED_WORK(&gl
->gl_work
, glock_work_func
);
715 INIT_WORK(&gl
->gl_delete
, delete_work_func
);
717 mapping
= gfs2_glock2aspace(gl
);
719 mapping
->a_ops
= &gfs2_meta_aops
;
720 mapping
->host
= s
->s_bdev
->bd_inode
;
722 mapping_set_gfp_mask(mapping
, GFP_NOFS
);
723 mapping
->private_data
= NULL
;
724 mapping
->writeback_index
= 0;
729 tmp
= rhashtable_lookup_get_insert_fast(&gl_hash_table
, &gl
->gl_node
,
739 if (lockref_get_not_dead(&tmp
->gl_lockref
)) {
748 kfree(gl
->gl_lksb
.sb_lvbptr
);
749 kmem_cache_free(cachep
, gl
);
750 atomic_dec(&sdp
->sd_glock_disposal
);
758 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
760 * @state: the state we're requesting
761 * @flags: the modifier flags
762 * @gh: the holder structure
766 void gfs2_holder_init(struct gfs2_glock
*gl
, unsigned int state
, u16 flags
,
767 struct gfs2_holder
*gh
)
769 INIT_LIST_HEAD(&gh
->gh_list
);
771 gh
->gh_ip
= _RET_IP_
;
772 gh
->gh_owner_pid
= get_pid(task_pid(current
));
773 gh
->gh_state
= state
;
774 gh
->gh_flags
= flags
;
781 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
782 * @state: the state we're requesting
783 * @flags: the modifier flags
784 * @gh: the holder structure
786 * Don't mess with the glock.
790 void gfs2_holder_reinit(unsigned int state
, u16 flags
, struct gfs2_holder
*gh
)
792 gh
->gh_state
= state
;
793 gh
->gh_flags
= flags
;
795 gh
->gh_ip
= _RET_IP_
;
796 put_pid(gh
->gh_owner_pid
);
797 gh
->gh_owner_pid
= get_pid(task_pid(current
));
801 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
802 * @gh: the holder structure
806 void gfs2_holder_uninit(struct gfs2_holder
*gh
)
808 put_pid(gh
->gh_owner_pid
);
809 gfs2_glock_put(gh
->gh_gl
);
810 gfs2_holder_mark_uninitialized(gh
);
815 * gfs2_glock_wait - wait on a glock acquisition
816 * @gh: the glock holder
818 * Returns: 0 on success
821 int gfs2_glock_wait(struct gfs2_holder
*gh
)
823 unsigned long time1
= jiffies
;
826 wait_on_bit(&gh
->gh_iflags
, HIF_WAIT
, TASK_UNINTERRUPTIBLE
);
827 if (time_after(jiffies
, time1
+ HZ
)) /* have we waited > a second? */
828 /* Lengthen the minimum hold time. */
829 gh
->gh_gl
->gl_hold_time
= min(gh
->gh_gl
->gl_hold_time
+
836 * handle_callback - process a demote request
838 * @state: the state the caller wants us to change to
840 * There are only two requests that we are going to see in actual
841 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
844 static void handle_callback(struct gfs2_glock
*gl
, unsigned int state
,
845 unsigned long delay
, bool remote
)
847 int bit
= delay
? GLF_PENDING_DEMOTE
: GLF_DEMOTE
;
849 set_bit(bit
, &gl
->gl_flags
);
850 if (gl
->gl_demote_state
== LM_ST_EXCLUSIVE
) {
851 gl
->gl_demote_state
= state
;
852 gl
->gl_demote_time
= jiffies
;
853 } else if (gl
->gl_demote_state
!= LM_ST_UNLOCKED
&&
854 gl
->gl_demote_state
!= state
) {
855 gl
->gl_demote_state
= LM_ST_UNLOCKED
;
857 if (gl
->gl_ops
->go_callback
)
858 gl
->gl_ops
->go_callback(gl
, remote
);
859 trace_gfs2_demote_rq(gl
, remote
);
862 void gfs2_print_dbg(struct seq_file
*seq
, const char *fmt
, ...)
864 struct va_format vaf
;
870 seq_vprintf(seq
, fmt
, args
);
882 * add_to_queue - Add a holder to the wait queue (but look for recursion)
883 * @gh: the holder structure to add
885 * Eventually we should move the recursive locking trap to a
886 * debugging option or something like that. This is the fast
887 * path and needs to have the minimum number of distractions.
