mm-only debug patch...
[mmotm.git] / fs / gfs2 / glock.c
blob12390765b57ddcfe240fc560f6fa8e99918637ac
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 <linux/rwsem.h>
23 #include <asm/uaccess.h>
24 #include <linux/seq_file.h>
25 #include <linux/debugfs.h>
26 #include <linux/kthread.h>
27 #include <linux/freezer.h>
28 #include <linux/workqueue.h>
29 #include <linux/jiffies.h>
31 #include "gfs2.h"
32 #include "incore.h"
33 #include "glock.h"
34 #include "glops.h"
35 #include "inode.h"
36 #include "lops.h"
37 #include "meta_io.h"
38 #include "quota.h"
39 #include "super.h"
40 #include "util.h"
41 #include "bmap.h"
43 #define CREATE_TRACE_POINTS
44 #include <trace/events/gfs2.h>
46 struct gfs2_gl_hash_bucket {
47 struct hlist_head hb_list;
50 struct gfs2_glock_iter {
51 int hash; /* hash bucket index */
52 struct gfs2_sbd *sdp; /* incore superblock */
53 struct gfs2_glock *gl; /* current glock struct */
54 char string[512]; /* scratch space */
57 typedef void (*glock_examiner) (struct gfs2_glock * gl);
59 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp);
60 static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl);
61 #define GLOCK_BUG_ON(gl,x) do { if (unlikely(x)) { __dump_glock(NULL, gl); BUG(); } } while(0)
62 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
64 static DECLARE_RWSEM(gfs2_umount_flush_sem);
65 static struct dentry *gfs2_root;
66 static struct workqueue_struct *glock_workqueue;
67 struct workqueue_struct *gfs2_delete_workqueue;
68 static LIST_HEAD(lru_list);
69 static atomic_t lru_count = ATOMIC_INIT(0);
70 static DEFINE_SPINLOCK(lru_lock);
72 #define GFS2_GL_HASH_SHIFT 15
73 #define GFS2_GL_HASH_SIZE (1 << GFS2_GL_HASH_SHIFT)
74 #define GFS2_GL_HASH_MASK (GFS2_GL_HASH_SIZE - 1)
76 static struct gfs2_gl_hash_bucket gl_hash_table[GFS2_GL_HASH_SIZE];
77 static struct dentry *gfs2_root;
80 * Despite what you might think, the numbers below are not arbitrary :-)
81 * They are taken from the ipv4 routing hash code, which is well tested
82 * and thus should be nearly optimal. Later on we might tweek the numbers
83 * but for now this should be fine.
85 * The reason for putting the locks in a separate array from the list heads
86 * is that we can have fewer locks than list heads and save memory. We use
87 * the same hash function for both, but with a different hash mask.
89 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
90 defined(CONFIG_PROVE_LOCKING)
92 #ifdef CONFIG_LOCKDEP
93 # define GL_HASH_LOCK_SZ 256
94 #else
95 # if NR_CPUS >= 32
96 # define GL_HASH_LOCK_SZ 4096
97 # elif NR_CPUS >= 16
98 # define GL_HASH_LOCK_SZ 2048
99 # elif NR_CPUS >= 8
100 # define GL_HASH_LOCK_SZ 1024
101 # elif NR_CPUS >= 4
102 # define GL_HASH_LOCK_SZ 512
103 # else
104 # define GL_HASH_LOCK_SZ 256
105 # endif
106 #endif
108 /* We never want more locks than chains */
109 #if GFS2_GL_HASH_SIZE < GL_HASH_LOCK_SZ
110 # undef GL_HASH_LOCK_SZ
111 # define GL_HASH_LOCK_SZ GFS2_GL_HASH_SIZE
112 #endif
114 static rwlock_t gl_hash_locks[GL_HASH_LOCK_SZ];
116 static inline rwlock_t *gl_lock_addr(unsigned int x)
118 return &gl_hash_locks[x & (GL_HASH_LOCK_SZ-1)];
120 #else /* not SMP, so no spinlocks required */
121 static inline rwlock_t *gl_lock_addr(unsigned int x)
123 return NULL;
125 #endif
128 * gl_hash() - Turn glock number into hash bucket number
129 * @lock: The glock number
131 * Returns: The number of the corresponding hash bucket
134 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
135 const struct lm_lockname *name)
137 unsigned int h;
139 h = jhash(&name->ln_number, sizeof(u64), 0);
140 h = jhash(&name->ln_type, sizeof(unsigned int), h);
141 h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
142 h &= GFS2_GL_HASH_MASK;
144 return h;
148 * glock_free() - Perform a few checks and then release struct gfs2_glock
149 * @gl: The glock to release
151 * Also calls lock module to release its internal structure for this glock.
155 static void glock_free(struct gfs2_glock *gl)
157 struct gfs2_sbd *sdp = gl->gl_sbd;
158 struct inode *aspace = gl->gl_aspace;
160 if (aspace)
161 gfs2_aspace_put(aspace);
162 trace_gfs2_glock_put(gl);
163 sdp->sd_lockstruct.ls_ops->lm_put_lock(gfs2_glock_cachep, gl);
167 * gfs2_glock_hold() - increment reference count on glock
168 * @gl: The glock to hold
172 void gfs2_glock_hold(struct gfs2_glock *gl)
174 GLOCK_BUG_ON(gl, atomic_read(&gl->gl_ref) == 0);
175 atomic_inc(&gl->gl_ref);
179 * demote_ok - Check to see if it's ok to unlock a glock
180 * @gl: the glock
182 * Returns: 1 if it's ok
185 static int demote_ok(const struct gfs2_glock *gl)
187 const struct gfs2_glock_operations *glops = gl->gl_ops;
189 if (gl->gl_state == LM_ST_UNLOCKED)
190 return 0;
191 if (!list_empty(&gl->gl_holders))
192 return 0;
193 if (glops->go_demote_ok)
194 return glops->go_demote_ok(gl);
195 return 1;
199 * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
200 * @gl: the glock
204 static void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
206 int may_reclaim;
207 may_reclaim = (demote_ok(gl) &&
208 (atomic_read(&gl->gl_ref) == 1 ||
209 (gl->gl_name.ln_type == LM_TYPE_INODE &&
210 atomic_read(&gl->gl_ref) <= 2)));
211 spin_lock(&lru_lock);
212 if (list_empty(&gl->gl_lru) && may_reclaim) {
213 list_add_tail(&gl->gl_lru, &lru_list);
214 atomic_inc(&lru_count);
216 spin_unlock(&lru_lock);
220 * gfs2_glock_put_nolock() - Decrement reference count on glock
221 * @gl: The glock to put
223 * This function should only be used if the caller has its own reference
224 * to the glock, in addition to the one it is dropping.
