Loading...
1/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
8 */
9
10#include <linux/sched.h>
11#include <linux/slab.h>
12#include <linux/spinlock.h>
13#include <linux/buffer_head.h>
14#include <linux/delay.h>
15#include <linux/sort.h>
16#include <linux/jhash.h>
17#include <linux/kallsyms.h>
18#include <linux/gfs2_ondisk.h>
19#include <linux/list.h>
20#include <linux/wait.h>
21#include <linux/module.h>
22#include <asm/uaccess.h>
23#include <linux/seq_file.h>
24#include <linux/debugfs.h>
25#include <linux/kthread.h>
26#include <linux/freezer.h>
27#include <linux/workqueue.h>
28#include <linux/jiffies.h>
29#include <linux/rcupdate.h>
30#include <linux/rculist_bl.h>
31#include <linux/bit_spinlock.h>
32
33#include "gfs2.h"
34#include "incore.h"
35#include "glock.h"
36#include "glops.h"
37#include "inode.h"
38#include "lops.h"
39#include "meta_io.h"
40#include "quota.h"
41#include "super.h"
42#include "util.h"
43#include "bmap.h"
44#define CREATE_TRACE_POINTS
45#include "trace_gfs2.h"
46
47struct gfs2_glock_iter {
48 int hash; /* hash bucket index */
49 struct gfs2_sbd *sdp; /* incore superblock */
50 struct gfs2_glock *gl; /* current glock struct */
51 char string[512]; /* scratch space */
52};
53
54typedef void (*glock_examiner) (struct gfs2_glock * gl);
55
56static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl);
57#define GLOCK_BUG_ON(gl,x) do { if (unlikely(x)) { __dump_glock(NULL, gl); BUG(); } } while(0)
58static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
59
60static struct dentry *gfs2_root;
61static struct workqueue_struct *glock_workqueue;
62struct workqueue_struct *gfs2_delete_workqueue;
63static LIST_HEAD(lru_list);
64static atomic_t lru_count = ATOMIC_INIT(0);
65static DEFINE_SPINLOCK(lru_lock);
66
67#define GFS2_GL_HASH_SHIFT 15
68#define GFS2_GL_HASH_SIZE (1 << GFS2_GL_HASH_SHIFT)
69#define GFS2_GL_HASH_MASK (GFS2_GL_HASH_SIZE - 1)
70
71static struct hlist_bl_head gl_hash_table[GFS2_GL_HASH_SIZE];
72static struct dentry *gfs2_root;
73
74/**
75 * gl_hash() - Turn glock number into hash bucket number
76 * @lock: The glock number
77 *
78 * Returns: The number of the corresponding hash bucket
79 */
80
81static unsigned int gl_hash(const struct gfs2_sbd *sdp,
82 const struct lm_lockname *name)
83{
84 unsigned int h;
85
86 h = jhash(&name->ln_number, sizeof(u64), 0);
87 h = jhash(&name->ln_type, sizeof(unsigned int), h);
88 h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
89 h &= GFS2_GL_HASH_MASK;
90
91 return h;
92}
93
94static inline void spin_lock_bucket(unsigned int hash)
95{
96 hlist_bl_lock(&gl_hash_table[hash]);
97}
98
99static inline void spin_unlock_bucket(unsigned int hash)
100{
101 hlist_bl_unlock(&gl_hash_table[hash]);
102}
103
104static void gfs2_glock_dealloc(struct rcu_head *rcu)
105{
106 struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
107
108 if (gl->gl_ops->go_flags & GLOF_ASPACE)
109 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
110 else
111 kmem_cache_free(gfs2_glock_cachep, gl);
112}
113
114void gfs2_glock_free(struct gfs2_glock *gl)
115{
116 struct gfs2_sbd *sdp = gl->gl_sbd;
117
118 call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
119 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
120 wake_up(&sdp->sd_glock_wait);
121}
122
123/**
124 * gfs2_glock_hold() - increment reference count on glock
125 * @gl: The glock to hold
126 *
127 */
128
129void gfs2_glock_hold(struct gfs2_glock *gl)
130{
131 GLOCK_BUG_ON(gl, atomic_read(&gl->gl_ref) == 0);
132 atomic_inc(&gl->gl_ref);
133}
134
135/**
136 * demote_ok - Check to see if it's ok to unlock a glock
137 * @gl: the glock
138 *
139 * Returns: 1 if it's ok
140 */
141
142static int demote_ok(const struct gfs2_glock *gl)
143{
144 const struct gfs2_glock_operations *glops = gl->gl_ops;
145
146 if (gl->gl_state == LM_ST_UNLOCKED)
147 return 0;
148 if (!list_empty(&gl->gl_holders))
149 return 0;
150 if (glops->go_demote_ok)
151 return glops->go_demote_ok(gl);
152 return 1;
153}
154
155
156void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
157{
158 spin_lock(&lru_lock);
159
160 if (!list_empty(&gl->gl_lru))
161 list_del_init(&gl->gl_lru);
162 else
163 atomic_inc(&lru_count);
164
165 list_add_tail(&gl->gl_lru, &lru_list);
166 set_bit(GLF_LRU, &gl->gl_flags);
167 spin_unlock(&lru_lock);
168}
169
170static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
171{
172 spin_lock(&lru_lock);
173 if (!list_empty(&gl->gl_lru)) {
174 list_del_init(&gl->gl_lru);
175 atomic_dec(&lru_count);
176 clear_bit(GLF_LRU, &gl->gl_flags);
177 }
178 spin_unlock(&lru_lock);
179}
180
181/**
182 * __gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
183 * @gl: the glock
184 *
185 * If the glock is demotable, then we add it (or move it) to the end
186 * of the glock LRU list.
187 */
188
189static void __gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
190{
191 if (demote_ok(gl))
192 gfs2_glock_add_to_lru(gl);
193}
194
195/**
196 * gfs2_glock_put_nolock() - Decrement reference count on glock
197 * @gl: The glock to put
198 *
199 * This function should only be used if the caller has its own reference
200 * to the glock, in addition to the one it is dropping.
201 */
202
203void gfs2_glock_put_nolock(struct gfs2_glock *gl)
204{
205 if (atomic_dec_and_test(&gl->gl_ref))
206 GLOCK_BUG_ON(gl, 1);
207}
208
209/**
210 * gfs2_glock_put() - Decrement reference count on glock
211 * @gl: The glock to put
212 *
213 */
214
215void gfs2_glock_put(struct gfs2_glock *gl)
216{
217 struct gfs2_sbd *sdp = gl->gl_sbd;
218 struct address_space *mapping = gfs2_glock2aspace(gl);
219
220 if (atomic_dec_and_test(&gl->gl_ref)) {
221 spin_lock_bucket(gl->gl_hash);
222 hlist_bl_del_rcu(&gl->gl_list);
223 spin_unlock_bucket(gl->gl_hash);
224 gfs2_glock_remove_from_lru(gl);
225 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
226 GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
227 trace_gfs2_glock_put(gl);
228 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
229 }
230}
231
232/**
233 * search_bucket() - Find struct gfs2_glock by lock number
234 * @bucket: the bucket to search
235 * @name: The lock name
236 *
237 * Returns: NULL, or the struct gfs2_glock with the requested number
238 */
239
240static struct gfs2_glock *search_bucket(unsigned int hash,
241 const struct gfs2_sbd *sdp,
242 const struct lm_lockname *name)
243{
244 struct gfs2_glock *gl;
245 struct hlist_bl_node *h;
246
247 hlist_bl_for_each_entry_rcu(gl, h, &gl_hash_table[hash], gl_list) {
248 if (!lm_name_equal(&gl->gl_name, name))
249 continue;
250 if (gl->gl_sbd != sdp)
251 continue;
252 if (atomic_inc_not_zero(&gl->gl_ref))
253 return gl;
254 }
255
256 return NULL;
257}
258
259/**
260 * may_grant - check if its ok to grant a new lock
261 * @gl: The glock
262 * @gh: The lock request which we wish to grant
263 *
264 * Returns: true if its ok to grant the lock
265 */
266
267static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
268{
269 const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
270 if ((gh->gh_state == LM_ST_EXCLUSIVE ||
271 gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
272 return 0;
273 if (gl->gl_state == gh->gh_state)
274 return 1;
275 if (gh->gh_flags & GL_EXACT)
276 return 0;
277 if (gl->gl_state == LM_ST_EXCLUSIVE) {
278 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
279 return 1;
280 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
281 return 1;
282 }
283 if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
284 return 1;
285 return 0;
286}
287
288static void gfs2_holder_wake(struct gfs2_holder *gh)
289{
290 clear_bit(HIF_WAIT, &gh->gh_iflags);
291 smp_mb__after_clear_bit();
292 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
293}
294
295/**
296 * do_error - Something unexpected has happened during a lock request
297 *
298 */
299
300static inline void do_error(struct gfs2_glock *gl, const int ret)
301{
302 struct gfs2_holder *gh, *tmp;
303
304 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
305 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
306 continue;
307 if (ret & LM_OUT_ERROR)
308 gh->gh_error = -EIO;
309 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
310 gh->gh_error = GLR_TRYFAILED;
311 else
312 continue;
313 list_del_init(&gh->gh_list);
314 trace_gfs2_glock_queue(gh, 0);
315 gfs2_holder_wake(gh);
316 }
317}
318
319/**
320 * do_promote - promote as many requests as possible on the current queue
321 * @gl: The glock
322 *
323 * Returns: 1 if there is a blocked holder at the head of the list, or 2
324 * if a type specific operation is underway.
325 */
326
327static int do_promote(struct gfs2_glock *gl)
328__releases(&gl->gl_spin)
329__acquires(&gl->gl_spin)
330{
331 const struct gfs2_glock_operations *glops = gl->gl_ops;
332 struct gfs2_holder *gh, *tmp;
333 int ret;
334
335restart:
336 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
337 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
338 continue;
339 if (may_grant(gl, gh)) {
340 if (gh->gh_list.prev == &gl->gl_holders &&
341 glops->go_lock) {
342 spin_unlock(&gl->gl_spin);
343 /* FIXME: eliminate this eventually */
344 ret = glops->go_lock(gh);
345 spin_lock(&gl->gl_spin);
346 if (ret) {
347 if (ret == 1)
348 return 2;
349 gh->gh_error = ret;
350 list_del_init(&gh->gh_list);
351 trace_gfs2_glock_queue(gh, 0);
352 gfs2_holder_wake(gh);
353 goto restart;
354 }
355 set_bit(HIF_HOLDER, &gh->gh_iflags);
356 trace_gfs2_promote(gh, 1);
357 gfs2_holder_wake(gh);
358 goto restart;
359 }
360 set_bit(HIF_HOLDER, &gh->gh_iflags);
361 trace_gfs2_promote(gh, 0);
362 gfs2_holder_wake(gh);
363 continue;
364 }
365 if (gh->gh_list.prev == &gl->gl_holders)
366 return 1;
367 do_error(gl, 0);
368 break;
369 }
370 return 0;
371}
372
373/**
374 * find_first_waiter - find the first gh that's waiting for the glock
375 * @gl: the glock
376 */
377
378static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
379{
380 struct gfs2_holder *gh;
381
382 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
383 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
384 return gh;
385 }
386 return NULL;
387}
388
389/**
390 * state_change - record that the glock is now in a different state
391 * @gl: the glock
392 * @new_state the new state
393 *
394 */
395
396static void state_change(struct gfs2_glock *gl, unsigned int new_state)
397{
398 int held1, held2;
399
400 held1 = (gl->gl_state != LM_ST_UNLOCKED);
401 held2 = (new_state != LM_ST_UNLOCKED);
402
403 if (held1 != held2) {
404 if (held2)
405 gfs2_glock_hold(gl);
406 else
407 gfs2_glock_put_nolock(gl);
408 }
409 if (held1 && held2 && list_empty(&gl->gl_holders))
410 clear_bit(GLF_QUEUED, &gl->gl_flags);
411
412 if (new_state != gl->gl_target)
413 /* shorten our minimum hold time */
414 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
415 GL_GLOCK_MIN_HOLD);
416 gl->gl_state = new_state;
417 gl->gl_tchange = jiffies;
418}
419
420static void gfs2_demote_wake(struct gfs2_glock *gl)
421{
422 gl->gl_demote_state = LM_ST_EXCLUSIVE;
423 clear_bit(GLF_DEMOTE, &gl->gl_flags);
424 smp_mb__after_clear_bit();
425 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
426}
427
428/**
429 * finish_xmote - The DLM has replied to one of our lock requests
430 * @gl: The glock
431 * @ret: The status from the DLM
432 *
433 */
434
435static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
436{
437 const struct gfs2_glock_operations *glops = gl->gl_ops;
438 struct gfs2_holder *gh;
439 unsigned state = ret & LM_OUT_ST_MASK;
440 int rv;
441
442 spin_lock(&gl->gl_spin);
443 trace_gfs2_glock_state_change(gl, state);
444 state_change(gl, state);
445 gh = find_first_waiter(gl);
446
447 /* Demote to UN request arrived during demote to SH or DF */
448 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
449 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
450 gl->gl_target = LM_ST_UNLOCKED;
451
452 /* Check for state != intended state */
453 if (unlikely(state != gl->gl_target)) {
454 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
455 /* move to back of queue and try next entry */
456 if (ret & LM_OUT_CANCELED) {
457 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
458 list_move_tail(&gh->gh_list, &gl->gl_holders);
459 gh = find_first_waiter(gl);
460 gl->gl_target = gh->gh_state;
461 goto retry;
462 }
463 /* Some error or failed "try lock" - report it */
464 if ((ret & LM_OUT_ERROR) ||
465 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
466 gl->gl_target = gl->gl_state;
467 do_error(gl, ret);
468 goto out;
469 }
470 }
471 switch(state) {
472 /* Unlocked due to conversion deadlock, try again */
473 case LM_ST_UNLOCKED:
474retry:
475 do_xmote(gl, gh, gl->gl_target);
476 break;
477 /* Conversion fails, unlock and try again */
478 case LM_ST_SHARED:
479 case LM_ST_DEFERRED:
480 do_xmote(gl, gh, LM_ST_UNLOCKED);
481 break;
