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_del(&gl->gl_lru);
216 list_add_tail(&gl->gl_lru, &lru_list);
217
218 if (!test_bit(GLF_LRU, &gl->gl_flags)) {
219 set_bit(GLF_LRU, &gl->gl_flags);
220 atomic_inc(&lru_count);
221 }
222
223 spin_unlock(&lru_lock);
224}
225
226static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
227{
228 if (!(gl->gl_ops->go_flags & GLOF_LRU))
229 return;
230
231 spin_lock(&lru_lock);
232 if (test_bit(GLF_LRU, &gl->gl_flags)) {
233 list_del_init(&gl->gl_lru);
234 atomic_dec(&lru_count);
235 clear_bit(GLF_LRU, &gl->gl_flags);
236 }
237 spin_unlock(&lru_lock);
238}
239
240/*
241 * Enqueue the glock on the work queue. Passes one glock reference on to the
242 * work queue.
243 */
244static void __gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
245 if (!queue_delayed_work(glock_workqueue, &gl->gl_work, delay)) {
246 /*
247 * We are holding the lockref spinlock, and the work was still
248 * queued above. The queued work (glock_work_func) takes that
249 * spinlock before dropping its glock reference(s), so it
250 * cannot have dropped them in the meantime.
251 */
252 GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2);
253 gl->gl_lockref.count--;
254 }
255}
256
257static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
258 spin_lock(&gl->gl_lockref.lock);
259 __gfs2_glock_queue_work(gl, delay);
260 spin_unlock(&gl->gl_lockref.lock);
261}
262
263static void __gfs2_glock_put(struct gfs2_glock *gl)
264{
265 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
266 struct address_space *mapping = gfs2_glock2aspace(gl);
267
268 lockref_mark_dead(&gl->gl_lockref);
269
270 gfs2_glock_remove_from_lru(gl);
271 spin_unlock(&gl->gl_lockref.lock);
272 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
273 GLOCK_BUG_ON(gl, mapping && mapping->nrpages && !gfs2_withdrawn(sdp));
274 trace_gfs2_glock_put(gl);
275 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
276}
277
278/*
279 * Cause the glock to be put in work queue context.
280 */
281void gfs2_glock_queue_put(struct gfs2_glock *gl)
282{
283 gfs2_glock_queue_work(gl, 0);
284}
285
286/**
287 * gfs2_glock_put() - Decrement reference count on glock
288 * @gl: The glock to put
289 *
290 */
291
292void gfs2_glock_put(struct gfs2_glock *gl)
293{
294 if (lockref_put_or_lock(&gl->gl_lockref))
295 return;
296
297 __gfs2_glock_put(gl);
298}
299
300/**
301 * may_grant - check if its ok to grant a new lock
302 * @gl: The glock
303 * @gh: The lock request which we wish to grant
304 *
305 * Returns: true if its ok to grant the lock
306 */
307
308static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
309{
310 const struct gfs2_holder *gh_head = list_first_entry(&gl->gl_holders, const struct gfs2_holder, gh_list);
311 if ((gh->gh_state == LM_ST_EXCLUSIVE ||
312 gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
313 return 0;
314 if (gl->gl_state == gh->gh_state)
315 return 1;
316 if (gh->gh_flags & GL_EXACT)
317 return 0;
318 if (gl->gl_state == LM_ST_EXCLUSIVE) {
319 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
320 return 1;
321 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
322 return 1;
323 }
324 if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
325 return 1;
326 return 0;
327}
328
329static void gfs2_holder_wake(struct gfs2_holder *gh)
330{
331 clear_bit(HIF_WAIT, &gh->gh_iflags);
332 smp_mb__after_atomic();
333 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
334 if (gh->gh_flags & GL_ASYNC) {
335 struct gfs2_sbd *sdp = gh->gh_gl->gl_name.ln_sbd;
336
337 wake_up(&sdp->sd_async_glock_wait);
338 }
339}
340
341/**
342 * do_error - Something unexpected has happened during a lock request
343 *
344 */
345
346static void do_error(struct gfs2_glock *gl, const int ret)
347{
348 struct gfs2_holder *gh, *tmp;
349
350 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
351 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
352 continue;
353 if (ret & LM_OUT_ERROR)
354 gh->gh_error = -EIO;
355 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
356 gh->gh_error = GLR_TRYFAILED;
357 else
358 continue;
359 list_del_init(&gh->gh_list);
360 trace_gfs2_glock_queue(gh, 0);
361 gfs2_holder_wake(gh);
362 }
363}
364
365/**
366 * do_promote - promote as many requests as possible on the current queue
367 * @gl: The glock
368 *
369 * Returns: 1 if there is a blocked holder at the head of the list, or 2
370 * if a type specific operation is underway.
371 */
372
373static int do_promote(struct gfs2_glock *gl)
374__releases(&gl->gl_lockref.lock)
375__acquires(&gl->gl_lockref.lock)
376{
377 const struct gfs2_glock_operations *glops = gl->gl_ops;
378 struct gfs2_holder *gh, *tmp;
379 int ret;
380
381restart:
382 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
383 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
384 continue;
385 if (may_grant(gl, gh)) {
386 if (gh->gh_list.prev == &gl->gl_holders &&
387 glops->go_lock) {
388 spin_unlock(&gl->gl_lockref.lock);
389 /* FIXME: eliminate this eventually */
390 ret = glops->go_lock(gh);
391 spin_lock(&gl->gl_lockref.lock);
392 if (ret) {
393 if (ret == 1)
394 return 2;
395 gh->gh_error = ret;
396 list_del_init(&gh->gh_list);
397 trace_gfs2_glock_queue(gh, 0);
398 gfs2_holder_wake(gh);
399 goto restart;
400 }
401 set_bit(HIF_HOLDER, &gh->gh_iflags);
402 trace_gfs2_promote(gh, 1);
403 gfs2_holder_wake(gh);
404 goto restart;
405 }
406 set_bit(HIF_HOLDER, &gh->gh_iflags);
407 trace_gfs2_promote(gh, 0);
408 gfs2_holder_wake(gh);
409 continue;
410 }
411 if (gh->gh_list.prev == &gl->gl_holders)
412 return 1;
413 do_error(gl, 0);
414 break;
415 }
416 return 0;
417}
418
419/**
420 * find_first_waiter - find the first gh that's waiting for the glock
421 * @gl: the glock
422 */
423
424static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
425{
426 struct gfs2_holder *gh;
427
428 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
429 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
430 return gh;
431 }
432 return NULL;
433}
434
435/**
436 * state_change - record that the glock is now in a different state
437 * @gl: the glock
438 * @new_state the new state
439 *
440 */
441
442static void state_change(struct gfs2_glock *gl, unsigned int new_state)
443{
444 int held1, held2;
445
446 held1 = (gl->gl_state != LM_ST_UNLOCKED);
447 held2 = (new_state != LM_ST_UNLOCKED);
448
449 if (held1 != held2) {
450 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
451 if (held2)
452 gl->gl_lockref.count++;
453 else
454 gl->gl_lockref.count--;
455 }
456 if (held1 && held2 && list_empty(&gl->gl_holders))
457 clear_bit(GLF_QUEUED, &gl->gl_flags);
458
459 if (new_state != gl->gl_target)
460 /* shorten our minimum hold time */
461 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
462 GL_GLOCK_MIN_HOLD);
463 gl->gl_state = new_state;
464 gl->gl_tchange = jiffies;
465}
466
467static void gfs2_set_demote(struct gfs2_glock *gl)
468{
469 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
470
471 set_bit(GLF_DEMOTE, &gl->gl_flags);
472 smp_mb();
473 wake_up(&sdp->sd_async_glock_wait);
474}
475
476static void gfs2_demote_wake(struct gfs2_glock *gl)
477{
478 gl->gl_demote_state = LM_ST_EXCLUSIVE;
479 clear_bit(GLF_DEMOTE, &gl->gl_flags);
480 smp_mb__after_atomic();
481 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
482}
483
484/**
485 * finish_xmote - The DLM has replied to one of our lock requests
486 * @gl: The glock
487 * @ret: The status from the DLM
488 *
489 */
490
491static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
492{
493 const struct gfs2_glock_operations *glops = gl->gl_ops;
494 struct gfs2_holder *gh;
495 unsigned state = ret & LM_OUT_ST_MASK;
496 int rv;
497
498 spin_lock(&gl->gl_lockref.lock);
499 trace_gfs2_glock_state_change(gl, state);
500 state_change(gl, state);
501 gh = find_first_waiter(gl);
502
503 /* Demote to UN request arrived during demote to SH or DF */
504 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
505 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
506 gl->gl_target = LM_ST_UNLOCKED;
507
508 /* Check for state != intended state */
509 if (unlikely(state != gl->gl_target)) {
510 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
511 /* move to back of queue and try next entry */
512 if (ret & LM_OUT_CANCELED) {
513 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
514 list_move_tail(&gh->gh_list, &gl->gl_holders);
515 gh = find_first_waiter(gl);
516 gl->gl_target = gh->gh_state;
517 goto retry;
518 }
519 /* Some error or failed "try lock" - report it */
520 if ((ret & LM_OUT_ERROR) ||
521 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
522 gl->gl_target = gl->gl_state;
523 do_error(gl, ret);
524 goto out;
525 }
526 }
527 switch(state) {
528 /* Unlocked due to conversion deadlock, try again */
529 case LM_ST_UNLOCKED:
530retry:
