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// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 */
6
7#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
9#include <linux/sched.h>
10#include <linux/slab.h>
11#include <linux/spinlock.h>
12#include <linux/buffer_head.h>
13#include <linux/delay.h>
14#include <linux/sort.h>
15#include <linux/hash.h>
16#include <linux/jhash.h>
17#include <linux/kallsyms.h>
18#include <linux/gfs2_ondisk.h>
19#include <linux/list.h>
20#include <linux/wait.h>
21#include <linux/module.h>
22#include <linux/uaccess.h>
23#include <linux/seq_file.h>
24#include <linux/debugfs.h>
25#include <linux/kthread.h>
26#include <linux/freezer.h>
27#include <linux/workqueue.h>
28#include <linux/jiffies.h>
29#include <linux/rcupdate.h>
30#include <linux/rculist_bl.h>
31#include <linux/bit_spinlock.h>
32#include <linux/percpu.h>
33#include <linux/list_sort.h>
34#include <linux/lockref.h>
35#include <linux/rhashtable.h>
36
37#include "gfs2.h"
38#include "incore.h"
39#include "glock.h"
40#include "glops.h"
41#include "inode.h"
42#include "lops.h"
43#include "meta_io.h"
44#include "quota.h"
45#include "super.h"
46#include "util.h"
47#include "bmap.h"
48#define CREATE_TRACE_POINTS
49#include "trace_gfs2.h"
50
51struct gfs2_glock_iter {
52 struct gfs2_sbd *sdp; /* incore superblock */
53 struct rhashtable_iter hti; /* rhashtable iterator */
54 struct gfs2_glock *gl; /* current glock struct */
55 loff_t last_pos; /* last position */
56};
57
58typedef void (*glock_examiner) (struct gfs2_glock * gl);
59
60static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
61
62static struct dentry *gfs2_root;
63static struct workqueue_struct *glock_workqueue;
64struct workqueue_struct *gfs2_delete_workqueue;
65static LIST_HEAD(lru_list);
66static atomic_t lru_count = ATOMIC_INIT(0);
67static DEFINE_SPINLOCK(lru_lock);
68
69#define GFS2_GL_HASH_SHIFT 15
70#define GFS2_GL_HASH_SIZE BIT(GFS2_GL_HASH_SHIFT)
71
72static const struct rhashtable_params ht_parms = {
73 .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4,
74 .key_len = offsetofend(struct lm_lockname, ln_type),
75 .key_offset = offsetof(struct gfs2_glock, gl_name),
76 .head_offset = offsetof(struct gfs2_glock, gl_node),
77};
78
79static struct rhashtable gl_hash_table;
80
81#define GLOCK_WAIT_TABLE_BITS 12
82#define GLOCK_WAIT_TABLE_SIZE (1 << GLOCK_WAIT_TABLE_BITS)
83static wait_queue_head_t glock_wait_table[GLOCK_WAIT_TABLE_SIZE] __cacheline_aligned;
84
85struct wait_glock_queue {
86 struct lm_lockname *name;
87 wait_queue_entry_t wait;
88};
89
90static int glock_wake_function(wait_queue_entry_t *wait, unsigned int mode,
91 int sync, void *key)
92{
93 struct wait_glock_queue *wait_glock =
94 container_of(wait, struct wait_glock_queue, wait);
95 struct lm_lockname *wait_name = wait_glock->name;
96 struct lm_lockname *wake_name = key;
97
98 if (wake_name->ln_sbd != wait_name->ln_sbd ||
99 wake_name->ln_number != wait_name->ln_number ||
100 wake_name->ln_type != wait_name->ln_type)
101 return 0;
102 return autoremove_wake_function(wait, mode, sync, key);
103}
104
105static wait_queue_head_t *glock_waitqueue(struct lm_lockname *name)
106{
107 u32 hash = jhash2((u32 *)name, ht_parms.key_len / 4, 0);
108
109 return glock_wait_table + hash_32(hash, GLOCK_WAIT_TABLE_BITS);
110}
111
112/**
113 * wake_up_glock - Wake up waiters on a glock
114 * @gl: the glock
115 */
116static void wake_up_glock(struct gfs2_glock *gl)
117{
118 wait_queue_head_t *wq = glock_waitqueue(&gl->gl_name);
119
120 if (waitqueue_active(wq))
121 __wake_up(wq, TASK_NORMAL, 1, &gl->gl_name);
122}
123
124static void gfs2_glock_dealloc(struct rcu_head *rcu)
125{
126 struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
127
128 kfree(gl->gl_lksb.sb_lvbptr);
129 if (gl->gl_ops->go_flags & GLOF_ASPACE)
130 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
131 else
132 kmem_cache_free(gfs2_glock_cachep, gl);
133}
134
135/**
136 * glock_blocked_by_withdraw - determine if we can still use a glock
137 * @gl: the glock
138 *
139 * We need to allow some glocks to be enqueued, dequeued, promoted, and demoted
140 * when we're withdrawn. For example, to maintain metadata integrity, we should
141 * disallow the use of inode and rgrp glocks when withdrawn. Other glocks, like
142 * iopen or the transaction glocks may be safely used because none of their
143 * metadata goes through the journal. So in general, we should disallow all
144 * glocks that are journaled, and allow all the others. One exception is:
145 * we need to allow our active journal to be promoted and demoted so others
146 * may recover it and we can reacquire it when they're done.
147 */
148static bool glock_blocked_by_withdraw(struct gfs2_glock *gl)
149{
150 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
151
152 if (likely(!gfs2_withdrawn(sdp)))
153 return false;
154 if (gl->gl_ops->go_flags & GLOF_NONDISK)
155 return false;
156 if (!sdp->sd_jdesc ||
157 gl->gl_name.ln_number == sdp->sd_jdesc->jd_no_addr)
158 return false;
159 return true;
160}
161
162void gfs2_glock_free(struct gfs2_glock *gl)
163{
164 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
165
166 gfs2_glock_assert_withdraw(gl, atomic_read(&gl->gl_revokes) == 0);
167 rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
168 smp_mb();
169 wake_up_glock(gl);
170 call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
171 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
172 wake_up(&sdp->sd_glock_wait);
173}
174
175/**
176 * gfs2_glock_hold() - increment reference count on glock
177 * @gl: The glock to hold
178 *
179 */
180
181void gfs2_glock_hold(struct gfs2_glock *gl)
182{
183 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
184 lockref_get(&gl->gl_lockref);
185}
186
187/**
188 * demote_ok - Check to see if it's ok to unlock a glock
189 * @gl: the glock
190 *
191 * Returns: 1 if it's ok
192 */
193
194static int demote_ok(const struct gfs2_glock *gl)
195{
196 const struct gfs2_glock_operations *glops = gl->gl_ops;
197
198 if (gl->gl_state == LM_ST_UNLOCKED)
199 return 0;
200 if (!list_empty(&gl->gl_holders))
201 return 0;
202 if (glops->go_demote_ok)
203 return glops->go_demote_ok(gl);
204 return 1;
205}
206
207
208void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
209{
210 if (!(gl->gl_ops->go_flags & GLOF_LRU))
211 return;
212
213 spin_lock(&lru_lock);
214
215 list_move_tail(&gl->gl_lru, &lru_list);
216
217 if (!test_bit(GLF_LRU, &gl->gl_flags)) {
218 set_bit(GLF_LRU, &gl->gl_flags);
219 atomic_inc(&lru_count);
220 }
221
222 spin_unlock(&lru_lock);
223}
224
225static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
226{
227 if (!(gl->gl_ops->go_flags & GLOF_LRU))
228 return;
229
230 spin_lock(&lru_lock);
231 if (test_bit(GLF_LRU, &gl->gl_flags)) {
232 list_del_init(&gl->gl_lru);
233 atomic_dec(&lru_count);
234 clear_bit(GLF_LRU, &gl->gl_flags);
235 }
236 spin_unlock(&lru_lock);
237}
238
239/*
240 * Enqueue the glock on the work queue. Passes one glock reference on to the
241 * work queue.
242 */
243static void __gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
244 if (!queue_delayed_work(glock_workqueue, &gl->gl_work, delay)) {
245 /*
246 * We are holding the lockref spinlock, and the work was still
247 * queued above. The queued work (glock_work_func) takes that
248 * spinlock before dropping its glock reference(s), so it
249 * cannot have dropped them in the meantime.
250 */
251 GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2);
252 gl->gl_lockref.count--;
253 }
254}
255
256static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
257 spin_lock(&gl->gl_lockref.lock);
258 __gfs2_glock_queue_work(gl, delay);
259 spin_unlock(&gl->gl_lockref.lock);
260}
261
262static void __gfs2_glock_put(struct gfs2_glock *gl)
263{
264 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
265 struct address_space *mapping = gfs2_glock2aspace(gl);
266
267 lockref_mark_dead(&gl->gl_lockref);
268
269 gfs2_glock_remove_from_lru(gl);
270 spin_unlock(&gl->gl_lockref.lock);
271 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
272 if (mapping) {
273 truncate_inode_pages_final(mapping);
274 if (!gfs2_withdrawn(sdp))
275 GLOCK_BUG_ON(gl, !mapping_empty(mapping));
276 }
277 trace_gfs2_glock_put(gl);
278 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
279}
280
281/*
282 * Cause the glock to be put in work queue context.
