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#include <linux/percpu.h>
33
34#include "gfs2.h"
35#include "incore.h"
36#include "glock.h"
37#include "glops.h"
38#include "inode.h"
39#include "lops.h"
40#include "meta_io.h"
41#include "quota.h"
42#include "super.h"
43#include "util.h"
44#include "bmap.h"
45#define CREATE_TRACE_POINTS
46#include "trace_gfs2.h"
47
48struct gfs2_glock_iter {
49 int hash; /* hash bucket index */
50 struct gfs2_sbd *sdp; /* incore superblock */
51 struct gfs2_glock *gl; /* current glock struct */
52 char string[512]; /* scratch space */
53};
54
55typedef void (*glock_examiner) (struct gfs2_glock * gl);
56
57static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl);
58#define GLOCK_BUG_ON(gl,x) do { if (unlikely(x)) { __dump_glock(NULL, gl); BUG(); } } while(0)
59static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
60
61static struct dentry *gfs2_root;
62static struct workqueue_struct *glock_workqueue;
63struct workqueue_struct *gfs2_delete_workqueue;
64static LIST_HEAD(lru_list);
65static atomic_t lru_count = ATOMIC_INIT(0);
66static DEFINE_SPINLOCK(lru_lock);
67
68#define GFS2_GL_HASH_SHIFT 15
69#define GFS2_GL_HASH_SIZE (1 << GFS2_GL_HASH_SHIFT)
70#define GFS2_GL_HASH_MASK (GFS2_GL_HASH_SIZE - 1)
71
72static struct hlist_bl_head gl_hash_table[GFS2_GL_HASH_SIZE];
73static struct dentry *gfs2_root;
74
75/**
76 * gl_hash() - Turn glock number into hash bucket number
77 * @lock: The glock number
78 *
79 * Returns: The number of the corresponding hash bucket
80 */
81
82static unsigned int gl_hash(const struct gfs2_sbd *sdp,
83 const struct lm_lockname *name)
84{
85 unsigned int h;
86
87 h = jhash(&name->ln_number, sizeof(u64), 0);
88 h = jhash(&name->ln_type, sizeof(unsigned int), h);
89 h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
90 h &= GFS2_GL_HASH_MASK;
91
92 return h;
93}
94
95static inline void spin_lock_bucket(unsigned int hash)
96{
97 hlist_bl_lock(&gl_hash_table[hash]);
98}
99
100static inline void spin_unlock_bucket(unsigned int hash)
101{
102 hlist_bl_unlock(&gl_hash_table[hash]);
103}
104
105static void gfs2_glock_dealloc(struct rcu_head *rcu)
106{
107 struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
108
109 if (gl->gl_ops->go_flags & GLOF_ASPACE)
110 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
111 else
112 kmem_cache_free(gfs2_glock_cachep, gl);
113}
114
115void gfs2_glock_free(struct gfs2_glock *gl)
116{
117 struct gfs2_sbd *sdp = gl->gl_sbd;
118
119 call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
120 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
121 wake_up(&sdp->sd_glock_wait);
122}
123
124/**
125 * gfs2_glock_hold() - increment reference count on glock
126 * @gl: The glock to hold
127 *
128 */
129
130void gfs2_glock_hold(struct gfs2_glock *gl)
131{
132 GLOCK_BUG_ON(gl, atomic_read(&gl->gl_ref) == 0);
133 atomic_inc(&gl->gl_ref);
134}
135
136/**
137 * demote_ok - Check to see if it's ok to unlock a glock
138 * @gl: the glock
139 *
140 * Returns: 1 if it's ok
141 */
142
143static int demote_ok(const struct gfs2_glock *gl)
144{
145 const struct gfs2_glock_operations *glops = gl->gl_ops;
146
147 if (gl->gl_state == LM_ST_UNLOCKED)
148 return 0;
149 if (!list_empty(&gl->gl_holders))
150 return 0;
151 if (glops->go_demote_ok)
152 return glops->go_demote_ok(gl);
153 return 1;
154}
155
156
157void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
158{
159 spin_lock(&lru_lock);
160
161 if (!list_empty(&gl->gl_lru))
162 list_del_init(&gl->gl_lru);
163 else
164 atomic_inc(&lru_count);
165
166 list_add_tail(&gl->gl_lru, &lru_list);
167 set_bit(GLF_LRU, &gl->gl_flags);
168 spin_unlock(&lru_lock);
169}
170
171static void __gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
172{
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}
179
180static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
181{
182 spin_lock(&lru_lock);
183 __gfs2_glock_remove_from_lru(gl);
184 spin_unlock(&lru_lock);
185}
186
187/**
188 * __gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
189 * @gl: the glock
190 *
191 * If the glock is demotable, then we add it (or move it) to the end
192 * of the glock LRU list.
193 */
194
195static void __gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
196{
197 if (demote_ok(gl))
198 gfs2_glock_add_to_lru(gl);
199}
200
201/**
202 * gfs2_glock_put_nolock() - Decrement reference count on glock
203 * @gl: The glock to put
204 *
205 * This function should only be used if the caller has its own reference
206 * to the glock, in addition to the one it is dropping.
207 */
208
209void gfs2_glock_put_nolock(struct gfs2_glock *gl)
210{
211 if (atomic_dec_and_test(&gl->gl_ref))
212 GLOCK_BUG_ON(gl, 1);
213}
214
215/**
216 * gfs2_glock_put() - Decrement reference count on glock
217 * @gl: The glock to put
218 *
219 */
220
221void gfs2_glock_put(struct gfs2_glock *gl)
222{
223 struct gfs2_sbd *sdp = gl->gl_sbd;
224 struct address_space *mapping = gfs2_glock2aspace(gl);
225
226 if (atomic_dec_and_lock(&gl->gl_ref, &lru_lock)) {
227 __gfs2_glock_remove_from_lru(gl);
228 spin_unlock(&lru_lock);
229 spin_lock_bucket(gl->gl_hash);
230 hlist_bl_del_rcu(&gl->gl_list);
231 spin_unlock_bucket(gl->gl_hash);
232 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
233 GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
234 trace_gfs2_glock_put(gl);
235 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
236 }
237}
238
239/**
240 * search_bucket() - Find struct gfs2_glock by lock number
241 * @bucket: the bucket to search
242 * @name: The lock name
243 *
244 * Returns: NULL, or the struct gfs2_glock with the requested number
245 */
246
247static struct gfs2_glock *search_bucket(unsigned int hash,
248 const struct gfs2_sbd *sdp,
249 const struct lm_lockname *name)
250{
251 struct gfs2_glock *gl;
252 struct hlist_bl_node *h;
253
254 hlist_bl_for_each_entry_rcu(gl, h, &gl_hash_table[hash], gl_list) {
255 if (!lm_name_equal(&gl->gl_name, name))
256 continue;
257 if (gl->gl_sbd != sdp)
258 continue;
259 if (atomic_inc_not_zero(&gl->gl_ref))
260 return gl;
261 }
262
263 return NULL;
264}
265
266/**
267 * may_grant - check if its ok to grant a new lock
268 * @gl: The glock
269 * @gh: The lock request which we wish to grant
270 *
271 * Returns: true if its ok to grant the lock
272 */
273
274static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
275{
276 const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
277 if ((gh->gh_state == LM_ST_EXCLUSIVE ||
278 gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
279 return 0;
280 if (gl->gl_state == gh->gh_state)
281 return 1;
282 if (gh->gh_flags & GL_EXACT)
283 return 0;
284 if (gl->gl_state == LM_ST_EXCLUSIVE) {
285 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
286 return 1;
287 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
288 return 1;
289 }
290 if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
291 return 1;
292 return 0;
293}
294
295static void gfs2_holder_wake(struct gfs2_holder *gh)
296{
297 clear_bit(HIF_WAIT, &gh->gh_iflags);
298 smp_mb__after_clear_bit();
299 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
300}
301
302/**
303 * do_error - Something unexpected has happened during a lock request
304 *
305 */
306
307static inline void do_error(struct gfs2_glock *gl, const int ret)
308{
309 struct gfs2_holder *gh, *tmp;
310
311 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
312 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
313 continue;
314 if (ret & LM_OUT_ERROR)
315 gh->gh_error = -EIO;
316 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
317 gh->gh_error = GLR_TRYFAILED;
318 else
319 continue;
320 list_del_init(&gh->gh_list);
321 trace_gfs2_glock_queue(gh, 0);
322 gfs2_holder_wake(gh);
323 }
324}
325
326/**
327 * do_promote - promote as many requests as possible on the current queue
328 * @gl: The glock
329 *
330 * Returns: 1 if there is a blocked holder at the head of the list, or 2
331 * if a type specific operation is underway.
332 */
333
334static int do_promote(struct gfs2_glock *gl)
335__releases(&gl->gl_spin)
336__acquires(&gl->gl_spin)
337{
338 const struct gfs2_glock_operations *glops = gl->gl_ops;
339 struct gfs2_holder *gh, *tmp;
340 int ret;
341
342restart:
343 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
344 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
345 continue;
346 if (may_grant(gl, gh)) {
347 if (gh->gh_list.prev == &gl->gl_holders &&
348 glops->go_lock) {
349 spin_unlock(&gl->gl_spin);
350 /* FIXME: eliminate this eventually */
351 ret = glops->go_lock(gh);
352 spin_lock(&gl->gl_spin);
353 if (ret) {
354 if (ret == 1)
355 return 2;
356 gh->gh_error = ret;
357 list_del_init(&gh->gh_list);
358 trace_gfs2_glock_queue(gh, 0);
359 gfs2_holder_wake(gh);
360 goto restart;
361 }
362 set_bit(HIF_HOLDER, &gh->gh_iflags);
363 trace_gfs2_promote(gh, 1);
364 gfs2_holder_wake(gh);
365 goto restart;
366 }
367 set_bit(HIF_HOLDER, &gh->gh_iflags);
368 trace_gfs2_promote(gh, 0);
369 gfs2_holder_wake(gh);
370 continue;
371 }
372 if (gh->gh_list.prev == &gl->gl_holders)
373 return 1;
374 do_error(gl, 0);
375 break;
376 }
377 return 0;
378}
379
380/**
381 * find_first_waiter - find the first gh that's waiting for the glock
382 * @gl: the glock
383 */
384
385static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
386{
387 struct gfs2_holder *gh;
388
389 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
390 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
391 return gh;
392 }
393 return NULL;
394}
395
396/**
397 * state_change - record that the glock is now in a different state
398 * @gl: the glock
399 * @new_state the new state
400 *
401 */
402
403static void state_change(struct gfs2_glock *gl, unsigned int new_state)
404{
405 int held1, held2;
406
407 held1 = (gl->gl_state != LM_ST_UNLOCKED);
408 held2 = (new_state != LM_ST_UNLOCKED);
409
410 if (held1 != held2) {
411 if (held2)
412 gfs2_glock_hold(gl);
413 else
414 gfs2_glock_put_nolock(gl);
415 }
416 if (held1 && held2 && list_empty(&gl->gl_holders))
417 clear_bit(GLF_QUEUED, &gl->gl_flags);
418
419 if (new_state != gl->gl_target)
420 /* shorten our minimum hold time */
421 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
422 GL_GLOCK_MIN_HOLD);
423 gl->gl_state = new_state;
424 gl->gl_tchange = jiffies;
425}
426
427static void gfs2_demote_wake(struct gfs2_glock *gl)
428{
429 gl->gl_demote_state = LM_ST_EXCLUSIVE;
430 clear_bit(GLF_DEMOTE, &gl->gl_flags);
431 smp_mb__after_clear_bit();
432 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
433}
434
435/**
436 * finish_xmote - The DLM has replied to one of our lock requests
437 * @gl: The glock
438 * @ret: The status from the DLM
439 *
440 */
441
442static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
443{
444 const struct gfs2_glock_operations *glops = gl->gl_ops;
445 struct gfs2_holder *gh;
446 unsigned state = ret & LM_OUT_ST_MASK;
447 int rv;
448
449 spin_lock(&gl->gl_spin);
450 trace_gfs2_glock_state_change(gl, state);
451 state_change(gl, state);
452 gh = find_first_waiter(gl);
453
454 /* Demote to UN request arrived during demote to SH or DF */
455 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
456 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
457 gl->gl_target = LM_ST_UNLOCKED;
458
459 /* Check for state != intended state */
460 if (unlikely(state != gl->gl_target)) {
461 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
462 /* move to back of queue and try next entry */
463 if (ret & LM_OUT_CANCELED) {
464 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
465 list_move_tail(&gh->gh_list, &gl->gl_holders);
466 gh = find_first_waiter(gl);
467 gl->gl_target = gh->gh_state;
468 goto retry;
469 }
470 /* Some error or failed "try lock" - report it */
471 if ((ret & LM_OUT_ERROR) ||
472 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
473 gl->gl_target = gl->gl_state;
474 do_error(gl, ret);
475 goto out;
476 }
477 }
478 switch(state) {
479 /* Unlocked due to conversion deadlock, try again */
480 case LM_ST_UNLOCKED:
481retry:
482 do_xmote(gl, gh, gl->gl_target);
483 break;
484 /* Conversion fails, unlock and try again */
485 case LM_ST_SHARED:
486 case LM_ST_DEFERRED:
487 do_xmote(gl, gh, LM_ST_UNLOCKED);
488 break;
489 default: /* Everything else */
490 printk(KERN_ERR "GFS2: wanted %u got %u\n", gl->gl_target, state);
491 GLOCK_BUG_ON(gl, 1);
492 }
493 spin_unlock(&gl->gl_spin);
494 return;
495 }
496
497 /* Fast path - we got what we asked for */
498 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
499 gfs2_demote_wake(gl);
500 if (state != LM_ST_UNLOCKED) {
501 if (glops->go_xmote_bh) {
502 spin_unlock(&gl->gl_spin);
503 rv = glops->go_xmote_bh(gl, gh);
504 spin_lock(&gl->gl_spin);
505 if (rv) {
506 do_error(gl, rv);
507 goto out;
508 }
509 }
510 rv = do_promote(gl);
511 if (rv == 2)
512 goto out_locked;
513 }
514out:
515 clear_bit(GLF_LOCK, &gl->gl_flags);
516out_locked:
517 spin_unlock(&gl->gl_spin);
518}
519
520/**
521 * do_xmote - Calls the DLM to change the state of a lock
522 * @gl: The lock state
523 * @gh: The holder (only for promotes)
524 * @target: The target lock state
525 *
526 */
527
528static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
529__releases(&gl->gl_spin)
530__acquires(&gl->gl_spin)
531{
532 const struct gfs2_glock_operations *glops = gl->gl_ops;
533 struct gfs2_sbd *sdp = gl->gl_sbd;
534 unsigned int lck_flags = gh ? gh->gh_flags : 0;
535 int ret;
536
537 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
538 LM_FLAG_PRIORITY);
539 GLOCK_BUG_ON(gl, gl->gl_state == target);
540 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
541 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
542 glops->go_inval) {
543 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
544 do_error(gl, 0); /* Fail queued try locks */
545 }
546 gl->gl_req = target;
547 set_bit(GLF_BLOCKING, &gl->gl_flags);
548 if ((gl->gl_req == LM_ST_UNLOCKED) ||
549 (gl->gl_state == LM_ST_EXCLUSIVE) ||
550 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
551 clear_bit(GLF_BLOCKING, &gl->gl_flags);
552 spin_unlock(&gl->gl_spin);
553 if (glops->go_xmote_th)
554 glops->go_xmote_th(gl);
555 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
556 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
557 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
558
559 gfs2_glock_hold(gl);
560 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
561 /* lock_dlm */
562 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
563 GLOCK_BUG_ON(gl, ret);
564 } else { /* lock_nolock */
565 finish_xmote(gl, target);
566 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
567 gfs2_glock_put(gl);
568 }
569
570 spin_lock(&gl->gl_spin);
571}
572
573/**
574 * find_first_holder - find the first "holder" gh
575 * @gl: the glock
576 */
577
578static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
579{
580 struct gfs2_holder *gh;
581
582 if (!list_empty(&gl->gl_holders)) {
583 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
584 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
585 return gh;
586 }
587 return NULL;
588}
589
590/**
591 * run_queue - do all outstanding tasks related to a glock
592 * @gl: The glock in question
593 * @nonblock: True if we must not block in run_queue
594 *
595 */
596
597static void run_queue(struct gfs2_glock *gl, const int nonblock)
598__releases(&gl->gl_spin)
599__acquires(&gl->gl_spin)
600{
601 struct gfs2_holder *gh = NULL;
602 int ret;
603
604 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
605 return;
606
607 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
608
609 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
610 gl->gl_demote_state != gl->gl_state) {
611 if (find_first_holder(gl))
612 goto out_unlock;
613 if (nonblock)
614 goto out_sched;
615 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
616 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
617 gl->gl_target = gl->gl_demote_state;
618 } else {
619 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
620 gfs2_demote_wake(gl);
621 ret = do_promote(gl);
622 if (ret == 0)
623 goto out_unlock;
624 if (ret == 2)
625 goto out;
626 gh = find_first_waiter(gl);
627 gl->gl_target = gh->gh_state;
628 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
629 do_error(gl, 0); /* Fail queued try locks */
630 }
631 do_xmote(gl, gh, gl->gl_target);
632out:
633 return;
634
635out_sched:
636 clear_bit(GLF_LOCK, &gl->gl_flags);
637 smp_mb__after_clear_bit();
638 gfs2_glock_hold(gl);
639 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
640 gfs2_glock_put_nolock(gl);
641 return;
642
643out_unlock:
644 clear_bit(GLF_LOCK, &gl->gl_flags);
645 smp_mb__after_clear_bit();
646 return;
647}
648
649static void delete_work_func(struct work_struct *work)
650{
651 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
652 struct gfs2_sbd *sdp = gl->gl_sbd;
653 struct gfs2_inode *ip;
654 struct inode *inode;
655 u64 no_addr = gl->gl_name.ln_number;
656
657 ip = gl->gl_object;
658 /* Note: Unsafe to dereference ip as we don't hold right refs/locks */
659
660 if (ip)
661 inode = gfs2_ilookup(sdp->sd_vfs, no_addr, 1);
662 else
663 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
664 if (inode && !IS_ERR(inode)) {
665 d_prune_aliases(inode);
666 iput(inode);
667 }
668 gfs2_glock_put(gl);
669}
670
671static void glock_work_func(struct work_struct *work)
672{
673 unsigned long delay = 0;
674 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
675 int drop_ref = 0;
676
677 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
678 finish_xmote(gl, gl->gl_reply);
679 drop_ref = 1;
680 }
681 spin_lock(&gl->gl_spin);
682 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
683 gl->gl_state != LM_ST_UNLOCKED &&
684 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
685 unsigned long holdtime, now = jiffies;
686
687 holdtime = gl->gl_tchange + gl->gl_hold_time;
688 if (time_before(now, holdtime))
689 delay = holdtime - now;
690
691 if (!delay) {
692 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
693 set_bit(GLF_DEMOTE, &gl->gl_flags);
694 }
695 }
696 run_queue(gl, 0);
697 spin_unlock(&gl->gl_spin);
698 if (!delay)
699 gfs2_glock_put(gl);
700 else {
701 if (gl->gl_name.ln_type != LM_TYPE_INODE)
702 delay = 0;
703 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
704 gfs2_glock_put(gl);
705 }
706 if (drop_ref)
707 gfs2_glock_put(gl);
708}
709
710/**
711 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
712 * @sdp: The GFS2 superblock
713 * @number: the lock number
714 * @glops: The glock_operations to use
715 * @create: If 0, don't create the glock if it doesn't exist
716 * @glp: the glock is returned here
717 *
718 * This does not lock a glock, just finds/creates structures for one.
