of: MSI: Simplify irqdomain lookup
[linux/fpc-iii.git] / fs / gfs2 / glock.c
blob32e74710b1aae4a68e24b2b417622fe417f76536
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 #include <linux/sched.h>
13 #include <linux/slab.h>
14 #include <linux/spinlock.h>
15 #include <linux/buffer_head.h>
16 #include <linux/delay.h>
17 #include <linux/sort.h>
18 #include <linux/jhash.h>
19 #include <linux/kallsyms.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <linux/list.h>
22 #include <linux/wait.h>
23 #include <linux/module.h>
24 #include <asm/uaccess.h>
25 #include <linux/seq_file.h>
26 #include <linux/debugfs.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/workqueue.h>
30 #include <linux/jiffies.h>
31 #include <linux/rcupdate.h>
32 #include <linux/rculist_bl.h>
33 #include <linux/bit_spinlock.h>
34 #include <linux/percpu.h>
35 #include <linux/list_sort.h>
36 #include <linux/lockref.h>
37 #include <linux/rhashtable.h>
39 #include "gfs2.h"
40 #include "incore.h"
41 #include "glock.h"
42 #include "glops.h"
43 #include "inode.h"
44 #include "lops.h"
45 #include "meta_io.h"
46 #include "quota.h"
47 #include "super.h"
48 #include "util.h"
49 #include "bmap.h"
50 #define CREATE_TRACE_POINTS
51 #include "trace_gfs2.h"
53 struct gfs2_glock_iter {
54 struct gfs2_sbd *sdp; /* incore superblock */
55 struct rhashtable_iter hti; /* rhashtable iterator */
56 struct gfs2_glock *gl; /* current glock struct */
57 loff_t last_pos; /* last position */
60 typedef void (*glock_examiner) (struct gfs2_glock * gl);
62 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
64 static struct dentry *gfs2_root;
65 static struct workqueue_struct *glock_workqueue;
66 struct workqueue_struct *gfs2_delete_workqueue;
67 static LIST_HEAD(lru_list);
68 static atomic_t lru_count = ATOMIC_INIT(0);
69 static DEFINE_SPINLOCK(lru_lock);
71 #define GFS2_GL_HASH_SHIFT 15
72 #define GFS2_GL_HASH_SIZE (1 << GFS2_GL_HASH_SHIFT)
74 static struct rhashtable_params ht_parms = {
75 .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4,
76 .key_len = sizeof(struct lm_lockname),
77 .key_offset = offsetof(struct gfs2_glock, gl_name),
78 .head_offset = offsetof(struct gfs2_glock, gl_node),
81 static struct rhashtable gl_hash_table;
83 void gfs2_glock_free(struct gfs2_glock *gl)
85 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
87 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
88 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
89 } else {
90 kfree(gl->gl_lksb.sb_lvbptr);
91 kmem_cache_free(gfs2_glock_cachep, gl);
93 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
94 wake_up(&sdp->sd_glock_wait);
97 /**
98 * gfs2_glock_hold() - increment reference count on glock
99 * @gl: The glock to hold
103 static void gfs2_glock_hold(struct gfs2_glock *gl)
105 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
106 lockref_get(&gl->gl_lockref);
110 * demote_ok - Check to see if it's ok to unlock a glock
111 * @gl: the glock
113 * Returns: 1 if it's ok
116 static int demote_ok(const struct gfs2_glock *gl)
118 const struct gfs2_glock_operations *glops = gl->gl_ops;
120 if (gl->gl_state == LM_ST_UNLOCKED)
121 return 0;
122 if (!list_empty(&gl->gl_holders))
123 return 0;
124 if (glops->go_demote_ok)
125 return glops->go_demote_ok(gl);
126 return 1;
130 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
132 spin_lock(&lru_lock);
134 if (!list_empty(&gl->gl_lru))
135 list_del_init(&gl->gl_lru);
136 else
137 atomic_inc(&lru_count);
139 list_add_tail(&gl->gl_lru, &lru_list);
140 set_bit(GLF_LRU, &gl->gl_flags);
141 spin_unlock(&lru_lock);
144 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
146 spin_lock(&lru_lock);
147 if (!list_empty(&gl->gl_lru)) {
148 list_del_init(&gl->gl_lru);
149 atomic_dec(&lru_count);
150 clear_bit(GLF_LRU, &gl->gl_flags);
152 spin_unlock(&lru_lock);
156 * gfs2_glock_put() - Decrement reference count on glock
157 * @gl: The glock to put
161 void gfs2_glock_put(struct gfs2_glock *gl)
163 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
164 struct address_space *mapping = gfs2_glock2aspace(gl);
166 if (lockref_put_or_lock(&gl->gl_lockref))
167 return;
169 lockref_mark_dead(&gl->gl_lockref);
171 gfs2_glock_remove_from_lru(gl);
172 spin_unlock(&gl->gl_lockref.lock);
173 rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
174 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
175 GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
176 trace_gfs2_glock_put(gl);
177 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
181 * may_grant - check if its ok to grant a new lock
182 * @gl: The glock
183 * @gh: The lock request which we wish to grant
185 * Returns: true if its ok to grant the lock
188 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
190 const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
191 if ((gh->gh_state == LM_ST_EXCLUSIVE ||
192 gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
193 return 0;
194 if (gl->gl_state == gh->gh_state)
195 return 1;
196 if (gh->gh_flags & GL_EXACT)
197 return 0;
198 if (gl->gl_state == LM_ST_EXCLUSIVE) {
199 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
200 return 1;
201 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
202 return 1;
204 if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
205 return 1;
206 return 0;
209 static void gfs2_holder_wake(struct gfs2_holder *gh)
211 clear_bit(HIF_WAIT, &gh->gh_iflags);
212 smp_mb__after_atomic();
213 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
217 * do_error - Something unexpected has happened during a lock request
221 static inline void do_error(struct gfs2_glock *gl, const int ret)
223 struct gfs2_holder *gh, *tmp;
225 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
226 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
227 continue;
228 if (ret & LM_OUT_ERROR)
229 gh->gh_error = -EIO;
230 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
231 gh->gh_error = GLR_TRYFAILED;
232 else
233 continue;
234 list_del_init(&gh->gh_list);
235 trace_gfs2_glock_queue(gh, 0);
236 gfs2_holder_wake(gh);
241 * do_promote - promote as many requests as possible on the current queue
242 * @gl: The glock
244 * Returns: 1 if there is a blocked holder at the head of the list, or 2
245 * if a type specific operation is underway.