891 static inline void add_to_queue(struct gfs2_holder
*gh
)
892 __releases(&gl
->gl_lockref
.lock
)
893 __acquires(&gl
->gl_lockref
.lock
)
895 struct gfs2_glock
*gl
= gh
->gh_gl
;
896 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
897 struct list_head
*insert_pt
= NULL
;
898 struct gfs2_holder
*gh2
;
901 BUG_ON(gh
->gh_owner_pid
== NULL
);
902 if (test_and_set_bit(HIF_WAIT
, &gh
->gh_iflags
))
905 if (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
)) {
906 if (test_bit(GLF_LOCK
, &gl
->gl_flags
))
907 try_futile
= !may_grant(gl
, gh
);
908 if (test_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
))
912 list_for_each_entry(gh2
, &gl
->gl_holders
, gh_list
) {
913 if (unlikely(gh2
->gh_owner_pid
== gh
->gh_owner_pid
&&
914 (gh
->gh_gl
->gl_ops
->go_type
!= LM_TYPE_FLOCK
)))
917 !(gh2
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
))) {
919 gh
->gh_error
= GLR_TRYFAILED
;
920 gfs2_holder_wake(gh
);
923 if (test_bit(HIF_HOLDER
, &gh2
->gh_iflags
))
925 if (unlikely((gh
->gh_flags
& LM_FLAG_PRIORITY
) && !insert_pt
))
926 insert_pt
= &gh2
->gh_list
;
928 set_bit(GLF_QUEUED
, &gl
->gl_flags
);
929 trace_gfs2_glock_queue(gh
, 1);
930 gfs2_glstats_inc(gl
, GFS2_LKS_QCOUNT
);
931 gfs2_sbstats_inc(gl
, GFS2_LKS_QCOUNT
);
932 if (likely(insert_pt
== NULL
)) {
933 list_add_tail(&gh
->gh_list
, &gl
->gl_holders
);
934 if (unlikely(gh
->gh_flags
& LM_FLAG_PRIORITY
))
938 list_add_tail(&gh
->gh_list
, insert_pt
);
940 gh
= list_entry(gl
->gl_holders
.next
, struct gfs2_holder
, gh_list
);
941 if (!(gh
->gh_flags
& LM_FLAG_PRIORITY
)) {
942 spin_unlock(&gl
->gl_lockref
.lock
);
943 if (sdp
->sd_lockstruct
.ls_ops
->lm_cancel
)
944 sdp
->sd_lockstruct
.ls_ops
->lm_cancel(gl
);
945 spin_lock(&gl
->gl_lockref
.lock
);
950 pr_err("original: %pSR\n", (void *)gh2
->gh_ip
);
951 pr_err("pid: %d\n", pid_nr(gh2
->gh_owner_pid
));
952 pr_err("lock type: %d req lock state : %d\n",
953 gh2
->gh_gl
->gl_name
.ln_type
, gh2
->gh_state
);
954 pr_err("new: %pSR\n", (void *)gh
->gh_ip
);
955 pr_err("pid: %d\n", pid_nr(gh
->gh_owner_pid
));
956 pr_err("lock type: %d req lock state : %d\n",
957 gh
->gh_gl
->gl_name
.ln_type
, gh
->gh_state
);
958 gfs2_dump_glock(NULL
, gl
);
963 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
964 * @gh: the holder structure
966 * if (gh->gh_flags & GL_ASYNC), this never returns an error
968 * Returns: 0, GLR_TRYFAILED, or errno on failure
971 int gfs2_glock_nq(struct gfs2_holder
*gh
)
973 struct gfs2_glock
*gl
= gh
->gh_gl
;
974 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
977 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)))
980 if (test_bit(GLF_LRU
, &gl
->gl_flags
))
981 gfs2_glock_remove_from_lru(gl
);
983 spin_lock(&gl
->gl_lockref
.lock
);
985 if (unlikely((LM_FLAG_NOEXP
& gh
->gh_flags
) &&
986 test_and_clear_bit(GLF_FROZEN
, &gl
->gl_flags
))) {
987 set_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
);
988 gl
->gl_lockref
.count
++;
989 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
990 gl
->gl_lockref
.count
--;
993 spin_unlock(&gl
->gl_lockref
.lock
);
995 if (!(gh
->gh_flags
& GL_ASYNC
))
996 error
= gfs2_glock_wait(gh
);
1002 * gfs2_glock_poll - poll to see if an async request has been completed
1005 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1008 int gfs2_glock_poll(struct gfs2_holder
*gh
)
1010 return test_bit(HIF_WAIT
, &gh
->gh_iflags
) ? 0 : 1;
1014 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1015 * @gh: the glock holder
1019 void gfs2_glock_dq(struct gfs2_holder
*gh
)
1021 struct gfs2_glock
*gl
= gh
->gh_gl
;
1022 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1026 spin_lock(&gl
->gl_lockref
.lock
);
1027 if (gh
->gh_flags
& GL_NOCACHE
)
1028 handle_callback(gl
, LM_ST_UNLOCKED
, 0, false);
1030 list_del_init(&gh
->gh_list
);
1031 clear_bit(HIF_HOLDER
, &gh
->gh_iflags
);
1032 if (find_first_holder(gl
) == NULL
) {
1033 if (glops
->go_unlock
) {
1034 GLOCK_BUG_ON(gl
, test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
));
1035 spin_unlock(&gl
->gl_lockref
.lock
);
1036 glops
->go_unlock(gh
);
1037 spin_lock(&gl
->gl_lockref
.lock
);
1038 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
1040 if (list_empty(&gl
->gl_holders
) &&
1041 !test_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
) &&
1042 !test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
1045 if (!test_bit(GLF_LFLUSH
, &gl
->gl_flags
) && demote_ok(gl
) &&
1046 (glops
->go_flags
& GLOF_LRU
))
1047 gfs2_glock_add_to_lru(gl
);
1049 trace_gfs2_glock_queue(gh
, 0);
1050 spin_unlock(&gl
->gl_lockref
.lock
);
1051 if (likely(fast_path
))
1054 gfs2_glock_hold(gl
);
1055 if (test_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
) &&
1056 !test_bit(GLF_DEMOTE
, &gl
->gl_flags
) &&
1057 gl
->gl_name
.ln_type
== LM_TYPE_INODE
)
1058 delay
= gl
->gl_hold_time
;
