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/hash.h>
19 #include <linux/jhash.h>
20 #include <linux/kallsyms.h>
21 #include <linux/gfs2_ondisk.h>
22 #include <linux/list.h>
23 #include <linux/wait.h>
24 #include <linux/module.h>
25 #include <linux/uaccess.h>
26 #include <linux/seq_file.h>
27 #include <linux/debugfs.h>
28 #include <linux/kthread.h>
29 #include <linux/freezer.h>
30 #include <linux/workqueue.h>
31 #include <linux/jiffies.h>
32 #include <linux/rcupdate.h>
33 #include <linux/rculist_bl.h>
34 #include <linux/bit_spinlock.h>
35 #include <linux/percpu.h>
36 #include <linux/list_sort.h>
37 #include <linux/lockref.h>
38 #include <linux/rhashtable.h>
51 #define CREATE_TRACE_POINTS
52 #include "trace_gfs2.h"
54 struct gfs2_glock_iter
{
55 struct gfs2_sbd
*sdp
; /* incore superblock */
56 struct rhashtable_iter hti
; /* rhashtable iterator */
57 struct gfs2_glock
*gl
; /* current glock struct */
58 loff_t last_pos
; /* last position */
61 typedef void (*glock_examiner
) (struct gfs2_glock
* gl
);
63 static void do_xmote(struct gfs2_glock
*gl
, struct gfs2_holder
*gh
, unsigned int target
);
65 static struct dentry
*gfs2_root
;
66 static struct workqueue_struct
*glock_workqueue
;
67 struct workqueue_struct
*gfs2_delete_workqueue
;
68 static LIST_HEAD(lru_list
);
69 static atomic_t lru_count
= ATOMIC_INIT(0);
70 static DEFINE_SPINLOCK(lru_lock
);
72 #define GFS2_GL_HASH_SHIFT 15
73 #define GFS2_GL_HASH_SIZE BIT(GFS2_GL_HASH_SHIFT)
75 static const struct rhashtable_params ht_parms
= {
76 .nelem_hint
= GFS2_GL_HASH_SIZE
* 3 / 4,
77 .key_len
= offsetofend(struct lm_lockname
, ln_type
),
78 .key_offset
= offsetof(struct gfs2_glock
, gl_name
),
79 .head_offset
= offsetof(struct gfs2_glock
, gl_node
),
82 static struct rhashtable gl_hash_table
;
84 #define GLOCK_WAIT_TABLE_BITS 12
85 #define GLOCK_WAIT_TABLE_SIZE (1 << GLOCK_WAIT_TABLE_BITS)
86 static wait_queue_head_t glock_wait_table
[GLOCK_WAIT_TABLE_SIZE
] __cacheline_aligned
;
88 struct wait_glock_queue
{
89 struct lm_lockname
*name
;
90 wait_queue_entry_t wait
;
93 static int glock_wake_function(wait_queue_entry_t
*wait
, unsigned int mode
,
96 struct wait_glock_queue
*wait_glock
=
97 container_of(wait
, struct wait_glock_queue
, wait
);
98 struct lm_lockname
*wait_name
= wait_glock
->name
;
99 struct lm_lockname
*wake_name
= key
;
101 if (wake_name
->ln_sbd
!= wait_name
->ln_sbd
||
102 wake_name
->ln_number
!= wait_name
->ln_number
||
103 wake_name
->ln_type
!= wait_name
->ln_type
)
105 return autoremove_wake_function(wait
, mode
, sync
, key
);
108 static wait_queue_head_t
*glock_waitqueue(struct lm_lockname
*name
)
110 u32 hash
= jhash2((u32
*)name
, ht_parms
.key_len
/ 4, 0);
112 return glock_wait_table
+ hash_32(hash
, GLOCK_WAIT_TABLE_BITS
);
116 * wake_up_glock - Wake up waiters on a glock
119 static void wake_up_glock(struct gfs2_glock
*gl
)
121 wait_queue_head_t
*wq
= glock_waitqueue(&gl
->gl_name
);
123 if (waitqueue_active(wq
))
124 __wake_up(wq
, TASK_NORMAL
, 1, &gl
->gl_name
);
127 static void gfs2_glock_dealloc(struct rcu_head
*rcu
)
129 struct gfs2_glock
*gl
= container_of(rcu
, struct gfs2_glock
, gl_rcu
);
131 if (gl
->gl_ops
->go_flags
& GLOF_ASPACE
) {
132 kmem_cache_free(gfs2_glock_aspace_cachep
, gl
);
134 kfree(gl
->gl_lksb
.sb_lvbptr
);
135 kmem_cache_free(gfs2_glock_cachep
, gl
);
139 void gfs2_glock_free(struct gfs2_glock
*gl
)
141 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
143 BUG_ON(atomic_read(&gl
->gl_revokes
));
144 rhashtable_remove_fast(&gl_hash_table
, &gl
->gl_node
, ht_parms
);
147 call_rcu(&gl
->gl_rcu
, gfs2_glock_dealloc
);
148 if (atomic_dec_and_test(&sdp
->sd_glock_disposal
))
149 wake_up(&sdp
->sd_glock_wait
);
153 * gfs2_glock_hold() - increment reference count on glock
154 * @gl: The glock to hold
158 void gfs2_glock_hold(struct gfs2_glock
*gl
)
160 GLOCK_BUG_ON(gl
, __lockref_is_dead(&gl
->gl_lockref
));
161 lockref_get(&gl
->gl_lockref
);
165 * demote_ok - Check to see if it's ok to unlock a glock
168 * Returns: 1 if it's ok
171 static int demote_ok(const struct gfs2_glock
*gl
)
173 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
175 if (gl
->gl_state
== LM_ST_UNLOCKED
)
177 if (!list_empty(&gl
->gl_holders
))
179 if (glops
->go_demote_ok
)
180 return glops
->go_demote_ok(gl
);
185 void gfs2_glock_add_to_lru(struct gfs2_glock
*gl
)
187 if (!(gl
->gl_ops
->go_flags
& GLOF_LRU
))
190 spin_lock(&lru_lock
);
192 list_del(&gl
->gl_lru
);
193 list_add_tail(&gl
->gl_lru
, &lru_list
);
195 if (!test_bit(GLF_LRU
, &gl
->gl_flags
)) {
196 set_bit(GLF_LRU
, &gl
->gl_flags
);
197 atomic_inc(&lru_count
);
200 spin_unlock(&lru_lock
);
203 static void gfs2_glock_remove_from_lru(struct gfs2_glock
*gl
)
205 if (!(gl
->gl_ops
->go_flags
& GLOF_LRU
))
208 spin_lock(&lru_lock
);
209 if (test_bit(GLF_LRU
, &gl
->gl_flags
)) {
210 list_del_init(&gl
->gl_lru
);
211 atomic_dec(&lru_count
);
212 clear_bit(GLF_LRU
, &gl
->gl_flags
);
214 spin_unlock(&lru_lock
);
218 * Enqueue the glock on the work queue. Passes one glock reference on to the
221 static void __gfs2_glock_queue_work(struct gfs2_glock
*gl
, unsigned long delay
) {
222 if (!queue_delayed_work(glock_workqueue
, &gl
->gl_work
, delay
)) {
224 * We are holding the lockref spinlock, and the work was still
225 * queued above. The queued work (glock_work_func) takes that
226 * spinlock before dropping its glock reference(s), so it
227 * cannot have dropped them in the meantime.
229 GLOCK_BUG_ON(gl
, gl
->gl_lockref
.count
< 2);
230 gl
->gl_lockref
.count
--;
234 static void gfs2_glock_queue_work(struct gfs2_glock
*gl
, unsigned long delay
) {
235 spin_lock(&gl
->gl_lockref
.lock
);
236 __gfs2_glock_queue_work(gl
, delay
);
237 spin_unlock(&gl
->gl_lockref
.lock
);
240 static void __gfs2_glock_put(struct gfs2_glock
*gl
)
242 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
243 struct address_space
*mapping
= gfs2_glock2aspace(gl
);
245 lockref_mark_dead(&gl
->gl_lockref
);
247 gfs2_glock_remove_from_lru(gl
);
248 spin_unlock(&gl
->gl_lockref
.lock
);
249 GLOCK_BUG_ON(gl
, !list_empty(&gl
->gl_holders
));
250 GLOCK_BUG_ON(gl
, mapping
&& mapping
->nrpages
);
251 trace_gfs2_glock_put(gl
);
252 sdp
->sd_lockstruct
.ls_ops
->lm_put_lock(gl
);
256 * Cause the glock to be put in work queue context.
