Loading...
1/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
4 *
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 */
9
10#include <linux/sched.h>
11#include <linux/slab.h>
12#include <linux/spinlock.h>
13#include <linux/buffer_head.h>
14#include <linux/delay.h>
15#include <linux/sort.h>
16#include <linux/jhash.h>
17#include <linux/kallsyms.h>
18#include <linux/gfs2_ondisk.h>
19#include <linux/list.h>
20#include <linux/wait.h>
21#include <linux/module.h>
22#include <asm/uaccess.h>
23#include <linux/seq_file.h>
24#include <linux/debugfs.h>
25#include <linux/kthread.h>
26#include <linux/freezer.h>
27#include <linux/workqueue.h>
28#include <linux/jiffies.h>
29#include <linux/rcupdate.h>
30#include <linux/rculist_bl.h>
31#include <linux/bit_spinlock.h>
32
33#include "gfs2.h"
34#include "incore.h"
35#include "glock.h"
36#include "glops.h"
37#include "inode.h"
38#include "lops.h"
39#include "meta_io.h"
40#include "quota.h"
41#include "super.h"
42#include "util.h"
43#include "bmap.h"
44#define CREATE_TRACE_POINTS
45#include "trace_gfs2.h"
46
47struct gfs2_glock_iter {
48 int hash; /* hash bucket index */
49 struct gfs2_sbd *sdp; /* incore superblock */
50 struct gfs2_glock *gl; /* current glock struct */
51 char string[512]; /* scratch space */
52};
53
54typedef void (*glock_examiner) (struct gfs2_glock * gl);
55
56static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl);
57#define GLOCK_BUG_ON(gl,x) do { if (unlikely(x)) { __dump_glock(NULL, gl); BUG(); } } while(0)
58static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
59
60static struct dentry *gfs2_root;
61static struct workqueue_struct *glock_workqueue;
62struct workqueue_struct *gfs2_delete_workqueue;
63static LIST_HEAD(lru_list);
64static atomic_t lru_count = ATOMIC_INIT(0);
65static DEFINE_SPINLOCK(lru_lock);
66
67#define GFS2_GL_HASH_SHIFT 15
68#define GFS2_GL_HASH_SIZE (1 << GFS2_GL_HASH_SHIFT)
69#define GFS2_GL_HASH_MASK (GFS2_GL_HASH_SIZE - 1)
70
71static struct hlist_bl_head gl_hash_table[GFS2_GL_HASH_SIZE];
72static struct dentry *gfs2_root;
73
74/**
75 * gl_hash() - Turn glock number into hash bucket number
76 * @lock: The glock number
77 *
78 * Returns: The number of the corresponding hash bucket
79 */
80
81static unsigned int gl_hash(const struct gfs2_sbd *sdp,
82 const struct lm_lockname *name)
83{
84 unsigned int h;
85
86 h = jhash(&name->ln_number, sizeof(u64), 0);
87 h = jhash(&name->ln_type, sizeof(unsigned int), h);
88 h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
89 h &= GFS2_GL_HASH_MASK;
90
91 return h;
92}
93
94static inline void spin_lock_bucket(unsigned int hash)
95{
96 hlist_bl_lock(&gl_hash_table[hash]);
97}
98
99static inline void spin_unlock_bucket(unsigned int hash)
100{
101 hlist_bl_unlock(&gl_hash_table[hash]);
102}
103
104static void gfs2_glock_dealloc(struct rcu_head *rcu)
105{
106 struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
107
108 if (gl->gl_ops->go_flags & GLOF_ASPACE)
109 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
110 else
111 kmem_cache_free(gfs2_glock_cachep, gl);
112}
113
114void gfs2_glock_free(struct gfs2_glock *gl)
115{
116 struct gfs2_sbd *sdp = gl->gl_sbd;
117
118 call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
119 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
120 wake_up(&sdp->sd_glock_wait);
121}
122
123/**
124 * gfs2_glock_hold() - increment reference count on glock
125 * @gl: The glock to hold
126 *
127 */
128
129void gfs2_glock_hold(struct gfs2_glock *gl)
130{
131 GLOCK_BUG_ON(gl, atomic_read(&gl->gl_ref) == 0);
132 atomic_inc(&gl->gl_ref);
133}
134
135/**
136 * demote_ok - Check to see if it's ok to unlock a glock
137 * @gl: the glock
138 *
139 * Returns: 1 if it's ok
140 */
141
142static int demote_ok(const struct gfs2_glock *gl)
143{
144 const struct gfs2_glock_operations *glops = gl->gl_ops;
145
146 if (gl->gl_state == LM_ST_UNLOCKED)
147 return 0;
148 if (!list_empty(&gl->gl_holders))
149 return 0;
150 if (glops->go_demote_ok)
151 return glops->go_demote_ok(gl);
152 return 1;
153}
154
155
156void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
157{
158 spin_lock(&lru_lock);
159
160 if (!list_empty(&gl->gl_lru))
161 list_del_init(&gl->gl_lru);
162 else
163 atomic_inc(&lru_count);
164
165 list_add_tail(&gl->gl_lru, &lru_list);
166 set_bit(GLF_LRU, &gl->gl_flags);
167 spin_unlock(&lru_lock);
168}
169
170static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
171{
172 spin_lock(&lru_lock);
173 if (!list_empty(&gl->gl_lru)) {
174 list_del_init(&gl->gl_lru);
175 atomic_dec(&lru_count);
176 clear_bit(GLF_LRU, &gl->gl_flags);
177 }
178 spin_unlock(&lru_lock);
179}
180
181/**
182 * __gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
183 * @gl: the glock
184 *
185 * If the glock is demotable, then we add it (or move it) to the end
186 * of the glock LRU list.
187 */
188
189static void __gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
190{
191 if (demote_ok(gl))
192 gfs2_glock_add_to_lru(gl);
193}
194
195/**
196 * gfs2_glock_put_nolock() - Decrement reference count on glock
197 * @gl: The glock to put
198 *
199 * This function should only be used if the caller has its own reference
200 * to the glock, in addition to the one it is dropping.
201 */
202
203void gfs2_glock_put_nolock(struct gfs2_glock *gl)
204{
205 if (atomic_dec_and_test(&gl->gl_ref))
206 GLOCK_BUG_ON(gl, 1);
207}
208
209/**
210 * gfs2_glock_put() - Decrement reference count on glock
211 * @gl: The glock to put
212 *
213 */
214
215void gfs2_glock_put(struct gfs2_glock *gl)
216{
217 struct gfs2_sbd *sdp = gl->gl_sbd;
218 struct address_space *mapping = gfs2_glock2aspace(gl);
219
220 if (atomic_dec_and_test(&gl->gl_ref)) {
221 spin_lock_bucket(gl->gl_hash);
222 hlist_bl_del_rcu(&gl->gl_list);
223 spin_unlock_bucket(gl->gl_hash);
224 gfs2_glock_remove_from_lru(gl);
225 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
226 GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
227 trace_gfs2_glock_put(gl);
228 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
229 }
230}
231
232/**
233 * search_bucket() - Find struct gfs2_glock by lock number
234 * @bucket: the bucket to search
235 * @name: The lock name
236 *
237 * Returns: NULL, or the struct gfs2_glock with the requested number
238 */
239
240static struct gfs2_glock *search_bucket(unsigned int hash,
241 const struct gfs2_sbd *sdp,
242 const struct lm_lockname *name)
243{
244 struct gfs2_glock *gl;
245 struct hlist_bl_node *h;
246
247 hlist_bl_for_each_entry_rcu(gl, h, &gl_hash_table[hash], gl_list) {
248 if (!lm_name_equal(&gl->gl_name, name))
249 continue;
250 if (gl->gl_sbd != sdp)
251 continue;
252 if (atomic_inc_not_zero(&gl->gl_ref))
253 return gl;
254 }
255
256 return NULL;
257}
258
259/**
260 * may_grant - check if its ok to grant a new lock
261 * @gl: The glock
262 * @gh: The lock request which we wish to grant
263 *
264 * Returns: true if its ok to grant the lock
265 */
266
267static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
268{
269 const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
270 if ((gh->gh_state == LM_ST_EXCLUSIVE ||
271 gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
272 return 0;
273 if (gl->gl_state == gh->gh_state)
274 return 1;
275 if (gh->gh_flags & GL_EXACT)
276 return 0;
277 if (gl->gl_state == LM_ST_EXCLUSIVE) {
278 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
279 return 1;
280 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
281 return 1;
282 }
283 if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
284 return 1;
285 return 0;
286}
287
288static void gfs2_holder_wake(struct gfs2_holder *gh)
289{
290 clear_bit(HIF_WAIT, &gh->gh_iflags);
291 smp_mb__after_clear_bit();
292 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
293}
294
295/**
296 * do_error - Something unexpected has happened during a lock request
297 *
298 */
299
300static inline void do_error(struct gfs2_glock *gl, const int ret)
301{
302 struct gfs2_holder *gh, *tmp;
303
304 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
305 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
306 continue;
307 if (ret & LM_OUT_ERROR)
308 gh->gh_error = -EIO;
309 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
310 gh->gh_error = GLR_TRYFAILED;
311 else
312 continue;
313 list_del_init(&gh->gh_list);
314 trace_gfs2_glock_queue(gh, 0);
315 gfs2_holder_wake(gh);
316 }
317}
318
319/**
320 * do_promote - promote as many requests as possible on the current queue
321 * @gl: The glock
322 *
323 * Returns: 1 if there is a blocked holder at the head of the list, or 2
324 * if a type specific operation is underway.
325 */
326
327static int do_promote(struct gfs2_glock *gl)
328__releases(&gl->gl_spin)
329__acquires(&gl->gl_spin)
330{
331 const struct gfs2_glock_operations *glops = gl->gl_ops;
332 struct gfs2_holder *gh, *tmp;
333 int ret;
334
335restart:
336 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
337 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
338 continue;
339 if (may_grant(gl, gh)) {
340 if (gh->gh_list.prev == &gl->gl_holders &&
341 glops->go_lock) {
342 spin_unlock(&gl->gl_spin);
343 /* FIXME: eliminate this eventually */
344 ret = glops->go_lock(gh);
345 spin_lock(&gl->gl_spin);
346 if (ret) {
347 if (ret == 1)
348 return 2;
349 gh->gh_error = ret;
350 list_del_init(&gh->gh_list);
351 trace_gfs2_glock_queue(gh, 0);
352 gfs2_holder_wake(gh);
353 goto restart;
354 }
355 set_bit(HIF_HOLDER, &gh->gh_iflags);
356 trace_gfs2_promote(gh, 1);
357 gfs2_holder_wake(gh);
358 goto restart;
359 }
360 set_bit(HIF_HOLDER, &gh->gh_iflags);
361 trace_gfs2_promote(gh, 0);
362 gfs2_holder_wake(gh);
363 continue;
364 }
365 if (gh->gh_list.prev == &gl->gl_holders)
366 return 1;
367 do_error(gl, 0);
368 break;
369 }
370 return 0;
371}
372
373/**
374 * find_first_waiter - find the first gh that's waiting for the glock
375 * @gl: the glock
376 */
377
378static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
379{
380 struct gfs2_holder *gh;
381
382 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
383 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
384 return gh;
385 }
386 return NULL;
387}
388
389/**
390 * state_change - record that the glock is now in a different state
391 * @gl: the glock
392 * @new_state the new state
393 *
394 */
395
396static void state_change(struct gfs2_glock *gl, unsigned int new_state)
397{
398 int held1, held2;
399
400 held1 = (gl->gl_state != LM_ST_UNLOCKED);
401 held2 = (new_state != LM_ST_UNLOCKED);
402
403 if (held1 != held2) {
404 if (held2)
405 gfs2_glock_hold(gl);
406 else
407 gfs2_glock_put_nolock(gl);
408 }
409 if (held1 && held2 && list_empty(&gl->gl_holders))
410 clear_bit(GLF_QUEUED, &gl->gl_flags);
411
412 if (new_state != gl->gl_target)
413 /* shorten our minimum hold time */
414 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
415 GL_GLOCK_MIN_HOLD);
416 gl->gl_state = new_state;
417 gl->gl_tchange = jiffies;
418}
419
420static void gfs2_demote_wake(struct gfs2_glock *gl)
421{
422 gl->gl_demote_state = LM_ST_EXCLUSIVE;
423 clear_bit(GLF_DEMOTE, &gl->gl_flags);
424 smp_mb__after_clear_bit();
425 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
426}
427
428/**
429 * finish_xmote - The DLM has replied to one of our lock requests
430 * @gl: The glock
431 * @ret: The status from the DLM
432 *
433 */
434
435static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
436{
437 const struct gfs2_glock_operations *glops = gl->gl_ops;
438 struct gfs2_holder *gh;
439 unsigned state = ret & LM_OUT_ST_MASK;
440 int rv;
441
442 spin_lock(&gl->gl_spin);
443 trace_gfs2_glock_state_change(gl, state);
444 state_change(gl, state);
445 gh = find_first_waiter(gl);
446
447 /* Demote to UN request arrived during demote to SH or DF */
448 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
449 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
450 gl->gl_target = LM_ST_UNLOCKED;
451
452 /* Check for state != intended state */
453 if (unlikely(state != gl->gl_target)) {
454 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
455 /* move to back of queue and try next entry */
456 if (ret & LM_OUT_CANCELED) {
457 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
458 list_move_tail(&gh->gh_list, &gl->gl_holders);
459 gh = find_first_waiter(gl);
460 gl->gl_target = gh->gh_state;
461 goto retry;
462 }
463 /* Some error or failed "try lock" - report it */
464 if ((ret & LM_OUT_ERROR) ||
465 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
466 gl->gl_target = gl->gl_state;
467 do_error(gl, ret);
468 goto out;
469 }
470 }
471 switch(state) {
472 /* Unlocked due to conversion deadlock, try again */
473 case LM_ST_UNLOCKED:
474retry:
475 do_xmote(gl, gh, gl->gl_target);
476 break;
477 /* Conversion fails, unlock and try again */
478 case LM_ST_SHARED:
479 case LM_ST_DEFERRED:
480 do_xmote(gl, gh, LM_ST_UNLOCKED);