1059 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, delay
) == 0)
1063 void gfs2_glock_dq_wait(struct gfs2_holder
*gh
)
1065 struct gfs2_glock
*gl
= gh
->gh_gl
;
1068 wait_on_bit(&gl
->gl_flags
, GLF_DEMOTE
, TASK_UNINTERRUPTIBLE
);
1072 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1073 * @gh: the holder structure
1077 void gfs2_glock_dq_uninit(struct gfs2_holder
*gh
)
1080 gfs2_holder_uninit(gh
);
1084 * gfs2_glock_nq_num - acquire a glock based on lock number
1085 * @sdp: the filesystem
1086 * @number: the lock number
1087 * @glops: the glock operations for the type of glock
1088 * @state: the state to acquire the glock in
1089 * @flags: modifier flags for the acquisition
1090 * @gh: the struct gfs2_holder
1095 int gfs2_glock_nq_num(struct gfs2_sbd
*sdp
, u64 number
,
1096 const struct gfs2_glock_operations
*glops
,
1097 unsigned int state
, u16 flags
, struct gfs2_holder
*gh
)
1099 struct gfs2_glock
*gl
;
1102 error
= gfs2_glock_get(sdp
, number
, glops
, CREATE
, &gl
);
1104 error
= gfs2_glock_nq_init(gl
, state
, flags
, gh
);
1112 * glock_compare - Compare two struct gfs2_glock structures for sorting
1113 * @arg_a: the first structure
1114 * @arg_b: the second structure
1118 static int glock_compare(const void *arg_a
, const void *arg_b
)
1120 const struct gfs2_holder
*gh_a
= *(const struct gfs2_holder
**)arg_a
;
1121 const struct gfs2_holder
*gh_b
= *(const struct gfs2_holder
**)arg_b
;
1122 const struct lm_lockname
*a
= &gh_a
->gh_gl
->gl_name
;
1123 const struct lm_lockname
*b
= &gh_b
->gh_gl
->gl_name
;
1125 if (a
->ln_number
> b
->ln_number
)
1127 if (a
->ln_number
< b
->ln_number
)
1129 BUG_ON(gh_a
->gh_gl
->gl_ops
->go_type
== gh_b
->gh_gl
->gl_ops
->go_type
);
1134 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1135 * @num_gh: the number of structures
1136 * @ghs: an array of struct gfs2_holder structures
1138 * Returns: 0 on success (all glocks acquired),
1139 * errno on failure (no glocks acquired)
1142 static int nq_m_sync(unsigned int num_gh
, struct gfs2_holder
*ghs
,
1143 struct gfs2_holder
**p
)
1148 for (x
= 0; x
< num_gh
; x
++)
1151 sort(p
, num_gh
, sizeof(struct gfs2_holder
*), glock_compare
, NULL
);
1153 for (x
= 0; x
< num_gh
; x
++) {
1154 p
[x
]->gh_flags
&= ~(LM_FLAG_TRY
| GL_ASYNC
);
1156 error
= gfs2_glock_nq(p
[x
]);
1159 gfs2_glock_dq(p
[x
]);
1168 * gfs2_glock_nq_m - acquire multiple glocks
1169 * @num_gh: the number of structures
1170 * @ghs: an array of struct gfs2_holder structures
1173 * Returns: 0 on success (all glocks acquired),
1174 * errno on failure (no glocks acquired)
1177 int gfs2_glock_nq_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1179 struct gfs2_holder
*tmp
[4];
1180 struct gfs2_holder
**pph
= tmp
;
1187 ghs
->gh_flags
&= ~(LM_FLAG_TRY
| GL_ASYNC
);
1188 return gfs2_glock_nq(ghs
);
1192 pph
= kmalloc(num_gh
* sizeof(struct gfs2_holder
*), GFP_NOFS
);
1197 error
= nq_m_sync(num_gh
, ghs
, pph
);
1206 * gfs2_glock_dq_m - release multiple glocks
1207 * @num_gh: the number of structures
1208 * @ghs: an array of struct gfs2_holder structures
1212 void gfs2_glock_dq_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1215 gfs2_glock_dq(&ghs
[num_gh
]);
1218 void gfs2_glock_cb(struct gfs2_glock
*gl
, unsigned int state
)
1220 unsigned long delay
= 0;
1221 unsigned long holdtime
;
1222 unsigned long now
= jiffies
;
1224 gfs2_glock_hold(gl
);
1225 holdtime
= gl
->gl_tchange
+ gl
->gl_hold_time
;
1226 if (test_bit(GLF_QUEUED
, &gl
->gl_flags
) &&
1227 gl
->gl_name
.ln_type
== LM_TYPE_INODE
) {
1228 if (time_before(now
, holdtime
))
1229 delay
= holdtime
- now
;
1230 if (test_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
))
1231 delay
= gl
->gl_hold_time
;
1234 spin_lock(&gl
->gl_lockref
.lock
);
1235 handle_callback(gl
, state
, delay
, true);
1236 spin_unlock(&gl
->gl_lockref
.lock
);
1237 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, delay
) == 0)
1242 * gfs2_should_freeze - Figure out if glock should be frozen
1243 * @gl: The glock in question
1245 * Glocks are not frozen if (a) the result of the dlm operation is
1246 * an error, (b) the locking operation was an unlock operation or
1247 * (c) if there is a "noexp" flagged request anywhere in the queue
1249 * Returns: 1 if freezing should occur, 0 otherwise
1252 static int gfs2_should_freeze(const struct gfs2_glock
*gl
)
1254 const struct gfs2_holder
*gh
;
1256 if (gl
->gl_reply
& ~LM_OUT_ST_MASK
)
1258 if (gl
->gl_target
== LM_ST_UNLOCKED
)
1261 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
) {
1262 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
1264 if (LM_FLAG_NOEXP
& gh
->gh_flags
)
1272 * gfs2_glock_complete - Callback used by locking
1273 * @gl: Pointer to the glock
1274 * @ret: The return value from the dlm
1276 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1277 * to use a bitfield shared with other glock state fields.