482 default: /* Everything else */
483 printk(KERN_ERR "GFS2: wanted %u got %u\n", gl->gl_target, state);
484 GLOCK_BUG_ON(gl, 1);
485 }
486 spin_unlock(&gl->gl_spin);
487 return;
488 }
489
490 /* Fast path - we got what we asked for */
491 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
492 gfs2_demote_wake(gl);
493 if (state != LM_ST_UNLOCKED) {
494 if (glops->go_xmote_bh) {
495 spin_unlock(&gl->gl_spin);
496 rv = glops->go_xmote_bh(gl, gh);
497 spin_lock(&gl->gl_spin);
498 if (rv) {
499 do_error(gl, rv);
500 goto out;
501 }
502 }
503 rv = do_promote(gl);
504 if (rv == 2)
505 goto out_locked;
506 }
507out:
508 clear_bit(GLF_LOCK, &gl->gl_flags);
509out_locked:
510 spin_unlock(&gl->gl_spin);
511}
512
513/**
514 * do_xmote - Calls the DLM to change the state of a lock
515 * @gl: The lock state
516 * @gh: The holder (only for promotes)
517 * @target: The target lock state
518 *
519 */
520
521static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
522__releases(&gl->gl_spin)
523__acquires(&gl->gl_spin)
524{
525 const struct gfs2_glock_operations *glops = gl->gl_ops;
526 struct gfs2_sbd *sdp = gl->gl_sbd;
527 unsigned int lck_flags = gh ? gh->gh_flags : 0;
528 int ret;
529
530 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
531 LM_FLAG_PRIORITY);
532 GLOCK_BUG_ON(gl, gl->gl_state == target);
533 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
534 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
535 glops->go_inval) {
536 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
537 do_error(gl, 0); /* Fail queued try locks */
538 }
539 gl->gl_req = target;
540 spin_unlock(&gl->gl_spin);
541 if (glops->go_xmote_th)
542 glops->go_xmote_th(gl);
543 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
544 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
545 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
546
547 gfs2_glock_hold(gl);
548 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
549 /* lock_dlm */
550 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
551 GLOCK_BUG_ON(gl, ret);
552 } else { /* lock_nolock */
553 finish_xmote(gl, target);
554 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
555 gfs2_glock_put(gl);
556 }
557
558 spin_lock(&gl->gl_spin);
559}
560
561/**
562 * find_first_holder - find the first "holder" gh
563 * @gl: the glock
564 */
565
566static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
567{
568 struct gfs2_holder *gh;
569
570 if (!list_empty(&gl->gl_holders)) {
571 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
572 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
573 return gh;
574 }
575 return NULL;
576}
577
578/**
579 * run_queue - do all outstanding tasks related to a glock
580 * @gl: The glock in question
581 * @nonblock: True if we must not block in run_queue
582 *
583 */
584
585static void run_queue(struct gfs2_glock *gl, const int nonblock)
586__releases(&gl->gl_spin)
587__acquires(&gl->gl_spin)
588{
589 struct gfs2_holder *gh = NULL;
590 int ret;
591
592 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
593 return;
594
595 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
596
597 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
598 gl->gl_demote_state != gl->gl_state) {
599 if (find_first_holder(gl))
600 goto out_unlock;
601 if (nonblock)
602 goto out_sched;
603 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
604 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
605 gl->gl_target = gl->gl_demote_state;
606 } else {
607 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
608 gfs2_demote_wake(gl);
609 ret = do_promote(gl);
610 if (ret == 0)
611 goto out_unlock;
612 if (ret == 2)
613 goto out;
614 gh = find_first_waiter(gl);
615 gl->gl_target = gh->gh_state;
616 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
617 do_error(gl, 0); /* Fail queued try locks */
618 }
619 do_xmote(gl, gh, gl->gl_target);
620out:
621 return;
622
623out_sched:
624 clear_bit(GLF_LOCK, &gl->gl_flags);
625 smp_mb__after_clear_bit();
626 gfs2_glock_hold(gl);
627 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
628 gfs2_glock_put_nolock(gl);
629 return;
630
631out_unlock:
632 clear_bit(GLF_LOCK, &gl->gl_flags);
633 smp_mb__after_clear_bit();
634 return;
635}
636
637static void delete_work_func(struct work_struct *work)
638{
639 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
640 struct gfs2_sbd *sdp = gl->gl_sbd;
641 struct gfs2_inode *ip;
642 struct inode *inode;
643 u64 no_addr = gl->gl_name.ln_number;
644
645 ip = gl->gl_object;
646 /* Note: Unsafe to dereference ip as we don't hold right refs/locks */
647
648 if (ip)
649 inode = gfs2_ilookup(sdp->sd_vfs, no_addr, 1);
650 else
651 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
652 if (inode && !IS_ERR(inode)) {
653 d_prune_aliases(inode);
654 iput(inode);
655 }
656 gfs2_glock_put(gl);
657}
658
659static void glock_work_func(struct work_struct *work)
660{
661 unsigned long delay = 0;
662 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
663 int drop_ref = 0;
664
665 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
666 finish_xmote(gl, gl->gl_reply);
667 drop_ref = 1;
668 }
669 spin_lock(&gl->gl_spin);
670 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
671 gl->gl_state != LM_ST_UNLOCKED &&
672 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
673 unsigned long holdtime, now = jiffies;
674
675 holdtime = gl->gl_tchange + gl->gl_hold_time;
676 if (time_before(now, holdtime))
677 delay = holdtime - now;
678
679 if (!delay) {
680 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
681 set_bit(GLF_DEMOTE, &gl->gl_flags);
682 }
683 }
684 run_queue(gl, 0);
685 spin_unlock(&gl->gl_spin);
686 if (!delay)
687 gfs2_glock_put(gl);
688 else {
689 if (gl->gl_name.ln_type != LM_TYPE_INODE)
690 delay = 0;
691 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
692 gfs2_glock_put(gl);
693 }
694 if (drop_ref)
695 gfs2_glock_put(gl);
696}
697
698/**
699 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
700 * @sdp: The GFS2 superblock
701 * @number: the lock number
702 * @glops: The glock_operations to use
703 * @create: If 0, don't create the glock if it doesn't exist
704 * @glp: the glock is returned here
705 *
706 * This does not lock a glock, just finds/creates structures for one.
707 *
708 * Returns: errno
709 */
710
711int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
712 const struct gfs2_glock_operations *glops, int create,
713 struct gfs2_glock **glp)
714{
715 struct super_block *s = sdp->sd_vfs;
716 struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
717 struct gfs2_glock *gl, *tmp;
718 unsigned int hash = gl_hash(sdp, &name);
719 struct address_space *mapping;
720 struct kmem_cache *cachep;
721
722 rcu_read_lock();
723 gl = search_bucket(hash, sdp, &name);
724 rcu_read_unlock();
725
726 *glp = gl;
727 if (gl)
728 return 0;
729 if (!create)
730 return -ENOENT;
731
732 if (glops->go_flags & GLOF_ASPACE)
733 cachep = gfs2_glock_aspace_cachep;
734 else
735 cachep = gfs2_glock_cachep;
736 gl = kmem_cache_alloc(cachep, GFP_KERNEL);
737 if (!gl)
738 return -ENOMEM;
739
740 atomic_inc(&sdp->sd_glock_disposal);
741 gl->gl_flags = 0;
742 gl->gl_name = name;
743 atomic_set(&gl->gl_ref, 1);
744 gl->gl_state = LM_ST_UNLOCKED;
745 gl->gl_target = LM_ST_UNLOCKED;
746 gl->gl_demote_state = LM_ST_EXCLUSIVE;
747 gl->gl_hash = hash;
748 gl->gl_ops = glops;
749 snprintf(gl->gl_strname, GDLM_STRNAME_BYTES, "%8x%16llx", name.ln_type, (unsigned long long)number);
750 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
751 gl->gl_lksb.sb_lvbptr = gl->gl_lvb;
752 gl->gl_tchange = jiffies;
753 gl->gl_object = NULL;
754 gl->gl_sbd = sdp;
755 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
756 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
757 INIT_WORK(&gl->gl_delete, delete_work_func);
758
759 mapping = gfs2_glock2aspace(gl);
760 if (mapping) {
761 mapping->a_ops = &gfs2_meta_aops;
762 mapping->host = s->s_bdev->bd_inode;
763 mapping->flags = 0;
764 mapping_set_gfp_mask(mapping, GFP_NOFS);
765 mapping->assoc_mapping = NULL;
766 mapping->backing_dev_info = s->s_bdi;
767 mapping->writeback_index = 0;
768 }
769
770 spin_lock_bucket(hash);
771 tmp = search_bucket(hash, sdp, &name);
772 if (tmp) {
773 spin_unlock_bucket(hash);
774 kmem_cache_free(cachep, gl);
775 atomic_dec(&sdp->sd_glock_disposal);
776 gl = tmp;
777 } else {
778 hlist_bl_add_head_rcu(&gl->gl_list, &gl_hash_table[hash]);
779 spin_unlock_bucket(hash);
780 }
781
782 *glp = gl;
783
784 return 0;
785}
786
787/**
788 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
789 * @gl: the glock
790 * @state: the state we're requesting
791 * @flags: the modifier flags
792 * @gh: the holder structure
793 *
794 */
795
796void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
797 struct gfs2_holder *gh)
798{
799 INIT_LIST_HEAD(&gh->gh_list);
800 gh->gh_gl = gl;
801 gh->gh_ip = (unsigned long)__builtin_return_address(0);
802 gh->gh_owner_pid = get_pid(task_pid(current));
803 gh->gh_state = state;
804 gh->gh_flags = flags;
805 gh->gh_error = 0;
806 gh->gh_iflags = 0;
807 gfs2_glock_hold(gl);
808}
809
810/**
811 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
812 * @state: the state we're requesting
813 * @flags: the modifier flags
814 * @gh: the holder structure
815 *
816 * Don't mess with the glock.
817 *
818 */
819
820void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
821{
822 gh->gh_state = state;
823 gh->gh_flags = flags;
824 gh->gh_iflags = 0;
825 gh->gh_ip = (unsigned long)__builtin_return_address(0);
826 if (gh->gh_owner_pid)
827 put_pid(gh->gh_owner_pid);
828 gh->gh_owner_pid = get_pid(task_pid(current));
829}
830
831/**
832 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
833 * @gh: the holder structure
834 *
835 */
836
837void gfs2_holder_uninit(struct gfs2_holder *gh)
838{
839 put_pid(gh->gh_owner_pid);
840 gfs2_glock_put(gh->gh_gl);
841 gh->gh_gl = NULL;
842 gh->gh_ip = 0;
843}
844
845/**
846 * gfs2_glock_holder_wait
847 * @word: unused
848 *
849 * This function and gfs2_glock_demote_wait both show up in the WCHAN
850 * field. Thus I've separated these otherwise identical functions in
851 * order to be more informative to the user.
852 */
853
854static int gfs2_glock_holder_wait(void *word)
855{
856 schedule();
857 return 0;
858}
859
860static int gfs2_glock_demote_wait(void *word)
861{
862 schedule();
863 return 0;
864}
865
866static void wait_on_holder(struct gfs2_holder *gh)
867{
868 unsigned long time1 = jiffies;
869
870 might_sleep();
871 wait_on_bit(&gh->gh_iflags, HIF_WAIT, gfs2_glock_holder_wait, TASK_UNINTERRUPTIBLE);
872 if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
873 /* Lengthen the minimum hold time. */
874 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
875 GL_GLOCK_HOLD_INCR,
876 GL_GLOCK_MAX_HOLD);
877}
878
879static void wait_on_demote(struct gfs2_glock *gl)
880{
881 might_sleep();
882 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, gfs2_glock_demote_wait, TASK_UNINTERRUPTIBLE);
883}
884
885/**
886 * handle_callback - process a demote request
887 * @gl: the glock
888 * @state: the state the caller wants us to change to
889 *
890 * There are only two requests that we are going to see in actual
891 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
892 */
893
894static void handle_callback(struct gfs2_glock *gl, unsigned int state,
895 unsigned long delay)
896{
897 int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
898
899 set_bit(bit, &gl->gl_flags);
900 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
901 gl->gl_demote_state = state;
902 gl->gl_demote_time = jiffies;
903 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
904 gl->gl_demote_state != state) {
905 gl->gl_demote_state = LM_ST_UNLOCKED;
906 }
907 if (gl->gl_ops->go_callback)
908 gl->gl_ops->go_callback(gl);
909 trace_gfs2_demote_rq(gl);
910}
911
912/**
913 * gfs2_glock_wait - wait on a glock acquisition
914 * @gh: the glock holder
915 *
916 * Returns: 0 on success
917 */
918
919int gfs2_glock_wait(struct gfs2_holder *gh)
920{
921 wait_on_holder(gh);
922 return gh->gh_error;
923}
924
925void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
926{
927 struct va_format vaf;
928 va_list args;
929
930 va_start(args, fmt);
931
932 if (seq) {
933 struct gfs2_glock_iter *gi = seq->private;
934 vsprintf(gi->string, fmt, args);
935 seq_printf(seq, gi->string);
936 } else {
937 vaf.fmt = fmt;
938 vaf.va = &args;
939
940 printk(KERN_ERR " %pV", &vaf);
941 }
942
943 va_end(args);
944}
945
946/**
947 * add_to_queue - Add a holder to the wait queue (but look for recursion)
948 * @gh: the holder structure to add
949 *
950 * Eventually we should move the recursive locking trap to a
951 * debugging option or something like that. This is the fast
952 * path and needs to have the minimum number of distractions.