248 static int do_promote(struct gfs2_glock *gl)
249 __releases(&gl->gl_lockref.lock)
250 __acquires(&gl->gl_lockref.lock)
252 const struct gfs2_glock_operations *glops = gl->gl_ops;
253 struct gfs2_holder *gh, *tmp;
254 int ret;
256 restart:
257 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
258 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
259 continue;
260 if (may_grant(gl, gh)) {
261 if (gh->gh_list.prev == &gl->gl_holders &&
262 glops->go_lock) {
263 spin_unlock(&gl->gl_lockref.lock);
264 /* FIXME: eliminate this eventually */
265 ret = glops->go_lock(gh);
266 spin_lock(&gl->gl_lockref.lock);
267 if (ret) {
268 if (ret == 1)
269 return 2;
270 gh->gh_error = ret;
271 list_del_init(&gh->gh_list);
272 trace_gfs2_glock_queue(gh, 0);
273 gfs2_holder_wake(gh);
274 goto restart;
276 set_bit(HIF_HOLDER, &gh->gh_iflags);
277 trace_gfs2_promote(gh, 1);
278 gfs2_holder_wake(gh);
279 goto restart;
281 set_bit(HIF_HOLDER, &gh->gh_iflags);
282 trace_gfs2_promote(gh, 0);
283 gfs2_holder_wake(gh);
284 continue;
286 if (gh->gh_list.prev == &gl->gl_holders)
287 return 1;
288 do_error(gl, 0);
289 break;
291 return 0;
295 * find_first_waiter - find the first gh that's waiting for the glock
296 * @gl: the glock
299 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
301 struct gfs2_holder *gh;
303 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
304 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
305 return gh;
307 return NULL;
311 * state_change - record that the glock is now in a different state
312 * @gl: the glock
313 * @new_state the new state
317 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
319 int held1, held2;
321 held1 = (gl->gl_state != LM_ST_UNLOCKED);
322 held2 = (new_state != LM_ST_UNLOCKED);
324 if (held1 != held2) {
325 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
326 if (held2)
327 gl->gl_lockref.count++;
328 else
329 gl->gl_lockref.count--;
331 if (held1 && held2 && list_empty(&gl->gl_holders))
332 clear_bit(GLF_QUEUED, &gl->gl_flags);
334 if (new_state != gl->gl_target)
335 /* shorten our minimum hold time */
336 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
337 GL_GLOCK_MIN_HOLD);
338 gl->gl_state = new_state;
339 gl->gl_tchange = jiffies;
342 static void gfs2_demote_wake(struct gfs2_glock *gl)
344 gl->gl_demote_state = LM_ST_EXCLUSIVE;
345 clear_bit(GLF_DEMOTE, &gl->gl_flags);
346 smp_mb__after_atomic();
347 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
351 * finish_xmote - The DLM has replied to one of our lock requests
352 * @gl: The glock
353 * @ret: The status from the DLM
357 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
359 const struct gfs2_glock_operations *glops = gl->gl_ops;
360 struct gfs2_holder *gh;
361 unsigned state = ret & LM_OUT_ST_MASK;
362 int rv;
364 spin_lock(&gl->gl_lockref.lock);
365 trace_gfs2_glock_state_change(gl, state);
366 state_change(gl, state);
367 gh = find_first_waiter(gl);
369 /* Demote to UN request arrived during demote to SH or DF */
370 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
371 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
372 gl->gl_target = LM_ST_UNLOCKED;
374 /* Check for state != intended state */
375 if (unlikely(state != gl->gl_target)) {
376 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
377 /* move to back of queue and try next entry */
378 if (ret & LM_OUT_CANCELED) {
379 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
380 list_move_tail(&gh->gh_list, &gl->gl_holders);
381 gh = find_first_waiter(gl);
382 gl->gl_target = gh->gh_state;
383 goto retry;
385 /* Some error or failed "try lock" - report it */
386 if ((ret & LM_OUT_ERROR) ||
387 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
388 gl->gl_target = gl->gl_state;
389 do_error(gl, ret);
390 goto out;
393 switch(state) {
394 /* Unlocked due to conversion deadlock, try again */
395 case LM_ST_UNLOCKED:
396 retry:
397 do_xmote(gl, gh, gl->gl_target);
398 break;
399 /* Conversion fails, unlock and try again */
400 case LM_ST_SHARED:
401 case LM_ST_DEFERRED:
402 do_xmote(gl, gh, LM_ST_UNLOCKED);
403 break;
404 default: /* Everything else */
405 pr_err("wanted %u got %u\n", gl->gl_target, state);
406 GLOCK_BUG_ON(gl, 1);
408 spin_unlock(&gl->gl_lockref.lock);
409 return;
412 /* Fast path - we got what we asked for */
413 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
414 gfs2_demote_wake(gl);
415 if (state != LM_ST_UNLOCKED) {
416 if (glops->go_xmote_bh) {
417 spin_unlock(&gl->gl_lockref.lock);
418 rv = glops->go_xmote_bh(gl, gh);
419 spin_lock(&gl->gl_lockref.lock);
420 if (rv) {
421 do_error(gl, rv);
422 goto out;
425 rv = do_promote(gl);
426 if (rv == 2)
427 goto out_locked;
429 out:
430 clear_bit(GLF_LOCK, &gl->gl_flags);
431 out_locked:
432 spin_unlock(&gl->gl_lockref.lock);
436 * do_xmote - Calls the DLM to change the state of a lock
437 * @gl: The lock state
438 * @gh: The holder (only for promotes)
439 * @target: The target lock state
443 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
444 __releases(&gl->gl_lockref.lock)
445 __acquires(&gl->gl_lockref.lock)
447 const struct gfs2_glock_operations *glops = gl->gl_ops;
448 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
449 unsigned int lck_flags = gh ? gh->gh_flags : 0;
450 int ret;
452 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
453 LM_FLAG_PRIORITY);
454 GLOCK_BUG_ON(gl, gl->gl_state == target);
455 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
456 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
457 glops->go_inval) {
458 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
459 do_error(gl, 0); /* Fail queued try locks */
461 gl->gl_req = target;
462 set_bit(GLF_BLOCKING, &gl->gl_flags);
463 if ((gl->gl_req == LM_ST_UNLOCKED) ||
464 (gl->gl_state == LM_ST_EXCLUSIVE) ||
465 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
466 clear_bit(GLF_BLOCKING, &gl->gl_flags);
467 spin_unlock(&gl->gl_lockref.lock);
468 if (glops->go_sync)
469 glops->go_sync(gl);
470 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
471 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
472 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
474 gfs2_glock_hold(gl);
475 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
476 /* lock_dlm */
477 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
478 if (ret) {
479 pr_err("lm_lock ret %d\n", ret);
480 GLOCK_BUG_ON(gl, 1);
482 } else { /* lock_nolock */
483 finish_xmote(gl, target);
484 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
485 gfs2_glock_put(gl);
488 spin_lock(&gl->gl_lockref.lock);
492 * find_first_holder - find the first "holder" gh
493 * @gl: the glock
496 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
498 struct gfs2_holder *gh;
500 if (!list_empty(&gl->gl_holders)) {
501 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
502 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
503 return gh;
505 return NULL;
509 * run_queue - do all outstanding tasks related to a glock
