1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 #include <linux/sched.h>
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/buffer_head.h>
13 #include <linux/delay.h>
14 #include <linux/sort.h>
15 #include <linux/hash.h>
16 #include <linux/jhash.h>
17 #include <linux/kallsyms.h>
18 #include <linux/gfs2_ondisk.h>
19 #include <linux/list.h>
20 #include <linux/wait.h>
21 #include <linux/module.h>
22 #include <linux/uaccess.h>
23 #include <linux/seq_file.h>
24 #include <linux/debugfs.h>
25 #include <linux/kthread.h>
26 #include <linux/freezer.h>
27 #include <linux/workqueue.h>
28 #include <linux/jiffies.h>
29 #include <linux/rcupdate.h>
30 #include <linux/rculist_bl.h>
31 #include <linux/bit_spinlock.h>
32 #include <linux/percpu.h>
33 #include <linux/list_sort.h>
34 #include <linux/lockref.h>
35 #include <linux/rhashtable.h>
48 #define CREATE_TRACE_POINTS
49 #include "trace_gfs2.h"
51 struct gfs2_glock_iter
{
52 struct gfs2_sbd
*sdp
; /* incore superblock */
53 struct rhashtable_iter hti
; /* rhashtable iterator */
54 struct gfs2_glock
*gl
; /* current glock struct */
55 loff_t last_pos
; /* last position */
58 typedef void (*glock_examiner
) (struct gfs2_glock
* gl
);
60 static void do_xmote(struct gfs2_glock
*gl
, struct gfs2_holder
*gh
, unsigned int target
);
62 static struct dentry
*gfs2_root
;
63 static struct workqueue_struct
*glock_workqueue
;
64 struct workqueue_struct
*gfs2_delete_workqueue
;
65 static LIST_HEAD(lru_list
);
66 static atomic_t lru_count
= ATOMIC_INIT(0);
67 static DEFINE_SPINLOCK(lru_lock
);
69 #define GFS2_GL_HASH_SHIFT 15
70 #define GFS2_GL_HASH_SIZE BIT(GFS2_GL_HASH_SHIFT)
72 static const struct rhashtable_params ht_parms
= {
73 .nelem_hint
= GFS2_GL_HASH_SIZE
* 3 / 4,
74 .key_len
= offsetofend(struct lm_lockname
, ln_type
),
75 .key_offset
= offsetof(struct gfs2_glock
, gl_name
),
76 .head_offset
= offsetof(struct gfs2_glock
, gl_node
),
79 static struct rhashtable gl_hash_table
;
81 #define GLOCK_WAIT_TABLE_BITS 12
82 #define GLOCK_WAIT_TABLE_SIZE (1 << GLOCK_WAIT_TABLE_BITS)
83 static wait_queue_head_t glock_wait_table
[GLOCK_WAIT_TABLE_SIZE
] __cacheline_aligned
;
85 struct wait_glock_queue
{
86 struct lm_lockname
*name
;
87 wait_queue_entry_t wait
;
90 static int glock_wake_function(wait_queue_entry_t
*wait
, unsigned int mode
,
93 struct wait_glock_queue
*wait_glock
=
94 container_of(wait
, struct wait_glock_queue
, wait
);
95 struct lm_lockname
*wait_name
= wait_glock
->name
;
96 struct lm_lockname
*wake_name
= key
;
98 if (wake_name
->ln_sbd
!= wait_name
->ln_sbd
||
99 wake_name
->ln_number
!= wait_name
->ln_number
||
100 wake_name
->ln_type
!= wait_name
->ln_type
)
102 return autoremove_wake_function(wait
, mode
, sync
, key
);
105 static wait_queue_head_t
*glock_waitqueue(struct lm_lockname
*name
)
107 u32 hash
= jhash2((u32
*)name
, ht_parms
.key_len
/ 4, 0);
109 return glock_wait_table
+ hash_32(hash
, GLOCK_WAIT_TABLE_BITS
);
113 * wake_up_glock - Wake up waiters on a glock
116 static void wake_up_glock(struct gfs2_glock
*gl
)
118 wait_queue_head_t
*wq
= glock_waitqueue(&gl
->gl_name
);
120 if (waitqueue_active(wq
))
121 __wake_up(wq
, TASK_NORMAL
, 1, &gl
->gl_name
);
124 static void gfs2_glock_dealloc(struct rcu_head
*rcu
)
126 struct gfs2_glock
*gl
= container_of(rcu
, struct gfs2_glock
, gl_rcu
);
128 kfree(gl
->gl_lksb
.sb_lvbptr
);
129 if (gl
->gl_ops
->go_flags
& GLOF_ASPACE
)
130 kmem_cache_free(gfs2_glock_aspace_cachep
, gl
);
132 kmem_cache_free(gfs2_glock_cachep
, gl
);
136 * glock_blocked_by_withdraw - determine if we can still use a glock
139 * We need to allow some glocks to be enqueued, dequeued, promoted, and demoted
140 * when we're withdrawn. For example, to maintain metadata integrity, we should
141 * disallow the use of inode and rgrp glocks when withdrawn. Other glocks, like
142 * iopen or the transaction glocks may be safely used because none of their
143 * metadata goes through the journal. So in general, we should disallow all
144 * glocks that are journaled, and allow all the others. One exception is:
145 * we need to allow our active journal to be promoted and demoted so others
146 * may recover it and we can reacquire it when they're done.
148 static bool glock_blocked_by_withdraw(struct gfs2_glock
*gl
)
150 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
152 if (likely(!gfs2_withdrawn(sdp
)))
154 if (gl
->gl_ops
->go_flags
& GLOF_NONDISK
)
156 if (!sdp
->sd_jdesc
||
157 gl
->gl_name
.ln_number
== sdp
->sd_jdesc
->jd_no_addr
)
162 void gfs2_glock_free(struct gfs2_glock
*gl
)
164 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
166 gfs2_glock_assert_withdraw(gl
, atomic_read(&gl
->gl_revokes
) == 0);
167 rhashtable_remove_fast(&gl_hash_table
, &gl
->gl_node
, ht_parms
);
170 call_rcu(&gl
->gl_rcu
, gfs2_glock_dealloc
);
171 if (atomic_dec_and_test(&sdp
->sd_glock_disposal
))
172 wake_up(&sdp
->sd_glock_wait
);
176 * gfs2_glock_hold() - increment reference count on glock
177 * @gl: The glock to hold
181 void gfs2_glock_hold(struct gfs2_glock
*gl
)
183 GLOCK_BUG_ON(gl
, __lockref_is_dead(&gl
->gl_lockref
));
184 lockref_get(&gl
->gl_lockref
);
188 * demote_ok - Check to see if it's ok to unlock a glock
191 * Returns: 1 if it's ok
194 static int demote_ok(const struct gfs2_glock
*gl
)
196 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
198 if (gl
->gl_state
== LM_ST_UNLOCKED
)
200 if (!list_empty(&gl
->gl_holders
))
202 if (glops
->go_demote_ok
)
203 return glops
->go_demote_ok(gl
);
208 void gfs2_glock_add_to_lru(struct gfs2_glock
*gl
)
210 if (!(gl
->gl_ops
->go_flags
& GLOF_LRU
))
213 spin_lock(&lru_lock
);
215 list_del(&gl
->gl_lru
);
216 list_add_tail(&gl
->gl_lru
, &lru_list
);
218 if (!test_bit(GLF_LRU
, &gl
->gl_flags
)) {
219 set_bit(GLF_LRU
, &gl
->gl_flags
);
220 atomic_inc(&lru_count
);
223 spin_unlock(&lru_lock
);
226 static void gfs2_glock_remove_from_lru(struct gfs2_glock
*gl
)
228 if (!(gl
->gl_ops
->go_flags
& GLOF_LRU
))
231 spin_lock(&lru_lock
);
232 if (test_bit(GLF_LRU
, &gl
->gl_flags
)) {
233 list_del_init(&gl
->gl_lru
);
234 atomic_dec(&lru_count
);
235 clear_bit(GLF_LRU
, &gl
->gl_flags
);
237 spin_unlock(&lru_lock
);
241 * Enqueue the glock on the work queue. Passes one glock reference on to the
244 static void __gfs2_glock_queue_work(struct gfs2_glock
*gl
, unsigned long delay
) {
245 if (!queue_delayed_work(glock_workqueue
, &gl
->gl_work
, delay
)) {
247 * We are holding the lockref spinlock, and the work was still
248 * queued above. The queued work (glock_work_func) takes that
249 * spinlock before dropping its glock reference(s), so it
250 * cannot have dropped them in the meantime.
252 GLOCK_BUG_ON(gl
, gl
->gl_lockref
.count
< 2);
253 gl
->gl_lockref
.count
--;
257 static void gfs2_glock_queue_work(struct gfs2_glock
*gl
, unsigned long delay
) {
258 spin_lock(&gl
->gl_lockref
.lock
);
259 __gfs2_glock_queue_work(gl
, delay
);
260 spin_unlock(&gl
->gl_lockref
.lock
);
263 static void __gfs2_glock_put(struct gfs2_glock
*gl
)
265 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
266 struct address_space
*mapping
= gfs2_glock2aspace(gl
);
268 lockref_mark_dead(&gl
->gl_lockref
);
270 gfs2_glock_remove_from_lru(gl
);
271 spin_unlock(&gl
->gl_lockref
.lock
);
272 GLOCK_BUG_ON(gl
, !list_empty(&gl
->gl_holders
));
273 GLOCK_BUG_ON(gl
, mapping
&& mapping
->nrpages
&& !gfs2_withdrawn(sdp
));
274 trace_gfs2_glock_put(gl
);
275 sdp
->sd_lockstruct
.ls_ops
->lm_put_lock(gl
);
279 * Cause the glock to be put in work queue context.