227 void gfs2_glock_put_nolock(struct gfs2_glock *gl)
229 if (atomic_dec_and_test(&gl->gl_ref))
230 GLOCK_BUG_ON(gl, 1);
231 gfs2_glock_schedule_for_reclaim(gl);
235 * gfs2_glock_put() - Decrement reference count on glock
236 * @gl: The glock to put
240 int gfs2_glock_put(struct gfs2_glock *gl)
242 int rv = 0;
244 write_lock(gl_lock_addr(gl->gl_hash));
245 if (atomic_dec_and_test(&gl->gl_ref)) {
246 hlist_del(&gl->gl_list);
247 write_unlock(gl_lock_addr(gl->gl_hash));
248 spin_lock(&lru_lock);
249 if (!list_empty(&gl->gl_lru)) {
250 list_del_init(&gl->gl_lru);
251 atomic_dec(&lru_count);
253 spin_unlock(&lru_lock);
254 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
255 glock_free(gl);
256 rv = 1;
257 goto out;
259 spin_lock(&gl->gl_spin);
260 gfs2_glock_schedule_for_reclaim(gl);
261 spin_unlock(&gl->gl_spin);
262 write_unlock(gl_lock_addr(gl->gl_hash));
263 out:
264 return rv;
268 * search_bucket() - Find struct gfs2_glock by lock number
269 * @bucket: the bucket to search
270 * @name: The lock name
272 * Returns: NULL, or the struct gfs2_glock with the requested number
275 static struct gfs2_glock *search_bucket(unsigned int hash,
276 const struct gfs2_sbd *sdp,
277 const struct lm_lockname *name)
279 struct gfs2_glock *gl;
280 struct hlist_node *h;
282 hlist_for_each_entry(gl, h, &gl_hash_table[hash].hb_list, gl_list) {
283 if (!lm_name_equal(&gl->gl_name, name))
284 continue;
285 if (gl->gl_sbd != sdp)
286 continue;
288 atomic_inc(&gl->gl_ref);
290 return gl;
293 return NULL;
297 * may_grant - check if its ok to grant a new lock
298 * @gl: The glock
299 * @gh: The lock request which we wish to grant
301 * Returns: true if its ok to grant the lock
304 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
306 const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
307 if ((gh->gh_state == LM_ST_EXCLUSIVE ||
308 gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
309 return 0;
310 if (gl->gl_state == gh->gh_state)
311 return 1;
312 if (gh->gh_flags & GL_EXACT)
313 return 0;
314 if (gl->gl_state == LM_ST_EXCLUSIVE) {
315 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
316 return 1;
317 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
318 return 1;
320 if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
321 return 1;
322 return 0;
325 static void gfs2_holder_wake(struct gfs2_holder *gh)
327 clear_bit(HIF_WAIT, &gh->gh_iflags);
328 smp_mb__after_clear_bit();
329 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
333 * do_promote - promote as many requests as possible on the current queue
334 * @gl: The glock
336 * Returns: 1 if there is a blocked holder at the head of the list, or 2
337 * if a type specific operation is underway.
340 static int do_promote(struct gfs2_glock *gl)
341 __releases(&gl->gl_spin)
342 __acquires(&gl->gl_spin)
344 const struct gfs2_glock_operations *glops = gl->gl_ops;
345 struct gfs2_holder *gh, *tmp;
346 int ret;
348 restart:
349 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
350 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
351 continue;
352 if (may_grant(gl, gh)) {
353 if (gh->gh_list.prev == &gl->gl_holders &&
354 glops->go_lock) {
355 spin_unlock(&gl->gl_spin);
356 /* FIXME: eliminate this eventually */
357 ret = glops->go_lock(gh);
358 spin_lock(&gl->gl_spin);
359 if (ret) {
360 if (ret == 1)
361 return 2;
362 gh->gh_error = ret;
363 list_del_init(&gh->gh_list);
364 trace_gfs2_glock_queue(gh, 0);
365 gfs2_holder_wake(gh);
366 goto restart;
368 set_bit(HIF_HOLDER, &gh->gh_iflags);
369 trace_gfs2_promote(gh, 1);
370 gfs2_holder_wake(gh);
371 goto restart;
373 set_bit(HIF_HOLDER, &gh->gh_iflags);
374 trace_gfs2_promote(gh, 0);
375 gfs2_holder_wake(gh);
376 continue;
378 if (gh->gh_list.prev == &gl->gl_holders)
379 return 1;
380 break;
382 return 0;
386 * do_error - Something unexpected has happened during a lock request
390 static inline void do_error(struct gfs2_glock *gl, const int ret)
392 struct gfs2_holder *gh, *tmp;
394 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
395 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
396 continue;
397 if (ret & LM_OUT_ERROR)
398 gh->gh_error = -EIO;
399 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
400 gh->gh_error = GLR_TRYFAILED;
401 else
402 continue;
403 list_del_init(&gh->gh_list);
404 trace_gfs2_glock_queue(gh, 0);
405 gfs2_holder_wake(gh);
410 * find_first_waiter - find the first gh that's waiting for the glock
411 * @gl: the glock
414 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
416 struct gfs2_holder *gh;
418 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
419 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
420 return gh;
422 return NULL;
426 * state_change - record that the glock is now in a different state
427 * @gl: the glock
428 * @new_state the new state
432 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
434 int held1, held2;
436 held1 = (gl->gl_state != LM_ST_UNLOCKED);
437 held2 = (new_state != LM_ST_UNLOCKED);
439 if (held1 != held2) {
440 if (held2)
441 gfs2_glock_hold(gl);
442 else
443 gfs2_glock_put_nolock(gl);
446 gl->gl_state = new_state;
447 gl->gl_tchange = jiffies;
450 static void gfs2_demote_wake(struct gfs2_glock *gl)
452 gl->gl_demote_state = LM_ST_EXCLUSIVE;
453 clear_bit(GLF_DEMOTE, &gl->gl_flags);
454 smp_mb__after_clear_bit();
455 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
459 * finish_xmote - The DLM has replied to one of our lock requests
460 * @gl: The glock
461 * @ret: The status from the DLM
465 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
467 const struct gfs2_glock_operations *glops = gl->gl_ops;
468 struct gfs2_holder *gh;
469 unsigned state = ret & LM_OUT_ST_MASK;
470 int rv;
472 spin_lock(&gl->gl_spin);
473 trace_gfs2_glock_state_change(gl, state);
474 state_change(gl, state);
475 gh = find_first_waiter(gl);
477 /* Demote to UN request arrived during demote to SH or DF */
478 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
479 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
480 gl->gl_target = LM_ST_UNLOCKED;
482 /* Check for state != intended state */
483 if (unlikely(state != gl->gl_target)) {
484 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
485 /* move to back of queue and try next entry */
486 if (ret & LM_OUT_CANCELED) {
487 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
488 list_move_tail(&gh->gh_list, &gl->gl_holders);
489 gh = find_first_waiter(gl);
490 gl->gl_target = gh->gh_state;
491 goto retry;
493 /* Some error or failed "try lock" - report it */
494 if ((ret & LM_OUT_ERROR) ||
495 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
496 gl->gl_target = gl->gl_state;
497 do_error(gl, ret);
498 goto out;
501 switch(state) {