510 * @gl: The glock in question
511 * @nonblock: True if we must not block in run_queue
515 static void run_queue(struct gfs2_glock *gl, const int nonblock)
516 __releases(&gl->gl_lockref.lock)
517 __acquires(&gl->gl_lockref.lock)
519 struct gfs2_holder *gh = NULL;
520 int ret;
522 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
523 return;
525 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
527 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
528 gl->gl_demote_state != gl->gl_state) {
529 if (find_first_holder(gl))
530 goto out_unlock;
531 if (nonblock)
532 goto out_sched;
533 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
534 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
535 gl->gl_target = gl->gl_demote_state;
536 } else {
537 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
538 gfs2_demote_wake(gl);
539 ret = do_promote(gl);
540 if (ret == 0)
541 goto out_unlock;
542 if (ret == 2)
543 goto out;
544 gh = find_first_waiter(gl);
545 gl->gl_target = gh->gh_state;
546 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
547 do_error(gl, 0); /* Fail queued try locks */
549 do_xmote(gl, gh, gl->gl_target);
550 out:
551 return;
553 out_sched:
554 clear_bit(GLF_LOCK, &gl->gl_flags);
555 smp_mb__after_atomic();
556 gl->gl_lockref.count++;
557 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
558 gl->gl_lockref.count--;
559 return;
561 out_unlock:
562 clear_bit(GLF_LOCK, &gl->gl_flags);
563 smp_mb__after_atomic();
564 return;
567 static void delete_work_func(struct work_struct *work)
569 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
570 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
571 struct gfs2_inode *ip;
572 struct inode *inode;
573 u64 no_addr = gl->gl_name.ln_number;
575 ip = gl->gl_object;
576 /* Note: Unsafe to dereference ip as we don't hold right refs/locks */
578 if (ip)
579 inode = gfs2_ilookup(sdp->sd_vfs, no_addr, 1);
580 else
581 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
582 if (inode && !IS_ERR(inode)) {
583 d_prune_aliases(inode);
584 iput(inode);
586 gfs2_glock_put(gl);
589 static void glock_work_func(struct work_struct *work)
591 unsigned long delay = 0;
592 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
593 int drop_ref = 0;
595 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
596 finish_xmote(gl, gl->gl_reply);
597 drop_ref = 1;
599 spin_lock(&gl->gl_lockref.lock);
600 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
601 gl->gl_state != LM_ST_UNLOCKED &&
602 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
603 unsigned long holdtime, now = jiffies;
605 holdtime = gl->gl_tchange + gl->gl_hold_time;
606 if (time_before(now, holdtime))
607 delay = holdtime - now;
609 if (!delay) {
610 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
611 set_bit(GLF_DEMOTE, &gl->gl_flags);
614 run_queue(gl, 0);
615 spin_unlock(&gl->gl_lockref.lock);
616 if (!delay)
617 gfs2_glock_put(gl);
618 else {
619 if (gl->gl_name.ln_type != LM_TYPE_INODE)
620 delay = 0;
621 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
622 gfs2_glock_put(gl);
624 if (drop_ref)
625 gfs2_glock_put(gl);
629 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
630 * @sdp: The GFS2 superblock
631 * @number: the lock number
632 * @glops: The glock_operations to use
633 * @create: If 0, don't create the glock if it doesn't exist
634 * @glp: the glock is returned here
636 * This does not lock a glock, just finds/creates structures for one.
638 * Returns: errno
641 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
642 const struct gfs2_glock_operations *glops, int create,
643 struct gfs2_glock **glp)
645 struct super_block *s = sdp->sd_vfs;
646 struct lm_lockname name = { .ln_number = number,
647 .ln_type = glops->go_type,
648 .ln_sbd = sdp };
649 struct gfs2_glock *gl, *tmp = NULL;
650 struct address_space *mapping;
651 struct kmem_cache *cachep;
652 int ret, tries = 0;
654 gl = rhashtable_lookup_fast(&gl_hash_table, &name, ht_parms);
655 if (gl && !lockref_get_not_dead(&gl->gl_lockref))
656 gl = NULL;
658 *glp = gl;
659 if (gl)
660 return 0;
661 if (!create)
662 return -ENOENT;
664 if (glops->go_flags & GLOF_ASPACE)
665 cachep = gfs2_glock_aspace_cachep;
666 else
667 cachep = gfs2_glock_cachep;
668 gl = kmem_cache_alloc(cachep, GFP_NOFS);
669 if (!gl)
670 return -ENOMEM;
672 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
674 if (glops->go_flags & GLOF_LVB) {
675 gl->gl_lksb.sb_lvbptr = kzalloc(GFS2_MIN_LVB_SIZE, GFP_NOFS);
676 if (!gl->gl_lksb.sb_lvbptr) {
677 kmem_cache_free(cachep, gl);
678 return -ENOMEM;
682 atomic_inc(&sdp->sd_glock_disposal);
683 gl->gl_node.next = NULL;
684 gl->gl_flags = 0;
685 gl->gl_name = name;
686 gl->gl_lockref.count = 1;
687 gl->gl_state = LM_ST_UNLOCKED;
688 gl->gl_target = LM_ST_UNLOCKED;
689 gl->gl_demote_state = LM_ST_EXCLUSIVE;
690 gl->gl_ops = glops;
691 gl->gl_dstamp = ktime_set(0, 0);
692 preempt_disable();
693 /* We use the global stats to estimate the initial per-glock stats */
694 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
695 preempt_enable();
696 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
697 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
698 gl->gl_tchange = jiffies;
699 gl->gl_object = NULL;
700 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
701 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
702 INIT_WORK(&gl->gl_delete, delete_work_func);
704 mapping = gfs2_glock2aspace(gl);
705 if (mapping) {
706 mapping->a_ops = &gfs2_meta_aops;
707 mapping->host = s->s_bdev->bd_inode;
708 mapping->flags = 0;
709 mapping_set_gfp_mask(mapping, GFP_NOFS);
710 mapping->private_data = NULL;
711 mapping->writeback_index = 0;
714 again:
715 ret = rhashtable_lookup_insert_fast(&gl_hash_table, &gl->gl_node,
716 ht_parms);
717 if (ret == 0) {
718 *glp = gl;
719 return 0;
722 if (ret == -EEXIST) {
723 ret = 0;
724 tmp = rhashtable_lookup_fast(&gl_hash_table, &name, ht_parms);
725 if (tmp == NULL || !lockref_get_not_dead(&tmp->gl_lockref)) {
726 if (++tries < 100) {
727 cond_resched();
728 goto again;
730 tmp = NULL;
731 ret = -ENOMEM;
733 } else {
734 WARN_ON_ONCE(ret);
736 kfree(gl->gl_lksb.sb_lvbptr);
737 kmem_cache_free(cachep, gl);
738 atomic_dec(&sdp->sd_glock_disposal);
739 *glp = tmp;
741 return ret;
745 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
746 * @gl: the glock
747 * @state: the state we're requesting
748 * @flags: the modifier flags
749 * @gh: the holder structure
753 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
754 struct gfs2_holder *gh)
756 INIT_LIST_HEAD(&gh->gh_list);
757 gh->gh_gl = gl;
758 gh->gh_ip = _RET_IP_;
759 gh->gh_owner_pid = get_pid(task_pid(current));
760 gh->gh_state = state;
761 gh->gh_flags = flags;
762 gh->gh_error = 0;
763 gh->gh_iflags = 0;
764 gfs2_glock_hold(gl);
768 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
769 * @state: the state we're requesting
770 * @flags: the modifier flags
771 * @gh: the holder structure
773 * Don't mess with the glock.