281 void gfs2_glock_queue_put(struct gfs2_glock
*gl
)
283 gfs2_glock_queue_work(gl
, 0);
287 * gfs2_glock_put() - Decrement reference count on glock
288 * @gl: The glock to put
292 void gfs2_glock_put(struct gfs2_glock
*gl
)
294 if (lockref_put_or_lock(&gl
->gl_lockref
))
297 __gfs2_glock_put(gl
);
301 * may_grant - check if its ok to grant a new lock
303 * @gh: The lock request which we wish to grant
305 * Returns: true if its ok to grant the lock
308 static inline int may_grant(const struct gfs2_glock
*gl
, const struct gfs2_holder
*gh
)
310 const struct gfs2_holder
*gh_head
= list_first_entry(&gl
->gl_holders
, const struct gfs2_holder
, gh_list
);
311 if ((gh
->gh_state
== LM_ST_EXCLUSIVE
||
312 gh_head
->gh_state
== LM_ST_EXCLUSIVE
) && gh
!= gh_head
)
314 if (gl
->gl_state
== gh
->gh_state
)
316 if (gh
->gh_flags
& GL_EXACT
)
318 if (gl
->gl_state
== LM_ST_EXCLUSIVE
) {
319 if (gh
->gh_state
== LM_ST_SHARED
&& gh_head
->gh_state
== LM_ST_SHARED
)
321 if (gh
->gh_state
== LM_ST_DEFERRED
&& gh_head
->gh_state
== LM_ST_DEFERRED
)
324 if (gl
->gl_state
!= LM_ST_UNLOCKED
&& (gh
->gh_flags
& LM_FLAG_ANY
))
329 static void gfs2_holder_wake(struct gfs2_holder
*gh
)
331 clear_bit(HIF_WAIT
, &gh
->gh_iflags
);
332 smp_mb__after_atomic();
333 wake_up_bit(&gh
->gh_iflags
, HIF_WAIT
);
334 if (gh
->gh_flags
& GL_ASYNC
) {
335 struct gfs2_sbd
*sdp
= gh
->gh_gl
->gl_name
.ln_sbd
;
337 wake_up(&sdp
->sd_async_glock_wait
);
342 * do_error - Something unexpected has happened during a lock request
346 static void do_error(struct gfs2_glock
*gl
, const int ret
)
348 struct gfs2_holder
*gh
, *tmp
;
350 list_for_each_entry_safe(gh
, tmp
, &gl
->gl_holders
, gh_list
) {
351 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
353 if (ret
& LM_OUT_ERROR
)
355 else if (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
))
356 gh
->gh_error
= GLR_TRYFAILED
;
359 list_del_init(&gh
->gh_list
);
360 trace_gfs2_glock_queue(gh
, 0);
361 gfs2_holder_wake(gh
);
366 * do_promote - promote as many requests as possible on the current queue
369 * Returns: 1 if there is a blocked holder at the head of the list, or 2
370 * if a type specific operation is underway.
373 static int do_promote(struct gfs2_glock
*gl
)
374 __releases(&gl
->gl_lockref
.lock
)
375 __acquires(&gl
->gl_lockref
.lock
)
377 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
378 struct gfs2_holder
*gh
, *tmp
;
382 list_for_each_entry_safe(gh
, tmp
, &gl
->gl_holders
, gh_list
) {
383 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
385 if (may_grant(gl
, gh
)) {
386 if (gh
->gh_list
.prev
== &gl
->gl_holders
&&
388 spin_unlock(&gl
->gl_lockref
.lock
);
389 /* FIXME: eliminate this eventually */
390 ret
= glops
->go_lock(gh
);
391 spin_lock(&gl
->gl_lockref
.lock
);
396 list_del_init(&gh
->gh_list
);
397 trace_gfs2_glock_queue(gh
, 0);
398 gfs2_holder_wake(gh
);
401 set_bit(HIF_HOLDER
, &gh
->gh_iflags
);
402 trace_gfs2_promote(gh
, 1);
403 gfs2_holder_wake(gh
);
406 set_bit(HIF_HOLDER
, &gh
->gh_iflags
);
407 trace_gfs2_promote(gh
, 0);
408 gfs2_holder_wake(gh
);
411 if (gh
->gh_list
.prev
== &gl
->gl_holders
)
420 * find_first_waiter - find the first gh that's waiting for the glock
424 static inline struct gfs2_holder
*find_first_waiter(const struct gfs2_glock
*gl
)
426 struct gfs2_holder
*gh
;
428 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
) {
429 if (!test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
436 * state_change - record that the glock is now in a different state
438 * @new_state the new state
442 static void state_change(struct gfs2_glock
*gl
, unsigned int new_state
)
446 held1
= (gl
->gl_state
!= LM_ST_UNLOCKED
);
447 held2
= (new_state
!= LM_ST_UNLOCKED
);
449 if (held1
!= held2
) {
450 GLOCK_BUG_ON(gl
, __lockref_is_dead(&gl
->gl_lockref
));
452 gl
->gl_lockref
.count
++;
454 gl
->gl_lockref
.count
--;
456 if (held1
&& held2
&& list_empty(&gl
->gl_holders
))
457 clear_bit(GLF_QUEUED
, &gl
->gl_flags
);
459 if (new_state
!= gl
->gl_target
)
460 /* shorten our minimum hold time */
461 gl
->gl_hold_time
= max(gl
->gl_hold_time
- GL_GLOCK_HOLD_DECR
,
463 gl
->gl_state
= new_state
;
464 gl
->gl_tchange
= jiffies
;
467 static void gfs2_set_demote(struct gfs2_glock
*gl
)
469 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
471 set_bit(GLF_DEMOTE
, &gl
->gl_flags
);
473 wake_up(&sdp
->sd_async_glock_wait
);
476 static void gfs2_demote_wake(struct gfs2_glock
*gl
)
478 gl
->gl_demote_state
= LM_ST_EXCLUSIVE
;
479 clear_bit(GLF_DEMOTE
, &gl
->gl_flags
);
480 smp_mb__after_atomic();
481 wake_up_bit(&gl
->gl_flags
, GLF_DEMOTE
);
485 * finish_xmote - The DLM has replied to one of our lock requests
487 * @ret: The status from the DLM
491 static void finish_xmote(struct gfs2_glock
*gl
, unsigned int ret
)
493 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
494 struct gfs2_holder
*gh
;
495 unsigned state
= ret
& LM_OUT_ST_MASK
;
498 spin_lock(&gl
->gl_lockref
.lock
);
499 trace_gfs2_glock_state_change(gl
, state
);
500 state_change(gl
, state
);
501 gh
= find_first_waiter(gl
);
503 /* Demote to UN request arrived during demote to SH or DF */
504 if (test_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
) &&
505 state
!= LM_ST_UNLOCKED
&& gl
->gl_demote_state
== LM_ST_UNLOCKED
)
506 gl
->gl_target
= LM_ST_UNLOCKED
;
508 /* Check for state != intended state */
509 if (unlikely(state
!= gl
->gl_target
)) {
510 if (gh
&& !test_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
)) {
511 /* move to back of queue and try next entry */
512 if (ret
& LM_OUT_CANCELED
) {
513 if ((gh
->gh_flags
& LM_FLAG_PRIORITY
) == 0)
514 list_move_tail(&gh
->gh_list
, &gl
->gl_holders
);
515 gh
= find_first_waiter(gl
);
516 gl
->gl_target
= gh
->gh_state
;
519 /* Some error or failed "try lock" - report it */
520 if ((ret
& LM_OUT_ERROR
) ||
521 (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
))) {
522 gl
->gl_target
= gl
->gl_state
;
528 /* Unlocked due to conversion deadlock, try again */
531 do_xmote(gl
, gh
, gl
->gl_target
);
533 /* Conversion fails, unlock and try again */
536 do_xmote(gl
, gh
, LM_ST_UNLOCKED
);
538 default: /* Everything else */
539 fs_err(gl
->gl_name
.ln_sbd
, "wanted %u got %u\n",
540 gl
->gl_target
, state
);
543 spin_unlock(&gl
->gl_lockref
.lock
);
547 /* Fast path - we got what we asked for */
548 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
))
549 gfs2_demote_wake(gl
);
550 if (state
!= LM_ST_UNLOCKED
) {
551 if (glops
->go_xmote_bh
) {
552 spin_unlock(&gl
->gl_lockref
.lock
);
553 rv
= glops
->go_xmote_bh(gl
, gh
);
554 spin_lock(&gl
->gl_lockref
.lock
);
565 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
567 spin_unlock(&gl
->gl_lockref
.lock
);
571 * do_xmote - Calls the DLM to change the state of a lock
572 * @gl: The lock state
573 * @gh: The holder (only for promotes)
574 * @target: The target lock state
578 static void do_xmote(struct gfs2_glock
*gl
, struct gfs2_holder
*gh
, unsigned int target
)
579 __releases(&gl
->gl_lockref
.lock
)
580 __acquires(&gl
->gl_lockref
.lock
)
582 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
583 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
584 unsigned int lck_flags
= (unsigned int)(gh
? gh
->gh_flags
: 0);
587 if (target
!= LM_ST_UNLOCKED
&& glock_blocked_by_withdraw(gl
) &&
588 gh
&& !(gh
->gh_flags
& LM_FLAG_NOEXP
))
590 lck_flags
&= (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
| LM_FLAG_NOEXP
|
592 GLOCK_BUG_ON(gl
, gl
->gl_state
== target
);
593 GLOCK_BUG_ON(gl
, gl
->gl_state
== gl
->gl_target
);
594 if ((target
== LM_ST_UNLOCKED
|| target
== LM_ST_DEFERRED
) &&
597 * If another process is already doing the invalidate, let that
598 * finish first. The glock state machine will get back to this
599 * holder again later.
601 if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS
,
604 do_error(gl
, 0); /* Fail queued try locks */
607 set_bit(GLF_BLOCKING
, &gl
->gl_flags
);
608 if ((gl
->gl_req
== LM_ST_UNLOCKED
) ||
609 (gl
->gl_state
== LM_ST_EXCLUSIVE
) ||
610 (lck_flags
& (LM_FLAG_TRY
|LM_FLAG_TRY_1CB
)))
611 clear_bit(GLF_BLOCKING
, &gl
->gl_flags
);
612 spin_unlock(&gl
->gl_lockref
.lock
);
613 if (glops
->go_sync
) {
614 ret
= glops
->go_sync(gl
);
615 /* If we had a problem syncing (due to io errors or whatever,
616 * we should not invalidate the metadata or tell dlm to
617 * release the glock to other nodes.
620 if (cmpxchg(&sdp
->sd_log_error
, 0, ret
)) {
621 fs_err(sdp
, "Error %d syncing glock \n", ret
);
622 gfs2_dump_glock(NULL
, gl
, true);
627 if (test_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
)) {
629 * The call to go_sync should have cleared out the ail list.