1280 void gfs2_glock_complete(struct gfs2_glock
*gl
, int ret
)
1282 struct lm_lockstruct
*ls
= &gl
->gl_name
.ln_sbd
->sd_lockstruct
;
1284 spin_lock(&gl
->gl_lockref
.lock
);
1287 if (unlikely(test_bit(DFL_BLOCK_LOCKS
, &ls
->ls_recover_flags
))) {
1288 if (gfs2_should_freeze(gl
)) {
1289 set_bit(GLF_FROZEN
, &gl
->gl_flags
);
1290 spin_unlock(&gl
->gl_lockref
.lock
);
1295 gl
->gl_lockref
.count
++;
1296 set_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
);
1297 spin_unlock(&gl
->gl_lockref
.lock
);
1299 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
1303 static int glock_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
1305 struct gfs2_glock
*gla
, *glb
;
1307 gla
= list_entry(a
, struct gfs2_glock
, gl_lru
);
1308 glb
= list_entry(b
, struct gfs2_glock
, gl_lru
);
1310 if (gla
->gl_name
.ln_number
> glb
->gl_name
.ln_number
)
1312 if (gla
->gl_name
.ln_number
< glb
->gl_name
.ln_number
)
1319 * gfs2_dispose_glock_lru - Demote a list of glocks
1320 * @list: The list to dispose of
1322 * Disposing of glocks may involve disk accesses, so that here we sort
1323 * the glocks by number (i.e. disk location of the inodes) so that if
1324 * there are any such accesses, they'll be sent in order (mostly).
1326 * Must be called under the lru_lock, but may drop and retake this
1327 * lock. While the lru_lock is dropped, entries may vanish from the
1328 * list, but no new entries will appear on the list (since it is
1332 static void gfs2_dispose_glock_lru(struct list_head
*list
)
1333 __releases(&lru_lock
)
1334 __acquires(&lru_lock
)
1336 struct gfs2_glock
*gl
;
1338 list_sort(NULL
, list
, glock_cmp
);
1340 while(!list_empty(list
)) {
1341 gl
= list_entry(list
->next
, struct gfs2_glock
, gl_lru
);
1342 list_del_init(&gl
->gl_lru
);
1343 if (!spin_trylock(&gl
->gl_lockref
.lock
)) {
1345 list_add(&gl
->gl_lru
, &lru_list
);
1346 atomic_inc(&lru_count
);
1349 if (test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
)) {
1350 spin_unlock(&gl
->gl_lockref
.lock
);
1351 goto add_back_to_lru
;
1353 clear_bit(GLF_LRU
, &gl
->gl_flags
);
1354 gl
->gl_lockref
.count
++;
1356 handle_callback(gl
, LM_ST_UNLOCKED
, 0, false);
1357 WARN_ON(!test_and_clear_bit(GLF_LOCK
, &gl
->gl_flags
));
1358 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
1359 gl
->gl_lockref
.count
--;
1360 spin_unlock(&gl
->gl_lockref
.lock
);
1361 cond_resched_lock(&lru_lock
);
1366 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1367 * @nr: The number of entries to scan
1369 * This function selects the entries on the LRU which are able to
1370 * be demoted, and then kicks off the process by calling
1371 * gfs2_dispose_glock_lru() above.