531 do_xmote(gl, gh, gl->gl_target);
532 break;
533 /* Conversion fails, unlock and try again */
534 case LM_ST_SHARED:
535 case LM_ST_DEFERRED:
536 do_xmote(gl, gh, LM_ST_UNLOCKED);
537 break;
538 default: /* Everything else */
539 fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n",
540 gl->gl_target, state);
541 GLOCK_BUG_ON(gl, 1);
542 }
543 spin_unlock(&gl->gl_lockref.lock);
544 return;
545 }
546
547 /* Fast path - we got what we asked for */
548 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
549 gfs2_demote_wake(gl);
550 if (state != LM_ST_UNLOCKED) {
551 if (glops->go_xmote_bh) {
552 spin_unlock(&gl->gl_lockref.lock);
553 rv = glops->go_xmote_bh(gl, gh);
554 spin_lock(&gl->gl_lockref.lock);
555 if (rv) {
556 do_error(gl, rv);
557 goto out;
558 }
559 }
560 rv = do_promote(gl);
561 if (rv == 2)
562 goto out_locked;
563 }
564out:
565 clear_bit(GLF_LOCK, &gl->gl_flags);
566out_locked:
567 spin_unlock(&gl->gl_lockref.lock);
568}
569
570/**
571 * do_xmote - Calls the DLM to change the state of a lock
572 * @gl: The lock state
573 * @gh: The holder (only for promotes)
574 * @target: The target lock state
575 *
576 */
577
578static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
579__releases(&gl->gl_lockref.lock)
580__acquires(&gl->gl_lockref.lock)
581{
582 const struct gfs2_glock_operations *glops = gl->gl_ops;
583 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
584 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
585 int ret;
586
587 if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) &&
588 gh && !(gh->gh_flags & LM_FLAG_NOEXP))
589 return;
590 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
591 LM_FLAG_PRIORITY);
592 GLOCK_BUG_ON(gl, gl->gl_state == target);
593 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
594 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
595 glops->go_inval) {
596 /*
597 * If another process is already doing the invalidate, let that
598 * finish first. The glock state machine will get back to this
599 * holder again later.
600 */
601 if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
602 &gl->gl_flags))
603 return;
604 do_error(gl, 0); /* Fail queued try locks */
605 }
606 gl->gl_req = target;
607 set_bit(GLF_BLOCKING, &gl->gl_flags);
608 if ((gl->gl_req == LM_ST_UNLOCKED) ||
609 (gl->gl_state == LM_ST_EXCLUSIVE) ||
610 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
611 clear_bit(GLF_BLOCKING, &gl->gl_flags);
612 spin_unlock(&gl->gl_lockref.lock);
613 if (glops->go_sync) {
614 ret = glops->go_sync(gl);
615 /* If we had a problem syncing (due to io errors or whatever,
616 * we should not invalidate the metadata or tell dlm to
617 * release the glock to other nodes.
618 */
619 if (ret) {
620 if (cmpxchg(&sdp->sd_log_error, 0, ret)) {
621 fs_err(sdp, "Error %d syncing glock \n", ret);
622 gfs2_dump_glock(NULL, gl, true);
623 }
624 goto skip_inval;
625 }
626 }
627 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) {
628 /*
629 * The call to go_sync should have cleared out the ail list.
630 * If there are still items, we have a problem. We ought to
631 * withdraw, but we can't because the withdraw code also uses
632 * glocks. Warn about the error, dump the glock, then fall
633 * through and wait for logd to do the withdraw for us.
634 */
635 if ((atomic_read(&gl->gl_ail_count) != 0) &&
636 (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) {
637 gfs2_glock_assert_warn(gl,
638 !atomic_read(&gl->gl_ail_count));
639 gfs2_dump_glock(NULL, gl, true);
640 }
641 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
642 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
643 }
644
645skip_inval:
646 gfs2_glock_hold(gl);
647 /*
648 * Check for an error encountered since we called go_sync and go_inval.
649 * If so, we can't withdraw from the glock code because the withdraw
650 * code itself uses glocks (see function signal_our_withdraw) to
651 * change the mount to read-only. Most importantly, we must not call
652 * dlm to unlock the glock until the journal is in a known good state
653 * (after journal replay) otherwise other nodes may use the object
654 * (rgrp or dinode) and then later, journal replay will corrupt the
655 * file system. The best we can do here is wait for the logd daemon
656 * to see sd_log_error and withdraw, and in the meantime, requeue the
657 * work for later.
658 *
659 * However, if we're just unlocking the lock (say, for unmount, when
660 * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
661 * then it's okay to tell dlm to unlock it.
662 */
663 if (unlikely(sdp->sd_log_error && !gfs2_withdrawn(sdp)))
664 gfs2_withdraw_delayed(sdp);
665 if (glock_blocked_by_withdraw(gl)) {
666 if (target != LM_ST_UNLOCKED ||
667 test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags)) {
668 gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD);
669 goto out;
670 }
671 }
672
673 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
674 /* lock_dlm */
675 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
676 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
677 target == LM_ST_UNLOCKED &&
678 test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
679 finish_xmote(gl, target);
680 gfs2_glock_queue_work(gl, 0);
681 } else if (ret) {
682 fs_err(sdp, "lm_lock ret %d\n", ret);
683 GLOCK_BUG_ON(gl, !gfs2_withdrawn(sdp));
684 }
685 } else { /* lock_nolock */
686 finish_xmote(gl, target);
687 gfs2_glock_queue_work(gl, 0);
688 }
689out:
690 spin_lock(&gl->gl_lockref.lock);
691}
692
693/**
694 * find_first_holder - find the first "holder" gh
695 * @gl: the glock
696 */
697
698static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
699{
700 struct gfs2_holder *gh;
701
702 if (!list_empty(&gl->gl_holders)) {
703 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
704 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
705 return gh;
706 }
707 return NULL;
708}
709
710/**
711 * run_queue - do all outstanding tasks related to a glock
712 * @gl: The glock in question
713 * @nonblock: True if we must not block in run_queue
714 *
715 */
716
717static void run_queue(struct gfs2_glock *gl, const int nonblock)
718__releases(&gl->gl_lockref.lock)
719__acquires(&gl->gl_lockref.lock)
720{
721 struct gfs2_holder *gh = NULL;
722 int ret;
723
724 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
725 return;
726
727 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
728
729 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
730 gl->gl_demote_state != gl->gl_state) {
731 if (find_first_holder(gl))
732 goto out_unlock;
733 if (nonblock)
734 goto out_sched;
735 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
736 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
737 gl->gl_target = gl->gl_demote_state;
738 } else {
739 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
740 gfs2_demote_wake(gl);
741 ret = do_promote(gl);
742 if (ret == 0)
743 goto out_unlock;
744 if (ret == 2)
745 goto out;
746 gh = find_first_waiter(gl);
747 gl->gl_target = gh->gh_state;
748 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
749 do_error(gl, 0); /* Fail queued try locks */
750 }
751 do_xmote(gl, gh, gl->gl_target);
752out:
753 return;
754
755out_sched:
756 clear_bit(GLF_LOCK, &gl->gl_flags);
757 smp_mb__after_atomic();
758 gl->gl_lockref.count++;
759 __gfs2_glock_queue_work(gl, 0);
760 return;
761
762out_unlock:
763 clear_bit(GLF_LOCK, &gl->gl_flags);
764 smp_mb__after_atomic();
765 return;
766}
767
768void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation)
769{
770 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
771
772 if (ri->ri_magic == 0)
773 ri->ri_magic = cpu_to_be32(GFS2_MAGIC);
774 if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC))
775 ri->ri_generation_deleted = cpu_to_be64(generation);
776}
777
778bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation)
779{
780 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
781
782 if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC))
783 return false;
784 return generation <= be64_to_cpu(ri->ri_generation_deleted);
785}
786
787static void gfs2_glock_poke(struct gfs2_glock *gl)
788{
789 int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP;
790 struct gfs2_holder gh;
791 int error;
792
793 gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh);
794 error = gfs2_glock_nq(&gh);
795 if (!error)
796 gfs2_glock_dq(&gh);
797 gfs2_holder_uninit(&gh);
798}
799
800static bool gfs2_try_evict(struct gfs2_glock *gl)
801{
802 struct gfs2_inode *ip;
803 bool evicted = false;
804
805 /*
806 * If there is contention on the iopen glock and we have an inode, try
807 * to grab and release the inode so that it can be evicted. This will
808 * allow the remote node to go ahead and delete the inode without us
809 * having to do it, which will avoid rgrp glock thrashing.