258 void gfs2_glock_queue_put(struct gfs2_glock
*gl
)
260 gfs2_glock_queue_work(gl
, 0);
264 * gfs2_glock_put() - Decrement reference count on glock
265 * @gl: The glock to put
269 void gfs2_glock_put(struct gfs2_glock
*gl
)
271 if (lockref_put_or_lock(&gl
->gl_lockref
))
274 __gfs2_glock_put(gl
);
278 * may_grant - check if its ok to grant a new lock
280 * @gh: The lock request which we wish to grant
282 * Returns: true if its ok to grant the lock
285 static inline int may_grant(const struct gfs2_glock
*gl
, const struct gfs2_holder
*gh
)
287 const struct gfs2_holder
*gh_head
= list_entry(gl
->gl_holders
.next
, const struct gfs2_holder
, gh_list
);
288 if ((gh
->gh_state
== LM_ST_EXCLUSIVE
||
289 gh_head
->gh_state
== LM_ST_EXCLUSIVE
) && gh
!= gh_head
)
291 if (gl
->gl_state
== gh
->gh_state
)
293 if (gh
->gh_flags
& GL_EXACT
)
295 if (gl
->gl_state
== LM_ST_EXCLUSIVE
) {
296 if (gh
->gh_state
== LM_ST_SHARED
&& gh_head
->gh_state
== LM_ST_SHARED
)
298 if (gh
->gh_state
== LM_ST_DEFERRED
&& gh_head
->gh_state
== LM_ST_DEFERRED
)
301 if (gl
->gl_state
!= LM_ST_UNLOCKED
&& (gh
->gh_flags
& LM_FLAG_ANY
))
306 static void gfs2_holder_wake(struct gfs2_holder
*gh
)
308 clear_bit(HIF_WAIT
, &gh
->gh_iflags
);
309 smp_mb__after_atomic();
310 wake_up_bit(&gh
->gh_iflags
, HIF_WAIT
);
314 * do_error - Something unexpected has happened during a lock request
318 static void do_error(struct gfs2_glock
*gl
, const int ret
)
320 struct gfs2_holder
*gh
, *tmp
;
322 list_for_each_entry_safe(gh
, tmp
, &gl
->gl_holders
, gh_list
) {
323 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
325 if (ret
& LM_OUT_ERROR
)
327 else if (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
))
328 gh
->gh_error
= GLR_TRYFAILED
;
331 list_del_init(&gh
->gh_list
);
332 trace_gfs2_glock_queue(gh
, 0);
333 gfs2_holder_wake(gh
);
338 * do_promote - promote as many requests as possible on the current queue
341 * Returns: 1 if there is a blocked holder at the head of the list, or 2
342 * if a type specific operation is underway.
345 static int do_promote(struct gfs2_glock
*gl
)
346 __releases(&gl
->gl_lockref
.lock
)
347 __acquires(&gl
->gl_lockref
.lock
)
349 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
350 struct gfs2_holder
*gh
, *tmp
;
354 list_for_each_entry_safe(gh
, tmp
, &gl
->gl_holders
, gh_list
) {
355 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
357 if (may_grant(gl
, gh
)) {
358 if (gh
->gh_list
.prev
== &gl
->gl_holders
&&
360 spin_unlock(&gl
->gl_lockref
.lock
);
361 /* FIXME: eliminate this eventually */
362 ret
= glops
->go_lock(gh
);
363 spin_lock(&gl
->gl_lockref
.lock
);
368 list_del_init(&gh
->gh_list
);
369 trace_gfs2_glock_queue(gh
, 0);
370 gfs2_holder_wake(gh
);
373 set_bit(HIF_HOLDER
, &gh
->gh_iflags
);
374 trace_gfs2_promote(gh
, 1);
375 gfs2_holder_wake(gh
);
378 set_bit(HIF_HOLDER
, &gh
->gh_iflags
);
379 trace_gfs2_promote(gh
, 0);
380 gfs2_holder_wake(gh
);
383 if (gh
->gh_list
.prev
== &gl
->gl_holders
)
392 * find_first_waiter - find the first gh that's waiting for the glock
396 static inline struct gfs2_holder
*find_first_waiter(const struct gfs2_glock
*gl
)
398 struct gfs2_holder
*gh
;
400 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
) {
401 if (!test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
408 * state_change - record that the glock is now in a different state
410 * @new_state the new state
414 static void state_change(struct gfs2_glock
*gl
, unsigned int new_state
)
418 held1
= (gl
->gl_state
!= LM_ST_UNLOCKED
);
419 held2
= (new_state
!= LM_ST_UNLOCKED
);
421 if (held1
!= held2
) {
422 GLOCK_BUG_ON(gl
, __lockref_is_dead(&gl
->gl_lockref
));
424 gl
->gl_lockref
.count
++;
426 gl
->gl_lockref
.count
--;
428 if (held1
&& held2
&& list_empty(&gl
->gl_holders
))
429 clear_bit(GLF_QUEUED
, &gl
->gl_flags
);
431 if (new_state
!= gl
->gl_target
)
432 /* shorten our minimum hold time */
433 gl
->gl_hold_time
= max(gl
->gl_hold_time
- GL_GLOCK_HOLD_DECR
,
435 gl
->gl_state
= new_state
;
436 gl
->gl_tchange
= jiffies
;
439 static void gfs2_demote_wake(struct gfs2_glock
*gl
)
441 gl
->gl_demote_state
= LM_ST_EXCLUSIVE
;
442 clear_bit(GLF_DEMOTE
, &gl
->gl_flags
);
443 smp_mb__after_atomic();
444 wake_up_bit(&gl
->gl_flags
, GLF_DEMOTE
);
448 * finish_xmote - The DLM has replied to one of our lock requests
450 * @ret: The status from the DLM
454 static void finish_xmote(struct gfs2_glock
*gl
, unsigned int ret
)
456 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
457 struct gfs2_holder
*gh
;
458 unsigned state
= ret
& LM_OUT_ST_MASK
;
461 spin_lock(&gl
->gl_lockref
.lock
);
462 trace_gfs2_glock_state_change(gl
, state
);
463 state_change(gl
, state
);
464 gh
= find_first_waiter(gl
);
466 /* Demote to UN request arrived during demote to SH or DF */
467 if (test_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
) &&
468 state
!= LM_ST_UNLOCKED
&& gl
->gl_demote_state
== LM_ST_UNLOCKED
)
469 gl
->gl_target
= LM_ST_UNLOCKED
;
471 /* Check for state != intended state */
472 if (unlikely(state
!= gl
->gl_target
)) {
473 if (gh
&& !test_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
)) {
474 /* move to back of queue and try next entry */
475 if (ret
& LM_OUT_CANCELED
) {
476 if ((gh
->gh_flags
& LM_FLAG_PRIORITY
) == 0)
477 list_move_tail(&gh
->gh_list
, &gl
->gl_holders
);
478 gh
= find_first_waiter(gl
);
479 gl
->gl_target
= gh
->gh_state
;
482 /* Some error or failed "try lock" - report it */
483 if ((ret
& LM_OUT_ERROR
) ||
484 (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
))) {
485 gl
->gl_target
= gl
->gl_state
;
491 /* Unlocked due to conversion deadlock, try again */
494 do_xmote(gl
, gh
, gl
->gl_target
);
496 /* Conversion fails, unlock and try again */
499 do_xmote(gl
, gh
, LM_ST_UNLOCKED
);
501 default: /* Everything else */
502 fs_err(gl
->gl_name
.ln_sbd
, "wanted %u got %u\n",
503 gl
->gl_target
, state
);
506 spin_unlock(&gl
->gl_lockref
.lock
);
510 /* Fast path - we got what we asked for */
511 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
))
512 gfs2_demote_wake(gl
);
513 if (state
!= LM_ST_UNLOCKED
) {
514 if (glops
->go_xmote_bh
) {
515 spin_unlock(&gl
->gl_lockref
.lock
);
516 rv
= glops
->go_xmote_bh(gl
, gh
);
517 spin_lock(&gl
->gl_lockref
.lock
);
528 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
530 spin_unlock(&gl
->gl_lockref
.lock
);
534 * do_xmote - Calls the DLM to change the state of a lock
535 * @gl: The lock state
536 * @gh: The holder (only for promotes)