630 * If there are still items, we have a problem. We ought to
631 * withdraw, but we can't because the withdraw code also uses
632 * glocks. Warn about the error, dump the glock, then fall
633 * through and wait for logd to do the withdraw for us.
635 if ((atomic_read(&gl
->gl_ail_count
) != 0) &&
636 (!cmpxchg(&sdp
->sd_log_error
, 0, -EIO
))) {
637 gfs2_glock_assert_warn(gl
,
638 !atomic_read(&gl
->gl_ail_count
));
639 gfs2_dump_glock(NULL
, gl
, true);
641 glops
->go_inval(gl
, target
== LM_ST_DEFERRED
? 0 : DIO_METADATA
);
642 clear_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
);
648 * Check for an error encountered since we called go_sync and go_inval.
649 * If so, we can't withdraw from the glock code because the withdraw
650 * code itself uses glocks (see function signal_our_withdraw) to
651 * change the mount to read-only. Most importantly, we must not call
652 * dlm to unlock the glock until the journal is in a known good state
653 * (after journal replay) otherwise other nodes may use the object
654 * (rgrp or dinode) and then later, journal replay will corrupt the
655 * file system. The best we can do here is wait for the logd daemon
656 * to see sd_log_error and withdraw, and in the meantime, requeue the
659 * However, if we're just unlocking the lock (say, for unmount, when
660 * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
661 * then it's okay to tell dlm to unlock it.
663 if (unlikely(sdp
->sd_log_error
&& !gfs2_withdrawn(sdp
)))
664 gfs2_withdraw_delayed(sdp
);
665 if (glock_blocked_by_withdraw(gl
)) {
666 if (target
!= LM_ST_UNLOCKED
||
667 test_bit(SDF_WITHDRAW_RECOVERY
, &sdp
->sd_flags
)) {
668 gfs2_glock_queue_work(gl
, GL_GLOCK_DFT_HOLD
);
673 if (sdp
->sd_lockstruct
.ls_ops
->lm_lock
) {
675 ret
= sdp
->sd_lockstruct
.ls_ops
->lm_lock(gl
, target
, lck_flags
);
676 if (ret
== -EINVAL
&& gl
->gl_target
== LM_ST_UNLOCKED
&&
677 target
== LM_ST_UNLOCKED
&&
678 test_bit(SDF_SKIP_DLM_UNLOCK
, &sdp
->sd_flags
)) {
679 finish_xmote(gl
, target
);
680 gfs2_glock_queue_work(gl
, 0);
682 fs_err(sdp
, "lm_lock ret %d\n", ret
);
683 GLOCK_BUG_ON(gl
, !gfs2_withdrawn(sdp
));
685 } else { /* lock_nolock */
686 finish_xmote(gl
, target
);
687 gfs2_glock_queue_work(gl
, 0);
690 spin_lock(&gl
->gl_lockref
.lock
);
694 * find_first_holder - find the first "holder" gh
698 static inline struct gfs2_holder
*find_first_holder(const struct gfs2_glock
*gl
)
700 struct gfs2_holder
*gh
;
702 if (!list_empty(&gl
->gl_holders
)) {
703 gh
= list_first_entry(&gl
->gl_holders
, struct gfs2_holder
, gh_list
);
704 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
711 * run_queue - do all outstanding tasks related to a glock
712 * @gl: The glock in question
713 * @nonblock: True if we must not block in run_queue
717 static void run_queue(struct gfs2_glock
*gl
, const int nonblock
)
718 __releases(&gl
->gl_lockref
.lock
)
719 __acquires(&gl
->gl_lockref
.lock
)
721 struct gfs2_holder
*gh
= NULL
;
724 if (test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
))
727 GLOCK_BUG_ON(gl
, test_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
));
729 if (test_bit(GLF_DEMOTE
, &gl
->gl_flags
) &&
730 gl
->gl_demote_state
!= gl
->gl_state
) {
731 if (find_first_holder(gl
))
735 set_bit(GLF_DEMOTE_IN_PROGRESS
, &gl
->gl_flags
);
736 GLOCK_BUG_ON(gl
, gl
->gl_demote_state
== LM_ST_EXCLUSIVE
);
737 gl
->gl_target
= gl
->gl_demote_state
;
739 if (test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
740 gfs2_demote_wake(gl
);
741 ret
= do_promote(gl
);
746 gh
= find_first_waiter(gl
);
747 gl
->gl_target
= gh
->gh_state
;
748 if (!(gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
)))
749 do_error(gl
, 0); /* Fail queued try locks */
751 do_xmote(gl
, gh
, gl
->gl_target
);
756 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
757 smp_mb__after_atomic();
758 gl
->gl_lockref
.count
++;
759 __gfs2_glock_queue_work(gl
, 0);
763 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
764 smp_mb__after_atomic();
768 void gfs2_inode_remember_delete(struct gfs2_glock
*gl
, u64 generation
)
770 struct gfs2_inode_lvb
*ri
= (void *)gl
->gl_lksb
.sb_lvbptr
;
772 if (ri
->ri_magic
== 0)
773 ri
->ri_magic
= cpu_to_be32(GFS2_MAGIC
);
774 if (ri
->ri_magic
== cpu_to_be32(GFS2_MAGIC
))
775 ri
->ri_generation_deleted
= cpu_to_be64(generation
);
778 bool gfs2_inode_already_deleted(struct gfs2_glock
*gl
, u64 generation
)
780 struct gfs2_inode_lvb
*ri
= (void *)gl
->gl_lksb
.sb_lvbptr
;
782 if (ri
->ri_magic
!= cpu_to_be32(GFS2_MAGIC
))
784 return generation
<= be64_to_cpu(ri
->ri_generation_deleted
);
787 static void gfs2_glock_poke(struct gfs2_glock
*gl
)
789 int flags
= LM_FLAG_TRY_1CB
| LM_FLAG_ANY
| GL_SKIP
;
790 struct gfs2_holder gh
;
793 gfs2_holder_init(gl
, LM_ST_SHARED
, flags
, &gh
);
794 error
= gfs2_glock_nq(&gh
);
797 gfs2_holder_uninit(&gh
);
800 static bool gfs2_try_evict(struct gfs2_glock
*gl
)
802 struct gfs2_inode
*ip
;
803 bool evicted
= false;
806 * If there is contention on the iopen glock and we have an inode, try
807 * to grab and release the inode so that it can be evicted. This will
808 * allow the remote node to go ahead and delete the inode without us
809 * having to do it, which will avoid rgrp glock thrashing.
811 * The remote node is likely still holding the corresponding inode
812 * glock, so it will run before we get to verify that the delete has
815 spin_lock(&gl
->gl_lockref
.lock
);
817 if (ip
&& !igrab(&ip
->i_inode
))
819 spin_unlock(&gl
->gl_lockref
.lock
);
821 struct gfs2_glock
*inode_gl
= NULL
;
823 gl
->gl_no_formal_ino
= ip
->i_no_formal_ino
;
824 set_bit(GIF_DEFERRED_DELETE
, &ip
->i_flags
);
825 d_prune_aliases(&ip
->i_inode
);
828 /* If the inode was evicted, gl->gl_object will now be NULL. */
829 spin_lock(&gl
->gl_lockref
.lock
);
833 lockref_get(&inode_gl
->gl_lockref
);
834 clear_bit(GIF_DEFERRED_DELETE
, &ip
->i_flags
);
836 spin_unlock(&gl
->gl_lockref
.lock
);
838 gfs2_glock_poke(inode_gl
);
839 gfs2_glock_put(inode_gl
);
846 static void delete_work_func(struct work_struct
*work
)
848 struct delayed_work
*dwork
= to_delayed_work(work
);
849 struct gfs2_glock
*gl
= container_of(dwork
, struct gfs2_glock
, gl_delete
);
850 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
852 u64 no_addr
= gl
->gl_name
.ln_number
;
854 spin_lock(&gl
->gl_lockref
.lock
);
855 clear_bit(GLF_PENDING_DELETE
, &gl
->gl_flags
);
856 spin_unlock(&gl
->gl_lockref
.lock
);
858 /* If someone's using this glock to create a new dinode, the block must
859 have been freed by another node, then re-used, in which case our
860 iopen callback is too late after the fact. Ignore it. */
861 if (test_bit(GLF_INODE_CREATING
, &gl
->gl_flags
))
864 if (test_bit(GLF_DEMOTE
, &gl
->gl_flags
)) {
866 * If we can evict the inode, give the remote node trying to
867 * delete the inode some time before verifying that the delete
868 * has happened. Otherwise, if we cause contention on the inode glock
869 * immediately, the remote node will think that we still have
870 * the inode in use, and so it will give up waiting.
872 * If we can't evict the inode, signal to the remote node that
873 * the inode is still in use. We'll later try to delete the
874 * inode locally in gfs2_evict_inode.
876 * FIXME: We only need to verify that the remote node has
877 * deleted the inode because nodes before this remote delete
878 * rework won't cooperate. At a later time, when we no longer
879 * care about compatibility with such nodes, we can skip this
882 if (gfs2_try_evict(gl
)) {
883 if (gfs2_queue_delete_work(gl
, 5 * HZ
))
889 inode
= gfs2_lookup_by_inum(sdp
, no_addr
, gl
->gl_no_formal_ino
,
890 GFS2_BLKST_UNLINKED
);
891 if (!IS_ERR_OR_NULL(inode
)) {
892 d_prune_aliases(inode
);
899 static void glock_work_func(struct work_struct
*work
)
901 unsigned long delay
= 0;
902 struct gfs2_glock
*gl
= container_of(work
, struct gfs2_glock
, gl_work
.work
);
903 unsigned int drop_refs
= 1;
905 if (test_and_clear_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
)) {
906 finish_xmote(gl
, gl
->gl_reply
);
909 spin_lock(&gl
->gl_lockref
.lock
);
910 if (test_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
) &&
911 gl
->gl_state
!= LM_ST_UNLOCKED
&&
912 gl
->gl_demote_state
!= LM_ST_EXCLUSIVE
) {
913 unsigned long holdtime
, now
= jiffies
;
915 holdtime
= gl
->gl_tchange
+ gl
->gl_hold_time
;
916 if (time_before(now
, holdtime
))
917 delay
= holdtime
- now
;
920 clear_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
);
926 /* Keep one glock reference for the work we requeue. */
928 if (gl
->gl_name
.ln_type
!= LM_TYPE_INODE
)
930 __gfs2_glock_queue_work(gl
, delay
);
934 * Drop the remaining glock references manually here. (Mind that
935 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
938 gl
->gl_lockref
.count
-= drop_refs
;
939 if (!gl
->gl_lockref
.count
) {
940 __gfs2_glock_put(gl
);
943 spin_unlock(&gl
->gl_lockref
.lock
);
946 static struct gfs2_glock
*find_insert_glock(struct lm_lockname
*name
,
947 struct gfs2_glock
*new)
949 struct wait_glock_queue wait
;
950 wait_queue_head_t
*wq
= glock_waitqueue(name
);
951 struct gfs2_glock
*gl
;
954 init_wait(&wait
.wait
);
955 wait
.wait
.func
= glock_wake_function
;
958 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
961 gl
= rhashtable_lookup_get_insert_fast(&gl_hash_table
,
962 &new->gl_node
, ht_parms
);
966 gl
= rhashtable_lookup_fast(&gl_hash_table
,
969 if (gl
&& !lockref_get_not_dead(&gl
->gl_lockref
)) {
976 finish_wait(wq
, &wait
.wait
);
981 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
982 * @sdp: The GFS2 superblock
983 * @number: the lock number
984 * @glops: The glock_operations to use
985 * @create: If 0, don't create the glock if it doesn't exist
986 * @glp: the glock is returned here
988 * This does not lock a glock, just finds/creates structures for one.