1374 static long gfs2_scan_glock_lru(int nr
)
1376 struct gfs2_glock
*gl
;
1381 spin_lock(&lru_lock
);
1382 while ((nr
-- >= 0) && !list_empty(&lru_list
)) {
1383 gl
= list_entry(lru_list
.next
, struct gfs2_glock
, gl_lru
);
1385 /* Test for being demotable */
1386 if (!test_bit(GLF_LOCK
, &gl
->gl_flags
)) {
1387 list_move(&gl
->gl_lru
, &dispose
);
1388 atomic_dec(&lru_count
);
1393 list_move(&gl
->gl_lru
, &skipped
);
1395 list_splice(&skipped
, &lru_list
);
1396 if (!list_empty(&dispose
))
1397 gfs2_dispose_glock_lru(&dispose
);
1398 spin_unlock(&lru_lock
);
1403 static unsigned long gfs2_glock_shrink_scan(struct shrinker
*shrink
,
1404 struct shrink_control
*sc
)
1406 if (!(sc
->gfp_mask
& __GFP_FS
))
1408 return gfs2_scan_glock_lru(sc
->nr_to_scan
);
1411 static unsigned long gfs2_glock_shrink_count(struct shrinker
*shrink
,
1412 struct shrink_control
*sc
)
1414 return vfs_pressure_ratio(atomic_read(&lru_count
));
1417 static struct shrinker glock_shrinker
= {
1418 .seeks
= DEFAULT_SEEKS
,
1419 .count_objects
= gfs2_glock_shrink_count
,
1420 .scan_objects
= gfs2_glock_shrink_scan
,
1424 * examine_bucket - Call a function for glock in a hash bucket
1425 * @examiner: the function
1426 * @sdp: the filesystem
1427 * @bucket: the bucket
1429 * Note that the function can be called multiple times on the same
1430 * object. So the user must ensure that the function can cope with
1434 static void glock_hash_walk(glock_examiner examiner
, const struct gfs2_sbd
*sdp
)
1436 struct gfs2_glock
*gl
;
1437 struct rhashtable_iter iter
;
1439 rhashtable_walk_enter(&gl_hash_table
, &iter
);
1442 gl
= ERR_PTR(rhashtable_walk_start(&iter
));
1446 while ((gl
= rhashtable_walk_next(&iter
)) && !IS_ERR(gl
))
1447 if ((gl
->gl_name
.ln_sbd
== sdp
) &&
1448 lockref_get_not_dead(&gl
->gl_lockref
))
1451 rhashtable_walk_stop(&iter
);
1452 } while (cond_resched(), gl
== ERR_PTR(-EAGAIN
));
1454 rhashtable_walk_exit(&iter
);
1458 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1459 * @gl: The glock to thaw
1463 static void thaw_glock(struct gfs2_glock
*gl
)
1465 if (!test_and_clear_bit(GLF_FROZEN
, &gl
->gl_flags
))
1467 set_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
);
1468 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0) {
1475 * clear_glock - look at a glock and see if we can free it from glock cache
1476 * @gl: the glock to look at
1480 static void clear_glock(struct gfs2_glock
*gl
)
1482 gfs2_glock_remove_from_lru(gl
);
1484 spin_lock(&gl
->gl_lockref
.lock
);
1485 if (gl
->gl_state
!= LM_ST_UNLOCKED
)
1486 handle_callback(gl
, LM_ST_UNLOCKED
, 0, false);
1487 spin_unlock(&gl
->gl_lockref
.lock
);
1488 if (queue_delayed_work(glock_workqueue
, &gl
->gl_work
, 0) == 0)
1493 * gfs2_glock_thaw - Thaw any frozen glocks
1494 * @sdp: The super block
1498 void gfs2_glock_thaw(struct gfs2_sbd
*sdp
)
1500 glock_hash_walk(thaw_glock
, sdp
);
1503 static void dump_glock(struct seq_file
*seq
, struct gfs2_glock
*gl
)
1505 spin_lock(&gl
->gl_lockref
.lock
);
1506 gfs2_dump_glock(seq
, gl
);
1507 spin_unlock(&gl
->gl_lockref
.lock
);
1510 static void dump_glock_func(struct gfs2_glock
*gl
)
1512 dump_glock(NULL
, gl
);
1516 * gfs2_gl_hash_clear - Empty out the glock hash table
1517 * @sdp: the filesystem
1518 * @wait: wait until it's all gone
1520 * Called when unmounting the filesystem.
1523 void gfs2_gl_hash_clear(struct gfs2_sbd
*sdp
)
1525 set_bit(SDF_SKIP_DLM_UNLOCK
, &sdp
->sd_flags
);
1526 flush_workqueue(glock_workqueue
);
1527 glock_hash_walk(clear_glock
, sdp
);
1528 flush_workqueue(glock_workqueue
);
1529 wait_event_timeout(sdp
->sd_glock_wait
,
1530 atomic_read(&sdp
->sd_glock_disposal
) == 0,
1532 glock_hash_walk(dump_glock_func
, sdp
);
1535 void gfs2_glock_finish_truncate(struct gfs2_inode
*ip
)
1537 struct gfs2_glock
*gl
= ip
->i_gl
;
1540 ret
= gfs2_truncatei_resume(ip
);
1541 gfs2_assert_withdraw(gl
->gl_name
.ln_sbd
, ret
== 0);
1543 spin_lock(&gl
->gl_lockref
.lock
);
1544 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