283 */
284void gfs2_glock_queue_put(struct gfs2_glock *gl)
285{
286 gfs2_glock_queue_work(gl, 0);
287}
288
289/**
290 * gfs2_glock_put() - Decrement reference count on glock
291 * @gl: The glock to put
292 *
293 */
294
295void gfs2_glock_put(struct gfs2_glock *gl)
296{
297 if (lockref_put_or_lock(&gl->gl_lockref))
298 return;
299
300 __gfs2_glock_put(gl);
301}
302
303/**
304 * may_grant - check if its ok to grant a new lock
305 * @gl: The glock
306 * @gh: The lock request which we wish to grant
307 *
308 * Returns: true if its ok to grant the lock
309 */
310
311static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
312{
313 const struct gfs2_holder *gh_head = list_first_entry(&gl->gl_holders, const struct gfs2_holder, gh_list);
314
315 if (gh != gh_head) {
316 /**
317 * Here we make a special exception to grant holders who agree
318 * to share the EX lock with other holders who also have the
319 * bit set. If the original holder has the LM_FLAG_NODE_SCOPE bit
320 * is set, we grant more holders with the bit set.
321 */
322 if (gh_head->gh_state == LM_ST_EXCLUSIVE &&
323 (gh_head->gh_flags & LM_FLAG_NODE_SCOPE) &&
324 gh->gh_state == LM_ST_EXCLUSIVE &&
325 (gh->gh_flags & LM_FLAG_NODE_SCOPE))
326 return 1;
327 if ((gh->gh_state == LM_ST_EXCLUSIVE ||
328 gh_head->gh_state == LM_ST_EXCLUSIVE))
329 return 0;
330 }
331 if (gl->gl_state == gh->gh_state)
332 return 1;
333 if (gh->gh_flags & GL_EXACT)
334 return 0;
335 if (gl->gl_state == LM_ST_EXCLUSIVE) {
336 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
337 return 1;
338 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
339 return 1;
340 }
341 if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
342 return 1;
343 return 0;
344}
345
346static void gfs2_holder_wake(struct gfs2_holder *gh)
347{
348 clear_bit(HIF_WAIT, &gh->gh_iflags);
349 smp_mb__after_atomic();
350 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
351 if (gh->gh_flags & GL_ASYNC) {
352 struct gfs2_sbd *sdp = gh->gh_gl->gl_name.ln_sbd;
353
354 wake_up(&sdp->sd_async_glock_wait);
355 }
356}
357
358/**
359 * do_error - Something unexpected has happened during a lock request
360 * @gl: The glock
361 * @ret: The status from the DLM
362 */
363
364static void do_error(struct gfs2_glock *gl, const int ret)
365{
366 struct gfs2_holder *gh, *tmp;
367
368 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
369 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
370 continue;
371 if (ret & LM_OUT_ERROR)
372 gh->gh_error = -EIO;
373 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
374 gh->gh_error = GLR_TRYFAILED;
375 else
376 continue;
377 list_del_init(&gh->gh_list);
378 trace_gfs2_glock_queue(gh, 0);
379 gfs2_holder_wake(gh);
380 }
381}
382
383/**
384 * do_promote - promote as many requests as possible on the current queue
385 * @gl: The glock
386 *
387 * Returns: 1 if there is a blocked holder at the head of the list, or 2
388 * if a type specific operation is underway.
389 */
390
391static int do_promote(struct gfs2_glock *gl)
392__releases(&gl->gl_lockref.lock)
393__acquires(&gl->gl_lockref.lock)
394{
395 const struct gfs2_glock_operations *glops = gl->gl_ops;
396 struct gfs2_holder *gh, *tmp;
397 int ret;
398
399restart:
400 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
401 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
402 continue;
403 if (may_grant(gl, gh)) {
404 if (gh->gh_list.prev == &gl->gl_holders &&
405 glops->go_lock) {
406 spin_unlock(&gl->gl_lockref.lock);
407 /* FIXME: eliminate this eventually */
408 ret = glops->go_lock(gh);
409 spin_lock(&gl->gl_lockref.lock);
410 if (ret) {
411 if (ret == 1)
412 return 2;
413 gh->gh_error = ret;
414 list_del_init(&gh->gh_list);
415 trace_gfs2_glock_queue(gh, 0);
416 gfs2_holder_wake(gh);
417 goto restart;
418 }
419 set_bit(HIF_HOLDER, &gh->gh_iflags);
420 trace_gfs2_promote(gh, 1);
421 gfs2_holder_wake(gh);
422 goto restart;
423 }
424 set_bit(HIF_HOLDER, &gh->gh_iflags);
425 trace_gfs2_promote(gh, 0);
426 gfs2_holder_wake(gh);
427 continue;
428 }
429 if (gh->gh_list.prev == &gl->gl_holders)
430 return 1;
431 do_error(gl, 0);
432 break;
433 }
434 return 0;
435}
436
437/**
438 * find_first_waiter - find the first gh that's waiting for the glock
439 * @gl: the glock
440 */
441
442static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
443{
444 struct gfs2_holder *gh;
445
446 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
447 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
448 return gh;
449 }
450 return NULL;
451}
452
453/**
454 * state_change - record that the glock is now in a different state
455 * @gl: the glock
456 * @new_state: the new state
457 */
458
459static void state_change(struct gfs2_glock *gl, unsigned int new_state)
460{
461 int held1, held2;
462
463 held1 = (gl->gl_state != LM_ST_UNLOCKED);
464 held2 = (new_state != LM_ST_UNLOCKED);
465
466 if (held1 != held2) {
467 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
468 if (held2)
469 gl->gl_lockref.count++;
470 else
471 gl->gl_lockref.count--;
472 }
473 if (new_state != gl->gl_target)
474 /* shorten our minimum hold time */
475 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
476 GL_GLOCK_MIN_HOLD);
477 gl->gl_state = new_state;
478 gl->gl_tchange = jiffies;
479}
480
481static void gfs2_set_demote(struct gfs2_glock *gl)
482{
483 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
484
485 set_bit(GLF_DEMOTE, &gl->gl_flags);
486 smp_mb();
487 wake_up(&sdp->sd_async_glock_wait);
488}
489
490static void gfs2_demote_wake(struct gfs2_glock *gl)
491{
492 gl->gl_demote_state = LM_ST_EXCLUSIVE;
493 clear_bit(GLF_DEMOTE, &gl->gl_flags);
494 smp_mb__after_atomic();
495 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
496}
497
498/**
499 * finish_xmote - The DLM has replied to one of our lock requests
500 * @gl: The glock
501 * @ret: The status from the DLM
502 *
503 */
504
505static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
506{
507 const struct gfs2_glock_operations *glops = gl->gl_ops;
508 struct gfs2_holder *gh;
509 unsigned state = ret & LM_OUT_ST_MASK;
510 int rv;
511
512 spin_lock(&gl->gl_lockref.lock);
513 trace_gfs2_glock_state_change(gl, state);
514 state_change(gl, state);
515 gh = find_first_waiter(gl);
516
517 /* Demote to UN request arrived during demote to SH or DF */
518 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
519 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
520 gl->gl_target = LM_ST_UNLOCKED;
521
522 /* Check for state != intended state */
523 if (unlikely(state != gl->gl_target)) {
524 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
525 /* move to back of queue and try next entry */
526 if (ret & LM_OUT_CANCELED) {
527 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
528 list_move_tail(&gh->gh_list, &gl->gl_holders);
529 gh = find_first_waiter(gl);
530 gl->gl_target = gh->gh_state;
531 goto retry;
532 }
533 /* Some error or failed "try lock" - report it */
534 if ((ret & LM_OUT_ERROR) ||
535 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
536 gl->gl_target = gl->gl_state;
537 do_error(gl, ret);
538 goto out;
539 }
540 }
541 switch(state) {
542 /* Unlocked due to conversion deadlock, try again */
543 case LM_ST_UNLOCKED:
544retry:
545 do_xmote(gl, gh, gl->gl_target);
546 break;
547 /* Conversion fails, unlock and try again */
548 case LM_ST_SHARED:
549 case LM_ST_DEFERRED:
550 do_xmote(gl, gh, LM_ST_UNLOCKED);
551 break;
552 default: /* Everything else */
553 fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n",
554 gl->gl_target, state);
555 GLOCK_BUG_ON(gl, 1);
556 }
557 spin_unlock(&gl->gl_lockref.lock);
558 return;
559 }
560
561 /* Fast path - we got what we asked for */
562 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
563 gfs2_demote_wake(gl);
564 if (state != LM_ST_UNLOCKED) {
565 if (glops->go_xmote_bh) {
566 spin_unlock(&gl->gl_lockref.lock);
567 rv = glops->go_xmote_bh(gl);
568 spin_lock(&gl->gl_lockref.lock);
569 if (rv) {
570 do_error(gl, rv);
571 goto out;
572 }
573 }
574 rv = do_promote(gl);
575 if (rv == 2)
576 goto out_locked;
577 }
578out:
579 clear_bit(GLF_LOCK, &gl->gl_flags);
580out_locked:
581 spin_unlock(&gl->gl_lockref.lock);
582}
583
584static bool is_system_glock(struct gfs2_glock *gl)
585{
586 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
587 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
588
589 if (gl == m_ip->i_gl)
590 return true;
591 return false;
592}
593
594/**
595 * do_xmote - Calls the DLM to change the state of a lock
596 * @gl: The lock state
597 * @gh: The holder (only for promotes)
598 * @target: The target lock state
599 *
600 */
601
602static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
603__releases(&gl->gl_lockref.lock)
604__acquires(&gl->gl_lockref.lock)
605{
606 const struct gfs2_glock_operations *glops = gl->gl_ops;
607 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
608 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
609 int ret;
610
611 if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) &&
612 gh && !(gh->gh_flags & LM_FLAG_NOEXP))
613 return;
614 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
615 LM_FLAG_PRIORITY);
616 GLOCK_BUG_ON(gl, gl->gl_state == target);
617 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
618 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
619 glops->go_inval) {
620 /*
621 * If another process is already doing the invalidate, let that
622 * finish first. The glock state machine will get back to this
623 * holder again later.