481 break;
482 default: /* Everything else */
483 printk(KERN_ERR "GFS2: wanted %u got %u\n", gl->gl_target, state);
484 GLOCK_BUG_ON(gl, 1);
485 }
486 spin_unlock(&gl->gl_spin);
487 return;
488 }
489
490 /* Fast path - we got what we asked for */
491 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
492 gfs2_demote_wake(gl);
493 if (state != LM_ST_UNLOCKED) {
494 if (glops->go_xmote_bh) {
495 spin_unlock(&gl->gl_spin);
496 rv = glops->go_xmote_bh(gl, gh);
497 spin_lock(&gl->gl_spin);
498 if (rv) {
499 do_error(gl, rv);
500 goto out;
501 }
502 }
503 rv = do_promote(gl);
504 if (rv == 2)
505 goto out_locked;
506 }
507out:
508 clear_bit(GLF_LOCK, &gl->gl_flags);
509out_locked:
510 spin_unlock(&gl->gl_spin);
511}
512
513/**
514 * do_xmote - Calls the DLM to change the state of a lock
515 * @gl: The lock state
516 * @gh: The holder (only for promotes)
517 * @target: The target lock state
518 *
519 */
520
521static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
522__releases(&gl->gl_spin)
523__acquires(&gl->gl_spin)
524{
525 const struct gfs2_glock_operations *glops = gl->gl_ops;
526 struct gfs2_sbd *sdp = gl->gl_sbd;
527 unsigned int lck_flags = gh ? gh->gh_flags : 0;
528 int ret;
529
530 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
531 LM_FLAG_PRIORITY);
532 GLOCK_BUG_ON(gl, gl->gl_state == target);
533 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
534 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
535 glops->go_inval) {
536 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
537 do_error(gl, 0); /* Fail queued try locks */
538 }
539 gl->gl_req = target;
540 spin_unlock(&gl->gl_spin);
541 if (glops->go_xmote_th)
542 glops->go_xmote_th(gl);
543 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
544 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
545 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
546
547 gfs2_glock_hold(gl);
548 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
549 /* lock_dlm */
550 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
551 GLOCK_BUG_ON(gl, ret);
552 } else { /* lock_nolock */
553 finish_xmote(gl, target);
554 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
555 gfs2_glock_put(gl);
556 }
557
558 spin_lock(&gl->gl_spin);
559}
560
561/**
562 * find_first_holder - find the first "holder" gh
563 * @gl: the glock
564 */
565
566static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
567{
568 struct gfs2_holder *gh;
569
570 if (!list_empty(&gl->gl_holders)) {
571 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
572 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
573 return gh;
574 }
575 return NULL;
576}
577
578/**
579 * run_queue - do all outstanding tasks related to a glock
580 * @gl: The glock in question
581 * @nonblock: True if we must not block in run_queue
582 *
583 */
584
585static void run_queue(struct gfs2_glock *gl, const int nonblock)
586__releases(&gl->gl_spin)
587__acquires(&gl->gl_spin)
588{
589 struct gfs2_holder *gh = NULL;
590 int ret;
591
592 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
593 return;
594
595 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
596
597 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
598 gl->gl_demote_state != gl->gl_state) {
599 if (find_first_holder(gl))
600 goto out_unlock;
601 if (nonblock)
602 goto out_sched;
603 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
604 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
605 gl->gl_target = gl->gl_demote_state;
606 } else {
607 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
608 gfs2_demote_wake(gl);
609 ret = do_promote(gl);
610 if (ret == 0)
611 goto out_unlock;
612 if (ret == 2)
613 goto out;
614 gh = find_first_waiter(gl);
615 gl->gl_target = gh->gh_state;
616 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
617 do_error(gl, 0); /* Fail queued try locks */
618 }
619 do_xmote(gl, gh, gl->gl_target);
620out:
621 return;
622
623out_sched:
624 clear_bit(GLF_LOCK, &gl->gl_flags);
625 smp_mb__after_clear_bit();
626 gfs2_glock_hold(gl);
627 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
628 gfs2_glock_put_nolock(gl);
629 return;
630
631out_unlock:
632 clear_bit(GLF_LOCK, &gl->gl_flags);
633 smp_mb__after_clear_bit();
634 return;
635}
636
637static void delete_work_func(struct work_struct *work)
638{
639 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
640 struct gfs2_sbd *sdp = gl->gl_sbd;
641 struct gfs2_inode *ip;
642 struct inode *inode;
643 u64 no_addr = gl->gl_name.ln_number;
644
645 ip = gl->gl_object;
646 /* Note: Unsafe to dereference ip as we don't hold right refs/locks */
647
648 if (ip)
649 inode = gfs2_ilookup(sdp->sd_vfs, no_addr, 1);
650 else
651 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
652 if (inode && !IS_ERR(inode)) {
653 d_prune_aliases(inode);
654 iput(inode);
655 }
656 gfs2_glock_put(gl);
657}
658
659static void glock_work_func(struct work_struct *work)
660{
661 unsigned long delay = 0;
662 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
663 int drop_ref = 0;
664
665 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
666 finish_xmote(gl, gl->gl_reply);
667 drop_ref = 1;
668 }
669 spin_lock(&gl->gl_spin);
670 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
671 gl->gl_state != LM_ST_UNLOCKED &&
672 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
673 unsigned long holdtime, now = jiffies;
674
675 holdtime = gl->gl_tchange + gl->gl_hold_time;
676 if (time_before(now, holdtime))
677 delay = holdtime - now;
678
679 if (!delay) {
680 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
681 set_bit(GLF_DEMOTE, &gl->gl_flags);
682 }
683 }
684 run_queue(gl, 0);
685 spin_unlock(&gl->gl_spin);
686 if (!delay)
687 gfs2_glock_put(gl);
688 else {
689 if (gl->gl_name.ln_type != LM_TYPE_INODE)
690 delay = 0;
691 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
692 gfs2_glock_put(gl);
693 }
694 if (drop_ref)
695 gfs2_glock_put(gl);
696}
697
698/**
699 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
700 * @sdp: The GFS2 superblock
701 * @number: the lock number
702 * @glops: The glock_operations to use
703 * @create: If 0, don't create the glock if it doesn't exist
704 * @glp: the glock is returned here
705 *
706 * This does not lock a glock, just finds/creates structures for one.
707 *
708 * Returns: errno
709 */
710
711int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
712 const struct gfs2_glock_operations *glops, int create,
713 struct gfs2_glock **glp)
714{
715 struct super_block *s = sdp->sd_vfs;
716 struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
717 struct gfs2_glock *gl, *tmp;
718 unsigned int hash = gl_hash(sdp, &name);
719 struct address_space *mapping;
720 struct kmem_cache *cachep;
721
722 rcu_read_lock();
723 gl = search_bucket(hash, sdp, &name);
724 rcu_read_unlock();
725
726 *glp = gl;
727 if (gl)
728 return 0;
729 if (!create)
730 return -ENOENT;
731
732 if (glops->go_flags & GLOF_ASPACE)
733 cachep = gfs2_glock_aspace_cachep;
734 else
735 cachep = gfs2_glock_cachep;
736 gl = kmem_cache_alloc(cachep, GFP_KERNEL);
737 if (!gl)
738 return -ENOMEM;
739
740 atomic_inc(&sdp->sd_glock_disposal);
741 gl->gl_flags = 0;
742 gl->gl_name = name;
743 atomic_set(&gl->gl_ref, 1);
744 gl->gl_state = LM_ST_UNLOCKED;
745 gl->gl_target = LM_ST_UNLOCKED;
746 gl->gl_demote_state = LM_ST_EXCLUSIVE;
747 gl->gl_hash = hash;
748 gl->gl_ops = glops;
749 snprintf(gl->gl_strname, GDLM_STRNAME_BYTES, "%8x%16llx", name.ln_type, (unsigned long long)number);
750 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
751 gl->gl_lksb.sb_lvbptr = gl->gl_lvb;
752 gl->gl_tchange = jiffies;
753 gl->gl_object = NULL;
754 gl->gl_sbd = sdp;
755 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
756 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
757 INIT_WORK(&gl->gl_delete, delete_work_func);
758
759 mapping = gfs2_glock2aspace(gl);
760 if (mapping) {
761 mapping->a_ops = &gfs2_meta_aops;
762 mapping->host = s->s_bdev->bd_inode;
763 mapping->flags = 0;
764 mapping_set_gfp_mask(mapping, GFP_NOFS);
765 mapping->assoc_mapping = NULL;
766 mapping->backing_dev_info = s->s_bdi;
767 mapping->writeback_index = 0;
768 }
769
770 spin_lock_bucket(hash);
771 tmp = search_bucket(hash, sdp, &name);
772 if (tmp) {
773 spin_unlock_bucket(hash);
774 kmem_cache_free(cachep, gl);
775 atomic_dec(&sdp->sd_glock_disposal);
776 gl = tmp;
777 } else {
778 hlist_bl_add_head_rcu(&gl->gl_list, &gl_hash_table[hash]);
779 spin_unlock_bucket(hash);
780 }
781
782 *glp = gl;
783
784 return 0;
785}
786
787/**
788 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
789 * @gl: the glock
790 * @state: the state we're requesting
791 * @flags: the modifier flags
792 * @gh: the holder structure
793 *
794 */
795
796void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
797 struct gfs2_holder *gh)
798{
799 INIT_LIST_HEAD(&gh->gh_list);
800 gh->gh_gl = gl;
801 gh->gh_ip = (unsigned long)__builtin_return_address(0);
802 gh->gh_owner_pid = get_pid(task_pid(current));
803 gh->gh_state = state;
804 gh->gh_flags = flags;
805 gh->gh_error = 0;
806 gh->gh_iflags = 0;
807 gfs2_glock_hold(gl);
808}
809
810/**
811 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
812 * @state: the state we're requesting
813 * @flags: the modifier flags
814 * @gh: the holder structure
815 *
816 * Don't mess with the glock.
817 *
818 */
819
820void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
821{
822 gh->gh_state = state;
823 gh->gh_flags = flags;
824 gh->gh_iflags = 0;
825 gh->gh_ip = (unsigned long)__builtin_return_address(0);
826 if (gh->gh_owner_pid)
827 put_pid(gh->gh_owner_pid);
828 gh->gh_owner_pid = get_pid(task_pid(current));
829}
830
831/**
832 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
833 * @gh: the holder structure
834 *
835 */
836
837void gfs2_holder_uninit(struct gfs2_holder *gh)
838{
839 put_pid(gh->gh_owner_pid);
840 gfs2_glock_put(gh->gh_gl);
841 gh->gh_gl = NULL;
842 gh->gh_ip = 0;
843}
844
845/**
846 * gfs2_glock_holder_wait
847 * @word: unused
848 *
849 * This function and gfs2_glock_demote_wait both show up in the WCHAN
850 * field. Thus I've separated these otherwise identical functions in
851 * order to be more informative to the user.
852 */
853
854static int gfs2_glock_holder_wait(void *word)
855{
856 schedule();
857 return 0;
858}
859
860static int gfs2_glock_demote_wait(void *word)
861{
862 schedule();
863 return 0;
864}
865
866static void wait_on_holder(struct gfs2_holder *gh)
867{
868 unsigned long time1 = jiffies;
869
870 might_sleep();
871 wait_on_bit(&gh->gh_iflags, HIF_WAIT, gfs2_glock_holder_wait, TASK_UNINTERRUPTIBLE);
872 if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
873 /* Lengthen the minimum hold time. */
874 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
875 GL_GLOCK_HOLD_INCR,
876 GL_GLOCK_MAX_HOLD);
877}
878
879static void wait_on_demote(struct gfs2_glock *gl)
880{
881 might_sleep();
882 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, gfs2_glock_demote_wait, TASK_UNINTERRUPTIBLE);
883}
884
885/**
886 * handle_callback - process a demote request
887 * @gl: the glock
888 * @state: the state the caller wants us to change to
889 *
890 * There are only two requests that we are going to see in actual
891 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
892 */
893
894static void handle_callback(struct gfs2_glock *gl, unsigned int state,
895 unsigned long delay)
896{
897 int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
898
899 set_bit(bit, &gl->gl_flags);
900 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
901 gl->gl_demote_state = state;
902 gl->gl_demote_time = jiffies;
903 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
904 gl->gl_demote_state != state) {
905 gl->gl_demote_state = LM_ST_UNLOCKED;
906 }
907 if (gl->gl_ops->go_callback)
908 gl->gl_ops->go_callback(gl);
909 trace_gfs2_demote_rq(gl);
910}
911
912/**
913 * gfs2_glock_wait - wait on a glock acquisition
914 * @gh: the glock holder
915 *
916 * Returns: 0 on success
917 */
918
919int gfs2_glock_wait(struct gfs2_holder *gh)
920{
921 wait_on_holder(gh);
922 return gh->gh_error;
923}
924
925void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
926{
927 struct va_format vaf;
928 va_list args;
929
930 va_start(args, fmt);
931
932 if (seq) {
933 struct gfs2_glock_iter *gi = seq->private;
934 vsprintf(gi->string, fmt, args);
935 seq_printf(seq, gi->string);
936 } else {
937 vaf.fmt = fmt;
938 vaf.va = &args;
939
940 printk(KERN_ERR " %pV", &vaf);
941 }
942
943 va_end(args);
944}
945
946/**
947 * add_to_queue - Add a holder to the wait queue (but look for recursion)
948 * @gh: the holder structure to add
949 *
950 * Eventually we should move the recursive locking trap to a
951 * debugging option or something like that. This is the fast
952 * path and needs to have the minimum number of distractions.
