?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Avoid beyond bounds copy while caching ACL
[zen-stable.git] / fs / gfs2 / glock.c
blob351a3e797789a802669a084d3eef1c771583c6e3
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
47 struct 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
54 typedef void (*glock_examiner) (struct gfs2_glock * gl);
55
56 static 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)
58 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
59
60 static struct dentry *gfs2_root;
61 static struct workqueue_struct *glock_workqueue;
62 struct workqueue_struct *gfs2_delete_workqueue;
63 static LIST_HEAD(lru_list);
64 static atomic_t lru_count = ATOMIC_INIT(0);
65 static 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
71 static struct hlist_bl_head gl_hash_table[GFS2_GL_HASH_SIZE];
72 static 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
81 static 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
94 static inline void spin_lock_bucket(unsigned int hash)
95 {
96 hlist_bl_lock(&gl_hash_table[hash]);
97 }
98
99 static inline void spin_unlock_bucket(unsigned int hash)
100 {
101 hlist_bl_unlock(&gl_hash_table[hash]);
102 }
103
104 static 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
114 void 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
129 void 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
142 static 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
156 void 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
170 static void __gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
171 {
172 if (!list_empty(&gl->gl_lru)) {
173 list_del_init(&gl->gl_lru);
174 atomic_dec(&lru_count);
175 clear_bit(GLF_LRU, &gl->gl_flags);
176 }
177 }
178
179 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
180 {
181 spin_lock(&lru_lock);
182 __gfs2_glock_remove_from_lru(gl);
183 spin_unlock(&lru_lock);
184 }
185
186 /**
187 * __gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
188 * @gl: the glock
189 *
190 * If the glock is demotable, then we add it (or move it) to the end
191 * of the glock LRU list.
192 */
193
194 static void __gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
195 {
196 if (demote_ok(gl))
197 gfs2_glock_add_to_lru(gl);
198 }
199
200 /**
201 * gfs2_glock_put_nolock() - Decrement reference count on glock
202 * @gl: The glock to put
203 *
204 * This function should only be used if the caller has its own reference
205 * to the glock, in addition to the one it is dropping.
206 */
207
208 void gfs2_glock_put_nolock(struct gfs2_glock *gl)
209 {
210 if (atomic_dec_and_test(&gl->gl_ref))
211 GLOCK_BUG_ON(gl, 1);
212 }
213
214 /**
215 * gfs2_glock_put() - Decrement reference count on glock
216 * @gl: The glock to put
217 *
218 */
219
220 void gfs2_glock_put(struct gfs2_glock *gl)
221 {
222 struct gfs2_sbd *sdp = gl->gl_sbd;
223 struct address_space *mapping = gfs2_glock2aspace(gl);
224
225 if (atomic_dec_and_lock(&gl->gl_ref, &lru_lock)) {
226 __gfs2_glock_remove_from_lru(gl);
227 spin_unlock(&lru_lock);
228 spin_lock_bucket(gl->gl_hash);
229 hlist_bl_del_rcu(&gl->gl_list);
230 spin_unlock_bucket(gl->gl_hash);
231 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
232 GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
233 trace_gfs2_glock_put(gl);
234 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
235 }
236 }
237
238 /**
239 * search_bucket() - Find struct gfs2_glock by lock number
240 * @bucket: the bucket to search
241 * @name: The lock name
242 *
243 * Returns: NULL, or the struct gfs2_glock with the requested number
244 */
245
246 static struct gfs2_glock *search_bucket(unsigned int hash,
247 const struct gfs2_sbd *sdp,
248 const struct lm_lockname *name)
249 {
250 struct gfs2_glock *gl;
251 struct hlist_bl_node *h;
252
253 hlist_bl_for_each_entry_rcu(gl, h, &gl_hash_table[hash], gl_list) {
254 if (!lm_name_equal(&gl->gl_name, name))
255 continue;
256 if (gl->gl_sbd != sdp)
257 continue;
258 if (atomic_inc_not_zero(&gl->gl_ref))
259 return gl;
260 }
261
262 return NULL;
263 }
264
265 /**
266 * may_grant - check if its ok to grant a new lock
267 * @gl: The glock
268 * @gh: The lock request which we wish to grant
269 *
270 * Returns: true if its ok to grant the lock
271 */
272
273 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
274 {
275 const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
276 if ((gh->gh_state == LM_ST_EXCLUSIVE ||
277 gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
278 return 0;
279 if (gl->gl_state == gh->gh_state)
280 return 1;
281 if (gh->gh_flags & GL_EXACT)
282 return 0;
283 if (gl->gl_state == LM_ST_EXCLUSIVE) {
284 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
285 return 1;
286 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
287 return 1;
288 }
289 if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
290 return 1;
291 return 0;
292 }
293
294 static void gfs2_holder_wake(struct gfs2_holder *gh)
295 {
296 clear_bit(HIF_WAIT, &gh->gh_iflags);
297 smp_mb__after_clear_bit();
298 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
299 }
300
301 /**
302 * do_error - Something unexpected has happened during a lock request
303 *
304 */
305
306 static inline void do_error(struct gfs2_glock *gl, const int ret)
307 {
308 struct gfs2_holder *gh, *tmp;
309
310 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
311 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
312 continue;
313 if (ret & LM_OUT_ERROR)
314 gh->gh_error = -EIO;
315 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
316 gh->gh_error = GLR_TRYFAILED;
317 else
318 continue;
319 list_del_init(&gh->gh_list);
320 trace_gfs2_glock_queue(gh, 0);
321 gfs2_holder_wake(gh);
322 }
323 }
324
325 /**
326 * do_promote - promote as many requests as possible on the current queue
327 * @gl: The glock
328 *
329 * Returns: 1 if there is a blocked holder at the head of the list, or 2
330 * if a type specific operation is underway.