719 *
720 * Returns: errno
721 */
722
723int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
724 const struct gfs2_glock_operations *glops, int create,
725 struct gfs2_glock **glp)
726{
727 struct super_block *s = sdp->sd_vfs;
728 struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
729 struct gfs2_glock *gl, *tmp;
730 unsigned int hash = gl_hash(sdp, &name);
731 struct address_space *mapping;
732 struct kmem_cache *cachep;
733
734 rcu_read_lock();
735 gl = search_bucket(hash, sdp, &name);
736 rcu_read_unlock();
737
738 *glp = gl;
739 if (gl)
740 return 0;
741 if (!create)
742 return -ENOENT;
743
744 if (glops->go_flags & GLOF_ASPACE)
745 cachep = gfs2_glock_aspace_cachep;
746 else
747 cachep = gfs2_glock_cachep;
748 gl = kmem_cache_alloc(cachep, GFP_KERNEL);
749 if (!gl)
750 return -ENOMEM;
751
752 atomic_inc(&sdp->sd_glock_disposal);
753 gl->gl_sbd = sdp;
754 gl->gl_flags = 0;
755 gl->gl_name = name;
756 atomic_set(&gl->gl_ref, 1);
757 gl->gl_state = LM_ST_UNLOCKED;
758 gl->gl_target = LM_ST_UNLOCKED;
759 gl->gl_demote_state = LM_ST_EXCLUSIVE;
760 gl->gl_hash = hash;
761 gl->gl_ops = glops;
762 gl->gl_dstamp = ktime_set(0, 0);
763 preempt_disable();
764 /* We use the global stats to estimate the initial per-glock stats */
765 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
766 preempt_enable();
767 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
768 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
769 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
770 gl->gl_lksb.sb_lvbptr = gl->gl_lvb;
771 gl->gl_tchange = jiffies;
772 gl->gl_object = NULL;
773 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
774 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
775 INIT_WORK(&gl->gl_delete, delete_work_func);
776
777 mapping = gfs2_glock2aspace(gl);
778 if (mapping) {
779 mapping->a_ops = &gfs2_meta_aops;
780 mapping->host = s->s_bdev->bd_inode;
781 mapping->flags = 0;
782 mapping_set_gfp_mask(mapping, GFP_NOFS);
783 mapping->assoc_mapping = NULL;
784 mapping->backing_dev_info = s->s_bdi;
785 mapping->writeback_index = 0;
786 }
787
788 spin_lock_bucket(hash);
789 tmp = search_bucket(hash, sdp, &name);
790 if (tmp) {
791 spin_unlock_bucket(hash);
792 kmem_cache_free(cachep, gl);
793 atomic_dec(&sdp->sd_glock_disposal);
794 gl = tmp;
795 } else {
796 hlist_bl_add_head_rcu(&gl->gl_list, &gl_hash_table[hash]);
797 spin_unlock_bucket(hash);
798 }
799
800 *glp = gl;
801
802 return 0;
803}
804
805/**
806 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
807 * @gl: the glock
808 * @state: the state we're requesting
809 * @flags: the modifier flags
810 * @gh: the holder structure
811 *
812 */
813
814void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
815 struct gfs2_holder *gh)
816{
817 INIT_LIST_HEAD(&gh->gh_list);
818 gh->gh_gl = gl;
819 gh->gh_ip = (unsigned long)__builtin_return_address(0);
820 gh->gh_owner_pid = get_pid(task_pid(current));
821 gh->gh_state = state;
822 gh->gh_flags = flags;
823 gh->gh_error = 0;
824 gh->gh_iflags = 0;
825 gfs2_glock_hold(gl);
826}
827
828/**
829 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
830 * @state: the state we're requesting
831 * @flags: the modifier flags
832 * @gh: the holder structure
833 *
834 * Don't mess with the glock.
835 *
836 */
837
838void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
839{
840 gh->gh_state = state;
841 gh->gh_flags = flags;
842 gh->gh_iflags = 0;
843 gh->gh_ip = (unsigned long)__builtin_return_address(0);
844 if (gh->gh_owner_pid)
845 put_pid(gh->gh_owner_pid);
846 gh->gh_owner_pid = get_pid(task_pid(current));
847}
848
849/**
850 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
851 * @gh: the holder structure
852 *
853 */
854
855void gfs2_holder_uninit(struct gfs2_holder *gh)
856{
857 put_pid(gh->gh_owner_pid);
858 gfs2_glock_put(gh->gh_gl);
859 gh->gh_gl = NULL;
860 gh->gh_ip = 0;
861}
862
863/**
864 * gfs2_glock_holder_wait
865 * @word: unused
866 *
867 * This function and gfs2_glock_demote_wait both show up in the WCHAN
868 * field. Thus I've separated these otherwise identical functions in
869 * order to be more informative to the user.
870 */
871
872static int gfs2_glock_holder_wait(void *word)
873{
874 schedule();
875 return 0;
876}
877
878static int gfs2_glock_demote_wait(void *word)
879{
880 schedule();
881 return 0;
882}
883
884static void wait_on_holder(struct gfs2_holder *gh)
885{
886 unsigned long time1 = jiffies;
887
888 might_sleep();
889 wait_on_bit(&gh->gh_iflags, HIF_WAIT, gfs2_glock_holder_wait, TASK_UNINTERRUPTIBLE);
890 if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
891 /* Lengthen the minimum hold time. */
892 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
893 GL_GLOCK_HOLD_INCR,
894 GL_GLOCK_MAX_HOLD);
895}
896
897static void wait_on_demote(struct gfs2_glock *gl)
898{
899 might_sleep();
900 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, gfs2_glock_demote_wait, TASK_UNINTERRUPTIBLE);
901}
902
903/**
904 * handle_callback - process a demote request
905 * @gl: the glock
906 * @state: the state the caller wants us to change to
907 *
908 * There are only two requests that we are going to see in actual
909 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
910 */
911
912static void handle_callback(struct gfs2_glock *gl, unsigned int state,
913 unsigned long delay)
914{
915 int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
916
917 set_bit(bit, &gl->gl_flags);
918 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
919 gl->gl_demote_state = state;
920 gl->gl_demote_time = jiffies;
921 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
922 gl->gl_demote_state != state) {
923 gl->gl_demote_state = LM_ST_UNLOCKED;
924 }
925 if (gl->gl_ops->go_callback)
926 gl->gl_ops->go_callback(gl);
927 trace_gfs2_demote_rq(gl);
928}
929
930/**
931 * gfs2_glock_wait - wait on a glock acquisition
932 * @gh: the glock holder
933 *
934 * Returns: 0 on success
935 */
936
937int gfs2_glock_wait(struct gfs2_holder *gh)
938{
939 wait_on_holder(gh);
940 return gh->gh_error;
941}
942
943void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
944{
945 struct va_format vaf;
946 va_list args;
947
948 va_start(args, fmt);
949
950 if (seq) {
951 struct gfs2_glock_iter *gi = seq->private;
952 vsprintf(gi->string, fmt, args);
953 seq_printf(seq, gi->string);
954 } else {
955 vaf.fmt = fmt;
956 vaf.va = &args;
957
958 printk(KERN_ERR " %pV", &vaf);
959 }
960
961 va_end(args);
962}
963
964/**
965 * add_to_queue - Add a holder to the wait queue (but look for recursion)
966 * @gh: the holder structure to add
967 *
968 * Eventually we should move the recursive locking trap to a
969 * debugging option or something like that. This is the fast
970 * path and needs to have the minimum number of distractions.
971 *
972 */
973
974static inline void add_to_queue(struct gfs2_holder *gh)
975__releases(&gl->gl_spin)
976__acquires(&gl->gl_spin)
977{
978 struct gfs2_glock *gl = gh->gh_gl;
979 struct gfs2_sbd *sdp = gl->gl_sbd;
980 struct list_head *insert_pt = NULL;
981 struct gfs2_holder *gh2;
982 int try_lock = 0;
983
984 BUG_ON(gh->gh_owner_pid == NULL);
985 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
986 BUG();
987
988 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
989 if (test_bit(GLF_LOCK, &gl->gl_flags))
990 try_lock = 1;
991 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
992 goto fail;
993 }
994
995 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
996 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
997 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
998 goto trap_recursive;
999 if (try_lock &&
1000 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) &&
1001 !may_grant(gl, gh)) {
1002fail:
1003 gh->gh_error = GLR_TRYFAILED;
1004 gfs2_holder_wake(gh);
1005 return;
1006 }
1007 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1008 continue;
1009 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1010 insert_pt = &gh2->gh_list;
1011 }
1012 set_bit(GLF_QUEUED, &gl->gl_flags);
1013 trace_gfs2_glock_queue(gh, 1);
1014 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1015 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1016 if (likely(insert_pt == NULL)) {
1017 list_add_tail(&gh->gh_list, &gl->gl_holders);
1018 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1019 goto do_cancel;
1020 return;
1021 }
1022 list_add_tail(&gh->gh_list, insert_pt);
1023do_cancel:
1024 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
1025 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1026 spin_unlock(&gl->gl_spin);
1027 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1028 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1029 spin_lock(&gl->gl_spin);
1030 }
1031 return;
1032
1033trap_recursive:
1034 print_symbol(KERN_ERR "original: %s\n", gh2->gh_ip);
1035 printk(KERN_ERR "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1036 printk(KERN_ERR "lock type: %d req lock state : %d\n",
1037 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1038 print_symbol(KERN_ERR "new: %s\n", gh->gh_ip);
1039 printk(KERN_ERR "pid: %d\n", pid_nr(gh->gh_owner_pid));
1040 printk(KERN_ERR "lock type: %d req lock state : %d\n",
1041 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1042 __dump_glock(NULL, gl);
1043 BUG();
1044}
1045
1046/**
1047 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1048 * @gh: the holder structure
1049 *
1050 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1051 *
1052 * Returns: 0, GLR_TRYFAILED, or errno on failure
1053 */
1054
1055int gfs2_glock_nq(struct gfs2_holder *gh)
1056{
1057 struct gfs2_glock *gl = gh->gh_gl;
1058 struct gfs2_sbd *sdp = gl->gl_sbd;
1059 int error = 0;
1060
1061 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1062 return -EIO;
1063
1064 if (test_bit(GLF_LRU, &gl->gl_flags))
1065 gfs2_glock_remove_from_lru(gl);
1066
1067 spin_lock(&gl->gl_spin);
1068 add_to_queue(gh);
1069 if ((LM_FLAG_NOEXP & gh->gh_flags) &&
1070 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1071 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1072 run_queue(gl, 1);
1073 spin_unlock(&gl->gl_spin);
1074
1075 if (!(gh->gh_flags & GL_ASYNC))
1076 error = gfs2_glock_wait(gh);
1077
1078 return error;
1079}
1080
1081/**
1082 * gfs2_glock_poll - poll to see if an async request has been completed
1083 * @gh: the holder
1084 *
1085 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1086 */
1087
1088int gfs2_glock_poll(struct gfs2_holder *gh)
1089{
1090 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1091}
1092
1093/**
1094 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1095 * @gh: the glock holder
1096 *
1097 */
1098
1099void gfs2_glock_dq(struct gfs2_holder *gh)
1100{
1101 struct gfs2_glock *gl = gh->gh_gl;
1102 const struct gfs2_glock_operations *glops = gl->gl_ops;
1103 unsigned delay = 0;
1104 int fast_path = 0;
1105
1106 spin_lock(&gl->gl_spin);
1107 if (gh->gh_flags & GL_NOCACHE)
1108 handle_callback(gl, LM_ST_UNLOCKED, 0);
1109
1110 list_del_init(&gh->gh_list);
1111 if (find_first_holder(gl) == NULL) {
1112 if (glops->go_unlock) {
1113 GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1114 spin_unlock(&gl->gl_spin);
1115 glops->go_unlock(gh);
1116 spin_lock(&gl->gl_spin);
1117 clear_bit(GLF_LOCK, &gl->gl_flags);
1118 }
1119 if (list_empty(&gl->gl_holders) &&
1120 !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1121 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1122 fast_path = 1;
1123 }
1124 if (!test_bit(GLF_LFLUSH, &gl->gl_flags))
1125 __gfs2_glock_schedule_for_reclaim(gl);
1126 trace_gfs2_glock_queue(gh, 0);
1127 spin_unlock(&gl->gl_spin);
1128 if (likely(fast_path))
1129 return;
1130
1131 gfs2_glock_hold(gl);
1132 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1133 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1134 gl->gl_name.ln_type == LM_TYPE_INODE)
1135 delay = gl->gl_hold_time;
1136 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1137 gfs2_glock_put(gl);
1138}
1139
1140void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1141{
1142 struct gfs2_glock *gl = gh->gh_gl;
1143 gfs2_glock_dq(gh);
1144 wait_on_demote(gl);
1145}
1146
1147/**
1148 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1149 * @gh: the holder structure
1150 *
1151 */
1152
1153void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1154{
1155 gfs2_glock_dq(gh);
1156 gfs2_holder_uninit(gh);
1157}
1158
1159/**
1160 * gfs2_glock_nq_num - acquire a glock based on lock number
1161 * @sdp: the filesystem
1162 * @number: the lock number
1163 * @glops: the glock operations for the type of glock
1164 * @state: the state to acquire the glock in
1165 * @flags: modifier flags for the acquisition
1166 * @gh: the struct gfs2_holder
1167 *
1168 * Returns: errno
1169 */
1170
1171int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1172 const struct gfs2_glock_operations *glops,
1173 unsigned int state, int flags, struct gfs2_holder *gh)
1174{
1175 struct gfs2_glock *gl;
1176 int error;
1177
1178 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1179 if (!error) {
1180 error = gfs2_glock_nq_init(gl, state, flags, gh);
1181 gfs2_glock_put(gl);
1182 }
1183
1184 return error;
1185}
1186
1187/**
1188 * glock_compare - Compare two struct gfs2_glock structures for sorting
1189 * @arg_a: the first structure
1190 * @arg_b: the second structure
1191 *
1192 */
1193
1194static int glock_compare(const void *arg_a, const void *arg_b)
1195{
1196 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1197 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1198 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1199 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1200
1201 if (a->ln_number > b->ln_number)
1202 return 1;
1203 if (a->ln_number < b->ln_number)
1204 return -1;
1205 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1206 return 0;
1207}
1208
1209/**
1210 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1211 * @num_gh: the number of structures
1212 * @ghs: an array of struct gfs2_holder structures
1213 *
1214 * Returns: 0 on success (all glocks acquired),
1215 * errno on failure (no glocks acquired)
1216 */
1217
1218static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1219 struct gfs2_holder **p)
1220{
1221 unsigned int x;
1222 int error = 0;
1223
1224 for (x = 0; x < num_gh; x++)
1225 p[x] = &ghs[x];
1226
1227 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1228
1229 for (x = 0; x < num_gh; x++) {
1230 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1231
1232 error = gfs2_glock_nq(p[x]);
1233 if (error) {
1234 while (x--)
1235 gfs2_glock_dq(p[x]);
1236 break;
1237 }
1238 }
1239
1240 return error;
1241}
1242
1243/**
1244 * gfs2_glock_nq_m - acquire multiple glocks
1245 * @num_gh: the number of structures
1246 * @ghs: an array of struct gfs2_holder structures
1247 *
1248 *
1249 * Returns: 0 on success (all glocks acquired),
1250 * errno on failure (no glocks acquired)
1251 */
1252
1253int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1254{
1255 struct gfs2_holder *tmp[4];
1256 struct gfs2_holder **pph = tmp;
1257 int error = 0;
1258
1259 switch(num_gh) {
1260 case 0:
1261 return 0;
1262 case 1:
1263 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1264 return gfs2_glock_nq(ghs);
1265 default:
1266 if (num_gh <= 4)
1267 break;
1268 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1269 if (!pph)
1270 return -ENOMEM;
1271 }
1272
1273 error = nq_m_sync(num_gh, ghs, pph);
1274
1275 if (pph != tmp)
1276 kfree(pph);
1277
1278 return error;
1279}
1280
1281/**
1282 * gfs2_glock_dq_m - release multiple glocks
1283 * @num_gh: the number of structures
1284 * @ghs: an array of struct gfs2_holder structures
1285 *
1286 */
1287
1288void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1289{
1290 while (num_gh--)
1291 gfs2_glock_dq(&ghs[num_gh]);
1292}
1293
1294/**
1295 * gfs2_glock_dq_uninit_m - release multiple glocks
1296 * @num_gh: the number of structures
1297 * @ghs: an array of struct gfs2_holder structures
1298 *
1299 */
1300
1301void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1302{
1303 while (num_gh--)
1304 gfs2_glock_dq_uninit(&ghs[num_gh]);
1305}
1306
1307void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1308{
1309 unsigned long delay = 0;
1310 unsigned long holdtime;
1311 unsigned long now = jiffies;
1312
1313 gfs2_glock_hold(gl);
1314 holdtime = gl->gl_tchange + gl->gl_hold_time;
1315 if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1316 gl->gl_name.ln_type == LM_TYPE_INODE) {
1317 if (time_before(now, holdtime))
1318 delay = holdtime - now;
1319 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1320 delay = gl->gl_hold_time;
1321 }
1322
1323 spin_lock(&gl->gl_spin);
1324 handle_callback(gl, state, delay);
1325 spin_unlock(&gl->gl_spin);
1326 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1327 gfs2_glock_put(gl);
1328}
1329
1330/**
1331 * gfs2_should_freeze - Figure out if glock should be frozen
1332 * @gl: The glock in question
1333 *
1334 * Glocks are not frozen if (a) the result of the dlm operation is
1335 * an error, (b) the locking operation was an unlock operation or
1336 * (c) if there is a "noexp" flagged request anywhere in the queue
1337 *
1338 * Returns: 1 if freezing should occur, 0 otherwise
1339 */
1340
1341static int gfs2_should_freeze(const struct gfs2_glock *gl)
1342{
1343 const struct gfs2_holder *gh;
1344
1345 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1346 return 0;
1347 if (gl->gl_target == LM_ST_UNLOCKED)
1348 return 0;
1349
1350 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1351 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1352 continue;
1353 if (LM_FLAG_NOEXP & gh->gh_flags)
1354 return 0;
1355 }
1356
1357 return 1;
1358}
1359
1360/**
1361 * gfs2_glock_complete - Callback used by locking
1362 * @gl: Pointer to the glock
1363 * @ret: The return value from the dlm
1364 *
1365 * The gl_reply field is under the gl_spin lock so that it is ok
1366 * to use a bitfield shared with other glock state fields.