624 */
625 if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
626 &gl->gl_flags))
627 return;
628 do_error(gl, 0); /* Fail queued try locks */
629 }
630 gl->gl_req = target;
631 set_bit(GLF_BLOCKING, &gl->gl_flags);
632 if ((gl->gl_req == LM_ST_UNLOCKED) ||
633 (gl->gl_state == LM_ST_EXCLUSIVE) ||
634 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
635 clear_bit(GLF_BLOCKING, &gl->gl_flags);
636 spin_unlock(&gl->gl_lockref.lock);
637 if (glops->go_sync) {
638 ret = glops->go_sync(gl);
639 /* If we had a problem syncing (due to io errors or whatever,
640 * we should not invalidate the metadata or tell dlm to
641 * release the glock to other nodes.
642 */
643 if (ret) {
644 if (cmpxchg(&sdp->sd_log_error, 0, ret)) {
645 fs_err(sdp, "Error %d syncing glock \n", ret);
646 gfs2_dump_glock(NULL, gl, true);
647 }
648 goto skip_inval;
649 }
650 }
651 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) {
652 /*
653 * The call to go_sync should have cleared out the ail list.
654 * If there are still items, we have a problem. We ought to
655 * withdraw, but we can't because the withdraw code also uses
656 * glocks. Warn about the error, dump the glock, then fall
657 * through and wait for logd to do the withdraw for us.
658 */
659 if ((atomic_read(&gl->gl_ail_count) != 0) &&
660 (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) {
661 gfs2_glock_assert_warn(gl,
662 !atomic_read(&gl->gl_ail_count));
663 gfs2_dump_glock(NULL, gl, true);
664 }
665 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
666 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
667 }
668
669skip_inval:
670 gfs2_glock_hold(gl);
671 /*
672 * Check for an error encountered since we called go_sync and go_inval.
673 * If so, we can't withdraw from the glock code because the withdraw
674 * code itself uses glocks (see function signal_our_withdraw) to
675 * change the mount to read-only. Most importantly, we must not call
676 * dlm to unlock the glock until the journal is in a known good state
677 * (after journal replay) otherwise other nodes may use the object
678 * (rgrp or dinode) and then later, journal replay will corrupt the
679 * file system. The best we can do here is wait for the logd daemon
680 * to see sd_log_error and withdraw, and in the meantime, requeue the
681 * work for later.
682 *
683 * We make a special exception for some system glocks, such as the
684 * system statfs inode glock, which needs to be granted before the
685 * gfs2_quotad daemon can exit, and that exit needs to finish before
686 * we can unmount the withdrawn file system.
687 *
688 * However, if we're just unlocking the lock (say, for unmount, when
689 * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
690 * then it's okay to tell dlm to unlock it.
691 */
692 if (unlikely(sdp->sd_log_error && !gfs2_withdrawn(sdp)))
693 gfs2_withdraw_delayed(sdp);
694 if (glock_blocked_by_withdraw(gl) &&
695 (target != LM_ST_UNLOCKED ||
696 test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags))) {
697 if (!is_system_glock(gl)) {
698 gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD);
699 goto out;
700 } else {
701 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
702 }
703 }
704
705 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
706 /* lock_dlm */
707 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
708 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
709 target == LM_ST_UNLOCKED &&
710 test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
711 finish_xmote(gl, target);
712 gfs2_glock_queue_work(gl, 0);
713 } else if (ret) {
714 fs_err(sdp, "lm_lock ret %d\n", ret);
715 GLOCK_BUG_ON(gl, !gfs2_withdrawn(sdp));
716 }
717 } else { /* lock_nolock */
718 finish_xmote(gl, target);
719 gfs2_glock_queue_work(gl, 0);
720 }
721out:
722 spin_lock(&gl->gl_lockref.lock);
723}
724
725/**
726 * find_first_holder - find the first "holder" gh
727 * @gl: the glock
728 */
729
730static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
731{
732 struct gfs2_holder *gh;
733
734 if (!list_empty(&gl->gl_holders)) {
735 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
736 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
737 return gh;
738 }
739 return NULL;
740}
741
742/**
743 * run_queue - do all outstanding tasks related to a glock
744 * @gl: The glock in question
745 * @nonblock: True if we must not block in run_queue
746 *
747 */
748
749static void run_queue(struct gfs2_glock *gl, const int nonblock)
750__releases(&gl->gl_lockref.lock)
751__acquires(&gl->gl_lockref.lock)
752{
753 struct gfs2_holder *gh = NULL;
754 int ret;
755
756 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
757 return;
758
759 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
760
761 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
762 gl->gl_demote_state != gl->gl_state) {
763 if (find_first_holder(gl))
764 goto out_unlock;
765 if (nonblock)
766 goto out_sched;
767 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
768 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
769 gl->gl_target = gl->gl_demote_state;
770 } else {
771 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
772 gfs2_demote_wake(gl);
773 ret = do_promote(gl);
774 if (ret == 0)
775 goto out_unlock;
776 if (ret == 2)
777 goto out;
778 gh = find_first_waiter(gl);
779 gl->gl_target = gh->gh_state;
780 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
781 do_error(gl, 0); /* Fail queued try locks */
782 }
783 do_xmote(gl, gh, gl->gl_target);
784out:
785 return;
786
787out_sched:
788 clear_bit(GLF_LOCK, &gl->gl_flags);
789 smp_mb__after_atomic();
790 gl->gl_lockref.count++;
791 __gfs2_glock_queue_work(gl, 0);
792 return;
793
794out_unlock:
795 clear_bit(GLF_LOCK, &gl->gl_flags);
796 smp_mb__after_atomic();
797 return;
798}
799
800void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation)
801{
802 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
803
804 if (ri->ri_magic == 0)
805 ri->ri_magic = cpu_to_be32(GFS2_MAGIC);
806 if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC))
807 ri->ri_generation_deleted = cpu_to_be64(generation);
808}
809
810bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation)
811{
812 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
813
814 if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC))
815 return false;
816 return generation <= be64_to_cpu(ri->ri_generation_deleted);
817}
818
819static void gfs2_glock_poke(struct gfs2_glock *gl)
820{
821 int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP;
822 struct gfs2_holder gh;
823 int error;
824
825 gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh);
826 error = gfs2_glock_nq(&gh);
827 if (!error)
828 gfs2_glock_dq(&gh);
829 gfs2_holder_uninit(&gh);
830}
831
832static bool gfs2_try_evict(struct gfs2_glock *gl)
833{
834 struct gfs2_inode *ip;
835 bool evicted = false;
836
837 /*
838 * If there is contention on the iopen glock and we have an inode, try
839 * to grab and release the inode so that it can be evicted. This will
840 * allow the remote node to go ahead and delete the inode without us
841 * having to do it, which will avoid rgrp glock thrashing.
842 *
843 * The remote node is likely still holding the corresponding inode
844 * glock, so it will run before we get to verify that the delete has
845 * happened below.
846 */
847 spin_lock(&gl->gl_lockref.lock);
848 ip = gl->gl_object;
849 if (ip && !igrab(&ip->i_inode))
850 ip = NULL;
851 spin_unlock(&gl->gl_lockref.lock);
852 if (ip) {
853 struct gfs2_glock *inode_gl = NULL;
854
855 gl->gl_no_formal_ino = ip->i_no_formal_ino;
856 set_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
857 d_prune_aliases(&ip->i_inode);
858 iput(&ip->i_inode);
859
860 /* If the inode was evicted, gl->gl_object will now be NULL. */
861 spin_lock(&gl->gl_lockref.lock);
862 ip = gl->gl_object;
863 if (ip) {
864 inode_gl = ip->i_gl;
865 lockref_get(&inode_gl->gl_lockref);
866 clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
867 }
868 spin_unlock(&gl->gl_lockref.lock);
869 if (inode_gl) {
870 gfs2_glock_poke(inode_gl);
871 gfs2_glock_put(inode_gl);
872 }
873 evicted = !ip;
874 }
875 return evicted;
876}
877
878static void delete_work_func(struct work_struct *work)
879{
880 struct delayed_work *dwork = to_delayed_work(work);
881 struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete);
882 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
883 struct inode *inode;
884 u64 no_addr = gl->gl_name.ln_number;
885
886 spin_lock(&gl->gl_lockref.lock);
887 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
888 spin_unlock(&gl->gl_lockref.lock);
889
890 if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
891 /*
892 * If we can evict the inode, give the remote node trying to
893 * delete the inode some time before verifying that the delete
894 * has happened. Otherwise, if we cause contention on the inode glock
895 * immediately, the remote node will think that we still have
896 * the inode in use, and so it will give up waiting.
897 *
898 * If we can't evict the inode, signal to the remote node that
899 * the inode is still in use. We'll later try to delete the
900 * inode locally in gfs2_evict_inode.
901 *
902 * FIXME: We only need to verify that the remote node has
903 * deleted the inode because nodes before this remote delete
904 * rework won't cooperate. At a later time, when we no longer
905 * care about compatibility with such nodes, we can skip this
906 * step entirely.