331 */
332
333 static int do_promote(struct gfs2_glock *gl)
334 __releases(&gl->gl_spin)
335 __acquires(&gl->gl_spin)
336 {
337 const struct gfs2_glock_operations *glops = gl->gl_ops;
338 struct gfs2_holder *gh, *tmp;
339 int ret;
340
341 restart:
342 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
343 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
344 continue;
345 if (may_grant(gl, gh)) {
346 if (gh->gh_list.prev == &gl->gl_holders &&
347 glops->go_lock) {
348 spin_unlock(&gl->gl_spin);
349 /* FIXME: eliminate this eventually */
350 ret = glops->go_lock(gh);
351 spin_lock(&gl->gl_spin);
352 if (ret) {
353 if (ret == 1)
354 return 2;
355 gh->gh_error = ret;
356 list_del_init(&gh->gh_list);
357 trace_gfs2_glock_queue(gh, 0);
358 gfs2_holder_wake(gh);
359 goto restart;
360 }
361 set_bit(HIF_HOLDER, &gh->gh_iflags);
362 trace_gfs2_promote(gh, 1);
363 gfs2_holder_wake(gh);
364 goto restart;
365 }
366 set_bit(HIF_HOLDER, &gh->gh_iflags);
367 trace_gfs2_promote(gh, 0);
368 gfs2_holder_wake(gh);
369 continue;
370 }
371 if (gh->gh_list.prev == &gl->gl_holders)
372 return 1;
373 do_error(gl, 0);
374 break;
375 }
376 return 0;
377 }
378
379 /**
380 * find_first_waiter - find the first gh that's waiting for the glock
381 * @gl: the glock
382 */
383
384 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
385 {
386 struct gfs2_holder *gh;
387
388 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
389 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
390 return gh;
391 }
392 return NULL;
393 }
394
395 /**
396 * state_change - record that the glock is now in a different state
397 * @gl: the glock
398 * @new_state the new state
399 *
400 */
401
402 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
403 {
404 int held1, held2;
405
406 held1 = (gl->gl_state != LM_ST_UNLOCKED);
407 held2 = (new_state != LM_ST_UNLOCKED);
408
409 if (held1 != held2) {
410 if (held2)
411 gfs2_glock_hold(gl);
412 else
413 gfs2_glock_put_nolock(gl);
414 }
415 if (held1 && held2 && list_empty(&gl->gl_holders))
416 clear_bit(GLF_QUEUED, &gl->gl_flags);
417
418 if (new_state != gl->gl_target)
419 /* shorten our minimum hold time */
420 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
421 GL_GLOCK_MIN_HOLD);
422 gl->gl_state = new_state;
423 gl->gl_tchange = jiffies;
424 }
425
426 static void gfs2_demote_wake(struct gfs2_glock *gl)
427 {
428 gl->gl_demote_state = LM_ST_EXCLUSIVE;
429 clear_bit(GLF_DEMOTE, &gl->gl_flags);
430 smp_mb__after_clear_bit();
431 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
432 }
433
434 /**
435 * finish_xmote - The DLM has replied to one of our lock requests
436 * @gl: The glock
437 * @ret: The status from the DLM
438 *
439 */
440
441 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
442 {
443 const struct gfs2_glock_operations *glops = gl->gl_ops;
444 struct gfs2_holder *gh;
445 unsigned state = ret & LM_OUT_ST_MASK;
446 int rv;
447
448 spin_lock(&gl->gl_spin);
449 trace_gfs2_glock_state_change(gl, state);
450 state_change(gl, state);
451 gh = find_first_waiter(gl);
452
453 /* Demote to UN request arrived during demote to SH or DF */
454 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
455 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
456 gl->gl_target = LM_ST_UNLOCKED;
457
458 /* Check for state != intended state */
459 if (unlikely(state != gl->gl_target)) {
460 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
461 /* move to back of queue and try next entry */
462 if (ret & LM_OUT_CANCELED) {
463 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
464 list_move_tail(&gh->gh_list, &gl->gl_holders);
465 gh = find_first_waiter(gl);
466 gl->gl_target = gh->gh_state;
467 goto retry;
468 }
469 /* Some error or failed "try lock" - report it */
470 if ((ret & LM_OUT_ERROR) ||
471 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
472 gl->gl_target = gl->gl_state;
473 do_error(gl, ret);
474 goto out;
475 }
476 }
477 switch(state) {
478 /* Unlocked due to conversion deadlock, try again */
479 case LM_ST_UNLOCKED:
480 retry:
481 do_xmote(gl, gh, gl->gl_target);
482 break;
483 /* Conversion fails, unlock and try again */
484 case LM_ST_SHARED:
485 case LM_ST_DEFERRED:
486 do_xmote(gl, gh, LM_ST_UNLOCKED);
487 break;
488 default: /* Everything else */
489 printk(KERN_ERR "GFS2: wanted %u got %u\n", gl->gl_target, state);
490 GLOCK_BUG_ON(gl, 1);
491 }
492 spin_unlock(&gl->gl_spin);
493 return;
494 }
495
496 /* Fast path - we got what we asked for */
497 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
498 gfs2_demote_wake(gl);
499 if (state != LM_ST_UNLOCKED) {
500 if (glops->go_xmote_bh) {
501 spin_unlock(&gl->gl_spin);
502 rv = glops->go_xmote_bh(gl, gh);
503 spin_lock(&gl->gl_spin);
504 if (rv) {
505 do_error(gl, rv);
506 goto out;
507 }
508 }
509 rv = do_promote(gl);
510 if (rv == 2)
511 goto out_locked;
512 }
513 out:
514 clear_bit(GLF_LOCK, &gl->gl_flags);
515 out_locked:
516 spin_unlock(&gl->gl_spin);
517 }
518
519 /**
520 * do_xmote - Calls the DLM to change the state of a lock
521 * @gl: The lock state
522 * @gh: The holder (only for promotes)
523 * @target: The target lock state
524 *
525 */
526
527 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
528 __releases(&gl->gl_spin)
529 __acquires(&gl->gl_spin)
530 {
531 const struct gfs2_glock_operations *glops = gl->gl_ops;
532 struct gfs2_sbd *sdp = gl->gl_sbd;
533 unsigned int lck_flags = gh ? gh->gh_flags : 0;
534 int ret;
535
536 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
537 LM_FLAG_PRIORITY);
538 GLOCK_BUG_ON(gl, gl->gl_state == target);
539 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
540 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
541 glops->go_inval) {
542 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
543 do_error(gl, 0); /* Fail queued try locks */
544 }
545 gl->gl_req = target;
546 spin_unlock(&gl->gl_spin);
547 if (glops->go_xmote_th)
548 glops->go_xmote_th(gl);
549 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
550 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
551 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
552
553 gfs2_glock_hold(gl);
554 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
555 /* lock_dlm */
556 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
557 GLOCK_BUG_ON(gl, ret);
558 } else { /* lock_nolock */
559 finish_xmote(gl, target);
560 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
561 gfs2_glock_put(gl);
562 }
563
564 spin_lock(&gl->gl_spin);
565 }
566
567 /**
568 * find_first_holder - find the first "holder" gh
569 * @gl: the glock
570 */
571
572 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
573 {
574 struct gfs2_holder *gh;
575
576 if (!list_empty(&gl->gl_holders)) {
577 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