810 *
811 * The remote node is likely still holding the corresponding inode
812 * glock, so it will run before we get to verify that the delete has
813 * happened below.
814 */
815 spin_lock(&gl->gl_lockref.lock);
816 ip = gl->gl_object;
817 if (ip && !igrab(&ip->i_inode))
818 ip = NULL;
819 spin_unlock(&gl->gl_lockref.lock);
820 if (ip) {
821 struct gfs2_glock *inode_gl = NULL;
822
823 gl->gl_no_formal_ino = ip->i_no_formal_ino;
824 set_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
825 d_prune_aliases(&ip->i_inode);
826 iput(&ip->i_inode);
827
828 /* If the inode was evicted, gl->gl_object will now be NULL. */
829 spin_lock(&gl->gl_lockref.lock);
830 ip = gl->gl_object;
831 if (ip) {
832 inode_gl = ip->i_gl;
833 lockref_get(&inode_gl->gl_lockref);
834 clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
835 }
836 spin_unlock(&gl->gl_lockref.lock);
837 if (inode_gl) {
838 gfs2_glock_poke(inode_gl);
839 gfs2_glock_put(inode_gl);
840 }
841 evicted = !ip;
842 }
843 return evicted;
844}
845
846static void delete_work_func(struct work_struct *work)
847{
848 struct delayed_work *dwork = to_delayed_work(work);
849 struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete);
850 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
851 struct inode *inode;
852 u64 no_addr = gl->gl_name.ln_number;
853
854 spin_lock(&gl->gl_lockref.lock);
855 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
856 spin_unlock(&gl->gl_lockref.lock);
857
858 /* If someone's using this glock to create a new dinode, the block must
859 have been freed by another node, then re-used, in which case our
860 iopen callback is too late after the fact. Ignore it. */
861 if (test_bit(GLF_INODE_CREATING, &gl->gl_flags))
862 goto out;
863
864 if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
865 /*
866 * If we can evict the inode, give the remote node trying to
867 * delete the inode some time before verifying that the delete
868 * has happened. Otherwise, if we cause contention on the inode glock
869 * immediately, the remote node will think that we still have
870 * the inode in use, and so it will give up waiting.
871 *
872 * If we can't evict the inode, signal to the remote node that
873 * the inode is still in use. We'll later try to delete the
874 * inode locally in gfs2_evict_inode.
875 *
876 * FIXME: We only need to verify that the remote node has
877 * deleted the inode because nodes before this remote delete
878 * rework won't cooperate. At a later time, when we no longer
879 * care about compatibility with such nodes, we can skip this
880 * step entirely.
881 */
882 if (gfs2_try_evict(gl)) {
883 if (gfs2_queue_delete_work(gl, 5 * HZ))
884 return;
885 }
886 goto out;
887 }
888
889 inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino,
890 GFS2_BLKST_UNLINKED);
891 if (!IS_ERR_OR_NULL(inode)) {
892 d_prune_aliases(inode);
893 iput(inode);
894 }
895out:
896 gfs2_glock_put(gl);
897}
898
899static void glock_work_func(struct work_struct *work)
900{
901 unsigned long delay = 0;
902 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
903 unsigned int drop_refs = 1;
904
905 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
906 finish_xmote(gl, gl->gl_reply);
907 drop_refs++;
908 }
909 spin_lock(&gl->gl_lockref.lock);
910 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
911 gl->gl_state != LM_ST_UNLOCKED &&
912 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
913 unsigned long holdtime, now = jiffies;
914
915 holdtime = gl->gl_tchange + gl->gl_hold_time;
916 if (time_before(now, holdtime))
917 delay = holdtime - now;
918
919 if (!delay) {
920 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
921 gfs2_set_demote(gl);
922 }
923 }
924 run_queue(gl, 0);
925 if (delay) {
926 /* Keep one glock reference for the work we requeue. */
927 drop_refs--;
928 if (gl->gl_name.ln_type != LM_TYPE_INODE)
929 delay = 0;
930 __gfs2_glock_queue_work(gl, delay);
931 }
932
933 /*
934 * Drop the remaining glock references manually here. (Mind that
935 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
936 * here as well.)
937 */
938 gl->gl_lockref.count -= drop_refs;
939 if (!gl->gl_lockref.count) {
940 __gfs2_glock_put(gl);
941 return;
942 }
943 spin_unlock(&gl->gl_lockref.lock);
944}
945
946static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
947 struct gfs2_glock *new)
948{
949 struct wait_glock_queue wait;
950 wait_queue_head_t *wq = glock_waitqueue(name);
951 struct gfs2_glock *gl;
952
953 wait.name = name;
954 init_wait(&wait.wait);
955 wait.wait.func = glock_wake_function;
956
957again:
958 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
959 rcu_read_lock();
960 if (new) {
961 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
962 &new->gl_node, ht_parms);
963 if (IS_ERR(gl))
964 goto out;
965 } else {
966 gl = rhashtable_lookup_fast(&gl_hash_table,
967 name, ht_parms);
968 }
969 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
970 rcu_read_unlock();
971 schedule();
972 goto again;
973 }
974out:
975 rcu_read_unlock();
976 finish_wait(wq, &wait.wait);
977 return gl;
978}
979
980/**
981 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
982 * @sdp: The GFS2 superblock
983 * @number: the lock number
984 * @glops: The glock_operations to use
985 * @create: If 0, don't create the glock if it doesn't exist
986 * @glp: the glock is returned here
987 *
988 * This does not lock a glock, just finds/creates structures for one.
989 *
990 * Returns: errno
991 */
992
993int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
994 const struct gfs2_glock_operations *glops, int create,
995 struct gfs2_glock **glp)
996{
997 struct super_block *s = sdp->sd_vfs;
998 struct lm_lockname name = { .ln_number = number,
999 .ln_type = glops->go_type,
1000 .ln_sbd = sdp };
1001 struct gfs2_glock *gl, *tmp;
1002 struct address_space *mapping;
1003 struct kmem_cache *cachep;
1004 int ret = 0;
1005
1006 gl = find_insert_glock(&name, NULL);
1007 if (gl) {
1008 *glp = gl;
1009 return 0;
1010 }
1011 if (!create)
1012 return -ENOENT;
1013
1014 if (glops->go_flags & GLOF_ASPACE)
1015 cachep = gfs2_glock_aspace_cachep;
1016 else
1017 cachep = gfs2_glock_cachep;
1018 gl = kmem_cache_alloc(cachep, GFP_NOFS);
1019 if (!gl)
1020 return -ENOMEM;
1021
1022 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
1023
1024 if (glops->go_flags & GLOF_LVB) {
1025 gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1026 if (!gl->gl_lksb.sb_lvbptr) {
1027 kmem_cache_free(cachep, gl);
1028 return -ENOMEM;
1029 }
1030 }
1031
1032 atomic_inc(&sdp->sd_glock_disposal);
1033 gl->gl_node.next = NULL;
1034 gl->gl_flags = 0;
1035 gl->gl_name = name;
1036 gl->gl_lockref.count = 1;
1037 gl->gl_state = LM_ST_UNLOCKED;
1038 gl->gl_target = LM_ST_UNLOCKED;
1039 gl->gl_demote_state = LM_ST_EXCLUSIVE;
1040 gl->gl_ops = glops;
1041 gl->gl_dstamp = 0;
1042 preempt_disable();
1043 /* We use the global stats to estimate the initial per-glock stats */
1044 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
1045 preempt_enable();
1046 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
1047 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
1048 gl->gl_tchange = jiffies;
1049 gl->gl_object = NULL;
1050 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
1051 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
1052 INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func);
1053
1054 mapping = gfs2_glock2aspace(gl);
1055 if (mapping) {
1056 mapping->a_ops = &gfs2_meta_aops;
1057 mapping->host = s->s_bdev->bd_inode;
1058 mapping->flags = 0;
1059 mapping_set_gfp_mask(mapping, GFP_NOFS);
1060 mapping->private_data = NULL;
1061 mapping->writeback_index = 0;
1062 }
1063
1064 tmp = find_insert_glock(&name, gl);
1065 if (!tmp) {
1066 *glp = gl;
1067 goto out;
1068 }
1069 if (IS_ERR(tmp)) {
1070 ret = PTR_ERR(tmp);
1071 goto out_free;
1072 }
1073 *glp = tmp;
1074
1075out_free:
1076 kfree(gl->gl_lksb.sb_lvbptr);
1077 kmem_cache_free(cachep, gl);
1078 atomic_dec(&sdp->sd_glock_disposal);
1079
1080out:
1081 return ret;
1082}
1083
1084/**
1085 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
1086 * @gl: the glock
1087 * @state: the state we're requesting
1088 * @flags: the modifier flags
1089 * @gh: the holder structure
1090 *
1091 */
1092
1093void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
1094 struct gfs2_holder *gh)
1095{
1096 INIT_LIST_HEAD(&gh->gh_list);
1097 gh->gh_gl = gl;
1098 gh->gh_ip = _RET_IP_;
1099 gh->gh_owner_pid = get_pid(task_pid(current));
1100 gh->gh_state = state;
1101 gh->gh_flags = flags;
1102 gh->gh_error = 0;
1103 gh->gh_iflags = 0;
1104 gfs2_glock_hold(gl);
1105}
1106
1107/**
1108 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1109 * @state: the state we're requesting
1110 * @flags: the modifier flags
1111 * @gh: the holder structure
1112 *
1113 * Don't mess with the glock.