953 *
954 */
955
956static inline void add_to_queue(struct gfs2_holder *gh)
957__releases(&gl->gl_spin)
958__acquires(&gl->gl_spin)
959{
960 struct gfs2_glock *gl = gh->gh_gl;
961 struct gfs2_sbd *sdp = gl->gl_sbd;
962 struct list_head *insert_pt = NULL;
963 struct gfs2_holder *gh2;
964 int try_lock = 0;
965
966 BUG_ON(gh->gh_owner_pid == NULL);
967 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
968 BUG();
969
970 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
971 if (test_bit(GLF_LOCK, &gl->gl_flags))
972 try_lock = 1;
973 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
974 goto fail;
975 }
976
977 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
978 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
979 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
980 goto trap_recursive;
981 if (try_lock &&
982 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) &&
983 !may_grant(gl, gh)) {
984fail:
985 gh->gh_error = GLR_TRYFAILED;
986 gfs2_holder_wake(gh);
987 return;
988 }
989 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
990 continue;
991 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
992 insert_pt = &gh2->gh_list;
993 }
994 set_bit(GLF_QUEUED, &gl->gl_flags);
995 trace_gfs2_glock_queue(gh, 1);
996 if (likely(insert_pt == NULL)) {
997 list_add_tail(&gh->gh_list, &gl->gl_holders);
998 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
999 goto do_cancel;
1000 return;
1001 }
1002 list_add_tail(&gh->gh_list, insert_pt);
1003do_cancel:
1004 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
1005 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1006 spin_unlock(&gl->gl_spin);
1007 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1008 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1009 spin_lock(&gl->gl_spin);
1010 }
1011 return;
1012
1013trap_recursive:
1014 print_symbol(KERN_ERR "original: %s\n", gh2->gh_ip);
1015 printk(KERN_ERR "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1016 printk(KERN_ERR "lock type: %d req lock state : %d\n",
1017 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1018 print_symbol(KERN_ERR "new: %s\n", gh->gh_ip);
1019 printk(KERN_ERR "pid: %d\n", pid_nr(gh->gh_owner_pid));
1020 printk(KERN_ERR "lock type: %d req lock state : %d\n",
1021 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1022 __dump_glock(NULL, gl);
1023 BUG();
1024}
1025
1026/**
1027 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1028 * @gh: the holder structure
1029 *
1030 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1031 *
1032 * Returns: 0, GLR_TRYFAILED, or errno on failure
1033 */
1034
1035int gfs2_glock_nq(struct gfs2_holder *gh)
1036{
1037 struct gfs2_glock *gl = gh->gh_gl;
1038 struct gfs2_sbd *sdp = gl->gl_sbd;
1039 int error = 0;
1040
1041 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1042 return -EIO;
1043
1044 if (test_bit(GLF_LRU, &gl->gl_flags))
1045 gfs2_glock_remove_from_lru(gl);
1046
1047 spin_lock(&gl->gl_spin);
1048 add_to_queue(gh);
1049 if ((LM_FLAG_NOEXP & gh->gh_flags) &&
1050 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1051 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1052 run_queue(gl, 1);
1053 spin_unlock(&gl->gl_spin);
1054
1055 if (!(gh->gh_flags & GL_ASYNC))
1056 error = gfs2_glock_wait(gh);
1057
1058 return error;
1059}
1060
1061/**
1062 * gfs2_glock_poll - poll to see if an async request has been completed
1063 * @gh: the holder
1064 *
1065 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1066 */
1067
1068int gfs2_glock_poll(struct gfs2_holder *gh)
1069{
1070 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1071}
1072
1073/**
1074 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1075 * @gh: the glock holder
1076 *
1077 */
1078
1079void gfs2_glock_dq(struct gfs2_holder *gh)
1080{
1081 struct gfs2_glock *gl = gh->gh_gl;
1082 const struct gfs2_glock_operations *glops = gl->gl_ops;
1083 unsigned delay = 0;
1084 int fast_path = 0;
1085
1086 spin_lock(&gl->gl_spin);
1087 if (gh->gh_flags & GL_NOCACHE)
1088 handle_callback(gl, LM_ST_UNLOCKED, 0);
1089
1090 list_del_init(&gh->gh_list);
1091 if (find_first_holder(gl) == NULL) {
1092 if (glops->go_unlock) {
1093 GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1094 spin_unlock(&gl->gl_spin);
1095 glops->go_unlock(gh);
1096 spin_lock(&gl->gl_spin);
1097 clear_bit(GLF_LOCK, &gl->gl_flags);
1098 }
1099 if (list_empty(&gl->gl_holders) &&
1100 !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1101 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1102 fast_path = 1;
1103 }
1104 if (!test_bit(GLF_LFLUSH, &gl->gl_flags))
1105 __gfs2_glock_schedule_for_reclaim(gl);
1106 trace_gfs2_glock_queue(gh, 0);
1107 spin_unlock(&gl->gl_spin);
1108 if (likely(fast_path))
1109 return;
1110
1111 gfs2_glock_hold(gl);
1112 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1113 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1114 gl->gl_name.ln_type == LM_TYPE_INODE)
1115 delay = gl->gl_hold_time;
1116 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1117 gfs2_glock_put(gl);
1118}
1119
1120void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1121{
1122 struct gfs2_glock *gl = gh->gh_gl;
1123 gfs2_glock_dq(gh);
1124 wait_on_demote(gl);
1125}
1126
1127/**
1128 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1129 * @gh: the holder structure
1130 *
1131 */
1132
1133void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1134{
1135 gfs2_glock_dq(gh);
1136 gfs2_holder_uninit(gh);
1137}
1138
1139/**
1140 * gfs2_glock_nq_num - acquire a glock based on lock number
1141 * @sdp: the filesystem
1142 * @number: the lock number
1143 * @glops: the glock operations for the type of glock
1144 * @state: the state to acquire the glock in
1145 * @flags: modifier flags for the acquisition
1146 * @gh: the struct gfs2_holder
1147 *
1148 * Returns: errno
1149 */
1150
1151int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1152 const struct gfs2_glock_operations *glops,
1153 unsigned int state, int flags, struct gfs2_holder *gh)
1154{
1155 struct gfs2_glock *gl;
1156 int error;
1157
1158 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1159 if (!error) {
1160 error = gfs2_glock_nq_init(gl, state, flags, gh);
1161 gfs2_glock_put(gl);
1162 }
1163
1164 return error;
1165}
1166
1167/**
1168 * glock_compare - Compare two struct gfs2_glock structures for sorting
1169 * @arg_a: the first structure
1170 * @arg_b: the second structure
1171 *
1172 */
1173
1174static int glock_compare(const void *arg_a, const void *arg_b)
1175{
1176 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1177 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1178 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1179 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1180
1181 if (a->ln_number > b->ln_number)
1182 return 1;
1183 if (a->ln_number < b->ln_number)
1184 return -1;
1185 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1186 return 0;
1187}
1188
1189/**
1190 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1191 * @num_gh: the number of structures
1192 * @ghs: an array of struct gfs2_holder structures
1193 *
1194 * Returns: 0 on success (all glocks acquired),
1195 * errno on failure (no glocks acquired)
1196 */
1197
1198static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1199 struct gfs2_holder **p)
1200{
1201 unsigned int x;
1202 int error = 0;
1203
1204 for (x = 0; x < num_gh; x++)
1205 p[x] = &ghs[x];
1206
1207 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1208
1209 for (x = 0; x < num_gh; x++) {
1210 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1211
1212 error = gfs2_glock_nq(p[x]);
1213 if (error) {
1214 while (x--)
1215 gfs2_glock_dq(p[x]);
1216 break;
1217 }
1218 }
1219
1220 return error;
1221}
1222
1223/**
1224 * gfs2_glock_nq_m - acquire multiple glocks
1225 * @num_gh: the number of structures
1226 * @ghs: an array of struct gfs2_holder structures
1227 *
1228 *
1229 * Returns: 0 on success (all glocks acquired),
1230 * errno on failure (no glocks acquired)
1231 */
1232
1233int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1234{
1235 struct gfs2_holder *tmp[4];
1236 struct gfs2_holder **pph = tmp;
1237 int error = 0;
1238
1239 switch(num_gh) {
1240 case 0:
1241 return 0;
1242 case 1:
1243 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1244 return gfs2_glock_nq(ghs);
1245 default:
1246 if (num_gh <= 4)
1247 break;
1248 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1249 if (!pph)
1250 return -ENOMEM;
1251 }
1252
1253 error = nq_m_sync(num_gh, ghs, pph);
1254
1255 if (pph != tmp)
1256 kfree(pph);
1257
1258 return error;
1259}
1260
1261/**
1262 * gfs2_glock_dq_m - release multiple glocks
1263 * @num_gh: the number of structures
1264 * @ghs: an array of struct gfs2_holder structures
1265 *
1266 */
1267
1268void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1269{
1270 while (num_gh--)
1271 gfs2_glock_dq(&ghs[num_gh]);
1272}
1273
1274/**
1275 * gfs2_glock_dq_uninit_m - release multiple glocks
1276 * @num_gh: the number of structures
1277 * @ghs: an array of struct gfs2_holder structures
1278 *
1279 */
1280
1281void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1282{
1283 while (num_gh--)
1284 gfs2_glock_dq_uninit(&ghs[num_gh]);
1285}
1286
1287void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1288{
1289 unsigned long delay = 0;
1290 unsigned long holdtime;
1291 unsigned long now = jiffies;
1292
1293 gfs2_glock_hold(gl);
1294 holdtime = gl->gl_tchange + gl->gl_hold_time;
1295 if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1296 gl->gl_name.ln_type == LM_TYPE_INODE) {
1297 if (time_before(now, holdtime))
1298 delay = holdtime - now;
1299 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1300 delay = gl->gl_hold_time;
1301 }
1302
1303 spin_lock(&gl->gl_spin);
1304 handle_callback(gl, state, delay);
1305 spin_unlock(&gl->gl_spin);
1306 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1307 gfs2_glock_put(gl);
1308}
1309
1310/**
1311 * gfs2_should_freeze - Figure out if glock should be frozen
1312 * @gl: The glock in question
1313 *
1314 * Glocks are not frozen if (a) the result of the dlm operation is
1315 * an error, (b) the locking operation was an unlock operation or
1316 * (c) if there is a "noexp" flagged request anywhere in the queue
1317 *
1318 * Returns: 1 if freezing should occur, 0 otherwise
1319 */
1320
1321static int gfs2_should_freeze(const struct gfs2_glock *gl)
1322{
1323 const struct gfs2_holder *gh;
1324
1325 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1326 return 0;
1327 if (gl->gl_target == LM_ST_UNLOCKED)
1328 return 0;
1329
1330 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1331 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1332 continue;
1333 if (LM_FLAG_NOEXP & gh->gh_flags)
1334 return 0;
1335 }
1336
1337 return 1;
1338}
1339
1340/**
1341 * gfs2_glock_complete - Callback used by locking
1342 * @gl: Pointer to the glock
1343 * @ret: The return value from the dlm
1344 *
1345 * The gl_reply field is under the gl_spin lock so that it is ok
1346 * to use a bitfield shared with other glock state fields.
1347 */
1348
1349void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1350{
1351 struct lm_lockstruct *ls = &gl->gl_sbd->sd_lockstruct;
1352
1353 spin_lock(&gl->gl_spin);
1354 gl->gl_reply = ret;
1355
1356 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_flags))) {
1357 if (gfs2_should_freeze(gl)) {
1358 set_bit(GLF_FROZEN, &gl->gl_flags);
1359 spin_unlock(&gl->gl_spin);
1360 return;
1361 }
1362 }
1363
1364 spin_unlock(&gl->gl_spin);
1365 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1366 smp_wmb();
1367 gfs2_glock_hold(gl);
1368 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1369 gfs2_glock_put(gl);
1370}
1371
1372
1373static int gfs2_shrink_glock_memory(struct shrinker *shrink,
1374 struct shrink_control *sc)
1375{
1376 struct gfs2_glock *gl;
1377 int may_demote;
1378 int nr_skipped = 0;
1379 int nr = sc->nr_to_scan;
1380 gfp_t gfp_mask = sc->gfp_mask;
1381 LIST_HEAD(skipped);
1382
1383 if (nr == 0)
1384 goto out;
1385
1386 if (!(gfp_mask & __GFP_FS))
1387 return -1;
1388
1389 spin_lock(&lru_lock);
1390 while(nr && !list_empty(&lru_list)) {
1391 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1392 list_del_init(&gl->gl_lru);
1393 clear_bit(GLF_LRU, &gl->gl_flags);
1394 atomic_dec(&lru_count);
1395
1396 /* Test for being demotable */
1397 if (!test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1398 gfs2_glock_hold(gl);
1399 spin_unlock(&lru_lock);
1400 spin_lock(&gl->gl_spin);
1401 may_demote = demote_ok(gl);
1402 if (may_demote) {
1403 handle_callback(gl, LM_ST_UNLOCKED, 0);
1404 nr--;
1405 }
1406 clear_bit(GLF_LOCK, &gl->gl_flags);
1407 smp_mb__after_clear_bit();
1408 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1409 gfs2_glock_put_nolock(gl);
1410 spin_unlock(&gl->gl_spin);
1411 spin_lock(&lru_lock);
1412 continue;
1413 }
1414 nr_skipped++;
1415 list_add(&gl->gl_lru, &skipped);
1416 set_bit(GLF_LRU, &gl->gl_flags);
1417 }
1418 list_splice(&skipped, &lru_list);
1419 atomic_add(nr_skipped, &lru_count);
1420 spin_unlock(&lru_lock);
1421out:
1422 return (atomic_read(&lru_count) / 100) * sysctl_vfs_cache_pressure;
1423}
1424
1425static struct shrinker glock_shrinker = {
1426 .shrink = gfs2_shrink_glock_memory,
1427 .seeks = DEFAULT_SEEKS,
1428};
1429
1430/**
1431 * examine_bucket - Call a function for glock in a hash bucket
1432 * @examiner: the function
1433 * @sdp: the filesystem
1434 * @bucket: the bucket
1435 *
1436 */
1437
1438static void examine_bucket(glock_examiner examiner, const struct gfs2_sbd *sdp,
1439 unsigned int hash)
1440{
1441 struct gfs2_glock *gl;
1442 struct hlist_bl_head *head = &gl_hash_table[hash];
1443 struct hlist_bl_node *pos;
1444
1445 rcu_read_lock();
1446 hlist_bl_for_each_entry_rcu(gl, pos, head, gl_list) {
1447 if ((gl->gl_sbd == sdp) && atomic_read(&gl->gl_ref))
1448 examiner(gl);
1449 }
1450 rcu_read_unlock();
1451 cond_resched();
1452}
1453
1454static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1455{
1456 unsigned x;
1457
1458 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1459 examine_bucket(examiner, sdp, x);
1460}
1461
1462
1463/**
1464 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1465 * @gl: The glock to thaw
1466 *
1467 * N.B. When we freeze a glock, we leave a ref to the glock outstanding,
1468 * so this has to result in the ref count being dropped by one.
1469 */
1470
1471static void thaw_glock(struct gfs2_glock *gl)
1472{
1473 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1474 return;
1475 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1476 gfs2_glock_hold(gl);
1477 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1478 gfs2_glock_put(gl);
1479}
1480
1481/**
1482 * clear_glock - look at a glock and see if we can free it from glock cache
1483 * @gl: the glock to look at
1484 *
1485 */
1486
1487static void clear_glock(struct gfs2_glock *gl)
1488{
1489 gfs2_glock_remove_from_lru(gl);
1490
1491 spin_lock(&gl->gl_spin);
1492 if (gl->gl_state != LM_ST_UNLOCKED)
1493 handle_callback(gl, LM_ST_UNLOCKED, 0);
1494 spin_unlock(&gl->gl_spin);
1495 gfs2_glock_hold(gl);
1496 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1497 gfs2_glock_put(gl);
1498}
1499
1500/**
1501 * gfs2_glock_thaw - Thaw any frozen glocks
1502 * @sdp: The super block
1503 *
1504 */
1505
1506void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1507{
1508 glock_hash_walk(thaw_glock, sdp);
1509}
1510
1511static int dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1512{
1513 int ret;
1514 spin_lock(&gl->gl_spin);
1515 ret = __dump_glock(seq, gl);
1516 spin_unlock(&gl->gl_spin);
1517 return ret;
1518}
1519
1520static void dump_glock_func(struct gfs2_glock *gl)
1521{
1522 dump_glock(NULL, gl);
1523}
1524
1525/**
1526 * gfs2_gl_hash_clear - Empty out the glock hash table
1527 * @sdp: the filesystem
1528 * @wait: wait until it's all gone
1529 *
1530 * Called when unmounting the filesystem.