993 int gfs2_glock_get(struct gfs2_sbd
*sdp
, u64 number
,
994 const struct gfs2_glock_operations
*glops
, int create
,
995 struct gfs2_glock
**glp
)
997 struct super_block
*s
= sdp
->sd_vfs
;
998 struct lm_lockname name
= { .ln_number
= number
,
999 .ln_type
= glops
->go_type
,
1001 struct gfs2_glock
*gl
, *tmp
;
1002 struct address_space
*mapping
;
1003 struct kmem_cache
*cachep
;
1006 gl
= find_insert_glock(&name
, NULL
);
1014 if (glops
->go_flags
& GLOF_ASPACE
)
1015 cachep
= gfs2_glock_aspace_cachep
;
1017 cachep
= gfs2_glock_cachep
;
1018 gl
= kmem_cache_alloc(cachep
, GFP_NOFS
);
1022 memset(&gl
->gl_lksb
, 0, sizeof(struct dlm_lksb
));
1024 if (glops
->go_flags
& GLOF_LVB
) {
1025 gl
->gl_lksb
.sb_lvbptr
= kzalloc(GDLM_LVB_SIZE
, GFP_NOFS
);
1026 if (!gl
->gl_lksb
.sb_lvbptr
) {
1027 kmem_cache_free(cachep
, gl
);
1032 atomic_inc(&sdp
->sd_glock_disposal
);
1033 gl
->gl_node
.next
= NULL
;
1036 gl
->gl_lockref
.count
= 1;
1037 gl
->gl_state
= LM_ST_UNLOCKED
;
1038 gl
->gl_target
= LM_ST_UNLOCKED
;
1039 gl
->gl_demote_state
= LM_ST_EXCLUSIVE
;
1043 /* We use the global stats to estimate the initial per-glock stats */
1044 gl
->gl_stats
= this_cpu_ptr(sdp
->sd_lkstats
)->lkstats
[glops
->go_type
];
1046 gl
->gl_stats
.stats
[GFS2_LKS_DCOUNT
] = 0;
1047 gl
->gl_stats
.stats
[GFS2_LKS_QCOUNT
] = 0;
1048 gl
->gl_tchange
= jiffies
;
1049 gl
->gl_object
= NULL
;
1050 gl
->gl_hold_time
= GL_GLOCK_DFT_HOLD
;
1051 INIT_DELAYED_WORK(&gl
->gl_work
, glock_work_func
);
1052 INIT_DELAYED_WORK(&gl
->gl_delete
, delete_work_func
);
1054 mapping
= gfs2_glock2aspace(gl
);
1056 mapping
->a_ops
= &gfs2_meta_aops
;
1057 mapping
->host
= s
->s_bdev
->bd_inode
;
1059 mapping_set_gfp_mask(mapping
, GFP_NOFS
);
1060 mapping
->private_data
= NULL
;
1061 mapping
->writeback_index
= 0;
1064 tmp
= find_insert_glock(&name
, gl
);
1076 kfree(gl
->gl_lksb
.sb_lvbptr
);
1077 kmem_cache_free(cachep
, gl
);
1078 atomic_dec(&sdp
->sd_glock_disposal
);
1085 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
1087 * @state: the state we're requesting
1088 * @flags: the modifier flags
1089 * @gh: the holder structure
1093 void gfs2_holder_init(struct gfs2_glock
*gl
, unsigned int state
, u16 flags
,
1094 struct gfs2_holder
*gh
)
1096 INIT_LIST_HEAD(&gh
->gh_list
);
1098 gh
->gh_ip
= _RET_IP_
;
1099 gh
->gh_owner_pid
= get_pid(task_pid(current
));
1100 gh
->gh_state
= state
;
1101 gh
->gh_flags
= flags
;
1104 gfs2_glock_hold(gl
);
1108 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1109 * @state: the state we're requesting
1110 * @flags: the modifier flags
1111 * @gh: the holder structure
1113 * Don't mess with the glock.
1117 void gfs2_holder_reinit(unsigned int state
, u16 flags
, struct gfs2_holder
*gh
)
1119 gh
->gh_state
= state
;
1120 gh
->gh_flags
= flags
;
1122 gh
->gh_ip
= _RET_IP_
;
1123 put_pid(gh
->gh_owner_pid
);
1124 gh
->gh_owner_pid
= get_pid(task_pid(current
));
1128 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1129 * @gh: the holder structure
1133 void gfs2_holder_uninit(struct gfs2_holder
*gh
)
1135 put_pid(gh
->gh_owner_pid
);
1136 gfs2_glock_put(gh
->gh_gl
);
1137 gfs2_holder_mark_uninitialized(gh
);
1141 static void gfs2_glock_update_hold_time(struct gfs2_glock
*gl
,
1142 unsigned long start_time
)
1144 /* Have we waited longer that a second? */
1145 if (time_after(jiffies
, start_time
+ HZ
)) {
1146 /* Lengthen the minimum hold time. */
1147 gl
->gl_hold_time
= min(gl
->gl_hold_time
+ GL_GLOCK_HOLD_INCR
,
1153 * gfs2_glock_wait - wait on a glock acquisition
1154 * @gh: the glock holder
1156 * Returns: 0 on success
1159 int gfs2_glock_wait(struct gfs2_holder
*gh
)
1161 unsigned long start_time
= jiffies
;
1164 wait_on_bit(&gh
->gh_iflags
, HIF_WAIT
, TASK_UNINTERRUPTIBLE
);
1165 gfs2_glock_update_hold_time(gh
->gh_gl
, start_time
);
1166 return gh
->gh_error
;
1169 static int glocks_pending(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1173 for (i
= 0; i
< num_gh
; i
++)
1174 if (test_bit(HIF_WAIT
, &ghs
[i
].gh_iflags
))
1180 * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1181 * @num_gh: the number of holders in the array
1182 * @ghs: the glock holder array
1184 * Returns: 0 on success, meaning all glocks have been granted and are held.
1185 * -ESTALE if the request timed out, meaning all glocks were released,
1186 * and the caller should retry the operation.
1189 int gfs2_glock_async_wait(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1191 struct gfs2_sbd
*sdp
= ghs
[0].gh_gl
->gl_name
.ln_sbd
;
1192 int i
, ret
= 0, timeout
= 0;
1193 unsigned long start_time
= jiffies
;
1198 * Total up the (minimum hold time * 2) of all glocks and use that to
1199 * determine the max amount of time we should wait.
1201 for (i
= 0; i
< num_gh
; i
++)
1202 timeout
+= ghs
[i
].gh_gl
->gl_hold_time
<< 1;
1205 if (!wait_event_timeout(sdp
->sd_async_glock_wait
,
1206 !glocks_pending(num_gh
, ghs
), timeout
))
1207 ret
= -ESTALE
; /* request timed out. */
1210 * If dlm granted all our requests, we need to adjust the glock
1211 * minimum hold time values according to how long we waited.
1213 * If our request timed out, we need to repeatedly release any held
1214 * glocks we acquired thus far to allow dlm to acquire the remaining
1215 * glocks without deadlocking. We cannot currently cancel outstanding
1216 * glock acquisitions.
1218 * The HIF_WAIT bit tells us which requests still need a response from
1221 * If dlm sent us any errors, we return the first error we find.
1223 keep_waiting
= false;
1224 for (i
= 0; i
< num_gh
; i
++) {
1225 /* Skip holders we have already dequeued below. */
1226 if (!gfs2_holder_queued(&ghs
[i
]))
1228 /* Skip holders with a pending DLM response. */
1229 if (test_bit(HIF_WAIT
, &ghs
[i
].gh_iflags
)) {
1230 keep_waiting
= true;
1234 if (test_bit(HIF_HOLDER
, &ghs
[i
].gh_iflags
)) {
1236 gfs2_glock_dq(&ghs
[i
]);
1238 gfs2_glock_update_hold_time(ghs
[i
].gh_gl
,
1242 ret
= ghs
[i
].gh_error
;
1249 * At this point, we've either acquired all locks or released them all.
1255 * handle_callback - process a demote request
1257 * @state: the state the caller wants us to change to
1259 * There are only two requests that we are going to see in actual
1260 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1263 static void handle_callback(struct gfs2_glock
*gl
, unsigned int state
,
1264 unsigned long delay
, bool remote
)
1267 set_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
);
1269 gfs2_set_demote(gl
);
1270 if (gl
->gl_demote_state
== LM_ST_EXCLUSIVE
) {
1271 gl
->gl_demote_state
= state
;
1272 gl
->gl_demote_time
= jiffies
;
1273 } else if (gl
->gl_demote_state
!= LM_ST_UNLOCKED
&&
1274 gl
->gl_demote_state
!= state
) {
1275 gl
->gl_demote_state
= LM_ST_UNLOCKED
;
1277 if (gl
->gl_ops
->go_callback
)
1278 gl
->gl_ops
->go_callback(gl
, remote
);
1279 trace_gfs2_demote_rq(gl
, remote
);
1282 void gfs2_print_dbg(struct seq_file
*seq
, const char *fmt
, ...)