1546 spin_unlock(&gl
->gl_lockref
.lock
);
1549 static const char *state2str(unsigned state
)
1552 case LM_ST_UNLOCKED
:
1556 case LM_ST_DEFERRED
:
1558 case LM_ST_EXCLUSIVE
:
1564 static const char *hflags2str(char *buf
, u16 flags
, unsigned long iflags
)
1567 if (flags
& LM_FLAG_TRY
)
1569 if (flags
& LM_FLAG_TRY_1CB
)
1571 if (flags
& LM_FLAG_NOEXP
)
1573 if (flags
& LM_FLAG_ANY
)
1575 if (flags
& LM_FLAG_PRIORITY
)
1577 if (flags
& GL_ASYNC
)
1579 if (flags
& GL_EXACT
)
1581 if (flags
& GL_NOCACHE
)
1583 if (test_bit(HIF_HOLDER
, &iflags
))
1585 if (test_bit(HIF_WAIT
, &iflags
))
1587 if (test_bit(HIF_FIRST
, &iflags
))
1594 * dump_holder - print information about a glock holder
1595 * @seq: the seq_file struct
1596 * @gh: the glock holder
1600 static void dump_holder(struct seq_file
*seq
, const struct gfs2_holder
*gh
)
1602 struct task_struct
*gh_owner
= NULL
;
1606 if (gh
->gh_owner_pid
)
1607 gh_owner
= pid_task(gh
->gh_owner_pid
, PIDTYPE_PID
);
1608 gfs2_print_dbg(seq
, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1609 state2str(gh
->gh_state
),
1610 hflags2str(flags_buf
, gh
->gh_flags
, gh
->gh_iflags
),
1612 gh
->gh_owner_pid
? (long)pid_nr(gh
->gh_owner_pid
) : -1,
1613 gh_owner
? gh_owner
->comm
: "(ended)",
1618 static const char *gflags2str(char *buf
, const struct gfs2_glock
*gl
)
1620 const unsigned long *gflags
= &gl
->gl_flags
;
1623 if (test_bit(GLF_LOCK
, gflags
))
1625 if (test_bit(GLF_DEMOTE
, gflags
))
1627 if (test_bit(GLF_PENDING_DEMOTE
, gflags
))
1629 if (test_bit(GLF_DEMOTE_IN_PROGRESS
, gflags
))
1631 if (test_bit(GLF_DIRTY
, gflags
))
1633 if (test_bit(GLF_LFLUSH
, gflags
))
1635 if (test_bit(GLF_INVALIDATE_IN_PROGRESS
, gflags
))
1637 if (test_bit(GLF_REPLY_PENDING
, gflags
))
1639 if (test_bit(GLF_INITIAL
, gflags
))
1641 if (test_bit(GLF_FROZEN
, gflags
))
1643 if (test_bit(GLF_QUEUED
, gflags
))
1645 if (test_bit(GLF_LRU
, gflags
))
1649 if (test_bit(GLF_BLOCKING
, gflags
))
1656 * gfs2_dump_glock - print information about a glock
1657 * @seq: The seq_file struct
1660 * The file format is as follows:
1661 * One line per object, capital letters are used to indicate objects
1662 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1663 * other objects are indented by a single space and follow the glock to
1664 * which they are related. Fields are indicated by lower case letters
1665 * followed by a colon and the field value, except for strings which are in
1666 * [] so that its possible to see if they are composed of spaces for
1667 * example. The field's are n = number (id of the object), f = flags,
1668 * t = type, s = state, r = refcount, e = error, p = pid.
1672 void gfs2_dump_glock(struct seq_file
*seq
, const struct gfs2_glock
*gl
)
1674 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1675 unsigned long long dtime
;
1676 const struct gfs2_holder
*gh
;
1677 char gflags_buf
[32];
1679 dtime
= jiffies
- gl
->gl_demote_time
;
1680 dtime
*= 1000000/HZ
; /* demote time in uSec */
1681 if (!test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
1683 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",
1684 state2str(gl
->gl_state
),
1685 gl
->gl_name
.ln_type
,
1686 (unsigned long long)gl
->gl_name
.ln_number
,
1687 gflags2str(gflags_buf
, gl
),
1688 state2str(gl
->gl_target
),
1689 state2str(gl
->gl_demote_state
), dtime
,
1690 atomic_read(&gl
->gl_ail_count
),
1691 atomic_read(&gl
->gl_revokes
),
1692 (int)gl
->gl_lockref
.count
, gl
->gl_hold_time
);
1694 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
)
1695 dump_holder(seq
, gh
);
1697 if (gl
->gl_state
!= LM_ST_UNLOCKED
&& glops
->go_dump
)
1698 glops
->go_dump(seq
, gl
);
1701 static int gfs2_glstats_seq_show(struct seq_file
*seq
, void *iter_ptr
)
1703 struct gfs2_glock
*gl
= iter_ptr
;
1705 seq_printf(seq
, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
1706 gl
->gl_name
.ln_type
,
1707 (unsigned long long)gl
->gl_name
.ln_number
,
1708 (unsigned long long)gl
->gl_stats
.stats
[GFS2_LKS_SRTT
],
1709 (unsigned long long)gl
->gl_stats
