bug: introduce BUILD_BUG_ON_INVALID() macro
[linux/fpc-iii.git] / fs / gfs2 / glock.c
blobdab2526071cc7b53bc18a9203deea61ab78aaae3
1 /*
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.
8 */
10 #include <linux/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/buffer_head.h>
14 #include <linux/delay.h>
15 #include <linux/sort.h>
16 #include <linux/jhash.h>
17 #include <linux/kallsyms.h>
18 #include <linux/gfs2_ondisk.h>
19 #include <linux/list.h>
20 #include <linux/wait.h>
21 #include <linux/module.h>
22 #include <asm/uaccess.h>
23 #include <linux/seq_file.h>
24 #include <linux/debugfs.h>
25 #include <linux/kthread.h>
26 #include <linux/freezer.h>
27 #include <linux/workqueue.h>
28 #include <linux/jiffies.h>
29 #include <linux/rcupdate.h>
30 #include <linux/rculist_bl.h>
31 #include <linux/bit_spinlock.h>
32 #include <linux/percpu.h>
34 #include "gfs2.h"
35 #include "incore.h"
36 #include "glock.h"
37 #include "glops.h"
38 #include "inode.h"
39 #include "lops.h"
40 #include "meta_io.h"
41 #include "quota.h"
42 #include "super.h"
43 #include "util.h"
44 #include "bmap.h"
45 #define CREATE_TRACE_POINTS
46 #include "trace_gfs2.h"
48 struct gfs2_glock_iter {
49 int hash; /* hash bucket index */
50 struct gfs2_sbd *sdp; /* incore superblock */
51 struct gfs2_glock *gl; /* current glock struct */
52 char string[512]; /* scratch space */
55 typedef void (*glock_examiner) (struct gfs2_glock * gl);
57 static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl);
58 #define GLOCK_BUG_ON(gl,x) do { if (unlikely(x)) { __dump_glock(NULL, gl); BUG(); } } while(0)
59 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
61 static struct dentry *gfs2_root;
62 static struct workqueue_struct *glock_workqueue;
63 struct workqueue_struct *gfs2_delete_workqueue;
64 static LIST_HEAD(lru_list);
65 static atomic_t lru_count = ATOMIC_INIT(0);
66 static DEFINE_SPINLOCK(lru_lock);
68 #define GFS2_GL_HASH_SHIFT 15
69 #define GFS2_GL_HASH_SIZE (1 << GFS2_GL_HASH_SHIFT)
70 #define GFS2_GL_HASH_MASK (GFS2_GL_HASH_SIZE - 1)
72 static struct hlist_bl_head gl_hash_table[GFS2_GL_HASH_SIZE];
73 static struct dentry *gfs2_root;
75 /**
76 * gl_hash() - Turn glock number into hash bucket number
77 * @lock: The glock number
79 * Returns: The number of the corresponding hash bucket
82 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
83 const struct lm_lockname *name)
85 unsigned int h;
87 h = jhash(&name->ln_number, sizeof(u64), 0);
88 h = jhash(&name->ln_type, sizeof(unsigned int), h);
89 h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
90 h &= GFS2_GL_HASH_MASK;
92 return h;
95 static inline void spin_lock_bucket(unsigned int hash)
97 hlist_bl_lock(&gl_hash_table[hash]);
100 static inline void spin_unlock_bucket(unsigned int hash)
102 hlist_bl_unlock(&gl_hash_table[hash]);
105 static void gfs2_glock_dealloc(struct rcu_head *rcu)
107 struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
109 if (gl->gl_ops->go_flags & GLOF_ASPACE)
110 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
111 else
112 kmem_cache_free(gfs2_glock_cachep, gl);
115 void gfs2_glock_free(struct gfs2_glock *gl)
117 struct gfs2_sbd *sdp = gl->gl_sbd;
119 call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
120 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
121 wake_up(&sdp->sd_glock_wait);
125 * gfs2_glock_hold() - increment reference count on glock
126 * @gl: The glock to hold
130 void gfs2_glock_hold(struct gfs2_glock *gl)
132 GLOCK_BUG_ON(gl, atomic_read(&gl->gl_ref) == 0);
133 atomic_inc(&gl->gl_ref);
137 * demote_ok - Check to see if it's ok to unlock a glock
138 * @gl: the glock
140 * Returns: 1 if it's ok
143 static int demote_ok(const struct gfs2_glock *gl)
145 const struct gfs2_glock_operations *glops = gl->gl_ops;
147 if (gl->gl_state == LM_ST_UNLOCKED)
148 return 0;
149 if (!list_empty(&gl->gl_holders))
150 return 0;
151 if (glops->go_demote_ok)
152 return glops->go_demote_ok(gl);
153 return 1;
157 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
159 spin_lock(&lru_lock);
161 if (!list_empty(&gl->gl_lru))
162 list_del_init(&gl->gl_lru);
163 else
164 atomic_inc(&lru_count);
166 list_add_tail(&gl->gl_lru, &lru_list);
167 set_bit(GLF_LRU, &gl->gl_flags);
168 spin_unlock(&lru_lock);
171 static void __gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
173 if (!list_empty(&gl->gl_lru)) {
174 list_del_init(&gl->gl_lru);
175 atomic_dec(&lru_count);
176 clear_bit(GLF_LRU, &gl->gl_flags);
180 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
182 spin_lock(&lru_lock);
183 __gfs2_glock_remove_from_lru(gl);
184 spin_unlock(&lru_lock);
188 * __gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
189 * @gl: the glock
191 * If the glock is demotable, then we add it (or move it) to the end
192 * of the glock LRU list.
195 static void __gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
197 if (demote_ok(gl))
198 gfs2_glock_add_to_lru(gl);
202 * gfs2_glock_put_nolock() - Decrement reference count on glock
203 * @gl: The glock to put
205 * This function should only be used if the caller has its own reference
206 * to the glock, in addition to the one it is dropping.