1114 *
1115 */
1116
1117void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
1118{
1119 gh->gh_state = state;
1120 gh->gh_flags = flags;
1121 gh->gh_iflags = 0;
1122 gh->gh_ip = _RET_IP_;
1123 put_pid(gh->gh_owner_pid);
1124 gh->gh_owner_pid = get_pid(task_pid(current));
1125}
1126
1127/**
1128 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1129 * @gh: the holder structure
1130 *
1131 */
1132
1133void gfs2_holder_uninit(struct gfs2_holder *gh)
1134{
1135 put_pid(gh->gh_owner_pid);
1136 gfs2_glock_put(gh->gh_gl);
1137 gfs2_holder_mark_uninitialized(gh);
1138 gh->gh_ip = 0;
1139}
1140
1141static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
1142 unsigned long start_time)
1143{
1144 /* Have we waited longer that a second? */
1145 if (time_after(jiffies, start_time + HZ)) {
1146 /* Lengthen the minimum hold time. */
1147 gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
1148 GL_GLOCK_MAX_HOLD);
1149 }
1150}
1151
1152/**
1153 * gfs2_glock_wait - wait on a glock acquisition
1154 * @gh: the glock holder
1155 *
1156 * Returns: 0 on success
1157 */
1158
1159int gfs2_glock_wait(struct gfs2_holder *gh)
1160{
1161 unsigned long start_time = jiffies;
1162
1163 might_sleep();
1164 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1165 gfs2_glock_update_hold_time(gh->gh_gl, start_time);
1166 return gh->gh_error;
1167}
1168
1169static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
1170{
1171 int i;
1172
1173 for (i = 0; i < num_gh; i++)
1174 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
1175 return 1;
1176 return 0;
1177}
1178
1179/**
1180 * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1181 * @num_gh: the number of holders in the array
1182 * @ghs: the glock holder array
1183 *
1184 * Returns: 0 on success, meaning all glocks have been granted and are held.
1185 * -ESTALE if the request timed out, meaning all glocks were released,
1186 * and the caller should retry the operation.
1187 */
1188
1189int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
1190{
1191 struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
1192 int i, ret = 0, timeout = 0;
1193 unsigned long start_time = jiffies;
1194 bool keep_waiting;
1195
1196 might_sleep();
1197 /*
1198 * Total up the (minimum hold time * 2) of all glocks and use that to
1199 * determine the max amount of time we should wait.
1200 */
1201 for (i = 0; i < num_gh; i++)
1202 timeout += ghs[i].gh_gl->gl_hold_time << 1;
1203
1204wait_for_dlm:
1205 if (!wait_event_timeout(sdp->sd_async_glock_wait,
1206 !glocks_pending(num_gh, ghs), timeout))
1207 ret = -ESTALE; /* request timed out. */
1208
1209 /*
1210 * If dlm granted all our requests, we need to adjust the glock
1211 * minimum hold time values according to how long we waited.
1212 *
1213 * If our request timed out, we need to repeatedly release any held
1214 * glocks we acquired thus far to allow dlm to acquire the remaining
1215 * glocks without deadlocking. We cannot currently cancel outstanding
1216 * glock acquisitions.
1217 *
1218 * The HIF_WAIT bit tells us which requests still need a response from
1219 * dlm.
1220 *
1221 * If dlm sent us any errors, we return the first error we find.
1222 */
1223 keep_waiting = false;
1224 for (i = 0; i < num_gh; i++) {
1225 /* Skip holders we have already dequeued below. */
1226 if (!gfs2_holder_queued(&ghs[i]))
1227 continue;
1228 /* Skip holders with a pending DLM response. */
1229 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags)) {
1230 keep_waiting = true;
1231 continue;
1232 }
1233
1234 if (test_bit(HIF_HOLDER, &ghs[i].gh_iflags)) {
1235 if (ret == -ESTALE)
1236 gfs2_glock_dq(&ghs[i]);
1237 else
1238 gfs2_glock_update_hold_time(ghs[i].gh_gl,
1239 start_time);
1240 }
1241 if (!ret)
1242 ret = ghs[i].gh_error;
1243 }
1244
1245 if (keep_waiting)
1246 goto wait_for_dlm;
1247
1248 /*
1249 * At this point, we've either acquired all locks or released them all.
1250 */
1251 return ret;
1252}
1253
1254/**
1255 * handle_callback - process a demote request
1256 * @gl: the glock
1257 * @state: the state the caller wants us to change to
1258 *
1259 * There are only two requests that we are going to see in actual
1260 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1261 */
1262
1263static void handle_callback(struct gfs2_glock *gl, unsigned int state,
1264 unsigned long delay, bool remote)
1265{
1266 if (delay)
1267 set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1268 else
1269 gfs2_set_demote(gl);
1270 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
1271 gl->gl_demote_state = state;
1272 gl->gl_demote_time = jiffies;
1273 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
1274 gl->gl_demote_state != state) {
1275 gl->gl_demote_state = LM_ST_UNLOCKED;
1276 }
1277 if (gl->gl_ops->go_callback)
1278 gl->gl_ops->go_callback(gl, remote);
1279 trace_gfs2_demote_rq(gl, remote);
1280}
1281
1282void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
1283{
1284 struct va_format vaf;
1285 va_list args;
1286
1287 va_start(args, fmt);
1288
1289 if (seq) {
1290 seq_vprintf(seq, fmt, args);
1291 } else {
1292 vaf.fmt = fmt;
1293 vaf.va = &args;
1294
1295 pr_err("%pV", &vaf);
1296 }
1297
1298 va_end(args);
1299}
1300
1301/**
1302 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1303 * @gh: the holder structure to add
1304 *
1305 * Eventually we should move the recursive locking trap to a
1306 * debugging option or something like that. This is the fast
1307 * path and needs to have the minimum number of distractions.