953 *
954 */
955
956static inline void add_to_queue(struct gfs2_holder *gh)
957__releases(&gl->gl_spin)
958__acquires(&gl->gl_spin)
959{
960 struct gfs2_glock *gl = gh->gh_gl;
961 struct gfs2_sbd *sdp = gl->gl_sbd;
962 struct list_head *insert_pt = NULL;
963 struct gfs2_holder *gh2;
964 int try_lock = 0;
965
966 BUG_ON(gh->gh_owner_pid == NULL);
967 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
968 BUG();
969
970 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
971 if (test_bit(GLF_LOCK, &gl->gl_flags))
972 try_lock = 1;
973 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
974 goto fail;
975 }
976
977 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
978 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
979 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
980 goto trap_recursive;
981 if (try_lock &&
982 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) &&
983 !may_grant(gl, gh)) {
984fail:
985 gh->gh_error = GLR_TRYFAILED;
986 gfs2_holder_wake(gh);
987 return;
988 }
989 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
990 continue;
991 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
992 insert_pt = &gh2->gh_list;
993 }
994 set_bit(GLF_QUEUED, &gl->gl_flags);
995 trace_gfs2_glock_queue(gh, 1);
996 if (likely(insert_pt == NULL)) {
997 list_add_tail(&gh->gh_list, &gl->gl_holders);
998 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
999 goto do_cancel;
1000 return;
1001 }
1002 list_add_tail(&gh->gh_list, insert_pt);
1003do_cancel:
1004 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
1005 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1006 spin_unlock(&gl->gl_spin);
1007 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1008 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1009 spin_lock(&gl->gl_spin);
1010 }
1011 return;
1012
1013trap_recursive:
1014 print_symbol(KERN_ERR "original: %s\n", gh2->gh_ip);
1015 printk(KERN_ERR "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1016 printk(KERN_ERR "lock type: %d req lock state : %d\n",
1017 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1018 print_symbol(KERN_ERR "new: %s\n", gh->gh_ip);
1019 printk(KERN_ERR "pid: %d\n", pid_nr(gh->gh_owner_pid));
1020 printk(KERN_ERR "lock type: %d req lock state : %d\n",
1021 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1022 __dump_glock(NULL, gl);
1023 BUG();
1024}
1025
1026/**
1027 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1028 * @gh: the holder structure
1029 *
1030 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1031 *
1032 * Returns: 0, GLR_TRYFAILED, or errno on failure
1033 */
1034
1035int gfs2_glock_nq(struct gfs2_holder *gh)
1036{
1037 struct gfs2_glock *gl = gh->gh_gl;
1038 struct gfs2_sbd *sdp = gl->gl_sbd;
1039 int error = 0;
1040
1041 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1042 return -EIO;
1043
1044 if (test_bit(GLF_LRU, &gl->gl_flags))
1045 gfs2_glock_remove_from_lru(gl);
1046
1047 spin_lock(&gl->gl_spin);
1048 add_to_queue(gh);
1049 if ((LM_FLAG_NOEXP & gh->gh_flags) &&
1050 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1051 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1052 run_queue(gl, 1);
1053 spin_unlock(&gl->gl_spin);
1054
1055 if (!(gh->gh_flags & GL_ASYNC))
1056 error = gfs2_glock_wait(gh);
1057
1058 return error;
1059}
1060
1061/**
1062 * gfs2_glock_poll - poll to see if an async request has been completed
1063 * @gh: the holder
1064 *
1065 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1066 */
1067
1068int gfs2_glock_poll(struct gfs2_holder *gh)
1069{
1070 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1071}
1072
1073/**
1074 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1075 * @gh: the glock holder
1076 *
1077 */
1078
1079void gfs2_glock_dq(struct gfs2_holder *gh)
1080{
1081 struct gfs2_glock *gl = gh->gh_gl;
1082 const struct gfs2_glock_operations *glops = gl->gl_ops;
1083 unsigned delay = 0;
1084 int fast_path = 0;
1085
1086 spin_lock(&gl->gl_spin);
1087 if (gh->gh_flags & GL_NOCACHE)
1088 handle_callback(gl, LM_ST_UNLOCKED, 0);
1089
1090 list_del_init(&gh->gh_list);
1091 if (find_first_holder(gl) == NULL) {
1092 if (glops->go_unlock) {
1093 GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1094 spin_unlock(&gl->gl_spin);
1095 glops->go_unlock(gh);
1096 spin_lock(&gl->gl_spin);
1097 clear_bit(GLF_LOCK, &gl->gl_flags);
1098 }
1099 if (list_empty(&gl->gl_holders) &&
1100 !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1101 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1102 fast_path = 1;
1103 }
1104 if (!test_bit(GLF_LFLUSH, &gl->gl_flags))
1105 __gfs2_glock_schedule_for_reclaim(gl);
1106 trace_gfs2_glock_queue(gh, 0);
1107 spin_unlock(&gl->gl_spin);
1108 if (likely(fast_path))
1109 return;
1110
1111 gfs2_glock_hold(gl);
1112 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1113 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1114 gl->gl_name.ln_type == LM_TYPE_INODE)
1115 delay = gl->gl_hold_time;
1116 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1117 gfs2_glock_put(gl);
1118}
1119
1120void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1121{
1122 struct gfs2_glock *gl = gh->gh_gl;
1123 gfs2_glock_dq(gh);
1124 wait_on_demote(gl);
1125}
1126
1127/**
1128 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1129 * @gh: the holder structure
1130 *
1131 */
1132
1133void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1134{
1135 gfs2_glock_dq(gh);
1136 gfs2_holder_uninit(gh);
1137}
1138
1139/**
1140 * gfs2_glock_nq_num - acquire a glock based on lock number
1141 * @sdp: the filesystem
1142 * @number: the lock number
1143 * @glops: the glock operations for the type of glock
1144 * @state: the state to acquire the glock in
1145 * @flags: modifier flags for the acquisition
1146 * @gh: the struct gfs2_holder
1147 *
1148 * Returns: errno
1149 */
1150
1151int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1152 const struct gfs2_glock_operations *glops,
1153 unsigned int state, int flags, struct gfs2_holder *gh)
1154{
1155 struct gfs2_glock *gl;
1156 int error;
1157
1158 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1159 if (!error) {
1160 error = gfs2_glock_nq_init(gl, state, flags, gh);
1161 gfs2_glock_put(gl);
1162 }
1163
1164 return error;
1165}
1166
1167/**
1168 * glock_compare - Compare two struct gfs2_glock structures for sorting
1169 * @arg_a: the first structure
1170 * @arg_b: the second structure
1171 *
1172 */
1173
1174static int glock_compare(const void *arg_a, const void *arg_b)
1175{
1176 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1177 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1178 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1179 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1180
1181 if (a->ln_number > b->ln_number)
1182 return 1;
1183 if (a->ln_number < b->ln_number)
1184 return -1;
1185 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1186 return 0;
1187}
1188
1189/**
1190 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1191 * @num_gh: the number of structures
1192 * @ghs: an array of struct gfs2_holder structures
1193 *
1194 * Returns: 0 on success (all glocks acquired),
1195 * errno on failure (no glocks acquired)
1196 */
1197
1198static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1199 struct gfs2_holder **p)
1200{
1201 unsigned int x;
1202 int error = 0;
1203
1204 for (x = 0; x < num_gh; x++)
1205 p[x] = &ghs[x];
1206
1207 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1208
1209 for (x = 0; x < num_gh; x++) {
1210 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1211
1212 error = gfs2_glock_nq(p[x]);
1213 if (error) {
1214 while (x--)
1215 gfs2_glock_dq(p[x]);
1216 break;
1217 }
1218 }
1219
1220 return error;
1221}
1222
1223/**
1224 * gfs2_glock_nq_m - acquire multiple glocks
1225 * @num_gh: the number of structures
1226 * @ghs: an array of struct gfs2_holder structures
1227 *
1228 *
1229 * Returns: 0 on success (all glocks acquired),
1230 * errno on failure (no glocks acquired)
1231 */
1232
1233int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1234{
1235 struct gfs2_holder *tmp[4];
1236 struct gfs2_holder **pph = tmp;
1237 int error = 0;
1238
1239 switch(num_gh) {
1240 case 0:
1241 return 0;
1242 case 1:
1243 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1244 return gfs2_glock_nq(ghs);
1245 default:
1246 if (num_gh <= 4)
1247 break;
1248 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1249 if (!pph)
1250 return -ENOMEM;
1251 }
1252
1253 error = nq_m_sync(num_gh, ghs, pph);
1254
1255 if (pph != tmp)
1256 kfree(pph);
1257
1258 return error;
1259}
1260
1261/**
1262 * gfs2_glock_dq_m - release multiple glocks
1263 * @num_gh: the number of structures
1264 * @ghs: an array of struct gfs2_holder structures
1265 *
1266 */
1267
1268void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1269{
1270 while (num_gh--)
1271 gfs2_glock_dq(&ghs[num_gh]);
1272}
1273
1274/**
1275 * gfs2_glock_dq_uninit_m - release multiple glocks
1276 * @num_gh: the number of structures
1277 * @ghs: an array of struct gfs2_holder structures
1278 *
1279 */
1280
1281void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1282{
1283 while (num_gh--)
1284 gfs2_glock_dq_uninit(&ghs[num_gh]);
1285}
1286
1287void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1288{
1289 unsigned long delay = 0;
1290 unsigned long holdtime;
1291 unsigned long now = jiffies;
1292
1293 gfs2_glock_hold(gl);
1294 holdtime = gl->gl_tchange + gl->gl_hold_time;
1295 if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1296 gl->gl_name.ln_type == LM_TYPE_INODE) {
1297 if (time_before(now, holdtime))
1298 delay = holdtime - now;
1299 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1300 delay = gl->gl_hold_time;
1301 }
1302
1303 spin_lock(&gl->gl_spin);
1304 handle_callback(gl, state, delay);
1305 spin_unlock(&gl->gl_spin);
1306 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1307 gfs2_glock_put(gl);
1308}
1309
1310/**
1311 * gfs2_should_freeze - Figure out if glock should be frozen
1312 * @gl: The glock in question
1313 *
1314 * Glocks are not frozen if (a) the result of the dlm operation is
1315 * an error, (b) the locking operation was an unlock operation or
1316 * (c) if there is a "noexp" flagged request anywhere in the queue
1317 *
1318 * Returns: 1 if freezing should occur, 0 otherwise
1319 */
1320
1321static int gfs2_should_freeze(const struct gfs2_glock *gl)
1322{
1323 const struct gfs2_holder *gh;
1324
1325 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1326 return 0;
1327 if (gl->gl_target == LM_ST_UNLOCKED)
1328 return 0;
1329
1330 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1331 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1332 continue;
1333 if (LM_FLAG_NOEXP & gh->gh_flags)
1334 return 0;
1335 }
1336
1337 return 1;
1338}
1339
1340/**
1341 * gfs2_glock_complete - Callback used by locking
1342 * @gl: Pointer to the glock
1343 * @ret: The return value from the dlm
1344 *
1345 * The gl_reply field is under the gl_spin lock so that it is ok
1346 * to use a bitfield shared with other glock state fields.
1347 */
1348
1349void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1350{
1351 struct lm_lockstruct *ls = &gl->gl_sbd->sd_lockstruct;
1352
1353 spin_lock(&gl->gl_spin);
1354 gl->gl_reply = ret;
1355
1356 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_flags))) {
1357 if (gfs2_should_freeze(gl)) {
1358 set_bit(GLF_FROZEN, &gl->gl_flags);
1359 spin_unlock(&gl->gl_spin);
1360 return;
1361 }
1362 }
1363
1364 spin_unlock(&gl->gl_spin);
1365 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1366 smp_wmb();
1367 gfs2_glock_hold(gl);
1368 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1369 gfs2_glock_put(gl);
1370}
1371
1372
1373static int gfs2_shrink_glock_memory(struct shrinker *shrink,
1374 struct shrink_control *sc)
1375{
1376 struct gfs2_glock *gl;
1377 int may_demote;
1378 int nr_skipped = 0;
1379 int nr = sc->nr_to_scan;
1380 gfp_t gfp_mask = sc->gfp_mask;
1381 LIST_HEAD(skipped);
1382
1383 if (nr == 0)
1384 goto out;
1385
1386 if (!(gfp_mask & __GFP_FS))
1387 return -1;
1388
1389 spin_lock(&lru_lock);
1390 while(nr && !list_empty(&lru_list)) {
1391 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1392 list_del_init(&gl->gl_lru);
1393 clear_bit(GLF_LRU, &gl->gl_flags);
1394 atomic_dec(&lru_count);
1395
1396 /* Test for being demotable */
1397 if (!test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1398 gfs2_glock_hold(gl);
1399 spin_unlock(&lru_lock);
1400 spin_lock(&gl->gl_spin);
1401 may_demote = demote_ok(gl);
1402 if (may_demote) {
1403 handle_callback(gl, LM_ST_UNLOCKED, 0);
1404 nr--;
1405 }
1406 clear_bit(GLF_LOCK, &gl->gl_flags);
1407 smp_mb__after_clear_bit();
1408 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1409 gfs2_glock_put_nolock(gl);
1410 spin_unlock(&gl->gl_spin);
1411 spin_lock(&lru_lock);
1412 continue;
1413 }
1414 nr_skipped++;
1415 list_add(&gl->gl_lru, &skipped);
1416 set_bit(GLF_LRU, &gl->gl_flags);
1417 }
1418 list_splice(&skipped, &lru_list);
1419 atomic_add(nr_skipped, &lru_count);
1420 spin_unlock(&lru_lock);
1421out:
1422 return (atomic_read(&lru_count) / 100) * sysctl_vfs_cache_pressure;
1423}
1424
1425static struct shrinker glock_shrinker = {
1426 .shrink = gfs2_shrink_glock_memory,
1427 .seeks = DEFAULT_SEEKS,
1428};
1429
1430/**
1431 * examine_bucket - Call a function for glock in a hash bucket
1432 * @examiner: the function
1433 * @sdp: the filesystem
1434 * @bucket: the bucket
1435 *
1436 */
1437
1438static void examine_bucket(glock_examiner examiner, const struct gfs2_sbd *sdp,
1439 unsigned int hash)
1440{
1441 struct gfs2_glock *gl;
1442 struct hlist_bl_head *head = &gl_hash_table[hash];
1443 struct hlist_bl_node *pos;
1444
1445 rcu_read_lock();
1446 hlist_bl_for_each_entry_rcu(gl, pos, head, gl_list) {
1447 if ((gl->gl_sbd == sdp) && atomic_read(&gl->gl_ref))
1448 examiner(gl);
1449 }
1450 rcu_read_unlock();
1451 cond_resched();
1452}
1453
1454static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1455{
1456 unsigned x;
1457
1458 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1459 examine_bucket(examiner, sdp, x);
1460}
1461
1462
1463/**
1464 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1465 * @gl: The glock to thaw
1466 *
1467 * N.B. When we freeze a glock, we leave a ref to the glock outstanding,
1468 * so this has to result in the ref count being dropped by one.
1469 */
1470
1471static void thaw_glock(struct gfs2_glock *gl)
1472{
1473 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1474 return;
1475 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1476 gfs2_glock_hold(gl);
1477 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1478 gfs2_glock_put(gl);
1479}
1480
1481/**
1482 * clear_glock - look at a glock and see if we can free it from glock cache
1483 * @gl: the glock to look at
1484 *
1485 */
1486
1487static void clear_glock(struct gfs2_glock *gl)
1488{
1489 gfs2_glock_remove_from_lru(gl);
1490
1491 spin_lock(&gl->gl_spin);
1492 if (gl->gl_state != LM_ST_UNLOCKED)
1493 handle_callback(gl, LM_ST_UNLOCKED, 0);
1494 spin_unlock(&gl->gl_spin);
1495 gfs2_glock_hold(gl);
1496 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1497 gfs2_glock_put(gl);
1498}
1499
1500/**
1501 * gfs2_glock_thaw - Thaw any frozen glocks
1502 * @sdp: The super block
1503 *
1504 */
1505
1506void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1507{
1508 glock_hash_walk(thaw_glock, sdp);
1509}
1510
1511static int dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1512{
1513 int ret;
1514 spin_lock(&gl->gl_spin);
1515 ret = __dump_glock(seq, gl);
1516 spin_unlock(&gl->gl_spin);
1517 return ret;
1518}
1519
1520static void dump_glock_func(struct gfs2_glock *gl)
1521{
1522 dump_glock(NULL, gl);
1523}
1524
1525/**
1526 * gfs2_gl_hash_clear - Empty out the glock hash table
1527 * @sdp: the filesystem
1528 * @wait: wait until it's all gone
1529 *
1530 * Called when unmounting the filesystem.