537 * @target: The target lock state
541 static void do_xmote(struct gfs2_glock
*gl
, struct gfs2_holder
*gh
, unsigned int target
)
542 __releases(&gl
->gl_lockref
.lock
)
543 __acquires(&gl
->gl_lockref
.lock
)
545 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
546 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
547 unsigned int lck_flags
= (unsigned int)(gh
? gh
->gh_flags
: 0);
550 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)) &&
551 target
!= LM_ST_UNLOCKED
)
553 lck_flags
&= (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
| LM_FLAG_NOEXP
|
555 GLOCK_BUG_ON(gl
, gl
->gl_state
== target
);
556 GLOCK_BUG_ON(gl
, gl
->gl_state
== gl
->gl_target
);
557 if ((target
== LM_ST_UNLOCKED
|| target
== LM_ST_DEFERRED
) &&
559 set_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
);
560 do_error(gl
, 0); /* Fail queued try locks */
563 set_bit(GLF_BLOCKING
, &gl
->gl_flags
);
564 if ((gl
->gl_req
== LM_ST_UNLOCKED
) ||
565 (gl
->gl_state
== LM_ST_EXCLUSIVE
) ||
566 (lck_flags
& (LM_FLAG_TRY
|LM_FLAG_TRY_1CB
)))
567 clear_bit(GLF_BLOCKING
, &gl
->gl_flags
);
568 spin_unlock(&gl
->gl_lockref
.lock
);
571 if (test_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
))
572 glops
->go_inval(gl
, target
== LM_ST_DEFERRED
? 0 : DIO_METADATA
);
573 clear_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
);
576 if (sdp
->sd_lockstruct
.ls_ops
->lm_lock
) {
578 ret
= sdp
->sd_lockstruct
.ls_ops
->lm_lock(gl
, target
, lck_flags
);
579 if (ret
== -EINVAL
&& gl
->gl_target
== LM_ST_UNLOCKED
&&
580 target
== LM_ST_UNLOCKED
&&
581 test_bit(SDF_SKIP_DLM_UNLOCK
, &sdp
->sd_flags
)) {
582 finish_xmote(gl
, target
);
583 gfs2_glock_queue_work(gl
, 0);
586 fs_err(sdp
, "lm_lock ret %d\n", ret
);
587 GLOCK_BUG_ON(gl
, !test_bit(SDF_SHUTDOWN
,
590 } else { /* lock_nolock */
591 finish_xmote(gl
, target
);
592 gfs2_glock_queue_work(gl
, 0);
595 spin_lock(&gl
->gl_lockref
.lock
);
599 * find_first_holder - find the first "holder" gh
603 static inline struct gfs2_holder
*find_first_holder(const struct gfs2_glock
*gl
)
605 struct gfs2_holder
*gh
;
607 if (!list_empty(&gl
->gl_holders
)) {
608 gh
= list_entry(gl
->gl_holders
.next
, struct gfs2_holder
, gh_list
);
609 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
616 * run_queue - do all outstanding tasks related to a glock
617 * @gl: The glock in question
618 * @nonblock: True if we must not block in run_queue
622 static void run_queue(struct gfs2_glock
*gl
, const int nonblock
)
623 __releases(&gl
->gl_lockref
.lock
)
624 __acquires(&gl
->gl_lockref
.lock
)
626 struct gfs2_holder
*gh
= NULL
;
629 if (test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
))
632 GLOCK_BUG_ON(gl
, test_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
));
634 if (test_bit(GLF_DEMOTE
, &gl
->gl_flags
) &&
635 gl
->gl_demote_state
!= gl
->gl_state
) {
636 if (find_first_holder(gl
))
640 set_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
);
641 GLOCK_BUG_ON(gl
, gl
->gl_demote_state
== LM_ST_EXCLUSIVE
);
642 gl
->gl_target
= gl
->gl_demote_state
;
644 if (test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
645 gfs2_demote_wake(gl
);
646 ret
= do_promote(gl
);
651 gh
= find_first_waiter(gl
);
652 gl
->gl_target
= gh
->gh_state
;
653 if (!(gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
)))
654 do_error(gl
, 0); /* Fail queued try locks */
656 do_xmote(gl
, gh
, gl
->gl_target
);
661 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
662 smp_mb__after_atomic();
663 gl
->gl_lockref
.count
++;
664 __gfs2_glock_queue_work(gl
, 0);
668 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
669 smp_mb__after_atomic();
673 static void delete_work_func(struct work_struct
*work
)
675 struct gfs2_glock
*gl
= container_of(work
, struct gfs2_glock
, gl_delete
);
676 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
678 u64 no_addr
= gl
->gl_name
.ln_number
;
680 /* If someone's using this glock to create a new dinode, the block must
681 have been freed by another node, then re-used, in which case our
682 iopen callback is too late after the fact. Ignore it. */
683 if (test_bit(GLF_INODE_CREATING
, &gl
->gl_flags
))
686 inode
= gfs2_lookup_by_inum(sdp
, no_addr
, NULL
, GFS2_BLKST_UNLINKED
);
687 if (inode
&& !IS_ERR(inode
)) {
688 d_prune_aliases(inode
);
695 static void glock_work_func(struct work_struct
*work
)
697 unsigned long delay
= 0;
698 struct gfs2_glock
*gl
= container_of(work
, struct gfs2_glock
, gl_work
.work
);
699 unsigned int drop_refs
= 1;
701 if (test_and_clear_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
)) {
702 finish_xmote(gl
, gl
->gl_reply
);
705 spin_lock(&gl
->gl_lockref
.lock
);
706 if (test_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
) &&
707 gl
->gl_state
!= LM_ST_UNLOCKED
&&
708 gl
->gl_demote_state
!= LM_ST_EXCLUSIVE
) {
709 unsigned long holdtime
, now
= jiffies
;
711 holdtime
= gl
->gl_tchange
+ gl
->gl_hold_time
;
712 if (time_before(now
, holdtime
))
713 delay
= holdtime
- now
;
716 clear_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
);
717 set_bit(GLF_DEMOTE
, &gl
->gl_flags
);
722 /* Keep one glock reference for the work we requeue. */
724 if (gl
->gl_name
.ln_type
!= LM_TYPE_INODE
)
726 __gfs2_glock_queue_work(gl
, delay
);
730 * Drop the remaining glock references manually here. (Mind that
731 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
734 gl
->gl_lockref
.count
-= drop_refs
;
735 if (!gl
->gl_lockref
.count
) {
736 __gfs2_glock_put(gl
);
739 spin_unlock(&gl
->gl_lockref
.lock
);
742 static struct gfs2_glock
*find_insert_glock(struct lm_lockname
*name
,
743 struct gfs2_glock
*new)
745 struct wait_glock_queue wait
;
746 wait_queue_head_t
*wq
= glock_waitqueue(name
);
747 struct gfs2_glock
*gl
;
750 init_wait(&wait
.wait
);
751 wait
.wait
.func
= glock_wake_function
;
754 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
757 gl
= rhashtable_lookup_get_insert_fast(&gl_hash_table
,
758 &new->gl_node
, ht_parms
);
762 gl
= rhashtable_lookup_fast(&gl_hash_table
,
765 if (gl
&& !lockref_get_not_dead(&gl
->gl_lockref
)) {
772 finish_wait(wq
, &wait
.wait
);
777 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
778 * @sdp: The GFS2 superblock
779 * @number: the lock number
780 * @glops: The glock_operations to use
781 * @create: If 0, don't create the glock if it doesn't exist
782 * @glp: the glock is returned here
784 * This does not lock a glock, just finds/creates structures for one.