209 void gfs2_glock_put_nolock(struct gfs2_glock *gl)
211 if (atomic_dec_and_test(&gl->gl_ref))
212 GLOCK_BUG_ON(gl, 1);
216 * gfs2_glock_put() - Decrement reference count on glock
217 * @gl: The glock to put
221 void gfs2_glock_put(struct gfs2_glock *gl)
223 struct gfs2_sbd *sdp = gl->gl_sbd;
224 struct address_space *mapping = gfs2_glock2aspace(gl);
226 if (atomic_dec_and_lock(&gl->gl_ref, &lru_lock)) {
227 __gfs2_glock_remove_from_lru(gl);
228 spin_unlock(&lru_lock);
229 spin_lock_bucket(gl->gl_hash);
230 hlist_bl_del_rcu(&gl->gl_list);
231 spin_unlock_bucket(gl->gl_hash);
232 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
233 GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
234 trace_gfs2_glock_put(gl);
235 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
240 * search_bucket() - Find struct gfs2_glock by lock number
241 * @bucket: the bucket to search
242 * @name: The lock name
244 * Returns: NULL, or the struct gfs2_glock with the requested number
247 static struct gfs2_glock *search_bucket(unsigned int hash,
248 const struct gfs2_sbd *sdp,
249 const struct lm_lockname *name)
251 struct gfs2_glock *gl;
252 struct hlist_bl_node *h;
254 hlist_bl_for_each_entry_rcu(gl, h, &gl_hash_table[hash], gl_list) {
255 if (!lm_name_equal(&gl->gl_name, name))
256 continue;
257 if (gl->gl_sbd != sdp)
258 continue;
259 if (atomic_inc_not_zero(&gl->gl_ref))
260 return gl;
263 return NULL;
267 * may_grant - check if its ok to grant a new lock
268 * @gl: The glock
269 * @gh: The lock request which we wish to grant
271 * Returns: true if its ok to grant the lock
274 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
276 const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
277 if ((gh->gh_state == LM_ST_EXCLUSIVE ||
278 gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
279 return 0;
280 if (gl->gl_state == gh->gh_state)
281 return 1;
282 if (gh->gh_flags & GL_EXACT)
283 return 0;
284 if (gl->gl_state == LM_ST_EXCLUSIVE) {
285 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
286 return 1;
287 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
288 return 1;
290 if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
291 return 1;
292 return 0;
295 static void gfs2_holder_wake(struct gfs2_holder *gh)
297 clear_bit(HIF_WAIT, &gh->gh_iflags);
298 smp_mb__after_clear_bit();
299 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
303 * do_error - Something unexpected has happened during a lock request
307 static inline void do_error(struct gfs2_glock *gl, const int ret)
309 struct gfs2_holder *gh, *tmp;
311 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
312 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
313 continue;
314 if (ret & LM_OUT_ERROR)
315 gh->gh_error = -EIO;
316 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
317 gh->gh_error = GLR_TRYFAILED;
318 else
319 continue;
320 list_del_init(&gh->gh_list);
321 trace_gfs2_glock_queue(gh, 0);
322 gfs2_holder_wake(gh);
327 * do_promote - promote as many requests as possible on the current queue
328 * @gl: The glock
330 * Returns: 1 if there is a blocked holder at the head of the list, or 2
331 * if a type specific operation is underway.
334 static int do_promote(struct gfs2_glock *gl)
335 __releases(&gl->gl_spin)
336 __acquires(&gl->gl_spin)
338 const struct gfs2_glock_operations *glops = gl->gl_ops;
339 struct gfs2_holder *gh, *tmp;
340 int ret;
342 restart:
343 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
344 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
345 continue;
346 if (may_grant(gl, gh)) {
347 if (gh->gh_list.prev == &gl->gl_holders &&
348 glops->go_lock) {
349 spin_unlock(&gl->gl_spin);
350 /* FIXME: eliminate this eventually */
351 ret = glops->go_lock(gh);
352 spin_lock(&gl->gl_spin);
353 if (ret) {
354 if (ret == 1)
355 return 2;
356 gh->gh_error = ret;
357 list_del_init(&gh->gh_list);
358 trace_gfs2_glock_queue(gh, 0);
359 gfs2_holder_wake(gh);
360 goto restart;
362 set_bit(HIF_HOLDER, &gh->gh_iflags);
363 trace_gfs2_promote(gh, 1);
364 gfs2_holder_wake(gh);
365 goto restart;
367 set_bit(HIF_HOLDER, &gh->gh_iflags);
368 trace_gfs2_promote(gh, 0);
369 gfs2_holder_wake(gh);
370 continue;
372 if (gh->gh_list.prev == &gl->gl_holders)
373 return 1;
374 do_error(gl, 0);
375 break;
377 return 0;
381 * find_first_waiter - find the first gh that's waiting for the glock
382 * @gl: the glock
385 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
387 struct gfs2_holder *gh;
389 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
390 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
391 return gh;
393 return NULL;
397 * state_change - record that the glock is now in a different state
398 * @gl: the glock
399 * @new_state the new state
403 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
405 int held1, held2;
407 held1 = (gl->gl_state != LM_ST_UNLOCKED);
408 held2 = (new_state != LM_ST_UNLOCKED);
410 if (held1 != held2) {
411 if (held2)
412 gfs2_glock_hold(gl);
413 else
414 gfs2_glock_put_nolock(gl);
416 if (held1 && held2 && list_empty(&gl->gl_holders))
417 clear_bit(GLF_QUEUED, &gl->gl_flags);
419 if (new_state != gl->gl_target)
420 /* shorten our minimum hold time */
421 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
422 GL_GLOCK_MIN_HOLD);
423 gl->gl_state = new_state;
424 gl->gl_tchange = jiffies;
427 static void gfs2_demote_wake(struct gfs2_glock *gl)
429 gl->gl_demote_state = LM_ST_EXCLUSIVE;
430 clear_bit(GLF_DEMOTE, &gl->gl_flags);
431 smp_mb__after_clear_bit();
432 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
436 * finish_xmote - The DLM has replied to one of our lock requests
437 * @gl: The glock
438 * @ret: The status from the DLM
442 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
444 const struct gfs2_glock_operations *glops = gl->gl_ops;
445 struct gfs2_holder *gh;
446 unsigned state = ret & LM_OUT_ST_MASK;
447 int rv;
449 spin_lock(&gl->gl_spin);
450 trace_gfs2_glock_state_change(gl, state);
451 state_change(gl, state);
452 gh = find_first_waiter(gl);
454 /* Demote to UN request arrived during demote to SH or DF */
455 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
456 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
457 gl->gl_target = LM_ST_UNLOCKED;
459 /* Check for state != intended state */
460 if (unlikely(state != gl->gl_target)) {
461 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
462 /* move to back of queue and try next entry */
463 if (ret & LM_OUT_CANCELED) {
464 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
465 list_move_tail(&gh->gh_list, &gl->gl_holders);
466 gh = find_first_waiter(gl);
467 gl->gl_target = gh->gh_state;
468 goto retry;
470 /* Some error or failed "try lock" - report it */
471 if ((ret & LM_OUT_ERROR) ||
472 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
473 gl->gl_target = gl->gl_state;
474 do_error(gl, ret);
475 goto out;
478 switch(state) {
479 /* Unlocked due to conversion deadlock, try again */
480 case LM_ST_UNLOCKED:
481 retry:
482 do_xmote(gl, gh, gl->gl_target);
483 break;
484 /* Conversion fails, unlock and try again */
485 case LM_ST_SHARED:
486 case LM_ST_DEFERRED:
487 do_xmote(gl, gh, LM_ST_UNLOCKED);
488 break;
489 default: /* Everything else */
490 printk(KERN_ERR "GFS2: wanted %u got %u\n", gl->gl_target, state);
491 GLOCK_BUG_ON(gl, 1);
493 spin_unlock(&gl->gl_spin);
494 return;
497 /* Fast path - we got what we asked for */
498 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
499 gfs2_demote_wake(gl);
500 if (state != LM_ST_UNLOCKED) {
501 if (glops->go_xmote_bh) {
502 spin_unlock(&gl->gl_spin);
503 rv = glops->go_xmote_bh(gl, gh);
504 spin_lock(&gl->gl_spin);
505 if (rv) {
506 do_error(gl, rv);
507 goto out;
510 rv = do_promote(gl);
511 if (rv == 2)
512 goto out_locked;
514 out:
515 clear_bit(GLF_LOCK, &gl->gl_flags);
516 out_locked:
517 spin_unlock(&gl->gl_spin);
521 * do_xmote - Calls the DLM to change the state of a lock
522 * @gl: The lock state
523 * @gh: The holder (only for promotes)
524 * @target: The target lock state
528 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
529 __releases(&gl->gl_spin)
530 __acquires(&gl->gl_spin)
532 const struct gfs2_glock_operations *glops = gl->gl_ops;
533 struct gfs2_sbd *sdp = gl->gl_sbd;