907 */
908 if (gfs2_try_evict(gl)) {
909 if (gfs2_queue_delete_work(gl, 5 * HZ))
910 return;
911 }
912 goto out;
913 }
914
915 inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino,
916 GFS2_BLKST_UNLINKED);
917 if (!IS_ERR_OR_NULL(inode)) {
918 d_prune_aliases(inode);
919 iput(inode);
920 }
921out:
922 gfs2_glock_put(gl);
923}
924
925static void glock_work_func(struct work_struct *work)
926{
927 unsigned long delay = 0;
928 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
929 unsigned int drop_refs = 1;
930
931 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
932 finish_xmote(gl, gl->gl_reply);
933 drop_refs++;
934 }
935 spin_lock(&gl->gl_lockref.lock);
936 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
937 gl->gl_state != LM_ST_UNLOCKED &&
938 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
939 unsigned long holdtime, now = jiffies;
940
941 holdtime = gl->gl_tchange + gl->gl_hold_time;
942 if (time_before(now, holdtime))
943 delay = holdtime - now;
944
945 if (!delay) {
946 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
947 gfs2_set_demote(gl);
948 }
949 }
950 run_queue(gl, 0);
951 if (delay) {
952 /* Keep one glock reference for the work we requeue. */
953 drop_refs--;
954 if (gl->gl_name.ln_type != LM_TYPE_INODE)
955 delay = 0;
956 __gfs2_glock_queue_work(gl, delay);
957 }
958
959 /*
960 * Drop the remaining glock references manually here. (Mind that
961 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
962 * here as well.)
963 */
964 gl->gl_lockref.count -= drop_refs;
965 if (!gl->gl_lockref.count) {
966 __gfs2_glock_put(gl);
967 return;
968 }
969 spin_unlock(&gl->gl_lockref.lock);
970}
971
972static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
973 struct gfs2_glock *new)
974{
975 struct wait_glock_queue wait;
976 wait_queue_head_t *wq = glock_waitqueue(name);
977 struct gfs2_glock *gl;
978
979 wait.name = name;
980 init_wait(&wait.wait);
981 wait.wait.func = glock_wake_function;
982
983again:
984 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
985 rcu_read_lock();
986 if (new) {
987 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
988 &new->gl_node, ht_parms);
989 if (IS_ERR(gl))
990 goto out;
991 } else {
992 gl = rhashtable_lookup_fast(&gl_hash_table,
993 name, ht_parms);
994 }
995 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
996 rcu_read_unlock();
997 schedule();
998 goto again;
999 }
1000out:
1001 rcu_read_unlock();
1002 finish_wait(wq, &wait.wait);
1003 return gl;
1004}
1005
1006/**
1007 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
1008 * @sdp: The GFS2 superblock
1009 * @number: the lock number
1010 * @glops: The glock_operations to use
1011 * @create: If 0, don't create the glock if it doesn't exist
1012 * @glp: the glock is returned here
1013 *
1014 * This does not lock a glock, just finds/creates structures for one.
1015 *
1016 * Returns: errno
1017 */
1018
1019int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
1020 const struct gfs2_glock_operations *glops, int create,
1021 struct gfs2_glock **glp)
1022{
1023 struct super_block *s = sdp->sd_vfs;
1024 struct lm_lockname name = { .ln_number = number,
1025 .ln_type = glops->go_type,
1026 .ln_sbd = sdp };
1027 struct gfs2_glock *gl, *tmp;
1028 struct address_space *mapping;
1029 struct kmem_cache *cachep;
1030 int ret = 0;
1031
1032 gl = find_insert_glock(&name, NULL);
1033 if (gl) {
1034 *glp = gl;
1035 return 0;
1036 }
1037 if (!create)
1038 return -ENOENT;
1039
1040 if (glops->go_flags & GLOF_ASPACE)
1041 cachep = gfs2_glock_aspace_cachep;
1042 else
1043 cachep = gfs2_glock_cachep;
1044 gl = kmem_cache_alloc(cachep, GFP_NOFS);
1045 if (!gl)
1046 return -ENOMEM;
1047
1048 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
1049
1050 if (glops->go_flags & GLOF_LVB) {
1051 gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1052 if (!gl->gl_lksb.sb_lvbptr) {
1053 kmem_cache_free(cachep, gl);
1054 return -ENOMEM;
1055 }
1056 }
1057
1058 atomic_inc(&sdp->sd_glock_disposal);
1059 gl->gl_node.next = NULL;
1060 gl->gl_flags = 0;
1061 gl->gl_name = name;
1062 lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass);
1063 gl->gl_lockref.count = 1;
1064 gl->gl_state = LM_ST_UNLOCKED;
1065 gl->gl_target = LM_ST_UNLOCKED;
1066 gl->gl_demote_state = LM_ST_EXCLUSIVE;
1067 gl->gl_ops = glops;
1068 gl->gl_dstamp = 0;
1069 preempt_disable();
1070 /* We use the global stats to estimate the initial per-glock stats */
1071 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
1072 preempt_enable();
1073 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
1074 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
1075 gl->gl_tchange = jiffies;
1076 gl->gl_object = NULL;
1077 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
1078 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
1079 if (gl->gl_name.ln_type == LM_TYPE_IOPEN)
1080 INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func);
1081
1082 mapping = gfs2_glock2aspace(gl);
1083 if (mapping) {
1084 mapping->a_ops = &gfs2_meta_aops;
1085 mapping->host = s->s_bdev->bd_inode;
1086 mapping->flags = 0;
1087 mapping_set_gfp_mask(mapping, GFP_NOFS);
1088 mapping->private_data = NULL;
1089 mapping->writeback_index = 0;
1090 }
1091
1092 tmp = find_insert_glock(&name, gl);
1093 if (!tmp) {
1094 *glp = gl;
1095 goto out;
1096 }
1097 if (IS_ERR(tmp)) {
1098 ret = PTR_ERR(tmp);
1099 goto out_free;
1100 }
1101 *glp = tmp;
1102
1103out_free:
1104 kfree(gl->gl_lksb.sb_lvbptr);
1105 kmem_cache_free(cachep, gl);
1106 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
1107 wake_up(&sdp->sd_glock_wait);
1108
1109out:
1110 return ret;
1111}
1112
1113/**
1114 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
1115 * @gl: the glock
1116 * @state: the state we're requesting
1117 * @flags: the modifier flags
1118 * @gh: the holder structure
1119 *
1120 */
1121
1122void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
1123 struct gfs2_holder *gh)
1124{
1125 INIT_LIST_HEAD(&gh->gh_list);
1126 gh->gh_gl = gl;
1127 gh->gh_ip = _RET_IP_;
1128 gh->gh_owner_pid = get_pid(task_pid(current));
1129 gh->gh_state = state;
1130 gh->gh_flags = flags;
1131 gh->gh_error = 0;
1132 gh->gh_iflags = 0;
1133 gfs2_glock_hold(gl);
1134}
1135
1136/**
1137 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1138 * @state: the state we're requesting
1139 * @flags: the modifier flags
1140 * @gh: the holder structure
1141 *
1142 * Don't mess with the glock.
1143 *
1144 */
1145
1146void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
1147{
1148 gh->gh_state = state;
1149 gh->gh_flags = flags;
1150 gh->gh_iflags = 0;
1151 gh->gh_ip = _RET_IP_;
1152 put_pid(gh->gh_owner_pid);
1153 gh->gh_owner_pid = get_pid(task_pid(current));
1154}
1155
1156/**
1157 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1158 * @gh: the holder structure
1159 *
1160 */
1161
1162void gfs2_holder_uninit(struct gfs2_holder *gh)
1163{
1164 put_pid(gh->gh_owner_pid);
1165 gfs2_glock_put(gh->gh_gl);
1166 gfs2_holder_mark_uninitialized(gh);
1167 gh->gh_ip = 0;
1168}
1169
1170static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
1171 unsigned long start_time)
1172{
1173 /* Have we waited longer that a second? */
1174 if (time_after(jiffies, start_time + HZ)) {
1175 /* Lengthen the minimum hold time. */
1176 gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
1177 GL_GLOCK_MAX_HOLD);
1178 }
1179}
1180
1181/**
1182 * gfs2_glock_wait - wait on a glock acquisition
1183 * @gh: the glock holder
1184 *
1185 * Returns: 0 on success
1186 */
1187
1188int gfs2_glock_wait(struct gfs2_holder *gh)
1189{
1190 unsigned long start_time = jiffies;
1191
1192 might_sleep();
1193 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1194 gfs2_glock_update_hold_time(gh->gh_gl, start_time);
1195 return gh->gh_error;
1196}
1197
1198static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
1199{
1200 int i;
1201
1202 for (i = 0; i < num_gh; i++)
1203 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
1204 return 1;
1205 return 0;
1206}
1207
1208/**
1209 * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1210 * @num_gh: the number of holders in the array
1211 * @ghs: the glock holder array
1212 *
1213 * Returns: 0 on success, meaning all glocks have been granted and are held.
1214 * -ESTALE if the request timed out, meaning all glocks were released,
1215 * and the caller should retry the operation.
1216 */
1217
1218int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
1219{
1220 struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
1221 int i, ret = 0, timeout = 0;
1222 unsigned long start_time = jiffies;
1223 bool keep_waiting;
1224
1225 might_sleep();
1226 /*
1227 * Total up the (minimum hold time * 2) of all glocks and use that to
1228 * determine the max amount of time we should wait.
1229 */
1230 for (i = 0; i < num_gh; i++)
1231 timeout += ghs[i].gh_gl->gl_hold_time << 1;
1232
1233wait_for_dlm:
1234 if (!wait_event_timeout(sdp->sd_async_glock_wait,
1235 !glocks_pending(num_gh, ghs), timeout))
1236 ret = -ESTALE; /* request timed out. */
1237
1238 /*
1239 * If dlm granted all our requests, we need to adjust the glock
1240 * minimum hold time values according to how long we waited.
1241 *
1242 * If our request timed out, we need to repeatedly release any held
1243 * glocks we acquired thus far to allow dlm to acquire the remaining
1244 * glocks without deadlocking. We cannot currently cancel outstanding
1245 * glock acquisitions.
1246 *
1247 * The HIF_WAIT bit tells us which requests still need a response from
1248 * dlm.
1249 *
1250 * If dlm sent us any errors, we return the first error we find.