1367 */
1368
1369void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1370{
1371 struct lm_lockstruct *ls = &gl->gl_sbd->sd_lockstruct;
1372
1373 spin_lock(&gl->gl_spin);
1374 gl->gl_reply = ret;
1375
1376 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1377 if (gfs2_should_freeze(gl)) {
1378 set_bit(GLF_FROZEN, &gl->gl_flags);
1379 spin_unlock(&gl->gl_spin);
1380 return;
1381 }
1382 }
1383
1384 spin_unlock(&gl->gl_spin);
1385 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1386 smp_wmb();
1387 gfs2_glock_hold(gl);
1388 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1389 gfs2_glock_put(gl);
1390}
1391
1392
1393static int gfs2_shrink_glock_memory(struct shrinker *shrink,
1394 struct shrink_control *sc)
1395{
1396 struct gfs2_glock *gl;
1397 int may_demote;
1398 int nr_skipped = 0;
1399 int nr = sc->nr_to_scan;
1400 gfp_t gfp_mask = sc->gfp_mask;
1401 LIST_HEAD(skipped);
1402
1403 if (nr == 0)
1404 goto out;
1405
1406 if (!(gfp_mask & __GFP_FS))
1407 return -1;
1408
1409 spin_lock(&lru_lock);
1410 while(nr && !list_empty(&lru_list)) {
1411 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1412 list_del_init(&gl->gl_lru);
1413 clear_bit(GLF_LRU, &gl->gl_flags);
1414 atomic_dec(&lru_count);
1415
1416 /* Test for being demotable */
1417 if (!test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1418 gfs2_glock_hold(gl);
1419 spin_unlock(&lru_lock);
1420 spin_lock(&gl->gl_spin);
1421 may_demote = demote_ok(gl);
1422 if (may_demote) {
1423 handle_callback(gl, LM_ST_UNLOCKED, 0);
1424 nr--;
1425 }
1426 clear_bit(GLF_LOCK, &gl->gl_flags);
1427 smp_mb__after_clear_bit();
1428 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1429 gfs2_glock_put_nolock(gl);
1430 spin_unlock(&gl->gl_spin);
1431 spin_lock(&lru_lock);
1432 continue;
1433 }
1434 nr_skipped++;
1435 list_add(&gl->gl_lru, &skipped);
1436 set_bit(GLF_LRU, &gl->gl_flags);
1437 }
1438 list_splice(&skipped, &lru_list);
1439 atomic_add(nr_skipped, &lru_count);
1440 spin_unlock(&lru_lock);
1441out:
1442 return (atomic_read(&lru_count) / 100) * sysctl_vfs_cache_pressure;
1443}
1444
1445static struct shrinker glock_shrinker = {
1446 .shrink = gfs2_shrink_glock_memory,
1447 .seeks = DEFAULT_SEEKS,
1448};
1449
1450/**
1451 * examine_bucket - Call a function for glock in a hash bucket
1452 * @examiner: the function
1453 * @sdp: the filesystem
1454 * @bucket: the bucket
1455 *
1456 */
1457
1458static void examine_bucket(glock_examiner examiner, const struct gfs2_sbd *sdp,
1459 unsigned int hash)
1460{
1461 struct gfs2_glock *gl;
1462 struct hlist_bl_head *head = &gl_hash_table[hash];
1463 struct hlist_bl_node *pos;
1464
1465 rcu_read_lock();
1466 hlist_bl_for_each_entry_rcu(gl, pos, head, gl_list) {
1467 if ((gl->gl_sbd == sdp) && atomic_read(&gl->gl_ref))
1468 examiner(gl);
1469 }
1470 rcu_read_unlock();
1471 cond_resched();
1472}
1473
1474static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1475{
1476 unsigned x;
1477
1478 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1479 examine_bucket(examiner, sdp, x);
1480}
1481
1482
1483/**
1484 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1485 * @gl: The glock to thaw
1486 *
1487 * N.B. When we freeze a glock, we leave a ref to the glock outstanding,
1488 * so this has to result in the ref count being dropped by one.
1489 */
1490
1491static void thaw_glock(struct gfs2_glock *gl)
1492{
1493 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1494 return;
1495 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1496 gfs2_glock_hold(gl);
1497 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1498 gfs2_glock_put(gl);
1499}
1500
1501/**
1502 * clear_glock - look at a glock and see if we can free it from glock cache
1503 * @gl: the glock to look at
1504 *
1505 */
1506
1507static void clear_glock(struct gfs2_glock *gl)
1508{
1509 gfs2_glock_remove_from_lru(gl);
1510
1511 spin_lock(&gl->gl_spin);
1512 if (gl->gl_state != LM_ST_UNLOCKED)
1513 handle_callback(gl, LM_ST_UNLOCKED, 0);
1514 spin_unlock(&gl->gl_spin);
1515 gfs2_glock_hold(gl);
1516 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1517 gfs2_glock_put(gl);
1518}
1519
1520/**
1521 * gfs2_glock_thaw - Thaw any frozen glocks
1522 * @sdp: The super block
1523 *
1524 */
1525
1526void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1527{
1528 glock_hash_walk(thaw_glock, sdp);
1529}
1530
1531static int dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1532{
1533 int ret;
1534 spin_lock(&gl->gl_spin);
1535 ret = __dump_glock(seq, gl);
1536 spin_unlock(&gl->gl_spin);
1537 return ret;
1538}
1539
1540static void dump_glock_func(struct gfs2_glock *gl)
1541{
1542 dump_glock(NULL, gl);
1543}
1544
1545/**
1546 * gfs2_gl_hash_clear - Empty out the glock hash table
1547 * @sdp: the filesystem
1548 * @wait: wait until it's all gone
1549 *
1550 * Called when unmounting the filesystem.
1551 */
1552
1553void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1554{
1555 glock_hash_walk(clear_glock, sdp);
1556 flush_workqueue(glock_workqueue);
1557 wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
1558 glock_hash_walk(dump_glock_func, sdp);
1559}
1560
1561void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1562{
1563 struct gfs2_glock *gl = ip->i_gl;
1564 int ret;
1565
1566 ret = gfs2_truncatei_resume(ip);
1567 gfs2_assert_withdraw(gl->gl_sbd, ret == 0);
1568
1569 spin_lock(&gl->gl_spin);
1570 clear_bit(GLF_LOCK, &gl->gl_flags);
1571 run_queue(gl, 1);
1572 spin_unlock(&gl->gl_spin);
1573}
1574
1575static const char *state2str(unsigned state)
1576{
1577 switch(state) {
1578 case LM_ST_UNLOCKED:
1579 return "UN";
1580 case LM_ST_SHARED:
1581 return "SH";
1582 case LM_ST_DEFERRED:
1583 return "DF";
1584 case LM_ST_EXCLUSIVE:
1585 return "EX";
1586 }
1587 return "??";
1588}
1589
1590static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1591{
1592 char *p = buf;
1593 if (flags & LM_FLAG_TRY)
1594 *p++ = 't';
1595 if (flags & LM_FLAG_TRY_1CB)
1596 *p++ = 'T';
1597 if (flags & LM_FLAG_NOEXP)
1598 *p++ = 'e';
1599 if (flags & LM_FLAG_ANY)
1600 *p++ = 'A';
1601 if (flags & LM_FLAG_PRIORITY)
1602 *p++ = 'p';
1603 if (flags & GL_ASYNC)
1604 *p++ = 'a';
1605 if (flags & GL_EXACT)
1606 *p++ = 'E';
1607 if (flags & GL_NOCACHE)
1608 *p++ = 'c';
1609 if (test_bit(HIF_HOLDER, &iflags))
1610 *p++ = 'H';
1611 if (test_bit(HIF_WAIT, &iflags))
1612 *p++ = 'W';
1613 if (test_bit(HIF_FIRST, &iflags))
1614 *p++ = 'F';
1615 *p = 0;
1616 return buf;
1617}
1618
1619/**
1620 * dump_holder - print information about a glock holder
1621 * @seq: the seq_file struct
1622 * @gh: the glock holder
1623 *
1624 * Returns: 0 on success, -ENOBUFS when we run out of space
1625 */
1626
1627static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1628{
1629 struct task_struct *gh_owner = NULL;
1630 char flags_buf[32];
1631
1632 if (gh->gh_owner_pid)
1633 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1634 gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1635 state2str(gh->gh_state),
1636 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1637 gh->gh_error,
1638 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1639 gh_owner ? gh_owner->comm : "(ended)",
1640 (void *)gh->gh_ip);
1641 return 0;
1642}
1643
1644static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1645{
1646 const unsigned long *gflags = &gl->gl_flags;
1647 char *p = buf;
1648
1649 if (test_bit(GLF_LOCK, gflags))
1650 *p++ = 'l';
1651 if (test_bit(GLF_DEMOTE, gflags))
1652 *p++ = 'D';
1653 if (test_bit(GLF_PENDING_DEMOTE, gflags))
1654 *p++ = 'd';
1655 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1656 *p++ = 'p';
1657 if (test_bit(GLF_DIRTY, gflags))
1658 *p++ = 'y';
1659 if (test_bit(GLF_LFLUSH, gflags))
1660 *p++ = 'f';
1661 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1662 *p++ = 'i';
1663 if (test_bit(GLF_REPLY_PENDING, gflags))
1664 *p++ = 'r';
1665 if (test_bit(GLF_INITIAL, gflags))
1666 *p++ = 'I';
1667 if (test_bit(GLF_FROZEN, gflags))
1668 *p++ = 'F';
1669 if (test_bit(GLF_QUEUED, gflags))
1670 *p++ = 'q';
1671 if (test_bit(GLF_LRU, gflags))
1672 *p++ = 'L';
1673 if (gl->gl_object)
1674 *p++ = 'o';
1675 if (test_bit(GLF_BLOCKING, gflags))
1676 *p++ = 'b';
1677 *p = 0;
1678 return buf;
1679}
1680
1681/**
1682 * __dump_glock - print information about a glock
1683 * @seq: The seq_file struct
1684 * @gl: the glock
1685 *
1686 * The file format is as follows:
1687 * One line per object, capital letters are used to indicate objects
1688 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1689 * other objects are indented by a single space and follow the glock to
1690 * which they are related. Fields are indicated by lower case letters
1691 * followed by a colon and the field value, except for strings which are in
1692 * [] so that its possible to see if they are composed of spaces for
1693 * example. The field's are n = number (id of the object), f = flags,
1694 * t = type, s = state, r = refcount, e = error, p = pid.
1695 *
1696 * Returns: 0 on success, -ENOBUFS when we run out of space
1697 */
1698
1699static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1700{
1701 const struct gfs2_glock_operations *glops = gl->gl_ops;
1702 unsigned long long dtime;
1703 const struct gfs2_holder *gh;
1704 char gflags_buf[32];
1705 int error = 0;
1706
1707 dtime = jiffies - gl->gl_demote_time;
1708 dtime *= 1000000/HZ; /* demote time in uSec */
1709 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1710 dtime = 0;
1711 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",
1712 state2str(gl->gl_state),
1713 gl->gl_name.ln_type,
1714 (unsigned long long)gl->gl_name.ln_number,
1715 gflags2str(gflags_buf, gl),
1716 state2str(gl->gl_target),
1717 state2str(gl->gl_demote_state), dtime,
1718 atomic_read(&gl->gl_ail_count),
1719 atomic_read(&gl->gl_revokes),
1720 atomic_read(&gl->gl_ref), gl->gl_hold_time);
1721
1722 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1723 error = dump_holder(seq, gh);
1724 if (error)
1725 goto out;
1726 }
1727 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1728 error = glops->go_dump(seq, gl);
1729out:
1730 return error;
1731}
1732
1733static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
1734{
1735 struct gfs2_glock *gl = iter_ptr;
1736
1737 seq_printf(seq, "G: n:%u/%llx rtt:%lld/%lld rttb:%lld/%lld irt:%lld/%lld dcnt: %lld qcnt: %lld\n",
1738 gl->gl_name.ln_type,
1739 (unsigned long long)gl->gl_name.ln_number,
1740 (long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
1741 (long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
1742 (long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
1743 (long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
1744 (long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
1745 (long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
1746 (long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
1747 (long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
1748 return 0;
1749}
1750
1751static const char *gfs2_gltype[] = {
1752 "type",
1753 "reserved",
1754 "nondisk",
1755 "inode",
1756 "rgrp",
1757 "meta",
1758 "iopen",
1759 "flock",
1760 "plock",
1761 "quota",
1762 "journal",
1763};
1764
1765static const char *gfs2_stype[] = {
1766 [GFS2_LKS_SRTT] = "srtt",
1767 [GFS2_LKS_SRTTVAR] = "srttvar",
1768 [GFS2_LKS_SRTTB] = "srttb",
1769 [GFS2_LKS_SRTTVARB] = "srttvarb",
1770 [GFS2_LKS_SIRT] = "sirt",
1771 [GFS2_LKS_SIRTVAR] = "sirtvar",
1772 [GFS2_LKS_DCOUNT] = "dlm",
1773 [GFS2_LKS_QCOUNT] = "queue",
1774};
1775
1776#define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1777
1778static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
1779{
1780 struct gfs2_glock_iter *gi = seq->private;
1781 struct gfs2_sbd *sdp = gi->sdp;
1782 unsigned index = gi->hash >> 3;
1783 unsigned subindex = gi->hash & 0x07;
1784 s64 value;
1785 int i;
1786
1787 if (index == 0 && subindex != 0)
1788 return 0;
1789
1790 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
1791 (index == 0) ? "cpu": gfs2_stype[subindex]);
1792
1793 for_each_possible_cpu(i) {
1794 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
1795 if (index == 0) {
1796 value = i;
1797 } else {
1798 value = lkstats->lkstats[index - 1].stats[subindex];
1799 }
1800 seq_printf(seq, " %15lld", (long long)value);
1801 }
1802 seq_putc(seq, '\n');
1803 return 0;
1804}
1805
1806int __init gfs2_glock_init(void)
1807{
1808 unsigned i;
1809 for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
1810 INIT_HLIST_BL_HEAD(&gl_hash_table[i]);
1811 }
1812
1813 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1814 WQ_HIGHPRI | WQ_FREEZABLE, 0);
1815 if (IS_ERR(glock_workqueue))
1816 return PTR_ERR(glock_workqueue);
1817 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1818 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1819 0);
1820 if (IS_ERR(gfs2_delete_workqueue)) {
1821 destroy_workqueue(glock_workqueue);
1822 return PTR_ERR(gfs2_delete_workqueue);
1823 }
1824
1825 register_shrinker(&glock_shrinker);
1826
1827 return 0;
1828}
1829
1830void gfs2_glock_exit(void)
1831{
1832 unregister_shrinker(&glock_shrinker);
1833 destroy_workqueue(glock_workqueue);
1834 destroy_workqueue(gfs2_delete_workqueue);
1835}
1836
1837static inline struct gfs2_glock *glock_hash_chain(unsigned hash)
1838{
1839 return hlist_bl_entry(hlist_bl_first_rcu(&gl_hash_table[hash]),
1840 struct gfs2_glock, gl_list);
1841}
1842
1843static inline struct gfs2_glock *glock_hash_next(struct gfs2_glock *gl)
1844{
1845 return hlist_bl_entry(rcu_dereference(gl->gl_list.next),
1846 struct gfs2_glock, gl_list);
1847}
1848
1849static int gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1850{
1851 struct gfs2_glock *gl;
1852
1853 do {
1854 gl = gi->gl;
1855 if (gl) {
1856 gi->gl = glock_hash_next(gl);
1857 } else {
1858 gi->gl = glock_hash_chain(gi->hash);
1859 }
1860 while (gi->gl == NULL) {
1861 gi->hash++;
1862 if (gi->hash >= GFS2_GL_HASH_SIZE) {
1863 rcu_read_unlock();
1864 return 1;
1865 }
1866 gi->gl = glock_hash_chain(gi->hash);
1867 }
1868 /* Skip entries for other sb and dead entries */
1869 } while (gi->sdp != gi->gl->gl_sbd || atomic_read(&gi->gl->gl_ref) == 0);
1870
1871 return 0;
1872}
1873
1874static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1875{
1876 struct gfs2_glock_iter *gi = seq->private;
1877 loff_t n = *pos;
1878
1879 gi->hash = 0;
1880 rcu_read_lock();
1881
1882 do {
1883 if (gfs2_glock_iter_next(gi))
1884 return NULL;
1885 } while (n--);
1886
1887 return gi->gl;
1888}
1889
1890static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1891 loff_t *pos)
1892{
1893 struct gfs2_glock_iter *gi = seq->private;
1894
1895 (*pos)++;
1896
1897 if (gfs2_glock_iter_next(gi))
1898 return NULL;
1899
1900 return gi->gl;
1901}
1902
1903static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1904{
1905 struct gfs2_glock_iter *gi = seq->private;
1906
1907 if (gi->gl)
1908 rcu_read_unlock();
1909 gi->gl = NULL;
1910}
1911
1912static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1913{
1914 return dump_glock(seq, iter_ptr);
1915}
1916
1917static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
1918{
1919 struct gfs2_glock_iter *gi = seq->private;
1920
1921 gi->hash = *pos;
1922 if (*pos >= GFS2_NR_SBSTATS)
1923 return NULL;
1924 preempt_disable();
1925 return SEQ_START_TOKEN;
1926}
1927
1928static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
1929 loff_t *pos)
1930{
1931 struct gfs2_glock_iter *gi = seq->private;
1932 (*pos)++;
1933 gi->hash++;
1934 if (gi->hash >= GFS2_NR_SBSTATS) {
1935 preempt_enable();
1936 return NULL;
1937 }
1938 return SEQ_START_TOKEN;
1939}
1940
1941static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
1942{
1943 preempt_enable();
1944}
1945
1946static const struct seq_operations gfs2_glock_seq_ops = {
1947 .start = gfs2_glock_seq_start,
1948 .next = gfs2_glock_seq_next,
1949 .stop = gfs2_glock_seq_stop,
1950 .show = gfs2_glock_seq_show,
1951};
1952
1953static const struct seq_operations gfs2_glstats_seq_ops = {
1954 .start = gfs2_glock_seq_start,
1955 .next = gfs2_glock_seq_next,
1956 .stop = gfs2_glock_seq_stop,
1957 .show = gfs2_glstats_seq_show,
1958};
1959
1960static const struct seq_operations gfs2_sbstats_seq_ops = {
1961 .start = gfs2_sbstats_seq_start,
1962 .next = gfs2_sbstats_seq_next,
1963 .stop = gfs2_sbstats_seq_stop,
1964 .show = gfs2_sbstats_seq_show,
1965};
1966
1967static int gfs2_glocks_open(struct inode *inode, struct file *file)
1968{
1969 int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1970 sizeof(struct gfs2_glock_iter));
1971 if (ret == 0) {
1972 struct seq_file *seq = file->private_data;
1973 struct gfs2_glock_iter *gi = seq->private;
1974 gi->sdp = inode->i_private;
1975 }
1976 return ret;
1977}
1978
1979static int gfs2_glstats_open(struct inode *inode, struct file *file)
1980{
1981 int ret = seq_open_private(file, &gfs2_glstats_seq_ops,