578 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
579 return gh;
580 }
581 return NULL;
582 }
583
584 /**
585 * run_queue - do all outstanding tasks related to a glock
586 * @gl: The glock in question
587 * @nonblock: True if we must not block in run_queue
588 *
589 */
590
591 static void run_queue(struct gfs2_glock *gl, const int nonblock)
592 __releases(&gl->gl_spin)
593 __acquires(&gl->gl_spin)
594 {
595 struct gfs2_holder *gh = NULL;
596 int ret;
597
598 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
599 return;
600
601 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
602
603 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
604 gl->gl_demote_state != gl->gl_state) {
605 if (find_first_holder(gl))
606 goto out_unlock;
607 if (nonblock)
608 goto out_sched;
609 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
610 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
611 gl->gl_target = gl->gl_demote_state;
612 } else {
613 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
614 gfs2_demote_wake(gl);
615 ret = do_promote(gl);
616 if (ret == 0)
617 goto out_unlock;
618 if (ret == 2)
619 goto out;
620 gh = find_first_waiter(gl);
621 gl->gl_target = gh->gh_state;
622 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
623 do_error(gl, 0); /* Fail queued try locks */
624 }
625 do_xmote(gl, gh, gl->gl_target);
626 out:
627 return;
628
629 out_sched:
630 clear_bit(GLF_LOCK, &gl->gl_flags);
631 smp_mb__after_clear_bit();
632 gfs2_glock_hold(gl);
633 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
634 gfs2_glock_put_nolock(gl);
635 return;
636
637 out_unlock:
638 clear_bit(GLF_LOCK, &gl->gl_flags);
639 smp_mb__after_clear_bit();
640 return;
641 }
642
643 static void delete_work_func(struct work_struct *work)
644 {
645 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
646 struct gfs2_sbd *sdp = gl->gl_sbd;
647 struct gfs2_inode *ip;
648 struct inode *inode;
649 u64 no_addr = gl->gl_name.ln_number;
650
651 ip = gl->gl_object;
652 /* Note: Unsafe to dereference ip as we don't hold right refs/locks */
653
654 if (ip)
655 inode = gfs2_ilookup(sdp->sd_vfs, no_addr, 1);
656 else
657 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
658 if (inode && !IS_ERR(inode)) {
659 d_prune_aliases(inode);
660 iput(inode);
661 }
662 gfs2_glock_put(gl);
663 }
664
665 static void glock_work_func(struct work_struct *work)
666 {
667 unsigned long delay = 0;
668 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
669 int drop_ref = 0;
670
671 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
672 finish_xmote(gl, gl->gl_reply);
673 drop_ref = 1;
674 }
675 spin_lock(&gl->gl_spin);
676 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
677 gl->gl_state != LM_ST_UNLOCKED &&
678 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
679 unsigned long holdtime, now = jiffies;
680
681 holdtime = gl->gl_tchange + gl->gl_hold_time;
682 if (time_before(now, holdtime))
683 delay = holdtime - now;
684
685 if (!delay) {
686 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
687 set_bit(GLF_DEMOTE, &gl->gl_flags);
688 }
689 }
690 run_queue(gl, 0);
691 spin_unlock(&gl->gl_spin);
692 if (!delay)
693 gfs2_glock_put(gl);
694 else {
695 if (gl->gl_name.ln_type != LM_TYPE_INODE)
696 delay = 0;
697 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
698 gfs2_glock_put(gl);
699 }
700 if (drop_ref)
701 gfs2_glock_put(gl);
702 }
703
704 /**
705 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
706 * @sdp: The GFS2 superblock
707 * @number: the lock number
708 * @glops: The glock_operations to use
709 * @create: If 0, don't create the glock if it doesn't exist
710 * @glp: the glock is returned here
711 *
712 * This does not lock a glock, just finds/creates structures for one.
713 *
714 * Returns: errno
715 */
716
717 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
718 const struct gfs2_glock_operations *glops, int create,
719 struct gfs2_glock **glp)
720 {
721 struct super_block *s = sdp->sd_vfs;
722 struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
723 struct gfs2_glock *gl, *tmp;
724 unsigned int hash = gl_hash(sdp, &name);
725 struct address_space *mapping;
726 struct kmem_cache *cachep;
727
728 rcu_read_lock();
729 gl = search_bucket(hash, sdp, &name);
730 rcu_read_unlock();
731
732 *glp = gl;
733 if (gl)
734 return 0;
735 if (!create)
736 return -ENOENT;
737
738 if (glops->go_flags & GLOF_ASPACE)
739 cachep = gfs2_glock_aspace_cachep;
740 else
741 cachep = gfs2_glock_cachep;
742 gl = kmem_cache_alloc(cachep, GFP_KERNEL);
743 if (!gl)
744 return -ENOMEM;
745
746 atomic_inc(&sdp->sd_glock_disposal);
747 gl->gl_flags = 0;
748 gl->gl_name = name;
749 atomic_set(&gl->gl_ref, 1);
750 gl->gl_state = LM_ST_UNLOCKED;
751 gl->gl_target = LM_ST_UNLOCKED;
752 gl->gl_demote_state = LM_ST_EXCLUSIVE;
753 gl->gl_hash = hash;
754 gl->gl_ops = glops;
755 snprintf(gl->gl_strname, GDLM_STRNAME_BYTES, "%8x%16llx", name.ln_type, (unsigned long long)number);
756 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
757 gl->gl_lksb.sb_lvbptr = gl->gl_lvb;
758 gl->gl_tchange = jiffies;
759 gl->gl_object = NULL;
760 gl->gl_sbd = sdp;
761 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
762 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
763 INIT_WORK(&gl->gl_delete, delete_work_func);
764
765 mapping = gfs2_glock2aspace(gl);
766 if (mapping) {
767 mapping->a_ops = &gfs2_meta_aops;
768 mapping->host = s->s_bdev->bd_inode;
769 mapping->flags = 0;
770 mapping_set_gfp_mask(mapping, GFP_NOFS);
771 mapping->assoc_mapping = NULL;
772 mapping->backing_dev_info = s->s_bdi;
773 mapping->writeback_index = 0;
774 }
775
776 spin_lock_bucket(hash);
777 tmp = search_bucket(hash, sdp, &name);
778 if (tmp) {
779 spin_unlock_bucket(hash);
780 kmem_cache_free(cachep, gl);
781 atomic_dec(&sdp->sd_glock_disposal);
782 gl = tmp;
783 } else {
784 hlist_bl_add_head_rcu(&gl->gl_list, &gl_hash_table[hash]);
785 spin_unlock_bucket(hash);
786 }
787
788 *glp = gl;
789
790 return 0;
791 }
792
793 /**
794 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
795 * @gl: the glock
796 * @state: the state we're requesting
797 * @flags: the modifier flags
798 * @gh: the holder structure
799 *
800 */
801
802 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
803 struct gfs2_holder *gh)
804 {
805 INIT_LIST_HEAD(&gh->gh_list);
806 gh->gh_gl = gl;
807 gh->gh_ip = (unsigned long)__builtin_return_address(0);
808 gh->gh_owner_pid = get_pid(task_pid(current));
809 gh->gh_state = state;
810 gh->gh_flags = flags;
811 gh->gh_error = 0;
812 gh->gh_iflags = 0;
813 gfs2_glock_hold(gl);
814 }
815
816 /**
817 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
818 * @state: the state we're requesting
819 * @flags: the modifier flags
820 * @gh: the holder structure
821 *
822 * Don't mess with the glock.