1251 */
1252 keep_waiting = false;
1253 for (i = 0; i < num_gh; i++) {
1254 /* Skip holders we have already dequeued below. */
1255 if (!gfs2_holder_queued(&ghs[i]))
1256 continue;
1257 /* Skip holders with a pending DLM response. */
1258 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags)) {
1259 keep_waiting = true;
1260 continue;
1261 }
1262
1263 if (test_bit(HIF_HOLDER, &ghs[i].gh_iflags)) {
1264 if (ret == -ESTALE)
1265 gfs2_glock_dq(&ghs[i]);
1266 else
1267 gfs2_glock_update_hold_time(ghs[i].gh_gl,
1268 start_time);
1269 }
1270 if (!ret)
1271 ret = ghs[i].gh_error;
1272 }
1273
1274 if (keep_waiting)
1275 goto wait_for_dlm;
1276
1277 /*
1278 * At this point, we've either acquired all locks or released them all.
1279 */
1280 return ret;
1281}
1282
1283/**
1284 * handle_callback - process a demote request
1285 * @gl: the glock
1286 * @state: the state the caller wants us to change to
1287 * @delay: zero to demote immediately; otherwise pending demote
1288 * @remote: true if this came from a different cluster node
1289 *
1290 * There are only two requests that we are going to see in actual
1291 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1292 */
1293
1294static void handle_callback(struct gfs2_glock *gl, unsigned int state,
1295 unsigned long delay, bool remote)
1296{
1297 if (delay)
1298 set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1299 else
1300 gfs2_set_demote(gl);
1301 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
1302 gl->gl_demote_state = state;
1303 gl->gl_demote_time = jiffies;
1304 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
1305 gl->gl_demote_state != state) {
1306 gl->gl_demote_state = LM_ST_UNLOCKED;
1307 }
1308 if (gl->gl_ops->go_callback)
1309 gl->gl_ops->go_callback(gl, remote);
1310 trace_gfs2_demote_rq(gl, remote);
1311}
1312
1313void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
1314{
1315 struct va_format vaf;
1316 va_list args;
1317
1318 va_start(args, fmt);
1319
1320 if (seq) {
1321 seq_vprintf(seq, fmt, args);
1322 } else {
1323 vaf.fmt = fmt;
1324 vaf.va = &args;
1325
1326 pr_err("%pV", &vaf);
1327 }
1328
1329 va_end(args);
1330}
1331
1332/**
1333 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1334 * @gh: the holder structure to add
1335 *
1336 * Eventually we should move the recursive locking trap to a
1337 * debugging option or something like that. This is the fast
1338 * path and needs to have the minimum number of distractions.
1339 *
1340 */
1341
1342static inline void add_to_queue(struct gfs2_holder *gh)
1343__releases(&gl->gl_lockref.lock)
1344__acquires(&gl->gl_lockref.lock)
1345{
1346 struct gfs2_glock *gl = gh->gh_gl;
1347 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1348 struct list_head *insert_pt = NULL;
1349 struct gfs2_holder *gh2;
1350 int try_futile = 0;
1351
1352 GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1353 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1354 GLOCK_BUG_ON(gl, true);
1355
1356 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1357 if (test_bit(GLF_LOCK, &gl->gl_flags))
1358 try_futile = !may_grant(gl, gh);
1359 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1360 goto fail;
1361 }
1362
1363 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1364 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
1365 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
1366 goto trap_recursive;
1367 if (try_futile &&
1368 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1369fail:
1370 gh->gh_error = GLR_TRYFAILED;
1371 gfs2_holder_wake(gh);
1372 return;
1373 }
1374 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1375 continue;
1376 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1377 insert_pt = &gh2->gh_list;
1378 }
1379 trace_gfs2_glock_queue(gh, 1);
1380 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1381 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1382 if (likely(insert_pt == NULL)) {
1383 list_add_tail(&gh->gh_list, &gl->gl_holders);
1384 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1385 goto do_cancel;
1386 return;
1387 }
1388 list_add_tail(&gh->gh_list, insert_pt);
1389do_cancel:
1390 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
1391 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1392 spin_unlock(&gl->gl_lockref.lock);
1393 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1394 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1395 spin_lock(&gl->gl_lockref.lock);
1396 }
1397 return;
1398
1399trap_recursive:
1400 fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1401 fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1402 fs_err(sdp, "lock type: %d req lock state : %d\n",
1403 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1404 fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1405 fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1406 fs_err(sdp, "lock type: %d req lock state : %d\n",
1407 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1408 gfs2_dump_glock(NULL, gl, true);
1409 BUG();
1410}
1411
1412/**
1413 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1414 * @gh: the holder structure
1415 *
1416 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1417 *
1418 * Returns: 0, GLR_TRYFAILED, or errno on failure
1419 */
1420
1421int gfs2_glock_nq(struct gfs2_holder *gh)
1422{
1423 struct gfs2_glock *gl = gh->gh_gl;
1424 int error = 0;
1425
1426 if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP))
1427 return -EIO;
1428
1429 if (test_bit(GLF_LRU, &gl->gl_flags))
1430 gfs2_glock_remove_from_lru(gl);
1431
1432 spin_lock(&gl->gl_lockref.lock);
1433 add_to_queue(gh);
1434 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1435 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1436 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1437 gl->gl_lockref.count++;
1438 __gfs2_glock_queue_work(gl, 0);
1439 }
1440 run_queue(gl, 1);
1441 spin_unlock(&gl->gl_lockref.lock);
1442
1443 if (!(gh->gh_flags & GL_ASYNC))
1444 error = gfs2_glock_wait(gh);
1445
1446 return error;
1447}
1448
1449/**
1450 * gfs2_glock_poll - poll to see if an async request has been completed
1451 * @gh: the holder
1452 *
1453 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1454 */
1455
1456int gfs2_glock_poll(struct gfs2_holder *gh)
1457{
1458 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1459}
1460
1461/**
1462 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1463 * @gh: the glock holder
1464 *
1465 */
1466
1467void gfs2_glock_dq(struct gfs2_holder *gh)
1468{
1469 struct gfs2_glock *gl = gh->gh_gl;
1470 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1471 unsigned delay = 0;
1472 int fast_path = 0;
1473
1474 spin_lock(&gl->gl_lockref.lock);
1475 /*
1476 * If we're in the process of file system withdraw, we cannot just
1477 * dequeue any glocks until our journal is recovered, lest we
1478 * introduce file system corruption. We need two exceptions to this
1479 * rule: We need to allow unlocking of nondisk glocks and the glock
1480 * for our own journal that needs recovery.
1481 */
1482 if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) &&
1483 glock_blocked_by_withdraw(gl) &&
1484 gh->gh_gl != sdp->sd_jinode_gl) {
1485 sdp->sd_glock_dqs_held++;
1486 spin_unlock(&gl->gl_lockref.lock);
1487 might_sleep();
1488 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
1489 TASK_UNINTERRUPTIBLE);
1490 spin_lock(&gl->gl_lockref.lock);
1491 }
1492 if (gh->gh_flags & GL_NOCACHE)
1493 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1494
1495 list_del_init(&gh->gh_list);
1496 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1497 if (find_first_holder(gl) == NULL) {
1498 if (list_empty(&gl->gl_holders) &&
1499 !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1500 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1501 fast_path = 1;
1502 }
1503 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1504 gfs2_glock_add_to_lru(gl);
1505
1506 trace_gfs2_glock_queue(gh, 0);
1507 if (unlikely(!fast_path)) {
1508 gl->gl_lockref.count++;
1509 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1510 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1511 gl->gl_name.ln_type == LM_TYPE_INODE)
1512 delay = gl->gl_hold_time;
1513 __gfs2_glock_queue_work(gl, delay);
1514 }
1515 spin_unlock(&gl->gl_lockref.lock);
1516}
1517
1518void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1519{
1520 struct gfs2_glock *gl = gh->gh_gl;
1521 gfs2_glock_dq(gh);
1522 might_sleep();
1523 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1524}
1525
1526/**
1527 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1528 * @gh: the holder structure
1529 *
1530 */
1531
1532void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1533{
1534 gfs2_glock_dq(gh);
1535 gfs2_holder_uninit(gh);
1536}
1537
1538/**
1539 * gfs2_glock_nq_num - acquire a glock based on lock number
1540 * @sdp: the filesystem
1541 * @number: the lock number
1542 * @glops: the glock operations for the type of glock
1543 * @state: the state to acquire the glock in
1544 * @flags: modifier flags for the acquisition
1545 * @gh: the struct gfs2_holder
1546 *
1547 * Returns: errno
1548 */
1549
1550int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1551 const struct gfs2_glock_operations *glops,
1552 unsigned int state, u16 flags, struct gfs2_holder *gh)