789 int gfs2_glock_get(struct gfs2_sbd
*sdp
, u64 number
,
790 const struct gfs2_glock_operations
*glops
, int create
,
791 struct gfs2_glock
**glp
)
793 struct super_block
*s
= sdp
->sd_vfs
;
794 struct lm_lockname name
= { .ln_number
= number
,
795 .ln_type
= glops
->go_type
,
797 struct gfs2_glock
*gl
, *tmp
;
798 struct address_space
*mapping
;
799 struct kmem_cache
*cachep
;
802 gl
= find_insert_glock(&name
, NULL
);
810 if (glops
->go_flags
& GLOF_ASPACE
)
811 cachep
= gfs2_glock_aspace_cachep
;
813 cachep
= gfs2_glock_cachep
;
814 gl
= kmem_cache_alloc(cachep
, GFP_NOFS
);
818 memset(&gl
->gl_lksb
, 0, sizeof(struct dlm_lksb
));
820 if (glops
->go_flags
& GLOF_LVB
) {
821 gl
->gl_lksb
.sb_lvbptr
= kzalloc(GFS2_MIN_LVB_SIZE
, GFP_NOFS
);
822 if (!gl
->gl_lksb
.sb_lvbptr
) {
823 kmem_cache_free(cachep
, gl
);
828 atomic_inc(&sdp
->sd_glock_disposal
);
829 gl
->gl_node
.next
= NULL
;
832 gl
->gl_lockref
.count
= 1;
833 gl
->gl_state
= LM_ST_UNLOCKED
;
834 gl
->gl_target
= LM_ST_UNLOCKED
;
835 gl
->gl_demote_state
= LM_ST_EXCLUSIVE
;
839 /* We use the global stats to estimate the initial per-glock stats */
840 gl
->gl_stats
= this_cpu_ptr(sdp
->sd_lkstats
)->lkstats
[glops
->go_type
];
842 gl
->gl_stats
.stats
[GFS2_LKS_DCOUNT
] = 0;
843 gl
->gl_stats
.stats
[GFS2_LKS_QCOUNT
] = 0;
844 gl
->gl_tchange
= jiffies
;
845 gl
->gl_object
= NULL
;
846 gl
->gl_hold_time
= GL_GLOCK_DFT_HOLD
;
847 INIT_DELAYED_WORK(&gl
->gl_work
, glock_work_func
);
848 INIT_WORK(&gl
->gl_delete
, delete_work_func
);
850 mapping
= gfs2_glock2aspace(gl
);
852 mapping
->a_ops
= &gfs2_meta_aops
;
853 mapping
->host
= s
->s_bdev
->bd_inode
;
855 mapping_set_gfp_mask(mapping
, GFP_NOFS
);
856 mapping
->private_data
= NULL
;
857 mapping
->writeback_index
= 0;
860 tmp
= find_insert_glock(&name
, gl
);
872 kfree(gl
->gl_lksb
.sb_lvbptr
);
873 kmem_cache_free(cachep
, gl
);
874 atomic_dec(&sdp
->sd_glock_disposal
);
881 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
883 * @state: the state we're requesting
884 * @flags: the modifier flags
885 * @gh: the holder structure
889 void gfs2_holder_init(struct gfs2_glock
*gl
, unsigned int state
, u16 flags
,
890 struct gfs2_holder
*gh
)
892 INIT_LIST_HEAD(&gh
->gh_list
);
894 gh
->gh_ip
= _RET_IP_
;
895 gh
->gh_owner_pid
= get_pid(task_pid(current
));
896 gh
->gh_state
= state
;
897 gh
->gh_flags
= flags
;
904 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
905 * @state: the state we're requesting
906 * @flags: the modifier flags
907 * @gh: the holder structure
909 * Don't mess with the glock.
913 void gfs2_holder_reinit(unsigned int state
, u16 flags
, struct gfs2_holder
*gh
)
915 gh
->gh_state
= state
;
916 gh
->gh_flags
= flags
;
918 gh
->gh_ip
= _RET_IP_
;
919 put_pid(gh
->gh_owner_pid
);
920 gh
->gh_owner_pid
= get_pid(task_pid(current
));
924 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
925 * @gh: the holder structure
929 void gfs2_holder_uninit(struct gfs2_holder
*gh
)
931 put_pid(gh
->gh_owner_pid
);
932 gfs2_glock_put(gh
->gh_gl
);
933 gfs2_holder_mark_uninitialized(gh
);
938 * gfs2_glock_wait - wait on a glock acquisition
939 * @gh: the glock holder
941 * Returns: 0 on success
944 int gfs2_glock_wait(struct gfs2_holder
*gh
)
946 unsigned long time1
= jiffies
;
949 wait_on_bit(&gh
->gh_iflags
, HIF_WAIT
, TASK_UNINTERRUPTIBLE
);
950 if (time_after(jiffies
, time1
+ HZ
)) /* have we waited > a second? */
951 /* Lengthen the minimum hold time. */
952 gh
->gh_gl
->gl_hold_time
= min(gh
->gh_gl
->gl_hold_time
+
959 * handle_callback - process a demote request
961 * @state: the state the caller wants us to change to
963 * There are only two requests that we are going to see in actual
964 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
967 static void handle_callback(struct gfs2_glock
*gl
, unsigned int state
,
968 unsigned long delay
, bool remote
)
970 int bit
= delay
? GLF_PENDING_DEMOTE
: GLF_DEMOTE
;
972 set_bit(bit
, &gl
->gl_flags
);
973 if (gl
->gl_demote_state
== LM_ST_EXCLUSIVE
) {
974 gl
->gl_demote_state
= state
;
975 gl
->gl_demote_time
= jiffies
;
976 } else if (gl
->gl_demote_state
!= LM_ST_UNLOCKED
&&
977 gl
->gl_demote_state
!= state
) {
978 gl
->gl_demote_state
= LM_ST_UNLOCKED
;
980 if (gl
->gl_ops
->go_callback
)
981 gl
->gl_ops
->go_callback(gl
, remote
);
982 trace_gfs2_demote_rq(gl
, remote
);
985 void gfs2_print_dbg(struct seq_file
*seq
, const char *fmt
, ...)
987 struct va_format vaf
;
993 seq_vprintf(seq
, fmt
, args
);
1005 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1006 * @gh: the holder structure to add
1008 * Eventually we should move the recursive locking trap to a
1009 * debugging option or something like that. This is the fast
1010 * path and needs to have the minimum number of distractions.
1014 static inline void add_to_queue(struct gfs2_holder
*gh
)
1015 __releases(&gl
->gl_lockref
.lock
)
1016 __acquires(&gl
->gl_lockref
.lock
)
1018 struct gfs2_glock
*gl
= gh
->gh_gl
;
1019 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
1020 struct list_head
*insert_pt
= NULL
;
1021 struct gfs2_holder
*gh2
;
1024 BUG_ON(gh
->gh_owner_pid
== NULL
);
1025 if (test_and_set_bit(HIF_WAIT
, &gh
->gh_iflags
))
1028 if (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
)) {
1029 if (test_bit(GLF_LOCK
, &gl
->gl_flags
))
1030 try_futile
= !may_grant(gl
, gh
);
1031 if (test_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
))
1035 list_for_each_entry(gh2
, &gl
->gl_holders
, gh_list
) {
1036 if (unlikely(gh2
->gh_owner_pid
== gh
->gh_owner_pid
&&
1037 (gh
->gh_gl
->gl_ops
->go_type
!= LM_TYPE_FLOCK
)))
1038 goto trap_recursive
;
1040 !(gh2
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
))) {
1042 gh
->gh_error
= GLR_TRYFAILED
;
1043 gfs2_holder_wake(gh
);
1046 if (test_bit(HIF_HOLDER
, &gh2
->gh_iflags
))
1048 if (unlikely((gh
->gh_flags
& LM_FLAG_PRIORITY
) && !insert_pt
))
1049 insert_pt
= &gh2
->gh_list
;
1051 set_bit(GLF_QUEUED
, &gl
->gl_flags
);
1052 trace_gfs2_glock_queue(gh
, 1);
1053 gfs2_glstats_inc(gl
, GFS2_LKS_QCOUNT
);
1054 gfs2_sbstats_inc(gl
, GFS2_LKS_QCOUNT
);
1055 if (likely(insert_pt
== NULL
)) {
1056 list_add_tail(&gh
->gh_list
, &gl
->gl_holders
);
1057 if (unlikely(gh
->gh_flags
& LM_FLAG_PRIORITY
))
1061 list_add_tail(&gh
->gh_list
, insert_pt
);
1063 gh
= list_entry(gl
->gl_holders
.next
, struct gfs2_holder
, gh_list
);
1064 if (!(gh
->gh_flags
& LM_FLAG_PRIORITY
)) {
1065 spin_unlock(&gl
->gl_lockref
.lock
);
1066 if (sdp
->sd_lockstruct
.ls_ops
->lm_cancel
)
1067 sdp
->sd_lockstruct
.ls_ops
->lm_cancel(gl
);
1068 spin_lock(&gl
->gl_lockref
.lock
);
1073 fs_err(sdp
, "original: %pSR\n", (void *)gh2
->gh_ip
);
1074 fs_err(sdp
, "pid: %d\n", pid_nr(gh2
->gh_owner_pid
));
1075 fs_err(sdp
, "lock type: %d req lock state : %d\n",
1076 gh2
->gh_gl
->gl_name
.ln_type
, gh2
->gh_state
);
1077 fs_err(sdp
, "new: %pSR\n", (void *)gh
->gh_ip
);
1078 fs_err(sdp
, "pid: %d\n", pid_nr(gh
->gh_owner_pid
));
1079 fs_err(sdp
, "lock type: %d req lock state : %d\n",
1080 gh
->gh_gl
->gl_name
.ln_type
, gh
->gh_state
);
1081 gfs2_dump_glock(NULL
, gl
);
1086 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1087 * @gh: the holder structure