777 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
779 gh->gh_state = state;
780 gh->gh_flags = flags;
781 gh->gh_iflags = 0;
782 gh->gh_ip = _RET_IP_;
783 put_pid(gh->gh_owner_pid);
784 gh->gh_owner_pid = get_pid(task_pid(current));
788 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
789 * @gh: the holder structure
793 void gfs2_holder_uninit(struct gfs2_holder *gh)
795 put_pid(gh->gh_owner_pid);
796 gfs2_glock_put(gh->gh_gl);
797 gh->gh_gl = NULL;
798 gh->gh_ip = 0;
802 * gfs2_glock_wait - wait on a glock acquisition
803 * @gh: the glock holder
805 * Returns: 0 on success
808 int gfs2_glock_wait(struct gfs2_holder *gh)
810 unsigned long time1 = jiffies;
812 might_sleep();
813 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
814 if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
815 /* Lengthen the minimum hold time. */
816 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
817 GL_GLOCK_HOLD_INCR,
818 GL_GLOCK_MAX_HOLD);
819 return gh->gh_error;
823 * handle_callback - process a demote request
824 * @gl: the glock
825 * @state: the state the caller wants us to change to
827 * There are only two requests that we are going to see in actual
828 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
831 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
832 unsigned long delay, bool remote)
834 int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
836 set_bit(bit, &gl->gl_flags);
837 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
838 gl->gl_demote_state = state;
839 gl->gl_demote_time = jiffies;
840 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
841 gl->gl_demote_state != state) {
842 gl->gl_demote_state = LM_ST_UNLOCKED;
844 if (gl->gl_ops->go_callback)
845 gl->gl_ops->go_callback(gl, remote);
846 trace_gfs2_demote_rq(gl, remote);
849 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
851 struct va_format vaf;
852 va_list args;
854 va_start(args, fmt);
856 if (seq) {
857 seq_vprintf(seq, fmt, args);
858 } else {
859 vaf.fmt = fmt;
860 vaf.va = &args;
862 pr_err("%pV", &vaf);
865 va_end(args);
869 * add_to_queue - Add a holder to the wait queue (but look for recursion)
870 * @gh: the holder structure to add
872 * Eventually we should move the recursive locking trap to a
873 * debugging option or something like that. This is the fast
874 * path and needs to have the minimum number of distractions.
878 static inline void add_to_queue(struct gfs2_holder *gh)
879 __releases(&gl->gl_lockref.lock)
880 __acquires(&gl->gl_lockref.lock)
882 struct gfs2_glock *gl = gh->gh_gl;
883 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
884 struct list_head *insert_pt = NULL;
885 struct gfs2_holder *gh2;
886 int try_futile = 0;
888 BUG_ON(gh->gh_owner_pid == NULL);
889 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
890 BUG();
892 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
893 if (test_bit(GLF_LOCK, &gl->gl_flags))
894 try_futile = !may_grant(gl, gh);
895 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
896 goto fail;
899 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
900 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
901 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
902 goto trap_recursive;
903 if (try_futile &&
904 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
905 fail:
906 gh->gh_error = GLR_TRYFAILED;
907 gfs2_holder_wake(gh);
908 return;
910 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
911 continue;
912 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
913 insert_pt = &gh2->gh_list;
915 set_bit(GLF_QUEUED, &gl->gl_flags);
916 trace_gfs2_glock_queue(gh, 1);
917 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
918 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
919 if (likely(insert_pt == NULL)) {
920 list_add_tail(&gh->gh_list, &gl->gl_holders);
921 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
922 goto do_cancel;
923 return;
925 list_add_tail(&gh->gh_list, insert_pt);
926 do_cancel:
927 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
928 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
929 spin_unlock(&gl->gl_lockref.lock);
930 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
931 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
932 spin_lock(&gl->gl_lockref.lock);
934 return;
936 trap_recursive:
937 pr_err("original: %pSR\n", (void *)gh2->gh_ip);
938 pr_err("pid: %d\n", pid_nr(gh2->gh_owner_pid));
939 pr_err("lock type: %d req lock state : %d\n",
940 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
941 pr_err("new: %pSR\n", (void *)gh->gh_ip);
942 pr_err("pid: %d\n", pid_nr(gh->gh_owner_pid));
943 pr_err("lock type: %d req lock state : %d\n",
944 gh->gh_gl->gl_name.ln_type, gh->gh_state);
945 gfs2_dump_glock(NULL, gl);
946 BUG();
950 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
951 * @gh: the holder structure
953 * if (gh->gh_flags & GL_ASYNC), this never returns an error
955 * Returns: 0, GLR_TRYFAILED, or errno on failure
958 int gfs2_glock_nq(struct gfs2_holder *gh)
960 struct gfs2_glock *gl = gh->gh_gl;
961 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
962 int error = 0;
964 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
965 return -EIO;
967 if (test_bit(GLF_LRU, &gl->gl_flags))
968 gfs2_glock_remove_from_lru(gl);
970 spin_lock(&gl->gl_lockref.lock);
971 add_to_queue(gh);
972 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
973 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
974 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
975 gl->gl_lockref.count++;
976 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
977 gl->gl_lockref.count--;
979 run_queue(gl, 1);
980 spin_unlock(&gl->gl_lockref.lock);
982 if (!(gh->gh_flags & GL_ASYNC))
983 error = gfs2_glock_wait(gh);
985 return error;
989 * gfs2_glock_poll - poll to see if an async request has been completed
990 * @gh: the holder
992 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
995 int gfs2_glock_poll(struct gfs2_holder *gh)
997 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1001 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1002 * @gh: the glock holder
1006 void gfs2_glock_dq(struct gfs2_holder *gh)
1008 struct gfs2_glock *gl = gh->gh_gl;
1009 const struct gfs2_glock_operations *glops = gl->gl_ops;
1010 unsigned delay = 0;
1011 int fast_path = 0;
1013 spin_lock(&gl->gl_lockref.lock);
1014 if (gh->gh_flags & GL_NOCACHE)
1015 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1017 list_del_init(&gh->gh_list);
1018 if (find_first_holder(gl) == NULL) {
1019 if (glops->go_unlock) {
1020 GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1021 spin_unlock(&gl->gl_lockref.lock);
1022 glops->go_unlock(gh);
1023 spin_lock(&gl->gl_lockref.lock);
1024 clear_bit(GLF_LOCK, &gl->gl_flags);
1026 if (list_empty(&gl->gl_holders) &&
1027 !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1028 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1029 fast_path = 1;
1031 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl) &&