534 unsigned int lck_flags = gh ? gh->gh_flags : 0;
535 int ret;
537 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
538 LM_FLAG_PRIORITY);
539 GLOCK_BUG_ON(gl, gl->gl_state == target);
540 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
541 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
542 glops->go_inval) {
543 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
544 do_error(gl, 0); /* Fail queued try locks */
546 gl->gl_req = target;
547 set_bit(GLF_BLOCKING, &gl->gl_flags);
548 if ((gl->gl_req == LM_ST_UNLOCKED) ||
549 (gl->gl_state == LM_ST_EXCLUSIVE) ||
550 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
551 clear_bit(GLF_BLOCKING, &gl->gl_flags);
552 spin_unlock(&gl->gl_spin);
553 if (glops->go_xmote_th)
554 glops->go_xmote_th(gl);
555 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
556 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
557 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
559 gfs2_glock_hold(gl);
560 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
561 /* lock_dlm */
562 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
563 GLOCK_BUG_ON(gl, ret);
564 } else { /* lock_nolock */
565 finish_xmote(gl, target);
566 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
567 gfs2_glock_put(gl);
570 spin_lock(&gl->gl_spin);
574 * find_first_holder - find the first "holder" gh
575 * @gl: the glock
578 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
580 struct gfs2_holder *gh;
582 if (!list_empty(&gl->gl_holders)) {
583 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
584 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
585 return gh;
587 return NULL;
591 * run_queue - do all outstanding tasks related to a glock
592 * @gl: The glock in question
593 * @nonblock: True if we must not block in run_queue
597 static void run_queue(struct gfs2_glock *gl, const int nonblock)
598 __releases(&gl->gl_spin)
599 __acquires(&gl->gl_spin)
601 struct gfs2_holder *gh = NULL;
602 int ret;
604 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
605 return;
607 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
609 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
610 gl->gl_demote_state != gl->gl_state) {
611 if (find_first_holder(gl))
612 goto out_unlock;
613 if (nonblock)
614 goto out_sched;
615 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
616 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
617 gl->gl_target = gl->gl_demote_state;
618 } else {
619 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
620 gfs2_demote_wake(gl);
621 ret = do_promote(gl);
622 if (ret == 0)
623 goto out_unlock;
624 if (ret == 2)
625 goto out;
626 gh = find_first_waiter(gl);
627 gl->gl_target = gh->gh_state;
628 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
629 do_error(gl, 0); /* Fail queued try locks */
631 do_xmote(gl, gh, gl->gl_target);
632 out:
633 return;
635 out_sched:
636 clear_bit(GLF_LOCK, &gl->gl_flags);
637 smp_mb__after_clear_bit();
638 gfs2_glock_hold(gl);
639 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
640 gfs2_glock_put_nolock(gl);
641 return;
643 out_unlock:
644 clear_bit(GLF_LOCK, &gl->gl_flags);
645 smp_mb__after_clear_bit();
646 return;
649 static void delete_work_func(struct work_struct *work)
651 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
652 struct gfs2_sbd *sdp = gl->gl_sbd;
653 struct gfs2_inode *ip;
654 struct inode *inode;
655 u64 no_addr = gl->gl_name.ln_number;
657 ip = gl->gl_object;
658 /* Note: Unsafe to dereference ip as we don't hold right refs/locks */
660 if (ip)
661 inode = gfs2_ilookup(sdp->sd_vfs, no_addr, 1);
662 else
663 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
664 if (inode && !IS_ERR(inode)) {
665 d_prune_aliases(inode);
666 iput(inode);
668 gfs2_glock_put(gl);
671 static void glock_work_func(struct work_struct *work)
673 unsigned long delay = 0;
674 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
675 int drop_ref = 0;
677 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
678 finish_xmote(gl, gl->gl_reply);
679 drop_ref = 1;
681 spin_lock(&gl->gl_spin);
682 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
683 gl->gl_state != LM_ST_UNLOCKED &&
684 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
685 unsigned long holdtime, now = jiffies;
687 holdtime = gl->gl_tchange + gl->gl_hold_time;
688 if (time_before(now, holdtime))
689 delay = holdtime - now;
691 if (!delay) {
692 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
693 set_bit(GLF_DEMOTE, &gl->gl_flags);
696 run_queue(gl, 0);
697 spin_unlock(&gl->gl_spin);
698 if (!delay)
699 gfs2_glock_put(gl);
700 else {
701 if (gl->gl_name.ln_type != LM_TYPE_INODE)
702 delay = 0;
703 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
704 gfs2_glock_put(gl);
706 if (drop_ref)
707 gfs2_glock_put(gl);
711 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
712 * @sdp: The GFS2 superblock
713 * @number: the lock number
714 * @glops: The glock_operations to use
715 * @create: If 0, don't create the glock if it doesn't exist
716 * @glp: the glock is returned here
718 * This does not lock a glock, just finds/creates structures for one.
720 * Returns: errno
723 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
724 const struct gfs2_glock_operations *glops, int create,
725 struct gfs2_glock **glp)
727 struct super_block *s = sdp->sd_vfs;
728 struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
729 struct gfs2_glock *gl, *tmp;
730 unsigned int hash = gl_hash(sdp, &name);
731 struct address_space *mapping;
732 struct kmem_cache *cachep;
734 rcu_read_lock();
735 gl = search_bucket(hash, sdp, &name);
736 rcu_read_unlock();
738 *glp = gl;
739 if (gl)
740 return 0;
741 if (!create)
742 return -ENOENT;
744 if (glops->go_flags & GLOF_ASPACE)
745 cachep = gfs2_glock_aspace_cachep;
746 else
747 cachep = gfs2_glock_cachep;
748 gl = kmem_cache_alloc(cachep, GFP_KERNEL);
749 if (!gl)
750 return -ENOMEM;
752 atomic_inc(&sdp->sd_glock_disposal);
753 gl->gl_sbd = sdp;
754 gl->gl_flags = 0;
755 gl->gl_name = name;
756 atomic_set(&gl->gl_ref, 1);
757 gl->gl_state = LM_ST_UNLOCKED;
758 gl->gl_target = LM_ST_UNLOCKED;
759 gl->gl_demote_state = LM_ST_EXCLUSIVE;
760 gl->gl_hash = hash;
761 gl->gl_ops = glops;
762 gl->gl_dstamp = ktime_set(0, 0);
763 preempt_disable();
764 /* We use the global stats to estimate the initial per-glock stats */
765 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
766 preempt_enable();
767 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
768 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
769 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
770 gl->gl_lksb.sb_lvbptr = gl->gl_lvb;
771 gl->gl_tchange = jiffies;
772 gl->gl_object = NULL;
773 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
774 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
775 INIT_WORK(&gl->gl_delete, delete_work_func);
777 mapping = gfs2_glock2aspace(gl);
778 if (mapping) {
779 mapping->a_ops = &gfs2_meta_aops;
780 mapping->host = s->s_bdev->bd_inode;
781 mapping->flags = 0;
782 mapping_set_gfp_mask(mapping, GFP_NOFS);
783 mapping->assoc_mapping = NULL;
784 mapping->backing_dev_info = s->s_bdi;
785 mapping->writeback_index = 0;
788 spin_lock_bucket(hash);
789 tmp = search_bucket(hash, sdp, &name);
790 if (tmp) {
791 spin_unlock_bucket(hash);
792 kmem_cache_free(cachep, gl);
793 atomic_dec(&sdp->sd_glock_disposal);
794 gl = tmp;
795 } else {
796 hlist_bl_add_head_rcu(&gl->gl_list, &gl_hash_table[hash]);
797 spin_unlock_bucket(hash);
800 *glp = gl;
802 return 0;
806 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
807 * @gl: the glock
808 * @state: the state we're requesting
809 * @flags: the modifier flags
810 * @gh: the holder structure
814 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
815 struct gfs2_holder *gh)
817 INIT_LIST_HEAD(&gh->gh_list);
818 gh->gh_gl = gl;
819 gh->gh_ip = (unsigned long)__builtin_return_address(0);
820 gh->gh_owner_pid = get_pid(task_pid(current));
821 gh->gh_state = state;
822 gh->gh_flags = flags;
823 gh->gh_error = 0;
824 gh->gh_iflags = 0;
825 gfs2_glock_hold(gl);
829 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
830 * @state: the state we're requesting
831 * @flags: the modifier flags
832 * @gh: the holder structure
834 * Don't mess with the glock.