1089 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1091 * Returns: 0, GLR_TRYFAILED, or errno on failure
1094 int gfs2_glock_nq(struct gfs2_holder
*gh
)
1096 struct gfs2_glock
*gl
= gh
->gh_gl
;
1097 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
1100 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)))
1103 if (test_bit(GLF_LRU
, &gl
->gl_flags
))
1104 gfs2_glock_remove_from_lru(gl
);
1106 spin_lock(&gl
->gl_lockref
.lock
);
1108 if (unlikely((LM_FLAG_NOEXP
& gh
->gh_flags
) &&
1109 test_and_clear_bit(GLF_FROZEN
, &gl
->gl_flags
))) {
1110 set_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
);
1111 gl
->gl_lockref
.count
++;
1112 __gfs2_glock_queue_work(gl
, 0);
1115 spin_unlock(&gl
->gl_lockref
.lock
);
1117 if (!(gh
->gh_flags
& GL_ASYNC
))
1118 error
= gfs2_glock_wait(gh
);
1124 * gfs2_glock_poll - poll to see if an async request has been completed
1127 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1130 int gfs2_glock_poll(struct gfs2_holder
*gh
)
1132 return test_bit(HIF_WAIT
, &gh
->gh_iflags
) ? 0 : 1;
1136 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1137 * @gh: the glock holder
1141 void gfs2_glock_dq(struct gfs2_holder
*gh
)
1143 struct gfs2_glock
*gl
= gh
->gh_gl
;
1144 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1148 spin_lock(&gl
->gl_lockref
.lock
);
1149 if (gh
->gh_flags
& GL_NOCACHE
)
1150 handle_callback(gl
, LM_ST_UNLOCKED
, 0, false);
1152 list_del_init(&gh
->gh_list
);
1153 clear_bit(HIF_HOLDER
, &gh
->gh_iflags
);
1154 if (find_first_holder(gl
) == NULL
) {
1155 if (glops
->go_unlock
) {
1156 GLOCK_BUG_ON(gl
, test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
));
1157 spin_unlock(&gl
->gl_lockref
.lock
);
1158 glops
->go_unlock(gh
);
1159 spin_lock(&gl
->gl_lockref
.lock
);
1160 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
1162 if (list_empty(&gl
->gl_holders
) &&
1163 !test_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
) &&
1164 !test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
1167 if (!test_bit(GLF_LFLUSH
, &gl
->gl_flags
) && demote_ok(gl
))
1168 gfs2_glock_add_to_lru(gl
);
1170 trace_gfs2_glock_queue(gh
, 0);
1171 if (unlikely(!fast_path
)) {
1172 gl
->gl_lockref
.count
++;
1173 if (test_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
) &&
1174 !test_bit(GLF_DEMOTE
, &gl
->gl_flags
) &&
1175 gl
->gl_name
.ln_type
== LM_TYPE_INODE
)
1176 delay
= gl
->gl_hold_time
;
1177 __gfs2_glock_queue_work(gl
, delay
);
1179 spin_unlock(&gl
->gl_lockref
.lock
);
1182 void gfs2_glock_dq_wait(struct gfs2_holder
*gh
)
1184 struct gfs2_glock
*gl
= gh
->gh_gl
;
1187 wait_on_bit(&gl
->gl_flags
, GLF_DEMOTE
, TASK_UNINTERRUPTIBLE
);
1191 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1192 * @gh: the holder structure
1196 void gfs2_glock_dq_uninit(struct gfs2_holder
*gh
)
1199 gfs2_holder_uninit(gh
);
1203 * gfs2_glock_nq_num - acquire a glock based on lock number
1204 * @sdp: the filesystem
1205 * @number: the lock number
1206 * @glops: the glock operations for the type of glock
1207 * @state: the state to acquire the glock in
1208 * @flags: modifier flags for the acquisition
1209 * @gh: the struct gfs2_holder
1214 int gfs2_glock_nq_num(struct gfs2_sbd
*sdp
, u64 number
,
1215 const struct gfs2_glock_operations
*glops
,
1216 unsigned int state
, u16 flags
, struct gfs2_holder
*gh
)
1218 struct gfs2_glock
*gl
;
1221 error
= gfs2_glock_get(sdp
, number
, glops
, CREATE
, &gl
);
1223 error
= gfs2_glock_nq_init(gl
, state
, flags
, gh
);
1231 * glock_compare - Compare two struct gfs2_glock structures for sorting
1232 * @arg_a: the first structure
1233 * @arg_b: the second structure
1237 static int glock_compare(const void *arg_a
, const void *arg_b
)
1239 const struct gfs2_holder
*gh_a
= *(const struct gfs2_holder
**)arg_a
;
1240 const struct gfs2_holder
*gh_b
= *(const struct gfs2_holder
**)arg_b
;
1241 const struct lm_lockname
*a
= &gh_a
->gh_gl
->gl_name
;
1242 const struct lm_lockname
*b
= &gh_b
->gh_gl
->gl_name
;
1244 if (a
->ln_number
> b
->ln_number
)
1246 if (a
->ln_number
< b
->ln_number
)
1248 BUG_ON(gh_a
->gh_gl
->gl_ops
->go_type
== gh_b
->gh_gl
->gl_ops
->go_type
);
1253 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1254 * @num_gh: the number of structures
1255 * @ghs: an array of struct gfs2_holder structures
1257 * Returns: 0 on success (all glocks acquired),
1258 * errno on failure (no glocks acquired)
1261 static int nq_m_sync(unsigned int num_gh
, struct gfs2_holder
*ghs
,
1262 struct gfs2_holder
**p
)
1267 for (x
= 0; x
< num_gh
; x
++)
1270 sort(p
, num_gh
, sizeof(struct gfs2_holder
*), glock_compare
, NULL
);
1272 for (x
= 0; x
< num_gh
; x
++) {
1273 p
[x
]->gh_flags
&= ~(LM_FLAG_TRY
| GL_ASYNC
);
1275 error
= gfs2_glock_nq(p
[x
]);
1278 gfs2_glock_dq(p
[x
]);
1287 * gfs2_glock_nq_m - acquire multiple glocks
1288 * @num_gh: the number of structures
1289 * @ghs: an array of struct gfs2_holder structures
1292 * Returns: 0 on success (all glocks acquired),
1293 * errno on failure (no glocks acquired)
1296 int gfs2_glock_nq_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1298 struct gfs2_holder
*tmp
[4];
1299 struct gfs2_holder
**pph
= tmp
;
1306 ghs
->gh_flags
&= ~(LM_FLAG_TRY
| GL_ASYNC
);
1307 return gfs2_glock_nq(ghs
);
1311 pph
= kmalloc_array(num_gh
, sizeof(struct gfs2_holder
*),
1317 error
= nq_m_sync(num_gh
, ghs
, pph
);
1326 * gfs2_glock_dq_m - release multiple glocks
1327 * @num_gh: the number of structures
1328 * @ghs: an array of struct gfs2_holder structures
1332 void gfs2_glock_dq_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1335 gfs2_glock_dq(&ghs
[num_gh
]);
1338 void gfs2_glock_cb(struct gfs2_glock
*gl
, unsigned int state
)
1340 unsigned long delay
= 0;
1341 unsigned long holdtime
;
1342 unsigned long now
= jiffies
;
1344 gfs2_glock_hold(gl
);
1345 holdtime
= gl
->gl_tchange
+ gl
->gl_hold_time
;
1346 if (test_bit(GLF_QUEUED
, &gl
->gl_flags
) &&
1347 gl
->gl_name
.ln_type
== LM_TYPE_INODE
) {
1348 if (time_before(now
, holdtime
))
1349 delay
= holdtime
- now
;
1350 if (test_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
))
1351 delay
= gl
->gl_hold_time
;
1354 spin_lock(&gl
->gl_lockref
.lock
);
1355 handle_callback(gl
, state
, delay
, true);
1356 __gfs2_glock_queue_work(gl
, delay
);
1357 spin_unlock(&gl
->gl_lockref
.lock
);
1361 * gfs2_should_freeze - Figure out if glock should be frozen
1362 * @gl: The glock in question
1364 * Glocks are not frozen if (a) the result of the dlm operation is
1365 * an error, (b) the locking operation was an unlock operation or
1366 * (c) if there is a "noexp" flagged request anywhere in the queue
1368 * Returns: 1 if freezing should occur, 0 otherwise
1371 static int gfs2_should_freeze(const struct gfs2_glock
*gl
)
1373 const struct gfs2_holder
*gh
;
1375 if (gl
->gl_reply
& ~LM_OUT_ST_MASK
)
1377 if (gl
->gl_target
== LM_ST_UNLOCKED
)
1380 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
) {
1381 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
1383 if (LM_FLAG_NOEXP
& gh
->gh_flags
)
1391 * gfs2_glock_complete - Callback used by locking
1392 * @gl: Pointer to the glock
1393 * @ret: The return value from the dlm
1395 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1396 * to use a bitfield shared with other glock state fields.