1531 */
1532
1533void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1534{
1535 glock_hash_walk(clear_glock, sdp);
1536 flush_workqueue(glock_workqueue);
1537 wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
1538 glock_hash_walk(dump_glock_func, sdp);
1539}
1540
1541void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1542{
1543 struct gfs2_glock *gl = ip->i_gl;
1544 int ret;
1545
1546 ret = gfs2_truncatei_resume(ip);
1547 gfs2_assert_withdraw(gl->gl_sbd, ret == 0);
1548
1549 spin_lock(&gl->gl_spin);
1550 clear_bit(GLF_LOCK, &gl->gl_flags);
1551 run_queue(gl, 1);
1552 spin_unlock(&gl->gl_spin);
1553}
1554
1555static const char *state2str(unsigned state)
1556{
1557 switch(state) {
1558 case LM_ST_UNLOCKED:
1559 return "UN";
1560 case LM_ST_SHARED:
1561 return "SH";
1562 case LM_ST_DEFERRED:
1563 return "DF";
1564 case LM_ST_EXCLUSIVE:
1565 return "EX";
1566 }
1567 return "??";
1568}
1569
1570static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1571{
1572 char *p = buf;
1573 if (flags & LM_FLAG_TRY)
1574 *p++ = 't';
1575 if (flags & LM_FLAG_TRY_1CB)
1576 *p++ = 'T';
1577 if (flags & LM_FLAG_NOEXP)
1578 *p++ = 'e';
1579 if (flags & LM_FLAG_ANY)
1580 *p++ = 'A';
1581 if (flags & LM_FLAG_PRIORITY)
1582 *p++ = 'p';
1583 if (flags & GL_ASYNC)
1584 *p++ = 'a';
1585 if (flags & GL_EXACT)
1586 *p++ = 'E';
1587 if (flags & GL_NOCACHE)
1588 *p++ = 'c';
1589 if (test_bit(HIF_HOLDER, &iflags))
1590 *p++ = 'H';
1591 if (test_bit(HIF_WAIT, &iflags))
1592 *p++ = 'W';
1593 if (test_bit(HIF_FIRST, &iflags))
1594 *p++ = 'F';
1595 *p = 0;
1596 return buf;
1597}
1598
1599/**
1600 * dump_holder - print information about a glock holder
1601 * @seq: the seq_file struct
1602 * @gh: the glock holder
1603 *
1604 * Returns: 0 on success, -ENOBUFS when we run out of space
1605 */
1606
1607static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1608{
1609 struct task_struct *gh_owner = NULL;
1610 char flags_buf[32];
1611
1612 if (gh->gh_owner_pid)
1613 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1614 gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1615 state2str(gh->gh_state),
1616 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1617 gh->gh_error,
1618 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1619 gh_owner ? gh_owner->comm : "(ended)",
1620 (void *)gh->gh_ip);
1621 return 0;
1622}
1623
1624static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1625{
1626 const unsigned long *gflags = &gl->gl_flags;
1627 char *p = buf;
1628
1629 if (test_bit(GLF_LOCK, gflags))
1630 *p++ = 'l';
1631 if (test_bit(GLF_DEMOTE, gflags))
1632 *p++ = 'D';
1633 if (test_bit(GLF_PENDING_DEMOTE, gflags))
1634 *p++ = 'd';
1635 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1636 *p++ = 'p';
1637 if (test_bit(GLF_DIRTY, gflags))
1638 *p++ = 'y';
1639 if (test_bit(GLF_LFLUSH, gflags))
1640 *p++ = 'f';
1641 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1642 *p++ = 'i';
1643 if (test_bit(GLF_REPLY_PENDING, gflags))
1644 *p++ = 'r';
1645 if (test_bit(GLF_INITIAL, gflags))
1646 *p++ = 'I';
1647 if (test_bit(GLF_FROZEN, gflags))
1648 *p++ = 'F';
1649 if (test_bit(GLF_QUEUED, gflags))
1650 *p++ = 'q';
1651 if (test_bit(GLF_LRU, gflags))
1652 *p++ = 'L';
1653 if (gl->gl_object)
1654 *p++ = 'o';
1655 *p = 0;
1656 return buf;
1657}
1658
1659/**
1660 * __dump_glock - print information about a glock
1661 * @seq: The seq_file struct
1662 * @gl: the glock
1663 *
1664 * The file format is as follows:
1665 * One line per object, capital letters are used to indicate objects
1666 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1667 * other objects are indented by a single space and follow the glock to
1668 * which they are related. Fields are indicated by lower case letters
1669 * followed by a colon and the field value, except for strings which are in
1670 * [] so that its possible to see if they are composed of spaces for
1671 * example. The field's are n = number (id of the object), f = flags,
1672 * t = type, s = state, r = refcount, e = error, p = pid.
1673 *
1674 * Returns: 0 on success, -ENOBUFS when we run out of space
1675 */
1676
1677static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1678{
1679 const struct gfs2_glock_operations *glops = gl->gl_ops;
1680 unsigned long long dtime;
1681 const struct gfs2_holder *gh;
1682 char gflags_buf[32];
1683 int error = 0;
1684
1685 dtime = jiffies - gl->gl_demote_time;
1686 dtime *= 1000000/HZ; /* demote time in uSec */
1687 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1688 dtime = 0;
1689 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",
1690 state2str(gl->gl_state),
1691 gl->gl_name.ln_type,
1692 (unsigned long long)gl->gl_name.ln_number,
1693 gflags2str(gflags_buf, gl),
1694 state2str(gl->gl_target),
1695 state2str(gl->gl_demote_state), dtime,
1696 atomic_read(&gl->gl_ail_count),
1697 atomic_read(&gl->gl_revokes),
1698 atomic_read(&gl->gl_ref), gl->gl_hold_time);
1699
1700 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1701 error = dump_holder(seq, gh);
1702 if (error)
1703 goto out;
1704 }
1705 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1706 error = glops->go_dump(seq, gl);
1707out:
1708 return error;
1709}
1710
1711
1712
1713
1714int __init gfs2_glock_init(void)
1715{
1716 unsigned i;
1717 for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
1718 INIT_HLIST_BL_HEAD(&gl_hash_table[i]);
1719 }
1720
1721 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1722 WQ_HIGHPRI | WQ_FREEZABLE, 0);
1723 if (IS_ERR(glock_workqueue))
1724 return PTR_ERR(glock_workqueue);
1725 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1726 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1727 0);
1728 if (IS_ERR(gfs2_delete_workqueue)) {
1729 destroy_workqueue(glock_workqueue);
1730 return PTR_ERR(gfs2_delete_workqueue);
1731 }
1732
1733 register_shrinker(&glock_shrinker);
1734
1735 return 0;
1736}
1737
1738void gfs2_glock_exit(void)
1739{
1740 unregister_shrinker(&glock_shrinker);
1741 destroy_workqueue(glock_workqueue);
1742 destroy_workqueue(gfs2_delete_workqueue);
1743}
1744
1745static inline struct gfs2_glock *glock_hash_chain(unsigned hash)
1746{
1747 return hlist_bl_entry(hlist_bl_first_rcu(&gl_hash_table[hash]),
1748 struct gfs2_glock, gl_list);
1749}
1750
1751static inline struct gfs2_glock *glock_hash_next(struct gfs2_glock *gl)
1752{
1753 return hlist_bl_entry(rcu_dereference(gl->gl_list.next),
1754 struct gfs2_glock, gl_list);
1755}
1756
1757static int gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1758{
1759 struct gfs2_glock *gl;
1760
1761 do {
1762 gl = gi->gl;
1763 if (gl) {
1764 gi->gl = glock_hash_next(gl);
1765 } else {
1766 gi->gl = glock_hash_chain(gi->hash);
1767 }
1768 while (gi->gl == NULL) {
1769 gi->hash++;
1770 if (gi->hash >= GFS2_GL_HASH_SIZE) {
1771 rcu_read_unlock();
1772 return 1;
1773 }
1774 gi->gl = glock_hash_chain(gi->hash);
1775 }
1776 /* Skip entries for other sb and dead entries */
1777 } while (gi->sdp != gi->gl->gl_sbd || atomic_read(&gi->gl->gl_ref) == 0);
1778
1779 return 0;
1780}
1781
1782static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1783{
1784 struct gfs2_glock_iter *gi = seq->private;
1785 loff_t n = *pos;
1786
1787 gi->hash = 0;
1788 rcu_read_lock();
1789
1790 do {
1791 if (gfs2_glock_iter_next(gi))
1792 return NULL;
1793 } while (n--);
1794
1795 return gi->gl;
1796}
1797
1798static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1799 loff_t *pos)
1800{
1801 struct gfs2_glock_iter *gi = seq->private;
1802
1803 (*pos)++;
1804
1805 if (gfs2_glock_iter_next(gi))
1806 return NULL;
1807
1808 return gi->gl;
1809}
1810
1811static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1812{
1813 struct gfs2_glock_iter *gi = seq->private;
1814
1815 if (gi->gl)
1816 rcu_read_unlock();
1817 gi->gl = NULL;
1818}
1819
1820static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1821{
1822 return dump_glock(seq, iter_ptr);
1823}
1824
1825static const struct seq_operations gfs2_glock_seq_ops = {
1826 .start = gfs2_glock_seq_start,
1827 .next = gfs2_glock_seq_next,
1828 .stop = gfs2_glock_seq_stop,
1829 .show = gfs2_glock_seq_show,
1830};
1831
1832static int gfs2_debugfs_open(struct inode *inode, struct file *file)
1833{
1834 int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1835 sizeof(struct gfs2_glock_iter));
1836 if (ret == 0) {
1837 struct seq_file *seq = file->private_data;
1838 struct gfs2_glock_iter *gi = seq->private;
1839 gi->sdp = inode->i_private;
1840 }
1841 return ret;
1842}
1843
1844static const struct file_operations gfs2_debug_fops = {
1845 .owner = THIS_MODULE,
1846 .open = gfs2_debugfs_open,
1847 .read = seq_read,
1848 .llseek = seq_lseek,
1849 .release = seq_release_private,
1850};
1851
1852int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
1853{
1854 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
1855 if (!sdp->debugfs_dir)
1856 return -ENOMEM;
1857 sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
1858 S_IFREG | S_IRUGO,
1859 sdp->debugfs_dir, sdp,
1860 &gfs2_debug_fops);
1861 if (!sdp->debugfs_dentry_glocks)
1862 return -ENOMEM;
1863
1864 return 0;
1865}
1866
1867void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
1868{
1869 if (sdp && sdp->debugfs_dir) {
1870 if (sdp->debugfs_dentry_glocks) {
1871 debugfs_remove(sdp->debugfs_dentry_glocks);
1872 sdp->debugfs_dentry_glocks = NULL;
1873 }
1874 debugfs_remove(sdp->debugfs_dir);
1875 sdp->debugfs_dir = NULL;
1876 }
1877}
1878
1879int gfs2_register_debugfs(void)
1880{
1881 gfs2_root = debugfs_create_dir("gfs2", NULL);
1882 return gfs2_root ? 0 : -ENOMEM;
1883}
1884
1885void gfs2_unregister_debugfs(void)
1886{
1887 debugfs_remove(gfs2_root);
1888 gfs2_root = NULL;
1889}
1/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
4 *
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 */
9
10#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11
12#include <linux/sched.h>
13#include <linux/slab.h>
14#include <linux/spinlock.h>
15#include <linux/buffer_head.h>
16#include <linux/delay.h>
17#include <linux/sort.h>
18#include <linux/hash.h>
19#include <linux/jhash.h>
20#include <linux/kallsyms.h>
21#include <linux/gfs2_ondisk.h>
22#include <linux/list.h>
23#include <linux/wait.h>
24#include <linux/module.h>
25#include <linux/uaccess.h>
26#include <linux/seq_file.h>
27#include <linux/debugfs.h>
28#include <linux/kthread.h>
29#include <linux/freezer.h>
30#include <linux/workqueue.h>
31#include <linux/jiffies.h>
32#include <linux/rcupdate.h>
33#include <linux/rculist_bl.h>
34#include <linux/bit_spinlock.h>
35#include <linux/percpu.h>
36#include <linux/list_sort.h>
37#include <linux/lockref.h>
38#include <linux/rhashtable.h>
39
40#include "gfs2.h"
41#include "incore.h"
42#include "glock.h"
43#include "glops.h"
44#include "inode.h"
45#include "lops.h"
46#include "meta_io.h"
47#include "quota.h"
48#include "super.h"
49#include "util.h"
50#include "bmap.h"
51#define CREATE_TRACE_POINTS
52#include "trace_gfs2.h"
53
54struct gfs2_glock_iter {
55 struct gfs2_sbd *sdp; /* incore superblock */
56 struct rhashtable_iter hti; /* rhashtable iterator */
57 struct gfs2_glock *gl; /* current glock struct */
58 loff_t last_pos; /* last position */
59};
60
61typedef void (*glock_examiner) (struct gfs2_glock * gl);
62
63static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
64
65static struct dentry *gfs2_root;
66static struct workqueue_struct *glock_workqueue;
67struct workqueue_struct *gfs2_delete_workqueue;
68static LIST_HEAD(lru_list);
69static atomic_t lru_count = ATOMIC_INIT(0);
70static DEFINE_SPINLOCK(lru_lock);
71
72#define GFS2_GL_HASH_SHIFT 15
73#define GFS2_GL_HASH_SIZE BIT(GFS2_GL_HASH_SHIFT)
74
75static const struct rhashtable_params ht_parms = {
76 .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4,
77 .key_len = offsetofend(struct lm_lockname, ln_type),
78 .key_offset = offsetof(struct gfs2_glock, gl_name),
79 .head_offset = offsetof(struct gfs2_glock, gl_node),
80};
81
82static struct rhashtable gl_hash_table;
83
84#define GLOCK_WAIT_TABLE_BITS 12
85#define GLOCK_WAIT_TABLE_SIZE (1 << GLOCK_WAIT_TABLE_BITS)
86static wait_queue_head_t glock_wait_table[GLOCK_WAIT_TABLE_SIZE] __cacheline_aligned;
87
88struct wait_glock_queue {
89 struct lm_lockname *name;
90 wait_queue_entry_t wait;
91};
92
93static int glock_wake_function(wait_queue_entry_t *wait, unsigned int mode,
94 int sync, void *key)
95{
96 struct wait_glock_queue *wait_glock =
97 container_of(wait, struct wait_glock_queue, wait);
98 struct lm_lockname *wait_name = wait_glock->name;
99 struct lm_lockname *wake_name = key;
100
101 if (wake_name->ln_sbd != wait_name->ln_sbd ||
102 wake_name->ln_number != wait_name->ln_number ||
103 wake_name->ln_type != wait_name->ln_type)
104 return 0;
105 return autoremove_wake_function(wait, mode, sync, key);
106}
107
108static wait_queue_head_t *glock_waitqueue(struct lm_lockname *name)
109{
110 u32 hash = jhash2((u32 *)name, sizeof(*name) / 4, 0);
111
112 return glock_wait_table + hash_32(hash, GLOCK_WAIT_TABLE_BITS);
113}
114
115/**
116 * wake_up_glock - Wake up waiters on a glock
117 * @gl: the glock
118 */
119static void wake_up_glock(struct gfs2_glock *gl)
120{
121 wait_queue_head_t *wq = glock_waitqueue(&gl->gl_name);
122
123 if (waitqueue_active(wq))
124 __wake_up(wq, TASK_NORMAL, 1, &gl->gl_name);
125}
126
127static void gfs2_glock_dealloc(struct rcu_head *rcu)
128{
129 struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
130
131 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
132 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
133 } else {
134 kfree(gl->gl_lksb.sb_lvbptr);
135 kmem_cache_free(gfs2_glock_cachep, gl);
136 }
137}
138
139void gfs2_glock_free(struct gfs2_glock *gl)
140{
141 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
142
143 rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
144 smp_mb();
145 wake_up_glock(gl);
146 call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
147 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
148 wake_up(&sdp->sd_glock_wait);
149}
150
151/**
152 * gfs2_glock_hold() - increment reference count on glock
153 * @gl: The glock to hold
154 *
155 */
156
157void gfs2_glock_hold(struct gfs2_glock *gl)
158{
159 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
160 lockref_get(&gl->gl_lockref);
161}
162
163/**
164 * demote_ok - Check to see if it's ok to unlock a glock
165 * @gl: the glock
166 *
167 * Returns: 1 if it's ok
168 */
169
170static int demote_ok(const struct gfs2_glock *gl)
171{
172 const struct gfs2_glock_operations *glops = gl->gl_ops;
173
174 if (gl->gl_state == LM_ST_UNLOCKED)
175 return 0;
176 if (!list_empty(&gl->gl_holders))
177 return 0;
178 if (glops->go_demote_ok)
179 return glops->go_demote_ok(gl);
180 return 1;
181}
182
183
184void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
185{
186 spin_lock(&lru_lock);
187
188 if (!list_empty(&gl->gl_lru))
189 list_del_init(&gl->gl_lru);
190 else
191 atomic_inc(&lru_count);
192
193 list_add_tail(&gl->gl_lru, &lru_list);
194 set_bit(GLF_LRU, &gl->gl_flags);
195 spin_unlock(&lru_lock);
196}
197
198static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
199{
200 if (!(gl->gl_ops->go_flags & GLOF_LRU))
201 return;
202
203 spin_lock(&lru_lock);
204 if (!list_empty(&gl->gl_lru)) {
205 list_del_init(&gl->gl_lru);
206 atomic_dec(&lru_count);
207 clear_bit(GLF_LRU, &gl->gl_flags);
208 }
209 spin_unlock(&lru_lock);
210}
211
212/*
213 * Enqueue the glock on the work queue. Passes one glock reference on to the
214 * work queue.