1032 (glops->go_flags & GLOF_LRU))
1033 gfs2_glock_add_to_lru(gl);
1035 trace_gfs2_glock_queue(gh, 0);
1036 spin_unlock(&gl->gl_lockref.lock);
1037 if (likely(fast_path))
1038 return;
1040 gfs2_glock_hold(gl);
1041 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1042 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1043 gl->gl_name.ln_type == LM_TYPE_INODE)
1044 delay = gl->gl_hold_time;
1045 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1046 gfs2_glock_put(gl);
1049 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1051 struct gfs2_glock *gl = gh->gh_gl;
1052 gfs2_glock_dq(gh);
1053 might_sleep();
1054 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1058 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1059 * @gh: the holder structure
1063 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1065 gfs2_glock_dq(gh);
1066 gfs2_holder_uninit(gh);
1070 * gfs2_glock_nq_num - acquire a glock based on lock number
1071 * @sdp: the filesystem
1072 * @number: the lock number
1073 * @glops: the glock operations for the type of glock
1074 * @state: the state to acquire the glock in
1075 * @flags: modifier flags for the acquisition
1076 * @gh: the struct gfs2_holder
1078 * Returns: errno
1081 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1082 const struct gfs2_glock_operations *glops,
1083 unsigned int state, int flags, struct gfs2_holder *gh)
1085 struct gfs2_glock *gl;
1086 int error;
1088 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1089 if (!error) {
1090 error = gfs2_glock_nq_init(gl, state, flags, gh);
1091 gfs2_glock_put(gl);
1094 return error;
1098 * glock_compare - Compare two struct gfs2_glock structures for sorting
1099 * @arg_a: the first structure
1100 * @arg_b: the second structure
1104 static int glock_compare(const void *arg_a, const void *arg_b)
1106 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1107 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1108 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1109 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1111 if (a->ln_number > b->ln_number)
1112 return 1;
1113 if (a->ln_number < b->ln_number)
1114 return -1;
1115 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1116 return 0;
1120 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1121 * @num_gh: the number of structures
1122 * @ghs: an array of struct gfs2_holder structures
1124 * Returns: 0 on success (all glocks acquired),
1125 * errno on failure (no glocks acquired)
1128 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1129 struct gfs2_holder **p)
1131 unsigned int x;
1132 int error = 0;
1134 for (x = 0; x < num_gh; x++)
1135 p[x] = &ghs[x];
1137 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1139 for (x = 0; x < num_gh; x++) {
1140 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1142 error = gfs2_glock_nq(p[x]);
1143 if (error) {
1144 while (x--)
1145 gfs2_glock_dq(p[x]);
1146 break;
1150 return error;
1154 * gfs2_glock_nq_m - acquire multiple glocks
1155 * @num_gh: the number of structures
1156 * @ghs: an array of struct gfs2_holder structures
1159 * Returns: 0 on success (all glocks acquired),
1160 * errno on failure (no glocks acquired)
1163 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1165 struct gfs2_holder *tmp[4];
1166 struct gfs2_holder **pph = tmp;
1167 int error = 0;
1169 switch(num_gh) {
1170 case 0:
1171 return 0;
1172 case 1:
1173 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1174 return gfs2_glock_nq(ghs);
1175 default:
1176 if (num_gh <= 4)
1177 break;
1178 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1179 if (!pph)
1180 return -ENOMEM;
1183 error = nq_m_sync(num_gh, ghs, pph);
1185 if (pph != tmp)
1186 kfree(pph);
1188 return error;
1192 * gfs2_glock_dq_m - release multiple glocks
1193 * @num_gh: the number of structures
1194 * @ghs: an array of struct gfs2_holder structures
1198 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1200 while (num_gh--)
1201 gfs2_glock_dq(&ghs[num_gh]);
1204 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1206 unsigned long delay = 0;
1207 unsigned long holdtime;
1208 unsigned long now = jiffies;
1210 gfs2_glock_hold(gl);
1211 holdtime = gl->gl_tchange + gl->gl_hold_time;
1212 if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1213 gl->gl_name.ln_type == LM_TYPE_INODE) {
1214 if (time_before(now, holdtime))
1215 delay = holdtime - now;
1216 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1217 delay = gl->gl_hold_time;
1220 spin_lock(&gl->gl_lockref.lock);
1221 handle_callback(gl, state, delay, true);
1222 spin_unlock(&gl->gl_lockref.lock);
1223 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1224 gfs2_glock_put(gl);
1228 * gfs2_should_freeze - Figure out if glock should be frozen
1229 * @gl: The glock in question
1231 * Glocks are not frozen if (a) the result of the dlm operation is
1232 * an error, (b) the locking operation was an unlock operation or
1233 * (c) if there is a "noexp" flagged request anywhere in the queue
1235 * Returns: 1 if freezing should occur, 0 otherwise
1238 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1240 const struct gfs2_holder *gh;
1242 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1243 return 0;
1244 if (gl->gl_target == LM_ST_UNLOCKED)
1245 return 0;
1247 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1248 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1249 continue;
1250 if (LM_FLAG_NOEXP & gh->gh_flags)
1251 return 0;
1254 return 1;
1258 * gfs2_glock_complete - Callback used by locking
1259 * @gl: Pointer to the glock
1260 * @ret: The return value from the dlm
1262 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1263 * to use a bitfield shared with other glock state fields.
1266 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1268 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1270 spin_lock(&gl->gl_lockref.lock);
1271 gl->gl_reply = ret;
1273 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1274 if (gfs2_should_freeze(gl)) {
1275 set_bit(GLF_FROZEN, &gl->gl_flags);
1276 spin_unlock(&gl->gl_lockref.lock);
1277 return;
1281 gl->gl_lockref.count++;
1282 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1283 spin_unlock(&gl->gl_lockref.lock);
1285 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1286 gfs2_glock_put(gl);
1289 static int glock_cmp(void *priv, struct list_head *a, struct list_head *b)
1291 struct gfs2_glock *gla, *glb;
1293 gla = list_entry(a, struct gfs2_glock, gl_lru);
1294 glb = list_entry(b, struct gfs2_glock, gl_lru);
1296 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1297 return 1;
1298 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1299 return -1;
1301 return 0;
1305 * gfs2_dispose_glock_lru - Demote a list of glocks
1306 * @list: The list to dispose of
1308 * Disposing of glocks may involve disk accesses, so that here we sort
1309 * the glocks by number (i.e. disk location of the inodes) so that if
1310 * there are any such accesses, they'll be sent in order (mostly).