823 *
824 */
825
826 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
827 {
828 gh->gh_state = state;
829 gh->gh_flags = flags;
830 gh->gh_iflags = 0;
831 gh->gh_ip = (unsigned long)__builtin_return_address(0);
832 if (gh->gh_owner_pid)
833 put_pid(gh->gh_owner_pid);
834 gh->gh_owner_pid = get_pid(task_pid(current));
835 }
836
837 /**
838 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
839 * @gh: the holder structure
840 *
841 */
842
843 void gfs2_holder_uninit(struct gfs2_holder *gh)
844 {
845 put_pid(gh->gh_owner_pid);
846 gfs2_glock_put(gh->gh_gl);
847 gh->gh_gl = NULL;
848 gh->gh_ip = 0;
849 }
850
851 /**
852 * gfs2_glock_holder_wait
853 * @word: unused
854 *
855 * This function and gfs2_glock_demote_wait both show up in the WCHAN
856 * field. Thus I've separated these otherwise identical functions in
857 * order to be more informative to the user.
858 */
859
860 static int gfs2_glock_holder_wait(void *word)
861 {
862 schedule();
863 return 0;
864 }
865
866 static int gfs2_glock_demote_wait(void *word)
867 {
868 schedule();
869 return 0;
870 }
871
872 static void wait_on_holder(struct gfs2_holder *gh)
873 {
874 unsigned long time1 = jiffies;
875
876 might_sleep();
877 wait_on_bit(&gh->gh_iflags, HIF_WAIT, gfs2_glock_holder_wait, TASK_UNINTERRUPTIBLE);
878 if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
879 /* Lengthen the minimum hold time. */
880 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
881 GL_GLOCK_HOLD_INCR,
882 GL_GLOCK_MAX_HOLD);
883 }
884
885 static void wait_on_demote(struct gfs2_glock *gl)
886 {
887 might_sleep();
888 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, gfs2_glock_demote_wait, TASK_UNINTERRUPTIBLE);
889 }
890
891 /**
892 * handle_callback - process a demote request
893 * @gl: the glock
894 * @state: the state the caller wants us to change to
895 *
896 * There are only two requests that we are going to see in actual
897 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
898 */
899
900 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
901 unsigned long delay)
902 {
903 int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
904
905 set_bit(bit, &gl->gl_flags);
906 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
907 gl->gl_demote_state = state;
908 gl->gl_demote_time = jiffies;
909 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
910 gl->gl_demote_state != state) {
911 gl->gl_demote_state = LM_ST_UNLOCKED;
912 }
913 if (gl->gl_ops->go_callback)
914 gl->gl_ops->go_callback(gl);
915 trace_gfs2_demote_rq(gl);
916 }
917
918 /**
919 * gfs2_glock_wait - wait on a glock acquisition
920 * @gh: the glock holder
921 *
922 * Returns: 0 on success
923 */
924
925 int gfs2_glock_wait(struct gfs2_holder *gh)
926 {
927 wait_on_holder(gh);
928 return gh->gh_error;
929 }
930
931 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
932 {
933 struct va_format vaf;
934 va_list args;
935
936 va_start(args, fmt);
937
938 if (seq) {
939 struct gfs2_glock_iter *gi = seq->private;
940 vsprintf(gi->string, fmt, args);
941 seq_printf(seq, gi->string);
942 } else {
943 vaf.fmt = fmt;
944 vaf.va = &args;
945
946 printk(KERN_ERR " %pV", &vaf);
947 }
948
949 va_end(args);
950 }
951
952 /**
953 * add_to_queue - Add a holder to the wait queue (but look for recursion)
954 * @gh: the holder structure to add
955 *
956 * Eventually we should move the recursive locking trap to a
957 * debugging option or something like that. This is the fast
958 * path and needs to have the minimum number of distractions.
959 *
960 */
961
962 static inline void add_to_queue(struct gfs2_holder *gh)
963 __releases(&gl->gl_spin)
964 __acquires(&gl->gl_spin)
965 {
966 struct gfs2_glock *gl = gh->gh_gl;
967 struct gfs2_sbd *sdp = gl->gl_sbd;
968 struct list_head *insert_pt = NULL;
969 struct gfs2_holder *gh2;
970 int try_lock = 0;
971
972 BUG_ON(gh->gh_owner_pid == NULL);
973 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
974 BUG();
975
976 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
977 if (test_bit(GLF_LOCK, &gl->gl_flags))
978 try_lock = 1;
979 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
980 goto fail;
981 }
982
983 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
984 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
985 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
986 goto trap_recursive;
987 if (try_lock &&
988 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) &&
989 !may_grant(gl, gh)) {
990 fail:
991 gh->gh_error = GLR_TRYFAILED;
992 gfs2_holder_wake(gh);
993 return;
994 }
995 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
996 continue;
997 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
998 insert_pt = &gh2->gh_list;
999 }
1000 set_bit(GLF_QUEUED, &gl->gl_flags);
1001 trace_gfs2_glock_queue(gh, 1);
1002 if (likely(insert_pt == NULL)) {
1003 list_add_tail(&gh->gh_list, &gl->gl_holders);
1004 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1005 goto do_cancel;
1006 return;
1007 }
1008 list_add_tail(&gh->gh_list, insert_pt);
1009 do_cancel:
1010 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
1011 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1012 spin_unlock(&gl->gl_spin);
1013 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1014 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1015 spin_lock(&gl->gl_spin);
1016 }
1017 return;
1018
1019 trap_recursive:
1020 print_symbol(KERN_ERR "original: %s\n", gh2->gh_ip);
1021 printk(KERN_ERR "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1022 printk(KERN_ERR "lock type: %d req lock state : %d\n",
1023 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1024 print_symbol(KERN_ERR "new: %s\n", gh->gh_ip);
1025 printk(KERN_ERR "pid: %d\n", pid_nr(gh->gh_owner_pid));
1026 printk(KERN_ERR "lock type: %d req lock state : %d\n",
1027 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1028 __dump_glock(NULL, gl);
1029 BUG();
1030 }
1031
1032 /**
1033 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1034 * @gh: the holder structure