1553{
1554 struct gfs2_glock *gl;
1555 int error;
1556
1557 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1558 if (!error) {
1559 error = gfs2_glock_nq_init(gl, state, flags, gh);
1560 gfs2_glock_put(gl);
1561 }
1562
1563 return error;
1564}
1565
1566/**
1567 * glock_compare - Compare two struct gfs2_glock structures for sorting
1568 * @arg_a: the first structure
1569 * @arg_b: the second structure
1570 *
1571 */
1572
1573static int glock_compare(const void *arg_a, const void *arg_b)
1574{
1575 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1576 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1577 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1578 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1579
1580 if (a->ln_number > b->ln_number)
1581 return 1;
1582 if (a->ln_number < b->ln_number)
1583 return -1;
1584 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1585 return 0;
1586}
1587
1588/**
1589 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1590 * @num_gh: the number of structures
1591 * @ghs: an array of struct gfs2_holder structures
1592 * @p: placeholder for the holder structure to pass back
1593 *
1594 * Returns: 0 on success (all glocks acquired),
1595 * errno on failure (no glocks acquired)
1596 */
1597
1598static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1599 struct gfs2_holder **p)
1600{
1601 unsigned int x;
1602 int error = 0;
1603
1604 for (x = 0; x < num_gh; x++)
1605 p[x] = &ghs[x];
1606
1607 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1608
1609 for (x = 0; x < num_gh; x++) {
1610 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1611
1612 error = gfs2_glock_nq(p[x]);
1613 if (error) {
1614 while (x--)
1615 gfs2_glock_dq(p[x]);
1616 break;
1617 }
1618 }
1619
1620 return error;
1621}
1622
1623/**
1624 * gfs2_glock_nq_m - acquire multiple glocks
1625 * @num_gh: the number of structures
1626 * @ghs: an array of struct gfs2_holder structures
1627 *
1628 *
1629 * Returns: 0 on success (all glocks acquired),
1630 * errno on failure (no glocks acquired)
1631 */
1632
1633int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1634{
1635 struct gfs2_holder *tmp[4];
1636 struct gfs2_holder **pph = tmp;
1637 int error = 0;
1638
1639 switch(num_gh) {
1640 case 0:
1641 return 0;
1642 case 1:
1643 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1644 return gfs2_glock_nq(ghs);
1645 default:
1646 if (num_gh <= 4)
1647 break;
1648 pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *),
1649 GFP_NOFS);
1650 if (!pph)
1651 return -ENOMEM;
1652 }
1653
1654 error = nq_m_sync(num_gh, ghs, pph);
1655
1656 if (pph != tmp)
1657 kfree(pph);
1658
1659 return error;
1660}
1661
1662/**
1663 * gfs2_glock_dq_m - release multiple glocks
1664 * @num_gh: the number of structures
1665 * @ghs: an array of struct gfs2_holder structures
1666 *
1667 */
1668
1669void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1670{
1671 while (num_gh--)
1672 gfs2_glock_dq(&ghs[num_gh]);
1673}
1674
1675void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1676{
1677 unsigned long delay = 0;
1678 unsigned long holdtime;
1679 unsigned long now = jiffies;
1680
1681 gfs2_glock_hold(gl);
1682 spin_lock(&gl->gl_lockref.lock);
1683 holdtime = gl->gl_tchange + gl->gl_hold_time;
1684 if (!list_empty(&gl->gl_holders) &&
1685 gl->gl_name.ln_type == LM_TYPE_INODE) {
1686 if (time_before(now, holdtime))
1687 delay = holdtime - now;
1688 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1689 delay = gl->gl_hold_time;
1690 }
1691 handle_callback(gl, state, delay, true);
1692 __gfs2_glock_queue_work(gl, delay);
1693 spin_unlock(&gl->gl_lockref.lock);
1694}
1695
1696/**
1697 * gfs2_should_freeze - Figure out if glock should be frozen
1698 * @gl: The glock in question
1699 *
1700 * Glocks are not frozen if (a) the result of the dlm operation is
1701 * an error, (b) the locking operation was an unlock operation or
1702 * (c) if there is a "noexp" flagged request anywhere in the queue
1703 *
1704 * Returns: 1 if freezing should occur, 0 otherwise
1705 */
1706
1707static int gfs2_should_freeze(const struct gfs2_glock *gl)
1708{
1709 const struct gfs2_holder *gh;
1710
1711 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1712 return 0;
1713 if (gl->gl_target == LM_ST_UNLOCKED)
1714 return 0;
1715
1716 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1717 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1718 continue;
1719 if (LM_FLAG_NOEXP & gh->gh_flags)
1720 return 0;
1721 }
1722
1723 return 1;
1724}
1725
1726/**
1727 * gfs2_glock_complete - Callback used by locking
1728 * @gl: Pointer to the glock
1729 * @ret: The return value from the dlm
1730 *
1731 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1732 * to use a bitfield shared with other glock state fields.
1733 */
1734
1735void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1736{
1737 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1738
1739 spin_lock(&gl->gl_lockref.lock);
1740 gl->gl_reply = ret;
1741
1742 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1743 if (gfs2_should_freeze(gl)) {
1744 set_bit(GLF_FROZEN, &gl->gl_flags);
1745 spin_unlock(&gl->gl_lockref.lock);
1746 return;
1747 }
1748 }
1749
1750 gl->gl_lockref.count++;
1751 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1752 __gfs2_glock_queue_work(gl, 0);
1753 spin_unlock(&gl->gl_lockref.lock);
1754}
1755
1756static int glock_cmp(void *priv, const struct list_head *a,
1757 const struct list_head *b)
1758{
1759 struct gfs2_glock *gla, *glb;
1760
1761 gla = list_entry(a, struct gfs2_glock, gl_lru);
1762 glb = list_entry(b, struct gfs2_glock, gl_lru);
1763
1764 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1765 return 1;
1766 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1767 return -1;
1768
1769 return 0;
1770}
1771
1772/**
1773 * gfs2_dispose_glock_lru - Demote a list of glocks
1774 * @list: The list to dispose of
1775 *
1776 * Disposing of glocks may involve disk accesses, so that here we sort
1777 * the glocks by number (i.e. disk location of the inodes) so that if
1778 * there are any such accesses, they'll be sent in order (mostly).
1779 *
1780 * Must be called under the lru_lock, but may drop and retake this
1781 * lock. While the lru_lock is dropped, entries may vanish from the
1782 * list, but no new entries will appear on the list (since it is
1783 * private)
1784 */
1785
1786static void gfs2_dispose_glock_lru(struct list_head *list)
1787__releases(&lru_lock)
1788__acquires(&lru_lock)
1789{
1790 struct gfs2_glock *gl;
1791
1792 list_sort(NULL, list, glock_cmp);
1793
1794 while(!list_empty(list)) {
1795 gl = list_first_entry(list, struct gfs2_glock, gl_lru);
1796 list_del_init(&gl->gl_lru);
1797 clear_bit(GLF_LRU, &gl->gl_flags);
1798 if (!spin_trylock(&gl->gl_lockref.lock)) {
1799add_back_to_lru:
1800 list_add(&gl->gl_lru, &lru_list);
1801 set_bit(GLF_LRU, &gl->gl_flags);
1802 atomic_inc(&lru_count);
1803 continue;
1804 }
1805 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1806 spin_unlock(&gl->gl_lockref.lock);
1807 goto add_back_to_lru;
1808 }
1809 gl->gl_lockref.count++;
1810 if (demote_ok(gl))
1811 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1812 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1813 __gfs2_glock_queue_work(gl, 0);
1814 spin_unlock(&gl->gl_lockref.lock);
1815 cond_resched_lock(&lru_lock);
1816 }
1817}
1818
1819/**
1820 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1821 * @nr: The number of entries to scan
1822 *
1823 * This function selects the entries on the LRU which are able to
1824 * be demoted, and then kicks off the process by calling
1825 * gfs2_dispose_glock_lru() above.
1826 */
1827
1828static long gfs2_scan_glock_lru(int nr)
1829{
1830 struct gfs2_glock *gl;
1831 LIST_HEAD(skipped);
1832 LIST_HEAD(dispose);
1833 long freed = 0;
1834
1835 spin_lock(&lru_lock);
1836 while ((nr-- >= 0) && !list_empty(&lru_list)) {
1837 gl = list_first_entry(&lru_list, struct gfs2_glock, gl_lru);
1838
1839 /* Test for being demotable */
1840 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1841 list_move(&gl->gl_lru, &dispose);
1842 atomic_dec(&lru_count);
1843 freed++;
1844 continue;
1845 }
1846
1847 list_move(&gl->gl_lru, &skipped);
1848 }
1849 list_splice(&skipped, &lru_list);
1850 if (!list_empty(&dispose))
1851 gfs2_dispose_glock_lru(&dispose);
1852 spin_unlock(&lru_lock);
1853
1854 return freed;
1855}
1856
1857static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1858 struct shrink_control *sc)
1859{
1860 if (!(sc->gfp_mask & __GFP_FS))
1861 return SHRINK_STOP;
1862 return gfs2_scan_glock_lru(sc->nr_to_scan);
1863}
1864
1865static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1866 struct shrink_control *sc)
1867{
1868 return vfs_pressure_ratio(atomic_read(&lru_count));
1869}
1870
1871static struct shrinker glock_shrinker = {
1872 .seeks = DEFAULT_SEEKS,
1873 .count_objects = gfs2_glock_shrink_count,
1874 .scan_objects = gfs2_glock_shrink_scan,
1875};
1876
1877/**
1878 * glock_hash_walk - Call a function for glock in a hash bucket
1879 * @examiner: the function
1880 * @sdp: the filesystem
1881 *
1882 * Note that the function can be called multiple times on the same
1883 * object. So the user must ensure that the function can cope with
1884 * that.