.stats
[GFS2_LKS_SRTTVAR
],
1710 (unsigned long long)gl
->gl_stats
.stats
[GFS2_LKS_SRTTB
],
1711 (unsigned long long)gl
->gl_stats
.stats
[GFS2_LKS_SRTTVARB
],
1712 (unsigned long long)gl
->gl_stats
.stats
[GFS2_LKS_SIRT
],
1713 (unsigned long long)gl
->gl_stats
.stats
[GFS2_LKS_SIRTVAR
],
1714 (unsigned long long)gl
->gl_stats
.stats
[GFS2_LKS_DCOUNT
],
1715 (unsigned long long)gl
->gl_stats
.stats
[GFS2_LKS_QCOUNT
]);
1719 static const char *gfs2_gltype
[] = {
1733 static const char *gfs2_stype
[] = {
1734 [GFS2_LKS_SRTT
] = "srtt",
1735 [GFS2_LKS_SRTTVAR
] = "srttvar",
1736 [GFS2_LKS_SRTTB
] = "srttb",
1737 [GFS2_LKS_SRTTVARB
] = "srttvarb",
1738 [GFS2_LKS_SIRT
] = "sirt",
1739 [GFS2_LKS_SIRTVAR
] = "sirtvar",
1740 [GFS2_LKS_DCOUNT
] = "dlm",
1741 [GFS2_LKS_QCOUNT
] = "queue",
1744 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1746 static int gfs2_sbstats_seq_show(struct seq_file
*seq
, void *iter_ptr
)
1748 struct gfs2_sbd
*sdp
= seq
->private;
1749 loff_t pos
= *(loff_t
*)iter_ptr
;
1750 unsigned index
= pos
>> 3;
1751 unsigned subindex
= pos
& 0x07;
1754 if (index
== 0 && subindex
!= 0)
1757 seq_printf(seq
, "%-10s %8s:", gfs2_gltype
[index
],
1758 (index
== 0) ? "cpu": gfs2_stype
[subindex
]);
1760 for_each_possible_cpu(i
) {
1761 const struct gfs2_pcpu_lkstats
*lkstats
= per_cpu_ptr(sdp
->sd_lkstats
, i
);
1764 seq_printf(seq
, " %15u", i
);
1766 seq_printf(seq
, " %15llu", (unsigned long long)lkstats
->
1767 lkstats
[index
- 1].stats
[subindex
]);
1769 seq_putc(seq
, '\n');
1773 int __init
gfs2_glock_init(void)
1777 ret
= rhashtable_init(&gl_hash_table
, &ht_parms
);
1781 glock_workqueue
= alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM
|
1782 WQ_HIGHPRI
| WQ_FREEZABLE
, 0);
1783 if (!glock_workqueue
) {
1784 rhashtable_destroy(&gl_hash_table
);
1787 gfs2_delete_workqueue
= alloc_workqueue("delete_workqueue",
1788 WQ_MEM_RECLAIM
| WQ_FREEZABLE
,
1790 if (!gfs2_delete_workqueue
) {
1791 destroy_workqueue(glock_workqueue
);
1792 rhashtable_destroy(&gl_hash_table
);
1796 ret
= register_shrinker(&glock_shrinker
);
1798 destroy_workqueue(gfs2_delete_workqueue
);
1799 destroy_workqueue(glock_workqueue
);
1800 rhashtable_destroy(&gl_hash_table
);
1807 void gfs2_glock_exit(void)
1809 unregister_shrinker(&glock_shrinker
);
1810 rhashtable_destroy(&gl_hash_table
);
1811 destroy_workqueue(glock_workqueue
);
1812 destroy_workqueue(gfs2_delete_workqueue
);
1815 static void gfs2_glock_iter_next(struct gfs2_glock_iter
*gi
)
1817 while ((gi
->gl
= rhashtable_walk_next(&gi
->hti
))) {
1818 if (IS_ERR(gi
->gl
)) {
1819 if (PTR_ERR(gi
->gl
) == -EAGAIN
)
1824 /* Skip entries for other sb and dead entries */
1825 if (gi
->sdp
== gi
->gl
->gl_name
.ln_sbd
&&
1826 !__lockref_is_dead(&gi
->gl
->gl_lockref
))
1831 static void *gfs2_glock_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1833 struct gfs2_glock_iter
*gi
= seq
->private;
1837 if (gi
->last_pos
<= *pos
)
1838 n
= (*pos
- gi
->last_pos
);
1840 ret
= rhashtable_walk_start(&gi
->hti
);
1845 gfs2_glock_iter_next(gi
);
1846 } while (gi
->gl
&& n
--);
1848 gi
->last_pos
= *pos
;
1852 static void *gfs2_glock_seq_next(struct seq_file
*seq
, void *iter_ptr
,
1855 struct gfs2_glock_iter
*gi
= seq
->private;
1858 gi
->last_pos
= *pos
;
1859 gfs2_glock_iter_next(gi
);
1863 static void gfs2_glock_seq_stop(struct seq_file
*seq
, void *iter_ptr
)
1865 struct gfs2_glock_iter
*gi
= seq
->private;
1868 rhashtable_walk_stop(&gi
->hti
);
1871 static int gfs2_glock_seq_show(struct seq_file
*seq
, void *iter_ptr
)
1873 dump_glock(seq
, iter_ptr
);
1877 static void *gfs2_sbstats_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1880 if (*pos
>= GFS2_NR_SBSTATS
)
1885 static void *gfs2_sbstats_seq_next(struct seq_file
*seq
, void *iter_ptr
,
1889 if (*pos
>= GFS2_NR_SBSTATS
)
1894 static void gfs2_sbstats_seq_stop(struct seq_file
*seq
, void *iter_ptr
)
1899 static const struct seq_operations gfs2_glock_seq_ops
= {
1900 .start
= gfs2_glock_seq_start
,
1901 .next
= gfs2_glock_seq_next
,
1902 .stop