1035 *
1036 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1037 *
1038 * Returns: 0, GLR_TRYFAILED, or errno on failure
1039 */
1040
1041 int gfs2_glock_nq(struct gfs2_holder *gh)
1042 {
1043 struct gfs2_glock *gl = gh->gh_gl;
1044 struct gfs2_sbd *sdp = gl->gl_sbd;
1045 int error = 0;
1046
1047 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1048 return -EIO;
1049
1050 if (test_bit(GLF_LRU, &gl->gl_flags))
1051 gfs2_glock_remove_from_lru(gl);
1052
1053 spin_lock(&gl->gl_spin);
1054 add_to_queue(gh);
1055 if ((LM_FLAG_NOEXP & gh->gh_flags) &&
1056 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1057 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1058 run_queue(gl, 1);
1059 spin_unlock(&gl->gl_spin);
1060
1061 if (!(gh->gh_flags & GL_ASYNC))
1062 error = gfs2_glock_wait(gh);
1063
1064 return error;
1065 }
1066
1067 /**
1068 * gfs2_glock_poll - poll to see if an async request has been completed
1069 * @gh: the holder
1070 *
1071 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1072 */
1073
1074 int gfs2_glock_poll(struct gfs2_holder *gh)
1075 {
1076 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1077 }
1078
1079 /**
1080 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1081 * @gh: the glock holder
1082 *
1083 */
1084
1085 void gfs2_glock_dq(struct gfs2_holder *gh)
1086 {
1087 struct gfs2_glock *gl = gh->gh_gl;
1088 const struct gfs2_glock_operations *glops = gl->gl_ops;
1089 unsigned delay = 0;
1090 int fast_path = 0;
1091
1092 spin_lock(&gl->gl_spin);
1093 if (gh->gh_flags & GL_NOCACHE)
1094 handle_callback(gl, LM_ST_UNLOCKED, 0);
1095
1096 list_del_init(&gh->gh_list);
1097 if (find_first_holder(gl) == NULL) {
1098 if (glops->go_unlock) {
1099 GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1100 spin_unlock(&gl->gl_spin);
1101 glops->go_unlock(gh);
1102 spin_lock(&gl->gl_spin);
1103 clear_bit(GLF_LOCK, &gl->gl_flags);
1104 }
1105 if (list_empty(&gl->gl_holders) &&
1106 !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1107 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1108 fast_path = 1;
1109 }
1110 if (!test_bit(GLF_LFLUSH, &gl->gl_flags))
1111 __gfs2_glock_schedule_for_reclaim(gl);
1112 trace_gfs2_glock_queue(gh, 0);
1113 spin_unlock(&gl->gl_spin);
1114 if (likely(fast_path))
1115 return;
1116
1117 gfs2_glock_hold(gl);
1118 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1119 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1120 gl->gl_name.ln_type == LM_TYPE_INODE)
1121 delay = gl->gl_hold_time;
1122 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1123 gfs2_glock_put(gl);
1124 }
1125
1126 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1127 {
1128 struct gfs2_glock *gl = gh->gh_gl;
1129 gfs2_glock_dq(gh);
1130 wait_on_demote(gl);
1131 }
1132
1133 /**
1134 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1135 * @gh: the holder structure
1136 *
1137 */
1138
1139 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1140 {
1141 gfs2_glock_dq(gh);
1142 gfs2_holder_uninit(gh);
1143 }
1144
1145 /**
1146 * gfs2_glock_nq_num - acquire a glock based on lock number
1147 * @sdp: the filesystem
1148 * @number: the lock number
1149 * @glops: the glock operations for the type of glock
1150 * @state: the state to acquire the glock in
1151 * @flags: modifier flags for the acquisition
1152 * @gh: the struct gfs2_holder
1153 *
1154 * Returns: errno
1155 */
1156
1157 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1158 const struct gfs2_glock_operations *glops,
1159 unsigned int state, int flags, struct gfs2_holder *gh)
1160 {
1161 struct gfs2_glock *gl;
1162 int error;
1163
1164 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1165 if (!error) {
1166 error = gfs2_glock_nq_init(gl, state, flags, gh);
1167 gfs2_glock_put(gl);
1168 }
1169
1170 return error;
1171 }
1172
1173 /**
1174 * glock_compare - Compare two struct gfs2_glock structures for sorting
1175 * @arg_a: the first structure
1176 * @arg_b: the second structure
1177 *
1178 */
1179
1180 static int glock_compare(const void *arg_a, const void *arg_b)
1181 {
1182 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1183 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1184 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1185 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1186
1187 if (a->ln_number > b->ln_number)
1188 return 1;
1189 if (a->ln_number < b->ln_number)
1190 return -1;
1191 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1192 return 0;
1193 }
1194
1195 /**
1196 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1197 * @num_gh: the number of structures
1198 * @ghs: an array of struct gfs2_holder structures
1199 *
1200 * Returns: 0 on success (all glocks acquired),
1201 * errno on failure (no glocks acquired)
1202 */
1203
1204 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1205 struct gfs2_holder **p)
1206 {
1207 unsigned int x;
1208 int error = 0;
1209
1210 for (x = 0; x < num_gh; x++)
1211 p[x] = &ghs[x];
1212
1213 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1214
1215 for (x = 0; x < num_gh; x++) {
1216 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1217
1218 error = gfs2_glock_nq(p[x]);
1219 if (error) {
1220 while (x--)
1221 gfs2_glock_dq(p[x]);
1222 break;
1223 }
1224 }
1225
1226 return error;
1227 }
1228
1229 /**
1230 * gfs2_glock_nq_m - acquire multiple glocks
1231 * @num_gh: the number of structures
1232 * @ghs: an array of struct gfs2_holder structures
1233 *
1234 *
1235 * Returns: 0 on success (all glocks acquired),
1236 * errno on failure (no glocks acquired)
1237 */
1238
1239 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1240 {
1241 struct gfs2_holder *tmp[4];
1242 struct gfs2_holder **pph = tmp;
1243 int error = 0;
1244
1245 switch(num_gh) {
1246 case 0:
1247 return 0;
1248 case 1:
1249 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1250 return gfs2_glock_nq(ghs);
1251 default:
1252 if (num_gh <= 4)
1253 break;
1254 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1255 if (!pph)