502 /* Unlocked due to conversion deadlock, try again */
503 case LM_ST_UNLOCKED:
504 retry:
505 do_xmote(gl, gh, gl->gl_target);
506 break;
507 /* Conversion fails, unlock and try again */
508 case LM_ST_SHARED:
509 case LM_ST_DEFERRED:
510 do_xmote(gl, gh, LM_ST_UNLOCKED);
511 break;
512 default: /* Everything else */
513 printk(KERN_ERR "GFS2: wanted %u got %u\n", gl->gl_target, state);
514 GLOCK_BUG_ON(gl, 1);
516 spin_unlock(&gl->gl_spin);
517 gfs2_glock_put(gl);
518 return;
521 /* Fast path - we got what we asked for */
522 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
523 gfs2_demote_wake(gl);
524 if (state != LM_ST_UNLOCKED) {
525 if (glops->go_xmote_bh) {
526 spin_unlock(&gl->gl_spin);
527 rv = glops->go_xmote_bh(gl, gh);
528 if (rv == -EAGAIN)
529 return;
530 spin_lock(&gl->gl_spin);
531 if (rv) {
532 do_error(gl, rv);
533 goto out;
536 rv = do_promote(gl);
537 if (rv == 2)
538 goto out_locked;
540 out:
541 clear_bit(GLF_LOCK, &gl->gl_flags);
542 out_locked:
543 spin_unlock(&gl->gl_spin);
544 gfs2_glock_put(gl);
547 static unsigned int gfs2_lm_lock(struct gfs2_sbd *sdp, void *lock,
548 unsigned int req_state,
549 unsigned int flags)
551 int ret = LM_OUT_ERROR;
553 if (!sdp->sd_lockstruct.ls_ops->lm_lock)
554 return req_state == LM_ST_UNLOCKED ? 0 : req_state;
556 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
557 ret = sdp->sd_lockstruct.ls_ops->lm_lock(lock,
558 req_state, flags);
559 return ret;
563 * do_xmote - Calls the DLM to change the state of a lock
564 * @gl: The lock state
565 * @gh: The holder (only for promotes)
566 * @target: The target lock state
570 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
571 __releases(&gl->gl_spin)
572 __acquires(&gl->gl_spin)
574 const struct gfs2_glock_operations *glops = gl->gl_ops;
575 struct gfs2_sbd *sdp = gl->gl_sbd;
576 unsigned int lck_flags = gh ? gh->gh_flags : 0;
577 int ret;
579 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
580 LM_FLAG_PRIORITY);
581 BUG_ON(gl->gl_state == target);
582 BUG_ON(gl->gl_state == gl->gl_target);
583 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
584 glops->go_inval) {
585 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
586 do_error(gl, 0); /* Fail queued try locks */
588 spin_unlock(&gl->gl_spin);
589 if (glops->go_xmote_th)
590 glops->go_xmote_th(gl);
591 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
592 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
593 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
595 gfs2_glock_hold(gl);
596 if (target != LM_ST_UNLOCKED && (gl->gl_state == LM_ST_SHARED ||
597 gl->gl_state == LM_ST_DEFERRED) &&
598 !(lck_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
599 lck_flags |= LM_FLAG_TRY_1CB;
600 ret = gfs2_lm_lock(sdp, gl, target, lck_flags);
602 if (!(ret & LM_OUT_ASYNC)) {
603 finish_xmote(gl, ret);
604 gfs2_glock_hold(gl);
605 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
606 gfs2_glock_put(gl);
607 } else {
608 GLOCK_BUG_ON(gl, ret != LM_OUT_ASYNC);
610 spin_lock(&gl->gl_spin);
614 * find_first_holder - find the first "holder" gh
615 * @gl: the glock
618 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
620 struct gfs2_holder *gh;
622 if (!list_empty(&gl->gl_holders)) {
623 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
624 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
625 return gh;
627 return NULL;
631 * run_queue - do all outstanding tasks related to a glock
632 * @gl: The glock in question
633 * @nonblock: True if we must not block in run_queue
637 static void run_queue(struct gfs2_glock *gl, const int nonblock)
638 __releases(&gl->gl_spin)
639 __acquires(&gl->gl_spin)
641 struct gfs2_holder *gh = NULL;
642 int ret;
644 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
645 return;
647 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
649 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
650 gl->gl_demote_state != gl->gl_state) {
651 if (find_first_holder(gl))
652 goto out_unlock;
653 if (nonblock)
654 goto out_sched;
655 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
656 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
657 gl->gl_target = gl->gl_demote_state;
658 } else {
659 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
660 gfs2_demote_wake(gl);
661 ret = do_promote(gl);
662 if (ret == 0)
663 goto out_unlock;
664 if (ret == 2)
665 goto out;
666 gh = find_first_waiter(gl);
667 gl->gl_target = gh->gh_state;
668 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
669 do_error(gl, 0); /* Fail queued try locks */
671 do_xmote(gl, gh, gl->gl_target);
672 out:
673 return;
675 out_sched:
676 clear_bit(GLF_LOCK, &gl->gl_flags);
677 smp_mb__after_clear_bit();
678 gfs2_glock_hold(gl);
679 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
680 gfs2_glock_put_nolock(gl);
681 return;
683 out_unlock:
684 clear_bit(GLF_LOCK, &gl->gl_flags);
685 smp_mb__after_clear_bit();
686 return;
689 static void delete_work_func(struct work_struct *work)
691 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
692 struct gfs2_sbd *sdp = gl->gl_sbd;
693 struct gfs2_inode *ip = NULL;
694 struct inode *inode;
695 u64 no_addr = 0;
697 spin_lock(&gl->gl_spin);
698 ip = (struct gfs2_inode *)gl->gl_object;
699 if (ip)
700 no_addr = ip->i_no_addr;
701 spin_unlock(&gl->gl_spin);
702 if (ip) {
703 inode = gfs2_ilookup(sdp->sd_vfs, no_addr);
704 if (inode) {
705 d_prune_aliases(inode);
706 iput(inode);
709 gfs2_glock_put(gl);
712 static void glock_work_func(struct work_struct *work)
714 unsigned long delay = 0;
715 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
717 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags))
718 finish_xmote(gl, gl->gl_reply);
719 down_read(&gfs2_umount_flush_sem);
720 spin_lock(&gl->gl_spin);
721 if (test_and_clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
722 gl->gl_state != LM_ST_UNLOCKED &&
723 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
724 unsigned long holdtime, now = jiffies;
725 holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
726 if (time_before(now, holdtime))
727 delay = holdtime - now;
728 set_bit(delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE, &gl->gl_flags);
730 run_queue(gl, 0);
731 spin_unlock(&gl->gl_spin);
732 up_read(&gfs2_umount_flush_sem);
733 if (!delay ||
734 queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
735 gfs2_glock_put(gl);
739 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
740 * @sdp: The GFS2 superblock
741 * @number: the lock number
742 * @glops: The glock_operations to use
743 * @create: If 0, don't create the glock if it doesn't exist
744 * @glp: the glock is returned here
746 * This does not lock a glock, just finds/creates structures for one.