1982 sizeof(struct gfs2_glock_iter));
1983 if (ret == 0) {
1984 struct seq_file *seq = file->private_data;
1985 struct gfs2_glock_iter *gi = seq->private;
1986 gi->sdp = inode->i_private;
1987 }
1988 return ret;
1989}
1990
1991static int gfs2_sbstats_open(struct inode *inode, struct file *file)
1992{
1993 int ret = seq_open_private(file, &gfs2_sbstats_seq_ops,
1994 sizeof(struct gfs2_glock_iter));
1995 if (ret == 0) {
1996 struct seq_file *seq = file->private_data;
1997 struct gfs2_glock_iter *gi = seq->private;
1998 gi->sdp = inode->i_private;
1999 }
2000 return ret;
2001}
2002
2003static const struct file_operations gfs2_glocks_fops = {
2004 .owner = THIS_MODULE,
2005 .open = gfs2_glocks_open,
2006 .read = seq_read,
2007 .llseek = seq_lseek,
2008 .release = seq_release_private,
2009};
2010
2011static const struct file_operations gfs2_glstats_fops = {
2012 .owner = THIS_MODULE,
2013 .open = gfs2_glstats_open,
2014 .read = seq_read,
2015 .llseek = seq_lseek,
2016 .release = seq_release_private,
2017};
2018
2019static const struct file_operations gfs2_sbstats_fops = {
2020 .owner = THIS_MODULE,
2021 .open = gfs2_sbstats_open,
2022 .read = seq_read,
2023 .llseek = seq_lseek,
2024 .release = seq_release_private,
2025};
2026
2027int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2028{
2029 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2030 if (!sdp->debugfs_dir)
2031 return -ENOMEM;
2032 sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
2033 S_IFREG | S_IRUGO,
2034 sdp->debugfs_dir, sdp,
2035 &gfs2_glocks_fops);
2036 if (!sdp->debugfs_dentry_glocks)
2037 goto fail;
2038
2039 sdp->debugfs_dentry_glstats = debugfs_create_file("glstats",
2040 S_IFREG | S_IRUGO,
2041 sdp->debugfs_dir, sdp,
2042 &gfs2_glstats_fops);
2043 if (!sdp->debugfs_dentry_glstats)
2044 goto fail;
2045
2046 sdp->debugfs_dentry_sbstats = debugfs_create_file("sbstats",
2047 S_IFREG | S_IRUGO,
2048 sdp->debugfs_dir, sdp,
2049 &gfs2_sbstats_fops);
2050 if (!sdp->debugfs_dentry_sbstats)
2051 goto fail;
2052
2053 return 0;
2054fail:
2055 gfs2_delete_debugfs_file(sdp);
2056 return -ENOMEM;
2057}
2058
2059void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2060{
2061 if (sdp->debugfs_dir) {
2062 if (sdp->debugfs_dentry_glocks) {
2063 debugfs_remove(sdp->debugfs_dentry_glocks);
2064 sdp->debugfs_dentry_glocks = NULL;
2065 }
2066 if (sdp->debugfs_dentry_glstats) {
2067 debugfs_remove(sdp->debugfs_dentry_glstats);
2068 sdp->debugfs_dentry_glstats = NULL;
2069 }
2070 if (sdp->debugfs_dentry_sbstats) {
2071 debugfs_remove(sdp->debugfs_dentry_sbstats);
2072 sdp->debugfs_dentry_sbstats = NULL;
2073 }
2074 debugfs_remove(sdp->debugfs_dir);
2075 sdp->debugfs_dir = NULL;
2076 }
2077}
2078
2079int gfs2_register_debugfs(void)
2080{
2081 gfs2_root = debugfs_create_dir("gfs2", NULL);
2082 return gfs2_root ? 0 : -ENOMEM;
2083}
2084
2085void gfs2_unregister_debugfs(void)
2086{
2087 debugfs_remove(gfs2_root);
2088 gfs2_root = NULL;
2089}
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#include <linux/pid_namespace.h>
37#include <linux/file.h>
38#include <linux/random.h>
39
40#include "gfs2.h"
41#include "incore.h"
42#include "glock.h"
43#include "glops.h"
44#include "inode.h"
45#include "lops.h"
46#include "meta_io.h"
47#include "quota.h"
48#include "super.h"
49#include "util.h"
50#include "bmap.h"
51#define CREATE_TRACE_POINTS
52#include "trace_gfs2.h"
53
54struct gfs2_glock_iter {
55 struct gfs2_sbd *sdp; /* incore superblock */
56 struct rhashtable_iter hti; /* rhashtable iterator */
57 struct gfs2_glock *gl; /* current glock struct */
58 loff_t last_pos; /* last position */
59};
60
61typedef void (*glock_examiner) (struct gfs2_glock * gl);
62
63static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
64static void request_demote(struct gfs2_glock *gl, unsigned int state,
65 unsigned long delay, bool remote);
66
67static struct dentry *gfs2_root;
68static LIST_HEAD(lru_list);
69static atomic_t lru_count = ATOMIC_INIT(0);
70static DEFINE_SPINLOCK(lru_lock);
71
72#define GFS2_GL_HASH_SHIFT 15
73#define GFS2_GL_HASH_SIZE BIT(GFS2_GL_HASH_SHIFT)
74
75static const struct rhashtable_params ht_parms = {
76 .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4,
77 .key_len = offsetofend(struct lm_lockname, ln_type),
78 .key_offset = offsetof(struct gfs2_glock, gl_name),
79 .head_offset = offsetof(struct gfs2_glock, gl_node),
80};
81
82static struct rhashtable gl_hash_table;
83
84#define GLOCK_WAIT_TABLE_BITS 12
85#define GLOCK_WAIT_TABLE_SIZE (1 << GLOCK_WAIT_TABLE_BITS)
86static wait_queue_head_t glock_wait_table[GLOCK_WAIT_TABLE_SIZE] __cacheline_aligned;
87
88struct wait_glock_queue {
89 struct lm_lockname *name;
90 wait_queue_entry_t wait;
91};
92
93static int glock_wake_function(wait_queue_entry_t *wait, unsigned int mode,
94 int sync, void *key)
95{
96 struct wait_glock_queue *wait_glock =
97 container_of(wait, struct wait_glock_queue, wait);
98 struct lm_lockname *wait_name = wait_glock->name;
99 struct lm_lockname *wake_name = key;
100
101 if (wake_name->ln_sbd != wait_name->ln_sbd ||
102 wake_name->ln_number != wait_name->ln_number ||
103 wake_name->ln_type != wait_name->ln_type)
104 return 0;
105 return autoremove_wake_function(wait, mode, sync, key);
106}
107
108static wait_queue_head_t *glock_waitqueue(struct lm_lockname *name)
109{
110 u32 hash = jhash2((u32 *)name, ht_parms.key_len / 4, 0);
111
112 return glock_wait_table + hash_32(hash, GLOCK_WAIT_TABLE_BITS);
113}
114
115/**
116 * wake_up_glock - Wake up waiters on a glock
117 * @gl: the glock
118 */
119static void wake_up_glock(struct gfs2_glock *gl)
120{
121 wait_queue_head_t *wq = glock_waitqueue(&gl->gl_name);
122
123 if (waitqueue_active(wq))
124 __wake_up(wq, TASK_NORMAL, 1, &gl->gl_name);
125}
126
127static void gfs2_glock_dealloc(struct rcu_head *rcu)
128{
129 struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
130
131 kfree(gl->gl_lksb.sb_lvbptr);
132 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
133 struct gfs2_glock_aspace *gla =
134 container_of(gl, struct gfs2_glock_aspace, glock);
135 kmem_cache_free(gfs2_glock_aspace_cachep, gla);
136 } else
137 kmem_cache_free(gfs2_glock_cachep, gl);
138}
139
140/**
141 * glock_blocked_by_withdraw - determine if we can still use a glock
142 * @gl: the glock
143 *
144 * We need to allow some glocks to be enqueued, dequeued, promoted, and demoted
145 * when we're withdrawn. For example, to maintain metadata integrity, we should
146 * disallow the use of inode and rgrp glocks when withdrawn. Other glocks like
147 * the iopen or freeze glock may be safely used because none of their
148 * metadata goes through the journal. So in general, we should disallow all
149 * glocks that are journaled, and allow all the others. One exception is:
150 * we need to allow our active journal to be promoted and demoted so others
151 * may recover it and we can reacquire it when they're done.
152 */
153static bool glock_blocked_by_withdraw(struct gfs2_glock *gl)
154{
155 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
156
157 if (!gfs2_withdrawing_or_withdrawn(sdp))
158 return false;
159 if (gl->gl_ops->go_flags & GLOF_NONDISK)
160 return false;
161 if (!sdp->sd_jdesc ||
162 gl->gl_name.ln_number == sdp->sd_jdesc->jd_no_addr)
163 return false;
164 return true;
165}
166
167static void __gfs2_glock_free(struct gfs2_glock *gl)
168{
169 rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
170 smp_mb();
171 wake_up_glock(gl);
172 call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
173}
174
175void gfs2_glock_free(struct gfs2_glock *gl) {
176 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
177
178 __gfs2_glock_free(gl);
179 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
180 wake_up(&sdp->sd_kill_wait);
181}
182
183void gfs2_glock_free_later(struct gfs2_glock *gl) {
184 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
185
186 spin_lock(&lru_lock);
187 list_add(&gl->gl_lru, &sdp->sd_dead_glocks);
188 spin_unlock(&lru_lock);
189 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
190 wake_up(&sdp->sd_kill_wait);
191}
192
193static void gfs2_free_dead_glocks(struct gfs2_sbd *sdp)
194{
195 struct list_head *list = &sdp->sd_dead_glocks;
196
197 while(!list_empty(list)) {
198 struct gfs2_glock *gl;
199
200 gl = list_first_entry(list, struct gfs2_glock, gl_lru);
201 list_del_init(&gl->gl_lru);
202 __gfs2_glock_free(gl);
203 }
204}
205
206/**
207 * gfs2_glock_hold() - increment reference count on glock
208 * @gl: The glock to hold
209 *
210 */
211
212struct gfs2_glock *gfs2_glock_hold(struct gfs2_glock *gl)
213{
214 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
215 lockref_get(&gl->gl_lockref);
216 return gl;
217}
218
219static void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
220{
221 spin_lock(&lru_lock);
222 list_move_tail(&gl->gl_lru, &lru_list);
223
224 if (!test_bit(GLF_LRU, &gl->gl_flags)) {
225 set_bit(GLF_LRU, &gl->gl_flags);
226 atomic_inc(&lru_count);
227 }
228
229 spin_unlock(&lru_lock);
230}
231
232static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
233{
234 spin_lock(&lru_lock);
235 if (test_bit(GLF_LRU, &gl->gl_flags)) {
236 list_del_init(&gl->gl_lru);
237 atomic_dec(&lru_count);
238 clear_bit(GLF_LRU, &gl->gl_flags);
239 }
240 spin_unlock(&lru_lock);
241}
242
243/*
244 * Enqueue the glock on the work queue. Passes one glock reference on to the
245 * work queue.
246 */
247static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
248 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
249
250 if (!queue_delayed_work(sdp->sd_glock_wq, &gl->gl_work, delay)) {
251 /*
252 * We are holding the lockref spinlock, and the work was still
253 * queued above. The queued work (glock_work_func) takes that
254 * spinlock before dropping its glock reference(s), so it
255 * cannot have dropped them in the meantime.
256 */
257 GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2);
258 gl->gl_lockref.count--;
259 }
260}
261
262static void __gfs2_glock_put(struct gfs2_glock *gl)
263{
264 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
265 struct address_space *mapping = gfs2_glock2aspace(gl);
266
267 lockref_mark_dead(&gl->gl_lockref);
268 spin_unlock(&gl->gl_lockref.lock);
269 gfs2_glock_remove_from_lru(gl);
270 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
271 if (mapping) {
272 truncate_inode_pages_final(mapping);
273 if (!gfs2_withdrawing_or_withdrawn(sdp))
274 GLOCK_BUG_ON(gl, !mapping_empty(mapping));
275 }
276 trace_gfs2_glock_put(gl);
277 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
278}
279
280static bool __gfs2_glock_put_or_lock(struct gfs2_glock *gl)
281{
282 if (lockref_put_or_lock(&gl->gl_lockref))
283 return true;
284 GLOCK_BUG_ON(gl, gl->gl_lockref.count != 1);
285 if (gl->gl_state != LM_ST_UNLOCKED) {
286 gl->gl_lockref.count--;
287 gfs2_glock_add_to_lru(gl);
288 spin_unlock(&gl->gl_lockref.lock);
289 return true;
290 }
291 return false;
292}
293
294/**
295 * gfs2_glock_put() - Decrement reference count on glock
296 * @gl: The glock to put
297 *
298 */
299
300void gfs2_glock_put(struct gfs2_glock *gl)
301{
302 if (__gfs2_glock_put_or_lock(gl))
303 return;
304
305 __gfs2_glock_put(gl);
306}
307
308/*
309 * gfs2_glock_put_async - Decrement reference count without sleeping
310 * @gl: The glock to put
311 *
312 * Decrement the reference count on glock immediately unless it is the last
313 * reference. Defer putting the last reference to work queue context.
314 */
315void gfs2_glock_put_async(struct gfs2_glock *gl)
316{
317 if (__gfs2_glock_put_or_lock(gl))
318 return;
319
320 gfs2_glock_queue_work(gl, 0);
321 spin_unlock(&gl->gl_lockref.lock);
322}
323
324/**
325 * may_grant - check if it's ok to grant a new lock
326 * @gl: The glock
327 * @current_gh: One of the current holders of @gl
328 * @gh: The lock request which we wish to grant
329 *
330 * With our current compatibility rules, if a glock has one or more active
331 * holders (HIF_HOLDER flag set), any of those holders can be passed in as
332 * @current_gh; they are all the same as far as compatibility with the new @gh
333 * goes.
334 *
335 * Returns true if it's ok to grant the lock.
336 */
337
338static inline bool may_grant(struct gfs2_glock *gl,
339 struct gfs2_holder *current_gh,
340 struct gfs2_holder *gh)
341{
342 if (current_gh) {
343 GLOCK_BUG_ON(gl, !test_bit(HIF_HOLDER, ¤t_gh->gh_iflags));
344
345 switch(current_gh->gh_state) {
346 case LM_ST_EXCLUSIVE:
347 /*
348 * Here we make a special exception to grant holders
349 * who agree to share the EX lock with other holders
350 * who also have the bit set. If the original holder
351 * has the LM_FLAG_NODE_SCOPE bit set, we grant more
352 * holders with the bit set.
353 */
354 return gh->gh_state == LM_ST_EXCLUSIVE &&
355 (current_gh->gh_flags & LM_FLAG_NODE_SCOPE) &&
356 (gh->gh_flags & LM_FLAG_NODE_SCOPE);
357
358 case LM_ST_SHARED:
359 case LM_ST_DEFERRED:
360 return gh->gh_state == current_gh->gh_state;
361
362 default:
363 return false;
364 }
365 }
366
367 if (gl->gl_state == gh->gh_state)
368 return true;
369 if (gh->gh_flags & GL_EXACT)
370 return false;
371 if (gl->gl_state == LM_ST_EXCLUSIVE) {
372 return gh->gh_state == LM_ST_SHARED ||
373 gh->gh_state == LM_ST_DEFERRED;
374 }
375 if (gh->gh_flags & LM_FLAG_ANY)
376 return gl->gl_state != LM_ST_UNLOCKED;
377 return false;
378}
379
380static void gfs2_holder_wake(struct gfs2_holder *gh)
381{
382 clear_bit(HIF_WAIT, &gh->gh_iflags);
383 smp_mb__after_atomic();
384 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
385 if (gh->gh_flags & GL_ASYNC) {
386 struct gfs2_sbd *sdp = gh->gh_gl->gl_name.ln_sbd;
387
388 wake_up(&sdp->sd_async_glock_wait);
389 }
390}
391
392/**
393 * do_error - Something unexpected has happened during a lock request
394 * @gl: The glock
395 * @ret: The status from the DLM
396 */
397
398static void do_error(struct gfs2_glock *gl, const int ret)
399{
400 struct gfs2_holder *gh, *tmp;
401
402 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
403 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
404 continue;
405 if (ret & LM_OUT_ERROR)
406 gh->gh_error = -EIO;
407 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
408 gh->gh_error = GLR_TRYFAILED;
409 else
410 continue;
411 list_del_init(&gh->gh_list);
412 trace_gfs2_glock_queue(gh, 0);
413 gfs2_holder_wake(gh);
414 }
415}
416
417/**
418 * find_first_holder - find the first "holder" gh
419 * @gl: the glock
420 */
421
422static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
423{
424 struct gfs2_holder *gh;
425
426 if (!list_empty(&gl->gl_holders)) {
427 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder,
428 gh_list);
429 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
430 return gh;
431 }
432 return NULL;
433}
434
435/*
436 * gfs2_instantiate - Call the glops instantiate function
437 * @gh: The glock holder
438 *
439 * Returns: 0 if instantiate was successful, or error.
440 */
441int gfs2_instantiate(struct gfs2_holder *gh)
442{
443 struct gfs2_glock *gl = gh->gh_gl;
444 const struct gfs2_glock_operations *glops = gl->gl_ops;
445 int ret;
446
447again:
448 if (!test_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags))
449 goto done;
450
451 /*
452 * Since we unlock the lockref lock, we set a flag to indicate
453 * instantiate is in progress.
454 */
455 if (test_and_set_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags)) {
456 wait_on_bit(&gl->gl_flags, GLF_INSTANTIATE_IN_PROG,
457 TASK_UNINTERRUPTIBLE);
458 /*
459 * Here we just waited for a different instantiate to finish.
460 * But that may not have been successful, as when a process
461 * locks an inode glock _before_ it has an actual inode to
462 * instantiate into. So we check again. This process might
463 * have an inode to instantiate, so might be successful.
464 */
465 goto again;
466 }
467
468 ret = glops->go_instantiate(gl);
469 if (!ret)
470 clear_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags);
471 clear_and_wake_up_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags);
472 if (ret)
473 return ret;
474
475done:
476 if (glops->go_held)
477 return glops->go_held(gh);
478 return 0;
479}
480
481/**
482 * do_promote - promote as many requests as possible on the current queue
483 * @gl: The glock
484 *
485 * Returns true on success (i.e., progress was made or there are no waiters).
486 */
487
488static bool do_promote(struct gfs2_glock *gl)
489{
490 struct gfs2_holder *gh, *current_gh;
491
492 current_gh = find_first_holder(gl);
493 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
494 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
495 continue;
496 if (!may_grant(gl, current_gh, gh)) {
497 /*
498 * If we get here, it means we may not grant this
499 * holder for some reason. If this holder is at the
500 * head of the list, it means we have a blocked holder
501 * at the head, so return false.