838 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
840 gh->gh_state = state;
841 gh->gh_flags = flags;
842 gh->gh_iflags = 0;
843 gh->gh_ip = (unsigned long)__builtin_return_address(0);
844 if (gh->gh_owner_pid)
845 put_pid(gh->gh_owner_pid);
846 gh->gh_owner_pid = get_pid(task_pid(current));
850 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
851 * @gh: the holder structure
855 void gfs2_holder_uninit(struct gfs2_holder *gh)
857 put_pid(gh->gh_owner_pid);
858 gfs2_glock_put(gh->gh_gl);
859 gh->gh_gl = NULL;
860 gh->gh_ip = 0;
864 * gfs2_glock_holder_wait
865 * @word: unused
867 * This function and gfs2_glock_demote_wait both show up in the WCHAN
868 * field. Thus I've separated these otherwise identical functions in
869 * order to be more informative to the user.
872 static int gfs2_glock_holder_wait(void *word)
874 schedule();
875 return 0;
878 static int gfs2_glock_demote_wait(void *word)
880 schedule();
881 return 0;
884 static void wait_on_holder(struct gfs2_holder *gh)
886 unsigned long time1 = jiffies;
888 might_sleep();
889 wait_on_bit(&gh->gh_iflags, HIF_WAIT, gfs2_glock_holder_wait, TASK_UNINTERRUPTIBLE);
890 if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
891 /* Lengthen the minimum hold time. */
892 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
893 GL_GLOCK_HOLD_INCR,
894 GL_GLOCK_MAX_HOLD);
897 static void wait_on_demote(struct gfs2_glock *gl)
899 might_sleep();
900 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, gfs2_glock_demote_wait, TASK_UNINTERRUPTIBLE);
904 * handle_callback - process a demote request
905 * @gl: the glock
906 * @state: the state the caller wants us to change to
908 * There are only two requests that we are going to see in actual
909 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
912 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
913 unsigned long delay)
915 int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
917 set_bit(bit, &gl->gl_flags);
918 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
919 gl->gl_demote_state = state;
920 gl->gl_demote_time = jiffies;
921 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
922 gl->gl_demote_state != state) {
923 gl->gl_demote_state = LM_ST_UNLOCKED;
925 if (gl->gl_ops->go_callback)
926 gl->gl_ops->go_callback(gl);
927 trace_gfs2_demote_rq(gl);
931 * gfs2_glock_wait - wait on a glock acquisition
932 * @gh: the glock holder
934 * Returns: 0 on success
937 int gfs2_glock_wait(struct gfs2_holder *gh)
939 wait_on_holder(gh);
940 return gh->gh_error;
943 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
945 struct va_format vaf;
946 va_list args;
948 va_start(args, fmt);
950 if (seq) {
951 struct gfs2_glock_iter *gi = seq->private;
952 vsprintf(gi->string, fmt, args);
953 seq_printf(seq, gi->string);
954 } else {
955 vaf.fmt = fmt;
956 vaf.va = &args;
958 printk(KERN_ERR " %pV", &vaf);
961 va_end(args);
965 * add_to_queue - Add a holder to the wait queue (but look for recursion)
966 * @gh: the holder structure to add
968 * Eventually we should move the recursive locking trap to a
969 * debugging option or something like that. This is the fast
970 * path and needs to have the minimum number of distractions.
974 static inline void add_to_queue(struct gfs2_holder *gh)
975 __releases(&gl->gl_spin)
976 __acquires(&gl->gl_spin)
978 struct gfs2_glock *gl = gh->gh_gl;
979 struct gfs2_sbd *sdp = gl->gl_sbd;
980 struct list_head *insert_pt = NULL;
981 struct gfs2_holder *gh2;
982 int try_lock = 0;
984 BUG_ON(gh->gh_owner_pid == NULL);
985 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
986 BUG();
988 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
989 if (test_bit(GLF_LOCK, &gl->gl_flags))
990 try_lock = 1;
991 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
992 goto fail;
995 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
996 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
997 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
998 goto trap_recursive;
999 if (try_lock &&
1000 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) &&
1001 !may_grant(gl, gh)) {
1002 fail:
1003 gh->gh_error = GLR_TRYFAILED;
1004 gfs2_holder_wake(gh);
1005 return;
1007 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1008 continue;
1009 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1010 insert_pt = &gh2->gh_list;
1012 set_bit(GLF_QUEUED, &gl->gl_flags);
1013 trace_gfs2_glock_queue(gh, 1);
1014 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1015 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1016 if (likely(insert_pt == NULL)) {
1017 list_add_tail(&gh->gh_list, &gl->gl_holders);
1018 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1019 goto do_cancel;
1020 return;
1022 list_add_tail(&gh->gh_list, insert_pt);
1023 do_cancel:
1024 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
1025 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1026 spin_unlock(&gl->gl_spin);
1027 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1028 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1029 spin_lock(&gl->gl_spin);
1031 return;
1033 trap_recursive:
1034 print_symbol(KERN_ERR "original: %s\n", gh2->gh_ip);
1035 printk(KERN_ERR "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1036 printk(KERN_ERR "lock type: %d req lock state : %d\n",
1037 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1038 print_symbol(KERN_ERR "new: %s\n", gh->gh_ip);
1039 printk(KERN_ERR "pid: %d\n", pid_nr(gh->gh_owner_pid));
1040 printk(KERN_ERR "lock type: %d req lock state : %d\n",
1041 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1042 __dump_glock(NULL, gl);
1043 BUG();
1047 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1048 * @gh: the holder structure
1050 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1052 * Returns: 0, GLR_TRYFAILED, or errno on failure
1055 int gfs2_glock_nq(struct gfs2_holder *gh)
1057 struct gfs2_glock *gl = gh->gh_gl;
1058 struct gfs2_sbd *sdp = gl->gl_sbd;
1059 int error = 0;
1061 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1062 return -EIO;
1064 if (test_bit(GLF_LRU, &gl->gl_flags))
1065 gfs2_glock_remove_from_lru(gl);
1067 spin_lock(&gl->gl_spin);
1068 add_to_queue(gh);
1069 if ((LM_FLAG_NOEXP & gh->gh_flags) &&
1070 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1071 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1072 run_queue(gl, 1);
1073 spin_unlock(&gl->gl_spin);
1075 if (!(gh->gh_flags & GL_ASYNC))
1076 error = gfs2_glock_wait(gh);
1078 return error;
1082 * gfs2_glock_poll - poll to see if an async request has been completed
1083 * @gh: the holder
1085 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1088 int gfs2_glock_poll(struct gfs2_holder *gh)
1090 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1094 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1095 * @gh: the glock holder