1308 *
1309 */
1310
1311static inline void add_to_queue(struct gfs2_holder *gh)
1312__releases(&gl->gl_lockref.lock)
1313__acquires(&gl->gl_lockref.lock)
1314{
1315 struct gfs2_glock *gl = gh->gh_gl;
1316 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1317 struct list_head *insert_pt = NULL;
1318 struct gfs2_holder *gh2;
1319 int try_futile = 0;
1320
1321 GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1322 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1323 GLOCK_BUG_ON(gl, true);
1324
1325 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1326 if (test_bit(GLF_LOCK, &gl->gl_flags))
1327 try_futile = !may_grant(gl, gh);
1328 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1329 goto fail;
1330 }
1331
1332 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1333 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
1334 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
1335 goto trap_recursive;
1336 if (try_futile &&
1337 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1338fail:
1339 gh->gh_error = GLR_TRYFAILED;
1340 gfs2_holder_wake(gh);
1341 return;
1342 }
1343 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1344 continue;
1345 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1346 insert_pt = &gh2->gh_list;
1347 }
1348 set_bit(GLF_QUEUED, &gl->gl_flags);
1349 trace_gfs2_glock_queue(gh, 1);
1350 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1351 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1352 if (likely(insert_pt == NULL)) {
1353 list_add_tail(&gh->gh_list, &gl->gl_holders);
1354 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1355 goto do_cancel;
1356 return;
1357 }
1358 list_add_tail(&gh->gh_list, insert_pt);
1359do_cancel:
1360 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
1361 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1362 spin_unlock(&gl->gl_lockref.lock);
1363 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1364 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1365 spin_lock(&gl->gl_lockref.lock);
1366 }
1367 return;
1368
1369trap_recursive:
1370 fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1371 fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1372 fs_err(sdp, "lock type: %d req lock state : %d\n",
1373 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1374 fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1375 fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1376 fs_err(sdp, "lock type: %d req lock state : %d\n",
1377 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1378 gfs2_dump_glock(NULL, gl, true);
1379 BUG();
1380}
1381
1382/**
1383 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1384 * @gh: the holder structure
1385 *
1386 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1387 *
1388 * Returns: 0, GLR_TRYFAILED, or errno on failure
1389 */
1390
1391int gfs2_glock_nq(struct gfs2_holder *gh)
1392{
1393 struct gfs2_glock *gl = gh->gh_gl;
1394 int error = 0;
1395
1396 if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP))
1397 return -EIO;
1398
1399 if (test_bit(GLF_LRU, &gl->gl_flags))
1400 gfs2_glock_remove_from_lru(gl);
1401
1402 spin_lock(&gl->gl_lockref.lock);
1403 add_to_queue(gh);
1404 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1405 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1406 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1407 gl->gl_lockref.count++;
1408 __gfs2_glock_queue_work(gl, 0);
1409 }
1410 run_queue(gl, 1);
1411 spin_unlock(&gl->gl_lockref.lock);
1412
1413 if (!(gh->gh_flags & GL_ASYNC))
1414 error = gfs2_glock_wait(gh);
1415
1416 return error;
1417}
1418
1419/**
1420 * gfs2_glock_poll - poll to see if an async request has been completed
1421 * @gh: the holder
1422 *
1423 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1424 */
1425
1426int gfs2_glock_poll(struct gfs2_holder *gh)
1427{
1428 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1429}
1430
1431/**
1432 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1433 * @gh: the glock holder
1434 *
1435 */
1436
1437void gfs2_glock_dq(struct gfs2_holder *gh)
1438{
1439 struct gfs2_glock *gl = gh->gh_gl;
1440 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1441 unsigned delay = 0;
1442 int fast_path = 0;
1443
1444 spin_lock(&gl->gl_lockref.lock);
1445 /*
1446 * If we're in the process of file system withdraw, we cannot just
1447 * dequeue any glocks until our journal is recovered, lest we
1448 * introduce file system corruption. We need two exceptions to this
1449 * rule: We need to allow unlocking of nondisk glocks and the glock
1450 * for our own journal that needs recovery.
1451 */
1452 if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) &&
1453 glock_blocked_by_withdraw(gl) &&
1454 gh->gh_gl != sdp->sd_jinode_gl) {
1455 sdp->sd_glock_dqs_held++;
1456 might_sleep();
1457 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
1458 TASK_UNINTERRUPTIBLE);
1459 }
1460 if (gh->gh_flags & GL_NOCACHE)
1461 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1462
1463 list_del_init(&gh->gh_list);
1464 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1465 if (find_first_holder(gl) == NULL) {
1466 if (list_empty(&gl->gl_holders) &&
1467 !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1468 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1469 fast_path = 1;
1470 }
1471 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1472 gfs2_glock_add_to_lru(gl);
1473
1474 trace_gfs2_glock_queue(gh, 0);
1475 if (unlikely(!fast_path)) {
1476 gl->gl_lockref.count++;
1477 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1478 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1479 gl->gl_name.ln_type == LM_TYPE_INODE)
1480 delay = gl->gl_hold_time;
1481 __gfs2_glock_queue_work(gl, delay);
1482 }
1483 spin_unlock(&gl->gl_lockref.lock);
1484}
1485
1486void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1487{
1488 struct gfs2_glock *gl = gh->gh_gl;
1489 gfs2_glock_dq(gh);
1490 might_sleep();
1491 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1492}
1493
1494/**
1495 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1496 * @gh: the holder structure
1497 *
1498 */
1499
1500void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1501{
1502 gfs2_glock_dq(gh);
1503 gfs2_holder_uninit(gh);
1504}
1505
1506/**
1507 * gfs2_glock_nq_num - acquire a glock based on lock number
1508 * @sdp: the filesystem
1509 * @number: the lock number
1510 * @glops: the glock operations for the type of glock
1511 * @state: the state to acquire the glock in
1512 * @flags: modifier flags for the acquisition
1513 * @gh: the struct gfs2_holder
1514 *
1515 * Returns: errno
1516 */
1517
1518int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1519 const struct gfs2_glock_operations *glops,
1520 unsigned int state, u16 flags, struct gfs2_holder *gh)
1521{
1522 struct gfs2_glock *gl;
1523 int error;
1524
1525 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1526 if (!error) {
1527 error = gfs2_glock_nq_init(gl, state, flags, gh);
1528 gfs2_glock_put(gl);
1529 }
1530
1531 return error;
1532}
1533
1534/**
1535 * glock_compare - Compare two struct gfs2_glock structures for sorting
1536 * @arg_a: the first structure
1537 * @arg_b: the second structure
1538 *
1539 */
1540
1541static int glock_compare(const void *arg_a, const void *arg_b)
1542{
1543 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1544 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1545 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1546 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1547
1548 if (a->ln_number > b->ln_number)
1549 return 1;
1550 if (a->ln_number < b->ln_number)
1551 return -1;
1552 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1553 return 0;
1554}
1555
1556/**
1557 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1558 * @num_gh: the number of structures
1559 * @ghs: an array of struct gfs2_holder structures
1560 *
1561 * Returns: 0 on success (all glocks acquired),
1562 * errno on failure (no glocks acquired)
1563 */
1564
1565static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1566 struct gfs2_holder **p)
1567{
1568 unsigned int x;
1569 int error = 0;
1570
1571 for (x = 0; x < num_gh; x++)
1572 p[x] = &ghs[x];
1573
1574 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1575
1576 for (x = 0; x < num_gh; x++) {
1577 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1578
1579 error = gfs2_glock_nq(p[x]);
1580 if (error) {
1581 while (x--)
1582 gfs2_glock_dq(p[x]);
1583 break;
1584 }
1585 }
1586
1587 return error;
1588}
1589
1590/**
1591 * gfs2_glock_nq_m - acquire multiple glocks
1592 * @num_gh: the number of structures
1593 * @ghs: an array of struct gfs2_holder structures
1594 *
1595 *
1596 * Returns: 0 on success (all glocks acquired),
1597 * errno on failure (no glocks acquired)
1598 */
1599
1600int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1601{
1602 struct gfs2_holder *tmp[4];
1603 struct gfs2_holder **pph = tmp;
1604 int error = 0;
1605
1606 switch(num_gh) {
1607 case 0:
1608 return 0;
1609 case 1:
1610 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1611 return gfs2_glock_nq(ghs);
1612 default:
1613 if (num_gh <= 4)
1614 break;
1615 pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *),
1616 GFP_NOFS);
1617 if (!pph)
1618 return -ENOMEM;
1619 }
1620
1621 error = nq_m_sync(num_gh, ghs, pph);
1622
1623 if (pph != tmp)
1624 kfree(pph);
1625
1626 return error;
1627}
1628
1629/**
1630 * gfs2_glock_dq_m - release multiple glocks
1631 * @num_gh: the number of structures
1632 * @ghs: an array of struct gfs2_holder structures
1633 *
1634 */
1635
1636void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1637{
1638 while (num_gh--)
1639 gfs2_glock_dq(&ghs[num_gh]);
1640}
1641
1642void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1643{
1644 unsigned long delay = 0;
1645 unsigned long holdtime;
1646 unsigned long now = jiffies;
1647
1648 gfs2_glock_hold(gl);
1649 holdtime = gl->gl_tchange + gl->gl_hold_time;
1650 if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1651 gl->gl_name.ln_type == LM_TYPE_INODE) {
1652 if (time_before(now, holdtime))
1653 delay = holdtime - now;
1654 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1655 delay = gl->gl_hold_time;
1656 }
1657
1658 spin_lock(&gl->gl_lockref.lock);
1659 handle_callback(gl, state, delay, true);
1660 __gfs2_glock_queue_work(gl, delay);
1661 spin_unlock(&gl->gl_lockref.lock);
1662}
1663
1664/**
1665 * gfs2_should_freeze - Figure out if glock should be frozen
1666 * @gl: The glock in question
1667 *
1668 * Glocks are not frozen if (a) the result of the dlm operation is
1669 * an error, (b) the locking operation was an unlock operation or
1670 * (c) if there is a "noexp" flagged request anywhere in the queue
1671 *
1672 * Returns: 1 if freezing should occur, 0 otherwise
1673 */
1674
1675static int gfs2_should_freeze(const struct gfs2_glock *gl)
1676{
1677 const struct gfs2_holder *gh;
1678
1679 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1680 return 0;
1681 if (gl->gl_target == LM_ST_UNLOCKED)
1682 return 0;
1683
1684 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1685 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1686 continue;
1687 if (LM_FLAG_NOEXP & gh->gh_flags)
1688 return 0;
1689 }
1690
1691 return 1;
1692}
1693
1694/**
1695 * gfs2_glock_complete - Callback used by locking
1696 * @gl: Pointer to the glock
1697 * @ret: The return value from the dlm
1698 *
1699 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1700 * to use a bitfield shared with other glock state fields.