1531 */
1532
1533void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1534{
1535 glock_hash_walk(clear_glock, sdp);
1536 flush_workqueue(glock_workqueue);
1537 wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
1538 glock_hash_walk(dump_glock_func, sdp);
1539}
1540
1541void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1542{
1543 struct gfs2_glock *gl = ip->i_gl;
1544 int ret;
1545
1546 ret = gfs2_truncatei_resume(ip);
1547 gfs2_assert_withdraw(gl->gl_sbd, ret == 0);
1548
1549 spin_lock(&gl->gl_spin);
1550 clear_bit(GLF_LOCK, &gl->gl_flags);
1551 run_queue(gl, 1);
1552 spin_unlock(&gl->gl_spin);
1553}
1554
1555static const char *state2str(unsigned state)
1556{
1557 switch(state) {
1558 case LM_ST_UNLOCKED:
1559 return "UN";
1560 case LM_ST_SHARED:
1561 return "SH";
1562 case LM_ST_DEFERRED:
1563 return "DF";
1564 case LM_ST_EXCLUSIVE:
1565 return "EX";
1566 }
1567 return "??";
1568}
1569
1570static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1571{
1572 char *p = buf;
1573 if (flags & LM_FLAG_TRY)
1574 *p++ = 't';
1575 if (flags & LM_FLAG_TRY_1CB)
1576 *p++ = 'T';
1577 if (flags & LM_FLAG_NOEXP)
1578 *p++ = 'e';
1579 if (flags & LM_FLAG_ANY)
1580 *p++ = 'A';
1581 if (flags & LM_FLAG_PRIORITY)
1582 *p++ = 'p';
1583 if (flags & GL_ASYNC)
1584 *p++ = 'a';
1585 if (flags & GL_EXACT)
1586 *p++ = 'E';
1587 if (flags & GL_NOCACHE)
1588 *p++ = 'c';
1589 if (test_bit(HIF_HOLDER, &iflags))
1590 *p++ = 'H';
1591 if (test_bit(HIF_WAIT, &iflags))
1592 *p++ = 'W';
1593 if (test_bit(HIF_FIRST, &iflags))
1594 *p++ = 'F';
1595 *p = 0;
1596 return buf;
1597}
1598
1599/**
1600 * dump_holder - print information about a glock holder
1601 * @seq: the seq_file struct
1602 * @gh: the glock holder
1603 *
1604 * Returns: 0 on success, -ENOBUFS when we run out of space
1605 */
1606
1607static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1608{
1609 struct task_struct *gh_owner = NULL;
1610 char flags_buf[32];
1611
1612 if (gh->gh_owner_pid)
1613 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1614 gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1615 state2str(gh->gh_state),
1616 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1617 gh->gh_error,
1618 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1619 gh_owner ? gh_owner->comm : "(ended)",
1620 (void *)gh->gh_ip);
1621 return 0;
1622}
1623
1624static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1625{
1626 const unsigned long *gflags = &gl->gl_flags;
1627 char *p = buf;
1628
1629 if (test_bit(GLF_LOCK, gflags))
1630 *p++ = 'l';
1631 if (test_bit(GLF_DEMOTE, gflags))
1632 *p++ = 'D';
1633 if (test_bit(GLF_PENDING_DEMOTE, gflags))
1634 *p++ = 'd';
1635 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1636 *p++ = 'p';
1637 if (test_bit(GLF_DIRTY, gflags))
1638 *p++ = 'y';
1639 if (test_bit(GLF_LFLUSH, gflags))
1640 *p++ = 'f';
1641 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1642 *p++ = 'i';
1643 if (test_bit(GLF_REPLY_PENDING, gflags))
1644 *p++ = 'r';
1645 if (test_bit(GLF_INITIAL, gflags))
1646 *p++ = 'I';
1647 if (test_bit(GLF_FROZEN, gflags))
1648 *p++ = 'F';
1649 if (test_bit(GLF_QUEUED, gflags))
1650 *p++ = 'q';
1651 if (test_bit(GLF_LRU, gflags))
1652 *p++ = 'L';
1653 if (gl->gl_object)
1654 *p++ = 'o';
1655 *p = 0;
1656 return buf;
1657}
1658
1659/**
1660 * __dump_glock - print information about a glock
1661 * @seq: The seq_file struct
1662 * @gl: the glock
1663 *
1664 * The file format is as follows:
1665 * One line per object, capital letters are used to indicate objects
1666 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1667 * other objects are indented by a single space and follow the glock to
1668 * which they are related. Fields are indicated by lower case letters
1669 * followed by a colon and the field value, except for strings which are in
1670 * [] so that its possible to see if they are composed of spaces for
1671 * example. The field's are n = number (id of the object), f = flags,
1672 * t = type, s = state, r = refcount, e = error, p = pid.
1673 *
1674 * Returns: 0 on success, -ENOBUFS when we run out of space
1675 */
1676
1677static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1678{
1679 const struct gfs2_glock_operations *glops = gl->gl_ops;
1680 unsigned long long dtime;
1681 const struct gfs2_holder *gh;
1682 char gflags_buf[32];
1683 int error = 0;
1684
1685 dtime = jiffies - gl->gl_demote_time;
1686 dtime *= 1000000/HZ; /* demote time in uSec */
1687 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1688 dtime = 0;
1689 gfs2_print_dbg(seq, "G: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d v:%d r:%d m:%ld\n",
1690 state2str(gl->gl_state),
1691 gl->gl_name.ln_type,
1692 (unsigned long long)gl->gl_name.ln_number,
1693 gflags2str(gflags_buf, gl),
1694 state2str(gl->gl_target),
1695 state2str(gl->gl_demote_state), dtime,
1696 atomic_read(&gl->gl_ail_count),
1697 atomic_read(&gl->gl_revokes),
1698 atomic_read(&gl->gl_ref), gl->gl_hold_time);
1699
1700 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1701 error = dump_holder(seq, gh);
1702 if (error)
1703 goto out;
1704 }
1705 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1706 error = glops->go_dump(seq, gl);
1707out:
1708 return error;
1709}
1710
1711
1712
1713
1714int __init gfs2_glock_init(void)
1715{
1716 unsigned i;
1717 for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
1718 INIT_HLIST_BL_HEAD(&gl_hash_table[i]);
1719 }
1720
1721 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1722 WQ_HIGHPRI | WQ_FREEZABLE, 0);
1723 if (IS_ERR(glock_workqueue))
1724 return PTR_ERR(glock_workqueue);
1725 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1726 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1727 0);
1728 if (IS_ERR(gfs2_delete_workqueue)) {
1729 destroy_workqueue(glock_workqueue);
1730 return PTR_ERR(gfs2_delete_workqueue);
1731 }
1732
1733 register_shrinker(&glock_shrinker);
1734
1735 return 0;
1736}
1737
1738void gfs2_glock_exit(void)
1739{
1740 unregister_shrinker(&glock_shrinker);
1741 destroy_workqueue(glock_workqueue);
1742 destroy_workqueue(gfs2_delete_workqueue);
1743}
1744
1745static inline struct gfs2_glock *glock_hash_chain(unsigned hash)
1746{
1747 return hlist_bl_entry(hlist_bl_first_rcu(&gl_hash_table[hash]),
1748 struct gfs2_glock, gl_list);
1749}
1750
1751static inline struct gfs2_glock *glock_hash_next(struct gfs2_glock *gl)
1752{
1753 return hlist_bl_entry(rcu_dereference(gl->gl_list.next),
1754 struct gfs2_glock, gl_list);
1755}
1756
1757static int gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1758{
1759 struct gfs2_glock *gl;
1760
1761 do {
1762 gl = gi->gl;
1763 if (gl) {
1764 gi->gl = glock_hash_next(gl);
1765 } else {
1766 gi->gl = glock_hash_chain(gi->hash);
1767 }
1768 while (gi->gl == NULL) {
1769 gi->hash++;
1770 if (gi->hash >= GFS2_GL_HASH_SIZE) {
1771 rcu_read_unlock();
1772 return 1;
1773 }
1774 gi->gl = glock_hash_chain(gi->hash);
1775 }
1776 /* Skip entries for other sb and dead entries */
1777 } while (gi->sdp != gi->gl->gl_sbd || atomic_read(&gi->gl->gl_ref) == 0);
1778
1779 return 0;
1780}
1781
1782static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1783{
1784 struct gfs2_glock_iter *gi = seq->private;
1785 loff_t n = *pos;
1786
1787 gi->hash = 0;
1788 rcu_read_lock();
1789
1790 do {
1791 if (gfs2_glock_iter_next(gi))
1792 return NULL;
1793 } while (n--);
1794
1795 return gi->gl;
1796}
1797
1798static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1799 loff_t *pos)
1800{
1801 struct gfs2_glock_iter *gi = seq->private;
1802
1803 (*pos)++;
1804
1805 if (gfs2_glock_iter_next(gi))
1806 return NULL;
1807
1808 return gi->gl;
1809}
1810
1811static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1812{
1813 struct gfs2_glock_iter *gi = seq->private;
1814
1815 if (gi->gl)
1816 rcu_read_unlock();
1817 gi->gl = NULL;
1818}
1819
1820static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1821{
1822 return dump_glock(seq, iter_ptr);
1823}
1824
1825static const struct seq_operations gfs2_glock_seq_ops = {
1826 .start = gfs2_glock_seq_start,
1827 .next = gfs2_glock_seq_next,
1828 .stop = gfs2_glock_seq_stop,
1829 .show = gfs2_glock_seq_show,
1830};
1831
1832static int gfs2_debugfs_open(struct inode *inode, struct file *file)
1833{
1834 int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1835 sizeof(struct gfs2_glock_iter));
1836 if (ret == 0) {
1837 struct seq_file *seq = file->private_data;
1838 struct gfs2_glock_iter *gi = seq->private;
1839 gi->sdp = inode->i_private;
1840 }
1841 return ret;
1842}
1843
1844static const struct file_operations gfs2_debug_fops = {
1845 .owner = THIS_MODULE,
1846 .open = gfs2_debugfs_open,
1847 .read = seq_read,
1848 .llseek = seq_lseek,
1849 .release = seq_release_private,
1850};
1851
1852int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
1853{
1854 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
1855 if (!sdp->debugfs_dir)
1856 return -ENOMEM;
1857 sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
1858 S_IFREG | S_IRUGO,
1859 sdp->debugfs_dir, sdp,
1860 &gfs2_debug_fops);
1861 if (!sdp->debugfs_dentry_glocks)
1862 return -ENOMEM;
1863
1864 return 0;
1865}
1866
1867void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
1868{
1869 if (sdp && sdp->debugfs_dir) {
1870 if (sdp->debugfs_dentry_glocks) {
1871 debugfs_remove(sdp->debugfs_dentry_glocks);
1872 sdp->debugfs_dentry_glocks = NULL;
1873 }
1874 debugfs_remove(sdp->debugfs_dir);
1875 sdp->debugfs_dir = NULL;
1876 }
1877}
1878
1879int gfs2_register_debugfs(void)
1880{
1881 gfs2_root = debugfs_create_dir("gfs2", NULL);
1882 return gfs2_root ? 0 : -ENOMEM;
1883}
1884
1885void gfs2_unregister_debugfs(void)
1886{
1887 debugfs_remove(gfs2_root);
1888 gfs2_root = NULL;
1889}
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 */
6
7#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
9#include <linux/sched.h>
10#include <linux/slab.h>
11#include <linux/spinlock.h>
12#include <linux/buffer_head.h>
13#include <linux/delay.h>
14#include <linux/sort.h>
15#include <linux/hash.h>
16#include <linux/jhash.h>
17#include <linux/kallsyms.h>
18#include <linux/gfs2_ondisk.h>
19#include <linux/list.h>
20#include <linux/wait.h>
21#include <linux/module.h>
22#include <linux/uaccess.h>
23#include <linux/seq_file.h>
24#include <linux/debugfs.h>
25#include <linux/kthread.h>
26#include <linux/freezer.h>
27#include <linux/workqueue.h>
28#include <linux/jiffies.h>
29#include <linux/rcupdate.h>
30#include <linux/rculist_bl.h>
31#include <linux/bit_spinlock.h>
32#include <linux/percpu.h>
33#include <linux/list_sort.h>
34#include <linux/lockref.h>
35#include <linux/rhashtable.h>
36#include <linux/pid_namespace.h>
37#include <linux/fdtable.h>
38#include <linux/file.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 __gfs2_glock_dq(struct gfs2_holder *gh);
65static void handle_callback(struct gfs2_glock *gl, unsigned int state,
66 unsigned long delay, bool remote);
67
68static struct dentry *gfs2_root;
69static struct workqueue_struct *glock_workqueue;
70static LIST_HEAD(lru_list);
71static atomic_t lru_count = ATOMIC_INIT(0);
72static DEFINE_SPINLOCK(lru_lock);
73
74#define GFS2_GL_HASH_SHIFT 15
75#define GFS2_GL_HASH_SIZE BIT(GFS2_GL_HASH_SHIFT)
76
77static const struct rhashtable_params ht_parms = {
78 .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4,
79 .key_len = offsetofend(struct lm_lockname, ln_type),
80 .key_offset = offsetof(struct gfs2_glock, gl_name),
81 .head_offset = offsetof(struct gfs2_glock, gl_node),
82};
83
84static struct rhashtable gl_hash_table;
85
86#define GLOCK_WAIT_TABLE_BITS 12
87#define GLOCK_WAIT_TABLE_SIZE (1 << GLOCK_WAIT_TABLE_BITS)
88static wait_queue_head_t glock_wait_table[GLOCK_WAIT_TABLE_SIZE] __cacheline_aligned;
89
90struct wait_glock_queue {
91 struct lm_lockname *name;
92 wait_queue_entry_t wait;
93};
94
95static int glock_wake_function(wait_queue_entry_t *wait, unsigned int mode,
96 int sync, void *key)
97{
98 struct wait_glock_queue *wait_glock =
99 container_of(wait, struct wait_glock_queue, wait);
100 struct lm_lockname *wait_name = wait_glock->name;
101 struct lm_lockname *wake_name = key;
102
103 if (wake_name->ln_sbd != wait_name->ln_sbd ||
104 wake_name->ln_number != wait_name->ln_number ||
105 wake_name->ln_type != wait_name->ln_type)
106 return 0;
107 return autoremove_wake_function(wait, mode, sync, key);
108}
109
110static wait_queue_head_t *glock_waitqueue(struct lm_lockname *name)
111{
112 u32 hash = jhash2((u32 *)name, ht_parms.key_len / 4, 0);
113
114 return glock_wait_table + hash_32(hash, GLOCK_WAIT_TABLE_BITS);
115}
116
117/**
118 * wake_up_glock - Wake up waiters on a glock
119 * @gl: the glock
120 */
121static void wake_up_glock(struct gfs2_glock *gl)
122{
123 wait_queue_head_t *wq = glock_waitqueue(&gl->gl_name);
124
125 if (waitqueue_active(wq))
126 __wake_up(wq, TASK_NORMAL, 1, &gl->gl_name);
127}
128
129static void gfs2_glock_dealloc(struct rcu_head *rcu)
130{
131 struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
132
133 kfree(gl->gl_lksb.sb_lvbptr);
134 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
135 struct gfs2_glock_aspace *gla =
136 container_of(gl, struct gfs2_glock_aspace, glock);
137 kmem_cache_free(gfs2_glock_aspace_cachep, gla);
138 } else
139 kmem_cache_free(gfs2_glock_cachep, gl);
140}
141
142/**
143 * glock_blocked_by_withdraw - determine if we can still use a glock
144 * @gl: the glock
145 *
146 * We need to allow some glocks to be enqueued, dequeued, promoted, and demoted
147 * when we're withdrawn. For example, to maintain metadata integrity, we should
148 * disallow the use of inode and rgrp glocks when withdrawn. Other glocks like
149 * the iopen or freeze glock may be safely used because none of their
150 * metadata goes through the journal. So in general, we should disallow all
151 * glocks that are journaled, and allow all the others. One exception is:
152 * we need to allow our active journal to be promoted and demoted so others
153 * may recover it and we can reacquire it when they're done.