1312 * Must be called under the lru_lock, but may drop and retake this
1313 * lock. While the lru_lock is dropped, entries may vanish from the
1314 * list, but no new entries will appear on the list (since it is
1315 * private)
1318 static void gfs2_dispose_glock_lru(struct list_head *list)
1319 __releases(&lru_lock)
1320 __acquires(&lru_lock)
1322 struct gfs2_glock *gl;
1324 list_sort(NULL, list, glock_cmp);
1326 while(!list_empty(list)) {
1327 gl = list_entry(list->next, struct gfs2_glock, gl_lru);
1328 list_del_init(&gl->gl_lru);
1329 if (!spin_trylock(&gl->gl_lockref.lock)) {
1330 add_back_to_lru:
1331 list_add(&gl->gl_lru, &lru_list);
1332 atomic_inc(&lru_count);
1333 continue;
1335 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1336 spin_unlock(&gl->gl_lockref.lock);
1337 goto add_back_to_lru;
1339 clear_bit(GLF_LRU, &gl->gl_flags);
1340 gl->gl_lockref.count++;
1341 if (demote_ok(gl))
1342 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1343 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1344 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1345 gl->gl_lockref.count--;
1346 spin_unlock(&gl->gl_lockref.lock);
1347 cond_resched_lock(&lru_lock);
1352 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1353 * @nr: The number of entries to scan
1355 * This function selects the entries on the LRU which are able to
1356 * be demoted, and then kicks off the process by calling
1357 * gfs2_dispose_glock_lru() above.
1360 static long gfs2_scan_glock_lru(int nr)
1362 struct gfs2_glock *gl;
1363 LIST_HEAD(skipped);
1364 LIST_HEAD(dispose);
1365 long freed = 0;
1367 spin_lock(&lru_lock);
1368 while ((nr-- >= 0) && !list_empty(&lru_list)) {
1369 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1371 /* Test for being demotable */
1372 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1373 list_move(&gl->gl_lru, &dispose);
1374 atomic_dec(&lru_count);
1375 freed++;
1376 continue;
1379 list_move(&gl->gl_lru, &skipped);
1381 list_splice(&skipped, &lru_list);
1382 if (!list_empty(&dispose))
1383 gfs2_dispose_glock_lru(&dispose);
1384 spin_unlock(&lru_lock);
1386 return freed;
1389 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1390 struct shrink_control *sc)
1392 if (!(sc->gfp_mask & __GFP_FS))
1393 return SHRINK_STOP;
1394 return gfs2_scan_glock_lru(sc->nr_to_scan);
1397 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1398 struct shrink_control *sc)
1400 return vfs_pressure_ratio(atomic_read(&lru_count));
1403 static struct shrinker glock_shrinker = {
1404 .seeks = DEFAULT_SEEKS,
1405 .count_objects = gfs2_glock_shrink_count,
1406 .scan_objects = gfs2_glock_shrink_scan,
1410 * examine_bucket - Call a function for glock in a hash bucket
1411 * @examiner: the function
1412 * @sdp: the filesystem
1413 * @bucket: the bucket
1417 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1419 struct gfs2_glock *gl;
1420 struct rhash_head *pos, *next;
1421 const struct bucket_table *tbl;
1422 int i;
1424 rcu_read_lock();
1425 tbl = rht_dereference_rcu(gl_hash_table.tbl, &gl_hash_table);
1426 for (i = 0; i < tbl->size; i++) {
1427 rht_for_each_entry_safe(gl, pos, next, tbl, i, gl_node) {
1428 if ((gl->gl_name.ln_sbd == sdp) &&
1429 lockref_get_not_dead(&gl->gl_lockref))
1430 examiner(gl);
1433 rcu_read_unlock();
1434 cond_resched();
1438 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1439 * @gl: The glock to thaw
1443 static void thaw_glock(struct gfs2_glock *gl)
1445 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1446 goto out;
1447 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1448 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0) {
1449 out:
1450 gfs2_glock_put(gl);
1455 * clear_glock - look at a glock and see if we can free it from glock cache
1456 * @gl: the glock to look at
1460 static void clear_glock(struct gfs2_glock *gl)
1462 gfs2_glock_remove_from_lru(gl);
1464 spin_lock(&gl->gl_lockref.lock);
1465 if (gl->gl_state != LM_ST_UNLOCKED)
1466 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1467 spin_unlock(&gl->gl_lockref.lock);
1468 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1469 gfs2_glock_put(gl);
1473 * gfs2_glock_thaw - Thaw any frozen glocks
1474 * @sdp: The super block
1478 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1480 glock_hash_walk(thaw_glock, sdp);
1483 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1485 spin_lock(&gl->gl_lockref.lock);
1486 gfs2_dump_glock(seq, gl);
1487 spin_unlock(&gl->gl_lockref.lock);
1490 static void dump_glock_func(struct gfs2_glock *gl)
1492 dump_glock(NULL, gl);
1496 * gfs2_gl_hash_clear - Empty out the glock hash table
1497 * @sdp: the filesystem
1498 * @wait: wait until it's all gone
1500 * Called when unmounting the filesystem.
1503 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1505 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1506 flush_workqueue(glock_workqueue);
1507 glock_hash_walk(clear_glock, sdp);
1508 flush_workqueue(glock_workqueue);
1509 wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
1510 glock_hash_walk(dump_glock_func, sdp);
1513 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1515 struct gfs2_glock *gl = ip->i_gl;
1516 int ret;
1518 ret = gfs2_truncatei_resume(ip);
1519 gfs2_assert_withdraw(gl->gl_name.ln_sbd, ret == 0);
1521 spin_lock(&gl->gl_lockref.lock);
1522 clear_bit(GLF_LOCK, &gl->gl_flags);
1523 run_queue(gl, 1);
1524 spin_unlock(&gl->gl_lockref.lock);
1527 static const char *state2str(unsigned state)
1529 switch(state) {
1530 case LM_ST_UNLOCKED:
1531 return "UN";
1532 case LM_ST_SHARED:
1533 return "SH";
1534 case LM_ST_DEFERRED:
1535 return "DF";
1536 case LM_ST_EXCLUSIVE:
1537 return "EX";
1539 return "??";
1542 static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1544 char *p = buf;
1545 if (flags & LM_FLAG_TRY)
1546 *p++ = 't';
1547 if (flags & LM_FLAG_TRY_1CB)
1548 *p++ = 'T';
1549 if (flags & LM_FLAG_NOEXP)
1550 *p++ = 'e';
1551 if (flags & LM_FLAG_ANY)
1552 *p++ = 'A';
1553 if (flags & LM_FLAG_PRIORITY)
1554 *p++ = 'p';
1555 if (flags & GL_ASYNC)
1556 *p++ = 'a';
1557 if (flags & GL_EXACT)
1558 *p++ = 'E';
1559 if (flags & GL_NOCACHE)
1560 *p++ = 'c';
1561 if (test_bit(HIF_HOLDER, &iflags))
1562 *p++ = 'H';
1563 if (test_bit(HIF_WAIT, &iflags))
1564 *p++ = 'W';
1565 if (test_bit(HIF_FIRST, &iflags))