502 */
503 if (list_is_first(&gh->gh_list, &gl->gl_holders))
504 return false;
505 do_error(gl, 0);
506 break;
507 }
508 set_bit(HIF_HOLDER, &gh->gh_iflags);
509 trace_gfs2_promote(gh);
510 gfs2_holder_wake(gh);
511 if (!current_gh)
512 current_gh = gh;
513 }
514 return true;
515}
516
517/**
518 * find_first_waiter - find the first gh that's waiting for the glock
519 * @gl: the glock
520 */
521
522static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
523{
524 struct gfs2_holder *gh;
525
526 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
527 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
528 return gh;
529 }
530 return NULL;
531}
532
533/**
534 * find_last_waiter - find the last gh that's waiting for the glock
535 * @gl: the glock
536 *
537 * This also is a fast way of finding out if there are any waiters.
538 */
539
540static inline struct gfs2_holder *find_last_waiter(const struct gfs2_glock *gl)
541{
542 struct gfs2_holder *gh;
543
544 if (list_empty(&gl->gl_holders))
545 return NULL;
546 gh = list_last_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
547 return test_bit(HIF_HOLDER, &gh->gh_iflags) ? NULL : gh;
548}
549
550/**
551 * state_change - record that the glock is now in a different state
552 * @gl: the glock
553 * @new_state: the new state
554 */
555
556static void state_change(struct gfs2_glock *gl, unsigned int new_state)
557{
558 if (new_state != gl->gl_target)
559 /* shorten our minimum hold time */
560 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
561 GL_GLOCK_MIN_HOLD);
562 gl->gl_state = new_state;
563 gl->gl_tchange = jiffies;
564}
565
566static void gfs2_set_demote(int nr, struct gfs2_glock *gl)
567{
568 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
569
570 set_bit(nr, &gl->gl_flags);
571 smp_mb();
572 wake_up(&sdp->sd_async_glock_wait);
573}
574
575static void gfs2_demote_wake(struct gfs2_glock *gl)
576{
577 gl->gl_demote_state = LM_ST_EXCLUSIVE;
578 clear_bit(GLF_DEMOTE, &gl->gl_flags);
579 smp_mb__after_atomic();
580 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
581}
582
583/**
584 * finish_xmote - The DLM has replied to one of our lock requests
585 * @gl: The glock
586 * @ret: The status from the DLM
587 *
588 */
589
590static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
591{
592 const struct gfs2_glock_operations *glops = gl->gl_ops;
593 struct gfs2_holder *gh;
594 unsigned state = ret & LM_OUT_ST_MASK;
595
596 trace_gfs2_glock_state_change(gl, state);
597 state_change(gl, state);
598 gh = find_first_waiter(gl);
599
600 /* Demote to UN request arrived during demote to SH or DF */
601 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
602 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
603 gl->gl_target = LM_ST_UNLOCKED;
604
605 /* Check for state != intended state */
606 if (unlikely(state != gl->gl_target)) {
607 if (gh && (ret & LM_OUT_CANCELED))
608 gfs2_holder_wake(gh);
609 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
610 /* move to back of queue and try next entry */
611 if (ret & LM_OUT_CANCELED) {
612 list_move_tail(&gh->gh_list, &gl->gl_holders);
613 gh = find_first_waiter(gl);
614 gl->gl_target = gh->gh_state;
615 if (do_promote(gl))
616 goto out;
617 goto retry;
618 }
619 /* Some error or failed "try lock" - report it */
620 if ((ret & LM_OUT_ERROR) ||
621 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
622 gl->gl_target = gl->gl_state;
623 do_error(gl, ret);
624 goto out;
625 }
626 }
627 switch(state) {
628 /* Unlocked due to conversion deadlock, try again */
629 case LM_ST_UNLOCKED:
630retry:
631 do_xmote(gl, gh, gl->gl_target);
632 break;
633 /* Conversion fails, unlock and try again */
634 case LM_ST_SHARED:
635 case LM_ST_DEFERRED:
636 do_xmote(gl, gh, LM_ST_UNLOCKED);
637 break;
638 default: /* Everything else */
639 fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n",
640 gl->gl_target, state);
641 GLOCK_BUG_ON(gl, 1);
642 }
643 return;
644 }
645
646 /* Fast path - we got what we asked for */
647 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
648 gfs2_demote_wake(gl);
649 if (state != LM_ST_UNLOCKED) {
650 if (glops->go_xmote_bh) {
651 int rv;
652
653 spin_unlock(&gl->gl_lockref.lock);
654 rv = glops->go_xmote_bh(gl);
655 spin_lock(&gl->gl_lockref.lock);
656 if (rv) {
657 do_error(gl, rv);
658 goto out;
659 }
660 }
661 do_promote(gl);
662 }
663out:
664 clear_bit(GLF_LOCK, &gl->gl_flags);
665}
666
667static bool is_system_glock(struct gfs2_glock *gl)
668{
669 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
670 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
671
672 if (gl == m_ip->i_gl)
673 return true;
674 return false;
675}
676
677/**
678 * do_xmote - Calls the DLM to change the state of a lock
679 * @gl: The lock state
680 * @gh: The holder (only for promotes)
681 * @target: The target lock state
682 *
683 */
684
685static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh,
686 unsigned int target)
687__releases(&gl->gl_lockref.lock)
688__acquires(&gl->gl_lockref.lock)
689{
690 const struct gfs2_glock_operations *glops = gl->gl_ops;
691 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
692 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
693 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
694 int ret;
695
696 if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) &&
697 gh && !(gh->gh_flags & LM_FLAG_NOEXP))
698 goto skip_inval;
699
700 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP);
701 GLOCK_BUG_ON(gl, gl->gl_state == target);
702 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
703 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
704 glops->go_inval) {
705 /*
706 * If another process is already doing the invalidate, let that
707 * finish first. The glock state machine will get back to this
708 * holder again later.
709 */
710 if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
711 &gl->gl_flags))
712 return;
713 do_error(gl, 0); /* Fail queued try locks */
714 }
715 gl->gl_req = target;
716 set_bit(GLF_BLOCKING, &gl->gl_flags);
717 if ((gl->gl_req == LM_ST_UNLOCKED) ||
718 (gl->gl_state == LM_ST_EXCLUSIVE) ||
719 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
720 clear_bit(GLF_BLOCKING, &gl->gl_flags);
721 if (!glops->go_inval && !glops->go_sync)
722 goto skip_inval;
723
724 spin_unlock(&gl->gl_lockref.lock);
725 if (glops->go_sync) {
726 ret = glops->go_sync(gl);
727 /* If we had a problem syncing (due to io errors or whatever,
728 * we should not invalidate the metadata or tell dlm to
729 * release the glock to other nodes.
730 */
731 if (ret) {
732 if (cmpxchg(&sdp->sd_log_error, 0, ret)) {
733 fs_err(sdp, "Error %d syncing glock \n", ret);
734 gfs2_dump_glock(NULL, gl, true);
735 }
736 spin_lock(&gl->gl_lockref.lock);
737 goto skip_inval;
738 }
739 }
740 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) {
741 /*
742 * The call to go_sync should have cleared out the ail list.
743 * If there are still items, we have a problem. We ought to
744 * withdraw, but we can't because the withdraw code also uses
745 * glocks. Warn about the error, dump the glock, then fall
746 * through and wait for logd to do the withdraw for us.
747 */
748 if ((atomic_read(&gl->gl_ail_count) != 0) &&
749 (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) {
750 gfs2_glock_assert_warn(gl,
751 !atomic_read(&gl->gl_ail_count));
752 gfs2_dump_glock(NULL, gl, true);
753 }
754 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
755 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
756 }
757 spin_lock(&gl->gl_lockref.lock);
758
759skip_inval:
760 gl->gl_lockref.count++;
761 /*
762 * Check for an error encountered since we called go_sync and go_inval.
763 * If so, we can't withdraw from the glock code because the withdraw
764 * code itself uses glocks (see function signal_our_withdraw) to
765 * change the mount to read-only. Most importantly, we must not call
766 * dlm to unlock the glock until the journal is in a known good state
767 * (after journal replay) otherwise other nodes may use the object
768 * (rgrp or dinode) and then later, journal replay will corrupt the
769 * file system. The best we can do here is wait for the logd daemon
770 * to see sd_log_error and withdraw, and in the meantime, requeue the
771 * work for later.
772 *
773 * We make a special exception for some system glocks, such as the
774 * system statfs inode glock, which needs to be granted before the
775 * gfs2_quotad daemon can exit, and that exit needs to finish before
776 * we can unmount the withdrawn file system.
777 *
778 * However, if we're just unlocking the lock (say, for unmount, when
779 * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
780 * then it's okay to tell dlm to unlock it.
781 */
782 if (unlikely(sdp->sd_log_error) && !gfs2_withdrawing_or_withdrawn(sdp))
783 gfs2_withdraw_delayed(sdp);
784 if (glock_blocked_by_withdraw(gl) &&
785 (target != LM_ST_UNLOCKED ||
786 test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags))) {
787 if (!is_system_glock(gl)) {
788 request_demote(gl, LM_ST_UNLOCKED, 0, false);
789 /*
790 * Ordinarily, we would call dlm and its callback would call
791 * finish_xmote, which would call state_change() to the new state.
792 * Since we withdrew, we won't call dlm, so call state_change
793 * manually, but to the UNLOCKED state we desire.
794 */
795 state_change(gl, LM_ST_UNLOCKED);
796 /*
797 * We skip telling dlm to do the locking, so we won't get a
798 * reply that would otherwise clear GLF_LOCK. So we clear it here.
799 */
800 clear_bit(GLF_LOCK, &gl->gl_flags);
801 clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
802 gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD);
803 return;
804 } else {
805 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
806 }
807 }
808
809 if (ls->ls_ops->lm_lock) {
810 spin_unlock(&gl->gl_lockref.lock);
811 ret = ls->ls_ops->lm_lock(gl, target, lck_flags);
812 spin_lock(&gl->gl_lockref.lock);
813
814 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
815 target == LM_ST_UNLOCKED &&
816 test_bit(DFL_UNMOUNT, &ls->ls_recover_flags)) {
817 /*
818 * The lockspace has been released and the lock has
819 * been unlocked implicitly.
820 */
821 } else if (ret) {
822 fs_err(sdp, "lm_lock ret %d\n", ret);
823 target = gl->gl_state | LM_OUT_ERROR;
824 } else {
825 /* The operation will be completed asynchronously. */
826 return;
827 }
828 }
829
830 /* Complete the operation now. */
831 finish_xmote(gl, target);
832 gfs2_glock_queue_work(gl, 0);
833}
834
835/**
836 * run_queue - do all outstanding tasks related to a glock
837 * @gl: The glock in question
838 * @nonblock: True if we must not block in run_queue
839 *
840 */
841
842static void run_queue(struct gfs2_glock *gl, const int nonblock)
843__releases(&gl->gl_lockref.lock)
844__acquires(&gl->gl_lockref.lock)
845{
846 struct gfs2_holder *gh = NULL;
847
848 if (test_bit(GLF_LOCK, &gl->gl_flags))
849 return;
850 set_bit(GLF_LOCK, &gl->gl_flags);
851
852 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
853
854 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
855 gl->gl_demote_state != gl->gl_state) {
856 if (find_first_holder(gl))
857 goto out_unlock;
858 if (nonblock)
859 goto out_sched;
860 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
861 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
862 gl->gl_target = gl->gl_demote_state;
863 } else {
864 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
865 gfs2_demote_wake(gl);
866 if (do_promote(gl))
867 goto out_unlock;
868 gh = find_first_waiter(gl);
869 gl->gl_target = gh->gh_state;
870 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
871 do_error(gl, 0); /* Fail queued try locks */
872 }
873 do_xmote(gl, gh, gl->gl_target);
874 return;
875
876out_sched:
877 clear_bit(GLF_LOCK, &gl->gl_flags);
878 smp_mb__after_atomic();
879 gl->gl_lockref.count++;
880 gfs2_glock_queue_work(gl, 0);
881 return;
882
883out_unlock:
884 clear_bit(GLF_LOCK, &gl->gl_flags);
885 smp_mb__after_atomic();
886}
887
888/**
889 * glock_set_object - set the gl_object field of a glock
890 * @gl: the glock
891 * @object: the object
892 */
893void glock_set_object(struct gfs2_glock *gl, void *object)
894{
895 void *prev_object;
896
897 spin_lock(&gl->gl_lockref.lock);
898 prev_object = gl->gl_object;
899 gl->gl_object = object;
900 spin_unlock(&gl->gl_lockref.lock);
901 if (gfs2_assert_warn(gl->gl_name.ln_sbd, prev_object == NULL)) {
902 pr_warn("glock=%u/%llx\n",
903 gl->gl_name.ln_type,
904 (unsigned long long)gl->gl_name.ln_number);
905 gfs2_dump_glock(NULL, gl, true);
906 }
907}
908
909/**
910 * glock_clear_object - clear the gl_object field of a glock
911 * @gl: the glock
912 * @object: object the glock currently points at
913 */
914void glock_clear_object(struct gfs2_glock *gl, void *object)
915{
916 void *prev_object;
917
918 spin_lock(&gl->gl_lockref.lock);
919 prev_object = gl->gl_object;
920 gl->gl_object = NULL;
921 spin_unlock(&gl->gl_lockref.lock);
922 if (gfs2_assert_warn(gl->gl_name.ln_sbd, prev_object == object)) {
923 pr_warn("glock=%u/%llx\n",
924 gl->gl_name.ln_type,
925 (unsigned long long)gl->gl_name.ln_number);
926 gfs2_dump_glock(NULL, gl, true);
927 }
928}
929
930void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation)
931{
932 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
933
934 if (ri->ri_magic == 0)
935 ri->ri_magic = cpu_to_be32(GFS2_MAGIC);
936 if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC))
937 ri->ri_generation_deleted = cpu_to_be64(generation);
938}
939
940bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation)
941{
942 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
943
944 if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC))
945 return false;
946 return generation <= be64_to_cpu(ri->ri_generation_deleted);
947}
948
949static void gfs2_glock_poke(struct gfs2_glock *gl)
950{
951 int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP;
952 struct gfs2_holder gh;
953 int error;
954
955 __gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh, _RET_IP_);
956 error = gfs2_glock_nq(&gh);
957 if (!error)
958 gfs2_glock_dq(&gh);
959 gfs2_holder_uninit(&gh);
960}
961
962static void gfs2_try_evict(struct gfs2_glock *gl)
963{
964 struct gfs2_inode *ip;
965
966 /*
967 * If there is contention on the iopen glock and we have an inode, try
968 * to grab and release the inode so that it can be evicted. The
969 * GIF_DEFER_DELETE flag indicates to gfs2_evict_inode() that the inode
970 * should not be deleted locally. This will allow the remote node to
971 * go ahead and delete the inode without us having to do it, which will
972 * avoid rgrp glock thrashing.
973 *
974 * The remote node is likely still holding the corresponding inode
975 * glock, so it will run before we get to verify that the delete has
976 * happened below. (Verification is triggered by the call to
977 * gfs2_queue_verify_delete() in gfs2_evict_inode().)
978 */
979 spin_lock(&gl->gl_lockref.lock);
980 ip = gl->gl_object;
981 if (ip && !igrab(&ip->i_inode))
982 ip = NULL;
983 spin_unlock(&gl->gl_lockref.lock);
984 if (ip) {
985 wait_on_inode(&ip->i_inode);
986 if (is_bad_inode(&ip->i_inode)) {
987 iput(&ip->i_inode);
988 ip = NULL;
989 }
990 }
991 if (ip) {
992 set_bit(GIF_DEFER_DELETE, &ip->i_flags);
993 d_prune_aliases(&ip->i_inode);
994 iput(&ip->i_inode);
995
996 /* If the inode was evicted, gl->gl_object will now be NULL. */
997 spin_lock(&gl->gl_lockref.lock);
998 ip = gl->gl_object;
999 if (ip) {
1000 clear_bit(GIF_DEFER_DELETE, &ip->i_flags);
1001 if (!igrab(&ip->i_inode))
1002 ip = NULL;
1003 }
1004 spin_unlock(&gl->gl_lockref.lock);
1005 if (ip) {
1006 gfs2_glock_poke(ip->i_gl);
1007 iput(&ip->i_inode);
1008 }
1009 }
1010}
1011
1012bool gfs2_queue_try_to_evict(struct gfs2_glock *gl)
1013{
1014 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1015
1016 if (test_and_set_bit(GLF_TRY_TO_EVICT, &gl->gl_flags))
1017 return false;
1018 return !mod_delayed_work(sdp->sd_delete_wq, &gl->gl_delete, 0);
1019}
1020
1021bool gfs2_queue_verify_delete(struct gfs2_glock *gl, bool later)
1022{
1023 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1024 unsigned long delay;
1025
1026 if (test_and_set_bit(GLF_VERIFY_DELETE, &gl->gl_flags))
1027 return false;
1028 delay = later ? HZ + get_random_long() % (HZ * 9) : 0;
1029 return queue_delayed_work(sdp->sd_delete_wq, &gl->gl_delete, delay);
1030}
1031
1032static void delete_work_func(struct work_struct *work)
1033{
1034 struct delayed_work *dwork = to_delayed_work(work);
1035 struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete);
1036 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1037 bool verify_delete = test_and_clear_bit(GLF_VERIFY_DELETE, &gl->gl_flags);
1038
1039 if (test_and_clear_bit(GLF_TRY_TO_EVICT, &gl->gl_flags))
1040 gfs2_try_evict(gl);
1041
1042 if (verify_delete) {
1043 u64 no_addr = gl->gl_name.ln_number;
1044 struct inode *inode;
1045
1046 inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino,
1047 GFS2_BLKST_UNLINKED);
1048 if (IS_ERR(inode)) {
1049 if (PTR_ERR(inode) == -EAGAIN &&
1050 !test_bit(SDF_KILL, &sdp->sd_flags) &&
1051 gfs2_queue_verify_delete(gl, true))
1052 return;
1053 } else {
1054 d_prune_aliases(inode);
1055 iput(inode);
1056 }
1057 }
1058
1059 gfs2_glock_put(gl);
1060}
1061
1062static void glock_work_func(struct work_struct *work)
1063{
1064 unsigned long delay = 0;
1065 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
1066 unsigned int drop_refs = 1;
1067
1068 spin_lock(&gl->gl_lockref.lock);
1069 if (test_bit(GLF_HAVE_REPLY, &gl->gl_flags)) {
1070 clear_bit(GLF_HAVE_REPLY, &gl->gl_flags);
1071 finish_xmote(gl, gl->gl_reply);
1072 drop_refs++;
1073 }
1074 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1075 gl->gl_state != LM_ST_UNLOCKED &&
1076 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
1077 if (gl->gl_name.ln_type == LM_TYPE_INODE) {
1078 unsigned long holdtime, now = jiffies;
1079
1080 holdtime = gl->gl_tchange + gl->gl_hold_time;
1081 if (time_before(now, holdtime))
1082 delay = holdtime - now;
1083 }
1084
1085 if (!delay) {
1086 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1087 gfs2_set_demote(GLF_DEMOTE, gl);
1088 }
1089 }
1090 run_queue(gl, 0);
1091 if (delay) {
1092 /* Keep one glock reference for the work we requeue. */
1093 drop_refs--;
1094 gfs2_glock_queue_work(gl, delay);
1095 }
1096
1097 /* Drop the remaining glock references manually. */
1098 GLOCK_BUG_ON(gl, gl->gl_lockref.count < drop_refs);
1099 gl->gl_lockref.count -= drop_refs;
1100 if (!gl->gl_lockref.count) {
1101 if (gl->gl_state == LM_ST_UNLOCKED) {
1102 __gfs2_glock_put(gl);
1103 return;
1104 }
1105 gfs2_glock_add_to_lru(gl);
1106 }
1107 spin_unlock(&gl->gl_lockref.lock);
1108}
1109
1110static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
1111 struct gfs2_glock *new)
1112{
1113 struct wait_glock_queue wait;
1114 wait_queue_head_t *wq = glock_waitqueue(name);
1115 struct gfs2_glock *gl;
1116
1117 wait.name = name;
1118 init_wait(&wait.wait);
1119 wait.wait.func = glock_wake_function;
1120
1121again:
1122 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1123 rcu_read_lock();
1124 if (new) {
1125 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
1126 &new->gl_node, ht_parms);
1127 if (IS_ERR(gl))
1128 goto out;
1129 } else {
1130 gl = rhashtable_lookup_fast(&gl_hash_table,
1131 name, ht_parms);
1132 }
1133 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
1134 rcu_read_unlock();
1135 schedule();
1136 goto again;
1137 }
1138out:
1139 rcu_read_unlock();
1140 finish_wait(wq, &wait.wait);
1141 if (gl)
1142 gfs2_glock_remove_from_lru(gl);
1143 return gl;
1144}
1145
1146/**
1147 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
1148 * @sdp: The GFS2 superblock
1149 * @number: the lock number
1150 * @glops: The glock_operations to use
1151 * @create: If 0, don't create the glock if it doesn't exist
1152 * @glp: the glock is returned here
1153 *
1154 * This does not lock a glock, just finds/creates structures for one.