1701 */
1702
1703void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1704{
1705 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1706
1707 spin_lock(&gl->gl_lockref.lock);
1708 gl->gl_reply = ret;
1709
1710 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1711 if (gfs2_should_freeze(gl)) {
1712 set_bit(GLF_FROZEN, &gl->gl_flags);
1713 spin_unlock(&gl->gl_lockref.lock);
1714 return;
1715 }
1716 }
1717
1718 gl->gl_lockref.count++;
1719 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1720 __gfs2_glock_queue_work(gl, 0);
1721 spin_unlock(&gl->gl_lockref.lock);
1722}
1723
1724static int glock_cmp(void *priv, struct list_head *a, struct list_head *b)
1725{
1726 struct gfs2_glock *gla, *glb;
1727
1728 gla = list_entry(a, struct gfs2_glock, gl_lru);
1729 glb = list_entry(b, struct gfs2_glock, gl_lru);
1730
1731 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1732 return 1;
1733 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1734 return -1;
1735
1736 return 0;
1737}
1738
1739/**
1740 * gfs2_dispose_glock_lru - Demote a list of glocks
1741 * @list: The list to dispose of
1742 *
1743 * Disposing of glocks may involve disk accesses, so that here we sort
1744 * the glocks by number (i.e. disk location of the inodes) so that if
1745 * there are any such accesses, they'll be sent in order (mostly).
1746 *
1747 * Must be called under the lru_lock, but may drop and retake this
1748 * lock. While the lru_lock is dropped, entries may vanish from the
1749 * list, but no new entries will appear on the list (since it is
1750 * private)
1751 */
1752
1753static void gfs2_dispose_glock_lru(struct list_head *list)
1754__releases(&lru_lock)
1755__acquires(&lru_lock)
1756{
1757 struct gfs2_glock *gl;
1758
1759 list_sort(NULL, list, glock_cmp);
1760
1761 while(!list_empty(list)) {
1762 gl = list_first_entry(list, struct gfs2_glock, gl_lru);
1763 list_del_init(&gl->gl_lru);
1764 if (!spin_trylock(&gl->gl_lockref.lock)) {
1765add_back_to_lru:
1766 list_add(&gl->gl_lru, &lru_list);
1767 set_bit(GLF_LRU, &gl->gl_flags);
1768 atomic_inc(&lru_count);
1769 continue;
1770 }
1771 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1772 spin_unlock(&gl->gl_lockref.lock);
1773 goto add_back_to_lru;
1774 }
1775 gl->gl_lockref.count++;
1776 if (demote_ok(gl))
1777 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1778 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1779 __gfs2_glock_queue_work(gl, 0);
1780 spin_unlock(&gl->gl_lockref.lock);
1781 cond_resched_lock(&lru_lock);
1782 }
1783}
1784
1785/**
1786 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1787 * @nr: The number of entries to scan
1788 *
1789 * This function selects the entries on the LRU which are able to
1790 * be demoted, and then kicks off the process by calling
1791 * gfs2_dispose_glock_lru() above.
1792 */
1793
1794static long gfs2_scan_glock_lru(int nr)
1795{
1796 struct gfs2_glock *gl;
1797 LIST_HEAD(skipped);
1798 LIST_HEAD(dispose);
1799 long freed = 0;
1800
1801 spin_lock(&lru_lock);
1802 while ((nr-- >= 0) && !list_empty(&lru_list)) {
1803 gl = list_first_entry(&lru_list, struct gfs2_glock, gl_lru);
1804
1805 /* Test for being demotable */
1806 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1807 list_move(&gl->gl_lru, &dispose);
1808 atomic_dec(&lru_count);
1809 clear_bit(GLF_LRU, &gl->gl_flags);
1810 freed++;
1811 continue;
1812 }
1813
1814 list_move(&gl->gl_lru, &skipped);
1815 }
1816 list_splice(&skipped, &lru_list);
1817 if (!list_empty(&dispose))
1818 gfs2_dispose_glock_lru(&dispose);
1819 spin_unlock(&lru_lock);
1820
1821 return freed;
1822}
1823
1824static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1825 struct shrink_control *sc)
1826{
1827 if (!(sc->gfp_mask & __GFP_FS))
1828 return SHRINK_STOP;
1829 return gfs2_scan_glock_lru(sc->nr_to_scan);
1830}
1831
1832static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1833 struct shrink_control *sc)
1834{
1835 return vfs_pressure_ratio(atomic_read(&lru_count));
1836}
1837
1838static struct shrinker glock_shrinker = {
1839 .seeks = DEFAULT_SEEKS,
1840 .count_objects = gfs2_glock_shrink_count,
1841 .scan_objects = gfs2_glock_shrink_scan,
1842};
1843
1844/**
1845 * examine_bucket - Call a function for glock in a hash bucket
1846 * @examiner: the function
1847 * @sdp: the filesystem
1848 * @bucket: the bucket
1849 *
1850 * Note that the function can be called multiple times on the same
1851 * object. So the user must ensure that the function can cope with
1852 * that.
1853 */
1854
1855static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1856{
1857 struct gfs2_glock *gl;
1858 struct rhashtable_iter iter;
1859
1860 rhashtable_walk_enter(&gl_hash_table, &iter);
1861
1862 do {
1863 rhashtable_walk_start(&iter);
1864
1865 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl))
1866 if (gl->gl_name.ln_sbd == sdp &&
1867 lockref_get_not_dead(&gl->gl_lockref))
1868 examiner(gl);
1869
1870 rhashtable_walk_stop(&iter);
1871 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
1872
1873 rhashtable_walk_exit(&iter);
1874}
1875
1876bool gfs2_queue_delete_work(struct gfs2_glock *gl, unsigned long delay)
1877{
1878 bool queued;
1879
1880 spin_lock(&gl->gl_lockref.lock);
1881 queued = queue_delayed_work(gfs2_delete_workqueue,
1882 &gl->gl_delete, delay);
1883 if (queued)
1884 set_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1885 spin_unlock(&gl->gl_lockref.lock);
1886 return queued;
1887}
1888
1889void gfs2_cancel_delete_work(struct gfs2_glock *gl)
1890{
1891 if (cancel_delayed_work_sync(&gl->gl_delete)) {
1892 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1893 gfs2_glock_put(gl);
1894 }
1895}
1896
1897bool gfs2_delete_work_queued(const struct gfs2_glock *gl)
1898{
1899 return test_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1900}
1901
1902static void flush_delete_work(struct gfs2_glock *gl)
1903{
1904 if (cancel_delayed_work(&gl->gl_delete)) {
1905 queue_delayed_work(gfs2_delete_workqueue,
1906 &gl->gl_delete, 0);
1907 }
1908 gfs2_glock_queue_work(gl, 0);
1909}
1910
1911void gfs2_flush_delete_work(struct gfs2_sbd *sdp)
1912{
1913 glock_hash_walk(flush_delete_work, sdp);
1914 flush_workqueue(gfs2_delete_workqueue);
1915}
1916
1917/**
1918 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1919 * @gl: The glock to thaw
1920 *
1921 */
1922
1923static void thaw_glock(struct gfs2_glock *gl)
1924{
1925 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags)) {
1926 gfs2_glock_put(gl);
1927 return;
1928 }
1929 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1930 gfs2_glock_queue_work(gl, 0);
1931}
1932
1933/**
1934 * clear_glock - look at a glock and see if we can free it from glock cache
1935 * @gl: the glock to look at
1936 *
1937 */
1938
1939static void clear_glock(struct gfs2_glock *gl)
1940{
1941 gfs2_glock_remove_from_lru(gl);
1942
1943 spin_lock(&gl->gl_lockref.lock);
1944 if (gl->gl_state != LM_ST_UNLOCKED)
1945 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1946 __gfs2_glock_queue_work(gl, 0);
1947 spin_unlock(&gl->gl_lockref.lock);
1948}
1949
1950/**
1951 * gfs2_glock_thaw - Thaw any frozen glocks
1952 * @sdp: The super block
1953 *
1954 */
1955
1956void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1957{
1958 glock_hash_walk(thaw_glock, sdp);
1959}
1960
1961static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
1962{
1963 spin_lock(&gl->gl_lockref.lock);
1964 gfs2_dump_glock(seq, gl, fsid);
1965 spin_unlock(&gl->gl_lockref.lock);
1966}
1967
1968static void dump_glock_func(struct gfs2_glock *gl)
1969{
1970 dump_glock(NULL, gl, true);
1971}
1972
1973/**
1974 * gfs2_gl_hash_clear - Empty out the glock hash table
1975 * @sdp: the filesystem
1976 * @wait: wait until it's all gone
1977 *
1978 * Called when unmounting the filesystem.