1099 void gfs2_glock_dq(struct gfs2_holder *gh)
1101 struct gfs2_glock *gl = gh->gh_gl;
1102 const struct gfs2_glock_operations *glops = gl->gl_ops;
1103 unsigned delay = 0;
1104 int fast_path = 0;
1106 spin_lock(&gl->gl_spin);
1107 if (gh->gh_flags & GL_NOCACHE)
1108 handle_callback(gl, LM_ST_UNLOCKED, 0);
1110 list_del_init(&gh->gh_list);
1111 if (find_first_holder(gl) == NULL) {
1112 if (glops->go_unlock) {
1113 GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1114 spin_unlock(&gl->gl_spin);
1115 glops->go_unlock(gh);
1116 spin_lock(&gl->gl_spin);
1117 clear_bit(GLF_LOCK, &gl->gl_flags);
1119 if (list_empty(&gl->gl_holders) &&
1120 !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1121 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1122 fast_path = 1;
1124 if (!test_bit(GLF_LFLUSH, &gl->gl_flags))
1125 __gfs2_glock_schedule_for_reclaim(gl);
1126 trace_gfs2_glock_queue(gh, 0);
1127 spin_unlock(&gl->gl_spin);
1128 if (likely(fast_path))
1129 return;
1131 gfs2_glock_hold(gl);
1132 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1133 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1134 gl->gl_name.ln_type == LM_TYPE_INODE)
1135 delay = gl->gl_hold_time;
1136 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1137 gfs2_glock_put(gl);
1140 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1142 struct gfs2_glock *gl = gh->gh_gl;
1143 gfs2_glock_dq(gh);
1144 wait_on_demote(gl);
1148 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1149 * @gh: the holder structure
1153 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1155 gfs2_glock_dq(gh);
1156 gfs2_holder_uninit(gh);
1160 * gfs2_glock_nq_num - acquire a glock based on lock number
1161 * @sdp: the filesystem
1162 * @number: the lock number
1163 * @glops: the glock operations for the type of glock
1164 * @state: the state to acquire the glock in
1165 * @flags: modifier flags for the acquisition
1166 * @gh: the struct gfs2_holder
1168 * Returns: errno
1171 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1172 const struct gfs2_glock_operations *glops,
1173 unsigned int state, int flags, struct gfs2_holder *gh)
1175 struct gfs2_glock *gl;
1176 int error;
1178 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1179 if (!error) {
1180 error = gfs2_glock_nq_init(gl, state, flags, gh);
1181 gfs2_glock_put(gl);
1184 return error;
1188 * glock_compare - Compare two struct gfs2_glock structures for sorting
1189 * @arg_a: the first structure
1190 * @arg_b: the second structure
1194 static int glock_compare(const void *arg_a, const void *arg_b)
1196 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1197 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1198 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1199 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1201 if (a->ln_number > b->ln_number)
1202 return 1;
1203 if (a->ln_number < b->ln_number)
1204 return -1;
1205 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1206 return 0;
1210 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1211 * @num_gh: the number of structures
1212 * @ghs: an array of struct gfs2_holder structures
1214 * Returns: 0 on success (all glocks acquired),
1215 * errno on failure (no glocks acquired)
1218 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1219 struct gfs2_holder **p)
1221 unsigned int x;
1222 int error = 0;
1224 for (x = 0; x < num_gh; x++)
1225 p[x] = &ghs[x];
1227 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1229 for (x = 0; x < num_gh; x++) {
1230 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1232 error = gfs2_glock_nq(p[x]);
1233 if (error) {
1234 while (x--)
1235 gfs2_glock_dq(p[x]);
1236 break;
1240 return error;
1244 * gfs2_glock_nq_m - acquire multiple glocks
1245 * @num_gh: the number of structures
1246 * @ghs: an array of struct gfs2_holder structures
1249 * Returns: 0 on success (all glocks acquired),
1250 * errno on failure (no glocks acquired)
1253 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1255 struct gfs2_holder *tmp[4];
1256 struct gfs2_holder **pph = tmp;
1257 int error = 0;
1259 switch(num_gh) {
1260 case 0:
1261 return 0;
1262 case 1:
1263 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1264 return gfs2_glock_nq(ghs);
1265 default:
1266 if (num_gh <= 4)
1267 break;
1268 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1269 if (!pph)
1270 return -ENOMEM;
1273 error = nq_m_sync(num_gh, ghs, pph);
1275 if (pph != tmp)
1276 kfree(pph);
1278 return error;
1282 * gfs2_glock_dq_m - release multiple glocks
1283 * @num_gh: the number of structures
1284 * @ghs: an array of struct gfs2_holder structures
1288 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1290 while (num_gh--)
1291 gfs2_glock_dq(&ghs[num_gh]);
1295 * gfs2_glock_dq_uninit_m - release multiple glocks
1296 * @num_gh: the number of structures
1297 * @ghs: an array of struct gfs2_holder structures
1301 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1303 while (num_gh--)
1304 gfs2_glock_dq_uninit(&ghs[num_gh]);
1307 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1309 unsigned long delay = 0;
1310 unsigned long holdtime;
1311 unsigned long now = jiffies;
1313 gfs2_glock_hold(gl);
1314 holdtime = gl->gl_tchange + gl->gl_hold_time;
1315 if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1316 gl->gl_name.ln_type == LM_TYPE_INODE) {
1317 if (time_before(now, holdtime))
1318 delay = holdtime - now;
1319 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1320 delay = gl->gl_hold_time;
1323 spin_lock(&gl->gl_spin);
1324 handle_callback(gl, state, delay);
1325 spin_unlock(&gl->gl_spin);
1326 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1327 gfs2_glock_put(gl);
1331 * gfs2_should_freeze - Figure out if glock should be frozen
1332 * @gl: The glock in question
1334 * Glocks are not frozen if (a) the result of the dlm operation is
1335 * an error, (b) the locking operation was an unlock operation or
1336 * (c) if there is a "noexp" flagged request anywhere in the queue
1338 * Returns: 1 if freezing should occur, 0 otherwise
1341 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1343 const struct gfs2_holder *gh;
1345 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1346 return 0;
1347 if (gl->gl_target == LM_ST_UNLOCKED)
1348 return 0;
1350 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1351 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1352 continue;
1353 if (LM_FLAG_NOEXP & gh->gh_flags)
1354 return 0;
1357 return 1;
1361 * gfs2_glock_complete - Callback used by locking
1362 * @gl: Pointer to the glock
1363 * @ret: The return value from the dlm
1365 * The gl_reply field is under the gl_spin lock so that it is ok
1366 * to use a bitfield shared with other glock state fields.