1284 struct va_format vaf
;
1287 va_start(args
, fmt
);
1290 seq_vprintf(seq
, fmt
, args
);
1295 pr_err("%pV", &vaf
);
1302 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1303 * @gh: the holder structure to add
1305 * Eventually we should move the recursive locking trap to a
1306 * debugging option or something like that. This is the fast
1307 * path and needs to have the minimum number of distractions.
1311 static inline void add_to_queue(struct gfs2_holder
*gh
)
1312 __releases(&gl
->gl_lockref
.lock
)
1313 __acquires(&gl
->gl_lockref
.lock
)
1315 struct gfs2_glock
*gl
= gh
->gh_gl
;
1316 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
1317 struct list_head
*insert_pt
= NULL
;
1318 struct gfs2_holder
*gh2
;
1321 GLOCK_BUG_ON(gl
, gh
->gh_owner_pid
== NULL
);
1322 if (test_and_set_bit(HIF_WAIT
, &gh
->gh_iflags
))
1323 GLOCK_BUG_ON(gl
, true);
1325 if (gh
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
)) {
1326 if (test_bit(GLF_LOCK
, &gl
->gl_flags
))
1327 try_futile
= !may_grant(gl
, gh
);
1328 if (test_bit(GLF_INVALIDATE_IN_PROGRESS
, &gl
->gl_flags
))
1332 list_for_each_entry(gh2
, &gl
->gl_holders
, gh_list
) {
1333 if (unlikely(gh2
->gh_owner_pid
== gh
->gh_owner_pid
&&
1334 (gh
->gh_gl
->gl_ops
->go_type
!= LM_TYPE_FLOCK
)))
1335 goto trap_recursive
;
1337 !(gh2
->gh_flags
& (LM_FLAG_TRY
| LM_FLAG_TRY_1CB
))) {
1339 gh
->gh_error
= GLR_TRYFAILED
;
1340 gfs2_holder_wake(gh
);
1343 if (test_bit(HIF_HOLDER
, &gh2
->gh_iflags
))
1345 if (unlikely((gh
->gh_flags
& LM_FLAG_PRIORITY
) && !insert_pt
))
1346 insert_pt
= &gh2
->gh_list
;
1348 set_bit(GLF_QUEUED
, &gl
->gl_flags
);
1349 trace_gfs2_glock_queue(gh
, 1);
1350 gfs2_glstats_inc(gl
, GFS2_LKS_QCOUNT
);
1351 gfs2_sbstats_inc(gl
, GFS2_LKS_QCOUNT
);
1352 if (likely(insert_pt
== NULL
)) {
1353 list_add_tail(&gh
->gh_list
, &gl
->gl_holders
);
1354 if (unlikely(gh
->gh_flags
& LM_FLAG_PRIORITY
))
1358 list_add_tail(&gh
->gh_list
, insert_pt
);
1360 gh
= list_first_entry(&gl
->gl_holders
, struct gfs2_holder
, gh_list
);
1361 if (!(gh
->gh_flags
& LM_FLAG_PRIORITY
)) {
1362 spin_unlock(&gl
->gl_lockref
.lock
);
1363 if (sdp
->sd_lockstruct
.ls_ops
->lm_cancel
)
1364 sdp
->sd_lockstruct
.ls_ops
->lm_cancel(gl
);
1365 spin_lock(&gl
->gl_lockref
.lock
);
1370 fs_err(sdp
, "original: %pSR\n", (void *)gh2
->gh_ip
);
1371 fs_err(sdp
, "pid: %d\n", pid_nr(gh2
->gh_owner_pid
));
1372 fs_err(sdp
, "lock type: %d req lock state : %d\n",
1373 gh2
->gh_gl
->gl_name
.ln_type
, gh2
->gh_state
);
1374 fs_err(sdp
, "new: %pSR\n", (void *)gh
->gh_ip
);
1375 fs_err(sdp
, "pid: %d\n", pid_nr(gh
->gh_owner_pid
));
1376 fs_err(sdp
, "lock type: %d req lock state : %d\n",
1377 gh
->gh_gl
->gl_name
.ln_type
, gh
->gh_state
);
1378 gfs2_dump_glock(NULL
, gl
, true);
1383 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1384 * @gh: the holder structure
1386 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1388 * Returns: 0, GLR_TRYFAILED, or errno on failure
1391 int gfs2_glock_nq(struct gfs2_holder
*gh
)
1393 struct gfs2_glock
*gl
= gh
->gh_gl
;
1396 if (glock_blocked_by_withdraw(gl
) && !(gh
->gh_flags
& LM_FLAG_NOEXP
))
1399 if (test_bit(GLF_LRU
, &gl
->gl_flags
))
1400 gfs2_glock_remove_from_lru(gl
);
1402 spin_lock(&gl
->gl_lockref
.lock
);
1404 if (unlikely((LM_FLAG_NOEXP
& gh
->gh_flags
) &&
1405 test_and_clear_bit(GLF_FROZEN
, &gl
->gl_flags
))) {
1406 set_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
);
1407 gl
->gl_lockref
.count
++;
1408 __gfs2_glock_queue_work(gl
, 0);
1411 spin_unlock(&gl
->gl_lockref
.lock
);
1413 if (!(gh
->gh_flags
& GL_ASYNC
))
1414 error
= gfs2_glock_wait(gh
);
1420 * gfs2_glock_poll - poll to see if an async request has been completed
1423 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1426 int gfs2_glock_poll(struct gfs2_holder
*gh
)
1428 return test_bit(HIF_WAIT
, &gh
->gh_iflags
) ? 0 : 1;
1432 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1433 * @gh: the glock holder
1437 void gfs2_glock_dq(struct gfs2_holder
*gh
)
1439 struct gfs2_glock
*gl
= gh
->gh_gl
;
1440 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
1444 spin_lock(&gl
->gl_lockref
.lock
);
1446 * If we're in the process of file system withdraw, we cannot just
1447 * dequeue any glocks until our journal is recovered, lest we
1448 * introduce file system corruption. We need two exceptions to this
1449 * rule: We need to allow unlocking of nondisk glocks and the glock
1450 * for our own journal that needs recovery.
1452 if (test_bit(SDF_WITHDRAW_RECOVERY
, &sdp
->sd_flags
) &&
1453 glock_blocked_by_withdraw(gl
) &&
1454 gh
->gh_gl
!= sdp
->sd_jinode_gl
) {
1455 sdp
->sd_glock_dqs_held
++;
1457 wait_on_bit(&sdp
->sd_flags
, SDF_WITHDRAW_RECOVERY
,
1458 TASK_UNINTERRUPTIBLE
);
1460 if (gh
->gh_flags
& GL_NOCACHE
)
1461 handle_callback(gl
, LM_ST_UNLOCKED
, 0, false);
1463 list_del_init(&gh
->gh_list
);
1464 clear_bit(HIF_HOLDER
, &gh
->gh_iflags
);
1465 if (find_first_holder(gl
) == NULL
) {
1466 if (list_empty(&gl
->gl_holders
) &&
1467 !test_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
) &&
1468 !test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
1471 if (!test_bit(GLF_LFLUSH
, &gl
->gl_flags
) && demote_ok(gl
))
1472 gfs2_glock_add_to_lru(gl
);
1474 trace_gfs2_glock_queue(gh
, 0);
1475 if (unlikely(!fast_path
)) {
1476 gl
->gl_lockref
.count
++;
1477 if (test_bit(GLF_PENDING_DEMOTE
, &gl
->gl_flags
) &&
1478 !test_bit(GLF_DEMOTE
, &gl
->gl_flags
) &&
1479 gl
->gl_name
.ln_type
== LM_TYPE_INODE
)
1480 delay
= gl
->gl_hold_time
;
1481 __gfs2_glock_queue_work(gl
, delay
);
1483 spin_unlock(&gl
->gl_lockref
.lock
);
1486 void gfs2_glock_dq_wait(struct gfs2_holder
*gh
)
1488 struct gfs2_glock
*gl
= gh
->gh_gl
;
1491 wait_on_bit(&gl
->gl_flags
, GLF_DEMOTE
, TASK_UNINTERRUPTIBLE
);
1495 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1496 * @gh: the holder structure
1500 void gfs2_glock_dq_uninit(struct gfs2_holder
*gh
)
1503 gfs2_holder_uninit(gh
);
1507 * gfs2_glock_nq_num - acquire a glock based on lock number
1508 * @sdp: the filesystem
1509 * @number: the lock number
1510 * @glops: the glock operations for the type of glock
1511 * @state: the state to acquire the glock in
1512 * @flags: modifier flags for the acquisition
1513 * @gh: the struct gfs2_holder
1518 int gfs2_glock_nq_num(struct gfs2_sbd
*sdp
, u64 number
,
1519 const struct gfs2_glock_operations
*glops
,
1520 unsigned int state
, u16 flags
, struct gfs2_holder
*gh
)
1522 struct gfs2_glock
*gl
;
1525 error
= gfs2_glock_get(sdp
, number
, glops
, CREATE
, &gl
);
1527 error
= gfs2_glock_nq_init(gl
, state
, flags
, gh
);
1535 * glock_compare - Compare two struct gfs2_glock structures for sorting
1536 * @arg_a: the first structure
1537 * @arg_b: the second structure
1541 static int glock_compare(const void *arg_a
, const void *arg_b
)
1543 const struct gfs2_holder
*gh_a
= *(const struct gfs2_holder
**)arg_a
;
1544 const struct gfs2_holder
*gh_b
= *(const struct gfs2_holder
**)arg_b
;
1545 const struct lm_lockname
*a
= &gh_a
->gh_gl
->gl_name
;
1546 const struct lm_lockname
*b
= &gh_b
->gh_gl
->gl_name
;
1548 if (a
->ln_number
> b
->ln_number
)
1550 if (a
->ln_number
< b
->ln_number
)
1552 BUG_ON(gh_a
->gh_gl
->gl_ops
->go_type
== gh_b
->gh_gl
->gl_ops
->go_type
);
1557 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1558 * @num_gh: the number of structures
1559 * @ghs: an array of struct gfs2_holder structures
1561 * Returns: 0 on success (all glocks acquired),
1562 * errno on failure (no glocks acquired)
1565 static int nq_m_sync(unsigned int num_gh
, struct gfs2_holder
*ghs
,
1566 struct gfs2_holder
**p
)
1571 for (x
= 0; x
< num_gh
; x
++)
1574 sort(p
, num_gh
, sizeof(struct gfs2_holder
*), glock_compare
, NULL
);
1576 for (x
= 0; x
< num_gh
; x
++) {
1577 p
[x
]->gh_flags
&= ~(LM_FLAG_TRY
| GL_ASYNC
);
1579 error
= gfs2_glock_nq(p
[x
]);
1582 gfs2_glock_dq(p
[x
]);
1591 * gfs2_glock_nq_m - acquire multiple glocks
1592 * @num_gh: the number of structures
1593 * @ghs: an array of struct gfs2_holder structures
1596 * Returns: 0 on success (all glocks acquired),
1597 * errno on failure (no glocks acquired)
1600 int gfs2_glock_nq_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1602 struct gfs2_holder
*tmp
[4];
1603 struct gfs2_holder
**pph
= tmp
;
1610 ghs
->gh_flags
&= ~(LM_FLAG_TRY
| GL_ASYNC
);
1611 return gfs2_glock_nq(ghs
);
1615 pph
= kmalloc_array(num_gh
, sizeof(struct gfs2_holder
*),
1621 error
= nq_m_sync(num_gh
, ghs
, pph
);
1630 * gfs2_glock_dq_m - release multiple glocks
1631 * @num_gh: the number of structures
1632 * @ghs: an array of struct gfs2_holder structures