748 * Returns: errno
751 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
752 const struct gfs2_glock_operations *glops, int create,
753 struct gfs2_glock **glp)
755 struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
756 struct gfs2_glock *gl, *tmp;
757 unsigned int hash = gl_hash(sdp, &name);
758 int error;
760 read_lock(gl_lock_addr(hash));
761 gl = search_bucket(hash, sdp, &name);
762 read_unlock(gl_lock_addr(hash));
764 *glp = gl;
765 if (gl)
766 return 0;
767 if (!create)
768 return -ENOENT;
770 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
771 if (!gl)
772 return -ENOMEM;
774 gl->gl_flags = 0;
775 gl->gl_name = name;
776 atomic_set(&gl->gl_ref, 1);
777 gl->gl_state = LM_ST_UNLOCKED;
778 gl->gl_target = LM_ST_UNLOCKED;
779 gl->gl_demote_state = LM_ST_EXCLUSIVE;
780 gl->gl_hash = hash;
781 gl->gl_ops = glops;
782 snprintf(gl->gl_strname, GDLM_STRNAME_BYTES, "%8x%16llx", name.ln_type, (unsigned long long)number);
783 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
784 gl->gl_lksb.sb_lvbptr = gl->gl_lvb;
785 gl->gl_tchange = jiffies;
786 gl->gl_object = NULL;
787 gl->gl_sbd = sdp;
788 gl->gl_aspace = NULL;
789 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
790 INIT_WORK(&gl->gl_delete, delete_work_func);
792 /* If this glock protects actual on-disk data or metadata blocks,
793 create a VFS inode to manage the pages/buffers holding them. */
794 if (glops == &gfs2_inode_glops || glops == &gfs2_rgrp_glops) {
795 gl->gl_aspace = gfs2_aspace_get(sdp);
796 if (!gl->gl_aspace) {
797 error = -ENOMEM;
798 goto fail;
802 write_lock(gl_lock_addr(hash));
803 tmp = search_bucket(hash, sdp, &name);
804 if (tmp) {
805 write_unlock(gl_lock_addr(hash));
806 glock_free(gl);
807 gl = tmp;
808 } else {
809 hlist_add_head(&gl->gl_list, &gl_hash_table[hash].hb_list);
810 write_unlock(gl_lock_addr(hash));
813 *glp = gl;
815 return 0;
817 fail:
818 kmem_cache_free(gfs2_glock_cachep, gl);
819 return error;
823 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
824 * @gl: the glock
825 * @state: the state we're requesting
826 * @flags: the modifier flags
827 * @gh: the holder structure
831 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
832 struct gfs2_holder *gh)
834 INIT_LIST_HEAD(&gh->gh_list);
835 gh->gh_gl = gl;
836 gh->gh_ip = (unsigned long)__builtin_return_address(0);
837 gh->gh_owner_pid = get_pid(task_pid(current));
838 gh->gh_state = state;
839 gh->gh_flags = flags;
840 gh->gh_error = 0;
841 gh->gh_iflags = 0;
842 gfs2_glock_hold(gl);
846 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
847 * @state: the state we're requesting
848 * @flags: the modifier flags
849 * @gh: the holder structure
851 * Don't mess with the glock.
855 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
857 gh->gh_state = state;
858 gh->gh_flags = flags;
859 gh->gh_iflags = 0;
860 gh->gh_ip = (unsigned long)__builtin_return_address(0);
864 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
865 * @gh: the holder structure
869 void gfs2_holder_uninit(struct gfs2_holder *gh)
871 put_pid(gh->gh_owner_pid);
872 gfs2_glock_put(gh->gh_gl);
873 gh->gh_gl = NULL;
874 gh->gh_ip = 0;
878 * gfs2_glock_holder_wait
879 * @word: unused
881 * This function and gfs2_glock_demote_wait both show up in the WCHAN
882 * field. Thus I've separated these otherwise identical functions in
883 * order to be more informative to the user.
886 static int gfs2_glock_holder_wait(void *word)
888 schedule();
889 return 0;
892 static int gfs2_glock_demote_wait(void *word)
894 schedule();
895 return 0;
898 static void wait_on_holder(struct gfs2_holder *gh)
900 might_sleep();
901 wait_on_bit(&gh->gh_iflags, HIF_WAIT, gfs2_glock_holder_wait, TASK_UNINTERRUPTIBLE);
904 static void wait_on_demote(struct gfs2_glock *gl)
906 might_sleep();
907 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, gfs2_glock_demote_wait, TASK_UNINTERRUPTIBLE);
911 * handle_callback - process a demote request
912 * @gl: the glock
913 * @state: the state the caller wants us to change to
915 * There are only two requests that we are going to see in actual
916 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
919 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
920 unsigned long delay)
922 int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
924 set_bit(bit, &gl->gl_flags);
925 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
926 gl->gl_demote_state = state;
927 gl->gl_demote_time = jiffies;
928 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
929 gl->gl_demote_state != state) {
930 gl->gl_demote_state = LM_ST_UNLOCKED;
932 if (gl->gl_ops->go_callback)
933 gl->gl_ops->go_callback(gl);
934 trace_gfs2_demote_rq(gl);
938 * gfs2_glock_wait - wait on a glock acquisition
939 * @gh: the glock holder
941 * Returns: 0 on success
944 int gfs2_glock_wait(struct gfs2_holder *gh)
946 wait_on_holder(gh);
947 return gh->gh_error;
950 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
952 va_list args;
954 va_start(args, fmt);
955 if (seq) {
956 struct gfs2_glock_iter *gi = seq->private;
957 vsprintf(gi->string, fmt, args);
958 seq_printf(seq, gi->string);
959 } else {
960 printk(KERN_ERR " ");
961 vprintk(fmt, args);
963 va_end(args);
967 * add_to_queue - Add a holder to the wait queue (but look for recursion)
968 * @gh: the holder structure to add
970 * Eventually we should move the recursive locking trap to a
971 * debugging option or something like that. This is the fast
972 * path and needs to have the minimum number of distractions.