1399 void gfs2_glock_complete(struct gfs2_glock
*gl
, int ret
)
1401 struct lm_lockstruct
*ls
= &gl
->gl_name
.ln_sbd
->sd_lockstruct
;
1403 spin_lock(&gl
->gl_lockref
.lock
);
1406 if (unlikely(test_bit(DFL_BLOCK_LOCKS
, &ls
->ls_recover_flags
))) {
1407 if (gfs2_should_freeze(gl
)) {
1408 set_bit(GLF_FROZEN
, &gl
->gl_flags
);
1409 spin_unlock(&gl
->gl_lockref
.lock
);
1414 gl
->gl_lockref
.count
++;
1415 set_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
);
1416 __gfs2_glock_queue_work(gl
, 0);
1417 spin_unlock(&gl
->gl_lockref
.lock
);
1420 static int glock_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
1422 struct gfs2_glock
*gla
, *glb
;
1424 gla
= list_entry(a
, struct gfs2_glock
, gl_lru
);
1425 glb
= list_entry(b
, struct gfs2_glock
, gl_lru
);
1427 if (gla
->gl_name
.ln_number
> glb
->gl_name
.ln_number
)
1429 if (gla
->gl_name
.ln_number
< glb
->gl_name
.ln_number
)
1436 * gfs2_dispose_glock_lru - Demote a list of glocks
1437 * @list: The list to dispose of
1439 * Disposing of glocks may involve disk accesses, so that here we sort
1440 * the glocks by number (i.e. disk location of the inodes) so that if
1441 * there are any such accesses, they'll be sent in order (mostly).
1443 * Must be called under the lru_lock, but may drop and retake this
1444 * lock. While the lru_lock is dropped, entries may vanish from the
1445 * list, but no new entries will appear on the list (since it is
1449 static void gfs2_dispose_glock_lru(struct list_head
*list
)
1450 __releases(&lru_lock
)
1451 __acquires(&lru_lock
)
1453 struct gfs2_glock
*gl
;
1455 list_sort(NULL
, list
, glock_cmp
);
1457 while(!list_empty(list
)) {
1458 gl
= list_entry(list
->next
, struct gfs2_glock
, gl_lru
);
1459 list_del_init(&gl
->gl_lru
);
1460 if (!spin_trylock(&gl
->gl_lockref
.lock
)) {
1462 list_add(&gl
->gl_lru
, &lru_list
);
1463 set_bit(GLF_LRU
, &gl
->gl_flags
);
1464 atomic_inc(&lru_count
);
1467 if (test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
)) {
1468 spin_unlock(&gl
->gl_lockref
.lock
);
1469 goto add_back_to_lru
;
1471 gl
->gl_lockref
.count
++;
1473 handle_callback(gl
, LM_ST_UNLOCKED
, 0, false);
1474 WARN_ON(!test_and_clear_bit(GLF_LOCK
, &gl
->gl_flags
));
1475 __gfs2_glock_queue_work(gl
, 0);
1476 spin_unlock(&gl
->gl_lockref
.lock
);
1477 cond_resched_lock(&lru_lock
);
1482 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1483 * @nr: The number of entries to scan
1485 * This function selects the entries on the LRU which are able to
1486 * be demoted, and then kicks off the process by calling
1487 * gfs2_dispose_glock_lru() above.
1490 static long gfs2_scan_glock_lru(int nr
)
1492 struct gfs2_glock
*gl
;
1497 spin_lock(&lru_lock
);
1498 while ((nr
-- >= 0) && !list_empty(&lru_list
)) {
1499 gl
= list_entry(lru_list
.next
, struct gfs2_glock
, gl_lru
);
1501 /* Test for being demotable */
1502 if (!test_bit(GLF_LOCK
, &gl
->gl_flags
)) {
1503 list_move(&gl
->gl_lru
, &dispose
);
1504 atomic_dec(&lru_count
);
1505 clear_bit(GLF_LRU
, &gl
->gl_flags
);
1510 list_move(&gl
->gl_lru
, &skipped
);
1512 list_splice(&skipped
, &lru_list
);
1513 if (!list_empty(&dispose
))
1514 gfs2_dispose_glock_lru(&dispose
);
1515 spin_unlock(&lru_lock
);
1520 static unsigned long gfs2_glock_shrink_scan(struct shrinker
*shrink
,
1521 struct shrink_control
*sc
)
1523 if (!(sc
->gfp_mask
& __GFP_FS
))
1525 return gfs2_scan_glock_lru(sc
->nr_to_scan
);
1528 static unsigned long gfs2_glock_shrink_count(struct shrinker
*shrink
,
1529 struct shrink_control
*sc
)
1531 return vfs_pressure_ratio(atomic_read(&lru_count
));
1534 static struct shrinker glock_shrinker
= {
1535 .seeks
= DEFAULT_SEEKS
,
1536 .count_objects
= gfs2_glock_shrink_count
,
1537 .scan_objects
= gfs2_glock_shrink_scan
,
1541 * examine_bucket - Call a function for glock in a hash bucket
1542 * @examiner: the function
1543 * @sdp: the filesystem
1544 * @bucket: the bucket
1546 * Note that the function can be called multiple times on the same
1547 * object. So the user must ensure that the function can cope with
1551 static void glock_hash_walk(glock_examiner examiner
, const struct gfs2_sbd
*sdp
)
1553 struct gfs2_glock
*gl
;
1554 struct rhashtable_iter iter
;
1556 rhashtable_walk_enter(&gl_hash_table
, &iter
);
1559 rhashtable_walk_start(&iter
);
1561 while ((gl
= rhashtable_walk_next(&iter
)) && !IS_ERR(gl
))
1562 if (gl
->gl_name
.ln_sbd
== sdp
&&
1563 lockref_get_not_dead(&gl
->gl_lockref
))
1566 rhashtable_walk_stop(&iter
);
1567 } while (cond_resched(), gl
== ERR_PTR(-EAGAIN
));
1569 rhashtable_walk_exit(&iter
);
1573 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1574 * @gl: The glock to thaw
1578 static void thaw_glock(struct gfs2_glock
*gl
)
1580 if (!test_and_clear_bit(GLF_FROZEN
, &gl
->gl_flags
)) {
1584 set_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
);
1585 gfs2_glock_queue_work(gl
, 0);
1589 * clear_glock - look at a glock and see if we can free it from glock cache
1590 * @gl: the glock to look at
1594 static void clear_glock(struct gfs2_glock
*gl
)
1596 gfs2_glock_remove_from_lru(gl
);
1598 spin_lock(&gl
->gl_lockref
.lock
);
1599 if (gl
->gl_state
!= LM_ST_UNLOCKED
)
1600 handle_callback(gl
, LM_ST_UNLOCKED
, 0, false);
1601 __gfs2_glock_queue_work(gl
, 0);
1602 spin_unlock(&gl
->gl_lockref
.lock
);
1606 * gfs2_glock_thaw - Thaw any frozen glocks
1607 * @sdp: The super block
1611 void gfs2_glock_thaw(struct gfs2_sbd
*sdp
)
1613 glock_hash_walk(thaw_glock
, sdp
);
1616 static void dump_glock(struct seq_file
*seq
, struct gfs2_glock
*gl
)
1618 spin_lock(&gl
->gl_lockref
.lock
);
1619 gfs2_dump_glock(seq
, gl
);
1620 spin_unlock(&gl
->gl_lockref
.lock
);
1623 static void dump_glock_func(struct gfs2_glock
*gl
)
1625 dump_glock(NULL
, gl
);
1629 * gfs2_gl_hash_clear - Empty out the glock hash table
1630 * @sdp: the filesystem
1631 * @wait: wait until it's all gone
1633 * Called when unmounting the filesystem.