1155 *
1156 * Returns: errno
1157 */
1158
1159int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
1160 const struct gfs2_glock_operations *glops, int create,
1161 struct gfs2_glock **glp)
1162{
1163 struct super_block *s = sdp->sd_vfs;
1164 struct lm_lockname name = { .ln_number = number,
1165 .ln_type = glops->go_type,
1166 .ln_sbd = sdp };
1167 struct gfs2_glock *gl, *tmp;
1168 struct address_space *mapping;
1169
1170 gl = find_insert_glock(&name, NULL);
1171 if (gl)
1172 goto found;
1173 if (!create)
1174 return -ENOENT;
1175
1176 if (glops->go_flags & GLOF_ASPACE) {
1177 struct gfs2_glock_aspace *gla =
1178 kmem_cache_alloc(gfs2_glock_aspace_cachep, GFP_NOFS);
1179 if (!gla)
1180 return -ENOMEM;
1181 gl = &gla->glock;
1182 } else {
1183 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_NOFS);
1184 if (!gl)
1185 return -ENOMEM;
1186 }
1187 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
1188 gl->gl_ops = glops;
1189
1190 if (glops->go_flags & GLOF_LVB) {
1191 gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1192 if (!gl->gl_lksb.sb_lvbptr) {
1193 gfs2_glock_dealloc(&gl->gl_rcu);
1194 return -ENOMEM;
1195 }
1196 }
1197
1198 atomic_inc(&sdp->sd_glock_disposal);
1199 gl->gl_node.next = NULL;
1200 gl->gl_flags = BIT(GLF_INITIAL);
1201 if (glops->go_instantiate)
1202 gl->gl_flags |= BIT(GLF_INSTANTIATE_NEEDED);
1203 gl->gl_name = name;
1204 lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass);
1205 gl->gl_lockref.count = 1;
1206 gl->gl_state = LM_ST_UNLOCKED;
1207 gl->gl_target = LM_ST_UNLOCKED;
1208 gl->gl_demote_state = LM_ST_EXCLUSIVE;
1209 gl->gl_dstamp = 0;
1210 preempt_disable();
1211 /* We use the global stats to estimate the initial per-glock stats */
1212 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
1213 preempt_enable();
1214 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
1215 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
1216 gl->gl_tchange = jiffies;
1217 gl->gl_object = NULL;
1218 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
1219 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
1220 if (gl->gl_name.ln_type == LM_TYPE_IOPEN)
1221 INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func);
1222
1223 mapping = gfs2_glock2aspace(gl);
1224 if (mapping) {
1225 mapping->a_ops = &gfs2_meta_aops;
1226 mapping->host = s->s_bdev->bd_mapping->host;
1227 mapping->flags = 0;
1228 mapping_set_gfp_mask(mapping, GFP_NOFS);
1229 mapping->i_private_data = NULL;
1230 mapping->writeback_index = 0;
1231 }
1232
1233 tmp = find_insert_glock(&name, gl);
1234 if (tmp) {
1235 gfs2_glock_dealloc(&gl->gl_rcu);
1236 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
1237 wake_up(&sdp->sd_kill_wait);
1238
1239 if (IS_ERR(tmp))
1240 return PTR_ERR(tmp);
1241 gl = tmp;
1242 }
1243
1244found:
1245 *glp = gl;
1246 return 0;
1247}
1248
1249/**
1250 * __gfs2_holder_init - initialize a struct gfs2_holder in the default way
1251 * @gl: the glock
1252 * @state: the state we're requesting
1253 * @flags: the modifier flags
1254 * @gh: the holder structure
1255 *
1256 */
1257
1258void __gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
1259 struct gfs2_holder *gh, unsigned long ip)
1260{
1261 INIT_LIST_HEAD(&gh->gh_list);
1262 gh->gh_gl = gfs2_glock_hold(gl);
1263 gh->gh_ip = ip;
1264 gh->gh_owner_pid = get_pid(task_pid(current));
1265 gh->gh_state = state;
1266 gh->gh_flags = flags;
1267 gh->gh_iflags = 0;
1268}
1269
1270/**
1271 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1272 * @state: the state we're requesting
1273 * @flags: the modifier flags
1274 * @gh: the holder structure
1275 *
1276 * Don't mess with the glock.
1277 *
1278 */
1279
1280void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
1281{
1282 gh->gh_state = state;
1283 gh->gh_flags = flags;
1284 gh->gh_iflags = 0;
1285 gh->gh_ip = _RET_IP_;
1286 put_pid(gh->gh_owner_pid);
1287 gh->gh_owner_pid = get_pid(task_pid(current));
1288}
1289
1290/**
1291 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1292 * @gh: the holder structure
1293 *
1294 */
1295
1296void gfs2_holder_uninit(struct gfs2_holder *gh)
1297{
1298 put_pid(gh->gh_owner_pid);
1299 gfs2_glock_put(gh->gh_gl);
1300 gfs2_holder_mark_uninitialized(gh);
1301 gh->gh_ip = 0;
1302}
1303
1304static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
1305 unsigned long start_time)
1306{
1307 /* Have we waited longer that a second? */
1308 if (time_after(jiffies, start_time + HZ)) {
1309 /* Lengthen the minimum hold time. */
1310 gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
1311 GL_GLOCK_MAX_HOLD);
1312 }
1313}
1314
1315/**
1316 * gfs2_glock_holder_ready - holder is ready and its error code can be collected
1317 * @gh: the glock holder
1318 *
1319 * Called when a glock holder no longer needs to be waited for because it is
1320 * now either held (HIF_HOLDER set; gh_error == 0), or acquiring the lock has
1321 * failed (gh_error != 0).
1322 */
1323
1324int gfs2_glock_holder_ready(struct gfs2_holder *gh)
1325{
1326 if (gh->gh_error || (gh->gh_flags & GL_SKIP))
1327 return gh->gh_error;
1328 gh->gh_error = gfs2_instantiate(gh);
1329 if (gh->gh_error)
1330 gfs2_glock_dq(gh);
1331 return gh->gh_error;
1332}
1333
1334/**
1335 * gfs2_glock_wait - wait on a glock acquisition
1336 * @gh: the glock holder
1337 *
1338 * Returns: 0 on success
1339 */
1340
1341int gfs2_glock_wait(struct gfs2_holder *gh)
1342{
1343 unsigned long start_time = jiffies;
1344
1345 might_sleep();
1346 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1347 gfs2_glock_update_hold_time(gh->gh_gl, start_time);
1348 return gfs2_glock_holder_ready(gh);
1349}
1350
1351static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
1352{
1353 int i;
1354
1355 for (i = 0; i < num_gh; i++)
1356 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
1357 return 1;
1358 return 0;
1359}
1360
1361/**
1362 * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1363 * @num_gh: the number of holders in the array
1364 * @ghs: the glock holder array
1365 *
1366 * Returns: 0 on success, meaning all glocks have been granted and are held.
1367 * -ESTALE if the request timed out, meaning all glocks were released,
1368 * and the caller should retry the operation.
1369 */
1370
1371int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
1372{
1373 struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
1374 int i, ret = 0, timeout = 0;
1375 unsigned long start_time = jiffies;
1376
1377 might_sleep();
1378 /*
1379 * Total up the (minimum hold time * 2) of all glocks and use that to
1380 * determine the max amount of time we should wait.
1381 */
1382 for (i = 0; i < num_gh; i++)
1383 timeout += ghs[i].gh_gl->gl_hold_time << 1;
1384
1385 if (!wait_event_timeout(sdp->sd_async_glock_wait,
1386 !glocks_pending(num_gh, ghs), timeout)) {
1387 ret = -ESTALE; /* request timed out. */
1388 goto out;
1389 }
1390
1391 for (i = 0; i < num_gh; i++) {
1392 struct gfs2_holder *gh = &ghs[i];
1393 int ret2;
1394
1395 if (test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1396 gfs2_glock_update_hold_time(gh->gh_gl,
1397 start_time);
1398 }
1399 ret2 = gfs2_glock_holder_ready(gh);
1400 if (!ret)
1401 ret = ret2;
1402 }
1403
1404out:
1405 if (ret) {
1406 for (i = 0; i < num_gh; i++) {
1407 struct gfs2_holder *gh = &ghs[i];
1408
1409 gfs2_glock_dq(gh);
1410 }
1411 }
1412 return ret;
1413}
1414
1415/**
1416 * request_demote - process a demote request
1417 * @gl: the glock
1418 * @state: the state the caller wants us to change to
1419 * @delay: zero to demote immediately; otherwise pending demote
1420 * @remote: true if this came from a different cluster node
1421 *
1422 * There are only two requests that we are going to see in actual
1423 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1424 */
1425
1426static void request_demote(struct gfs2_glock *gl, unsigned int state,
1427 unsigned long delay, bool remote)
1428{
1429 gfs2_set_demote(delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE, gl);
1430 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
1431 gl->gl_demote_state = state;
1432 gl->gl_demote_time = jiffies;
1433 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
1434 gl->gl_demote_state != state) {
1435 gl->gl_demote_state = LM_ST_UNLOCKED;
1436 }
1437 if (gl->gl_ops->go_callback)
1438 gl->gl_ops->go_callback(gl, remote);
1439 trace_gfs2_demote_rq(gl, remote);
1440}
1441
1442void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
1443{
1444 struct va_format vaf;
1445 va_list args;
1446
1447 va_start(args, fmt);
1448
1449 if (seq) {
1450 seq_vprintf(seq, fmt, args);
1451 } else {
1452 vaf.fmt = fmt;
1453 vaf.va = &args;
1454
1455 pr_err("%pV", &vaf);
1456 }
1457
1458 va_end(args);
1459}
1460
1461static inline bool pid_is_meaningful(const struct gfs2_holder *gh)
1462{
1463 if (!(gh->gh_flags & GL_NOPID))
1464 return true;
1465 if (gh->gh_state == LM_ST_UNLOCKED)
1466 return true;
1467 return false;
1468}
1469
1470/**
1471 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1472 * @gh: the holder structure to add
1473 *
1474 * Eventually we should move the recursive locking trap to a
1475 * debugging option or something like that. This is the fast
1476 * path and needs to have the minimum number of distractions.
1477 *
1478 */
1479
1480static inline void add_to_queue(struct gfs2_holder *gh)
1481__releases(&gl->gl_lockref.lock)
1482__acquires(&gl->gl_lockref.lock)
1483{
1484 struct gfs2_glock *gl = gh->gh_gl;
1485 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1486 struct list_head *insert_pt = NULL;
1487 struct gfs2_holder *gh2;
1488 int try_futile = 0;
1489
1490 GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1491 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1492 GLOCK_BUG_ON(gl, true);
1493
1494 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1495 if (test_bit(GLF_LOCK, &gl->gl_flags)) {
1496 struct gfs2_holder *current_gh;
1497
1498 current_gh = find_first_holder(gl);
1499 try_futile = !may_grant(gl, current_gh, gh);
1500 }
1501 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1502 goto fail;
1503 }
1504
1505 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1506 if (likely(gh2->gh_owner_pid != gh->gh_owner_pid))
1507 continue;
1508 if (gh->gh_gl->gl_ops->go_type == LM_TYPE_FLOCK)
1509 continue;
1510 if (!pid_is_meaningful(gh2))
1511 continue;
1512 goto trap_recursive;
1513 }
1514 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1515 if (try_futile &&
1516 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1517fail:
1518 gh->gh_error = GLR_TRYFAILED;
1519 gfs2_holder_wake(gh);
1520 return;
1521 }
1522 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1523 continue;
1524 }
1525 trace_gfs2_glock_queue(gh, 1);
1526 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1527 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1528 if (likely(insert_pt == NULL)) {
1529 list_add_tail(&gh->gh_list, &gl->gl_holders);
1530 return;
1531 }
1532 list_add_tail(&gh->gh_list, insert_pt);
1533 spin_unlock(&gl->gl_lockref.lock);
1534 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1535 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1536 spin_lock(&gl->gl_lockref.lock);
1537 return;
1538
1539trap_recursive:
1540 fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1541 fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1542 fs_err(sdp, "lock type: %d req lock state : %d\n",
1543 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1544 fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1545 fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1546 fs_err(sdp, "lock type: %d req lock state : %d\n",
1547 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1548 gfs2_dump_glock(NULL, gl, true);
1549 BUG();
1550}
1551
1552/**
1553 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1554 * @gh: the holder structure
1555 *
1556 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1557 *
1558 * Returns: 0, GLR_TRYFAILED, or errno on failure
1559 */
1560
1561int gfs2_glock_nq(struct gfs2_holder *gh)
1562{
1563 struct gfs2_glock *gl = gh->gh_gl;
1564 int error;
1565
1566 if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP))
1567 return -EIO;
1568
1569 if (gh->gh_flags & GL_NOBLOCK) {
1570 struct gfs2_holder *current_gh;
1571
1572 error = -ECHILD;
1573 spin_lock(&gl->gl_lockref.lock);
1574 if (find_last_waiter(gl))
1575 goto unlock;
1576 current_gh = find_first_holder(gl);
1577 if (!may_grant(gl, current_gh, gh))
1578 goto unlock;
1579 set_bit(HIF_HOLDER, &gh->gh_iflags);
1580 list_add_tail(&gh->gh_list, &gl->gl_holders);
1581 trace_gfs2_promote(gh);
1582 error = 0;
1583unlock:
1584 spin_unlock(&gl->gl_lockref.lock);
1585 return error;
1586 }
1587
1588 gh->gh_error = 0;
1589 spin_lock(&gl->gl_lockref.lock);
1590 add_to_queue(gh);
1591 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1592 test_and_clear_bit(GLF_HAVE_FROZEN_REPLY, &gl->gl_flags))) {
1593 set_bit(GLF_HAVE_REPLY, &gl->gl_flags);
1594 gl->gl_lockref.count++;
1595 gfs2_glock_queue_work(gl, 0);
1596 }
1597 run_queue(gl, 1);
1598 spin_unlock(&gl->gl_lockref.lock);
1599
1600 error = 0;
1601 if (!(gh->gh_flags & GL_ASYNC))
1602 error = gfs2_glock_wait(gh);
1603
1604 return error;
1605}
1606
1607/**
1608 * gfs2_glock_poll - poll to see if an async request has been completed
1609 * @gh: the holder
1610 *
1611 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1612 */
1613
1614int gfs2_glock_poll(struct gfs2_holder *gh)
1615{
1616 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1617}
1618
1619static void __gfs2_glock_dq(struct gfs2_holder *gh)
1620{
1621 struct gfs2_glock *gl = gh->gh_gl;
1622 unsigned delay = 0;
1623 int fast_path = 0;
1624
1625 /*
1626 * This holder should not be cached, so mark it for demote.
1627 * Note: this should be done before the glock_needs_demote
1628 * check below.
1629 */
1630 if (gh->gh_flags & GL_NOCACHE)
1631 request_demote(gl, LM_ST_UNLOCKED, 0, false);
1632
1633 list_del_init(&gh->gh_list);
1634 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1635 trace_gfs2_glock_queue(gh, 0);
1636
1637 /*
1638 * If there hasn't been a demote request we are done.
1639 * (Let the remaining holders, if any, keep holding it.)
1640 */
1641 if (!glock_needs_demote(gl)) {
1642 if (list_empty(&gl->gl_holders))
1643 fast_path = 1;
1644 }
1645
1646 if (unlikely(!fast_path)) {
1647 gl->gl_lockref.count++;
1648 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1649 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1650 gl->gl_name.ln_type == LM_TYPE_INODE)
1651 delay = gl->gl_hold_time;
1652 gfs2_glock_queue_work(gl, delay);
1653 }
1654}
1655
1656/**
1657 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1658 * @gh: the glock holder
1659 *
1660 */
1661void gfs2_glock_dq(struct gfs2_holder *gh)
1662{
1663 struct gfs2_glock *gl = gh->gh_gl;
1664 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1665
1666 spin_lock(&gl->gl_lockref.lock);
1667 if (!gfs2_holder_queued(gh)) {
1668 /*
1669 * May have already been dequeued because the locking request
1670 * was GL_ASYNC and it has failed in the meantime.
1671 */
1672 goto out;
1673 }
1674
1675 if (list_is_first(&gh->gh_list, &gl->gl_holders) &&
1676 !test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1677 spin_unlock(&gl->gl_lockref.lock);
1678 gl->gl_name.ln_sbd->sd_lockstruct.ls_ops->lm_cancel(gl);
1679 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1680 spin_lock(&gl->gl_lockref.lock);
1681 }
1682
1683 /*
1684 * If we're in the process of file system withdraw, we cannot just
1685 * dequeue any glocks until our journal is recovered, lest we introduce
1686 * file system corruption. We need two exceptions to this rule: We need
1687 * to allow unlocking of nondisk glocks and the glock for our own
1688 * journal that needs recovery.