976 static inline void add_to_queue(struct gfs2_holder *gh)
977 __releases(&gl->gl_spin)
978 __acquires(&gl->gl_spin)
980 struct gfs2_glock *gl = gh->gh_gl;
981 struct gfs2_sbd *sdp = gl->gl_sbd;
982 struct list_head *insert_pt = NULL;
983 struct gfs2_holder *gh2;
984 int try_lock = 0;
986 BUG_ON(gh->gh_owner_pid == NULL);
987 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
988 BUG();
990 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
991 if (test_bit(GLF_LOCK, &gl->gl_flags))
992 try_lock = 1;
993 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
994 goto fail;
997 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
998 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
999 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
1000 goto trap_recursive;
1001 if (try_lock &&
1002 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) &&
1003 !may_grant(gl, gh)) {
1004 fail:
1005 gh->gh_error = GLR_TRYFAILED;
1006 gfs2_holder_wake(gh);
1007 return;
1009 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1010 continue;
1011 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1012 insert_pt = &gh2->gh_list;
1014 if (likely(insert_pt == NULL)) {
1015 list_add_tail(&gh->gh_list, &gl->gl_holders);
1016 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1017 goto do_cancel;
1018 return;
1020 trace_gfs2_glock_queue(gh, 1);
1021 list_add_tail(&gh->gh_list, insert_pt);
1022 do_cancel:
1023 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
1024 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1025 spin_unlock(&gl->gl_spin);
1026 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1027 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1028 spin_lock(&gl->gl_spin);
1030 return;
1032 trap_recursive:
1033 print_symbol(KERN_ERR "original: %s\n", gh2->gh_ip);
1034 printk(KERN_ERR "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1035 printk(KERN_ERR "lock type: %d req lock state : %d\n",
1036 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1037 print_symbol(KERN_ERR "new: %s\n", gh->gh_ip);
1038 printk(KERN_ERR "pid: %d\n", pid_nr(gh->gh_owner_pid));
1039 printk(KERN_ERR "lock type: %d req lock state : %d\n",
1040 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1041 __dump_glock(NULL, gl);
1042 BUG();
1046 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1047 * @gh: the holder structure
1049 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1051 * Returns: 0, GLR_TRYFAILED, or errno on failure
1054 int gfs2_glock_nq(struct gfs2_holder *gh)
1056 struct gfs2_glock *gl = gh->gh_gl;
1057 struct gfs2_sbd *sdp = gl->gl_sbd;
1058 int error = 0;
1060 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1061 return -EIO;
1063 spin_lock(&gl->gl_spin);
1064 add_to_queue(gh);
1065 run_queue(gl, 1);
1066 spin_unlock(&gl->gl_spin);
1068 if (!(gh->gh_flags & GL_ASYNC))
1069 error = gfs2_glock_wait(gh);
1071 return error;
1075 * gfs2_glock_poll - poll to see if an async request has been completed
1076 * @gh: the holder
1078 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1081 int gfs2_glock_poll(struct gfs2_holder *gh)
1083 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1087 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1088 * @gh: the glock holder
1092 void gfs2_glock_dq(struct gfs2_holder *gh)
1094 struct gfs2_glock *gl = gh->gh_gl;
1095 const struct gfs2_glock_operations *glops = gl->gl_ops;
1096 unsigned delay = 0;
1097 int fast_path = 0;
1099 spin_lock(&gl->gl_spin);
1100 if (gh->gh_flags & GL_NOCACHE)
1101 handle_callback(gl, LM_ST_UNLOCKED, 0);
1103 list_del_init(&gh->gh_list);
1104 if (find_first_holder(gl) == NULL) {
1105 if (glops->go_unlock) {
1106 GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1107 spin_unlock(&gl->gl_spin);
1108 glops->go_unlock(gh);
1109 spin_lock(&gl->gl_spin);
1110 clear_bit(GLF_LOCK, &gl->gl_flags);
1112 if (list_empty(&gl->gl_holders) &&
1113 !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1114 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1115 fast_path = 1;
1117 trace_gfs2_glock_queue(gh, 0);
1118 spin_unlock(&gl->gl_spin);
1119 if (likely(fast_path))
1120 return;
1122 gfs2_glock_hold(gl);
1123 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1124 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1125 delay = gl->gl_ops->go_min_hold_time;
1126 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1127 gfs2_glock_put(gl);
1130 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1132 struct gfs2_glock *gl = gh->gh_gl;
1133 gfs2_glock_dq(gh);
1134 wait_on_demote(gl);
1138 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1139 * @gh: the holder structure
1143 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1145 gfs2_glock_dq(gh);
1146 gfs2_holder_uninit(gh);
1150 * gfs2_glock_nq_num - acquire a glock based on lock number
1151 * @sdp: the filesystem
1152 * @number: the lock number
1153 * @glops: the glock operations for the type of glock
1154 * @state: the state to acquire the glock in
1155 * @flags: modifier flags for the aquisition
1156 * @gh: the struct gfs2_holder
1158 * Returns: errno
1161 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1162 const struct gfs2_glock_operations *glops,
1163 unsigned int state, int flags, struct gfs2_holder *gh)
1165 struct gfs2_glock *gl;
1166 int error;
1168 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1169 if (!error) {
1170 error = gfs2_glock_nq_init(gl, state, flags, gh);
1171 gfs2_glock_put(gl);
1174 return error;
1178 * glock_compare - Compare two struct gfs2_glock structures for sorting
1179 * @arg_a: the first structure
1180 * @arg_b: the second structure
1184 static int glock_compare(const void *arg_a, const void *arg_b)
1186 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1187 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1188 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1189 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1191 if (a->ln_number > b->ln_number)
1192 return 1;
1193 if (a->ln_number < b->ln_number)
1194 return -1;
1195 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1196 return 0;
1200 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1201 * @num_gh: the number of structures