1256 return -ENOMEM;
1257 }
1258
1259 error = nq_m_sync(num_gh, ghs, pph);
1260
1261 if (pph != tmp)
1262 kfree(pph);
1263
1264 return error;
1265 }
1266
1267 /**
1268 * gfs2_glock_dq_m - release multiple glocks
1269 * @num_gh: the number of structures
1270 * @ghs: an array of struct gfs2_holder structures
1271 *
1272 */
1273
1274 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1275 {
1276 while (num_gh--)
1277 gfs2_glock_dq(&ghs[num_gh]);
1278 }
1279
1280 /**
1281 * gfs2_glock_dq_uninit_m - release multiple glocks
1282 * @num_gh: the number of structures
1283 * @ghs: an array of struct gfs2_holder structures
1284 *
1285 */
1286
1287 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1288 {
1289 while (num_gh--)
1290 gfs2_glock_dq_uninit(&ghs[num_gh]);
1291 }
1292
1293 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1294 {
1295 unsigned long delay = 0;
1296 unsigned long holdtime;
1297 unsigned long now = jiffies;
1298
1299 gfs2_glock_hold(gl);
1300 holdtime = gl->gl_tchange + gl->gl_hold_time;
1301 if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1302 gl->gl_name.ln_type == LM_TYPE_INODE) {
1303 if (time_before(now, holdtime))
1304 delay = holdtime - now;
1305 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1306 delay = gl->gl_hold_time;
1307 }
1308
1309 spin_lock(&gl->gl_spin);
1310 handle_callback(gl, state, delay);
1311 spin_unlock(&gl->gl_spin);
1312 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1313 gfs2_glock_put(gl);
1314 }
1315
1316 /**
1317 * gfs2_should_freeze - Figure out if glock should be frozen
1318 * @gl: The glock in question
1319 *
1320 * Glocks are not frozen if (a) the result of the dlm operation is
1321 * an error, (b) the locking operation was an unlock operation or
1322 * (c) if there is a "noexp" flagged request anywhere in the queue
1323 *
1324 * Returns: 1 if freezing should occur, 0 otherwise
1325 */
1326
1327 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1328 {
1329 const struct gfs2_holder *gh;
1330
1331 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1332 return 0;
1333 if (gl->gl_target == LM_ST_UNLOCKED)
1334 return 0;
1335
1336 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1337 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1338 continue;
1339 if (LM_FLAG_NOEXP & gh->gh_flags)
1340 return 0;
1341 }
1342
1343 return 1;
1344 }
1345
1346 /**
1347 * gfs2_glock_complete - Callback used by locking
1348 * @gl: Pointer to the glock
1349 * @ret: The return value from the dlm
1350 *
1351 * The gl_reply field is under the gl_spin lock so that it is ok
1352 * to use a bitfield shared with other glock state fields.
1353 */
1354
1355 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1356 {
1357 struct lm_lockstruct *ls = &gl->gl_sbd->sd_lockstruct;
1358
1359 spin_lock(&gl->gl_spin);
1360 gl->gl_reply = ret;
1361
1362 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1363 if (gfs2_should_freeze(gl)) {
1364 set_bit(GLF_FROZEN, &gl->gl_flags);
1365 spin_unlock(&gl->gl_spin);
1366 return;
1367 }
1368 }
1369
1370 spin_unlock(&gl->gl_spin);
1371 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1372 smp_wmb();
1373 gfs2_glock_hold(gl);
1374 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1375 gfs2_glock_put(gl);
1376 }
1377
1378
1379 static int gfs2_shrink_glock_memory(struct shrinker *shrink,
1380 struct shrink_control *sc)
1381 {
1382 struct gfs2_glock *gl;
1383 int may_demote;
1384 int nr_skipped = 0;
1385 int nr = sc->nr_to_scan;
1386 gfp_t gfp_mask = sc->gfp_mask;
1387 LIST_HEAD(skipped);
1388
1389 if (nr == 0)
1390 goto out;
1391
1392 if (!(gfp_mask & __GFP_FS))
1393 return -1;
1394
1395 spin_lock(&lru_lock);
1396 while(nr && !list_empty(&lru_list)) {
1397 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1398 list_del_init(&gl->gl_lru);
1399 clear_bit(GLF_LRU, &gl->gl_flags);
1400 atomic_dec(&lru_count);
1401
1402 /* Test for being demotable */
1403 if (!test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1404 gfs2_glock_hold(gl);
1405 spin_unlock(&lru_lock);
1406 spin_lock(&gl->gl_spin);
1407 may_demote = demote_ok(gl);
1408 if (may_demote) {
1409 handle_callback(gl, LM_ST_UNLOCKED, 0);
1410 nr--;
1411 }
1412 clear_bit(GLF_LOCK, &gl->gl_flags);
1413 smp_mb__after_clear_bit();
1414 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1415 gfs2_glock_put_nolock(gl);
1416 spin_unlock(&gl->gl_spin);
1417 spin_lock(&lru_lock);
1418 continue;
1419 }
1420 nr_skipped++;
1421 list_add(&gl->gl_lru, &skipped);
1422 set_bit(GLF_LRU, &gl->gl_flags);
1423 }
1424 list_splice(&skipped, &lru_list);
1425 atomic_add(nr_skipped, &lru_count);
1426 spin_unlock(&lru_lock);
1427 out:
1428 return (atomic_read(&lru_count) / 100) * sysctl_vfs_cache_pressure;
1429 }
1430
1431 static struct shrinker glock_shrinker = {
1432 .shrink = gfs2_shrink_glock_memory,
1433 .seeks = DEFAULT_SEEKS,
1434 };
1435
1436 /**
1437 * examine_bucket - Call a function for glock in a hash bucket
1438 * @examiner: the function
1439 * @sdp: the filesystem
1440 * @bucket: the bucket
1441 *
1442 */
1443
1444 static void examine_bucket(glock_examiner examiner, const struct gfs2_sbd *sdp,
1445 unsigned int hash)
1446 {
1447 struct gfs2_glock *gl;
1448 struct hlist_bl_head *head = &gl_hash_table[hash];
1449 struct hlist_bl_node *pos;
1450
1451 rcu_read_lock();
1452 hlist_bl_for_each_entry_rcu(gl, pos, head, gl_list) {
1453 if ((gl->gl_sbd == sdp) && atomic_read(&gl->gl_ref))
1454 examiner(gl);
1455 }
1456 rcu_read_unlock();
1457 cond_resched();
1458 }
1459
1460 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1461 {
1462 unsigned x;
1463
1464 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1465 examine_bucket(examiner, sdp, x);
1466 }
1467
1468
1469 /**
1470 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1471 * @gl: The glock to thaw
1472 *
1473 * N.B. When we freeze a glock, we leave a ref to the glock outstanding,
1474 * so this has to result in the ref count being dropped by one.