1636 void gfs2_glock_dq_m(unsigned int num_gh
, struct gfs2_holder
*ghs
)
1639 gfs2_glock_dq(&ghs
[num_gh
]);
1642 void gfs2_glock_cb(struct gfs2_glock
*gl
, unsigned int state
)
1644 unsigned long delay
= 0;
1645 unsigned long holdtime
;
1646 unsigned long now
= jiffies
;
1648 gfs2_glock_hold(gl
);
1649 holdtime
= gl
->gl_tchange
+ gl
->gl_hold_time
;
1650 if (test_bit(GLF_QUEUED
, &gl
->gl_flags
) &&
1651 gl
->gl_name
.ln_type
== LM_TYPE_INODE
) {
1652 if (time_before(now
, holdtime
))
1653 delay
= holdtime
- now
;
1654 if (test_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
))
1655 delay
= gl
->gl_hold_time
;
1658 spin_lock(&gl
->gl_lockref
.lock
);
1659 handle_callback(gl
, state
, delay
, true);
1660 __gfs2_glock_queue_work(gl
, delay
);
1661 spin_unlock(&gl
->gl_lockref
.lock
);
1665 * gfs2_should_freeze - Figure out if glock should be frozen
1666 * @gl: The glock in question
1668 * Glocks are not frozen if (a) the result of the dlm operation is
1669 * an error, (b) the locking operation was an unlock operation or
1670 * (c) if there is a "noexp" flagged request anywhere in the queue
1672 * Returns: 1 if freezing should occur, 0 otherwise
1675 static int gfs2_should_freeze(const struct gfs2_glock
*gl
)
1677 const struct gfs2_holder
*gh
;
1679 if (gl
->gl_reply
& ~LM_OUT_ST_MASK
)
1681 if (gl
->gl_target
== LM_ST_UNLOCKED
)
1684 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
) {
1685 if (test_bit(HIF_HOLDER
, &gh
->gh_iflags
))
1687 if (LM_FLAG_NOEXP
& gh
->gh_flags
)
1695 * gfs2_glock_complete - Callback used by locking
1696 * @gl: Pointer to the glock
1697 * @ret: The return value from the dlm
1699 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1700 * to use a bitfield shared with other glock state fields.
1703 void gfs2_glock_complete(struct gfs2_glock
*gl
, int ret
)
1705 struct lm_lockstruct
*ls
= &gl
->gl_name
.ln_sbd
->sd_lockstruct
;
1707 spin_lock(&gl
->gl_lockref
.lock
);
1710 if (unlikely(test_bit(DFL_BLOCK_LOCKS
, &ls
->ls_recover_flags
))) {
1711 if (gfs2_should_freeze(gl
)) {
1712 set_bit(GLF_FROZEN
, &gl
->gl_flags
);
1713 spin_unlock(&gl
->gl_lockref
.lock
);
1718 gl
->gl_lockref
.count
++;
1719 set_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
);
1720 __gfs2_glock_queue_work(gl
, 0);
1721 spin_unlock(&gl
->gl_lockref
.lock
);
1724 static int glock_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
1726 struct gfs2_glock
*gla
, *glb
;
1728 gla
= list_entry(a
, struct gfs2_glock
, gl_lru
);
1729 glb
= list_entry(b
, struct gfs2_glock
, gl_lru
);
1731 if (gla
->gl_name
.ln_number
> glb
->gl_name
.ln_number
)
1733 if (gla
->gl_name
.ln_number
< glb
->gl_name
.ln_number
)
1740 * gfs2_dispose_glock_lru - Demote a list of glocks
1741 * @list: The list to dispose of
1743 * Disposing of glocks may involve disk accesses, so that here we sort
1744 * the glocks by number (i.e. disk location of the inodes) so that if
1745 * there are any such accesses, they'll be sent in order (mostly).
1747 * Must be called under the lru_lock, but may drop and retake this
1748 * lock. While the lru_lock is dropped, entries may vanish from the
1749 * list, but no new entries will appear on the list (since it is
1753 static void gfs2_dispose_glock_lru(struct list_head
*list
)
1754 __releases(&lru_lock
)
1755 __acquires(&lru_lock
)
1757 struct gfs2_glock
*gl
;
1759 list_sort(NULL
, list
, glock_cmp
);
1761 while(!list_empty(list
)) {
1762 gl
= list_first_entry(list
, struct gfs2_glock
, gl_lru
);
1763 list_del_init(&gl
->gl_lru
);
1764 if (!spin_trylock(&gl
->gl_lockref
.lock
)) {
1766 list_add(&gl
->gl_lru
, &lru_list
);
1767 set_bit(GLF_LRU
, &gl
->gl_flags
);
1768 atomic_inc(&lru_count
);
1771 if (test_and_set_bit(GLF_LOCK
, &gl
->gl_flags
)) {
1772 spin_unlock(&gl
->gl_lockref
.lock
);
1773 goto add_back_to_lru
;
1775 gl
->gl_lockref
.count
++;
1777 handle_callback(gl
, LM_ST_UNLOCKED
, 0, false);
1778 WARN_ON(!test_and_clear_bit(GLF_LOCK
, &gl
->gl_flags
));
1779 __gfs2_glock_queue_work(gl
, 0);
1780 spin_unlock(&gl
->gl_lockref
.lock
);
1781 cond_resched_lock(&lru_lock
);
1786 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1787 * @nr: The number of entries to scan
1789 * This function selects the entries on the LRU which are able to
1790 * be demoted, and then kicks off the process by calling
1791 * gfs2_dispose_glock_lru() above.
1794 static long gfs2_scan_glock_lru(int nr
)
1796 struct gfs2_glock
*gl
;
1801 spin_lock(&lru_lock
);
1802 while ((nr
-- >= 0) && !list_empty(&lru_list
)) {
1803 gl
= list_first_entry(&lru_list
, struct gfs2_glock
, gl_lru
);
1805 /* Test for being demotable */
1806 if (!test_bit(GLF_LOCK
, &gl
->gl_flags
)) {
1807 list_move(&gl
->gl_lru
, &dispose
);
1808 atomic_dec(&lru_count
);
1809 clear_bit(GLF_LRU
, &gl
->gl_flags
);
1814 list_move(&gl
->gl_lru
, &skipped
);
1816 list_splice(&skipped
, &lru_list
);
1817 if (!list_empty(&dispose
))
1818 gfs2_dispose_glock_lru(&dispose
);
1819 spin_unlock(&lru_lock
);
1824 static unsigned long gfs2_glock_shrink_scan(struct shrinker
*shrink
,
1825 struct shrink_control
*sc
)
1827 if (!(sc
->gfp_mask
& __GFP_FS
))
1829 return gfs2_scan_glock_lru(sc
->nr_to_scan
);
1832 static unsigned long gfs2_glock_shrink_count(struct shrinker
*shrink
,
1833 struct shrink_control
*sc
)
1835 return vfs_pressure_ratio(atomic_read(&lru_count
));
1838 static struct shrinker glock_shrinker
= {
1839 .seeks
= DEFAULT_SEEKS
,
1840 .count_objects
= gfs2_glock_shrink_count
,
1841 .scan_objects
= gfs2_glock_shrink_scan
,
1845 * examine_bucket - Call a function for glock in a hash bucket
1846 * @examiner: the function
1847 * @sdp: the filesystem
1848 * @bucket: the bucket
1850 * Note that the function can be called multiple times on the same
1851 * object. So the user must ensure that the function can cope with
1855 static void glock_hash_walk(glock_examiner examiner
, const struct gfs2_sbd
*sdp
)
1857 struct gfs2_glock
*gl
;
1858 struct rhashtable_iter iter
;
1860 rhashtable_walk_enter(&gl_hash_table
, &iter
);
1863 rhashtable_walk_start(&iter
);
1865 while ((gl
= rhashtable_walk_next(&iter
)) && !IS_ERR(gl
))
1866 if (gl
->gl_name
.ln_sbd
== sdp
&&
1867 lockref_get_not_dead(&gl
->gl_lockref
))
1870 rhashtable_walk_stop(&iter
);
1871 } while (cond_resched(), gl
== ERR_PTR(-EAGAIN
));
1873 rhashtable_walk_exit(&iter
);
1876 bool gfs2_queue_delete_work(struct gfs2_glock
*gl
, unsigned long delay
)
1880 spin_lock(&gl
->gl_lockref
.lock
);
1881 queued
= queue_delayed_work(gfs2_delete_workqueue
,
1882 &gl
->gl_delete
, delay
);
1884 set_bit(GLF_PENDING_DELETE
, &gl
->gl_flags
);
1885 spin_unlock(&gl
->gl_lockref
.lock
);
1889 void gfs2_cancel_delete_work(struct gfs2_glock
*gl
)
1891 if (cancel_delayed_work_sync(&gl
->gl_delete
)) {
1892 clear_bit(GLF_PENDING_DELETE
, &gl
->gl_flags
);
1897 bool gfs2_delete_work_queued(const struct gfs2_glock
*gl
)
1899 return test_bit(GLF_PENDING_DELETE
, &gl
->gl_flags
);
1902 static void flush_delete_work(struct gfs2_glock
*gl
)
1904 if (cancel_delayed_work(&gl
->gl_delete
)) {
1905 queue_delayed_work(gfs2_delete_workqueue
,
1908 gfs2_glock_queue_work(gl
, 0);
1911 void gfs2_flush_delete_work(struct gfs2_sbd
*sdp
)
1913 glock_hash_walk(flush_delete_work
, sdp
);
1914 flush_workqueue(gfs2_delete_workqueue
);
1918 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1919 * @gl: The glock to thaw
1923 static void thaw_glock(struct gfs2_glock
*gl
)
1925 if (!test_and_clear_bit(GLF_FROZEN
, &gl
->gl_flags
)) {
1929 set_bit(GLF_REPLY_PENDING
, &gl
->gl_flags
);
1930 gfs2_glock_queue_work(gl
, 0);
1934 * clear_glock - look at a glock and see if we can free it from glock cache
1935 * @gl: the glock to look at
1939 static void clear_glock(struct gfs2_glock
*gl
)
1941 gfs2_glock_remove_from_lru(gl
);
1943 spin_lock(&gl
->gl_lockref
.lock
);
1944 if (gl
->gl_state
!= LM_ST_UNLOCKED
)
1945 handle_callback(gl
, LM_ST_UNLOCKED
, 0, false);
1946 __gfs2_glock_queue_work(gl
, 0);
1947 spin_unlock(&gl
->gl_lockref
.lock
);
1951 * gfs2_glock_thaw - Thaw any frozen glocks
1952 * @sdp: The super block
1956 void gfs2_glock_thaw(struct gfs2_sbd
*sdp
)
1958 glock_hash_walk(thaw_glock
, sdp
);
1961 static void dump_glock(struct seq_file
*seq
, struct gfs2_glock
*gl
, bool fsid
)
1963 spin_lock(&gl
->gl_lockref
.lock
);
1964 gfs2_dump_glock(seq
, gl
, fsid
);
1965 spin_unlock(&gl
->gl_lockref
.lock
);
1968 static void dump_glock_func(struct gfs2_glock
*gl
)
1970 dump_glock(NULL
, gl
, true);
1974 * gfs2_gl_hash_clear - Empty out the glock hash table
1975 * @sdp: the filesystem
1976 * @wait: wait until it's all gone
1978 * Called when unmounting the filesystem.