1202 * @ghs: an array of struct gfs2_holder structures
1204 * Returns: 0 on success (all glocks acquired),
1205 * errno on failure (no glocks acquired)
1208 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1209 struct gfs2_holder **p)
1211 unsigned int x;
1212 int error = 0;
1214 for (x = 0; x < num_gh; x++)
1215 p[x] = &ghs[x];
1217 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1219 for (x = 0; x < num_gh; x++) {
1220 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1222 error = gfs2_glock_nq(p[x]);
1223 if (error) {
1224 while (x--)
1225 gfs2_glock_dq(p[x]);
1226 break;
1230 return error;
1234 * gfs2_glock_nq_m - acquire multiple glocks
1235 * @num_gh: the number of structures
1236 * @ghs: an array of struct gfs2_holder structures
1239 * Returns: 0 on success (all glocks acquired),
1240 * errno on failure (no glocks acquired)
1243 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1245 struct gfs2_holder *tmp[4];
1246 struct gfs2_holder **pph = tmp;
1247 int error = 0;
1249 switch(num_gh) {
1250 case 0:
1251 return 0;
1252 case 1:
1253 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1254 return gfs2_glock_nq(ghs);
1255 default:
1256 if (num_gh <= 4)
1257 break;
1258 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1259 if (!pph)
1260 return -ENOMEM;
1263 error = nq_m_sync(num_gh, ghs, pph);
1265 if (pph != tmp)
1266 kfree(pph);
1268 return error;
1272 * gfs2_glock_dq_m - release multiple glocks
1273 * @num_gh: the number of structures
1274 * @ghs: an array of struct gfs2_holder structures
1278 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1280 unsigned int x;
1282 for (x = 0; x < num_gh; x++)
1283 gfs2_glock_dq(&ghs[x]);
1287 * gfs2_glock_dq_uninit_m - release multiple glocks
1288 * @num_gh: the number of structures
1289 * @ghs: an array of struct gfs2_holder structures
1293 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1295 unsigned int x;
1297 for (x = 0; x < num_gh; x++)
1298 gfs2_glock_dq_uninit(&ghs[x]);
1301 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1303 unsigned long delay = 0;
1304 unsigned long holdtime;
1305 unsigned long now = jiffies;
1307 gfs2_glock_hold(gl);
1308 holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
1309 if (time_before(now, holdtime))
1310 delay = holdtime - now;
1311 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1312 delay = gl->gl_ops->go_min_hold_time;
1314 spin_lock(&gl->gl_spin);
1315 handle_callback(gl, state, delay);
1316 spin_unlock(&gl->gl_spin);
1317 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1318 gfs2_glock_put(gl);
1322 * gfs2_glock_complete - Callback used by locking
1323 * @gl: Pointer to the glock
1324 * @ret: The return value from the dlm
1328 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1330 struct lm_lockstruct *ls = &gl->gl_sbd->sd_lockstruct;
1331 gl->gl_reply = ret;
1332 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_flags))) {
1333 struct gfs2_holder *gh;
1334 spin_lock(&gl->gl_spin);
1335 gh = find_first_waiter(gl);
1336 if ((!(gh && (gh->gh_flags & LM_FLAG_NOEXP)) &&
1337 (gl->gl_target != LM_ST_UNLOCKED)) ||
1338 ((ret & ~LM_OUT_ST_MASK) != 0))
1339 set_bit(GLF_FROZEN, &gl->gl_flags);
1340 spin_unlock(&gl->gl_spin);
1341 if (test_bit(GLF_FROZEN, &gl->gl_flags))
1342 return;
1344 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1345 gfs2_glock_hold(gl);
1346 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1347 gfs2_glock_put(gl);
1351 static int gfs2_shrink_glock_memory(int nr, gfp_t gfp_mask)
1353 struct gfs2_glock *gl;
1354 int may_demote;
1355 int nr_skipped = 0;
1356 LIST_HEAD(skipped);
1358 if (nr == 0)
1359 goto out;
1361 if (!(gfp_mask & __GFP_FS))
1362 return -1;
1364 spin_lock(&lru_lock);
1365 while(nr && !list_empty(&lru_list)) {
1366 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1367 list_del_init(&gl->gl_lru);
1368 atomic_dec(&lru_count);
1370 /* Check if glock is about to be freed */
1371 if (atomic_read(&gl->gl_ref) == 0)
1372 continue;
1374 /* Test for being demotable */
1375 if (!test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1376 gfs2_glock_hold(gl);
1377 spin_unlock(&lru_lock);
1378 spin_lock(&gl->gl_spin);
1379 may_demote = demote_ok(gl);
1380 if (may_demote) {
1381 handle_callback(gl, LM_ST_UNLOCKED, 0);
1382 nr--;
1384 clear_bit(GLF_LOCK, &gl->gl_flags);
1385 smp_mb__after_clear_bit();
1386 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1387 gfs2_glock_put_nolock(gl);
1388 spin_unlock(&gl->gl_spin);
1389 spin_lock(&lru_lock);
1390 continue;
1392 nr_skipped++;
1393 list_add(&gl->gl_lru, &skipped);
1395 list_splice(&skipped, &lru_list);
1396 atomic_add(nr_skipped, &lru_count);
1397 spin_unlock(&lru_lock);
1398 out:
1399 return (atomic_read(&lru_count) / 100) * sysctl_vfs_cache_pressure;
1402 static struct shrinker glock_shrinker = {
1403 .shrink = gfs2_shrink_glock_memory,
1404 .seeks = DEFAULT_SEEKS,
1408 * examine_bucket - Call a function for glock in a hash bucket
1409 * @examiner: the function
1410 * @sdp: the filesystem
1411 * @bucket: the bucket
1413 * Returns: 1 if the bucket has entries
1416 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1417 unsigned int hash)
1419 struct gfs2_glock *gl, *prev = NULL;
1420 int has_entries = 0;
1421 struct hlist_head *head = &gl_hash_table[hash].hb_list;
1423 read_lock(gl_lock_addr(hash));
1424 /* Can't use hlist_for_each_entry - don't want prefetch here */
1425 if (hlist_empty(head))
1426 goto out;
1427 gl = list_entry(head->first, struct gfs2_glock, gl_list);
1428 while(1) {
1429 if (!sdp || gl->gl_sbd == sdp) {
1430 gfs2_glock_hold(gl);
1431 read_unlock(gl_lock_addr(hash));
1432 if (prev)
1433 gfs2_glock_put(prev);
1434 prev = gl;
1435 examiner(gl);
1436 has_entries = 1;
1437 read_lock(gl_lock_addr(hash));
1439 if (gl->gl_list.next == NULL)
1440 break;
1441 gl = list_entry(gl->gl_list.next, struct gfs2_glock, gl_list);
1443 out:
1444 read_unlock(gl_lock_addr(hash));
1445 if (prev)
1446 gfs2_glock_put(prev);
1447 cond_resched();
1448 return has_entries;
1453 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1454 * @gl: The glock to thaw
1456 * N.B. When we freeze a glock, we leave a ref to the glock outstanding,
1457 * so this has to result in the ref count being dropped by one.