1369 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1371 struct lm_lockstruct *ls = &gl->gl_sbd->sd_lockstruct;
1373 spin_lock(&gl->gl_spin);
1374 gl->gl_reply = ret;
1376 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1377 if (gfs2_should_freeze(gl)) {
1378 set_bit(GLF_FROZEN, &gl->gl_flags);
1379 spin_unlock(&gl->gl_spin);
1380 return;
1384 spin_unlock(&gl->gl_spin);
1385 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1386 smp_wmb();
1387 gfs2_glock_hold(gl);
1388 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1389 gfs2_glock_put(gl);
1393 static int gfs2_shrink_glock_memory(struct shrinker *shrink,
1394 struct shrink_control *sc)
1396 struct gfs2_glock *gl;
1397 int may_demote;
1398 int nr_skipped = 0;
1399 int nr = sc->nr_to_scan;
1400 gfp_t gfp_mask = sc->gfp_mask;
1401 LIST_HEAD(skipped);
1403 if (nr == 0)
1404 goto out;
1406 if (!(gfp_mask & __GFP_FS))
1407 return -1;
1409 spin_lock(&lru_lock);
1410 while(nr && !list_empty(&lru_list)) {
1411 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1412 list_del_init(&gl->gl_lru);
1413 clear_bit(GLF_LRU, &gl->gl_flags);
1414 atomic_dec(&lru_count);
1416 /* Test for being demotable */
1417 if (!test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1418 gfs2_glock_hold(gl);
1419 spin_unlock(&lru_lock);
1420 spin_lock(&gl->gl_spin);
1421 may_demote = demote_ok(gl);
1422 if (may_demote) {
1423 handle_callback(gl, LM_ST_UNLOCKED, 0);
1424 nr--;
1426 clear_bit(GLF_LOCK, &gl->gl_flags);
1427 smp_mb__after_clear_bit();
1428 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1429 gfs2_glock_put_nolock(gl);
1430 spin_unlock(&gl->gl_spin);
1431 spin_lock(&lru_lock);
1432 continue;
1434 nr_skipped++;
1435 list_add(&gl->gl_lru, &skipped);
1436 set_bit(GLF_LRU, &gl->gl_flags);
1438 list_splice(&skipped, &lru_list);
1439 atomic_add(nr_skipped, &lru_count);
1440 spin_unlock(&lru_lock);
1441 out:
1442 return (atomic_read(&lru_count) / 100) * sysctl_vfs_cache_pressure;
1445 static struct shrinker glock_shrinker = {
1446 .shrink = gfs2_shrink_glock_memory,
1447 .seeks = DEFAULT_SEEKS,
1451 * examine_bucket - Call a function for glock in a hash bucket
1452 * @examiner: the function
1453 * @sdp: the filesystem
1454 * @bucket: the bucket
1458 static void examine_bucket(glock_examiner examiner, const struct gfs2_sbd *sdp,
1459 unsigned int hash)
1461 struct gfs2_glock *gl;
1462 struct hlist_bl_head *head = &gl_hash_table[hash];
1463 struct hlist_bl_node *pos;
1465 rcu_read_lock();
1466 hlist_bl_for_each_entry_rcu(gl, pos, head, gl_list) {
1467 if ((gl->gl_sbd == sdp) && atomic_read(&gl->gl_ref))
1468 examiner(gl);
1470 rcu_read_unlock();
1471 cond_resched();
1474 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1476 unsigned x;
1478 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1479 examine_bucket(examiner, sdp, x);
1484 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1485 * @gl: The glock to thaw
1487 * N.B. When we freeze a glock, we leave a ref to the glock outstanding,
1488 * so this has to result in the ref count being dropped by one.
1491 static void thaw_glock(struct gfs2_glock *gl)
1493 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1494 return;
1495 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1496 gfs2_glock_hold(gl);
1497 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1498 gfs2_glock_put(gl);
1502 * clear_glock - look at a glock and see if we can free it from glock cache
1503 * @gl: the glock to look at
1507 static void clear_glock(struct gfs2_glock *gl)
1509 gfs2_glock_remove_from_lru(gl);
1511 spin_lock(&gl->gl_spin);
1512 if (gl->gl_state != LM_ST_UNLOCKED)
1513 handle_callback(gl, LM_ST_UNLOCKED, 0);
1514 spin_unlock(&gl->gl_spin);
1515 gfs2_glock_hold(gl);
1516 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1517 gfs2_glock_put(gl);
1521 * gfs2_glock_thaw - Thaw any frozen glocks
1522 * @sdp: The super block
1526 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1528 glock_hash_walk(thaw_glock, sdp);
1531 static int dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1533 int ret;
1534 spin_lock(&gl->gl_spin);
1535 ret = __dump_glock(seq, gl);
1536 spin_unlock(&gl->gl_spin);
1537 return ret;
1540 static void dump_glock_func(struct gfs2_glock *gl)
1542 dump_glock(NULL, gl);
1546 * gfs2_gl_hash_clear - Empty out the glock hash table
1547 * @sdp: the filesystem
1548 * @wait: wait until it's all gone
1550 * Called when unmounting the filesystem.