= gfs2_glock_seq_stop
,
1903 .show
= gfs2_glock_seq_show
,
1906 static const struct seq_operations gfs2_glstats_seq_ops
= {
1907 .start
= gfs2_glock_seq_start
,
1908 .next
= gfs2_glock_seq_next
,
1909 .stop
= gfs2_glock_seq_stop
,
1910 .show
= gfs2_glstats_seq_show
,
1913 static const struct seq_operations gfs2_sbstats_seq_ops
= {
1914 .start
= gfs2_sbstats_seq_start
,
1915 .next
= gfs2_sbstats_seq_next
,
1916 .stop
= gfs2_sbstats_seq_stop
,
1917 .show
= gfs2_sbstats_seq_show
,
1920 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
1922 static int __gfs2_glocks_open(struct inode
*inode
, struct file
*file
,
1923 const struct seq_operations
*ops
)
1925 int ret
= seq_open_private(file
, ops
, sizeof(struct gfs2_glock_iter
));
1927 struct seq_file
*seq
= file
->private_data
;
1928 struct gfs2_glock_iter
*gi
= seq
->private;
1930 gi
->sdp
= inode
->i_private
;
1932 seq
->buf
= kmalloc(GFS2_SEQ_GOODSIZE
, GFP_KERNEL
| __GFP_NOWARN
);
1934 seq
->size
= GFS2_SEQ_GOODSIZE
;
1936 rhashtable_walk_enter(&gl_hash_table
, &gi
->hti
);
1941 static int gfs2_glocks_open(struct inode
*inode
, struct file
*file
)
1943 return __gfs2_glocks_open(inode
, file
, &gfs2_glock_seq_ops
);
1946 static int gfs2_glocks_release(struct inode
*inode
, struct file
*file
)
1948 struct seq_file
*seq
= file
->private_data
;
1949 struct gfs2_glock_iter
*gi
= seq
->private;
1952 rhashtable_walk_exit(&gi
->hti
);
1953 return seq_release_private(inode
, file
);
1956 static int gfs2_glstats_open(struct inode
*inode
, struct file
*file
)
1958 return __gfs2_glocks_open(inode
, file
, &gfs2_glstats_seq_ops
);
1961 static int gfs2_sbstats_open(struct inode
*inode
, struct file
*file
)
1963 int ret
= seq_open(file
, &gfs2_sbstats_seq_ops
);
1965 struct seq_file
*seq
= file
->private_data
;
1966 seq
->private = inode
->i_private
; /* sdp */
1971 static const struct file_operations gfs2_glocks_fops
= {
1972 .owner
= THIS_MODULE
,
1973 .open
= gfs2_glocks_open
,
1975 .llseek
= seq_lseek
,
1976 .release
= gfs2_glocks_release
,
1979 static const struct file_operations gfs2_glstats_fops
= {
1980 .owner
= THIS_MODULE
,
1981 .open
= gfs2_glstats_open
,
1983 .llseek
= seq_lseek
,
1984 .release
= gfs2_glocks_release
,
1987 static const struct file_operations gfs2_sbstats_fops
= {
1988 .owner
= THIS_MODULE
,
1989 .open
= gfs2_sbstats_open
,
1991 .llseek
= seq_lseek
,
1992 .release
= seq_release
,
1995 int gfs2_create_debugfs_file(struct gfs2_sbd
*sdp
)
1997 struct dentry
*dent
;
1999 dent
= debugfs_create_dir(sdp
->sd_table_name
, gfs2_root
);
2000 if (IS_ERR_OR_NULL(dent
))
2002 sdp
->debugfs_dir
= dent
;
2004 dent
= debugfs_create_file("glocks",
2006 sdp
->debugfs_dir
, sdp
,
2008 if (IS_ERR_OR_NULL(dent
))
2010 sdp
->debugfs_dentry_glocks
= dent
;
2012 dent
= debugfs_create_file("glstats",
2014 sdp
->debugfs_dir
, sdp
,
2015 &gfs2_glstats_fops
);
2016 if (IS_ERR_OR_NULL(dent
))
2018 sdp
->debugfs_dentry_glstats
= dent
;
2020 dent
= debugfs_create_file("sbstats",
2022 sdp
->debugfs_dir
, sdp
,
2023 &gfs2_sbstats_fops
);
2024 if (IS_ERR_OR_NULL(dent
))
2026 sdp
->debugfs_dentry_sbstats
= dent
;
2030 gfs2_delete_debugfs_file(sdp
);
2031 return dent
? PTR_ERR(dent
) : -ENOMEM
;
2034 void gfs2_delete_debugfs_file(struct gfs2_sbd
*sdp
)
2036 if (sdp
->debugfs_dir
) {
2037 if (sdp
->debugfs_dentry_glocks
) {
2038 debugfs_remove(sdp
->debugfs_dentry_glocks
);
2039 sdp
->debugfs_dentry_glocks
= NULL
;
2041 if (sdp
->debugfs_dentry_glstats
) {
2042 debugfs_remove(sdp
->debugfs_dentry_glstats
);
2043 sdp
->debugfs_dentry_glstats
= NULL
;
2045 if (sdp
->debugfs_dentry_sbstats
) {
2046 debugfs_remove(sdp
->debugfs_dentry_sbstats
);
2047 sdp
->debugfs_dentry_sbstats
= NULL
;
2049 debugfs_remove(sdp
->debugfs_dir
);
2050 sdp
->debugfs_dir
= NULL
;
2054 int gfs2_register_debugfs(void)
2056 gfs2_root
= debugfs_create_dir("gfs2", NULL
);
2057 if (IS_ERR(gfs2_root
))
2058 return PTR_ERR(gfs2_root
);
2059 return gfs2_root
? 0 : -ENOMEM
;
2062 void gfs2_unregister_debugfs(void)
2064 debugfs_remove(gfs2_root
);