154 */
155static bool glock_blocked_by_withdraw(struct gfs2_glock *gl)
156{
157 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
158
159 if (!gfs2_withdrawing_or_withdrawn(sdp))
160 return false;
161 if (gl->gl_ops->go_flags & GLOF_NONDISK)
162 return false;
163 if (!sdp->sd_jdesc ||
164 gl->gl_name.ln_number == sdp->sd_jdesc->jd_no_addr)
165 return false;
166 return true;
167}
168
169static void __gfs2_glock_free(struct gfs2_glock *gl)
170{
171 rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
172 smp_mb();
173 wake_up_glock(gl);
174 call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
175}
176
177void gfs2_glock_free(struct gfs2_glock *gl) {
178 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
179
180 __gfs2_glock_free(gl);
181 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
182 wake_up(&sdp->sd_kill_wait);
183}
184
185void gfs2_glock_free_later(struct gfs2_glock *gl) {
186 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
187
188 spin_lock(&lru_lock);
189 list_add(&gl->gl_lru, &sdp->sd_dead_glocks);
190 spin_unlock(&lru_lock);
191 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
192 wake_up(&sdp->sd_kill_wait);
193}
194
195static void gfs2_free_dead_glocks(struct gfs2_sbd *sdp)
196{
197 struct list_head *list = &sdp->sd_dead_glocks;
198
199 while(!list_empty(list)) {
200 struct gfs2_glock *gl;
201
202 gl = list_first_entry(list, struct gfs2_glock, gl_lru);
203 list_del_init(&gl->gl_lru);
204 __gfs2_glock_free(gl);
205 }
206}
207
208/**
209 * gfs2_glock_hold() - increment reference count on glock
210 * @gl: The glock to hold
211 *
212 */
213
214struct gfs2_glock *gfs2_glock_hold(struct gfs2_glock *gl)
215{
216 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
217 lockref_get(&gl->gl_lockref);
218 return gl;
219}
220
221/**
222 * demote_ok - Check to see if it's ok to unlock a glock
223 * @gl: the glock
224 *
225 * Returns: 1 if it's ok
226 */
227
228static int demote_ok(const struct gfs2_glock *gl)
229{
230 const struct gfs2_glock_operations *glops = gl->gl_ops;
231
232 if (gl->gl_state == LM_ST_UNLOCKED)
233 return 0;
234 if (!list_empty(&gl->gl_holders))
235 return 0;
236 if (glops->go_demote_ok)
237 return glops->go_demote_ok(gl);
238 return 1;
239}
240
241
242void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
243{
244 if (!(gl->gl_ops->go_flags & GLOF_LRU))
245 return;
246
247 spin_lock(&lru_lock);
248
249 list_move_tail(&gl->gl_lru, &lru_list);
250
251 if (!test_bit(GLF_LRU, &gl->gl_flags)) {
252 set_bit(GLF_LRU, &gl->gl_flags);
253 atomic_inc(&lru_count);
254 }
255
256 spin_unlock(&lru_lock);
257}
258
259static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
260{
261 if (!(gl->gl_ops->go_flags & GLOF_LRU))
262 return;
263
264 spin_lock(&lru_lock);
265 if (test_bit(GLF_LRU, &gl->gl_flags)) {
266 list_del_init(&gl->gl_lru);
267 atomic_dec(&lru_count);
268 clear_bit(GLF_LRU, &gl->gl_flags);
269 }
270 spin_unlock(&lru_lock);
271}
272
273/*
274 * Enqueue the glock on the work queue. Passes one glock reference on to the
275 * work queue.
276 */
277static void __gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
278 if (!queue_delayed_work(glock_workqueue, &gl->gl_work, delay)) {
279 /*
280 * We are holding the lockref spinlock, and the work was still
281 * queued above. The queued work (glock_work_func) takes that
282 * spinlock before dropping its glock reference(s), so it
283 * cannot have dropped them in the meantime.
284 */
285 GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2);
286 gl->gl_lockref.count--;
287 }
288}
289
290static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
291 spin_lock(&gl->gl_lockref.lock);
292 __gfs2_glock_queue_work(gl, delay);
293 spin_unlock(&gl->gl_lockref.lock);
294}
295
296static void __gfs2_glock_put(struct gfs2_glock *gl)
297{
298 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
299 struct address_space *mapping = gfs2_glock2aspace(gl);
300
301 lockref_mark_dead(&gl->gl_lockref);
302 spin_unlock(&gl->gl_lockref.lock);
303 gfs2_glock_remove_from_lru(gl);
304 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
305 if (mapping) {
306 truncate_inode_pages_final(mapping);
307 if (!gfs2_withdrawing_or_withdrawn(sdp))
308 GLOCK_BUG_ON(gl, !mapping_empty(mapping));
309 }
310 trace_gfs2_glock_put(gl);
311 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
312}
313
314/*
315 * Cause the glock to be put in work queue context.
316 */
317void gfs2_glock_queue_put(struct gfs2_glock *gl)
318{
319 gfs2_glock_queue_work(gl, 0);
320}
321
322/**
323 * gfs2_glock_put() - Decrement reference count on glock
324 * @gl: The glock to put
325 *
326 */
327
328void gfs2_glock_put(struct gfs2_glock *gl)
329{
330 if (lockref_put_or_lock(&gl->gl_lockref))
331 return;
332
333 __gfs2_glock_put(gl);
334}
335
336/**
337 * may_grant - check if it's ok to grant a new lock
338 * @gl: The glock
339 * @current_gh: One of the current holders of @gl
340 * @gh: The lock request which we wish to grant
341 *
342 * With our current compatibility rules, if a glock has one or more active
343 * holders (HIF_HOLDER flag set), any of those holders can be passed in as
344 * @current_gh; they are all the same as far as compatibility with the new @gh
345 * goes.
346 *
347 * Returns true if it's ok to grant the lock.
348 */
349
350static inline bool may_grant(struct gfs2_glock *gl,
351 struct gfs2_holder *current_gh,
352 struct gfs2_holder *gh)
353{
354 if (current_gh) {
355 GLOCK_BUG_ON(gl, !test_bit(HIF_HOLDER, ¤t_gh->gh_iflags));
356
357 switch(current_gh->gh_state) {
358 case LM_ST_EXCLUSIVE:
359 /*
360 * Here we make a special exception to grant holders
361 * who agree to share the EX lock with other holders
362 * who also have the bit set. If the original holder
363 * has the LM_FLAG_NODE_SCOPE bit set, we grant more
364 * holders with the bit set.
365 */
366 return gh->gh_state == LM_ST_EXCLUSIVE &&
367 (current_gh->gh_flags & LM_FLAG_NODE_SCOPE) &&
368 (gh->gh_flags & LM_FLAG_NODE_SCOPE);
369
370 case LM_ST_SHARED:
371 case LM_ST_DEFERRED:
372 return gh->gh_state == current_gh->gh_state;
373
374 default:
375 return false;
376 }
377 }
378
379 if (gl->gl_state == gh->gh_state)
380 return true;
381 if (gh->gh_flags & GL_EXACT)
382 return false;
383 if (gl->gl_state == LM_ST_EXCLUSIVE) {
384 return gh->gh_state == LM_ST_SHARED ||
385 gh->gh_state == LM_ST_DEFERRED;
386 }
387 if (gh->gh_flags & LM_FLAG_ANY)
388 return gl->gl_state != LM_ST_UNLOCKED;
389 return false;
390}
391
392static void gfs2_holder_wake(struct gfs2_holder *gh)
393{
394 clear_bit(HIF_WAIT, &gh->gh_iflags);
395 smp_mb__after_atomic();
396 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
397 if (gh->gh_flags & GL_ASYNC) {
398 struct gfs2_sbd *sdp = gh->gh_gl->gl_name.ln_sbd;
399
400 wake_up(&sdp->sd_async_glock_wait);
401 }
402}
403
404/**
405 * do_error - Something unexpected has happened during a lock request
406 * @gl: The glock
407 * @ret: The status from the DLM
408 */
409
410static void do_error(struct gfs2_glock *gl, const int ret)
411{
412 struct gfs2_holder *gh, *tmp;
413
414 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
415 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
416 continue;
417 if (ret & LM_OUT_ERROR)
418 gh->gh_error = -EIO;
419 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
420 gh->gh_error = GLR_TRYFAILED;
421 else
422 continue;
423 list_del_init(&gh->gh_list);
424 trace_gfs2_glock_queue(gh, 0);
425 gfs2_holder_wake(gh);
426 }
427}
428
429/**
430 * find_first_holder - find the first "holder" gh
431 * @gl: the glock
432 */
433
434static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
435{
436 struct gfs2_holder *gh;
437
438 if (!list_empty(&gl->gl_holders)) {
439 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder,
440 gh_list);
441 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
442 return gh;
443 }
444 return NULL;
445}
446
447/*
448 * gfs2_instantiate - Call the glops instantiate function
449 * @gh: The glock holder
450 *
451 * Returns: 0 if instantiate was successful, or error.
452 */
453int gfs2_instantiate(struct gfs2_holder *gh)
454{
455 struct gfs2_glock *gl = gh->gh_gl;
456 const struct gfs2_glock_operations *glops = gl->gl_ops;
457 int ret;
458
459again:
460 if (!test_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags))
461 goto done;
462
463 /*
464 * Since we unlock the lockref lock, we set a flag to indicate
465 * instantiate is in progress.
466 */
467 if (test_and_set_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags)) {
468 wait_on_bit(&gl->gl_flags, GLF_INSTANTIATE_IN_PROG,
469 TASK_UNINTERRUPTIBLE);
470 /*
471 * Here we just waited for a different instantiate to finish.
472 * But that may not have been successful, as when a process
473 * locks an inode glock _before_ it has an actual inode to
474 * instantiate into. So we check again. This process might
475 * have an inode to instantiate, so might be successful.
476 */
477 goto again;
478 }
479
480 ret = glops->go_instantiate(gl);
481 if (!ret)
482 clear_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags);
483 clear_and_wake_up_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags);
484 if (ret)
485 return ret;
486
487done:
488 if (glops->go_held)
489 return glops->go_held(gh);
490 return 0;
491}
492
493/**
494 * do_promote - promote as many requests as possible on the current queue
495 * @gl: The glock
496 *
497 * Returns true on success (i.e., progress was made or there are no waiters).
498 */
499
500static bool do_promote(struct gfs2_glock *gl)
501{
502 struct gfs2_holder *gh, *current_gh;
503
504 current_gh = find_first_holder(gl);
505 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
506 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
507 continue;
508 if (!may_grant(gl, current_gh, gh)) {
509 /*
510 * If we get here, it means we may not grant this
511 * holder for some reason. If this holder is at the
512 * head of the list, it means we have a blocked holder
513 * at the head, so return false.
514 */
515 if (list_is_first(&gh->gh_list, &gl->gl_holders))
516 return false;
517 do_error(gl, 0);
518 break;
519 }
520 set_bit(HIF_HOLDER, &gh->gh_iflags);
521 trace_gfs2_promote(gh);
522 gfs2_holder_wake(gh);
523 if (!current_gh)
524 current_gh = gh;
525 }
526 return true;
527}
528
529/**
530 * find_first_waiter - find the first gh that's waiting for the glock
531 * @gl: the glock
532 */
533
534static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
535{
536 struct gfs2_holder *gh;
537
538 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
539 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
540 return gh;
541 }
542 return NULL;
543}
544
545/**
546 * find_last_waiter - find the last gh that's waiting for the glock
547 * @gl: the glock
548 *
549 * This also is a fast way of finding out if there are any waiters.
550 */
551
552static inline struct gfs2_holder *find_last_waiter(const struct gfs2_glock *gl)
553{
554 struct gfs2_holder *gh;
555
556 if (list_empty(&gl->gl_holders))
557 return NULL;
558 gh = list_last_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
559 return test_bit(HIF_HOLDER, &gh->gh_iflags) ? NULL : gh;
560}
561
562/**
563 * state_change - record that the glock is now in a different state
564 * @gl: the glock
565 * @new_state: the new state
566 */
567
568static void state_change(struct gfs2_glock *gl, unsigned int new_state)
569{
570 int held1, held2;
571
572 held1 = (gl->gl_state != LM_ST_UNLOCKED);
573 held2 = (new_state != LM_ST_UNLOCKED);
574
575 if (held1 != held2) {
576 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
577 if (held2)
578 gl->gl_lockref.count++;
579 else
580 gl->gl_lockref.count--;
581 }
582 if (new_state != gl->gl_target)
583 /* shorten our minimum hold time */
584 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
585 GL_GLOCK_MIN_HOLD);
586 gl->gl_state = new_state;
587 gl->gl_tchange = jiffies;
588}
589
590static void gfs2_set_demote(struct gfs2_glock *gl)
591{
592 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
593
594 set_bit(GLF_DEMOTE, &gl->gl_flags);
595 smp_mb();
596 wake_up(&sdp->sd_async_glock_wait);
597}
598
599static void gfs2_demote_wake(struct gfs2_glock *gl)
600{
601 gl->gl_demote_state = LM_ST_EXCLUSIVE;
602 clear_bit(GLF_DEMOTE, &gl->gl_flags);
603 smp_mb__after_atomic();
604 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
605}
606
607/**
608 * finish_xmote - The DLM has replied to one of our lock requests
609 * @gl: The glock
610 * @ret: The status from the DLM
611 *
612 */
613
614static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
615{
616 const struct gfs2_glock_operations *glops = gl->gl_ops;
617 struct gfs2_holder *gh;
618 unsigned state = ret & LM_OUT_ST_MASK;
619
620 trace_gfs2_glock_state_change(gl, state);
621 state_change(gl, state);
622 gh = find_first_waiter(gl);
623
624 /* Demote to UN request arrived during demote to SH or DF */
625 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
626 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
627 gl->gl_target = LM_ST_UNLOCKED;
628
629 /* Check for state != intended state */
630 if (unlikely(state != gl->gl_target)) {
631 if (gh && (ret & LM_OUT_CANCELED))
632 gfs2_holder_wake(gh);
633 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
634 /* move to back of queue and try next entry */
635 if (ret & LM_OUT_CANCELED) {
636 list_move_tail(&gh->gh_list, &gl->gl_holders);
637 gh = find_first_waiter(gl);
638 gl->gl_target = gh->gh_state;
639 if (do_promote(gl))
640 goto out;
641 goto retry;
642 }
643 /* Some error or failed "try lock" - report it */
644 if ((ret & LM_OUT_ERROR) ||
645 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
646 gl->gl_target = gl->gl_state;
647 do_error(gl, ret);
648 goto out;
649 }
650 }
651 switch(state) {
652 /* Unlocked due to conversion deadlock, try again */
653 case LM_ST_UNLOCKED:
654retry:
655 do_xmote(gl, gh, gl->gl_target);
656 break;
657 /* Conversion fails, unlock and try again */
658 case LM_ST_SHARED:
659 case LM_ST_DEFERRED:
660 do_xmote(gl, gh, LM_ST_UNLOCKED);
661 break;
662 default: /* Everything else */
663 fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n",
664 gl->gl_target, state);
665 GLOCK_BUG_ON(gl, 1);
666 }
667 return;
668 }
669
670 /* Fast path - we got what we asked for */
671 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
672 gfs2_demote_wake(gl);
673 if (state != LM_ST_UNLOCKED) {
674 if (glops->go_xmote_bh) {
675 int rv;
676
677 spin_unlock(&gl->gl_lockref.lock);
678 rv = glops->go_xmote_bh(gl);
679 spin_lock(&gl->gl_lockref.lock);
680 if (rv) {
681 do_error(gl, rv);
682 goto out;
683 }
684 }
685 do_promote(gl);
686 }
687out:
688 clear_bit(GLF_LOCK, &gl->gl_flags);
689}
690
691static bool is_system_glock(struct gfs2_glock *gl)
692{
693 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
694 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
695
696 if (gl == m_ip->i_gl)
697 return true;
698 return false;
699}
700
701/**
702 * do_xmote - Calls the DLM to change the state of a lock
703 * @gl: The lock state
704 * @gh: The holder (only for promotes)
705 * @target: The target lock state
706 *
707 */
708
709static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh,
710 unsigned int target)
711__releases(&gl->gl_lockref.lock)
712__acquires(&gl->gl_lockref.lock)
713{
714 const struct gfs2_glock_operations *glops = gl->gl_ops;
715 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
716 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
717 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
718 int ret;
719
720 if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) &&
721 gh && !(gh->gh_flags & LM_FLAG_NOEXP))
722 goto skip_inval;
723
724 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP);
725 GLOCK_BUG_ON(gl, gl->gl_state == target);
726 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
727 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
728 glops->go_inval) {
729 /*
730 * If another process is already doing the invalidate, let that
731 * finish first. The glock state machine will get back to this
732 * holder again later.