1475 */
1476
1477 static void thaw_glock(struct gfs2_glock *gl)
1478 {
1479 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1480 return;
1481 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1482 gfs2_glock_hold(gl);
1483 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1484 gfs2_glock_put(gl);
1485 }
1486
1487 /**
1488 * clear_glock - look at a glock and see if we can free it from glock cache
1489 * @gl: the glock to look at
1490 *
1491 */
1492
1493 static void clear_glock(struct gfs2_glock *gl)
1494 {
1495 gfs2_glock_remove_from_lru(gl);
1496
1497 spin_lock(&gl->gl_spin);
1498 if (gl->gl_state != LM_ST_UNLOCKED)
1499 handle_callback(gl, LM_ST_UNLOCKED, 0);
1500 spin_unlock(&gl->gl_spin);
1501 gfs2_glock_hold(gl);
1502 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1503 gfs2_glock_put(gl);
1504 }
1505
1506 /**
1507 * gfs2_glock_thaw - Thaw any frozen glocks
1508 * @sdp: The super block
1509 *
1510 */
1511
1512 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1513 {
1514 glock_hash_walk(thaw_glock, sdp);
1515 }
1516
1517 static int dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1518 {
1519 int ret;
1520 spin_lock(&gl->gl_spin);
1521 ret = __dump_glock(seq, gl);
1522 spin_unlock(&gl->gl_spin);
1523 return ret;
1524 }
1525
1526 static void dump_glock_func(struct gfs2_glock *gl)
1527 {
1528 dump_glock(NULL, gl);
1529 }
1530
1531 /**
1532 * gfs2_gl_hash_clear - Empty out the glock hash table
1533 * @sdp: the filesystem
1534 * @wait: wait until it's all gone
1535 *
1536 * Called when unmounting the filesystem.
1537 */
1538
1539 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1540 {
1541 glock_hash_walk(clear_glock, sdp);
1542 flush_workqueue(glock_workqueue);
1543 wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
1544 glock_hash_walk(dump_glock_func, sdp);
1545 }
1546
1547 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1548 {
1549 struct gfs2_glock *gl = ip->i_gl;
1550 int ret;
1551
1552 ret = gfs2_truncatei_resume(ip);
1553 gfs2_assert_withdraw(gl->gl_sbd, ret == 0);
1554
1555 spin_lock(&gl->gl_spin);
1556 clear_bit(GLF_LOCK, &gl->gl_flags);
1557 run_queue(gl, 1);
1558 spin_unlock(&gl->gl_spin);
1559 }
1560
1561 static const char *state2str(unsigned state)
1562 {
1563 switch(state) {
1564 case LM_ST_UNLOCKED:
1565 return "UN";
1566 case LM_ST_SHARED:
1567 return "SH";
1568 case LM_ST_DEFERRED:
1569 return "DF";
1570 case LM_ST_EXCLUSIVE:
1571 return "EX";
1572 }
1573 return "??";
1574 }
1575
1576 static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1577 {
1578 char *p = buf;
1579 if (flags & LM_FLAG_TRY)
1580 *p++ = 't';
1581 if (flags & LM_FLAG_TRY_1CB)
1582 *p++ = 'T';
1583 if (flags & LM_FLAG_NOEXP)
1584 *p++ = 'e';
1585 if (flags & LM_FLAG_ANY)
1586 *p++ = 'A';
1587 if (flags & LM_FLAG_PRIORITY)
1588 *p++ = 'p';
1589 if (flags & GL_ASYNC)
1590 *p++ = 'a';
1591 if (flags & GL_EXACT)
1592 *p++ = 'E';
1593 if (flags & GL_NOCACHE)
1594 *p++ = 'c';
1595 if (test_bit(HIF_HOLDER, &iflags))
1596 *p++ = 'H';
1597 if (test_bit(HIF_WAIT, &iflags))
1598 *p++ = 'W';
1599 if (test_bit(HIF_FIRST, &iflags))
1600 *p++ = 'F';
1601 *p = 0;
1602 return buf;
1603 }
1604
1605 /**
1606 * dump_holder - print information about a glock holder
1607 * @seq: the seq_file struct
1608 * @gh: the glock holder
1609 *
1610 * Returns: 0 on success, -ENOBUFS when we run out of space
1611 */
1612
1613 static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1614 {
1615 struct task_struct *gh_owner = NULL;
1616 char flags_buf[32];
1617
1618 if (gh->gh_owner_pid)
1619 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1620 gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1621 state2str(gh->gh_state),
1622 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1623 gh->gh_error,
1624 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1625 gh_owner ? gh_owner->comm : "(ended)",
1626 (void *)gh->gh_ip);
1627 return 0;
1628 }
1629
1630 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1631 {
1632 const unsigned long *gflags = &gl->gl_flags;
1633 char *p = buf;
1634
1635 if (test_bit(GLF_LOCK, gflags))
1636 *p++ = 'l';
1637 if (test_bit(GLF_DEMOTE, gflags))
1638 *p++ = 'D';
1639 if (test_bit(GLF_PENDING_DEMOTE, gflags))
1640 *p++ = 'd';
1641 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1642 *p++ = 'p';
1643 if (test_bit(GLF_DIRTY, gflags))
1644 *p++ = 'y';
1645 if (test_bit(GLF_LFLUSH, gflags))
1646 *p++ = 'f';
1647 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1648 *p++ = 'i';
1649 if (test_bit(GLF_REPLY_PENDING, gflags))
1650 *p++ = 'r';
1651 if (test_bit(GLF_INITIAL, gflags))
1652 *p++ = 'I';
1653 if (test_bit(GLF_FROZEN, gflags))
1654 *p++ = 'F';
1655 if (test_bit(GLF_QUEUED, gflags))
1656 *p++ = 'q';
1657 if (test_bit(GLF_LRU, gflags))
1658 *p++ = 'L';
1659 if (gl->gl_object)
1660 *p++ = 'o';
1661 *p = 0;
1662 return buf;
1663 }
1664
1665 /**
1666 * __dump_glock - print information about a glock
1667 * @seq: The seq_file struct
1668 * @gl: the glock
1669 *
1670 * The file format is as follows:
1671 * One line per object, capital letters are used to indicate objects
1672 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1673 * other objects are indented by a single space and follow the glock to
1674 * which they are related. Fields are indicated by lower case letters
1675 * followed by a colon and the field value, except for strings which are in
1676 * [] so that its possible to see if they are composed of spaces for
1677 * example. The field's are n = number (id of the object), f = flags,
1678 * t = type, s = state, r = refcount, e = error, p = pid.