1460 static void thaw_glock(struct gfs2_glock *gl)
1462 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1463 return;
1464 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1465 gfs2_glock_hold(gl);
1466 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1467 gfs2_glock_put(gl);
1471 * clear_glock - look at a glock and see if we can free it from glock cache
1472 * @gl: the glock to look at
1476 static void clear_glock(struct gfs2_glock *gl)
1478 spin_lock(&lru_lock);
1479 if (!list_empty(&gl->gl_lru)) {
1480 list_del_init(&gl->gl_lru);
1481 atomic_dec(&lru_count);
1483 spin_unlock(&lru_lock);
1485 spin_lock(&gl->gl_spin);
1486 if (find_first_holder(gl) == NULL && gl->gl_state != LM_ST_UNLOCKED)
1487 handle_callback(gl, LM_ST_UNLOCKED, 0);
1488 spin_unlock(&gl->gl_spin);
1489 gfs2_glock_hold(gl);
1490 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1491 gfs2_glock_put(gl);
1495 * gfs2_glock_thaw - Thaw any frozen glocks
1496 * @sdp: The super block
1500 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1502 unsigned x;
1504 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1505 examine_bucket(thaw_glock, sdp, x);
1509 * gfs2_gl_hash_clear - Empty out the glock hash table
1510 * @sdp: the filesystem
1511 * @wait: wait until it's all gone
1513 * Called when unmounting the filesystem.
1516 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1518 unsigned long t;
1519 unsigned int x;
1520 int cont;
1522 t = jiffies;
1524 for (;;) {
1525 cont = 0;
1526 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1527 if (examine_bucket(clear_glock, sdp, x))
1528 cont = 1;
1531 if (!cont)
1532 break;
1534 if (time_after_eq(jiffies,
1535 t + gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
1536 fs_warn(sdp, "Unmount seems to be stalled. "
1537 "Dumping lock state...\n");
1538 gfs2_dump_lockstate(sdp);
1539 t = jiffies;
1542 down_write(&gfs2_umount_flush_sem);
1543 invalidate_inodes(sdp->sd_vfs);
1544 up_write(&gfs2_umount_flush_sem);
1545 msleep(10);
1549 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1551 struct gfs2_glock *gl = ip->i_gl;
1552 int ret;
1554 ret = gfs2_truncatei_resume(ip);
1555 gfs2_assert_withdraw(gl->gl_sbd, ret == 0);
1557 spin_lock(&gl->gl_spin);
1558 clear_bit(GLF_LOCK, &gl->gl_flags);
1559 run_queue(gl, 1);
1560 spin_unlock(&gl->gl_spin);
1563 static const char *state2str(unsigned state)
1565 switch(state) {
1566 case LM_ST_UNLOCKED:
1567 return "UN";
1568 case LM_ST_SHARED:
1569 return "SH";
1570 case LM_ST_DEFERRED:
1571 return "DF";
1572 case LM_ST_EXCLUSIVE:
1573 return "EX";
1575 return "??";
1578 static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1580 char *p = buf;
1581 if (flags & LM_FLAG_TRY)
1582 *p++ = 't';
1583 if (flags & LM_FLAG_TRY_1CB)
1584 *p++ = 'T';
1585 if (flags & LM_FLAG_NOEXP)
1586 *p++ = 'e';
1587 if (flags & LM_FLAG_ANY)
1588 *p++ = 'A';
1589 if (flags & LM_FLAG_PRIORITY)
1590 *p++ = 'p';
1591 if (flags & GL_ASYNC)
1592 *p++ = 'a';
1593 if (flags & GL_EXACT)
1594 *p++ = 'E';
1595 if (flags & GL_NOCACHE)
1596 *p++ = 'c';
1597 if (test_bit(HIF_HOLDER, &iflags))
1598 *p++ = 'H';
1599 if (test_bit(HIF_WAIT, &iflags))
1600 *p++ = 'W';
1601 if (test_bit(HIF_FIRST, &iflags))
1602 *p++ = 'F';
1603 *p = 0;
1604 return buf;
1608 * dump_holder - print information about a glock holder
1609 * @seq: the seq_file struct
1610 * @gh: the glock holder
1612 * Returns: 0 on success, -ENOBUFS when we run out of space
1615 static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1617 struct task_struct *gh_owner = NULL;
1618 char buffer[KSYM_SYMBOL_LEN];
1619 char flags_buf[32];
1621 sprint_symbol(buffer, gh->gh_ip);
1622 if (gh->gh_owner_pid)
1623 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1624 gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %s\n",
1625 state2str(gh->gh_state),
1626 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1627 gh->gh_error,
1628 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1629 gh_owner ? gh_owner->comm : "(ended)", buffer);
1630 return 0;
1633 static const char *gflags2str(char *buf, const unsigned long *gflags)
1635 char *p = buf;
1636 if (test_bit(GLF_LOCK, gflags))
1637 *p++ = 'l';
1638 if (test_bit(GLF_DEMOTE, gflags))
1639 *p++ = 'D';
1640 if (test_bit(GLF_PENDING_DEMOTE, gflags))
1641 *p++ = 'd';
1642 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1643 *p++ = 'p';
1644 if (test_bit(GLF_DIRTY, gflags))
1645 *p++ = 'y';
1646 if (test_bit(GLF_LFLUSH, gflags))
1647 *p++ = 'f';
1648 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1649 *p++ = 'i';
1650 if (test_bit(GLF_REPLY_PENDING, gflags))
1651 *p++ = 'r';
1652 if (test_bit(GLF_INITIAL, gflags))
1653 *p++ = 'I';
1654 if (test_bit(GLF_FROZEN, gflags))
1655 *p++ = 'F';
1656 *p = 0;
1657 return buf;
1661 * __dump_glock - print information about a glock
1662 * @seq: The seq_file struct
1663 * @gl: the glock
1665 * The file format is as follows:
1666 * One line per object, capital letters are used to indicate objects
1667 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1668 * other objects are indented by a single space and follow the glock to
1669 * which they are related. Fields are indicated by lower case letters
1670 * followed by a colon and the field value, except for strings which are in
1671 * [] so that its possible to see if they are composed of spaces for
1672 * example. The field's are n = number (id of the object), f = flags,
1673 * t = type, s = state, r = refcount, e = error, p = pid.