1979 */
1980
1981void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1982{
1983 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1984 flush_workqueue(glock_workqueue);
1985 glock_hash_walk(clear_glock, sdp);
1986 flush_workqueue(glock_workqueue);
1987 wait_event_timeout(sdp->sd_glock_wait,
1988 atomic_read(&sdp->sd_glock_disposal) == 0,
1989 HZ * 600);
1990 glock_hash_walk(dump_glock_func, sdp);
1991}
1992
1993void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1994{
1995 struct gfs2_glock *gl = ip->i_gl;
1996 int ret;
1997
1998 ret = gfs2_truncatei_resume(ip);
1999 gfs2_glock_assert_withdraw(gl, ret == 0);
2000
2001 spin_lock(&gl->gl_lockref.lock);
2002 clear_bit(GLF_LOCK, &gl->gl_flags);
2003 run_queue(gl, 1);
2004 spin_unlock(&gl->gl_lockref.lock);
2005}
2006
2007static const char *state2str(unsigned state)
2008{
2009 switch(state) {
2010 case LM_ST_UNLOCKED:
2011 return "UN";
2012 case LM_ST_SHARED:
2013 return "SH";
2014 case LM_ST_DEFERRED:
2015 return "DF";
2016 case LM_ST_EXCLUSIVE:
2017 return "EX";
2018 }
2019 return "??";
2020}
2021
2022static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
2023{
2024 char *p = buf;
2025 if (flags & LM_FLAG_TRY)
2026 *p++ = 't';
2027 if (flags & LM_FLAG_TRY_1CB)
2028 *p++ = 'T';
2029 if (flags & LM_FLAG_NOEXP)
2030 *p++ = 'e';
2031 if (flags & LM_FLAG_ANY)
2032 *p++ = 'A';
2033 if (flags & LM_FLAG_PRIORITY)
2034 *p++ = 'p';
2035 if (flags & GL_ASYNC)
2036 *p++ = 'a';
2037 if (flags & GL_EXACT)
2038 *p++ = 'E';
2039 if (flags & GL_NOCACHE)
2040 *p++ = 'c';
2041 if (test_bit(HIF_HOLDER, &iflags))
2042 *p++ = 'H';
2043 if (test_bit(HIF_WAIT, &iflags))
2044 *p++ = 'W';
2045 if (test_bit(HIF_FIRST, &iflags))
2046 *p++ = 'F';
2047 *p = 0;
2048 return buf;
2049}
2050
2051/**
2052 * dump_holder - print information about a glock holder
2053 * @seq: the seq_file struct
2054 * @gh: the glock holder
2055 * @fs_id_buf: pointer to file system id (if requested)
2056 *
2057 */
2058
2059static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
2060 const char *fs_id_buf)
2061{
2062 struct task_struct *gh_owner = NULL;
2063 char flags_buf[32];
2064
2065 rcu_read_lock();
2066 if (gh->gh_owner_pid)
2067 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
2068 gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2069 fs_id_buf, state2str(gh->gh_state),
2070 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
2071 gh->gh_error,
2072 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
2073 gh_owner ? gh_owner->comm : "(ended)",
2074 (void *)gh->gh_ip);
2075 rcu_read_unlock();
2076}
2077
2078static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
2079{
2080 const unsigned long *gflags = &gl->gl_flags;
2081 char *p = buf;
2082
2083 if (test_bit(GLF_LOCK, gflags))
2084 *p++ = 'l';
2085 if (test_bit(GLF_DEMOTE, gflags))
2086 *p++ = 'D';
2087 if (test_bit(GLF_PENDING_DEMOTE, gflags))
2088 *p++ = 'd';
2089 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
2090 *p++ = 'p';
2091 if (test_bit(GLF_DIRTY, gflags))
2092 *p++ = 'y';
2093 if (test_bit(GLF_LFLUSH, gflags))
2094 *p++ = 'f';
2095 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
2096 *p++ = 'i';
2097 if (test_bit(GLF_REPLY_PENDING, gflags))
2098 *p++ = 'r';
2099 if (test_bit(GLF_INITIAL, gflags))
2100 *p++ = 'I';
2101 if (test_bit(GLF_FROZEN, gflags))
2102 *p++ = 'F';
2103 if (test_bit(GLF_QUEUED, gflags))
2104 *p++ = 'q';
2105 if (test_bit(GLF_LRU, gflags))
2106 *p++ = 'L';
2107 if (gl->gl_object)
2108 *p++ = 'o';
2109 if (test_bit(GLF_BLOCKING, gflags))
2110 *p++ = 'b';
2111 if (test_bit(GLF_INODE_CREATING, gflags))
2112 *p++ = 'c';
2113 if (test_bit(GLF_PENDING_DELETE, gflags))
2114 *p++ = 'P';
2115 if (test_bit(GLF_FREEING, gflags))
2116 *p++ = 'x';
2117 *p = 0;
2118 return buf;
2119}
2120
2121/**
2122 * gfs2_dump_glock - print information about a glock
2123 * @seq: The seq_file struct
2124 * @gl: the glock
2125 * @fsid: If true, also dump the file system id
2126 *
2127 * The file format is as follows:
2128 * One line per object, capital letters are used to indicate objects
2129 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2130 * other objects are indented by a single space and follow the glock to
2131 * which they are related. Fields are indicated by lower case letters
2132 * followed by a colon and the field value, except for strings which are in
2133 * [] so that its possible to see if they are composed of spaces for
2134 * example. The field's are n = number (id of the object), f = flags,
2135 * t = type, s = state, r = refcount, e = error, p = pid.
2136 *
2137 */
2138
2139void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2140{
2141 const struct gfs2_glock_operations *glops = gl->gl_ops;
2142 unsigned long long dtime;
2143 const struct gfs2_holder *gh;
2144 char gflags_buf[32];
2145 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2146 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2147 unsigned long nrpages = 0;
2148
2149 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
2150 struct address_space *mapping = gfs2_glock2aspace(gl);
2151
2152 nrpages = mapping->nrpages;
2153 }
2154 memset(fs_id_buf, 0, sizeof(fs_id_buf));
2155 if (fsid && sdp) /* safety precaution */
2156 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2157 dtime = jiffies - gl->gl_demote_time;
2158 dtime *= 1000000/HZ; /* demote time in uSec */
2159 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
2160 dtime = 0;
2161 gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2162 "v:%d r:%d m:%ld p:%lu\n",
2163 fs_id_buf, state2str(gl->gl_state),
2164 gl->gl_name.ln_type,
2165 (unsigned long long)gl->gl_name.ln_number,
2166 gflags2str(gflags_buf, gl),
2167 state2str(gl->gl_target),
2168 state2str(gl->gl_demote_state), dtime,
2169 atomic_read(&gl->gl_ail_count),
2170 atomic_read(&gl->gl_revokes),
2171 (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages);
2172
2173 list_for_each_entry(gh, &gl->gl_holders, gh_list)
2174 dump_holder(seq, gh, fs_id_buf);
2175
2176 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
2177 glops->go_dump(seq, gl, fs_id_buf);
2178}
2179
2180static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
2181{
2182 struct gfs2_glock *gl = iter_ptr;
2183
2184 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2185 gl->gl_name.ln_type,
2186 (unsigned long long)gl->gl_name.ln_number,
2187 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
2188 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
2189 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
2190 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
2191 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
2192 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
2193 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
2194 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
2195 return 0;
2196}
2197
2198static const char *gfs2_gltype[] = {
2199 "type",
2200 "reserved",
2201 "nondisk",
2202 "inode",
2203 "rgrp",
2204 "meta",
2205 "iopen",
2206 "flock",
2207 "plock",
2208 "quota",
2209 "journal",
2210};
2211
2212static const char *gfs2_stype[] = {
2213 [GFS2_LKS_SRTT] = "srtt",
2214 [GFS2_LKS_SRTTVAR] = "srttvar",
2215 [GFS2_LKS_SRTTB] = "srttb",
2216 [GFS2_LKS_SRTTVARB] = "srttvarb",
2217 [GFS2_LKS_SIRT] = "sirt",
2218 [GFS2_LKS_SIRTVAR] = "sirtvar",
2219 [GFS2_LKS_DCOUNT] = "dlm",
2220 [GFS2_LKS_QCOUNT] = "queue",
2221};
2222
2223#define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2224
2225static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
2226{
2227 struct gfs2_sbd *sdp = seq->private;
2228 loff_t pos = *(loff_t *)iter_ptr;
2229 unsigned index = pos >> 3;
2230 unsigned subindex = pos & 0x07;
2231 int i;
2232
2233 if (index == 0 && subindex != 0)
2234 return 0;
2235