1566 *p++ = 'F';
1567 *p = 0;
1568 return buf;
1572 * dump_holder - print information about a glock holder
1573 * @seq: the seq_file struct
1574 * @gh: the glock holder
1578 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1580 struct task_struct *gh_owner = NULL;
1581 char flags_buf[32];
1583 rcu_read_lock();
1584 if (gh->gh_owner_pid)
1585 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1586 gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1587 state2str(gh->gh_state),
1588 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1589 gh->gh_error,
1590 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1591 gh_owner ? gh_owner->comm : "(ended)",
1592 (void *)gh->gh_ip);
1593 rcu_read_unlock();
1596 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1598 const unsigned long *gflags = &gl->gl_flags;
1599 char *p = buf;
1601 if (test_bit(GLF_LOCK, gflags))
1602 *p++ = 'l';
1603 if (test_bit(GLF_DEMOTE, gflags))
1604 *p++ = 'D';
1605 if (test_bit(GLF_PENDING_DEMOTE, gflags))
1606 *p++ = 'd';
1607 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1608 *p++ = 'p';
1609 if (test_bit(GLF_DIRTY, gflags))
1610 *p++ = 'y';
1611 if (test_bit(GLF_LFLUSH, gflags))
1612 *p++ = 'f';
1613 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1614 *p++ = 'i';
1615 if (test_bit(GLF_REPLY_PENDING, gflags))
1616 *p++ = 'r';
1617 if (test_bit(GLF_INITIAL, gflags))
1618 *p++ = 'I';
1619 if (test_bit(GLF_FROZEN, gflags))
1620 *p++ = 'F';
1621 if (test_bit(GLF_QUEUED, gflags))
1622 *p++ = 'q';
1623 if (test_bit(GLF_LRU, gflags))
1624 *p++ = 'L';
1625 if (gl->gl_object)
1626 *p++ = 'o';
1627 if (test_bit(GLF_BLOCKING, gflags))
1628 *p++ = 'b';
1629 *p = 0;
1630 return buf;
1634 * gfs2_dump_glock - print information about a glock
1635 * @seq: The seq_file struct
1636 * @gl: the glock
1638 * The file format is as follows:
1639 * One line per object, capital letters are used to indicate objects
1640 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1641 * other objects are indented by a single space and follow the glock to
1642 * which they are related. Fields are indicated by lower case letters
1643 * followed by a colon and the field value, except for strings which are in
1644 * [] so that its possible to see if they are composed of spaces for
1645 * example. The field's are n = number (id of the object), f = flags,
1646 * t = type, s = state, r = refcount, e = error, p = pid.
1650 void gfs2_dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1652 const struct gfs2_glock_operations *glops = gl->gl_ops;
1653 unsigned long long dtime;
1654 const struct gfs2_holder *gh;
1655 char gflags_buf[32];
1657 dtime = jiffies - gl->gl_demote_time;
1658 dtime *= 1000000/HZ; /* demote time in uSec */
1659 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1660 dtime = 0;
1661 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",
1662 state2str(gl->gl_state),
1663 gl->gl_name.ln_type,
1664 (unsigned long long)gl->gl_name.ln_number,
1665 gflags2str(gflags_buf, gl),
1666 state2str(gl->gl_target),
1667 state2str(gl->gl_demote_state), dtime,
1668 atomic_read(&gl->gl_ail_count),
1669 atomic_read(&gl->gl_revokes),
1670 (int)gl->gl_lockref.count, gl->gl_hold_time);
1672 list_for_each_entry(gh, &gl->gl_holders, gh_list)
1673 dump_holder(seq, gh);
1675 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1676 glops->go_dump(seq, gl);
1679 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
1681 struct gfs2_glock *gl = iter_ptr;
1683 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
1684 gl->gl_name.ln_type,
1685 (unsigned long long)gl->gl_name.ln_number,
1686 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
1687 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
1688 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
1689 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
1690 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
1691 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
1692 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
1693 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
1694 return 0;
1697 static const char *gfs2_gltype[] = {
1698 "type",
1699 "reserved",
1700 "nondisk",
1701 "inode",
1702 "rgrp",
1703 "meta",
1704 "iopen",
1705 "flock",
1706 "plock",
1707 "quota",
1708 "journal",
1711 static const char *gfs2_stype[] = {
1712 [GFS2_LKS_SRTT] = "srtt",
1713 [GFS2_LKS_SRTTVAR] = "srttvar",
1714 [GFS2_LKS_SRTTB] = "srttb",
1715 [GFS2_LKS_SRTTVARB] = "srttvarb",
1716 [GFS2_LKS_SIRT] = "sirt",
1717 [GFS2_LKS_SIRTVAR] = "sirtvar",
1718 [GFS2_LKS_DCOUNT] = "dlm",
1719 [GFS2_LKS_QCOUNT] = "queue",
1722 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1724 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
1726 struct gfs2_sbd *sdp = seq->private;
1727 loff_t pos = *(loff_t *)iter_ptr;
1728 unsigned index = pos >> 3;
1729 unsigned subindex = pos & 0x07;
1730 int i;
1732 if (index == 0 && subindex != 0)
1733 return 0;
1735 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
1736 (index == 0) ? "cpu": gfs2_stype[subindex]);
1738 for_each_possible_cpu(i) {
1739 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
1741 if (index == 0)
1742 seq_printf(seq, " %15u", i);
1743 else
1744 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
1745 lkstats[index - 1].stats[subindex]);
1747 seq_putc(seq, '\n');
1748 return 0;
1751 int __init gfs2_glock_init(void)
1753 int ret;
1755 ret = rhashtable_init(&gl_hash_table, &ht_parms);
1756 if (ret < 0)
1757 return ret;
1759 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1760 WQ_HIGHPRI | WQ_FREEZABLE, 0);
1761 if (!glock_workqueue) {
1762 rhashtable_destroy(&gl_hash_table);
1763 return -ENOMEM;
1765 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1766 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1768 if (!gfs2_delete_workqueue) {
1769 destroy_workqueue(glock_workqueue);
1770 rhashtable_destroy(&gl_hash_table);
1771 return -ENOMEM;
1774 register_shrinker(&glock_shrinker);
1776 return 0;
1779 void gfs2_glock_exit(void)
1781 unregister_shrinker(&glock_shrinker);
1782 rhashtable_destroy(&gl_hash_table);
1783 destroy_workqueue(glock_workqueue);
1784 destroy_workqueue(gfs2_delete_workqueue);
1787 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1789 do {