733 */
734 if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
735 &gl->gl_flags))
736 return;
737 do_error(gl, 0); /* Fail queued try locks */
738 }
739 gl->gl_req = target;
740 set_bit(GLF_BLOCKING, &gl->gl_flags);
741 if ((gl->gl_req == LM_ST_UNLOCKED) ||
742 (gl->gl_state == LM_ST_EXCLUSIVE) ||
743 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
744 clear_bit(GLF_BLOCKING, &gl->gl_flags);
745 if (!glops->go_inval && !glops->go_sync)
746 goto skip_inval;
747
748 spin_unlock(&gl->gl_lockref.lock);
749 if (glops->go_sync) {
750 ret = glops->go_sync(gl);
751 /* If we had a problem syncing (due to io errors or whatever,
752 * we should not invalidate the metadata or tell dlm to
753 * release the glock to other nodes.
754 */
755 if (ret) {
756 if (cmpxchg(&sdp->sd_log_error, 0, ret)) {
757 fs_err(sdp, "Error %d syncing glock \n", ret);
758 gfs2_dump_glock(NULL, gl, true);
759 }
760 spin_lock(&gl->gl_lockref.lock);
761 goto skip_inval;
762 }
763 }
764 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) {
765 /*
766 * The call to go_sync should have cleared out the ail list.
767 * If there are still items, we have a problem. We ought to
768 * withdraw, but we can't because the withdraw code also uses
769 * glocks. Warn about the error, dump the glock, then fall
770 * through and wait for logd to do the withdraw for us.
771 */
772 if ((atomic_read(&gl->gl_ail_count) != 0) &&
773 (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) {
774 gfs2_glock_assert_warn(gl,
775 !atomic_read(&gl->gl_ail_count));
776 gfs2_dump_glock(NULL, gl, true);
777 }
778 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
779 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
780 }
781 spin_lock(&gl->gl_lockref.lock);
782
783skip_inval:
784 gl->gl_lockref.count++;
785 /*
786 * Check for an error encountered since we called go_sync and go_inval.
787 * If so, we can't withdraw from the glock code because the withdraw
788 * code itself uses glocks (see function signal_our_withdraw) to
789 * change the mount to read-only. Most importantly, we must not call
790 * dlm to unlock the glock until the journal is in a known good state
791 * (after journal replay) otherwise other nodes may use the object
792 * (rgrp or dinode) and then later, journal replay will corrupt the
793 * file system. The best we can do here is wait for the logd daemon
794 * to see sd_log_error and withdraw, and in the meantime, requeue the
795 * work for later.
796 *
797 * We make a special exception for some system glocks, such as the
798 * system statfs inode glock, which needs to be granted before the
799 * gfs2_quotad daemon can exit, and that exit needs to finish before
800 * we can unmount the withdrawn file system.
801 *
802 * However, if we're just unlocking the lock (say, for unmount, when
803 * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
804 * then it's okay to tell dlm to unlock it.
805 */
806 if (unlikely(sdp->sd_log_error) && !gfs2_withdrawing_or_withdrawn(sdp))
807 gfs2_withdraw_delayed(sdp);
808 if (glock_blocked_by_withdraw(gl) &&
809 (target != LM_ST_UNLOCKED ||
810 test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags))) {
811 if (!is_system_glock(gl)) {
812 handle_callback(gl, LM_ST_UNLOCKED, 0, false); /* sets demote */
813 /*
814 * Ordinarily, we would call dlm and its callback would call
815 * finish_xmote, which would call state_change() to the new state.
816 * Since we withdrew, we won't call dlm, so call state_change
817 * manually, but to the UNLOCKED state we desire.
818 */
819 state_change(gl, LM_ST_UNLOCKED);
820 /*
821 * We skip telling dlm to do the locking, so we won't get a
822 * reply that would otherwise clear GLF_LOCK. So we clear it here.
823 */
824 clear_bit(GLF_LOCK, &gl->gl_flags);
825 clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
826 __gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD);
827 return;
828 } else {
829 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
830 }
831 }
832
833 if (ls->ls_ops->lm_lock) {
834 spin_unlock(&gl->gl_lockref.lock);
835 ret = ls->ls_ops->lm_lock(gl, target, lck_flags);
836 spin_lock(&gl->gl_lockref.lock);
837
838 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
839 target == LM_ST_UNLOCKED &&
840 test_bit(DFL_UNMOUNT, &ls->ls_recover_flags)) {
841 /*
842 * The lockspace has been released and the lock has
843 * been unlocked implicitly.
844 */
845 } else if (ret) {
846 fs_err(sdp, "lm_lock ret %d\n", ret);
847 target = gl->gl_state | LM_OUT_ERROR;
848 } else {
849 /* The operation will be completed asynchronously. */
850 return;
851 }
852 }
853
854 /* Complete the operation now. */
855 finish_xmote(gl, target);
856 __gfs2_glock_queue_work(gl, 0);
857}
858
859/**
860 * run_queue - do all outstanding tasks related to a glock
861 * @gl: The glock in question
862 * @nonblock: True if we must not block in run_queue
863 *
864 */
865
866static void run_queue(struct gfs2_glock *gl, const int nonblock)
867__releases(&gl->gl_lockref.lock)
868__acquires(&gl->gl_lockref.lock)
869{
870 struct gfs2_holder *gh = NULL;
871
872 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
873 return;
874
875 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
876
877 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
878 gl->gl_demote_state != gl->gl_state) {
879 if (find_first_holder(gl))
880 goto out_unlock;
881 if (nonblock)
882 goto out_sched;
883 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
884 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
885 gl->gl_target = gl->gl_demote_state;
886 } else {
887 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
888 gfs2_demote_wake(gl);
889 if (do_promote(gl))
890 goto out_unlock;
891 gh = find_first_waiter(gl);
892 gl->gl_target = gh->gh_state;
893 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
894 do_error(gl, 0); /* Fail queued try locks */
895 }
896 do_xmote(gl, gh, gl->gl_target);
897 return;
898
899out_sched:
900 clear_bit(GLF_LOCK, &gl->gl_flags);
901 smp_mb__after_atomic();
902 gl->gl_lockref.count++;
903 __gfs2_glock_queue_work(gl, 0);
904 return;
905
906out_unlock:
907 clear_bit(GLF_LOCK, &gl->gl_flags);
908 smp_mb__after_atomic();
909 return;
910}
911
912/**
913 * glock_set_object - set the gl_object field of a glock
914 * @gl: the glock
915 * @object: the object
916 */
917void glock_set_object(struct gfs2_glock *gl, void *object)
918{
919 void *prev_object;
920
921 spin_lock(&gl->gl_lockref.lock);
922 prev_object = gl->gl_object;
923 gl->gl_object = object;
924 spin_unlock(&gl->gl_lockref.lock);
925 if (gfs2_assert_warn(gl->gl_name.ln_sbd, prev_object == NULL)) {
926 pr_warn("glock=%u/%llx\n",
927 gl->gl_name.ln_type,
928 (unsigned long long)gl->gl_name.ln_number);
929 gfs2_dump_glock(NULL, gl, true);
930 }
931}
932
933/**
934 * glock_clear_object - clear the gl_object field of a glock
935 * @gl: the glock
936 * @object: object the glock currently points at
937 */
938void glock_clear_object(struct gfs2_glock *gl, void *object)
939{
940 void *prev_object;
941
942 spin_lock(&gl->gl_lockref.lock);
943 prev_object = gl->gl_object;
944 gl->gl_object = NULL;
945 spin_unlock(&gl->gl_lockref.lock);
946 if (gfs2_assert_warn(gl->gl_name.ln_sbd, prev_object == object)) {
947 pr_warn("glock=%u/%llx\n",
948 gl->gl_name.ln_type,
949 (unsigned long long)gl->gl_name.ln_number);
950 gfs2_dump_glock(NULL, gl, true);
951 }
952}
953
954void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation)
955{
956 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
957
958 if (ri->ri_magic == 0)
959 ri->ri_magic = cpu_to_be32(GFS2_MAGIC);
960 if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC))
961 ri->ri_generation_deleted = cpu_to_be64(generation);
962}
963
964bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation)
965{
966 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
967
968 if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC))
969 return false;
970 return generation <= be64_to_cpu(ri->ri_generation_deleted);
971}
972
973static void gfs2_glock_poke(struct gfs2_glock *gl)
974{
975 int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP;
976 struct gfs2_holder gh;
977 int error;
978
979 __gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh, _RET_IP_);
980 error = gfs2_glock_nq(&gh);
981 if (!error)
982 gfs2_glock_dq(&gh);
983 gfs2_holder_uninit(&gh);
984}
985
986static bool gfs2_try_evict(struct gfs2_glock *gl)
987{
988 struct gfs2_inode *ip;
989 bool evicted = false;
990
991 /*
992 * If there is contention on the iopen glock and we have an inode, try
993 * to grab and release the inode so that it can be evicted. This will
994 * allow the remote node to go ahead and delete the inode without us
995 * having to do it, which will avoid rgrp glock thrashing.
996 *
997 * The remote node is likely still holding the corresponding inode
998 * glock, so it will run before we get to verify that the delete has
999 * happened below.
1000 */
1001 spin_lock(&gl->gl_lockref.lock);
1002 ip = gl->gl_object;
1003 if (ip && !igrab(&ip->i_inode))
1004 ip = NULL;
1005 spin_unlock(&gl->gl_lockref.lock);
1006 if (ip) {
1007 gl->gl_no_formal_ino = ip->i_no_formal_ino;
1008 set_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
1009 d_prune_aliases(&ip->i_inode);
1010 iput(&ip->i_inode);
1011
1012 /* If the inode was evicted, gl->gl_object will now be NULL. */
1013 spin_lock(&gl->gl_lockref.lock);
1014 ip = gl->gl_object;
1015 if (ip) {
1016 clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
1017 if (!igrab(&ip->i_inode))
1018 ip = NULL;
1019 }
1020 spin_unlock(&gl->gl_lockref.lock);
1021 if (ip) {
1022 gfs2_glock_poke(ip->i_gl);
1023 iput(&ip->i_inode);
1024 }
1025 evicted = !ip;
1026 }
1027 return evicted;
1028}
1029
1030bool gfs2_queue_try_to_evict(struct gfs2_glock *gl)
1031{
1032 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1033
1034 if (test_and_set_bit(GLF_TRY_TO_EVICT, &gl->gl_flags))
1035 return false;
1036 return queue_delayed_work(sdp->sd_delete_wq,
1037 &gl->gl_delete, 0);
1038}
1039
1040static bool gfs2_queue_verify_evict(struct gfs2_glock *gl)
1041{
1042 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1043
1044 if (test_and_set_bit(GLF_VERIFY_EVICT, &gl->gl_flags))
1045 return false;
1046 return queue_delayed_work(sdp->sd_delete_wq,
1047 &gl->gl_delete, 5 * HZ);
1048}
1049
1050static void delete_work_func(struct work_struct *work)
1051{
1052 struct delayed_work *dwork = to_delayed_work(work);
1053 struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete);
1054 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1055 struct inode *inode;
1056 u64 no_addr = gl->gl_name.ln_number;
1057
1058 if (test_and_clear_bit(GLF_TRY_TO_EVICT, &gl->gl_flags)) {
1059 /*
1060 * If we can evict the inode, give the remote node trying to
1061 * delete the inode some time before verifying that the delete
1062 * has happened. Otherwise, if we cause contention on the inode glock
1063 * immediately, the remote node will think that we still have
1064 * the inode in use, and so it will give up waiting.
1065 *
1066 * If we can't evict the inode, signal to the remote node that
1067 * the inode is still in use. We'll later try to delete the
1068 * inode locally in gfs2_evict_inode.
1069 *
1070 * FIXME: We only need to verify that the remote node has
1071 * deleted the inode because nodes before this remote delete
1072 * rework won't cooperate. At a later time, when we no longer
1073 * care about compatibility with such nodes, we can skip this
1074 * step entirely.
1075 */
1076 if (gfs2_try_evict(gl)) {
1077 if (test_bit(SDF_KILL, &sdp->sd_flags))
1078 goto out;
1079 if (gfs2_queue_verify_evict(gl))
1080 return;
1081 }
1082 goto out;
1083 }
1084
1085 if (test_and_clear_bit(GLF_VERIFY_EVICT, &gl->gl_flags)) {
1086 inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino,
1087 GFS2_BLKST_UNLINKED);
1088 if (IS_ERR(inode)) {
1089 if (PTR_ERR(inode) == -EAGAIN &&
1090 !test_bit(SDF_KILL, &sdp->sd_flags) &&
1091 gfs2_queue_verify_evict(gl))
1092 return;
1093 } else {
1094 d_prune_aliases(inode);
1095 iput(inode);
1096 }
1097 }
1098
1099out:
1100 gfs2_glock_put(gl);
1101}
1102
1103static void glock_work_func(struct work_struct *work)
1104{
1105 unsigned long delay = 0;
1106 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
1107 unsigned int drop_refs = 1;
1108
1109 spin_lock(&gl->gl_lockref.lock);
1110 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
1111 clear_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1112 finish_xmote(gl, gl->gl_reply);
1113 drop_refs++;
1114 }
1115 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1116 gl->gl_state != LM_ST_UNLOCKED &&
1117 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
1118 unsigned long holdtime, now = jiffies;
1119
1120 holdtime = gl->gl_tchange + gl->gl_hold_time;
1121 if (time_before(now, holdtime))
1122 delay = holdtime - now;
1123
1124 if (!delay) {
1125 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1126 gfs2_set_demote(gl);
1127 }
1128 }
1129 run_queue(gl, 0);
1130 if (delay) {
1131 /* Keep one glock reference for the work we requeue. */
1132 drop_refs--;
1133 if (gl->gl_name.ln_type != LM_TYPE_INODE)
1134 delay = 0;
1135 __gfs2_glock_queue_work(gl, delay);
1136 }
1137
1138 /*
1139 * Drop the remaining glock references manually here. (Mind that
1140 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
1141 * here as well.)