1679 *
1680 * Returns: 0 on success, -ENOBUFS when we run out of space
1681 */
1682
1683 static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1684 {
1685 const struct gfs2_glock_operations *glops = gl->gl_ops;
1686 unsigned long long dtime;
1687 const struct gfs2_holder *gh;
1688 char gflags_buf[32];
1689 int error = 0;
1690
1691 dtime = jiffies - gl->gl_demote_time;
1692 dtime *= 1000000/HZ; /* demote time in uSec */
1693 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1694 dtime = 0;
1695 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",
1696 state2str(gl->gl_state),
1697 gl->gl_name.ln_type,
1698 (unsigned long long)gl->gl_name.ln_number,
1699 gflags2str(gflags_buf, gl),
1700 state2str(gl->gl_target),
1701 state2str(gl->gl_demote_state), dtime,
1702 atomic_read(&gl->gl_ail_count),
1703 atomic_read(&gl->gl_revokes),
1704 atomic_read(&gl->gl_ref), gl->gl_hold_time);
1705
1706 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1707 error = dump_holder(seq, gh);
1708 if (error)
1709 goto out;
1710 }
1711 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1712 error = glops->go_dump(seq, gl);
1713 out:
1714 return error;
1715 }
1716
1717
1718
1719
1720 int __init gfs2_glock_init(void)
1721 {
1722 unsigned i;
1723 for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
1724 INIT_HLIST_BL_HEAD(&gl_hash_table[i]);
1725 }
1726
1727 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1728 WQ_HIGHPRI | WQ_FREEZABLE, 0);
1729 if (IS_ERR(glock_workqueue))
1730 return PTR_ERR(glock_workqueue);
1731 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1732 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1733 0);
1734 if (IS_ERR(gfs2_delete_workqueue)) {
1735 destroy_workqueue(glock_workqueue);
1736 return PTR_ERR(gfs2_delete_workqueue);
1737 }
1738
1739 register_shrinker(&glock_shrinker);
1740
1741 return 0;
1742 }
1743
1744 void gfs2_glock_exit(void)
1745 {
1746 unregister_shrinker(&glock_shrinker);
1747 destroy_workqueue(glock_workqueue);
1748 destroy_workqueue(gfs2_delete_workqueue);
1749 }
1750
1751 static inline struct gfs2_glock *glock_hash_chain(unsigned hash)
1752 {
1753 return hlist_bl_entry(hlist_bl_first_rcu(&gl_hash_table[hash]),
1754 struct gfs2_glock, gl_list);
1755 }
1756
1757 static inline struct gfs2_glock *glock_hash_next(struct gfs2_glock *gl)
1758 {
1759 return hlist_bl_entry(rcu_dereference(gl->gl_list.next),
1760 struct gfs2_glock, gl_list);
1761 }
1762
1763 static int gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1764 {
1765 struct gfs2_glock *gl;
1766
1767 do {
1768 gl = gi->gl;
1769 if (gl) {
1770 gi->gl = glock_hash_next(gl);
1771 } else {
1772 gi->gl = glock_hash_chain(gi->hash);
1773 }
1774 while (gi->gl == NULL) {
1775 gi->hash++;
1776 if (gi->hash >= GFS2_GL_HASH_SIZE) {
1777 rcu_read_unlock();
1778 return 1;
1779 }
1780 gi->gl = glock_hash_chain(gi->hash);
1781 }
1782 /* Skip entries for other sb and dead entries */
1783 } while (gi->sdp != gi->gl->gl_sbd || atomic_read(&gi->gl->gl_ref) == 0);
1784
1785 return 0;
1786 }
1787
1788 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1789 {
1790 struct gfs2_glock_iter *gi = seq->private;
1791 loff_t n = *pos;
1792
1793 gi->hash = 0;
1794 rcu_read_lock();
1795
1796 do {
1797 if (gfs2_glock_iter_next(gi))
1798 return NULL;
1799 } while (n--);
1800
1801 return gi->gl;
1802 }
1803
1804 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1805 loff_t *pos)
1806 {
1807 struct gfs2_glock_iter *gi = seq->private;
1808
1809 (*pos)++;
1810
1811 if (gfs2_glock_iter_next(gi))
1812 return NULL;
1813
1814 return gi->gl;
1815 }
1816
1817 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1818 {
1819 struct gfs2_glock_iter *gi = seq->private;
1820
1821 if (gi->gl)
1822 rcu_read_unlock();
1823 gi->gl = NULL;
1824 }
1825
1826 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1827 {
1828 return dump_glock(seq, iter_ptr);
1829 }
1830
1831 static const struct seq_operations gfs2_glock_seq_ops = {
1832 .start = gfs2_glock_seq_start,
1833 .next = gfs2_glock_seq_next,
1834 .stop = gfs2_glock_seq_stop,
1835 .show = gfs2_glock_seq_show,
1836 };
1837
1838 static int gfs2_debugfs_open(struct inode *inode, struct file *file)
1839 {
1840 int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1841 sizeof(struct gfs2_glock_iter));
1842 if (ret == 0) {
1843 struct seq_file *seq = file->private_data;
1844 struct gfs2_glock_iter *gi = seq->private;
1845 gi->sdp = inode->i_private;
1846 }
1847 return ret;
1848 }
1849
1850 static const struct file_operations gfs2_debug_fops = {
1851 .owner = THIS_MODULE,
1852 .open = gfs2_debugfs_open,
1853 .read = seq_read,
1854 .llseek = seq_lseek,
1855 .release = seq_release_private,
1856 };
1857
1858 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
1859 {
1860 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
1861 if (!sdp->debugfs_dir)
1862 return -ENOMEM;
1863 sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
1864 S_IFREG | S_IRUGO,
1865 sdp->debugfs_dir, sdp,
1866 &gfs2_debug_fops);
1867 if (!sdp->debugfs_dentry_glocks)
1868 return -ENOMEM;
1869
1870 return 0;
1871 }
1872
1873 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
1874 {
1875 if (sdp && sdp->debugfs_dir) {
1876 if (sdp->debugfs_dentry_glocks) {
1877 debugfs_remove(sdp->debugfs_dentry_glocks);
1878 sdp->debugfs_dentry_glocks = NULL;
1879 }
1880 debugfs_remove(sdp->debugfs_dir);
1881 sdp->debugfs_dir = NULL;
1882 }
1883 }
1884
1885 int gfs2_register_debugfs(void)
1886 {
1887 gfs2_root = debugfs_create_dir("gfs2", NULL);
1888 return gfs2_root ? 0 : -ENOMEM;
1889 }
1890
1891 void gfs2_unregister_debugfs(void)
1892 {
1893 debugfs_remove(gfs2_root);
1894 gfs2_root = NULL;
1895 }