1981 void gfs2_gl_hash_clear(struct gfs2_sbd
*sdp
)
1983 set_bit(SDF_SKIP_DLM_UNLOCK
, &sdp
->sd_flags
);
1984 flush_workqueue(glock_workqueue
);
1985 glock_hash_walk(clear_glock
, sdp
);
1986 flush_workqueue(glock_workqueue
);
1987 wait_event_timeout(sdp
->sd_glock_wait
,
1988 atomic_read(&sdp
->sd_glock_disposal
) == 0,
1990 glock_hash_walk(dump_glock_func
, sdp
);
1993 void gfs2_glock_finish_truncate(struct gfs2_inode
*ip
)
1995 struct gfs2_glock
*gl
= ip
->i_gl
;
1998 ret
= gfs2_truncatei_resume(ip
);
1999 gfs2_glock_assert_withdraw(gl
, ret
== 0);
2001 spin_lock(&gl
->gl_lockref
.lock
);
2002 clear_bit(GLF_LOCK
, &gl
->gl_flags
);
2004 spin_unlock(&gl
->gl_lockref
.lock
);
2007 static const char *state2str(unsigned state
)
2010 case LM_ST_UNLOCKED
:
2014 case LM_ST_DEFERRED
:
2016 case LM_ST_EXCLUSIVE
:
2022 static const char *hflags2str(char *buf
, u16 flags
, unsigned long iflags
)
2025 if (flags
& LM_FLAG_TRY
)
2027 if (flags
& LM_FLAG_TRY_1CB
)
2029 if (flags
& LM_FLAG_NOEXP
)
2031 if (flags
& LM_FLAG_ANY
)
2033 if (flags
& LM_FLAG_PRIORITY
)
2035 if (flags
& GL_ASYNC
)
2037 if (flags
& GL_EXACT
)
2039 if (flags
& GL_NOCACHE
)
2041 if (test_bit(HIF_HOLDER
, &iflags
))
2043 if (test_bit(HIF_WAIT
, &iflags
))
2045 if (test_bit(HIF_FIRST
, &iflags
))
2052 * dump_holder - print information about a glock holder
2053 * @seq: the seq_file struct
2054 * @gh: the glock holder
2055 * @fs_id_buf: pointer to file system id (if requested)
2059 static void dump_holder(struct seq_file
*seq
, const struct gfs2_holder
*gh
,
2060 const char *fs_id_buf
)
2062 struct task_struct
*gh_owner
= NULL
;
2066 if (gh
->gh_owner_pid
)
2067 gh_owner
= pid_task(gh
->gh_owner_pid
, PIDTYPE_PID
);
2068 gfs2_print_dbg(seq
, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2069 fs_id_buf
, state2str(gh
->gh_state
),
2070 hflags2str(flags_buf
, gh
->gh_flags
, gh
->gh_iflags
),
2072 gh
->gh_owner_pid
? (long)pid_nr(gh
->gh_owner_pid
) : -1,
2073 gh_owner
? gh_owner
->comm
: "(ended)",
2078 static const char *gflags2str(char *buf
, const struct gfs2_glock
*gl
)
2080 const unsigned long *gflags
= &gl
->gl_flags
;
2083 if (test_bit(GLF_LOCK
, gflags
))
2085 if (test_bit(GLF_DEMOTE
, gflags
))
2087 if (test_bit(GLF_PENDING_DEMOTE
, gflags
))
2089 if (test_bit(GLF_DEMOTE_IN_PROGRESS
, gflags
))
2091 if (test_bit(GLF_DIRTY
, gflags
))
2093 if (test_bit(GLF_LFLUSH
, gflags
))
2095 if (test_bit(GLF_INVALIDATE_IN_PROGRESS
, gflags
))
2097 if (test_bit(GLF_REPLY_PENDING
, gflags
))
2099 if (test_bit(GLF_INITIAL
, gflags
))
2101 if (test_bit(GLF_FROZEN
, gflags
))
2103 if (test_bit(GLF_QUEUED
, gflags
))
2105 if (test_bit(GLF_LRU
, gflags
))
2109 if (test_bit(GLF_BLOCKING
, gflags
))
2111 if (test_bit(GLF_INODE_CREATING
, gflags
))
2113 if (test_bit(GLF_PENDING_DELETE
, gflags
))
2115 if (test_bit(GLF_FREEING
, gflags
))
2122 * gfs2_dump_glock - print information about a glock
2123 * @seq: The seq_file struct
2125 * @fsid: If true, also dump the file system id
2127 * The file format is as follows:
2128 * One line per object, capital letters are used to indicate objects
2129 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2130 * other objects are indented by a single space and follow the glock to
2131 * which they are related. Fields are indicated by lower case letters
2132 * followed by a colon and the field value, except for strings which are in
2133 * [] so that its possible to see if they are composed of spaces for
2134 * example. The field's are n = number (id of the object), f = flags,
2135 * t = type, s = state, r = refcount, e = error, p = pid.
2139 void gfs2_dump_glock(struct seq_file
*seq
, struct gfs2_glock
*gl
, bool fsid
)
2141 const struct gfs2_glock_operations
*glops
= gl
->gl_ops
;
2142 unsigned long long dtime
;
2143 const struct gfs2_holder
*gh
;
2144 char gflags_buf
[32];
2145 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
2146 char fs_id_buf
[sizeof(sdp
->sd_fsname
) + 7];
2147 unsigned long nrpages
= 0;
2149 if (gl
->gl_ops
->go_flags
& GLOF_ASPACE
) {
2150 struct address_space
*mapping
= gfs2_glock2aspace(gl
);
2152 nrpages
= mapping
->nrpages
;
2154 memset(fs_id_buf
, 0, sizeof(fs_id_buf
));
2155 if (fsid
&& sdp
) /* safety precaution */
2156 sprintf(fs_id_buf
, "fsid=%s: ", sdp
->sd_fsname
);
2157 dtime
= jiffies
- gl
->gl_demote_time
;
2158 dtime
*= 1000000/HZ
; /* demote time in uSec */
2159 if (!test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
2161 gfs2_print_dbg(seq
, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2162 "v:%d r:%d m:%ld p:%lu\n",
2163 fs_id_buf
, state2str(gl
->gl_state
),
2164 gl
->gl_name
.ln_type
,
2165 (unsigned long long)gl
->gl_name
.ln_number
,
2166 gflags2str(gflags_buf
, gl
),
2167 state2str(gl
->gl_target
),
2168 state2str(gl
->gl_demote_state
), dtime
,
2169 atomic_read(&gl
->gl_ail_count
),
2170 atomic_read(&gl
->gl_revokes
),
2171 (int)gl
->gl_lockref
.count
, gl
->gl_hold_time
, nrpages
);
2173 list_for_each_entry(gh
, &gl
->gl_holders
, gh_list
)
2174 dump_holder(seq
, gh
, fs_id_buf
);
2176 if (gl
->gl_state
!= LM_ST_UNLOCKED
&& glops
->go_dump
)
2177 glops
->go_dump(seq
, gl
, fs_id_buf
);
2180 static int gfs2_glstats_seq_show(struct seq_file
*seq
, void *iter_ptr
)
2182 struct gfs2_glock
*gl
= iter_ptr
;
2184 seq_printf(seq
, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2185 gl
->gl_name
.ln_type
,
2186 (unsigned long long)gl
->gl_name
.ln_number
,
2187 (unsigned long long)gl
->gl_stats
.stats
[GFS2_LKS_SRTT
],
2188 (unsigned long long)gl
->gl_stats
.stats
[GFS2_LKS_SRTTVAR
],
2189 (unsigned long long)gl
->gl_stats
.stats
[GFS2_LKS_SRTTB
],
2190 (unsigned long long)gl
->gl_stats
.stats
[GFS2_LKS_SRTTVARB
],
2191 (unsigned long long)gl
->gl_stats
.stats
[GFS2_LKS_SIRT
],
2192 (unsigned long long)gl
->gl_stats
.stats
[GFS2_LKS_SIRTVAR
],
2193 (unsigned long long)gl
->gl_stats
.stats
[GFS2_LKS_DCOUNT
],
2194 (unsigned long long)gl
->gl_stats
.stats
[GFS2_LKS_QCOUNT
]);
2198 static const char *gfs2_gltype
[] = {
2212 static const char *gfs2_stype
[] = {
2213 [GFS2_LKS_SRTT
] = "srtt",
2214 [GFS2_LKS_SRTTVAR
] = "srttvar",
2215 [GFS2_LKS_SRTTB
] = "srttb",
2216 [GFS2_LKS_SRTTVARB
] = "srttvarb",
2217 [GFS2_LKS_SIRT
] = "sirt",
2218 [GFS2_LKS_SIRTVAR
] = "sirtvar",
2219 [GFS2_LKS_DCOUNT
] = "dlm",
2220 [GFS2_LKS_QCOUNT
] = "queue",
2223 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2225 static int gfs2_sbstats_seq_show(struct seq_file
*seq
, void *iter_ptr
)
2227 struct gfs2_sbd
*sdp
= seq
->private;
2228 loff_t pos
= *(loff_t
*)iter_ptr
;
2229 unsigned index
= pos
>> 3;
2230 unsigned subindex
= pos
& 0x07;
2233 if (index