215 */
216static void __gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
217 if (!queue_delayed_work(glock_workqueue, &gl->gl_work, delay)) {
218 /*
219 * We are holding the lockref spinlock, and the work was still
220 * queued above. The queued work (glock_work_func) takes that
221 * spinlock before dropping its glock reference(s), so it
222 * cannot have dropped them in the meantime.
223 */
224 GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2);
225 gl->gl_lockref.count--;
226 }
227}
228
229static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
230 spin_lock(&gl->gl_lockref.lock);
231 __gfs2_glock_queue_work(gl, delay);
232 spin_unlock(&gl->gl_lockref.lock);
233}
234
235static void __gfs2_glock_put(struct gfs2_glock *gl)
236{
237 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
238 struct address_space *mapping = gfs2_glock2aspace(gl);
239
240 lockref_mark_dead(&gl->gl_lockref);
241
242 gfs2_glock_remove_from_lru(gl);
243 spin_unlock(&gl->gl_lockref.lock);
244 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
245 GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
246 trace_gfs2_glock_put(gl);
247 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
248}
249
250/*
251 * Cause the glock to be put in work queue context.
252 */
253void gfs2_glock_queue_put(struct gfs2_glock *gl)
254{
255 gfs2_glock_queue_work(gl, 0);
256}
257
258/**
259 * gfs2_glock_put() - Decrement reference count on glock
260 * @gl: The glock to put
261 *
262 */
263
264void gfs2_glock_put(struct gfs2_glock *gl)
265{
266 if (lockref_put_or_lock(&gl->gl_lockref))
267 return;
268
269 __gfs2_glock_put(gl);
270}
271
272/**
273 * may_grant - check if its ok to grant a new lock
274 * @gl: The glock
275 * @gh: The lock request which we wish to grant
276 *
277 * Returns: true if its ok to grant the lock
278 */
279
280static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
281{
282 const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
283 if ((gh->gh_state == LM_ST_EXCLUSIVE ||
284 gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
285 return 0;
286 if (gl->gl_state == gh->gh_state)
287 return 1;
288 if (gh->gh_flags & GL_EXACT)
289 return 0;
290 if (gl->gl_state == LM_ST_EXCLUSIVE) {
291 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
292 return 1;
293 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
294 return 1;
295 }
296 if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
297 return 1;
298 return 0;
299}
300
301static void gfs2_holder_wake(struct gfs2_holder *gh)
302{
303 clear_bit(HIF_WAIT, &gh->gh_iflags);
304 smp_mb__after_atomic();
305 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
306}
307
308/**
309 * do_error - Something unexpected has happened during a lock request
310 *
311 */
312
313static void do_error(struct gfs2_glock *gl, const int ret)
314{
315 struct gfs2_holder *gh, *tmp;
316
317 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
318 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
319 continue;
320 if (ret & LM_OUT_ERROR)
321 gh->gh_error = -EIO;
322 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
323 gh->gh_error = GLR_TRYFAILED;
324 else
325 continue;
326 list_del_init(&gh->gh_list);
327 trace_gfs2_glock_queue(gh, 0);
328 gfs2_holder_wake(gh);
329 }
330}
331
332/**
333 * do_promote - promote as many requests as possible on the current queue
334 * @gl: The glock
335 *
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.
338 */
339
340static int do_promote(struct gfs2_glock *gl)
341__releases(&gl->gl_lockref.lock)
342__acquires(&gl->gl_lockref.lock)
343{
344 const struct gfs2_glock_operations *glops = gl->gl_ops;
345 struct gfs2_holder *gh, *tmp;
346 int ret;
347
348restart:
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_lockref.lock);
356 /* FIXME: eliminate this eventually */
357 ret = glops->go_lock(gh);
358 spin_lock(&gl->gl_lockref.lock);
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;
367 }
368 set_bit(HIF_HOLDER, &gh->gh_iflags);
369 trace_gfs2_promote(gh, 1);
370 gfs2_holder_wake(gh);
371 goto restart;
372 }
373 set_bit(HIF_HOLDER, &gh->gh_iflags);
374 trace_gfs2_promote(gh, 0);
375 gfs2_holder_wake(gh);
376 continue;
377 }
378 if (gh->gh_list.prev == &gl->gl_holders)
379 return 1;
380 do_error(gl, 0);
381 break;
382 }
383 return 0;
384}
385
386/**
387 * find_first_waiter - find the first gh that's waiting for the glock
388 * @gl: the glock
389 */
390
391static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
392{
393 struct gfs2_holder *gh;
394
395 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
396 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
397 return gh;
398 }
399 return NULL;
400}
401
402/**
403 * state_change - record that the glock is now in a different state
404 * @gl: the glock
405 * @new_state the new state
406 *
407 */
408
409static void state_change(struct gfs2_glock *gl, unsigned int new_state)
410{
411 int held1, held2;
412
413 held1 = (gl->gl_state != LM_ST_UNLOCKED);
414 held2 = (new_state != LM_ST_UNLOCKED);
415
416 if (held1 != held2) {
417 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
418 if (held2)
419 gl->gl_lockref.count++;
420 else
421 gl->gl_lockref.count--;
422 }
423 if (held1 && held2 && list_empty(&gl->gl_holders))
424 clear_bit(GLF_QUEUED, &gl->gl_flags);
425
426 if (new_state != gl->gl_target)
427 /* shorten our minimum hold time */
428 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
429 GL_GLOCK_MIN_HOLD);
430 gl->gl_state = new_state;
431 gl->gl_tchange = jiffies;
432}
433
434static void gfs2_demote_wake(struct gfs2_glock *gl)
435{
436 gl->gl_demote_state = LM_ST_EXCLUSIVE;
437 clear_bit(GLF_DEMOTE, &gl->gl_flags);
438 smp_mb__after_atomic();
439 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
440}
441
442/**
443 * finish_xmote - The DLM has replied to one of our lock requests
444 * @gl: The glock
445 * @ret: The status from the DLM
446 *
447 */
448
449static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
450{
451 const struct gfs2_glock_operations *glops = gl->gl_ops;
452 struct gfs2_holder *gh;
453 unsigned state = ret & LM_OUT_ST_MASK;
454 int rv;
455
456 spin_lock(&gl->gl_lockref.lock);
457 trace_gfs2_glock_state_change(gl, state);
458 state_change(gl, state);
459 gh = find_first_waiter(gl);
460
461 /* Demote to UN request arrived during demote to SH or DF */
462 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
463 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
464 gl->gl_target = LM_ST_UNLOCKED;
465
466 /* Check for state != intended state */
467 if (unlikely(state != gl->gl_target)) {
468 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
469 /* move to back of queue and try next entry */
470 if (ret & LM_OUT_CANCELED) {
471 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
472 list_move_tail(&gh->gh_list, &gl->gl_holders);
473 gh = find_first_waiter(gl);
474 gl->gl_target = gh->gh_state;
475 goto retry;
476 }
477 /* Some error or failed "try lock" - report it */
478 if ((ret & LM_OUT_ERROR) ||
479 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
480 gl->gl_target = gl->gl_state;
481 do_error(gl, ret);
482 goto out;
483 }
484 }
485 switch(state) {
486 /* Unlocked due to conversion deadlock, try again */
487 case LM_ST_UNLOCKED:
488retry:
489 do_xmote(gl, gh, gl->gl_target);
490 break;
491 /* Conversion fails, unlock and try again */
492 case LM_ST_SHARED:
493 case LM_ST_DEFERRED:
494 do_xmote(gl, gh, LM_ST_UNLOCKED);
495 break;
496 default: /* Everything else */
497 pr_err("wanted %u got %u\n", gl->gl_target, state);
498 GLOCK_BUG_ON(gl, 1);
499 }
500 spin_unlock(&gl->gl_lockref.lock);
501 return;
502 }
503
504 /* Fast path - we got what we asked for */
505 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
506 gfs2_demote_wake(gl);
507 if (state != LM_ST_UNLOCKED) {
508 if (glops->go_xmote_bh) {
509 spin_unlock(&gl->gl_lockref.lock);
510 rv = glops->go_xmote_bh(gl, gh);
511 spin_lock(&gl->gl_lockref.lock);
512 if (rv) {
513 do_error(gl, rv);
514 goto out;
515 }
516 }
517 rv = do_promote(gl);
518 if (rv == 2)
519 goto out_locked;
520 }
521out:
522 clear_bit(GLF_LOCK, &gl->gl_flags);
523out_locked:
524 spin_unlock(&gl->gl_lockref.lock);
525}
526
527/**
528 * do_xmote - Calls the DLM to change the state of a lock
529 * @gl: The lock state
530 * @gh: The holder (only for promotes)
531 * @target: The target lock state
532 *
533 */
534
535static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
536__releases(&gl->gl_lockref.lock)
537__acquires(&gl->gl_lockref.lock)
538{
539 const struct gfs2_glock_operations *glops = gl->gl_ops;
540 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
541 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
542 int ret;
543
544 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
545 target != LM_ST_UNLOCKED)
546 return;
547 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
548 LM_FLAG_PRIORITY);
549 GLOCK_BUG_ON(gl, gl->gl_state == target);
550 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
551 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
552 glops->go_inval) {
553 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
554 do_error(gl, 0); /* Fail queued try locks */
555 }
556 gl->gl_req = target;
557 set_bit(GLF_BLOCKING, &gl->gl_flags);
558 if ((gl->gl_req == LM_ST_UNLOCKED) ||
559 (gl->gl_state == LM_ST_EXCLUSIVE) ||
560 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
561 clear_bit(GLF_BLOCKING, &gl->gl_flags);
562 spin_unlock(&gl->gl_lockref.lock);
563 if (glops->go_sync)
564 glops->go_sync(gl);
565 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
566 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
567 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
568
569 gfs2_glock_hold(gl);
570 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
571 /* lock_dlm */
572 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
573 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
574 target == LM_ST_UNLOCKED &&
575 test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
576 finish_xmote(gl, target);
577 gfs2_glock_queue_work(gl, 0);
578 }
579 else if (ret) {
580 pr_err("lm_lock ret %d\n", ret);
581 GLOCK_BUG_ON(gl, !test_bit(SDF_SHUTDOWN,
582 &sdp->sd_flags));
583 }
584 } else { /* lock_nolock */
585 finish_xmote(gl, target);
586 gfs2_glock_queue_work(gl, 0);
587 }
588
589 spin_lock(&gl->gl_lockref.lock);
590}
591
592/**
593 * find_first_holder - find the first "holder" gh
594 * @gl: the glock
595 */
596
597static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
598{
599 struct gfs2_holder *gh;
600
601 if (!list_empty(&gl->gl_holders)) {
602 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
603 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
604 return gh;
605 }
606 return NULL;
607}
608
609/**
610 * run_queue - do all outstanding tasks related to a glock
611 * @gl: The glock in question
612 * @nonblock: True if we must not block in run_queue
613 *
614 */
615
616static void run_queue(struct gfs2_glock *gl, const int nonblock)
617__releases(&gl->gl_lockref.lock)
618__acquires(&gl->gl_lockref.lock)
619{
620 struct gfs2_holder *gh = NULL;
621 int ret;
622
623 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
624 return;
625
626 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
627
628 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
629 gl->gl_demote_state != gl->gl_state) {
630 if (find_first_holder(gl))
631 goto out_unlock;
632 if (nonblock)
633 goto out_sched;
634 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
635 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
636 gl->gl_target = gl->gl_demote_state;
637 } else {
638 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
639 gfs2_demote_wake(gl);
640 ret = do_promote(gl);
641 if (ret == 0)
642 goto out_unlock;
643 if (ret == 2)
644 goto out;
645 gh = find_first_waiter(gl);
646 gl->gl_target = gh->gh_state;
647 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
648 do_error(gl, 0); /* Fail queued try locks */
649 }
650 do_xmote(gl, gh, gl->gl_target);
651out:
652 return;
653
654out_sched:
655 clear_bit(GLF_LOCK, &gl->gl_flags);
656 smp_mb__after_atomic();
657 gl->gl_lockref.count++;
658 __gfs2_glock_queue_work(gl, 0);
659 return;
660
661out_unlock:
662 clear_bit(GLF_LOCK, &gl->gl_flags);
663 smp_mb__after_atomic();
664 return;
665}
666
667static void delete_work_func(struct work_struct *work)
668{
669 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
670 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
671 struct inode *inode;
672 u64 no_addr = gl->gl_name.ln_number;
673
674 /* If someone's using this glock to create a new dinode, the block must
675 have been freed by another node, then re-used, in which case our
676 iopen callback is too late after the fact. Ignore it. */
677 if (test_bit(GLF_INODE_CREATING, &gl->gl_flags))
678 goto out;
679
680 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
681 if (inode && !IS_ERR(inode)) {
682 d_prune_aliases(inode);
683 iput(inode);
684 }
685out:
686 gfs2_glock_put(gl);
687}
688
689static void glock_work_func(struct work_struct *work)
690{
691 unsigned long delay = 0;
692 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
693 unsigned int drop_refs = 1;
694
695 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
696 finish_xmote(gl, gl->gl_reply);
697 drop_refs++;
698 }
699 spin_lock(&gl->gl_lockref.lock);
700 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
701 gl->gl_state != LM_ST_UNLOCKED &&
702 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
703 unsigned long holdtime, now = jiffies;
704
705 holdtime = gl->gl_tchange + gl->gl_hold_time;
706 if (time_before(now, holdtime))
707 delay = holdtime - now;
708
709 if (!delay) {
710 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
711 set_bit(GLF_DEMOTE, &gl->gl_flags);
712 }
713 }
714 run_queue(gl, 0);
715 if (delay) {
716 /* Keep one glock reference for the work we requeue. */
717 drop_refs--;
718 if (gl->gl_name.ln_type != LM_TYPE_INODE)
719 delay = 0;
720 __gfs2_glock_queue_work(gl, delay);
721 }
722
723 /*
724 * Drop the remaining glock references manually here. (Mind that
725 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
726 * here as well.)