2236 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
2237 (index == 0) ? "cpu": gfs2_stype[subindex]);
2238
2239 for_each_possible_cpu(i) {
2240 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
2241
2242 if (index == 0)
2243 seq_printf(seq, " %15u", i);
2244 else
2245 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
2246 lkstats[index - 1].stats[subindex]);
2247 }
2248 seq_putc(seq, '\n');
2249 return 0;
2250}
2251
2252int __init gfs2_glock_init(void)
2253{
2254 int i, ret;
2255
2256 ret = rhashtable_init(&gl_hash_table, &ht_parms);
2257 if (ret < 0)
2258 return ret;
2259
2260 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
2261 WQ_HIGHPRI | WQ_FREEZABLE, 0);
2262 if (!glock_workqueue) {
2263 rhashtable_destroy(&gl_hash_table);
2264 return -ENOMEM;
2265 }
2266 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
2267 WQ_MEM_RECLAIM | WQ_FREEZABLE,
2268 0);
2269 if (!gfs2_delete_workqueue) {
2270 destroy_workqueue(glock_workqueue);
2271 rhashtable_destroy(&gl_hash_table);
2272 return -ENOMEM;
2273 }
2274
2275 ret = register_shrinker(&glock_shrinker);
2276 if (ret) {
2277 destroy_workqueue(gfs2_delete_workqueue);
2278 destroy_workqueue(glock_workqueue);
2279 rhashtable_destroy(&gl_hash_table);
2280 return ret;
2281 }
2282
2283 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2284 init_waitqueue_head(glock_wait_table + i);
2285
2286 return 0;
2287}
2288
2289void gfs2_glock_exit(void)
2290{
2291 unregister_shrinker(&glock_shrinker);
2292 rhashtable_destroy(&gl_hash_table);
2293 destroy_workqueue(glock_workqueue);
2294 destroy_workqueue(gfs2_delete_workqueue);
2295}
2296
2297static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2298{
2299 struct gfs2_glock *gl = gi->gl;
2300
2301 if (gl) {
2302 if (n == 0)
2303 return;
2304 if (!lockref_put_not_zero(&gl->gl_lockref))
2305 gfs2_glock_queue_put(gl);
2306 }
2307 for (;;) {
2308 gl = rhashtable_walk_next(&gi->hti);
2309 if (IS_ERR_OR_NULL(gl)) {
2310 if (gl == ERR_PTR(-EAGAIN)) {
2311 n = 1;
2312 continue;
2313 }
2314 gl = NULL;
2315 break;
2316 }
2317 if (gl->gl_name.ln_sbd != gi->sdp)
2318 continue;
2319 if (n <= 1) {
2320 if (!lockref_get_not_dead(&gl->gl_lockref))
2321 continue;
2322 break;
2323 } else {
2324 if (__lockref_is_dead(&gl->gl_lockref))
2325 continue;
2326 n--;
2327 }
2328 }
2329 gi->gl = gl;
2330}
2331
2332static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2333 __acquires(RCU)
2334{
2335 struct gfs2_glock_iter *gi = seq->private;
2336 loff_t n;
2337
2338 /*
2339 * We can either stay where we are, skip to the next hash table
2340 * entry, or start from the beginning.
2341 */
2342 if (*pos < gi->last_pos) {
2343 rhashtable_walk_exit(&gi->hti);
2344 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2345 n = *pos + 1;
2346 } else {
2347 n = *pos - gi->last_pos;
2348 }
2349
2350 rhashtable_walk_start(&gi->hti);
2351
2352 gfs2_glock_iter_next(gi, n);
2353 gi->last_pos = *pos;
2354 return gi->gl;
2355}
2356
2357static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2358 loff_t *pos)
2359{
2360 struct gfs2_glock_iter *gi = seq->private;
2361
2362 (*pos)++;
2363 gi->last_pos = *pos;
2364 gfs2_glock_iter_next(gi, 1);
2365 return gi->gl;
2366}
2367
2368static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2369 __releases(RCU)
2370{
2371 struct gfs2_glock_iter *gi = seq->private;
2372
2373 rhashtable_walk_stop(&gi->hti);
2374}
2375
2376static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2377{
2378 dump_glock(seq, iter_ptr, false);
2379 return 0;
2380}
2381
2382static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2383{
2384 preempt_disable();
2385 if (*pos >= GFS2_NR_SBSTATS)
2386 return NULL;
2387 return pos;
2388}
2389
2390static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2391 loff_t *pos)
2392{
2393 (*pos)++;
2394 if (*pos >= GFS2_NR_SBSTATS)
2395 return NULL;
2396 return pos;
2397}
2398
2399static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2400{
2401 preempt_enable();
2402}
2403
2404static const struct seq_operations gfs2_glock_seq_ops = {
2405 .start = gfs2_glock_seq_start,
2406 .next = gfs2_glock_seq_next,
2407 .stop = gfs2_glock_seq_stop,
2408 .show = gfs2_glock_seq_show,
2409};
2410
2411static const struct seq_operations gfs2_glstats_seq_ops = {
2412 .start = gfs2_glock_seq_start,
2413 .next = gfs2_glock_seq_next,
2414 .stop = gfs2_glock_seq_stop,
2415 .show = gfs2_glstats_seq_show,
2416};
2417
2418static const struct seq_operations gfs2_sbstats_seq_ops = {
2419 .start = gfs2_sbstats_seq_start,
2420 .next = gfs2_sbstats_seq_next,
2421 .stop = gfs2_sbstats_seq_stop,
2422 .show = gfs2_sbstats_seq_show,
2423};
2424
2425#define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2426
2427static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2428 const struct seq_operations *ops)
2429{
2430 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2431 if (ret == 0) {
2432 struct seq_file *seq = file->private_data;
2433 struct gfs2_glock_iter *gi = seq->private;
2434
2435 gi->sdp = inode->i_private;
2436 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2437 if (seq->buf)
2438 seq->size = GFS2_SEQ_GOODSIZE;
2439 /*
2440 * Initially, we are "before" the first hash table entry; the
2441 * first call to rhashtable_walk_next gets us the first entry.
2442 */
2443 gi->last_pos = -1;
2444 gi->gl = NULL;
2445 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2446 }
2447 return ret;
2448}
2449
2450static int gfs2_glocks_open(struct inode *inode, struct file *file)
2451{
2452 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2453}
2454
2455static int gfs2_glocks_release(struct inode *inode, struct file *file)
2456{
2457 struct seq_file *seq = file->private_data;
2458 struct gfs2_glock_iter *gi = seq->private;
2459
2460 if (gi->gl)
2461 gfs2_glock_put(gi->gl);
2462 rhashtable_walk_exit(&gi->hti);
2463 return seq_release_private(inode, file);
2464}
2465
2466static int gfs2_glstats_open(struct inode *inode, struct file *file)
2467{
2468 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2469}
2470
2471static int gfs2_sbstats_open(struct inode *inode, struct file *file)
2472{
2473 int ret = seq_open(file, &gfs2_sbstats_seq_ops);
2474 if (ret == 0) {
2475 struct seq_file *seq = file->private_data;
2476 seq->private = inode->i_private; /* sdp */
2477 }
2478 return ret;
2479}
2480
2481static const struct file_operations gfs2_glocks_fops = {
2482 .owner = THIS_MODULE,
2483 .open = gfs2_glocks_open,
2484 .read = seq_read,
2485 .llseek = seq_lseek,
2486 .release = gfs2_glocks_release,
2487};
2488
2489static const struct file_operations gfs2_glstats_fops = {
2490 .owner = THIS_MODULE,
2491 .open = gfs2_glstats_open,
2492 .read = seq_read,
2493 .llseek = seq_lseek,
2494 .release = gfs2_glocks_release,
2495};
2496
2497static const struct file_operations gfs2_sbstats_fops = {
2498 .owner = THIS_MODULE,
2499 .open = gfs2_sbstats_open,
2500 .read = seq_read,
2501 .llseek = seq_lseek,
2502 .release = seq_release,
2503};
2504
2505void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2506{
2507 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2508
2509 debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2510 &gfs2_glocks_fops);
2511
2512 debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2513 &gfs2_glstats_fops);
2514
2515 debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2516 &gfs2_sbstats_fops);
2517}
2518
2519void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2520{
2521 debugfs_remove_recursive(sdp->debugfs_dir);
2522 sdp->debugfs_dir = NULL;
2523}
2524
2525void gfs2_register_debugfs(void)
2526{
2527 gfs2_root = debugfs_create_dir("gfs2", NULL);
2528}
2529
2530void gfs2_unregister_debugfs(void)
2531{
2532 debugfs_remove(gfs2_root);
2533 gfs2_root = NULL;
2534}