1790 gi->gl = rhashtable_walk_next(&gi->hti);
1791 if (IS_ERR(gi->gl)) {
1792 if (PTR_ERR(gi->gl) == -EAGAIN)
1793 continue;
1794 gi->gl = NULL;
1796 /* Skip entries for other sb and dead entries */
1797 } while ((gi->gl) && ((gi->sdp != gi->gl->gl_name.ln_sbd) ||
1798 __lockref_is_dead(&gi->gl->gl_lockref)));
1801 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1803 struct gfs2_glock_iter *gi = seq->private;
1804 loff_t n = *pos;
1805 int ret;
1807 if (gi->last_pos <= *pos)
1808 n = (*pos - gi->last_pos);
1810 ret = rhashtable_walk_start(&gi->hti);
1811 if (ret)
1812 return NULL;
1814 do {
1815 gfs2_glock_iter_next(gi);
1816 } while (gi->gl && n--);
1818 gi->last_pos = *pos;
1819 return gi->gl;
1822 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1823 loff_t *pos)
1825 struct gfs2_glock_iter *gi = seq->private;
1827 (*pos)++;
1828 gi->last_pos = *pos;
1829 gfs2_glock_iter_next(gi);
1830 return gi->gl;
1833 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1835 struct gfs2_glock_iter *gi = seq->private;
1837 gi->gl = NULL;
1838 rhashtable_walk_stop(&gi->hti);
1841 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1843 dump_glock(seq, iter_ptr);
1844 return 0;
1847 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
1849 preempt_disable();
1850 if (*pos >= GFS2_NR_SBSTATS)
1851 return NULL;
1852 return pos;
1855 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
1856 loff_t *pos)
1858 (*pos)++;
1859 if (*pos >= GFS2_NR_SBSTATS)
1860 return NULL;
1861 return pos;
1864 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
1866 preempt_enable();
1869 static const struct seq_operations gfs2_glock_seq_ops = {
1870 .start = gfs2_glock_seq_start,
1871 .next = gfs2_glock_seq_next,
1872 .stop = gfs2_glock_seq_stop,
1873 .show = gfs2_glock_seq_show,
1876 static const struct seq_operations gfs2_glstats_seq_ops = {
1877 .start = gfs2_glock_seq_start,
1878 .next = gfs2_glock_seq_next,
1879 .stop = gfs2_glock_seq_stop,
1880 .show = gfs2_glstats_seq_show,
1883 static const struct seq_operations gfs2_sbstats_seq_ops = {
1884 .start = gfs2_sbstats_seq_start,
1885 .next = gfs2_sbstats_seq_next,
1886 .stop = gfs2_sbstats_seq_stop,
1887 .show = gfs2_sbstats_seq_show,
1890 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
1892 static int gfs2_glocks_open(struct inode *inode, struct file *file)
1894 int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1895 sizeof(struct gfs2_glock_iter));
1896 if (ret == 0) {
1897 struct seq_file *seq = file->private_data;
1898 struct gfs2_glock_iter *gi = seq->private;
1900 gi->sdp = inode->i_private;
1901 gi->last_pos = 0;
1902 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
1903 if (seq->buf)
1904 seq->size = GFS2_SEQ_GOODSIZE;
1905 gi->gl = NULL;
1906 ret = rhashtable_walk_init(&gl_hash_table, &gi->hti);
1908 return ret;
1911 static int gfs2_glocks_release(struct inode *inode, struct file *file)
1913 struct seq_file *seq = file->private_data;
1914 struct gfs2_glock_iter *gi = seq->private;
1916 gi->gl = NULL;
1917 rhashtable_walk_exit(&gi->hti);
1918 return seq_release_private(inode, file);
1921 static int gfs2_glstats_open(struct inode *inode, struct file *file)
1923 int ret = seq_open_private(file, &gfs2_glstats_seq_ops,
1924 sizeof(struct gfs2_glock_iter));
1925 if (ret == 0) {
1926 struct seq_file *seq = file->private_data;
1927 struct gfs2_glock_iter *gi = seq->private;
1928 gi->sdp = inode->i_private;
1929 gi->last_pos = 0;
1930 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
1931 if (seq->buf)
1932 seq->size = GFS2_SEQ_GOODSIZE;
1933 gi->gl = NULL;
1934 ret = rhashtable_walk_init(&gl_hash_table, &gi->hti);
1936 return ret;
1939 static int gfs2_sbstats_open(struct inode *inode, struct file *file)
1941 int ret = seq_open(file, &gfs2_sbstats_seq_ops);
1942 if (ret == 0) {
1943 struct seq_file *seq = file->private_data;
1944 seq->private = inode->i_private; /* sdp */
1946 return ret;
1949 static const struct file_operations gfs2_glocks_fops = {
1950 .owner = THIS_MODULE,
1951 .open = gfs2_glocks_open,
1952 .read = seq_read,
1953 .llseek = seq_lseek,
1954 .release = gfs2_glocks_release,
1957 static const struct file_operations gfs2_glstats_fops = {
1958 .owner = THIS_MODULE,
1959 .open = gfs2_glstats_open,
1960 .read = seq_read,
1961 .llseek = seq_lseek,
1962 .release = gfs2_glocks_release,
1965 static const struct file_operations gfs2_sbstats_fops = {
1966 .owner = THIS_MODULE,
1967 .open = gfs2_sbstats_open,
1968 .read = seq_read,
1969 .llseek = seq_lseek,
1970 .release = seq_release,
1973 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
1975 struct dentry *dent;
1977 dent = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
1978 if (IS_ERR_OR_NULL(dent))
1979 goto fail;
1980 sdp->debugfs_dir = dent;
1982 dent = debugfs_create_file("glocks",
1983 S_IFREG | S_IRUGO,
1984 sdp->debugfs_dir, sdp,
1985 &gfs2_glocks_fops);
1986 if (IS_ERR_OR_NULL(dent))
1987 goto fail;
1988 sdp->debugfs_dentry_glocks = dent;
1990 dent = debugfs_create_file("glstats",
1991 S_IFREG | S_IRUGO,
1992 sdp->debugfs_dir, sdp,
1993 &gfs2_glstats_fops);
1994 if (IS_ERR_OR_NULL(dent))
1995 goto fail;
1996 sdp->debugfs_dentry_glstats = dent;
1998 dent = debugfs_create_file("sbstats",
1999 S_IFREG | S_IRUGO,
2000 sdp->debugfs_dir, sdp,
2001 &gfs2_sbstats_fops);
2002 if (IS_ERR_OR_NULL(dent))
2003 goto fail;
2004 sdp->debugfs_dentry_sbstats = dent;
2006 return 0;
2007 fail:
2008 gfs2_delete_debugfs_file(sdp);
2009 return dent ? PTR_ERR(dent) : -ENOMEM;
2012 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2014 if (sdp->debugfs_dir) {
2015 if (sdp->debugfs_dentry_glocks) {
2016 debugfs_remove(sdp->debugfs_dentry_glocks);
2017 sdp->debugfs_dentry_glocks = NULL;
2019 if (sdp->debugfs_dentry_glstats) {
2020 debugfs_remove(sdp->debugfs_dentry_glstats);
2021 sdp->debugfs_dentry_glstats = NULL;
2023 if (sdp->debugfs_dentry_sbstats) {
2024 debugfs_remove(sdp->debugfs_dentry_sbstats);
2025 sdp->debugfs_dentry_sbstats = NULL;
2027 debugfs_remove(sdp->debugfs_dir);
2028 sdp->debugfs_dir = NULL;
2032 int gfs2_register_debugfs(void)
2034 gfs2_root = debugfs_create_dir("gfs2", NULL);
2035 if (IS_ERR(gfs2_root))
2036 return PTR_ERR(gfs2_root);
2037 return gfs2_root ? 0 : -ENOMEM;
2040 void gfs2_unregister_debugfs(void)
2042 debugfs_remove(gfs2_root);
2043 gfs2_root = NULL;