727 */
728 gl->gl_lockref.count -= drop_refs;
729 if (!gl->gl_lockref.count) {
730 __gfs2_glock_put(gl);
731 return;
732 }
733 spin_unlock(&gl->gl_lockref.lock);
734}
735
736static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
737 struct gfs2_glock *new)
738{
739 struct wait_glock_queue wait;
740 wait_queue_head_t *wq = glock_waitqueue(name);
741 struct gfs2_glock *gl;
742
743 wait.name = name;
744 init_wait(&wait.wait);
745 wait.wait.func = glock_wake_function;
746
747again:
748 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
749 rcu_read_lock();
750 if (new) {
751 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
752 &new->gl_node, ht_parms);
753 if (IS_ERR(gl))
754 goto out;
755 } else {
756 gl = rhashtable_lookup_fast(&gl_hash_table,
757 name, ht_parms);
758 }
759 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
760 rcu_read_unlock();
761 schedule();
762 goto again;
763 }
764out:
765 rcu_read_unlock();
766 finish_wait(wq, &wait.wait);
767 return gl;
768}
769
770/**
771 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
772 * @sdp: The GFS2 superblock
773 * @number: the lock number
774 * @glops: The glock_operations to use
775 * @create: If 0, don't create the glock if it doesn't exist
776 * @glp: the glock is returned here
777 *
778 * This does not lock a glock, just finds/creates structures for one.
779 *
780 * Returns: errno
781 */
782
783int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
784 const struct gfs2_glock_operations *glops, int create,
785 struct gfs2_glock **glp)
786{
787 struct super_block *s = sdp->sd_vfs;
788 struct lm_lockname name = { .ln_number = number,
789 .ln_type = glops->go_type,
790 .ln_sbd = sdp };
791 struct gfs2_glock *gl, *tmp;
792 struct address_space *mapping;
793 struct kmem_cache *cachep;
794 int ret = 0;
795
796 gl = find_insert_glock(&name, NULL);
797 if (gl) {
798 *glp = gl;
799 return 0;
800 }
801 if (!create)
802 return -ENOENT;
803
804 if (glops->go_flags & GLOF_ASPACE)
805 cachep = gfs2_glock_aspace_cachep;
806 else
807 cachep = gfs2_glock_cachep;
808 gl = kmem_cache_alloc(cachep, GFP_NOFS);
809 if (!gl)
810 return -ENOMEM;
811
812 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
813
814 if (glops->go_flags & GLOF_LVB) {
815 gl->gl_lksb.sb_lvbptr = kzalloc(GFS2_MIN_LVB_SIZE, GFP_NOFS);
816 if (!gl->gl_lksb.sb_lvbptr) {
817 kmem_cache_free(cachep, gl);
818 return -ENOMEM;
819 }
820 }
821
822 atomic_inc(&sdp->sd_glock_disposal);
823 gl->gl_node.next = NULL;
824 gl->gl_flags = 0;
825 gl->gl_name = name;
826 gl->gl_lockref.count = 1;
827 gl->gl_state = LM_ST_UNLOCKED;
828 gl->gl_target = LM_ST_UNLOCKED;
829 gl->gl_demote_state = LM_ST_EXCLUSIVE;
830 gl->gl_ops = glops;
831 gl->gl_dstamp = 0;
832 preempt_disable();
833 /* We use the global stats to estimate the initial per-glock stats */
834 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
835 preempt_enable();
836 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
837 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
838 gl->gl_tchange = jiffies;
839 gl->gl_object = NULL;
840 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
841 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
842 INIT_WORK(&gl->gl_delete, delete_work_func);
843
844 mapping = gfs2_glock2aspace(gl);
845 if (mapping) {
846 mapping->a_ops = &gfs2_meta_aops;
847 mapping->host = s->s_bdev->bd_inode;
848 mapping->flags = 0;
849 mapping_set_gfp_mask(mapping, GFP_NOFS);
850 mapping->private_data = NULL;
851 mapping->writeback_index = 0;
852 }
853
854 tmp = find_insert_glock(&name, gl);
855 if (!tmp) {
856 *glp = gl;
857 goto out;
858 }
859 if (IS_ERR(tmp)) {
860 ret = PTR_ERR(tmp);
861 goto out_free;
862 }
863 *glp = tmp;
864
865out_free:
866 kfree(gl->gl_lksb.sb_lvbptr);
867 kmem_cache_free(cachep, gl);
868 atomic_dec(&sdp->sd_glock_disposal);
869
870out:
871 return ret;
872}
873
874/**
875 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
876 * @gl: the glock
877 * @state: the state we're requesting
878 * @flags: the modifier flags
879 * @gh: the holder structure
880 *
881 */
882
883void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
884 struct gfs2_holder *gh)
885{
886 INIT_LIST_HEAD(&gh->gh_list);
887 gh->gh_gl = gl;
888 gh->gh_ip = _RET_IP_;
889 gh->gh_owner_pid = get_pid(task_pid(current));
890 gh->gh_state = state;
891 gh->gh_flags = flags;
892 gh->gh_error = 0;
893 gh->gh_iflags = 0;
894 gfs2_glock_hold(gl);
895}
896
897/**
898 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
899 * @state: the state we're requesting
900 * @flags: the modifier flags
901 * @gh: the holder structure
902 *
903 * Don't mess with the glock.
904 *
905 */
906
907void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
908{
909 gh->gh_state = state;
910 gh->gh_flags = flags;
911 gh->gh_iflags = 0;
912 gh->gh_ip = _RET_IP_;
913 put_pid(gh->gh_owner_pid);
914 gh->gh_owner_pid = get_pid(task_pid(current));
915}
916
917/**
918 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
919 * @gh: the holder structure
920 *
921 */
922
923void gfs2_holder_uninit(struct gfs2_holder *gh)
924{
925 put_pid(gh->gh_owner_pid);
926 gfs2_glock_put(gh->gh_gl);
927 gfs2_holder_mark_uninitialized(gh);
928 gh->gh_ip = 0;
929}
930
931/**
932 * gfs2_glock_wait - wait on a glock acquisition
933 * @gh: the glock holder
934 *
935 * Returns: 0 on success
936 */
937
938int gfs2_glock_wait(struct gfs2_holder *gh)
939{
940 unsigned long time1 = jiffies;
941
942 might_sleep();
943 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
944 if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
945 /* Lengthen the minimum hold time. */
946 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
947 GL_GLOCK_HOLD_INCR,
948 GL_GLOCK_MAX_HOLD);
949 return gh->gh_error;
950}
951
952/**
953 * handle_callback - process a demote request
954 * @gl: the glock
955 * @state: the state the caller wants us to change to
956 *
957 * There are only two requests that we are going to see in actual
958 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
959 */
960
961static void handle_callback(struct gfs2_glock *gl, unsigned int state,
962 unsigned long delay, bool remote)
963{
964 int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
965
966 set_bit(bit, &gl->gl_flags);
967 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
968 gl->gl_demote_state = state;
969 gl->gl_demote_time = jiffies;
970 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
971 gl->gl_demote_state != state) {
972 gl->gl_demote_state = LM_ST_UNLOCKED;
973 }
974 if (gl->gl_ops->go_callback)
975 gl->gl_ops->go_callback(gl, remote);
976 trace_gfs2_demote_rq(gl, remote);
977}
978
979void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
980{
981 struct va_format vaf;
982 va_list args;
983
984 va_start(args, fmt);
985
986 if (seq) {
987 seq_vprintf(seq, fmt, args);
988 } else {
989 vaf.fmt = fmt;
990 vaf.va = &args;
991
992 pr_err("%pV", &vaf);
993 }
994
995 va_end(args);
996}
997
998/**
999 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1000 * @gh: the holder structure to add
1001 *
1002 * Eventually we should move the recursive locking trap to a
1003 * debugging option or something like that. This is the fast
1004 * path and needs to have the minimum number of distractions.
1005 *
1006 */
1007
1008static inline void add_to_queue(struct gfs2_holder *gh)
1009__releases(&gl->gl_lockref.lock)
1010__acquires(&gl->gl_lockref.lock)
1011{
1012 struct gfs2_glock *gl = gh->gh_gl;
1013 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1014 struct list_head *insert_pt = NULL;
1015 struct gfs2_holder *gh2;
1016 int try_futile = 0;
1017
1018 BUG_ON(gh->gh_owner_pid == NULL);
1019 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1020 BUG();
1021
1022 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1023 if (test_bit(GLF_LOCK, &gl->gl_flags))
1024 try_futile = !may_grant(gl, gh);
1025 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1026 goto fail;
1027 }
1028
1029 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1030 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
1031 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
1032 goto trap_recursive;
1033 if (try_futile &&
1034 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1035fail:
1036 gh->gh_error = GLR_TRYFAILED;
1037 gfs2_holder_wake(gh);
1038 return;
1039 }
1040 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1041 continue;
1042 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1043 insert_pt = &gh2->gh_list;
1044 }
1045 set_bit(GLF_QUEUED, &gl->gl_flags);
1046 trace_gfs2_glock_queue(gh, 1);
1047 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1048 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1049 if (likely(insert_pt == NULL)) {
1050 list_add_tail(&gh->gh_list, &gl->gl_holders);
1051 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1052 goto do_cancel;
1053 return;
1054 }
1055 list_add_tail(&gh->gh_list, insert_pt);
1056do_cancel:
1057 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
1058 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1059 spin_unlock(&gl->gl_lockref.lock);
1060 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1061 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1062 spin_lock(&gl->gl_lockref.lock);
1063 }
1064 return;
1065
1066trap_recursive:
1067 pr_err("original: %pSR\n", (void *)gh2->gh_ip);
1068 pr_err("pid: %d\n", pid_nr(gh2->gh_owner_pid));
1069 pr_err("lock type: %d req lock state : %d\n",
1070 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1071 pr_err("new: %pSR\n", (void *)gh->gh_ip);
1072 pr_err("pid: %d\n", pid_nr(gh->gh_owner_pid));
1073 pr_err("lock type: %d req lock state : %d\n",
1074 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1075 gfs2_dump_glock(NULL, gl);
1076 BUG();
1077}
1078
1079/**
1080 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1081 * @gh: the holder structure
1082 *
1083 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1084 *
1085 * Returns: 0, GLR_TRYFAILED, or errno on failure
1086 */
1087
1088int gfs2_glock_nq(struct gfs2_holder *gh)
1089{
1090 struct gfs2_glock *gl = gh->gh_gl;
1091 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1092 int error = 0;
1093
1094 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1095 return -EIO;
1096
1097 if (test_bit(GLF_LRU, &gl->gl_flags))
1098 gfs2_glock_remove_from_lru(gl);
1099
1100 spin_lock(&gl->gl_lockref.lock);
1101 add_to_queue(gh);
1102 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1103 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1104 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1105 gl->gl_lockref.count++;
1106 __gfs2_glock_queue_work(gl, 0);
1107 }
1108 run_queue(gl, 1);
1109 spin_unlock(&gl->gl_lockref.lock);
1110
1111 if (!(gh->gh_flags & GL_ASYNC))
1112 error = gfs2_glock_wait(gh);
1113
1114 return error;
1115}
1116
1117/**
1118 * gfs2_glock_poll - poll to see if an async request has been completed
1119 * @gh: the holder
1120 *
1121 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1122 */
1123
1124int gfs2_glock_poll(struct gfs2_holder *gh)
1125{
1126 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1127}
1128
1129/**
1130 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1131 * @gh: the glock holder
1132 *
1133 */
1134
1135void gfs2_glock_dq(struct gfs2_holder *gh)
1136{
1137 struct gfs2_glock *gl = gh->gh_gl;
1138 const struct gfs2_glock_operations *glops = gl->gl_ops;
1139 unsigned delay = 0;
1140 int fast_path = 0;
1141
1142 spin_lock(&gl->gl_lockref.lock);
1143 if (gh->gh_flags & GL_NOCACHE)
1144 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1145
1146 list_del_init(&gh->gh_list);
1147 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1148 if (find_first_holder(gl) == NULL) {
1149 if (glops->go_unlock) {
1150 GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1151 spin_unlock(&gl->gl_lockref.lock);
1152 glops->go_unlock(gh);
1153 spin_lock(&gl->gl_lockref.lock);
1154 clear_bit(GLF_LOCK, &gl->gl_flags);
1155 }
1156 if (list_empty(&gl->gl_holders) &&
1157 !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1158 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1159 fast_path = 1;
1160 }
1161 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl) &&
1162 (glops->go_flags & GLOF_LRU))
1163 gfs2_glock_add_to_lru(gl);
1164
1165 trace_gfs2_glock_queue(gh, 0);
1166 if (unlikely(!fast_path)) {
1167 gl->gl_lockref.count++;
1168 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1169 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1170 gl->gl_name.ln_type == LM_TYPE_INODE)
1171 delay = gl->gl_hold_time;
1172 __gfs2_glock_queue_work(gl, delay);
1173 }
1174 spin_unlock(&gl->gl_lockref.lock);
1175}
1176
1177void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1178{
1179 struct gfs2_glock *gl = gh->gh_gl;
1180 gfs2_glock_dq(gh);
1181 might_sleep();
1182 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1183}
1184
1185/**
1186 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1187 * @gh: the holder structure
1188 *
1189 */
1190
1191void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1192{
1193 gfs2_glock_dq(gh);
1194 gfs2_holder_uninit(gh);
1195}
1196
1197/**
1198 * gfs2_glock_nq_num - acquire a glock based on lock number
1199 * @sdp: the filesystem
1200 * @number: the lock number
1201 * @glops: the glock operations for the type of glock
1202 * @state: the state to acquire the glock in
1203 * @flags: modifier flags for the acquisition
1204 * @gh: the struct gfs2_holder
1205 *
1206 * Returns: errno
1207 */
1208
1209int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1210 const struct gfs2_glock_operations *glops,
1211 unsigned int state, u16 flags, struct gfs2_holder *gh)
1212{
1213 struct gfs2_glock *gl;
1214 int error;
1215
1216 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1217 if (!error) {
1218 error = gfs2_glock_nq_init(gl, state, flags, gh);
1219 gfs2_glock_put(gl);
1220 }
1221
1222 return error;
1223}
1224
1225/**
1226 * glock_compare - Compare two struct gfs2_glock structures for sorting
1227 * @arg_a: the first structure
1228 * @arg_b: the second structure
1229 *
1230 */
1231
1232static int glock_compare(const void *arg_a, const void *arg_b)
1233{
1234 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1235 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1236 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1237 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1238
1239 if (a->ln_number > b->ln_number)
1240 return 1;
1241 if (a->ln_number < b->ln_number)
1242 return -1;
1243 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1244 return 0;