1142 */
1143 gl->gl_lockref.count -= drop_refs;
1144 if (!gl->gl_lockref.count) {
1145 __gfs2_glock_put(gl);
1146 return;
1147 }
1148 spin_unlock(&gl->gl_lockref.lock);
1149}
1150
1151static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
1152 struct gfs2_glock *new)
1153{
1154 struct wait_glock_queue wait;
1155 wait_queue_head_t *wq = glock_waitqueue(name);
1156 struct gfs2_glock *gl;
1157
1158 wait.name = name;
1159 init_wait(&wait.wait);
1160 wait.wait.func = glock_wake_function;
1161
1162again:
1163 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1164 rcu_read_lock();
1165 if (new) {
1166 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
1167 &new->gl_node, ht_parms);
1168 if (IS_ERR(gl))
1169 goto out;
1170 } else {
1171 gl = rhashtable_lookup_fast(&gl_hash_table,
1172 name, ht_parms);
1173 }
1174 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
1175 rcu_read_unlock();
1176 schedule();
1177 goto again;
1178 }
1179out:
1180 rcu_read_unlock();
1181 finish_wait(wq, &wait.wait);
1182 return gl;
1183}
1184
1185/**
1186 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
1187 * @sdp: The GFS2 superblock
1188 * @number: the lock number
1189 * @glops: The glock_operations to use
1190 * @create: If 0, don't create the glock if it doesn't exist
1191 * @glp: the glock is returned here
1192 *
1193 * This does not lock a glock, just finds/creates structures for one.
1194 *
1195 * Returns: errno
1196 */
1197
1198int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
1199 const struct gfs2_glock_operations *glops, int create,
1200 struct gfs2_glock **glp)
1201{
1202 struct super_block *s = sdp->sd_vfs;
1203 struct lm_lockname name = { .ln_number = number,
1204 .ln_type = glops->go_type,
1205 .ln_sbd = sdp };
1206 struct gfs2_glock *gl, *tmp;
1207 struct address_space *mapping;
1208 int ret = 0;
1209
1210 gl = find_insert_glock(&name, NULL);
1211 if (gl) {
1212 *glp = gl;
1213 return 0;
1214 }
1215 if (!create)
1216 return -ENOENT;
1217
1218 if (glops->go_flags & GLOF_ASPACE) {
1219 struct gfs2_glock_aspace *gla =
1220 kmem_cache_alloc(gfs2_glock_aspace_cachep, GFP_NOFS);
1221 if (!gla)
1222 return -ENOMEM;
1223 gl = &gla->glock;
1224 } else {
1225 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_NOFS);
1226 if (!gl)
1227 return -ENOMEM;
1228 }
1229 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
1230 gl->gl_ops = glops;
1231
1232 if (glops->go_flags & GLOF_LVB) {
1233 gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1234 if (!gl->gl_lksb.sb_lvbptr) {
1235 gfs2_glock_dealloc(&gl->gl_rcu);
1236 return -ENOMEM;
1237 }
1238 }
1239
1240 atomic_inc(&sdp->sd_glock_disposal);
1241 gl->gl_node.next = NULL;
1242 gl->gl_flags = glops->go_instantiate ? BIT(GLF_INSTANTIATE_NEEDED) : 0;
1243 gl->gl_name = name;
1244 lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass);
1245 gl->gl_lockref.count = 1;
1246 gl->gl_state = LM_ST_UNLOCKED;
1247 gl->gl_target = LM_ST_UNLOCKED;
1248 gl->gl_demote_state = LM_ST_EXCLUSIVE;
1249 gl->gl_dstamp = 0;
1250 preempt_disable();
1251 /* We use the global stats to estimate the initial per-glock stats */
1252 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
1253 preempt_enable();
1254 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
1255 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
1256 gl->gl_tchange = jiffies;
1257 gl->gl_object = NULL;
1258 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
1259 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
1260 if (gl->gl_name.ln_type == LM_TYPE_IOPEN)
1261 INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func);
1262
1263 mapping = gfs2_glock2aspace(gl);
1264 if (mapping) {
1265 mapping->a_ops = &gfs2_meta_aops;
1266 mapping->host = s->s_bdev->bd_inode;
1267 mapping->flags = 0;
1268 mapping_set_gfp_mask(mapping, GFP_NOFS);
1269 mapping->i_private_data = NULL;
1270 mapping->writeback_index = 0;
1271 }
1272
1273 tmp = find_insert_glock(&name, gl);
1274 if (!tmp) {
1275 *glp = gl;
1276 goto out;
1277 }
1278 if (IS_ERR(tmp)) {
1279 ret = PTR_ERR(tmp);
1280 goto out_free;
1281 }
1282 *glp = tmp;
1283
1284out_free:
1285 gfs2_glock_dealloc(&gl->gl_rcu);
1286 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
1287 wake_up(&sdp->sd_kill_wait);
1288
1289out:
1290 return ret;
1291}
1292
1293/**
1294 * __gfs2_holder_init - initialize a struct gfs2_holder in the default way
1295 * @gl: the glock
1296 * @state: the state we're requesting
1297 * @flags: the modifier flags
1298 * @gh: the holder structure
1299 *
1300 */
1301
1302void __gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
1303 struct gfs2_holder *gh, unsigned long ip)
1304{
1305 INIT_LIST_HEAD(&gh->gh_list);
1306 gh->gh_gl = gfs2_glock_hold(gl);
1307 gh->gh_ip = ip;
1308 gh->gh_owner_pid = get_pid(task_pid(current));
1309 gh->gh_state = state;
1310 gh->gh_flags = flags;
1311 gh->gh_iflags = 0;
1312}
1313
1314/**
1315 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1316 * @state: the state we're requesting
1317 * @flags: the modifier flags
1318 * @gh: the holder structure
1319 *
1320 * Don't mess with the glock.
1321 *
1322 */
1323
1324void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
1325{
1326 gh->gh_state = state;
1327 gh->gh_flags = flags;
1328 gh->gh_iflags = 0;
1329 gh->gh_ip = _RET_IP_;
1330 put_pid(gh->gh_owner_pid);
1331 gh->gh_owner_pid = get_pid(task_pid(current));
1332}
1333
1334/**
1335 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1336 * @gh: the holder structure
1337 *
1338 */
1339
1340void gfs2_holder_uninit(struct gfs2_holder *gh)
1341{
1342 put_pid(gh->gh_owner_pid);
1343 gfs2_glock_put(gh->gh_gl);
1344 gfs2_holder_mark_uninitialized(gh);
1345 gh->gh_ip = 0;
1346}
1347
1348static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
1349 unsigned long start_time)
1350{
1351 /* Have we waited longer that a second? */
1352 if (time_after(jiffies, start_time + HZ)) {
1353 /* Lengthen the minimum hold time. */
1354 gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
1355 GL_GLOCK_MAX_HOLD);
1356 }
1357}
1358
1359/**
1360 * gfs2_glock_holder_ready - holder is ready and its error code can be collected
1361 * @gh: the glock holder
1362 *
1363 * Called when a glock holder no longer needs to be waited for because it is
1364 * now either held (HIF_HOLDER set; gh_error == 0), or acquiring the lock has
1365 * failed (gh_error != 0).
1366 */
1367
1368int gfs2_glock_holder_ready(struct gfs2_holder *gh)
1369{
1370 if (gh->gh_error || (gh->gh_flags & GL_SKIP))
1371 return gh->gh_error;
1372 gh->gh_error = gfs2_instantiate(gh);
1373 if (gh->gh_error)
1374 gfs2_glock_dq(gh);
1375 return gh->gh_error;
1376}
1377
1378/**
1379 * gfs2_glock_wait - wait on a glock acquisition
1380 * @gh: the glock holder
1381 *
1382 * Returns: 0 on success
1383 */
1384
1385int gfs2_glock_wait(struct gfs2_holder *gh)
1386{
1387 unsigned long start_time = jiffies;
1388
1389 might_sleep();
1390 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1391 gfs2_glock_update_hold_time(gh->gh_gl, start_time);
1392 return gfs2_glock_holder_ready(gh);
1393}
1394
1395static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
1396{
1397 int i;
1398
1399 for (i = 0; i < num_gh; i++)
1400 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
1401 return 1;
1402 return 0;
1403}
1404
1405/**
1406 * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1407 * @num_gh: the number of holders in the array
1408 * @ghs: the glock holder array
1409 *
1410 * Returns: 0 on success, meaning all glocks have been granted and are held.
1411 * -ESTALE if the request timed out, meaning all glocks were released,
1412 * and the caller should retry the operation.
1413 */
1414
1415int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
1416{
1417 struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
1418 int i, ret = 0, timeout = 0;
1419 unsigned long start_time = jiffies;
1420
1421 might_sleep();
1422 /*
1423 * Total up the (minimum hold time * 2) of all glocks and use that to
1424 * determine the max amount of time we should wait.
1425 */
1426 for (i = 0; i < num_gh; i++)
1427 timeout += ghs[i].gh_gl->gl_hold_time << 1;
1428
1429 if (!wait_event_timeout(sdp->sd_async_glock_wait,
1430 !glocks_pending(num_gh, ghs), timeout)) {
1431 ret = -ESTALE; /* request timed out. */
1432 goto out;
1433 }
1434
1435 for (i = 0; i < num_gh; i++) {
1436 struct gfs2_holder *gh = &ghs[i];
1437 int ret2;
1438
1439 if (test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1440 gfs2_glock_update_hold_time(gh->gh_gl,
1441 start_time);
1442 }
1443 ret2 = gfs2_glock_holder_ready(gh);
1444 if (!ret)
1445 ret = ret2;
1446 }
1447
1448out:
1449 if (ret) {
1450 for (i = 0; i < num_gh; i++) {
1451 struct gfs2_holder *gh = &ghs[i];
1452
1453 gfs2_glock_dq(gh);
1454 }
1455 }
1456 return ret;
1457}
1458
1459/**
1460 * handle_callback - process a demote request
1461 * @gl: the glock
1462 * @state: the state the caller wants us to change to
1463 * @delay: zero to demote immediately; otherwise pending demote
1464 * @remote: true if this came from a different cluster node
1465 *
1466 * There are only two requests that we are going to see in actual
1467 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1468 */
1469
1470static void handle_callback(struct gfs2_glock *gl, unsigned int state,
1471 unsigned long delay, bool remote)
1472{
1473 if (delay)
1474 set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1475 else
1476 gfs2_set_demote(gl);
1477 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
1478 gl->gl_demote_state = state;
1479 gl->gl_demote_time = jiffies;
1480 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
1481 gl->gl_demote_state != state) {
1482 gl->gl_demote_state = LM_ST_UNLOCKED;
1483 }
1484 if (gl->gl_ops->go_callback)
1485 gl->gl_ops->go_callback(gl, remote);
1486 trace_gfs2_demote_rq(gl, remote);
1487}
1488
1489void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
1490{
1491 struct va_format vaf;
1492 va_list args;
1493
1494 va_start(args, fmt);
1495
1496 if (seq) {
1497 seq_vprintf(seq, fmt, args);
1498 } else {
1499 vaf.fmt = fmt;
1500 vaf.va = &args;
1501
1502 pr_err("%pV", &vaf);
1503 }
1504
1505 va_end(args);
1506}
1507
1508static inline bool pid_is_meaningful(const struct gfs2_holder *gh)
1509{
1510 if (!(gh->gh_flags & GL_NOPID))
1511 return true;
1512 if (gh->gh_state == LM_ST_UNLOCKED)
1513 return true;
1514 return false;
1515}
1516
1517/**
1518 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1519 * @gh: the holder structure to add
1520 *
1521 * Eventually we should move the recursive locking trap to a
1522 * debugging option or something like that. This is the fast
1523 * path and needs to have the minimum number of distractions.
1524 *
1525 */
1526
1527static inline void add_to_queue(struct gfs2_holder *gh)
1528__releases(&gl->gl_lockref.lock)
1529__acquires(&gl->gl_lockref.lock)
1530{
1531 struct gfs2_glock *gl = gh->gh_gl;
1532 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1533 struct list_head *insert_pt = NULL;
1534 struct gfs2_holder *gh2;
1535 int try_futile = 0;
1536
1537 GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1538 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1539 GLOCK_BUG_ON(gl, true);
1540
1541 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1542 if (test_bit(GLF_LOCK, &gl->gl_flags)) {
1543 struct gfs2_holder *current_gh;
1544
1545 current_gh = find_first_holder(gl);
1546 try_futile = !may_grant(gl, current_gh, gh);
1547 }
1548 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1549 goto fail;
1550 }
1551
1552 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1553 if (likely(gh2->gh_owner_pid != gh->gh_owner_pid))
1554 continue;
1555 if (gh->gh_gl->gl_ops->go_type == LM_TYPE_FLOCK)
1556 continue;
1557 if (!pid_is_meaningful(gh2))
1558 continue;
1559 goto trap_recursive;
1560 }
1561 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1562 if (try_futile &&
1563 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1564fail:
1565 gh->gh_error = GLR_TRYFAILED;
1566 gfs2_holder_wake(gh);
1567 return;
1568 }
1569 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1570 continue;
1571 }
1572 trace_gfs2_glock_queue(gh, 1);
1573 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1574 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1575 if (likely(insert_pt == NULL)) {
1576 list_add_tail(&gh->gh_list, &gl->gl_holders);
1577 return;
1578 }
1579 list_add_tail(&gh->gh_list, insert_pt);
1580 spin_unlock(&gl->gl_lockref.lock);
1581 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1582 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1583 spin_lock(&gl->gl_lockref.lock);
1584 return;
1585
1586trap_recursive:
1587 fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1588 fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1589 fs_err(sdp, "lock type: %d req lock state : %d\n",
1590 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1591 fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1592 fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1593 fs_err(sdp, "lock type: %d req lock state : %d\n",
1594 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1595 gfs2_dump_glock(NULL, gl, true);
1596 BUG();
1597}
1598
1599/**
1600 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1601 * @gh: the holder structure
1602 *
1603 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1604 *
1605 * Returns: 0, GLR_TRYFAILED, or errno on failure
1606 */
1607
1608int gfs2_glock_nq(struct gfs2_holder *gh)
1609{
1610 struct gfs2_glock *gl = gh->gh_gl;
1611 int error;
1612
1613 if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP))
1614 return -EIO;
1615
1616 if (gh->gh_flags & GL_NOBLOCK) {
1617 struct gfs2_holder *current_gh;
1618
1619 error = -ECHILD;
1620 spin_lock(&gl->gl_lockref.lock);
1621 if (find_last_waiter(gl))
1622 goto unlock;
1623 current_gh = find_first_holder(gl);
1624 if (!may_grant(gl, current_gh, gh))
1625 goto unlock;
1626 set_bit(HIF_HOLDER, &gh->gh_iflags);
1627 list_add_tail(&gh->gh_list, &gl->gl_holders);
1628 trace_gfs2_promote(gh);
1629 error = 0;
1630unlock:
1631 spin_unlock(&gl->gl_lockref.lock);
1632 return error;
1633 }
1634
1635 if (test_bit(GLF_LRU, &gl->gl_flags))
1636 gfs2_glock_remove_from_lru(gl);
1637
1638 gh->gh_error = 0;
1639 spin_lock(&gl->gl_lockref.lock);
1640 add_to_queue(gh);
1641 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1642 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1643 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1644 gl->gl_lockref.count++;
1645 __gfs2_glock_queue_work(gl, 0);
1646 }
1647 run_queue(gl, 1);
1648 spin_unlock(&gl->gl_lockref.lock);
1649
1650 error = 0;
1651 if (!(gh->gh_flags & GL_ASYNC))
1652 error = gfs2_glock_wait(gh);
1653
1654 return error;
1655}
1656
1657/**
1658 * gfs2_glock_poll - poll to see if an async request has been completed
1659 * @gh: the holder
1660 *
1661 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1662 */
1663
1664int gfs2_glock_poll(struct gfs2_holder *gh)
1665{
1666 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1667}
1668
1669static inline bool needs_demote(struct gfs2_glock *gl)
1670{
1671 return (test_bit(GLF_DEMOTE, &gl->gl_flags) ||
1672 test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags));
1673}
1674
1675static void __gfs2_glock_dq(struct gfs2_holder *gh)
1676{
1677 struct gfs2_glock *gl = gh->gh_gl;
1678 unsigned delay = 0;
1679 int fast_path = 0;
1680
1681 /*
1682 * This holder should not be cached, so mark it for demote.