1675 * Returns: 0 on success, -ENOBUFS when we run out of space
1678 static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1680 const struct gfs2_glock_operations *glops = gl->gl_ops;
1681 unsigned long long dtime;
1682 const struct gfs2_holder *gh;
1683 char gflags_buf[32];
1684 int error = 0;
1686 dtime = jiffies - gl->gl_demote_time;
1687 dtime *= 1000000/HZ; /* demote time in uSec */
1688 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1689 dtime = 0;
1690 gfs2_print_dbg(seq, "G: s:%s n:%u/%llu f:%s t:%s d:%s/%llu a:%d r:%d\n",
1691 state2str(gl->gl_state),
1692 gl->gl_name.ln_type,
1693 (unsigned long long)gl->gl_name.ln_number,
1694 gflags2str(gflags_buf, &gl->gl_flags),
1695 state2str(gl->gl_target),
1696 state2str(gl->gl_demote_state), dtime,
1697 atomic_read(&gl->gl_ail_count),
1698 atomic_read(&gl->gl_ref));
1700 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1701 error = dump_holder(seq, gh);
1702 if (error)
1703 goto out;
1705 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1706 error = glops->go_dump(seq, gl);
1707 out:
1708 return error;
1711 static int dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1713 int ret;
1714 spin_lock(&gl->gl_spin);
1715 ret = __dump_glock(seq, gl);
1716 spin_unlock(&gl->gl_spin);
1717 return ret;
1721 * gfs2_dump_lockstate - print out the current lockstate
1722 * @sdp: the filesystem
1723 * @ub: the buffer to copy the information into
1725 * If @ub is NULL, dump the lockstate to the console.
1729 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
1731 struct gfs2_glock *gl;
1732 struct hlist_node *h;
1733 unsigned int x;
1734 int error = 0;
1736 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1738 read_lock(gl_lock_addr(x));
1740 hlist_for_each_entry(gl, h, &gl_hash_table[x].hb_list, gl_list) {
1741 if (gl->gl_sbd != sdp)
1742 continue;
1744 error = dump_glock(NULL, gl);
1745 if (error)
1746 break;
1749 read_unlock(gl_lock_addr(x));
1751 if (error)
1752 break;
1756 return error;
1760 int __init gfs2_glock_init(void)
1762 unsigned i;
1763 for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
1764 INIT_HLIST_HEAD(&gl_hash_table[i].hb_list);
1766 #ifdef GL_HASH_LOCK_SZ
1767 for(i = 0; i < GL_HASH_LOCK_SZ; i++) {
1768 rwlock_init(&gl_hash_locks[i]);
1770 #endif
1772 glock_workqueue = create_workqueue("glock_workqueue");
1773 if (IS_ERR(glock_workqueue))
1774 return PTR_ERR(glock_workqueue);
1775 gfs2_delete_workqueue = create_workqueue("delete_workqueue");
1776 if (IS_ERR(gfs2_delete_workqueue)) {
1777 destroy_workqueue(glock_workqueue);
1778 return PTR_ERR(gfs2_delete_workqueue);
1781 register_shrinker(&glock_shrinker);
1783 return 0;
1786 void gfs2_glock_exit(void)
1788 unregister_shrinker(&glock_shrinker);
1789 destroy_workqueue(glock_workqueue);
1790 destroy_workqueue(gfs2_delete_workqueue);
1793 static int gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1795 struct gfs2_glock *gl;
1797 restart:
1798 read_lock(gl_lock_addr(gi->hash));
1799 gl = gi->gl;
1800 if (gl) {
1801 gi->gl = hlist_entry(gl->gl_list.next,
1802 struct gfs2_glock, gl_list);
1803 } else {
1804 gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
1805 struct gfs2_glock, gl_list);
1807 if (gi->gl)
1808 gfs2_glock_hold(gi->gl);
1809 read_unlock(gl_lock_addr(gi->hash));
1810 if (gl)
1811 gfs2_glock_put(gl);
1812 while (gi->gl == NULL) {
1813 gi->hash++;
1814 if (gi->hash >= GFS2_GL_HASH_SIZE)
1815 return 1;
1816 read_lock(gl_lock_addr(gi->hash));
1817 gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
1818 struct gfs2_glock, gl_list);
1819 if (gi->gl)
1820 gfs2_glock_hold(gi->gl);
1821 read_unlock(gl_lock_addr(gi->hash));
1824 if (gi->sdp != gi->gl->gl_sbd)
1825 goto restart;
1827 return 0;
1830 static void gfs2_glock_iter_free(struct gfs2_glock_iter *gi)
1832 if (gi->gl)
1833 gfs2_glock_put(gi->gl);
1834 gi->gl = NULL;
1837 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1839 struct gfs2_glock_iter *gi = seq->private;
1840 loff_t n = *pos;
1842 gi->hash = 0;
1844 do {
1845 if (gfs2_glock_iter_next(gi)) {
1846 gfs2_glock_iter_free(gi);
1847 return NULL;
1849 } while (n--);
1851 return gi->gl;
1854 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1855 loff_t *pos)
1857 struct gfs2_glock_iter *gi = seq->private;
1859 (*pos)++;
1861 if (gfs2_glock_iter_next(gi)) {
1862 gfs2_glock_iter_free(gi);
1863 return NULL;
1866 return gi->gl;
1869 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1871 struct gfs2_glock_iter *gi = seq->private;
1872 gfs2_glock_iter_free(gi);
1875 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1877 return dump_glock(seq, iter_ptr);
1880 static const struct seq_operations gfs2_glock_seq_ops = {
1881 .start = gfs2_glock_seq_start,
1882 .next = gfs2_glock_seq_next,
1883 .stop = gfs2_glock_seq_stop,
1884 .show = gfs2_glock_seq_show,
1887 static int gfs2_debugfs_open(struct inode *inode, struct file *file)
1889 int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1890 sizeof(struct gfs2_glock_iter));
1891 if (ret == 0) {
1892 struct seq_file *seq = file->private_data;
1893 struct gfs2_glock_iter *gi = seq->private;
1894 gi->sdp = inode->i_private;
1896 return ret;
1899 static const struct file_operations gfs2_debug_fops = {
1900 .owner = THIS_MODULE,
1901 .open = gfs2_debugfs_open,
1902 .read = seq_read,
1903 .llseek = seq_lseek,
1904 .release = seq_release_private,
1907 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
1909 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
1910 if (!sdp->debugfs_dir)
1911 return -ENOMEM;
1912 sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
1913 S_IFREG | S_IRUGO,
1914 sdp->debugfs_dir, sdp,
1915 &gfs2_debug_fops);
1916 if (!sdp->debugfs_dentry_glocks)
1917 return -ENOMEM;
1919 return 0;
1922 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
1924 if (sdp && sdp->debugfs_dir) {
1925 if (sdp->debugfs_dentry_glocks) {
1926 debugfs_remove(sdp->debugfs_dentry_glocks);
1927 sdp->debugfs_dentry_glocks = NULL;
1929 debugfs_remove(sdp->debugfs_dir);
1930 sdp->debugfs_dir = NULL;
1934 int gfs2_register_debugfs(void)
1936 gfs2_root = debugfs_create_dir("gfs2", NULL);
1937 return gfs2_root ? 0 : -ENOMEM;
1940 void gfs2_unregister_debugfs(void)
1942 debugfs_remove(gfs2_root);
1943 gfs2_root = NULL;