1689 */
1690 if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) &&
1691 glock_blocked_by_withdraw(gl) &&
1692 gh->gh_gl != sdp->sd_jinode_gl) {
1693 sdp->sd_glock_dqs_held++;
1694 spin_unlock(&gl->gl_lockref.lock);
1695 might_sleep();
1696 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
1697 TASK_UNINTERRUPTIBLE);
1698 spin_lock(&gl->gl_lockref.lock);
1699 }
1700
1701 __gfs2_glock_dq(gh);
1702out:
1703 spin_unlock(&gl->gl_lockref.lock);
1704}
1705
1706void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1707{
1708 struct gfs2_glock *gl = gh->gh_gl;
1709 gfs2_glock_dq(gh);
1710 might_sleep();
1711 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1712}
1713
1714/**
1715 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1716 * @gh: the holder structure
1717 *
1718 */
1719
1720void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1721{
1722 gfs2_glock_dq(gh);
1723 gfs2_holder_uninit(gh);
1724}
1725
1726/**
1727 * gfs2_glock_nq_num - acquire a glock based on lock number
1728 * @sdp: the filesystem
1729 * @number: the lock number
1730 * @glops: the glock operations for the type of glock
1731 * @state: the state to acquire the glock in
1732 * @flags: modifier flags for the acquisition
1733 * @gh: the struct gfs2_holder
1734 *
1735 * Returns: errno
1736 */
1737
1738int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1739 const struct gfs2_glock_operations *glops,
1740 unsigned int state, u16 flags, struct gfs2_holder *gh)
1741{
1742 struct gfs2_glock *gl;
1743 int error;
1744
1745 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1746 if (!error) {
1747 error = gfs2_glock_nq_init(gl, state, flags, gh);
1748 gfs2_glock_put(gl);
1749 }
1750
1751 return error;
1752}
1753
1754/**
1755 * glock_compare - Compare two struct gfs2_glock structures for sorting
1756 * @arg_a: the first structure
1757 * @arg_b: the second structure
1758 *
1759 */
1760
1761static int glock_compare(const void *arg_a, const void *arg_b)
1762{
1763 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1764 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1765 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1766 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1767
1768 if (a->ln_number > b->ln_number)
1769 return 1;
1770 if (a->ln_number < b->ln_number)
1771 return -1;
1772 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1773 return 0;
1774}
1775
1776/**
1777 * nq_m_sync - synchronously acquire more than one glock in deadlock free order
1778 * @num_gh: the number of structures
1779 * @ghs: an array of struct gfs2_holder structures
1780 * @p: placeholder for the holder structure to pass back
1781 *
1782 * Returns: 0 on success (all glocks acquired),
1783 * errno on failure (no glocks acquired)
1784 */
1785
1786static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1787 struct gfs2_holder **p)
1788{
1789 unsigned int x;
1790 int error = 0;
1791
1792 for (x = 0; x < num_gh; x++)
1793 p[x] = &ghs[x];
1794
1795 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1796
1797 for (x = 0; x < num_gh; x++) {
1798 error = gfs2_glock_nq(p[x]);
1799 if (error) {
1800 while (x--)
1801 gfs2_glock_dq(p[x]);
1802 break;
1803 }
1804 }
1805
1806 return error;
1807}
1808
1809/**
1810 * gfs2_glock_nq_m - acquire multiple glocks
1811 * @num_gh: the number of structures
1812 * @ghs: an array of struct gfs2_holder structures
1813 *
1814 * Returns: 0 on success (all glocks acquired),
1815 * errno on failure (no glocks acquired)
1816 */
1817
1818int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1819{
1820 struct gfs2_holder *tmp[4];
1821 struct gfs2_holder **pph = tmp;
1822 int error = 0;
1823
1824 switch(num_gh) {
1825 case 0:
1826 return 0;
1827 case 1:
1828 return gfs2_glock_nq(ghs);
1829 default:
1830 if (num_gh <= 4)
1831 break;
1832 pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *),
1833 GFP_NOFS);
1834 if (!pph)
1835 return -ENOMEM;
1836 }
1837
1838 error = nq_m_sync(num_gh, ghs, pph);
1839
1840 if (pph != tmp)
1841 kfree(pph);
1842
1843 return error;
1844}
1845
1846/**
1847 * gfs2_glock_dq_m - release multiple glocks
1848 * @num_gh: the number of structures
1849 * @ghs: an array of struct gfs2_holder structures
1850 *
1851 */
1852
1853void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1854{
1855 while (num_gh--)
1856 gfs2_glock_dq(&ghs[num_gh]);
1857}
1858
1859void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1860{
1861 unsigned long delay = 0;
1862
1863 gfs2_glock_hold(gl);
1864 spin_lock(&gl->gl_lockref.lock);
1865 if (!list_empty(&gl->gl_holders) &&
1866 gl->gl_name.ln_type == LM_TYPE_INODE) {
1867 unsigned long now = jiffies;
1868 unsigned long holdtime;
1869
1870 holdtime = gl->gl_tchange + gl->gl_hold_time;
1871
1872 if (time_before(now, holdtime))
1873 delay = holdtime - now;
1874 if (test_bit(GLF_HAVE_REPLY, &gl->gl_flags))
1875 delay = gl->gl_hold_time;
1876 }
1877 request_demote(gl, state, delay, true);
1878 gfs2_glock_queue_work(gl, delay);
1879 spin_unlock(&gl->gl_lockref.lock);
1880}
1881
1882/**
1883 * gfs2_should_freeze - Figure out if glock should be frozen
1884 * @gl: The glock in question
1885 *
1886 * Glocks are not frozen if (a) the result of the dlm operation is
1887 * an error, (b) the locking operation was an unlock operation or
1888 * (c) if there is a "noexp" flagged request anywhere in the queue
1889 *
1890 * Returns: 1 if freezing should occur, 0 otherwise
1891 */
1892
1893static int gfs2_should_freeze(const struct gfs2_glock *gl)
1894{
1895 const struct gfs2_holder *gh;
1896
1897 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1898 return 0;
1899 if (gl->gl_target == LM_ST_UNLOCKED)
1900 return 0;
1901
1902 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1903 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1904 continue;
1905 if (LM_FLAG_NOEXP & gh->gh_flags)
1906 return 0;
1907 }
1908
1909 return 1;
1910}
1911
1912/**
1913 * gfs2_glock_complete - Callback used by locking
1914 * @gl: Pointer to the glock
1915 * @ret: The return value from the dlm
1916 *
1917 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1918 * to use a bitfield shared with other glock state fields.
1919 */
1920
1921void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1922{
1923 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1924
1925 spin_lock(&gl->gl_lockref.lock);
1926 gl->gl_reply = ret;
1927
1928 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1929 if (gfs2_should_freeze(gl)) {
1930 set_bit(GLF_HAVE_FROZEN_REPLY, &gl->gl_flags);
1931 spin_unlock(&gl->gl_lockref.lock);
1932 return;
1933 }
1934 }
1935
1936 gl->gl_lockref.count++;
1937 set_bit(GLF_HAVE_REPLY, &gl->gl_flags);
1938 gfs2_glock_queue_work(gl, 0);
1939 spin_unlock(&gl->gl_lockref.lock);
1940}
1941
1942static int glock_cmp(void *priv, const struct list_head *a,
1943 const struct list_head *b)
1944{
1945 struct gfs2_glock *gla, *glb;
1946
1947 gla = list_entry(a, struct gfs2_glock, gl_lru);
1948 glb = list_entry(b, struct gfs2_glock, gl_lru);
1949
1950 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1951 return 1;
1952 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1953 return -1;
1954
1955 return 0;
1956}
1957
1958static bool can_free_glock(struct gfs2_glock *gl)
1959{
1960 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1961
1962 return !test_bit(GLF_LOCK, &gl->gl_flags) &&
1963 !gl->gl_lockref.count &&
1964 (!test_bit(GLF_LFLUSH, &gl->gl_flags) ||
1965 test_bit(SDF_KILL, &sdp->sd_flags));
1966}
1967
1968/**
1969 * gfs2_dispose_glock_lru - Demote a list of glocks
1970 * @list: The list to dispose of
1971 *
1972 * Disposing of glocks may involve disk accesses, so that here we sort
1973 * the glocks by number (i.e. disk location of the inodes) so that if
1974 * there are any such accesses, they'll be sent in order (mostly).
1975 *
1976 * Must be called under the lru_lock, but may drop and retake this
1977 * lock. While the lru_lock is dropped, entries may vanish from the
1978 * list, but no new entries will appear on the list (since it is
1979 * private)
1980 */
1981
1982static unsigned long gfs2_dispose_glock_lru(struct list_head *list)
1983__releases(&lru_lock)
1984__acquires(&lru_lock)
1985{
1986 struct gfs2_glock *gl;
1987 unsigned long freed = 0;
1988
1989 list_sort(NULL, list, glock_cmp);
1990
1991 while(!list_empty(list)) {
1992 gl = list_first_entry(list, struct gfs2_glock, gl_lru);
1993 if (!spin_trylock(&gl->gl_lockref.lock)) {
1994add_back_to_lru:
1995 list_move(&gl->gl_lru, &lru_list);
1996 continue;
1997 }
1998 if (!can_free_glock(gl)) {
1999 spin_unlock(&gl->gl_lockref.lock);
2000 goto add_back_to_lru;
2001 }
2002 list_del_init(&gl->gl_lru);
2003 atomic_dec(&lru_count);
2004 clear_bit(GLF_LRU, &gl->gl_flags);
2005 freed++;
2006 gl->gl_lockref.count++;
2007 if (gl->gl_state != LM_ST_UNLOCKED)
2008 request_demote(gl, LM_ST_UNLOCKED, 0, false);
2009 gfs2_glock_queue_work(gl, 0);
2010 spin_unlock(&gl->gl_lockref.lock);
2011 cond_resched_lock(&lru_lock);
2012 }
2013 return freed;
2014}
2015
2016/**
2017 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
2018 * @nr: The number of entries to scan
2019 *
2020 * This function selects the entries on the LRU which are able to
2021 * be demoted, and then kicks off the process by calling
2022 * gfs2_dispose_glock_lru() above.
2023 */
2024
2025static unsigned long gfs2_scan_glock_lru(unsigned long nr)
2026{
2027 struct gfs2_glock *gl, *next;
2028 LIST_HEAD(dispose);
2029 unsigned long freed = 0;
2030
2031 spin_lock(&lru_lock);
2032 list_for_each_entry_safe(gl, next, &lru_list, gl_lru) {
2033 if (!nr--)
2034 break;
2035 if (can_free_glock(gl))
2036 list_move(&gl->gl_lru, &dispose);
2037 }
2038 if (!list_empty(&dispose))
2039 freed = gfs2_dispose_glock_lru(&dispose);
2040 spin_unlock(&lru_lock);
2041
2042 return freed;
2043}
2044
2045static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
2046 struct shrink_control *sc)
2047{
2048 if (!(sc->gfp_mask & __GFP_FS))
2049 return SHRINK_STOP;
2050 return gfs2_scan_glock_lru(sc->nr_to_scan);
2051}
2052
2053static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
2054 struct shrink_control *sc)
2055{
2056 return vfs_pressure_ratio(atomic_read(&lru_count));
2057}
2058
2059static struct shrinker *glock_shrinker;
2060
2061/**
2062 * glock_hash_walk - Call a function for glock in a hash bucket
2063 * @examiner: the function
2064 * @sdp: the filesystem
2065 *
2066 * Note that the function can be called multiple times on the same
2067 * object. So the user must ensure that the function can cope with
2068 * that.
2069 */
2070
2071static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
2072{
2073 struct gfs2_glock *gl;
2074 struct rhashtable_iter iter;
2075
2076 rhashtable_walk_enter(&gl_hash_table, &iter);
2077
2078 do {
2079 rhashtable_walk_start(&iter);
2080
2081 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl)) {
2082 if (gl->gl_name.ln_sbd == sdp)
2083 examiner(gl);
2084 }
2085
2086 rhashtable_walk_stop(&iter);
2087 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
2088
2089 rhashtable_walk_exit(&iter);
2090}
2091
2092void gfs2_cancel_delete_work(struct gfs2_glock *gl)
2093{
2094 clear_bit(GLF_TRY_TO_EVICT, &gl->gl_flags);
2095 clear_bit(GLF_VERIFY_DELETE, &gl->gl_flags);
2096 if (cancel_delayed_work(&gl->gl_delete))
2097 gfs2_glock_put(gl);
2098}
2099
2100static void flush_delete_work(struct gfs2_glock *gl)
2101{
2102 if (gl->gl_name.ln_type == LM_TYPE_IOPEN) {
2103 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2104
2105 if (cancel_delayed_work(&gl->gl_delete)) {
2106 queue_delayed_work(sdp->sd_delete_wq,
2107 &gl->gl_delete, 0);
2108 }
2109 }
2110}
2111
2112void gfs2_flush_delete_work(struct gfs2_sbd *sdp)
2113{
2114 glock_hash_walk(flush_delete_work, sdp);
2115 flush_workqueue(sdp->sd_delete_wq);
2116}
2117
2118/**
2119 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
2120 * @gl: The glock to thaw
2121 *
2122 */
2123
2124static void thaw_glock(struct gfs2_glock *gl)
2125{
2126 if (!test_and_clear_bit(GLF_HAVE_FROZEN_REPLY, &gl->gl_flags))
2127 return;
2128 if (!lockref_get_not_dead(&gl->gl_lockref))
2129 return;
2130
2131 gfs2_glock_remove_from_lru(gl);
2132 spin_lock(&gl->gl_lockref.lock);
2133 set_bit(GLF_HAVE_REPLY, &gl->gl_flags);
2134 gfs2_glock_queue_work(gl, 0);
2135 spin_unlock(&gl->gl_lockref.lock);
2136}
2137
2138/**
2139 * clear_glock - look at a glock and see if we can free it from glock cache
2140 * @gl: the glock to look at
2141 *
2142 */
2143
2144static void clear_glock(struct gfs2_glock *gl)
2145{
2146 gfs2_glock_remove_from_lru(gl);
2147
2148 spin_lock(&gl->gl_lockref.lock);
2149 if (!__lockref_is_dead(&gl->gl_lockref)) {
2150 gl->gl_lockref.count++;
2151 if (gl->gl_state != LM_ST_UNLOCKED)
2152 request_demote(gl, LM_ST_UNLOCKED, 0, false);
2153 gfs2_glock_queue_work(gl, 0);
2154 }
2155 spin_unlock(&gl->gl_lockref.lock);
2156}
2157
2158/**
2159 * gfs2_glock_thaw - Thaw any frozen glocks
2160 * @sdp: The super block
2161 *
2162 */
2163
2164void gfs2_glock_thaw(struct gfs2_sbd *sdp)
2165{
2166 glock_hash_walk(thaw_glock, sdp);
2167}
2168
2169static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2170{
2171 spin_lock(&gl->gl_lockref.lock);
2172 gfs2_dump_glock(seq, gl, fsid);
2173 spin_unlock(&gl->gl_lockref.lock);
2174}
2175
2176static void dump_glock_func(struct gfs2_glock *gl)
2177{
2178 dump_glock(NULL, gl, true);
2179}
2180
2181static void withdraw_dq(struct gfs2_glock *gl)
2182{
2183 spin_lock(&gl->gl_lockref.lock);
2184 if (!__lockref_is_dead(&gl->gl_lockref) &&
2185 glock_blocked_by_withdraw(gl))
2186 do_error(gl, LM_OUT_ERROR); /* remove pending waiters */
2187 spin_unlock(&gl->gl_lockref.lock);
2188}
2189
2190void gfs2_gl_dq_holders(struct gfs2_sbd *sdp)
2191{
2192 glock_hash_walk(withdraw_dq, sdp);
2193}
2194
2195/**
2196 * gfs2_gl_hash_clear - Empty out the glock hash table
2197 * @sdp: the filesystem
2198 *
2199 * Called when unmounting the filesystem.
2200 */
2201
2202void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
2203{
2204 unsigned long start = jiffies;
2205 bool timed_out = false;
2206
2207 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
2208 flush_workqueue(sdp->sd_glock_wq);
2209 glock_hash_walk(clear_glock, sdp);
2210 flush_workqueue(sdp->sd_glock_wq);
2211
2212 while (!timed_out) {
2213 wait_event_timeout(sdp->sd_kill_wait,
2214 !atomic_read(&sdp->sd_glock_disposal),
2215 HZ * 60);
2216 if (!atomic_read(&sdp->sd_glock_disposal))
2217 break;
2218 timed_out = time_after(jiffies, start + (HZ * 600));
2219 fs_warn(sdp, "%u glocks left after %u seconds%s\n",
2220 atomic_read(&sdp->sd_glock_disposal),
2221 jiffies_to_msecs(jiffies - start) / 1000,
2222 timed_out ? ":" : "; still waiting");
2223 }
2224 gfs2_lm_unmount(sdp);
2225 gfs2_free_dead_glocks(sdp);
2226 glock_hash_walk(dump_glock_func, sdp);
2227 destroy_workqueue(sdp->sd_glock_wq);
2228 sdp->sd_glock_wq = NULL;
2229}
2230
2231static const char *state2str(unsigned state)
2232{
2233 switch(state) {
2234 case LM_ST_UNLOCKED:
2235 return "UN";
2236 case LM_ST_SHARED:
2237 return "SH";
2238 case LM_ST_DEFERRED:
2239 return "DF";
2240 case LM_ST_EXCLUSIVE:
2241 return "EX";
2242 }
2243 return "??";
2244}
2245
2246static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
2247{
2248 char *p = buf;
2249 if (flags & LM_FLAG_TRY)
2250 *p++ = 't';
2251 if (flags & LM_FLAG_TRY_1CB)
2252 *p++ = 'T';
2253 if (flags & LM_FLAG_NOEXP)
2254 *p++ = 'e';
2255 if (flags & LM_FLAG_ANY)
2256 *p++ = 'A';
2257 if (flags & LM_FLAG_NODE_SCOPE)
2258 *p++ = 'n';
2259 if (flags & GL_ASYNC)
2260 *p++ = 'a';
2261 if (flags & GL_EXACT)
2262 *p++ = 'E';
2263 if (flags & GL_NOCACHE)
2264 *p++ = 'c';
2265 if (test_bit(HIF_HOLDER, &iflags))
2266 *p++ = 'H';
2267 if (test_bit(HIF_WAIT, &iflags))
2268 *p++ = 'W';
2269 if (flags & GL_SKIP)
2270 *p++ = 's';
2271 *p = 0;
2272 return buf;
2273}
2274
2275/**
2276 * dump_holder - print information about a glock holder
2277 * @seq: the seq_file struct
2278 * @gh: the glock holder
2279 * @fs_id_buf: pointer to file system id (if requested)
2280 *
2281 */
2282
2283static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
2284 const char *fs_id_buf)
2285{
2286 const char *comm = "(none)";
2287 pid_t owner_pid = 0;
2288 char flags_buf[32];
2289
2290 rcu_read_lock();
2291 if (pid_is_meaningful(gh)) {
2292 struct task_struct *gh_owner;
2293
2294 comm = "(ended)";
2295 owner_pid = pid_nr(gh->gh_owner_pid);
2296 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
2297 if (gh_owner)
2298 comm = gh_owner->comm;
2299 }
2300 gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2301 fs_id_buf, state2str(gh->gh_state),
2302 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
2303 gh->gh_error, (long)owner_pid, comm, (void *)gh->gh_ip);
2304 rcu_read_unlock();
2305}
2306
2307static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
2308{
2309 const unsigned long *gflags = &gl->gl_flags;
2310 char *p = buf;
2311
2312 if (test_bit(GLF_LOCK, gflags))
2313 *p++ = 'l';
2314 if (test_bit(GLF_DEMOTE, gflags))
2315 *p++ = 'D';
2316 if (test_bit(GLF_PENDING_DEMOTE, gflags))
2317 *p++ = 'd';
2318 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
2319 *p++ = 'p';
2320 if (test_bit(GLF_DIRTY, gflags))
2321 *p++ = 'y';
2322 if (test_bit(GLF_LFLUSH, gflags))
2323 *p++ = 'f';
2324 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
2325 *p++ = 'i';
2326 if (test_bit(GLF_HAVE_REPLY, gflags))
2327 *p++ = 'r';
2328 if (test_bit(GLF_INITIAL, gflags))
2329 *p++ = 'a';
2330 if (test_bit(GLF_HAVE_FROZEN_REPLY, gflags))
2331 *p++ = 'F';
2332 if (!list_empty(&gl->gl_holders))
2333 *p++ = 'q';
2334 if (test_bit(GLF_LRU, gflags))
2335 *p++ = 'L';
2336 if (gl->gl_object)
2337 *p++ = 'o';
2338 if (test_bit(GLF_BLOCKING, gflags))
2339 *p++ = 'b';
2340 if (test_bit(GLF_UNLOCKED, gflags))
2341 *p++ = 'x';
2342 if (test_bit(GLF_INSTANTIATE_NEEDED, gflags))
2343 *p++ = 'n';
2344 if (test_bit(GLF_INSTANTIATE_IN_PROG, gflags))
2345 *p++ = 'N';
2346 if (test_bit(GLF_TRY_TO_EVICT, gflags))
2347 *p++ = 'e';
2348 if (test_bit(GLF_VERIFY_DELETE, gflags))
2349 *p++ = 'E';
2350 *p = 0;
2351 return buf;
2352}
2353
2354/**
2355 * gfs2_dump_glock - print information about a glock
2356 * @seq: The seq_file struct
2357 * @gl: the glock
2358 * @fsid: If true, also dump the file system id
2359 *
2360 * The file format is as follows:
2361 * One line per object, capital letters are used to indicate objects
2362 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2363 * other objects are indented by a single space and follow the glock to
2364 * which they are related. Fields are indicated by lower case letters
2365 * followed by a colon and the field value, except for strings which are in
2366 * [] so that its possible to see if they are composed of spaces for
2367 * example. The field's are n = number (id of the object), f = flags,
2368 * t = type, s = state, r = refcount, e = error, p = pid.