1245}
1246
1247/**
1248 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1249 * @num_gh: the number of structures
1250 * @ghs: an array of struct gfs2_holder structures
1251 *
1252 * Returns: 0 on success (all glocks acquired),
1253 * errno on failure (no glocks acquired)
1254 */
1255
1256static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1257 struct gfs2_holder **p)
1258{
1259 unsigned int x;
1260 int error = 0;
1261
1262 for (x = 0; x < num_gh; x++)
1263 p[x] = &ghs[x];
1264
1265 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1266
1267 for (x = 0; x < num_gh; x++) {
1268 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1269
1270 error = gfs2_glock_nq(p[x]);
1271 if (error) {
1272 while (x--)
1273 gfs2_glock_dq(p[x]);
1274 break;
1275 }
1276 }
1277
1278 return error;
1279}
1280
1281/**
1282 * gfs2_glock_nq_m - acquire multiple glocks
1283 * @num_gh: the number of structures
1284 * @ghs: an array of struct gfs2_holder structures
1285 *
1286 *
1287 * Returns: 0 on success (all glocks acquired),
1288 * errno on failure (no glocks acquired)
1289 */
1290
1291int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1292{
1293 struct gfs2_holder *tmp[4];
1294 struct gfs2_holder **pph = tmp;
1295 int error = 0;
1296
1297 switch(num_gh) {
1298 case 0:
1299 return 0;
1300 case 1:
1301 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1302 return gfs2_glock_nq(ghs);
1303 default:
1304 if (num_gh <= 4)
1305 break;
1306 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1307 if (!pph)
1308 return -ENOMEM;
1309 }
1310
1311 error = nq_m_sync(num_gh, ghs, pph);
1312
1313 if (pph != tmp)
1314 kfree(pph);
1315
1316 return error;
1317}
1318
1319/**
1320 * gfs2_glock_dq_m - release multiple glocks
1321 * @num_gh: the number of structures
1322 * @ghs: an array of struct gfs2_holder structures
1323 *
1324 */
1325
1326void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1327{
1328 while (num_gh--)
1329 gfs2_glock_dq(&ghs[num_gh]);
1330}
1331
1332void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1333{
1334 unsigned long delay = 0;
1335 unsigned long holdtime;
1336 unsigned long now = jiffies;
1337
1338 gfs2_glock_hold(gl);
1339 holdtime = gl->gl_tchange + gl->gl_hold_time;
1340 if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1341 gl->gl_name.ln_type == LM_TYPE_INODE) {
1342 if (time_before(now, holdtime))
1343 delay = holdtime - now;
1344 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1345 delay = gl->gl_hold_time;
1346 }
1347
1348 spin_lock(&gl->gl_lockref.lock);
1349 handle_callback(gl, state, delay, true);
1350 __gfs2_glock_queue_work(gl, delay);
1351 spin_unlock(&gl->gl_lockref.lock);
1352}
1353
1354/**
1355 * gfs2_should_freeze - Figure out if glock should be frozen
1356 * @gl: The glock in question
1357 *
1358 * Glocks are not frozen if (a) the result of the dlm operation is
1359 * an error, (b) the locking operation was an unlock operation or
1360 * (c) if there is a "noexp" flagged request anywhere in the queue
1361 *
1362 * Returns: 1 if freezing should occur, 0 otherwise
1363 */
1364
1365static int gfs2_should_freeze(const struct gfs2_glock *gl)
1366{
1367 const struct gfs2_holder *gh;
1368
1369 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1370 return 0;
1371 if (gl->gl_target == LM_ST_UNLOCKED)
1372 return 0;
1373
1374 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1375 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1376 continue;
1377 if (LM_FLAG_NOEXP & gh->gh_flags)
1378 return 0;
1379 }
1380
1381 return 1;
1382}
1383
1384/**
1385 * gfs2_glock_complete - Callback used by locking
1386 * @gl: Pointer to the glock
1387 * @ret: The return value from the dlm
1388 *
1389 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1390 * to use a bitfield shared with other glock state fields.
1391 */
1392
1393void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1394{
1395 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1396
1397 spin_lock(&gl->gl_lockref.lock);
1398 gl->gl_reply = ret;
1399
1400 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1401 if (gfs2_should_freeze(gl)) {
1402 set_bit(GLF_FROZEN, &gl->gl_flags);
1403 spin_unlock(&gl->gl_lockref.lock);
1404 return;
1405 }
1406 }
1407
1408 gl->gl_lockref.count++;
1409 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1410 __gfs2_glock_queue_work(gl, 0);
1411 spin_unlock(&gl->gl_lockref.lock);
1412}
1413
1414static int glock_cmp(void *priv, struct list_head *a, struct list_head *b)
1415{
1416 struct gfs2_glock *gla, *glb;
1417
1418 gla = list_entry(a, struct gfs2_glock, gl_lru);
1419 glb = list_entry(b, struct gfs2_glock, gl_lru);
1420
1421 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1422 return 1;
1423 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1424 return -1;
1425
1426 return 0;
1427}
1428
1429/**
1430 * gfs2_dispose_glock_lru - Demote a list of glocks
1431 * @list: The list to dispose of
1432 *
1433 * Disposing of glocks may involve disk accesses, so that here we sort
1434 * the glocks by number (i.e. disk location of the inodes) so that if
1435 * there are any such accesses, they'll be sent in order (mostly).
1436 *
1437 * Must be called under the lru_lock, but may drop and retake this
1438 * lock. While the lru_lock is dropped, entries may vanish from the
1439 * list, but no new entries will appear on the list (since it is
1440 * private)
1441 */
1442
1443static void gfs2_dispose_glock_lru(struct list_head *list)
1444__releases(&lru_lock)
1445__acquires(&lru_lock)
1446{
1447 struct gfs2_glock *gl;
1448
1449 list_sort(NULL, list, glock_cmp);
1450
1451 while(!list_empty(list)) {
1452 gl = list_entry(list->next, struct gfs2_glock, gl_lru);
1453 list_del_init(&gl->gl_lru);
1454 if (!spin_trylock(&gl->gl_lockref.lock)) {
1455add_back_to_lru:
1456 list_add(&gl->gl_lru, &lru_list);
1457 atomic_inc(&lru_count);
1458 continue;
1459 }
1460 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1461 spin_unlock(&gl->gl_lockref.lock);
1462 goto add_back_to_lru;
1463 }
1464 clear_bit(GLF_LRU, &gl->gl_flags);
1465 gl->gl_lockref.count++;
1466 if (demote_ok(gl))
1467 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1468 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1469 __gfs2_glock_queue_work(gl, 0);
1470 spin_unlock(&gl->gl_lockref.lock);
1471 cond_resched_lock(&lru_lock);
1472 }
1473}
1474
1475/**
1476 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1477 * @nr: The number of entries to scan
1478 *
1479 * This function selects the entries on the LRU which are able to
1480 * be demoted, and then kicks off the process by calling
1481 * gfs2_dispose_glock_lru() above.
1482 */
1483
1484static long gfs2_scan_glock_lru(int nr)
1485{
1486 struct gfs2_glock *gl;
1487 LIST_HEAD(skipped);
1488 LIST_HEAD(dispose);
1489 long freed = 0;
1490
1491 spin_lock(&lru_lock);
1492 while ((nr-- >= 0) && !list_empty(&lru_list)) {
1493 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1494
1495 /* Test for being demotable */
1496 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1497 list_move(&gl->gl_lru, &dispose);
1498 atomic_dec(&lru_count);
1499 freed++;
1500 continue;
1501 }
1502
1503 list_move(&gl->gl_lru, &skipped);
1504 }
1505 list_splice(&skipped, &lru_list);
1506 if (!list_empty(&dispose))
1507 gfs2_dispose_glock_lru(&dispose);
1508 spin_unlock(&lru_lock);
1509
1510 return freed;
1511}
1512
1513static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1514 struct shrink_control *sc)
1515{
1516 if (!(sc->gfp_mask & __GFP_FS))
1517 return SHRINK_STOP;
1518 return gfs2_scan_glock_lru(sc->nr_to_scan);
1519}
1520
1521static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1522 struct shrink_control *sc)
1523{
1524 return vfs_pressure_ratio(atomic_read(&lru_count));
1525}
1526
1527static struct shrinker glock_shrinker = {
1528 .seeks = DEFAULT_SEEKS,
1529 .count_objects = gfs2_glock_shrink_count,
1530 .scan_objects = gfs2_glock_shrink_scan,
1531};
1532
1533/**
1534 * examine_bucket - Call a function for glock in a hash bucket
1535 * @examiner: the function
1536 * @sdp: the filesystem
1537 * @bucket: the bucket
1538 *
1539 * Note that the function can be called multiple times on the same
1540 * object. So the user must ensure that the function can cope with
1541 * that.
1542 */
1543
1544static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1545{
1546 struct gfs2_glock *gl;
1547 struct rhashtable_iter iter;
1548
1549 rhashtable_walk_enter(&gl_hash_table, &iter);
1550
1551 do {
1552 rhashtable_walk_start(&iter);
1553
1554 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl))
1555 if (gl->gl_name.ln_sbd == sdp &&
1556 lockref_get_not_dead(&gl->gl_lockref))
1557 examiner(gl);
1558
1559 rhashtable_walk_stop(&iter);
1560 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
1561
1562 rhashtable_walk_exit(&iter);
1563}
1564
1565/**
1566 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1567 * @gl: The glock to thaw
1568 *
1569 */
1570
1571static void thaw_glock(struct gfs2_glock *gl)
1572{
1573 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags)) {
1574 gfs2_glock_put(gl);
1575 return;
1576 }
1577 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1578 gfs2_glock_queue_work(gl, 0);
1579}
1580
1581/**
1582 * clear_glock - look at a glock and see if we can free it from glock cache
1583 * @gl: the glock to look at
1584 *
1585 */
1586
1587static void clear_glock(struct gfs2_glock *gl)
1588{
1589 gfs2_glock_remove_from_lru(gl);
1590
1591 spin_lock(&gl->gl_lockref.lock);
1592 if (gl->gl_state != LM_ST_UNLOCKED)
1593 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1594 __gfs2_glock_queue_work(gl, 0);
1595 spin_unlock(&gl->gl_lockref.lock);
1596}
1597
1598/**
1599 * gfs2_glock_thaw - Thaw any frozen glocks
1600 * @sdp: The super block
1601 *
1602 */
1603
1604void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1605{
1606 glock_hash_walk(thaw_glock, sdp);
1607}
1608
1609static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1610{
1611 spin_lock(&gl->gl_lockref.lock);
1612 gfs2_dump_glock(seq, gl);
1613 spin_unlock(&gl->gl_lockref.lock);
1614}
1615
1616static void dump_glock_func(struct gfs2_glock *gl)
1617{
1618 dump_glock(NULL, gl);
1619}
1620
1621/**
1622 * gfs2_gl_hash_clear - Empty out the glock hash table
1623 * @sdp: the filesystem
1624 * @wait: wait until it's all gone
1625 *
1626 * Called when unmounting the filesystem.
1627 */
1628
1629void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1630{
1631 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1632 flush_workqueue(glock_workqueue);
1633 glock_hash_walk(clear_glock, sdp);
1634 flush_workqueue(glock_workqueue);
1635 wait_event_timeout(sdp->sd_glock_wait,
1636 atomic_read(&sdp->sd_glock_disposal) == 0,
1637 HZ * 600);
1638 glock_hash_walk(dump_glock_func, sdp);
1639}
1640
1641void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1642{
1643 struct gfs2_glock *gl = ip->i_gl;
1644 int ret;
1645
1646 ret = gfs2_truncatei_resume(ip);
1647 gfs2_assert_withdraw(gl->gl_name.ln_sbd, ret == 0);
1648
1649 spin_lock(&gl->gl_lockref.lock);
1650 clear_bit(GLF_LOCK, &gl->gl_flags);
1651 run_queue(gl, 1);
1652 spin_unlock(&gl->gl_lockref.lock);
1653}
1654
1655static const char *state2str(unsigned state)
1656{
1657 switch(state) {
1658 case LM_ST_UNLOCKED:
1659 return "UN";
1660 case LM_ST_SHARED:
1661 return "SH";
1662 case LM_ST_DEFERRED:
1663 return "DF";
1664 case LM_ST_EXCLUSIVE:
1665 return "EX";
1666 }
1667 return "??";
1668}
1669
1670static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
1671{
1672 char *p = buf;
1673 if (flags & LM_FLAG_TRY)
1674 *p++ = 't';
1675 if (flags & LM_FLAG_TRY_1CB)
1676 *p++ = 'T';
1677 if (flags & LM_FLAG_NOEXP)
1678 *p++ = 'e';
1679 if (flags & LM_FLAG_ANY)
1680 *p++ = 'A';
1681 if (flags & LM_FLAG_PRIORITY)
1682 *p++ = 'p';
1683 if (flags & GL_ASYNC)
1684 *p++ = 'a';
1685 if (flags & GL_EXACT)
1686 *p++ = 'E';
1687 if (flags & GL_NOCACHE)
1688 *p++ = 'c';
1689 if (test_bit(HIF_HOLDER, &iflags))
1690 *p++ = 'H';
1691 if (test_bit(HIF_WAIT, &iflags))
1692 *p++ = 'W';
1693 if (test_bit(HIF_FIRST, &iflags))
1694 *p++ = 'F';
1695 *p = 0;
1696 return buf;
1697}
1698
1699/**
1700 * dump_holder - print information about a glock holder
1701 * @seq: the seq_file struct
1702 * @gh: the glock holder
1703 *
1704 */
1705
1706static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1707{
1708 struct task_struct *gh_owner = NULL;
1709 char flags_buf[32];
1710
1711 rcu_read_lock();
1712 if (gh->gh_owner_pid)
1713 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1714 gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1715 state2str(gh->gh_state),
1716 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1717 gh->gh_error,
1718 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1719 gh_owner ? gh_owner->comm : "(ended)",
1720 (void *)gh->gh_ip);
1721 rcu_read_unlock();
1722}
1723
1724static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1725{
1726 const unsigned long *gflags = &gl->gl_flags;
1727 char *p = buf;
1728
1729 if (test_bit(GLF_LOCK, gflags))
1730 *p++ = 'l';
1731 if (test_bit(GLF_DEMOTE, gflags))
1732 *p++ = 'D';
1733 if (test_bit(GLF_PENDING_DEMOTE, gflags))
1734 *p++ = 'd';
1735 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1736 *p++ = 'p';
1737 if (test_bit(GLF_DIRTY, gflags))
1738 *p++ = 'y';
1739 if (test_bit(GLF_LFLUSH, gflags))
1740 *p++ = 'f';
1741 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1742 *p++ = 'i';
1743 if (test_bit(GLF_REPLY_PENDING, gflags))
1744 *p++ = 'r';
1745 if (test_bit(GLF_INITIAL, gflags))
1746 *p++ = 'I';
1747 if (test_bit(GLF_FROZEN, gflags))
1748 *p++ = 'F';
1749 if (test_bit(GLF_QUEUED, gflags))
1750 *p++ = 'q';
1751 if (test_bit(GLF_LRU, gflags))
1752 *p++ = 'L';
1753 if (gl->gl_object)
1754 *p++ = 'o';
1755 if (test_bit(GLF_BLOCKING, gflags))
1756 *p++ = 'b';
1757 *p = 0;
1758 return buf;
1759}
1760
1761/**
1762 * gfs2_dump_glock - print information about a glock
1763 * @seq: The seq_file struct
1764 * @gl: the glock
1765 *
1766 * The file format is as follows:
1767 * One line per object, capital letters are used to indicate objects
1768 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1769 * other objects are indented by a single space and follow the glock to
1770 * which they are related. Fields are indicated by lower case letters
1771 * followed by a colon and the field value, except for strings which are in
1772 * [] so that its possible to see if they are composed of spaces for
1773 * example. The field's are n = number (id of the object), f = flags,
1774 * t = type, s = state, r = refcount, e = error, p = pid.