1636 void gfs2_gl_hash_clear(struct gfs2_sbd
*sdp
)
1638 set_bit(SDF_SKIP_DLM_UNLOCK
, &sdp
->sd_flags
);
1639 flush_workqueue(glock_workqueue
);
1640 glock_hash_walk(clear_glock
, sdp
);
1641 flush_workqueue(glock_workqueue
);
1642 wait_event_timeout(sdp
->sd_glock_wait
,
1643 atomic_read(&sdp
->sd_glock_disposal
) == 0,
1645 glock_hash_walk(dump_glock_func
, sdp
);
1648 void gfs2_glock_finish_truncate(struct gfs2_inode
*ip
)
1650 struct gfs2_glock
*gl
= ip
->i_gl
;
1653 ret
= gfs2_truncatei_resume(ip
);
1654 gfs2_assert_withdraw(gl
->gl_name
.ln_sbd
, ret
== 0);
1656 spin_lock(&gl
->gl_lockref
.lock
);
1657 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
1659 spin_unlock(&gl
->gl_lockref
.lock
);
1662 static const char *state2str(unsigned state
)
1665 case LM_ST_UNLOCKED
:
1669 case LM_ST_DEFERRED
:
1671 case LM_ST_EXCLUSIVE
:
1677 static const char *hflags2str(char *buf
, u16 flags
, unsigned long iflags
)
1680 if (flags
& LM_FLAG_TRY
)
1682 if (flags
& LM_FLAG_TRY_1CB
)
1684 if (flags
& LM_FLAG_NOEXP
)
1686 if (flags
& LM_FLAG_ANY
)
1688 if (flags
& LM_FLAG_PRIORITY
)
1690 if (flags
& GL_ASYNC
)
1692 if (flags
& GL_EXACT
)
1694 if (flags
& GL_NOCACHE
)
1696 if (test_bit(HIF_HOLDER
, &iflags
))
1698 if (test_bit(HIF_WAIT
, &iflags
))
1700 if (test_bit(HIF_FIRST
, &iflags
))
1707 * dump_holder - print information about a glock holder
1708 * @seq: the seq_file struct
1709 * @gh: the glock holder
1713 static void dump_holder(struct seq_file
*seq
, const struct gfs2_holder
*gh
)
1715 struct task_struct
*gh_owner
= NULL
;
1719 if (gh
->gh_owner_pid
)
1720 gh_owner
= pid_task(gh
->gh_owner_pid
, PIDTYPE_PID
);
1721 gfs2_print_dbg(seq
, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1722 state2str(gh
->gh_state
),
1723 hflags2str(flags_buf
, gh
->gh_flags
, gh
->gh_iflags
),
1725 gh
->gh_owner_pid
? (long)pid_nr(gh
->gh_owner_pid
) : -1,
1726 gh_owner
? gh_owner
->comm
: "(ended)",
1731 static const char *gflags2str(char *buf
, const struct gfs2_glock
*gl
)
1733 const unsigned long *gflags
= &gl
->gl_flags
;
1736 if (test_bit(GLF_LOCK
, gflags
))
1738 if (test_bit(GLF_DEMOTE
, gflags
))
1740 if (test_bit(GLF_PENDING_DEMOTE
, gflags
))
1742 if (test_bit(GLF_DEMOTE_IN_PROGRESS
, gflags
))
1744 if (test_bit(GLF_DIRTY
, gflags
))
1746 if (test_bit(GLF_LFLUSH
, gflags
))
1748 if (test_bit(GLF_INVALIDATE_IN_PROGRESS
, gflags
))
1750 if (test_bit(GLF_REPLY_PENDING
, gflags
))
1752 if (test_bit(GLF_INITIAL
, gflags
))
1754 if (test_bit(GLF_FROZEN
, gflags
))
1756 if (test_bit(GLF_QUEUED
, gflags
))
1758 if (test_bit(GLF_LRU
, gflags
))
1762 if (test_bit(GLF_BLOCKING
, gflags
))
1769 * gfs2_dump_glock - print information about a glock
1770 * @seq: The seq_file struct
1773 * The file format is as follows:
1774 * One line per object, capital letters are used to indicate objects
1775 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1776 * other objects are indented by a single space and follow the glock to
1777 * which they are related. Fields are indicated by lower case letters
1778 * followed by a colon and the field value, except for strings which are in
1779 * [] so that its possible to see if they are composed of spaces for
1780 * example. The field's are n = number (id of the object), f = flags,
1781 * t = type, s = state, r = refcount, e = error, p = pid.
1785 void gfs2_dump_glock(struct seq_file
*seq
, struct gfs2_glock
*gl
)
1787 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
1788 unsigned long long dtime
;
1789 const struct gfs2_holder
*gh
;
1790 char gflags_buf
[32];
1792 dtime
= jiffies
- gl
->gl_demote_time
;
1793 dtime
*= 1000000/HZ
; /* demote time in uSec */
1794 if (!test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
1796 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",
1797 state2str(gl
->gl_state
),
1798 gl
->gl_name
.ln_type
,
1799 (unsigned long long)gl
->gl_name
.ln_number
,
1800 gflags2str(gflags_buf
, gl
),
1801 state2str(gl
->gl_target
),
1802 state2str(gl
->gl_demote_state
), dtime
,
1803 atomic_read(&gl
->gl_ail_count
),
1804 atomic_read(&gl
->gl_revokes
),
1805 (int)gl
->gl_lockref
.count
, gl
->gl_hold_time
);
1807 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
)
1808 dump_holder(seq
, gh
);
1810 if (gl
->gl_state
!= LM_ST_UNLOCKED
&& glops
->go_dump
)
1811 glops
->go_dump(seq
, gl
);
1814 static int gfs2_glstats_seq_show(struct seq_file
*seq
, void *iter_ptr
)
1816 struct gfs2_glock
*gl
= iter_ptr
;
1818 seq_printf(seq
, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
1819 gl
->gl_name
.ln_type
,
1820 (unsigned long long)gl
->gl_name
.ln_number
,
1821 (unsigned long long)gl
->gl_stats
.stats
[GFS2_LKS_SRTT
],
1822 (unsigned long long)gl
->gl_stats
.stats
[GFS2_LKS_SRTTVAR
],
1823 (unsigned long long)gl
->gl_stats
.stats
[GFS2_LKS_SRTTB
],
1824 (unsigned long long)gl
->gl_stats
.stats
[GFS2_LKS_SRTTVARB
],
1825 (unsigned long long)gl
->gl_stats
.stats
[GFS2_LKS_SIRT
],
1826 (unsigned long long)gl
->gl_stats
.stats
[GFS2_LKS_SIRTVAR
],
1827 (unsigned long long)gl
->gl_stats
.stats
[GFS2_LKS_DCOUNT
],
1828 (unsigned long long)gl
->gl_stats
.stats
[GFS2_LKS_QCOUNT
]);
1832 static const char *gfs2_gltype
[] = {
1846 static const char *gfs2_stype
[] = {
1847 [GFS2_LKS_SRTT
] = "srtt",
1848 [GFS2_LKS_SRTTVAR
] = "srttvar",
1849 [GFS2_LKS_SRTTB
] = "srttb",
1850 [GFS2_LKS_SRTTVARB
] = "srttvarb",
1851 [GFS2_LKS_SIRT
] = "sirt",
1852 [GFS2_LKS_SIRTVAR
] = "sirtvar",
1853 [GFS2_LKS_DCOUNT
] = "dlm",
1854 [GFS2_LKS_QCOUNT
] = "queue",
1857 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1859 static int gfs2_sbstats_seq_show(struct seq_file
*seq
, void *iter_ptr
)
1861 struct gfs2_sbd
*sdp
= seq
->private;
1862 loff_t pos
= *(loff_t
*)iter_ptr
;
1863 unsigned index
= pos
>> 3;
1864 unsigned subindex
= pos
& 0x07;
1867 if (index
== 0 && subindex
!= 0)
1870 seq_printf(seq
, "%-10s %8s:", gfs2_gltype
[index
],
1871 (index
== 0) ? "cpu": gfs2_stype
[subindex
]);
1873 for_each_possible_cpu(i
) {
1874 const struct gfs2_pcpu_lkstats
*lkstats
= per_cpu_ptr(sdp
->sd_lkstats
, i
);
1877 seq_printf(seq
, " %15u", i
);
1879 seq_printf(seq
, " %15llu", (unsigned long long)lkstats
->
1880 lkstats
[index
- 1].stats
[subindex
]);
1882 seq_putc(seq
, '\n');
1886 int __init
gfs2_glock_init(void)
1890 ret
= rhashtable_init(&gl_hash_table
, &ht_parms
);
1894 glock_workqueue
= alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM
|
1895 WQ_HIGHPRI
| WQ_FREEZABLE
, 0);
1896 if (!glock_workqueue
) {
1897 rhashtable_destroy(&gl_hash_table
);
1900 gfs2_delete_workqueue
= alloc_workqueue("delete_workqueue",
1901 WQ_MEM_RECLAIM
| WQ_FREEZABLE
,
1903 if (!gfs2_delete_workqueue
) {
1904 destroy_workqueue(glock_workqueue
);
1905 rhashtable_destroy(&gl_hash_table
);
1909 ret
= register_shrinker(&glock_shrinker
);
1911 destroy_workqueue(gfs2_delete_workqueue
);
1912 destroy_workqueue(glock_workqueue
);
1913 rhashtable_destroy(&gl_hash_table
);
1917 for (i
= 0; i
< GLOCK_WAIT_TABLE_SIZE
; i
++)
1918 init_waitqueue_head(glock_wait_table
+ i
);
1923 void gfs2_glock_exit(void)
1925 unregister_shrinker(&glock_shrinker
);
1926 rhashtable_destroy(&gl_hash_table
);
1927 destroy_workqueue(glock_workqueue
);
1928 destroy_workqueue(gfs2_delete_workqueue
);
1931 static void gfs2_glock_iter_next(struct gfs2_glock_iter
*gi
, loff_t n
)
1933 struct gfs2_glock
*gl
= gi
->gl
;
1938 if (!lockref_put_not_zero(&gl
->gl_lockref
))
1939 gfs2_glock_queue_put(gl
);
1942 gl
= rhashtable_walk_next(&gi
->hti
);
1943 if (IS_ERR_OR_NULL(gl
)) {
1944 if (gl
== ERR_PTR(-EAGAIN
)) {
1951 if (gl
->gl_name
.ln_sbd
!= gi
->sdp
)
1954 if (!lockref_get_not_dead(&gl
->gl_lockref
))
1958 if (__lockref_is_dead(&gl
->gl_lockref
))
1966 static void *gfs2_glock_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1969 struct gfs2_glock_iter
*gi
= seq
->private;
1973 * We can either stay where we are, skip to the next hash table
1974 * entry, or start from the beginning.