1553 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1555 glock_hash_walk(clear_glock, sdp);
1556 flush_workqueue(glock_workqueue);
1557 wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
1558 glock_hash_walk(dump_glock_func, sdp);
1561 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1563 struct gfs2_glock *gl = ip->i_gl;
1564 int ret;
1566 ret = gfs2_truncatei_resume(ip);
1567 gfs2_assert_withdraw(gl->gl_sbd, ret == 0);
1569 spin_lock(&gl->gl_spin);
1570 clear_bit(GLF_LOCK, &gl->gl_flags);
1571 run_queue(gl, 1);
1572 spin_unlock(&gl->gl_spin);
1575 static const char *state2str(unsigned state)
1577 switch(state) {
1578 case LM_ST_UNLOCKED:
1579 return "UN";
1580 case LM_ST_SHARED:
1581 return "SH";
1582 case LM_ST_DEFERRED:
1583 return "DF";
1584 case LM_ST_EXCLUSIVE:
1585 return "EX";
1587 return "??";
1590 static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1592 char *p = buf;
1593 if (flags & LM_FLAG_TRY)
1594 *p++ = 't';
1595 if (flags & LM_FLAG_TRY_1CB)
1596 *p++ = 'T';
1597 if (flags & LM_FLAG_NOEXP)
1598 *p++ = 'e';
1599 if (flags & LM_FLAG_ANY)
1600 *p++ = 'A';
1601 if (flags & LM_FLAG_PRIORITY)
1602 *p++ = 'p';
1603 if (flags & GL_ASYNC)
1604 *p++ = 'a';
1605 if (flags & GL_EXACT)
1606 *p++ = 'E';
1607 if (flags & GL_NOCACHE)
1608 *p++ = 'c';
1609 if (test_bit(HIF_HOLDER, &iflags))
1610 *p++ = 'H';
1611 if (test_bit(HIF_WAIT, &iflags))
1612 *p++ = 'W';
1613 if (test_bit(HIF_FIRST, &iflags))
1614 *p++ = 'F';
1615 *p = 0;
1616 return buf;
1620 * dump_holder - print information about a glock holder
1621 * @seq: the seq_file struct
1622 * @gh: the glock holder
1624 * Returns: 0 on success, -ENOBUFS when we run out of space
1627 static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1629 struct task_struct *gh_owner = NULL;
1630 char flags_buf[32];
1632 if (gh->gh_owner_pid)
1633 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1634 gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1635 state2str(gh->gh_state),
1636 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1637 gh->gh_error,
1638 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1639 gh_owner ? gh_owner->comm : "(ended)",
1640 (void *)gh->gh_ip);
1641 return 0;
1644 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1646 const unsigned long *gflags = &gl->gl_flags;
1647 char *p = buf;
1649 if (test_bit(GLF_LOCK, gflags))
1650 *p++ = 'l';
1651 if (test_bit(GLF_DEMOTE, gflags))
1652 *p++ = 'D';
1653 if (test_bit(GLF_PENDING_DEMOTE, gflags))
1654 *p++ = 'd';
1655 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1656 *p++ = 'p';
1657 if (test_bit(GLF_DIRTY, gflags))
1658 *p++ = 'y';
1659 if (test_bit(GLF_LFLUSH, gflags))
1660 *p++ = 'f';
1661 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1662 *p++ = 'i';
1663 if (test_bit(GLF_REPLY_PENDING, gflags))
1664 *p++ = 'r';
1665 if (test_bit(GLF_INITIAL, gflags))
1666 *p++ = 'I';
1667 if (test_bit(GLF_FROZEN, gflags))
1668 *p++ = 'F';
1669 if (test_bit(GLF_QUEUED, gflags))
1670 *p++ = 'q';
1671 if (test_bit(GLF_LRU, gflags))
1672 *p++ = 'L';
1673 if (gl->gl_object)
1674 *p++ = 'o';
1675 if (test_bit(GLF_BLOCKING, gflags))
1676 *p++ = 'b';
1677 *p = 0;
1678 return buf;
1682 * __dump_glock - print information about a glock
1683 * @seq: The seq_file struct
1684 * @gl: the glock
1686 * The file format is as follows:
1687 * One line per object, capital letters are used to indicate objects
1688 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1689 * other objects are indented by a single space and follow the glock to
1690 * which they are related. Fields are indicated by lower case letters
1691 * followed by a colon and the field value, except for strings which are in
1692 * [] so that its possible to see if they are composed of spaces for
1693 * example. The field's are n = number (id of the object), f = flags,
1694 * t = type, s = state, r = refcount, e = error, p = pid.
1696 * Returns: 0 on success, -ENOBUFS when we run out of space
1699 static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1701 const struct gfs2_glock_operations *glops = gl->gl_ops;
1702 unsigned long long dtime;
1703 const struct gfs2_holder *gh;
1704 char gflags_buf[32];
1705 int error = 0;
1707 dtime = jiffies - gl->gl_demote_time;
1708 dtime *= 1000000/HZ; /* demote time in uSec */
1709 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1710 dtime = 0;
1711 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",
1712 state2str(gl->gl_state),
1713 gl->gl_name.ln_type,
1714 (unsigned long long)gl->gl_name.ln_number,
1715 gflags2str(gflags_buf, gl),
1716 state2str(gl->gl_target),
1717 state2str(gl->gl_demote_state), dtime,
1718 atomic_read(&gl->gl_ail_count),
1719 atomic_read(&gl->gl_revokes),
1720 atomic_read(&gl->gl_ref), gl->gl_hold_time);
1722 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1723 error = dump_holder(seq, gh);
1724 if (error)
1725 goto out;
1727 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1728 error = glops->go_dump(seq, gl);
1729 out:
1730 return error;
1733 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
1735 struct gfs2_glock *gl = iter_ptr;
1737 seq_printf(seq, "G: n:%u/%llx rtt:%lld/%lld rttb:%lld/%lld irt:%lld/%lld dcnt: %lld qcnt: %lld\n",
1738 gl->gl_name.ln_type,
1739 (unsigned long long)gl->gl_name.ln_number,
1740 (long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
1741 (long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
1742 (long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
1743 (long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
1744 (long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
1745 (long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
1746 (long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
1747 (long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
1748 return 0;
1751 static const char *gfs2_gltype[] = {
1752 "type",
1753 "reserved",
1754 "nondisk",
1755 "inode",
1756 "rgrp",
1757 "meta",
1758 "iopen",
1759 "flock",
1760 "plock",
1761 "quota",
1762 "journal",
1765 static const char *gfs2_stype[] = {
1766 [GFS2_LKS_SRTT] = "srtt",
1767 [GFS2_LKS_SRTTVAR] = "srttvar",
1768 [GFS2_LKS_SRTTB] = "srttb",
1769 [GFS2_LKS_SRTTVARB] = "srttvarb",
1770 [GFS2_LKS_SIRT] = "sirt",
1771 [GFS2_LKS_SIRTVAR] = "sirtvar",
1772 [GFS2_LKS_DCOUNT] = "dlm",
1773 [GFS2_LKS_QCOUNT] = "queue",
1776 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1778 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
1780 struct gfs2_glock_iter *gi = seq->private;
1781 struct gfs2_sbd *sdp = gi->sdp;
1782 unsigned index = gi->hash >> 3;
1783 unsigned subindex = gi->hash & 0x07;
1784 s64 value;
1785 int i;
1787 if (index == 0 && subindex != 0)
1788 return 0;
1790 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
1791 (index == 0) ? "cpu": gfs2_stype[subindex]);
1793 for_each_possible_cpu(i) {
1794 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
1795 if (index == 0) {
1796 value = i;
1797 } else {
1798 value = lkstats->lkstats[index - 1].stats[subindex];