1775 *
1776 */
1777
1778void gfs2_dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1779{
1780 const struct gfs2_glock_operations *glops = gl->gl_ops;
1781 unsigned long long dtime;
1782 const struct gfs2_holder *gh;
1783 char gflags_buf[32];
1784
1785 dtime = jiffies - gl->gl_demote_time;
1786 dtime *= 1000000/HZ; /* demote time in uSec */
1787 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1788 dtime = 0;
1789 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",
1790 state2str(gl->gl_state),
1791 gl->gl_name.ln_type,
1792 (unsigned long long)gl->gl_name.ln_number,
1793 gflags2str(gflags_buf, gl),
1794 state2str(gl->gl_target),
1795 state2str(gl->gl_demote_state), dtime,
1796 atomic_read(&gl->gl_ail_count),
1797 atomic_read(&gl->gl_revokes),
1798 (int)gl->gl_lockref.count, gl->gl_hold_time);
1799
1800 list_for_each_entry(gh, &gl->gl_holders, gh_list)
1801 dump_holder(seq, gh);
1802
1803 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1804 glops->go_dump(seq, gl);
1805}
1806
1807static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
1808{
1809 struct gfs2_glock *gl = iter_ptr;
1810
1811 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
1812 gl->gl_name.ln_type,
1813 (unsigned long long)gl->gl_name.ln_number,
1814 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
1815 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
1816 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
1817 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
1818 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
1819 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
1820 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
1821 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
1822 return 0;
1823}
1824
1825static const char *gfs2_gltype[] = {
1826 "type",
1827 "reserved",
1828 "nondisk",
1829 "inode",
1830 "rgrp",
1831 "meta",
1832 "iopen",
1833 "flock",
1834 "plock",
1835 "quota",
1836 "journal",
1837};
1838
1839static const char *gfs2_stype[] = {
1840 [GFS2_LKS_SRTT] = "srtt",
1841 [GFS2_LKS_SRTTVAR] = "srttvar",
1842 [GFS2_LKS_SRTTB] = "srttb",
1843 [GFS2_LKS_SRTTVARB] = "srttvarb",
1844 [GFS2_LKS_SIRT] = "sirt",
1845 [GFS2_LKS_SIRTVAR] = "sirtvar",
1846 [GFS2_LKS_DCOUNT] = "dlm",
1847 [GFS2_LKS_QCOUNT] = "queue",
1848};
1849
1850#define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1851
1852static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
1853{
1854 struct gfs2_sbd *sdp = seq->private;
1855 loff_t pos = *(loff_t *)iter_ptr;
1856 unsigned index = pos >> 3;
1857 unsigned subindex = pos & 0x07;
1858 int i;
1859
1860 if (index == 0 && subindex != 0)
1861 return 0;
1862
1863 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
1864 (index == 0) ? "cpu": gfs2_stype[subindex]);
1865
1866 for_each_possible_cpu(i) {
1867 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
1868
1869 if (index == 0)
1870 seq_printf(seq, " %15u", i);
1871 else
1872 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
1873 lkstats[index - 1].stats[subindex]);
1874 }
1875 seq_putc(seq, '\n');
1876 return 0;
1877}
1878
1879int __init gfs2_glock_init(void)
1880{
1881 int i, ret;
1882
1883 ret = rhashtable_init(&gl_hash_table, &ht_parms);
1884 if (ret < 0)
1885 return ret;
1886
1887 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1888 WQ_HIGHPRI | WQ_FREEZABLE, 0);
1889 if (!glock_workqueue) {
1890 rhashtable_destroy(&gl_hash_table);
1891 return -ENOMEM;
1892 }
1893 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1894 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1895 0);
1896 if (!gfs2_delete_workqueue) {
1897 destroy_workqueue(glock_workqueue);
1898 rhashtable_destroy(&gl_hash_table);
1899 return -ENOMEM;
1900 }
1901
1902 ret = register_shrinker(&glock_shrinker);
1903 if (ret) {
1904 destroy_workqueue(gfs2_delete_workqueue);
1905 destroy_workqueue(glock_workqueue);
1906 rhashtable_destroy(&gl_hash_table);
1907 return ret;
1908 }
1909
1910 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
1911 init_waitqueue_head(glock_wait_table + i);
1912
1913 return 0;
1914}
1915
1916void gfs2_glock_exit(void)
1917{
1918 unregister_shrinker(&glock_shrinker);
1919 rhashtable_destroy(&gl_hash_table);
1920 destroy_workqueue(glock_workqueue);
1921 destroy_workqueue(gfs2_delete_workqueue);
1922}
1923
1924static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
1925{
1926 struct gfs2_glock *gl = gi->gl;
1927
1928 if (gl) {
1929 if (n == 0)
1930 return;
1931 if (!lockref_put_not_zero(&gl->gl_lockref))
1932 gfs2_glock_queue_put(gl);
1933 }
1934 for (;;) {
1935 gl = rhashtable_walk_next(&gi->hti);
1936 if (IS_ERR_OR_NULL(gl)) {
1937 if (gl == ERR_PTR(-EAGAIN)) {
1938 n = 1;
1939 continue;
1940 }
1941 gl = NULL;
1942 break;
1943 }
1944 if (gl->gl_name.ln_sbd != gi->sdp)
1945 continue;
1946 if (n <= 1) {
1947 if (!lockref_get_not_dead(&gl->gl_lockref))
1948 continue;
1949 break;
1950 } else {
1951 if (__lockref_is_dead(&gl->gl_lockref))
1952 continue;
1953 n--;
1954 }
1955 }
1956 gi->gl = gl;
1957}
1958
1959static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1960 __acquires(RCU)
1961{
1962 struct gfs2_glock_iter *gi = seq->private;
1963 loff_t n;
1964
1965 /*
1966 * We can either stay where we are, skip to the next hash table
1967 * entry, or start from the beginning.
1968 */
1969 if (*pos < gi->last_pos) {
1970 rhashtable_walk_exit(&gi->hti);
1971 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
1972 n = *pos + 1;
1973 } else {
1974 n = *pos - gi->last_pos;
1975 }
1976
1977 rhashtable_walk_start(&gi->hti);
1978
1979 gfs2_glock_iter_next(gi, n);
1980 gi->last_pos = *pos;
1981 return gi->gl;
1982}
1983
1984static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1985 loff_t *pos)
1986{
1987 struct gfs2_glock_iter *gi = seq->private;
1988
1989 (*pos)++;
1990 gi->last_pos = *pos;
1991 gfs2_glock_iter_next(gi, 1);
1992 return gi->gl;
1993}
1994
1995static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1996 __releases(RCU)
1997{
1998 struct gfs2_glock_iter *gi = seq->private;
1999
2000 rhashtable_walk_stop(&gi->hti);
2001}
2002
2003static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2004{
2005 dump_glock(seq, iter_ptr);
2006 return 0;
2007}
2008
2009static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2010{
2011 preempt_disable();
2012 if (*pos >= GFS2_NR_SBSTATS)
2013 return NULL;
2014 return pos;
2015}
2016
2017static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2018 loff_t *pos)
2019{
2020 (*pos)++;
2021 if (*pos >= GFS2_NR_SBSTATS)
2022 return NULL;
2023 return pos;
2024}
2025
2026static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2027{
2028 preempt_enable();
2029}
2030
2031static const struct seq_operations gfs2_glock_seq_ops = {
2032 .start = gfs2_glock_seq_start,
2033 .next = gfs2_glock_seq_next,
2034 .stop = gfs2_glock_seq_stop,
2035 .show = gfs2_glock_seq_show,
2036};
2037
2038static const struct seq_operations gfs2_glstats_seq_ops = {
2039 .start = gfs2_glock_seq_start,
2040 .next = gfs2_glock_seq_next,
2041 .stop = gfs2_glock_seq_stop,
2042 .show = gfs2_glstats_seq_show,
2043};
2044
2045static const struct seq_operations gfs2_sbstats_seq_ops = {
2046 .start = gfs2_sbstats_seq_start,
2047 .next = gfs2_sbstats_seq_next,
2048 .stop = gfs2_sbstats_seq_stop,
2049 .show = gfs2_sbstats_seq_show,
2050};
2051
2052#define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2053
2054static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2055 const struct seq_operations *ops)
2056{
2057 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2058 if (ret == 0) {
2059 struct seq_file *seq = file->private_data;
2060 struct gfs2_glock_iter *gi = seq->private;
2061
2062 gi->sdp = inode->i_private;
2063 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2064 if (seq->buf)
2065 seq->size = GFS2_SEQ_GOODSIZE;
2066 /*
2067 * Initially, we are "before" the first hash table entry; the
2068 * first call to rhashtable_walk_next gets us the first entry.
2069 */
2070 gi->last_pos = -1;
2071 gi->gl = NULL;
2072 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2073 }
2074 return ret;
2075}
2076
2077static int gfs2_glocks_open(struct inode *inode, struct file *file)
2078{
2079 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2080}
2081
2082static int gfs2_glocks_release(struct inode *inode, struct file *file)
2083{
2084 struct seq_file *seq = file->private_data;
2085 struct gfs2_glock_iter *gi = seq->private;
2086
2087 if (gi->gl)
2088 gfs2_glock_put(gi->gl);
2089 rhashtable_walk_exit(&gi->hti);
2090 return seq_release_private(inode, file);
2091}
2092
2093static int gfs2_glstats_open(struct inode *inode, struct file *file)
2094{
2095 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2096}
2097
2098static int gfs2_sbstats_open(struct inode *inode, struct file *file)
2099{
2100 int ret = seq_open(file, &gfs2_sbstats_seq_ops);
2101 if (ret == 0) {
2102 struct seq_file *seq = file->private_data;
2103 seq->private = inode->i_private; /* sdp */
2104 }
2105 return ret;
2106}
2107
2108static const struct file_operations gfs2_glocks_fops = {
2109 .owner = THIS_MODULE,
2110 .open = gfs2_glocks_open,
2111 .read = seq_read,
2112 .llseek = seq_lseek,
2113 .release = gfs2_glocks_release,
2114};
2115
2116static const struct file_operations gfs2_glstats_fops = {
2117 .owner = THIS_MODULE,
2118 .open = gfs2_glstats_open,
2119 .read = seq_read,
2120 .llseek = seq_lseek,
2121 .release = gfs2_glocks_release,
2122};
2123
2124static const struct file_operations gfs2_sbstats_fops = {
2125 .owner = THIS_MODULE,
2126 .open = gfs2_sbstats_open,
2127 .read = seq_read,
2128 .llseek = seq_lseek,
2129 .release = seq_release,
2130};
2131
2132int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2133{
2134 struct dentry *dent;
2135
2136 dent = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2137 if (IS_ERR_OR_NULL(dent))
2138 goto fail;
2139 sdp->debugfs_dir = dent;
2140
2141 dent = debugfs_create_file("glocks",
2142 S_IFREG | S_IRUGO,
2143 sdp->debugfs_dir, sdp,
2144 &gfs2_glocks_fops);
2145 if (IS_ERR_OR_NULL(dent))
2146 goto fail;
2147 sdp->debugfs_dentry_glocks = dent;
2148
2149 dent = debugfs_create_file("glstats",
2150 S_IFREG | S_IRUGO,
2151 sdp->debugfs_dir, sdp,
2152 &gfs2_glstats_fops);
2153 if (IS_ERR_OR_NULL(dent))
2154 goto fail;
2155 sdp->debugfs_dentry_glstats = dent;
2156
2157 dent = debugfs_create_file("sbstats",
2158 S_IFREG | S_IRUGO,
2159 sdp->debugfs_dir, sdp,
2160 &gfs2_sbstats_fops);
2161 if (IS_ERR_OR_NULL(dent))
2162 goto fail;
2163 sdp->debugfs_dentry_sbstats = dent;
2164
2165 return 0;
2166fail:
2167 gfs2_delete_debugfs_file(sdp);
2168 return dent ? PTR_ERR(dent) : -ENOMEM;
2169}
2170
2171void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2172{
2173 if (sdp->debugfs_dir) {
2174 if (sdp->debugfs_dentry_glocks) {
2175 debugfs_remove(sdp->debugfs_dentry_glocks);
2176 sdp->debugfs_dentry_glocks = NULL;
2177 }
2178 if (sdp->debugfs_dentry_glstats) {
2179 debugfs_remove(sdp->debugfs_dentry_glstats);
2180 sdp->debugfs_dentry_glstats = NULL;
2181 }
2182 if (sdp->debugfs_dentry_sbstats) {
2183 debugfs_remove(sdp->debugfs_dentry_sbstats);
2184 sdp->debugfs_dentry_sbstats = NULL;
2185 }
2186 debugfs_remove(sdp->debugfs_dir);
2187 sdp->debugfs_dir = NULL;
2188 }
2189}
2190
2191int gfs2_register_debugfs(void)
2192{
2193 gfs2_root = debugfs_create_dir("gfs2", NULL);
2194 if (IS_ERR(gfs2_root))
2195 return PTR_ERR(gfs2_root);
2196 return gfs2_root ? 0 : -ENOMEM;
2197}
2198
2199void gfs2_unregister_debugfs(void)
2200{
2201 debugfs_remove(gfs2_root);
2202 gfs2_root = NULL;
2203}