1885 */
1886
1887static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1888{
1889 struct gfs2_glock *gl;
1890 struct rhashtable_iter iter;
1891
1892 rhashtable_walk_enter(&gl_hash_table, &iter);
1893
1894 do {
1895 rhashtable_walk_start(&iter);
1896
1897 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl))
1898 if (gl->gl_name.ln_sbd == sdp &&
1899 lockref_get_not_dead(&gl->gl_lockref))
1900 examiner(gl);
1901
1902 rhashtable_walk_stop(&iter);
1903 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
1904
1905 rhashtable_walk_exit(&iter);
1906}
1907
1908bool gfs2_queue_delete_work(struct gfs2_glock *gl, unsigned long delay)
1909{
1910 bool queued;
1911
1912 spin_lock(&gl->gl_lockref.lock);
1913 queued = queue_delayed_work(gfs2_delete_workqueue,
1914 &gl->gl_delete, delay);
1915 if (queued)
1916 set_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1917 spin_unlock(&gl->gl_lockref.lock);
1918 return queued;
1919}
1920
1921void gfs2_cancel_delete_work(struct gfs2_glock *gl)
1922{
1923 if (cancel_delayed_work_sync(&gl->gl_delete)) {
1924 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1925 gfs2_glock_put(gl);
1926 }
1927}
1928
1929bool gfs2_delete_work_queued(const struct gfs2_glock *gl)
1930{
1931 return test_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1932}
1933
1934static void flush_delete_work(struct gfs2_glock *gl)
1935{
1936 if (gl->gl_name.ln_type == LM_TYPE_IOPEN) {
1937 if (cancel_delayed_work(&gl->gl_delete)) {
1938 queue_delayed_work(gfs2_delete_workqueue,
1939 &gl->gl_delete, 0);
1940 }
1941 }
1942 gfs2_glock_queue_work(gl, 0);
1943}
1944
1945void gfs2_flush_delete_work(struct gfs2_sbd *sdp)
1946{
1947 glock_hash_walk(flush_delete_work, sdp);
1948 flush_workqueue(gfs2_delete_workqueue);
1949}
1950
1951/**
1952 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1953 * @gl: The glock to thaw
1954 *
1955 */
1956
1957static void thaw_glock(struct gfs2_glock *gl)
1958{
1959 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags)) {
1960 gfs2_glock_put(gl);
1961 return;
1962 }
1963 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1964 gfs2_glock_queue_work(gl, 0);
1965}
1966
1967/**
1968 * clear_glock - look at a glock and see if we can free it from glock cache
1969 * @gl: the glock to look at
1970 *
1971 */
1972
1973static void clear_glock(struct gfs2_glock *gl)
1974{
1975 gfs2_glock_remove_from_lru(gl);
1976
1977 spin_lock(&gl->gl_lockref.lock);
1978 if (gl->gl_state != LM_ST_UNLOCKED)
1979 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1980 __gfs2_glock_queue_work(gl, 0);
1981 spin_unlock(&gl->gl_lockref.lock);
1982}
1983
1984/**
1985 * gfs2_glock_thaw - Thaw any frozen glocks
1986 * @sdp: The super block
1987 *
1988 */
1989
1990void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1991{
1992 glock_hash_walk(thaw_glock, sdp);
1993}
1994
1995static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
1996{
1997 spin_lock(&gl->gl_lockref.lock);
1998 gfs2_dump_glock(seq, gl, fsid);
1999 spin_unlock(&gl->gl_lockref.lock);
2000}
2001
2002static void dump_glock_func(struct gfs2_glock *gl)
2003{
2004 dump_glock(NULL, gl, true);
2005}
2006
2007/**
2008 * gfs2_gl_hash_clear - Empty out the glock hash table
2009 * @sdp: the filesystem
2010 *
2011 * Called when unmounting the filesystem.
2012 */
2013
2014void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
2015{
2016 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
2017 flush_workqueue(glock_workqueue);
2018 glock_hash_walk(clear_glock, sdp);
2019 flush_workqueue(glock_workqueue);
2020 wait_event_timeout(sdp->sd_glock_wait,
2021 atomic_read(&sdp->sd_glock_disposal) == 0,
2022 HZ * 600);
2023 glock_hash_walk(dump_glock_func, sdp);
2024}
2025
2026void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
2027{
2028 struct gfs2_glock *gl = ip->i_gl;
2029 int ret;
2030
2031 ret = gfs2_truncatei_resume(ip);
2032 gfs2_glock_assert_withdraw(gl, ret == 0);
2033
2034 spin_lock(&gl->gl_lockref.lock);
2035 clear_bit(GLF_LOCK, &gl->gl_flags);
2036 run_queue(gl, 1);
2037 spin_unlock(&gl->gl_lockref.lock);
2038}
2039
2040static const char *state2str(unsigned state)
2041{
2042 switch(state) {
2043 case LM_ST_UNLOCKED:
2044 return "UN";
2045 case LM_ST_SHARED:
2046 return "SH";
2047 case LM_ST_DEFERRED:
2048 return "DF";
2049 case LM_ST_EXCLUSIVE:
2050 return "EX";
2051 }
2052 return "??";
2053}
2054
2055static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
2056{
2057 char *p = buf;
2058 if (flags & LM_FLAG_TRY)
2059 *p++ = 't';
2060 if (flags & LM_FLAG_TRY_1CB)
2061 *p++ = 'T';
2062 if (flags & LM_FLAG_NOEXP)
2063 *p++ = 'e';
2064 if (flags & LM_FLAG_ANY)
2065 *p++ = 'A';
2066 if (flags & LM_FLAG_PRIORITY)
2067 *p++ = 'p';
2068 if (flags & LM_FLAG_NODE_SCOPE)
2069 *p++ = 'n';
2070 if (flags & GL_ASYNC)
2071 *p++ = 'a';
2072 if (flags & GL_EXACT)
2073 *p++ = 'E';
2074 if (flags & GL_NOCACHE)
2075 *p++ = 'c';
2076 if (test_bit(HIF_HOLDER, &iflags))
2077 *p++ = 'H';
2078 if (test_bit(HIF_WAIT, &iflags))
2079 *p++ = 'W';
2080 if (test_bit(HIF_FIRST, &iflags))
2081 *p++ = 'F';
2082 *p = 0;
2083 return buf;
2084}
2085
2086/**
2087 * dump_holder - print information about a glock holder
2088 * @seq: the seq_file struct
2089 * @gh: the glock holder
2090 * @fs_id_buf: pointer to file system id (if requested)
2091 *
2092 */
2093
2094static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
2095 const char *fs_id_buf)
2096{
2097 struct task_struct *gh_owner = NULL;
2098 char flags_buf[32];
2099
2100 rcu_read_lock();
2101 if (gh->gh_owner_pid)
2102 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
2103 gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2104 fs_id_buf, state2str(gh->gh_state),
2105 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
2106 gh->gh_error,
2107 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
2108 gh_owner ? gh_owner->comm : "(ended)",
2109 (void *)gh->gh_ip);
2110 rcu_read_unlock();
2111}
2112
2113static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
2114{
2115 const unsigned long *gflags = &gl->gl_flags;
2116 char *p = buf;
2117
2118 if (test_bit(GLF_LOCK, gflags))
2119 *p++ = 'l';
2120 if (test_bit(GLF_DEMOTE, gflags))
2121 *p++ = 'D';
2122 if (test_bit(GLF_PENDING_DEMOTE, gflags))
2123 *p++ = 'd';
2124 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
2125 *p++ = 'p';
2126 if (test_bit(GLF_DIRTY, gflags))
2127 *p++ = 'y';
2128 if (test_bit(GLF_LFLUSH, gflags))
2129 *p++ = 'f';
2130 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
2131 *p++ = 'i';
2132 if (test_bit(GLF_REPLY_PENDING, gflags))
2133 *p++ = 'r';
2134 if (test_bit(GLF_INITIAL, gflags))
2135 *p++ = 'I';
2136 if (test_bit(GLF_FROZEN, gflags))
2137 *p++ = 'F';
2138 if (!list_empty(&gl->gl_holders))
2139 *p++ = 'q';
2140 if (test_bit(GLF_LRU, gflags))
2141 *p++ = 'L';
2142 if (gl->gl_object)
2143 *p++ = 'o';
2144 if (test_bit(GLF_BLOCKING, gflags))
2145 *p++ = 'b';
2146 if (test_bit(GLF_PENDING_DELETE, gflags))
2147 *p++ = 'P';
2148 if (test_bit(GLF_FREEING, gflags))
2149 *p++ = 'x';
2150 *p = 0;
2151 return buf;
2152}
2153
2154/**
2155 * gfs2_dump_glock - print information about a glock
2156 * @seq: The seq_file struct
2157 * @gl: the glock
2158 * @fsid: If true, also dump the file system id
2159 *
2160 * The file format is as follows:
2161 * One line per object, capital letters are used to indicate objects
2162 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2163 * other objects are indented by a single space and follow the glock to
2164 * which they are related. Fields are indicated by lower case letters
2165 * followed by a colon and the field value, except for strings which are in
2166 * [] so that its possible to see if they are composed of spaces for
2167 * example. The field's are n = number (id of the object), f = flags,
2168 * t = type, s = state, r = refcount, e = error, p = pid.