1800 seq_printf(seq, " %15lld", (long long)value);
1802 seq_putc(seq, '\n');
1803 return 0;
1806 int __init gfs2_glock_init(void)
1808 unsigned i;
1809 for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
1810 INIT_HLIST_BL_HEAD(&gl_hash_table[i]);
1813 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1814 WQ_HIGHPRI | WQ_FREEZABLE, 0);
1815 if (IS_ERR(glock_workqueue))
1816 return PTR_ERR(glock_workqueue);
1817 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1818 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1820 if (IS_ERR(gfs2_delete_workqueue)) {
1821 destroy_workqueue(glock_workqueue);
1822 return PTR_ERR(gfs2_delete_workqueue);
1825 register_shrinker(&glock_shrinker);
1827 return 0;
1830 void gfs2_glock_exit(void)
1832 unregister_shrinker(&glock_shrinker);
1833 destroy_workqueue(glock_workqueue);
1834 destroy_workqueue(gfs2_delete_workqueue);
1837 static inline struct gfs2_glock *glock_hash_chain(unsigned hash)
1839 return hlist_bl_entry(hlist_bl_first_rcu(&gl_hash_table[hash]),
1840 struct gfs2_glock, gl_list);
1843 static inline struct gfs2_glock *glock_hash_next(struct gfs2_glock *gl)
1845 return hlist_bl_entry(rcu_dereference(gl->gl_list.next),
1846 struct gfs2_glock, gl_list);
1849 static int gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1851 struct gfs2_glock *gl;
1853 do {
1854 gl = gi->gl;
1855 if (gl) {
1856 gi->gl = glock_hash_next(gl);
1857 } else {
1858 gi->gl = glock_hash_chain(gi->hash);
1860 while (gi->gl == NULL) {
1861 gi->hash++;
1862 if (gi->hash >= GFS2_GL_HASH_SIZE) {
1863 rcu_read_unlock();
1864 return 1;
1866 gi->gl = glock_hash_chain(gi->hash);
1868 /* Skip entries for other sb and dead entries */
1869 } while (gi->sdp != gi->gl->gl_sbd || atomic_read(&gi->gl->gl_ref) == 0);
1871 return 0;
1874 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1876 struct gfs2_glock_iter *gi = seq->private;
1877 loff_t n = *pos;
1879 gi->hash = 0;
1880 rcu_read_lock();
1882 do {
1883 if (gfs2_glock_iter_next(gi))
1884 return NULL;
1885 } while (n--);
1887 return gi->gl;
1890 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1891 loff_t *pos)
1893 struct gfs2_glock_iter *gi = seq->private;
1895 (*pos)++;
1897 if (gfs2_glock_iter_next(gi))
1898 return NULL;
1900 return gi->gl;
1903 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1905 struct gfs2_glock_iter *gi = seq->private;
1907 if (gi->gl)
1908 rcu_read_unlock();
1909 gi->gl = NULL;
1912 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1914 return dump_glock(seq, iter_ptr);
1917 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
1919 struct gfs2_glock_iter *gi = seq->private;
1921 gi->hash = *pos;
1922 if (*pos >= GFS2_NR_SBSTATS)
1923 return NULL;
1924 preempt_disable();
1925 return SEQ_START_TOKEN;
1928 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
1929 loff_t *pos)
1931 struct gfs2_glock_iter *gi = seq->private;
1932 (*pos)++;
1933 gi->hash++;
1934 if (gi->hash >= GFS2_NR_SBSTATS) {
1935 preempt_enable();
1936 return NULL;
1938 return SEQ_START_TOKEN;
1941 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
1943 preempt_enable();
1946 static const struct seq_operations gfs2_glock_seq_ops = {
1947 .start = gfs2_glock_seq_start,
1948 .next = gfs2_glock_seq_next,
1949 .stop = gfs2_glock_seq_stop,
1950 .show = gfs2_glock_seq_show,
1953 static const struct seq_operations gfs2_glstats_seq_ops = {
1954 .start = gfs2_glock_seq_start,
1955 .next = gfs2_glock_seq_next,
1956 .stop = gfs2_glock_seq_stop,
1957 .show = gfs2_glstats_seq_show,
1960 static const struct seq_operations gfs2_sbstats_seq_ops = {
1961 .start = gfs2_sbstats_seq_start,
1962 .next = gfs2_sbstats_seq_next,
1963 .stop = gfs2_sbstats_seq_stop,
1964 .show = gfs2_sbstats_seq_show,
1967 static int gfs2_glocks_open(struct inode *inode, struct file *file)
1969 int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1970 sizeof(struct gfs2_glock_iter));
1971 if (ret == 0) {
1972 struct seq_file *seq = file->private_data;
1973 struct gfs2_glock_iter *gi = seq->private;
1974 gi->sdp = inode->i_private;
1976 return ret;
1979 static int gfs2_glstats_open(struct inode *inode, struct file *file)
1981 int ret = seq_open_private(file, &gfs2_glstats_seq_ops,
1982 sizeof(struct gfs2_glock_iter));
1983 if (ret == 0) {
1984 struct seq_file *seq = file->private_data;
1985 struct gfs2_glock_iter *gi = seq->private;
1986 gi->sdp = inode->i_private;
1988 return ret;
1991 static int gfs2_sbstats_open(struct inode *inode, struct file *file)
1993 int ret = seq_open_private(file, &gfs2_sbstats_seq_ops,
1994 sizeof(struct gfs2_glock_iter));
1995 if (ret == 0) {
1996 struct seq_file *seq = file->private_data;
1997 struct gfs2_glock_iter *gi = seq->private;
1998 gi->sdp = inode->i_private;
2000 return ret;
2003 static const struct file_operations gfs2_glocks_fops = {
2004 .owner = THIS_MODULE,
2005 .open = gfs2_glocks_open,
2006 .read = seq_read,
2007 .llseek = seq_lseek,
2008 .release = seq_release_private,
2011 static const struct file_operations gfs2_glstats_fops = {
2012 .owner = THIS_MODULE,
2013 .open = gfs2_glstats_open,
2014 .read = seq_read,
2015 .llseek = seq_lseek,
2016 .release = seq_release_private,
2019 static const struct file_operations gfs2_sbstats_fops = {
2020 .owner = THIS_MODULE,
2021 .open = gfs2_sbstats_open,
2022 .read = seq_read,
2023 .llseek = seq_lseek,
2024 .release = seq_release_private,
2027 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2029 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2030 if (!sdp->debugfs_dir)
2031 return -ENOMEM;
2032 sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
2033 S_IFREG | S_IRUGO,
2034 sdp->debugfs_dir, sdp,
2035 &gfs2_glocks_fops);
2036 if (!sdp->debugfs_dentry_glocks)
2037 goto fail;
2039 sdp->debugfs_dentry_glstats = debugfs_create_file("glstats",
2040 S_IFREG | S_IRUGO,
2041 sdp->debugfs_dir, sdp,
2042 &gfs2_glstats_fops);
2043 if (!sdp->debugfs_dentry_glstats)
2044 goto fail;
2046 sdp->debugfs_dentry_sbstats = debugfs_create_file("sbstats",
2047 S_IFREG | S_IRUGO,
2048 sdp->debugfs_dir, sdp,
2049 &gfs2_sbstats_fops);
2050 if (!sdp->debugfs_dentry_sbstats)
2051 goto fail;
2053 return 0;
2054 fail:
2055 gfs2_delete_debugfs_file(sdp);
2056 return -ENOMEM;
2059 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2061 if (sdp->debugfs_dir) {
2062 if (sdp->debugfs_dentry_glocks) {
2063 debugfs_remove(sdp->debugfs_dentry_glocks);
2064 sdp->debugfs_dentry_glocks = NULL;
2066 if (sdp->debugfs_dentry_glstats) {
2067 debugfs_remove(sdp->debugfs_dentry_glstats);
2068 sdp->debugfs_dentry_glstats = NULL;
2070 if (sdp->debugfs_dentry_sbstats) {
2071 debugfs_remove(sdp->debugfs_dentry_sbstats);
2072 sdp->debugfs_dentry_sbstats = NULL;
2074 debugfs_remove(sdp->debugfs_dir);
2075 sdp->debugfs_dir = NULL;
2079 int gfs2_register_debugfs(void)
2081 gfs2_root = debugfs_create_dir("gfs2", NULL);
2082 return gfs2_root ? 0 : -ENOMEM;
2085 void gfs2_unregister_debugfs(void)
2087 debugfs_remove(gfs2_root);
2088 gfs2_root = NULL;