== 0 && subindex
!= 0)
2236 seq_printf(seq
, "%-10s %8s:", gfs2_gltype
[index
],
2237 (index
== 0) ? "cpu": gfs2_stype
[subindex
]);
2239 for_each_possible_cpu(i
) {
2240 const struct gfs2_pcpu_lkstats
*lkstats
= per_cpu_ptr(sdp
->sd_lkstats
, i
);
2243 seq_printf(seq
, " %15u", i
);
2245 seq_printf(seq
, " %15llu", (unsigned long long)lkstats
->
2246 lkstats
[index
- 1].stats
[subindex
]);
2248 seq_putc(seq
, '\n');
2252 int __init
gfs2_glock_init(void)
2256 ret
= rhashtable_init(&gl_hash_table
, &ht_parms
);
2260 glock_workqueue
= alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM
|
2261 WQ_HIGHPRI
| WQ_FREEZABLE
, 0);
2262 if (!glock_workqueue
) {
2263 rhashtable_destroy(&gl_hash_table
);
2266 gfs2_delete_workqueue
= alloc_workqueue("delete_workqueue",
2267 WQ_MEM_RECLAIM
| WQ_FREEZABLE
,
2269 if (!gfs2_delete_workqueue
) {
2270 destroy_workqueue(glock_workqueue
);
2271 rhashtable_destroy(&gl_hash_table
);
2275 ret
= register_shrinker(&glock_shrinker
);
2277 destroy_workqueue(gfs2_delete_workqueue
);
2278 destroy_workqueue(glock_workqueue
);
2279 rhashtable_destroy(&gl_hash_table
);
2283 for (i
= 0; i
< GLOCK_WAIT_TABLE_SIZE
; i
++)
2284 init_waitqueue_head(glock_wait_table
+ i
);
2289 void gfs2_glock_exit(void)
2291 unregister_shrinker(&glock_shrinker
);
2292 rhashtable_destroy(&gl_hash_table
);
2293 destroy_workqueue(glock_workqueue
);
2294 destroy_workqueue(gfs2_delete_workqueue
);
2297 static void gfs2_glock_iter_next(struct gfs2_glock_iter
*gi
, loff_t n
)
2299 struct gfs2_glock
*gl
= gi
->gl
;
2304 if (!lockref_put_not_zero(&gl
->gl_lockref
))
2305 gfs2_glock_queue_put(gl
);
2308 gl
= rhashtable_walk_next(&gi
->hti
);
2309 if (IS_ERR_OR_NULL(gl
)) {
2310 if (gl
== ERR_PTR(-EAGAIN
)) {
2317 if (gl
->gl_name
.ln_sbd
!= gi
->sdp
)
2320 if (!lockref_get_not_dead(&gl
->gl_lockref
))
2324 if (__lockref_is_dead(&gl
->gl_lockref
))
2332 static void *gfs2_glock_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2335 struct gfs2_glock_iter
*gi
= seq
->private;
2339 * We can either stay where we are, skip to the next hash table
2340 * entry, or start from the beginning.
2342 if (*pos
< gi
->last_pos
) {
2343 rhashtable_walk_exit(&gi
->hti
);
2344 rhashtable_walk_enter(&gl_hash_table
, &gi
->hti
);
2347 n
= *pos
- gi
->last_pos
;
2350 rhashtable_walk_start(&gi
->hti
);
2352 gfs2_glock_iter_next(gi
, n
);
2353 gi
->last_pos
= *pos
;
2357 static void *gfs2_glock_seq_next(struct seq_file
*seq
, void *iter_ptr
,
2360 struct gfs2_glock_iter
*gi
= seq
->private;
2363 gi
->last_pos
= *pos
;
2364 gfs2_glock_iter_next(gi
, 1);
2368 static void gfs2_glock_seq_stop(struct seq_file
*seq
, void *iter_ptr
)
2371 struct gfs2_glock_iter
*gi
= seq
->private;
2373 rhashtable_walk_stop(&gi
->hti
);
2376 static int gfs2_glock_seq_show(struct seq_file
*seq
, void *iter_ptr
)
2378 dump_glock(seq
, iter_ptr
, false);
2382 static void *gfs2_sbstats_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2385 if (*pos
>= GFS2_NR_SBSTATS
)
2390 static void *gfs2_sbstats_seq_next(struct seq_file
*seq
, void *iter_ptr
,
2394 if (*pos
>= GFS2_NR_SBSTATS
)
2399 static void gfs2_sbstats_seq_stop(struct seq_file
*seq
, void *iter_ptr
)
2404 static const struct seq_operations gfs2_glock_seq_ops
= {
2405 .start
= gfs2_glock_seq_start
,
2406 .next
= gfs2_glock_seq_next
,
2407 .stop
= gfs2_glock_seq_stop
,
2408 .show
= gfs2_glock_seq_show
,
2411 static const struct seq_operations gfs2_glstats_seq_ops
= {
2412 .start
= gfs2_glock_seq_start
,
2413 .next
= gfs2_glock_seq_next
,
2414 .stop
= gfs2_glock_seq_stop
,
2415 .show
= gfs2_glstats_seq_show
,
2418 static const struct seq_operations gfs2_sbstats_seq_ops
= {
2419 .start
= gfs2_sbstats_seq_start
,
2420 .next
= gfs2_sbstats_seq_next
,
2421 .stop
= gfs2_sbstats_seq_stop
,
2422 .show
= gfs2_sbstats_seq_show
,
2425 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2427 static int __gfs2_glocks_open(struct inode
*inode
, struct file
*file
,
2428 const struct seq_operations
*ops
)
2430 int ret
= seq_open_private(file
, ops
, sizeof(struct gfs2_glock_iter
));
2432 struct seq_file
*seq
= file
->private_data
;
2433 struct gfs2_glock_iter
*gi
= seq
->private;
2435 gi
->sdp
= inode
->i_private
;
2436 seq
->buf
= kmalloc(GFS2_SEQ_GOODSIZE
, GFP_KERNEL
| __GFP_NOWARN
);
2438 seq
->size
= GFS2_SEQ_GOODSIZE
;
2440 * Initially, we are "before" the first hash table entry; the
2441 * first call to rhashtable_walk_next gets us the first entry.
2445 rhashtable_walk_enter(&gl_hash_table
, &gi
->hti
);
2450 static int gfs2_glocks_open(struct inode
*inode
, struct file
*file
)
2452 return __gfs2_glocks_open(inode
, file
, &gfs2_glock_seq_ops
);
2455 static int gfs2_glocks_release(struct inode
*inode
, struct file
*file
)
2457 struct seq_file
*seq
= file
->private_data
;
2458 struct gfs2_glock_iter
*gi
= seq
->private;
2461 gfs2_glock_put(gi
->gl
);
2462 rhashtable_walk_exit(&gi
->hti
);
2463 return seq_release_private(inode
, file
);
2466 static int gfs2_glstats_open(struct inode
*inode
, struct file
*file
)
2468 return __gfs2_glocks_open(inode
, file
, &gfs2_glstats_seq_ops
);
2471 static int gfs2_sbstats_open(struct inode
*inode
, struct file
*file
)
2473 int ret
= seq_open(file
, &gfs2_sbstats_seq_ops
);
2475 struct seq_file
*seq
= file
->private_data
;
2476 seq
->private = inode
->i_private
; /* sdp */
2481 static const struct file_operations gfs2_glocks_fops
= {
2482 .owner
= THIS_MODULE
,
2483 .open
= gfs2_glocks_open
,
2485 .llseek
= seq_lseek
,
2486 .release
= gfs2_glocks_release
,
2489 static const struct file_operations gfs2_glstats_fops
= {
2490 .owner
= THIS_MODULE
,
2491 .open
= gfs2_glstats_open
,
2493 .llseek
= seq_lseek
,
2494 .release
= gfs2_glocks_release
,
2497 static const struct file_operations gfs2_sbstats_fops
= {
2498 .owner
= THIS_MODULE
,
2499 .open
= gfs2_sbstats_open
,
2501 .llseek
= seq_lseek
,
2502 .release
= seq_release
,
2505 void gfs2_create_debugfs_file(struct gfs2_sbd
*sdp
)
2507 sdp
->debugfs_dir
= debugfs_create_dir(sdp
->sd_table_name
, gfs2_root
);
2509 debugfs_create_file("glocks", S_IFREG
| S_IRUGO
, sdp
->debugfs_dir
, sdp
,
2512 debugfs_create_file("glstats", S_IFREG
| S_IRUGO
, sdp
->debugfs_dir
, sdp
,
2513 &gfs2_glstats_fops
);
2515 debugfs_create_file("sbstats", S_IFREG
| S_IRUGO
, sdp
->debugfs_dir
, sdp
,
2516 &gfs2_sbstats_fops
);
2519 void gfs2_delete_debugfs_file(struct gfs2_sbd
*sdp
)
2521 debugfs_remove_recursive(sdp
->debugfs_dir
);
2522 sdp
->debugfs_dir
= NULL
;
2525 void gfs2_register_debugfs(void)
2527 gfs2_root
= debugfs_create_dir("gfs2", NULL
);
2530 void gfs2_unregister_debugfs(void)
2532 debugfs_remove(gfs2_root
);