2169 *
2170 */
2171
2172void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2173{
2174 const struct gfs2_glock_operations *glops = gl->gl_ops;
2175 unsigned long long dtime;
2176 const struct gfs2_holder *gh;
2177 char gflags_buf[32];
2178 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2179 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2180 unsigned long nrpages = 0;
2181
2182 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
2183 struct address_space *mapping = gfs2_glock2aspace(gl);
2184
2185 nrpages = mapping->nrpages;
2186 }
2187 memset(fs_id_buf, 0, sizeof(fs_id_buf));
2188 if (fsid && sdp) /* safety precaution */
2189 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2190 dtime = jiffies - gl->gl_demote_time;
2191 dtime *= 1000000/HZ; /* demote time in uSec */
2192 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
2193 dtime = 0;
2194 gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2195 "v:%d r:%d m:%ld p:%lu\n",
2196 fs_id_buf, state2str(gl->gl_state),
2197 gl->gl_name.ln_type,
2198 (unsigned long long)gl->gl_name.ln_number,
2199 gflags2str(gflags_buf, gl),
2200 state2str(gl->gl_target),
2201 state2str(gl->gl_demote_state), dtime,
2202 atomic_read(&gl->gl_ail_count),
2203 atomic_read(&gl->gl_revokes),
2204 (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages);
2205
2206 list_for_each_entry(gh, &gl->gl_holders, gh_list)
2207 dump_holder(seq, gh, fs_id_buf);
2208
2209 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
2210 glops->go_dump(seq, gl, fs_id_buf);
2211}
2212
2213static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
2214{
2215 struct gfs2_glock *gl = iter_ptr;
2216
2217 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2218 gl->gl_name.ln_type,
2219 (unsigned long long)gl->gl_name.ln_number,
2220 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
2221 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
2222 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
2223 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
2224 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
2225 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
2226 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
2227 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
2228 return 0;
2229}
2230
2231static const char *gfs2_gltype[] = {
2232 "type",
2233 "reserved",
2234 "nondisk",
2235 "inode",
2236 "rgrp",
2237 "meta",
2238 "iopen",
2239 "flock",
2240 "plock",
2241 "quota",
2242 "journal",
2243};
2244
2245static const char *gfs2_stype[] = {
2246 [GFS2_LKS_SRTT] = "srtt",
2247 [GFS2_LKS_SRTTVAR] = "srttvar",
2248 [GFS2_LKS_SRTTB] = "srttb",
2249 [GFS2_LKS_SRTTVARB] = "srttvarb",
2250 [GFS2_LKS_SIRT] = "sirt",
2251 [GFS2_LKS_SIRTVAR] = "sirtvar",
2252 [GFS2_LKS_DCOUNT] = "dlm",
2253 [GFS2_LKS_QCOUNT] = "queue",
2254};
2255
2256#define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2257
2258static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
2259{
2260 struct gfs2_sbd *sdp = seq->private;
2261 loff_t pos = *(loff_t *)iter_ptr;
2262 unsigned index = pos >> 3;
2263 unsigned subindex = pos & 0x07;
2264 int i;
2265
2266 if (index == 0 && subindex != 0)
2267 return 0;
2268
2269 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
2270 (index == 0) ? "cpu": gfs2_stype[subindex]);
2271
2272 for_each_possible_cpu(i) {
2273 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
2274
2275 if (index == 0)
2276 seq_printf(seq, " %15u", i);
2277 else
2278 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
2279 lkstats[index - 1].stats[subindex]);
2280 }
2281 seq_putc(seq, '\n');
2282 return 0;
2283}
2284
2285int __init gfs2_glock_init(void)
2286{
2287 int i, ret;
2288
2289 ret = rhashtable_init(&gl_hash_table, &ht_parms);
2290 if (ret < 0)
2291 return ret;
2292
2293 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
2294 WQ_HIGHPRI | WQ_FREEZABLE, 0);
2295 if (!glock_workqueue) {
2296 rhashtable_destroy(&gl_hash_table);
2297 return -ENOMEM;
2298 }
2299 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
2300 WQ_MEM_RECLAIM | WQ_FREEZABLE,
2301 0);
2302 if (!gfs2_delete_workqueue) {
2303 destroy_workqueue(glock_workqueue);
2304 rhashtable_destroy(&gl_hash_table);
2305 return -ENOMEM;
2306 }
2307
2308 ret = register_shrinker(&glock_shrinker);
2309 if (ret) {
2310 destroy_workqueue(gfs2_delete_workqueue);
2311 destroy_workqueue(glock_workqueue);
2312 rhashtable_destroy(&gl_hash_table);
2313 return ret;
2314 }
2315
2316 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2317 init_waitqueue_head(glock_wait_table + i);
2318
2319 return 0;
2320}
2321
2322void gfs2_glock_exit(void)
2323{
2324 unregister_shrinker(&glock_shrinker);
2325 rhashtable_destroy(&gl_hash_table);
2326 destroy_workqueue(glock_workqueue);
2327 destroy_workqueue(gfs2_delete_workqueue);
2328}
2329
2330static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2331{
2332 struct gfs2_glock *gl = gi->gl;
2333
2334 if (gl) {
2335 if (n == 0)
2336 return;
2337 if (!lockref_put_not_zero(&gl->gl_lockref))
2338 gfs2_glock_queue_put(gl);
2339 }
2340 for (;;) {
2341 gl = rhashtable_walk_next(&gi->hti);
2342 if (IS_ERR_OR_NULL(gl)) {
2343 if (gl == ERR_PTR(-EAGAIN)) {
2344 n = 1;
2345 continue;
2346 }
2347 gl = NULL;
2348 break;
2349 }
2350 if (gl->gl_name.ln_sbd != gi->sdp)
2351 continue;
2352 if (n <= 1) {
2353 if (!lockref_get_not_dead(&gl->gl_lockref))
2354 continue;
2355 break;
2356 } else {
2357 if (__lockref_is_dead(&gl->gl_lockref))
2358 continue;
2359 n--;
2360 }
2361 }
2362 gi->gl = gl;
2363}
2364
2365static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2366 __acquires(RCU)
2367{
2368 struct gfs2_glock_iter *gi = seq->private;
2369 loff_t n;
2370
2371 /*
2372 * We can either stay where we are, skip to the next hash table
2373 * entry, or start from the beginning.
2374 */
2375 if (*pos < gi->last_pos) {
2376 rhashtable_walk_exit(&gi->hti);
2377 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2378 n = *pos + 1;
2379 } else {
2380 n = *pos - gi->last_pos;
2381 }
2382
2383 rhashtable_walk_start(&gi->hti);
2384
2385 gfs2_glock_iter_next(gi, n);
2386 gi->last_pos = *pos;
2387 return gi->gl;
2388}
2389
2390static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2391 loff_t *pos)
2392{
2393 struct gfs2_glock_iter *gi = seq->private;
2394
2395 (*pos)++;
2396 gi->last_pos = *pos;
2397 gfs2_glock_iter_next(gi, 1);
2398 return gi->gl;
2399}
2400
2401static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2402 __releases(RCU)
2403{
2404 struct gfs2_glock_iter *gi = seq->private;
2405
2406 rhashtable_walk_stop(&gi->hti);
2407}
2408
2409static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2410{
2411 dump_glock(seq, iter_ptr, false);
2412 return 0;
2413}
2414
2415static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2416{
2417 preempt_disable();
2418 if (*pos >= GFS2_NR_SBSTATS)
2419 return NULL;
2420 return pos;
2421}
2422
2423static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2424 loff_t *pos)
2425{
2426 (*pos)++;
2427 if (*pos >= GFS2_NR_SBSTATS)
2428 return NULL;
2429 return pos;
2430}
2431
2432static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2433{
2434 preempt_enable();
2435}
2436
2437static const struct seq_operations gfs2_glock_seq_ops = {
2438 .start = gfs2_glock_seq_start,
2439 .next = gfs2_glock_seq_next,
2440 .stop = gfs2_glock_seq_stop,
2441 .show = gfs2_glock_seq_show,
2442};
2443
2444static const struct seq_operations gfs2_glstats_seq_ops = {
2445 .start = gfs2_glock_seq_start,
2446 .next = gfs2_glock_seq_next,
2447 .stop = gfs2_glock_seq_stop,
2448 .show = gfs2_glstats_seq_show,
2449};
2450
2451static const struct seq_operations gfs2_sbstats_sops = {
2452 .start = gfs2_sbstats_seq_start,
2453 .next = gfs2_sbstats_seq_next,
2454 .stop = gfs2_sbstats_seq_stop,
2455 .show = gfs2_sbstats_seq_show,
2456};
2457
2458#define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2459
2460static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2461 const struct seq_operations *ops)
2462{
2463 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2464 if (ret == 0) {
2465 struct seq_file *seq = file->private_data;
2466 struct gfs2_glock_iter *gi = seq->private;
2467
2468 gi->sdp = inode->i_private;
2469 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2470 if (seq->buf)
2471 seq->size = GFS2_SEQ_GOODSIZE;
2472 /*
2473 * Initially, we are "before" the first hash table entry; the
2474 * first call to rhashtable_walk_next gets us the first entry.
2475 */
2476 gi->last_pos = -1;
2477 gi->gl = NULL;
2478 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2479 }
2480 return ret;
2481}
2482
2483static int gfs2_glocks_open(struct inode *inode, struct file *file)
2484{
2485 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2486}
2487
2488static int gfs2_glocks_release(struct inode *inode, struct file *file)
2489{
2490 struct seq_file *seq = file->private_data;
2491 struct gfs2_glock_iter *gi = seq->private;
2492
2493 if (gi->gl)
2494 gfs2_glock_put(gi->gl);
2495 rhashtable_walk_exit(&gi->hti);
2496 return seq_release_private(inode, file);
2497}
2498
2499static int gfs2_glstats_open(struct inode *inode, struct file *file)
2500{
2501 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2502}
2503
2504static const struct file_operations gfs2_glocks_fops = {
2505 .owner = THIS_MODULE,
2506 .open = gfs2_glocks_open,
2507 .read = seq_read,
2508 .llseek = seq_lseek,
2509 .release = gfs2_glocks_release,
2510};
2511
2512static const struct file_operations gfs2_glstats_fops = {
2513 .owner = THIS_MODULE,
2514 .open = gfs2_glstats_open,
2515 .read = seq_read,
2516 .llseek = seq_lseek,
2517 .release = gfs2_glocks_release,
2518};
2519
2520DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats);
2521
2522void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2523{
2524 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2525
2526 debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2527 &gfs2_glocks_fops);
2528
2529 debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2530 &gfs2_glstats_fops);
2531
2532 debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2533 &gfs2_sbstats_fops);
2534}
2535
2536void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2537{
2538 debugfs_remove_recursive(sdp->debugfs_dir);
2539 sdp->debugfs_dir = NULL;
2540}
2541
2542void gfs2_register_debugfs(void)
2543{
2544 gfs2_root = debugfs_create_dir("gfs2", NULL);
2545}
2546
2547void gfs2_unregister_debugfs(void)
2548{
2549 debugfs_remove(gfs2_root);
2550 gfs2_root = NULL;
2551}