1976 if (*pos
< gi
->last_pos
) {
1977 rhashtable_walk_exit(&gi
->hti
);
1978 rhashtable_walk_enter(&gl_hash_table
, &gi
->hti
);
1981 n
= *pos
- gi
->last_pos
;
1984 rhashtable_walk_start(&gi
->hti
);
1986 gfs2_glock_iter_next(gi
, n
);
1987 gi
->last_pos
= *pos
;
1991 static void *gfs2_glock_seq_next(struct seq_file
*seq
, void *iter_ptr
,
1994 struct gfs2_glock_iter
*gi
= seq
->private;
1997 gi
->last_pos
= *pos
;
1998 gfs2_glock_iter_next(gi
, 1);
2002 static void gfs2_glock_seq_stop(struct seq_file
*seq
, void *iter_ptr
)
2005 struct gfs2_glock_iter
*gi
= seq
->private;
2007 rhashtable_walk_stop(&gi
->hti
);
2010 static int gfs2_glock_seq_show(struct seq_file
*seq
, void *iter_ptr
)
2012 dump_glock(seq
, iter_ptr
);
2016 static void *gfs2_sbstats_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2019 if (*pos
>= GFS2_NR_SBSTATS
)
2024 static void *gfs2_sbstats_seq_next(struct seq_file
*seq
, void *iter_ptr
,
2028 if (*pos
>= GFS2_NR_SBSTATS
)
2033 static void gfs2_sbstats_seq_stop(struct seq_file
*seq
, void *iter_ptr
)
2038 static const struct seq_operations gfs2_glock_seq_ops
= {
2039 .start
= gfs2_glock_seq_start
,
2040 .next
= gfs2_glock_seq_next
,
2041 .stop
= gfs2_glock_seq_stop
,
2042 .show
= gfs2_glock_seq_show
,
2045 static const struct seq_operations gfs2_glstats_seq_ops
= {
2046 .start
= gfs2_glock_seq_start
,
2047 .next
= gfs2_glock_seq_next
,
2048 .stop
= gfs2_glock_seq_stop
,
2049 .show
= gfs2_glstats_seq_show
,
2052 static const struct seq_operations gfs2_sbstats_seq_ops
= {
2053 .start
= gfs2_sbstats_seq_start
,
2054 .next
= gfs2_sbstats_seq_next
,
2055 .stop
= gfs2_sbstats_seq_stop
,
2056 .show
= gfs2_sbstats_seq_show
,
2059 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2061 static int __gfs2_glocks_open(struct inode
*inode
, struct file
*file
,
2062 const struct seq_operations
*ops
)
2064 int ret
= seq_open_private(file
, ops
, sizeof(struct gfs2_glock_iter
));
2066 struct seq_file
*seq
= file
->private_data
;
2067 struct gfs2_glock_iter
*gi
= seq
->private;
2069 gi
->sdp
= inode
->i_private
;
2070 seq
->buf
= kmalloc(GFS2_SEQ_GOODSIZE
, GFP_KERNEL
| __GFP_NOWARN
);
2072 seq
->size
= GFS2_SEQ_GOODSIZE
;
2074 * Initially, we are "before" the first hash table entry; the
2075 * first call to rhashtable_walk_next gets us the first entry.
2079 rhashtable_walk_enter(&gl_hash_table
, &gi
->hti
);
2084 static int gfs2_glocks_open(struct inode
*inode
, struct file
*file
)
2086 return __gfs2_glocks_open(inode
, file
, &gfs2_glock_seq_ops
);
2089 static int gfs2_glocks_release(struct inode
*inode
, struct file
*file
)
2091 struct seq_file
*seq
= file
->private_data
;
2092 struct gfs2_glock_iter
*gi
= seq
->private;
2095 gfs2_glock_put(gi
->gl
);
2096 rhashtable_walk_exit(&gi
->hti
);
2097 return seq_release_private(inode
, file
);
2100 static int gfs2_glstats_open(struct inode
*inode
, struct file
*file
)
2102 return __gfs2_glocks_open(inode
, file
, &gfs2_glstats_seq_ops
);
2105 static int gfs2_sbstats_open(struct inode
*inode
, struct file
*file
)
2107 int ret
= seq_open(file
, &gfs2_sbstats_seq_ops
);
2109 struct seq_file
*seq
= file
->private_data
;
2110 seq
->private = inode
->i_private
; /* sdp */
2115 static const struct file_operations gfs2_glocks_fops
= {
2116 .owner
= THIS_MODULE
,
2117 .open
= gfs2_glocks_open
,
2119 .llseek
= seq_lseek
,
2120 .release
= gfs2_glocks_release
,
2123 static const struct file_operations gfs2_glstats_fops
= {
2124 .owner
= THIS_MODULE
,
2125 .open
= gfs2_glstats_open
,
2127 .llseek
= seq_lseek
,
2128 .release
= gfs2_glocks_release
,
2131 static const struct file_operations gfs2_sbstats_fops
= {
2132 .owner
= THIS_MODULE
,
2133 .open
= gfs2_sbstats_open
,
2135 .llseek
= seq_lseek
,
2136 .release
= seq_release
,
2139 void gfs2_create_debugfs_file(struct gfs2_sbd
*sdp
)
2141 sdp
->debugfs_dir
= debugfs_create_dir(sdp
->sd_table_name
, gfs2_root
);
2143 debugfs_create_file("glocks", S_IFREG
| S_IRUGO
, sdp
->debugfs_dir
, sdp
,
2146 debugfs_create_file("glstats", S_IFREG
| S_IRUGO
, sdp
->debugfs_dir
, sdp
,
2147 &gfs2_glstats_fops
);
2149 debugfs_create_file("sbstats", S_IFREG
| S_IRUGO
, sdp
->debugfs_dir
, sdp
,
2150 &gfs2_sbstats_fops
);
2153 void gfs2_delete_debugfs_file(struct gfs2_sbd
*sdp
)
2155 debugfs_remove_recursive(sdp
->debugfs_dir
);
2156 sdp
->debugfs_dir
= NULL
;
2159 void gfs2_register_debugfs(void)
2161 gfs2_root
= debugfs_create_dir("gfs2", NULL
);
2164 void gfs2_unregister_debugfs(void)
2166 debugfs_remove(gfs2_root
);