2369 *
2370 */
2371
2372void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2373{
2374 const struct gfs2_glock_operations *glops = gl->gl_ops;
2375 unsigned long long dtime;
2376 const struct gfs2_holder *gh;
2377 char gflags_buf[32];
2378 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2379 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2380 unsigned long nrpages = 0;
2381
2382 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
2383 struct address_space *mapping = gfs2_glock2aspace(gl);
2384
2385 nrpages = mapping->nrpages;
2386 }
2387 memset(fs_id_buf, 0, sizeof(fs_id_buf));
2388 if (fsid && sdp) /* safety precaution */
2389 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2390 dtime = jiffies - gl->gl_demote_time;
2391 dtime *= 1000000/HZ; /* demote time in uSec */
2392 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
2393 dtime = 0;
2394 gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2395 "v:%d r:%d m:%ld p:%lu\n",
2396 fs_id_buf, state2str(gl->gl_state),
2397 gl->gl_name.ln_type,
2398 (unsigned long long)gl->gl_name.ln_number,
2399 gflags2str(gflags_buf, gl),
2400 state2str(gl->gl_target),
2401 state2str(gl->gl_demote_state), dtime,
2402 atomic_read(&gl->gl_ail_count),
2403 atomic_read(&gl->gl_revokes),
2404 (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages);
2405
2406 list_for_each_entry(gh, &gl->gl_holders, gh_list)
2407 dump_holder(seq, gh, fs_id_buf);
2408
2409 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
2410 glops->go_dump(seq, gl, fs_id_buf);
2411}
2412
2413static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
2414{
2415 struct gfs2_glock *gl = iter_ptr;
2416
2417 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2418 gl->gl_name.ln_type,
2419 (unsigned long long)gl->gl_name.ln_number,
2420 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
2421 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
2422 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
2423 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
2424 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
2425 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
2426 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
2427 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
2428 return 0;
2429}
2430
2431static const char *gfs2_gltype[] = {
2432 "type",
2433 "reserved",
2434 "nondisk",
2435 "inode",
2436 "rgrp",
2437 "meta",
2438 "iopen",
2439 "flock",
2440 "plock",
2441 "quota",
2442 "journal",
2443};
2444
2445static const char *gfs2_stype[] = {
2446 [GFS2_LKS_SRTT] = "srtt",
2447 [GFS2_LKS_SRTTVAR] = "srttvar",
2448 [GFS2_LKS_SRTTB] = "srttb",
2449 [GFS2_LKS_SRTTVARB] = "srttvarb",
2450 [GFS2_LKS_SIRT] = "sirt",
2451 [GFS2_LKS_SIRTVAR] = "sirtvar",
2452 [GFS2_LKS_DCOUNT] = "dlm",
2453 [GFS2_LKS_QCOUNT] = "queue",
2454};
2455
2456#define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2457
2458static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
2459{
2460 struct gfs2_sbd *sdp = seq->private;
2461 loff_t pos = *(loff_t *)iter_ptr;
2462 unsigned index = pos >> 3;
2463 unsigned subindex = pos & 0x07;
2464 int i;
2465
2466 if (index == 0 && subindex != 0)
2467 return 0;
2468
2469 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
2470 (index == 0) ? "cpu": gfs2_stype[subindex]);
2471
2472 for_each_possible_cpu(i) {
2473 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
2474
2475 if (index == 0)
2476 seq_printf(seq, " %15u", i);
2477 else
2478 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
2479 lkstats[index - 1].stats[subindex]);
2480 }
2481 seq_putc(seq, '\n');
2482 return 0;
2483}
2484
2485int __init gfs2_glock_init(void)
2486{
2487 int i, ret;
2488
2489 ret = rhashtable_init(&gl_hash_table, &ht_parms);
2490 if (ret < 0)
2491 return ret;
2492
2493 glock_shrinker = shrinker_alloc(0, "gfs2-glock");
2494 if (!glock_shrinker) {
2495 rhashtable_destroy(&gl_hash_table);
2496 return -ENOMEM;
2497 }
2498
2499 glock_shrinker->count_objects = gfs2_glock_shrink_count;
2500 glock_shrinker->scan_objects = gfs2_glock_shrink_scan;
2501
2502 shrinker_register(glock_shrinker);
2503
2504 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2505 init_waitqueue_head(glock_wait_table + i);
2506
2507 return 0;
2508}
2509
2510void gfs2_glock_exit(void)
2511{
2512 shrinker_free(glock_shrinker);
2513 rhashtable_destroy(&gl_hash_table);
2514}
2515
2516static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2517{
2518 struct gfs2_glock *gl = gi->gl;
2519
2520 if (gl) {
2521 if (n == 0)
2522 return;
2523 gfs2_glock_put_async(gl);
2524 }
2525 for (;;) {
2526 gl = rhashtable_walk_next(&gi->hti);
2527 if (IS_ERR_OR_NULL(gl)) {
2528 if (gl == ERR_PTR(-EAGAIN)) {
2529 n = 1;
2530 continue;
2531 }
2532 gl = NULL;
2533 break;
2534 }
2535 if (gl->gl_name.ln_sbd != gi->sdp)
2536 continue;
2537 if (n <= 1) {
2538 if (!lockref_get_not_dead(&gl->gl_lockref))
2539 continue;
2540 break;
2541 } else {
2542 if (__lockref_is_dead(&gl->gl_lockref))
2543 continue;
2544 n--;
2545 }
2546 }
2547 gi->gl = gl;
2548}
2549
2550static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2551 __acquires(RCU)
2552{
2553 struct gfs2_glock_iter *gi = seq->private;
2554 loff_t n;
2555
2556 /*
2557 * We can either stay where we are, skip to the next hash table
2558 * entry, or start from the beginning.
2559 */
2560 if (*pos < gi->last_pos) {
2561 rhashtable_walk_exit(&gi->hti);
2562 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2563 n = *pos + 1;
2564 } else {
2565 n = *pos - gi->last_pos;
2566 }
2567
2568 rhashtable_walk_start(&gi->hti);
2569
2570 gfs2_glock_iter_next(gi, n);
2571 gi->last_pos = *pos;
2572 return gi->gl;
2573}
2574
2575static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2576 loff_t *pos)
2577{
2578 struct gfs2_glock_iter *gi = seq->private;
2579
2580 (*pos)++;
2581 gi->last_pos = *pos;
2582 gfs2_glock_iter_next(gi, 1);
2583 return gi->gl;
2584}
2585
2586static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2587 __releases(RCU)
2588{
2589 struct gfs2_glock_iter *gi = seq->private;
2590
2591 rhashtable_walk_stop(&gi->hti);
2592}
2593
2594static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2595{
2596 dump_glock(seq, iter_ptr, false);
2597 return 0;
2598}
2599
2600static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2601{
2602 preempt_disable();
2603 if (*pos >= GFS2_NR_SBSTATS)
2604 return NULL;
2605 return pos;
2606}
2607
2608static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2609 loff_t *pos)
2610{
2611 (*pos)++;
2612 if (*pos >= GFS2_NR_SBSTATS)
2613 return NULL;
2614 return pos;
2615}
2616
2617static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2618{
2619 preempt_enable();
2620}
2621
2622static const struct seq_operations gfs2_glock_seq_ops = {
2623 .start = gfs2_glock_seq_start,
2624 .next = gfs2_glock_seq_next,
2625 .stop = gfs2_glock_seq_stop,
2626 .show = gfs2_glock_seq_show,
2627};
2628
2629static const struct seq_operations gfs2_glstats_seq_ops = {
2630 .start = gfs2_glock_seq_start,
2631 .next = gfs2_glock_seq_next,
2632 .stop = gfs2_glock_seq_stop,
2633 .show = gfs2_glstats_seq_show,
2634};
2635
2636static const struct seq_operations gfs2_sbstats_sops = {
2637 .start = gfs2_sbstats_seq_start,
2638 .next = gfs2_sbstats_seq_next,
2639 .stop = gfs2_sbstats_seq_stop,
2640 .show = gfs2_sbstats_seq_show,
2641};
2642
2643#define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2644
2645static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2646 const struct seq_operations *ops)
2647{
2648 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2649 if (ret == 0) {
2650 struct seq_file *seq = file->private_data;
2651 struct gfs2_glock_iter *gi = seq->private;
2652
2653 gi->sdp = inode->i_private;
2654 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2655 if (seq->buf)
2656 seq->size = GFS2_SEQ_GOODSIZE;
2657 /*
2658 * Initially, we are "before" the first hash table entry; the
2659 * first call to rhashtable_walk_next gets us the first entry.
2660 */
2661 gi->last_pos = -1;
2662 gi->gl = NULL;
2663 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2664 }
2665 return ret;
2666}
2667
2668static int gfs2_glocks_open(struct inode *inode, struct file *file)
2669{
2670 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2671}
2672
2673static int gfs2_glocks_release(struct inode *inode, struct file *file)
2674{
2675 struct seq_file *seq = file->private_data;
2676 struct gfs2_glock_iter *gi = seq->private;
2677
2678 if (gi->gl)
2679 gfs2_glock_put(gi->gl);
2680 rhashtable_walk_exit(&gi->hti);
2681 return seq_release_private(inode, file);
2682}
2683
2684static int gfs2_glstats_open(struct inode *inode, struct file *file)
2685{
2686 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2687}
2688
2689static const struct file_operations gfs2_glocks_fops = {
2690 .owner = THIS_MODULE,
2691 .open = gfs2_glocks_open,
2692 .read = seq_read,
2693 .llseek = seq_lseek,
2694 .release = gfs2_glocks_release,
2695};
2696
2697static const struct file_operations gfs2_glstats_fops = {
2698 .owner = THIS_MODULE,
2699 .open = gfs2_glstats_open,
2700 .read = seq_read,
2701 .llseek = seq_lseek,
2702 .release = gfs2_glocks_release,
2703};
2704
2705struct gfs2_glockfd_iter {
2706 struct super_block *sb;
2707 unsigned int tgid;
2708 struct task_struct *task;
2709 unsigned int fd;
2710 struct file *file;
2711};
2712
2713static struct task_struct *gfs2_glockfd_next_task(struct gfs2_glockfd_iter *i)
2714{
2715 struct pid_namespace *ns = task_active_pid_ns(current);
2716 struct pid *pid;
2717
2718 if (i->task)
2719 put_task_struct(i->task);
2720
2721 rcu_read_lock();
2722retry:
2723 i->task = NULL;
2724 pid = find_ge_pid(i->tgid, ns);
2725 if (pid) {
2726 i->tgid = pid_nr_ns(pid, ns);
2727 i->task = pid_task(pid, PIDTYPE_TGID);
2728 if (!i->task) {
2729 i->tgid++;
2730 goto retry;
2731 }
2732 get_task_struct(i->task);
2733 }
2734 rcu_read_unlock();
2735 return i->task;
2736}
2737
2738static struct file *gfs2_glockfd_next_file(struct gfs2_glockfd_iter *i)
2739{
2740 if (i->file) {
2741 fput(i->file);
2742 i->file = NULL;
2743 }
2744
2745 for(;; i->fd++) {
2746 i->file = fget_task_next(i->task, &i->fd);
2747 if (!i->file) {
2748 i->fd = 0;
2749 break;
2750 }
2751
2752 if (file_inode(i->file)->i_sb == i->sb)
2753 break;
2754
2755 fput(i->file);
2756 }
2757 return i->file;
2758}
2759
2760static void *gfs2_glockfd_seq_start(struct seq_file *seq, loff_t *pos)
2761{
2762 struct gfs2_glockfd_iter *i = seq->private;
2763
2764 if (*pos)
2765 return NULL;
2766 while (gfs2_glockfd_next_task(i)) {
2767 if (gfs2_glockfd_next_file(i))
2768 return i;
2769 i->tgid++;
2770 }
2771 return NULL;
2772}
2773
2774static void *gfs2_glockfd_seq_next(struct seq_file *seq, void *iter_ptr,
2775 loff_t *pos)
2776{
2777 struct gfs2_glockfd_iter *i = seq->private;
2778
2779 (*pos)++;
2780 i->fd++;
2781 do {
2782 if (gfs2_glockfd_next_file(i))
2783 return i;
2784 i->tgid++;
2785 } while (gfs2_glockfd_next_task(i));
2786 return NULL;
2787}
2788
2789static void gfs2_glockfd_seq_stop(struct seq_file *seq, void *iter_ptr)
2790{
2791 struct gfs2_glockfd_iter *i = seq->private;
2792
2793 if (i->file)
2794 fput(i->file);
2795 if (i->task)
2796 put_task_struct(i->task);
2797}
2798
2799static void gfs2_glockfd_seq_show_flock(struct seq_file *seq,
2800 struct gfs2_glockfd_iter *i)
2801{
2802 struct gfs2_file *fp = i->file->private_data;
2803 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
2804 struct lm_lockname gl_name = { .ln_type = LM_TYPE_RESERVED };
2805
2806 if (!READ_ONCE(fl_gh->gh_gl))
2807 return;
2808
2809 spin_lock(&i->file->f_lock);
2810 if (gfs2_holder_initialized(fl_gh))
2811 gl_name = fl_gh->gh_gl->gl_name;
2812 spin_unlock(&i->file->f_lock);
2813
2814 if (gl_name.ln_type != LM_TYPE_RESERVED) {
2815 seq_printf(seq, "%d %u %u/%llx\n",
2816 i->tgid, i->fd, gl_name.ln_type,
2817 (unsigned long long)gl_name.ln_number);
2818 }
2819}
2820
2821static int gfs2_glockfd_seq_show(struct seq_file *seq, void *iter_ptr)
2822{
2823 struct gfs2_glockfd_iter *i = seq->private;
2824 struct inode *inode = file_inode(i->file);
2825 struct gfs2_glock *gl;
2826
2827 inode_lock_shared(inode);
2828 gl = GFS2_I(inode)->i_iopen_gh.gh_gl;
2829 if (gl) {
2830 seq_printf(seq, "%d %u %u/%llx\n",
2831 i->tgid, i->fd, gl->gl_name.ln_type,
2832 (unsigned long long)gl->gl_name.ln_number);
2833 }
2834 gfs2_glockfd_seq_show_flock(seq, i);
2835 inode_unlock_shared(inode);
2836 return 0;
2837}
2838
2839static const struct seq_operations gfs2_glockfd_seq_ops = {
2840 .start = gfs2_glockfd_seq_start,
2841 .next = gfs2_glockfd_seq_next,
2842 .stop = gfs2_glockfd_seq_stop,
2843 .show = gfs2_glockfd_seq_show,
2844};
2845
2846static int gfs2_glockfd_open(struct inode *inode, struct file *file)
2847{
2848 struct gfs2_glockfd_iter *i;
2849 struct gfs2_sbd *sdp = inode->i_private;
2850
2851 i = __seq_open_private(file, &gfs2_glockfd_seq_ops,
2852 sizeof(struct gfs2_glockfd_iter));
2853 if (!i)
2854 return -ENOMEM;
2855 i->sb = sdp->sd_vfs;
2856 return 0;
2857}
2858
2859static const struct file_operations gfs2_glockfd_fops = {
2860 .owner = THIS_MODULE,
2861 .open = gfs2_glockfd_open,
2862 .read = seq_read,
2863 .llseek = seq_lseek,
2864 .release = seq_release_private,
2865};
2866
2867DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats);
2868
2869void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2870{
2871 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2872
2873 debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2874 &gfs2_glocks_fops);
2875
2876 debugfs_create_file("glockfd", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2877 &gfs2_glockfd_fops);
2878
2879 debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2880 &gfs2_glstats_fops);
2881
2882 debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2883 &gfs2_sbstats_fops);
2884}
2885
2886void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2887{
2888 debugfs_remove_recursive(sdp->debugfs_dir);
2889 sdp->debugfs_dir = NULL;
2890}
2891
2892void gfs2_register_debugfs(void)
2893{
2894 gfs2_root = debugfs_create_dir("gfs2", NULL);
2895}
2896
2897void gfs2_unregister_debugfs(void)
2898{
2899 debugfs_remove(gfs2_root);
2900 gfs2_root = NULL;
2901}