1683 * Note: this should be done before the check for needs_demote
1684 * below.
1685 */
1686 if (gh->gh_flags & GL_NOCACHE)
1687 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1688
1689 list_del_init(&gh->gh_list);
1690 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1691 trace_gfs2_glock_queue(gh, 0);
1692
1693 /*
1694 * If there hasn't been a demote request we are done.
1695 * (Let the remaining holders, if any, keep holding it.)
1696 */
1697 if (!needs_demote(gl)) {
1698 if (list_empty(&gl->gl_holders))
1699 fast_path = 1;
1700 }
1701
1702 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1703 gfs2_glock_add_to_lru(gl);
1704
1705 if (unlikely(!fast_path)) {
1706 gl->gl_lockref.count++;
1707 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1708 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1709 gl->gl_name.ln_type == LM_TYPE_INODE)
1710 delay = gl->gl_hold_time;
1711 __gfs2_glock_queue_work(gl, delay);
1712 }
1713}
1714
1715/**
1716 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1717 * @gh: the glock holder
1718 *
1719 */
1720void gfs2_glock_dq(struct gfs2_holder *gh)
1721{
1722 struct gfs2_glock *gl = gh->gh_gl;
1723 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1724
1725 spin_lock(&gl->gl_lockref.lock);
1726 if (!gfs2_holder_queued(gh)) {
1727 /*
1728 * May have already been dequeued because the locking request
1729 * was GL_ASYNC and it has failed in the meantime.
1730 */
1731 goto out;
1732 }
1733
1734 if (list_is_first(&gh->gh_list, &gl->gl_holders) &&
1735 !test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1736 spin_unlock(&gl->gl_lockref.lock);
1737 gl->gl_name.ln_sbd->sd_lockstruct.ls_ops->lm_cancel(gl);
1738 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1739 spin_lock(&gl->gl_lockref.lock);
1740 }
1741
1742 /*
1743 * If we're in the process of file system withdraw, we cannot just
1744 * dequeue any glocks until our journal is recovered, lest we introduce
1745 * file system corruption. We need two exceptions to this rule: We need
1746 * to allow unlocking of nondisk glocks and the glock for our own
1747 * journal that needs recovery.
1748 */
1749 if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) &&
1750 glock_blocked_by_withdraw(gl) &&
1751 gh->gh_gl != sdp->sd_jinode_gl) {
1752 sdp->sd_glock_dqs_held++;
1753 spin_unlock(&gl->gl_lockref.lock);
1754 might_sleep();
1755 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
1756 TASK_UNINTERRUPTIBLE);
1757 spin_lock(&gl->gl_lockref.lock);
1758 }
1759
1760 __gfs2_glock_dq(gh);
1761out:
1762 spin_unlock(&gl->gl_lockref.lock);
1763}
1764
1765void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1766{
1767 struct gfs2_glock *gl = gh->gh_gl;
1768 gfs2_glock_dq(gh);
1769 might_sleep();
1770 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1771}
1772
1773/**
1774 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1775 * @gh: the holder structure
1776 *
1777 */
1778
1779void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1780{
1781 gfs2_glock_dq(gh);
1782 gfs2_holder_uninit(gh);
1783}
1784
1785/**
1786 * gfs2_glock_nq_num - acquire a glock based on lock number
1787 * @sdp: the filesystem
1788 * @number: the lock number
1789 * @glops: the glock operations for the type of glock
1790 * @state: the state to acquire the glock in
1791 * @flags: modifier flags for the acquisition
1792 * @gh: the struct gfs2_holder
1793 *
1794 * Returns: errno
1795 */
1796
1797int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1798 const struct gfs2_glock_operations *glops,
1799 unsigned int state, u16 flags, struct gfs2_holder *gh)
1800{
1801 struct gfs2_glock *gl;
1802 int error;
1803
1804 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1805 if (!error) {
1806 error = gfs2_glock_nq_init(gl, state, flags, gh);
1807 gfs2_glock_put(gl);
1808 }
1809
1810 return error;
1811}
1812
1813/**
1814 * glock_compare - Compare two struct gfs2_glock structures for sorting
1815 * @arg_a: the first structure
1816 * @arg_b: the second structure
1817 *
1818 */
1819
1820static int glock_compare(const void *arg_a, const void *arg_b)
1821{
1822 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1823 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1824 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1825 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1826
1827 if (a->ln_number > b->ln_number)
1828 return 1;
1829 if (a->ln_number < b->ln_number)
1830 return -1;
1831 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1832 return 0;
1833}
1834
1835/**
1836 * nq_m_sync - synchronously acquire more than one glock in deadlock free order
1837 * @num_gh: the number of structures
1838 * @ghs: an array of struct gfs2_holder structures
1839 * @p: placeholder for the holder structure to pass back
1840 *
1841 * Returns: 0 on success (all glocks acquired),
1842 * errno on failure (no glocks acquired)
1843 */
1844
1845static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1846 struct gfs2_holder **p)
1847{
1848 unsigned int x;
1849 int error = 0;
1850
1851 for (x = 0; x < num_gh; x++)
1852 p[x] = &ghs[x];
1853
1854 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1855
1856 for (x = 0; x < num_gh; x++) {
1857 error = gfs2_glock_nq(p[x]);
1858 if (error) {
1859 while (x--)
1860 gfs2_glock_dq(p[x]);
1861 break;
1862 }
1863 }
1864
1865 return error;
1866}
1867
1868/**
1869 * gfs2_glock_nq_m - acquire multiple glocks
1870 * @num_gh: the number of structures
1871 * @ghs: an array of struct gfs2_holder structures
1872 *
1873 * Returns: 0 on success (all glocks acquired),
1874 * errno on failure (no glocks acquired)
1875 */
1876
1877int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1878{
1879 struct gfs2_holder *tmp[4];
1880 struct gfs2_holder **pph = tmp;
1881 int error = 0;
1882
1883 switch(num_gh) {
1884 case 0:
1885 return 0;
1886 case 1:
1887 return gfs2_glock_nq(ghs);
1888 default:
1889 if (num_gh <= 4)
1890 break;
1891 pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *),
1892 GFP_NOFS);
1893 if (!pph)
1894 return -ENOMEM;
1895 }
1896
1897 error = nq_m_sync(num_gh, ghs, pph);
1898
1899 if (pph != tmp)
1900 kfree(pph);
1901
1902 return error;
1903}
1904
1905/**
1906 * gfs2_glock_dq_m - release multiple glocks
1907 * @num_gh: the number of structures
1908 * @ghs: an array of struct gfs2_holder structures
1909 *
1910 */
1911
1912void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1913{
1914 while (num_gh--)
1915 gfs2_glock_dq(&ghs[num_gh]);
1916}
1917
1918void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1919{
1920 unsigned long delay = 0;
1921 unsigned long holdtime;
1922 unsigned long now = jiffies;
1923
1924 gfs2_glock_hold(gl);
1925 spin_lock(&gl->gl_lockref.lock);
1926 holdtime = gl->gl_tchange + gl->gl_hold_time;
1927 if (!list_empty(&gl->gl_holders) &&
1928 gl->gl_name.ln_type == LM_TYPE_INODE) {
1929 if (time_before(now, holdtime))
1930 delay = holdtime - now;
1931 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1932 delay = gl->gl_hold_time;
1933 }
1934 handle_callback(gl, state, delay, true);
1935 __gfs2_glock_queue_work(gl, delay);
1936 spin_unlock(&gl->gl_lockref.lock);
1937}
1938
1939/**
1940 * gfs2_should_freeze - Figure out if glock should be frozen
1941 * @gl: The glock in question
1942 *
1943 * Glocks are not frozen if (a) the result of the dlm operation is
1944 * an error, (b) the locking operation was an unlock operation or
1945 * (c) if there is a "noexp" flagged request anywhere in the queue
1946 *
1947 * Returns: 1 if freezing should occur, 0 otherwise
1948 */
1949
1950static int gfs2_should_freeze(const struct gfs2_glock *gl)
1951{
1952 const struct gfs2_holder *gh;
1953
1954 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1955 return 0;
1956 if (gl->gl_target == LM_ST_UNLOCKED)
1957 return 0;
1958
1959 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1960 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1961 continue;
1962 if (LM_FLAG_NOEXP & gh->gh_flags)
1963 return 0;
1964 }
1965
1966 return 1;
1967}
1968
1969/**
1970 * gfs2_glock_complete - Callback used by locking
1971 * @gl: Pointer to the glock
1972 * @ret: The return value from the dlm
1973 *
1974 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1975 * to use a bitfield shared with other glock state fields.
1976 */
1977
1978void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1979{
1980 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1981
1982 spin_lock(&gl->gl_lockref.lock);
1983 gl->gl_reply = ret;
1984
1985 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1986 if (gfs2_should_freeze(gl)) {
1987 set_bit(GLF_FROZEN, &gl->gl_flags);
1988 spin_unlock(&gl->gl_lockref.lock);
1989 return;
1990 }
1991 }
1992
1993 gl->gl_lockref.count++;
1994 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1995 __gfs2_glock_queue_work(gl, 0);
1996 spin_unlock(&gl->gl_lockref.lock);
1997}
1998
1999static int glock_cmp(void *priv, const struct list_head *a,
2000 const struct list_head *b)
2001{
2002 struct gfs2_glock *gla, *glb;
2003
2004 gla = list_entry(a, struct gfs2_glock, gl_lru);
2005 glb = list_entry(b, struct gfs2_glock, gl_lru);
2006
2007 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
2008 return 1;
2009 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
2010 return -1;
2011
2012 return 0;
2013}
2014
2015/**
2016 * gfs2_dispose_glock_lru - Demote a list of glocks
2017 * @list: The list to dispose of
2018 *
2019 * Disposing of glocks may involve disk accesses, so that here we sort
2020 * the glocks by number (i.e. disk location of the inodes) so that if
2021 * there are any such accesses, they'll be sent in order (mostly).
2022 *
2023 * Must be called under the lru_lock, but may drop and retake this
2024 * lock. While the lru_lock is dropped, entries may vanish from the
2025 * list, but no new entries will appear on the list (since it is
2026 * private)
2027 */
2028
2029static void gfs2_dispose_glock_lru(struct list_head *list)
2030__releases(&lru_lock)
2031__acquires(&lru_lock)
2032{
2033 struct gfs2_glock *gl;
2034
2035 list_sort(NULL, list, glock_cmp);
2036
2037 while(!list_empty(list)) {
2038 gl = list_first_entry(list, struct gfs2_glock, gl_lru);
2039 list_del_init(&gl->gl_lru);
2040 clear_bit(GLF_LRU, &gl->gl_flags);
2041 if (!spin_trylock(&gl->gl_lockref.lock)) {
2042add_back_to_lru:
2043 list_add(&gl->gl_lru, &lru_list);
2044 set_bit(GLF_LRU, &gl->gl_flags);
2045 atomic_inc(&lru_count);
2046 continue;
2047 }
2048 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
2049 spin_unlock(&gl->gl_lockref.lock);
2050 goto add_back_to_lru;
2051 }
2052 gl->gl_lockref.count++;
2053 if (demote_ok(gl))
2054 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2055 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
2056 __gfs2_glock_queue_work(gl, 0);
2057 spin_unlock(&gl->gl_lockref.lock);
2058 cond_resched_lock(&lru_lock);
2059 }
2060}
2061
2062/**
2063 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
2064 * @nr: The number of entries to scan
2065 *
2066 * This function selects the entries on the LRU which are able to
2067 * be demoted, and then kicks off the process by calling
2068 * gfs2_dispose_glock_lru() above.
2069 */
2070
2071static long gfs2_scan_glock_lru(int nr)
2072{
2073 struct gfs2_glock *gl, *next;
2074 LIST_HEAD(dispose);
2075 long freed = 0;
2076
2077 spin_lock(&lru_lock);
2078 list_for_each_entry_safe(gl, next, &lru_list, gl_lru) {
2079 if (nr-- <= 0)
2080 break;
2081 /* Test for being demotable */
2082 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
2083 if (!spin_trylock(&gl->gl_lockref.lock))
2084 continue;
2085 if (gl->gl_lockref.count <= 1 &&
2086 (gl->gl_state == LM_ST_UNLOCKED ||
2087 demote_ok(gl))) {
2088 list_move(&gl->gl_lru, &dispose);
2089 atomic_dec(&lru_count);
2090 freed++;
2091 }
2092 spin_unlock(&gl->gl_lockref.lock);
2093 }
2094 }
2095 if (!list_empty(&dispose))
2096 gfs2_dispose_glock_lru(&dispose);
2097 spin_unlock(&lru_lock);
2098
2099 return freed;
2100}
2101
2102static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
2103 struct shrink_control *sc)
2104{
2105 if (!(sc->gfp_mask & __GFP_FS))
2106 return SHRINK_STOP;
2107 return gfs2_scan_glock_lru(sc->nr_to_scan);
2108}
2109
2110static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
2111 struct shrink_control *sc)
2112{
2113 return vfs_pressure_ratio(atomic_read(&lru_count));
2114}
2115
2116static struct shrinker *glock_shrinker;
2117
2118/**
2119 * glock_hash_walk - Call a function for glock in a hash bucket
2120 * @examiner: the function
2121 * @sdp: the filesystem
2122 *
2123 * Note that the function can be called multiple times on the same
2124 * object. So the user must ensure that the function can cope with
2125 * that.
2126 */
2127
2128static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
2129{
2130 struct gfs2_glock *gl;
2131 struct rhashtable_iter iter;
2132
2133 rhashtable_walk_enter(&gl_hash_table, &iter);
2134
2135 do {
2136 rhashtable_walk_start(&iter);
2137
2138 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl)) {
2139 if (gl->gl_name.ln_sbd == sdp)
2140 examiner(gl);
2141 }
2142
2143 rhashtable_walk_stop(&iter);
2144 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
2145
2146 rhashtable_walk_exit(&iter);
2147}
2148
2149void gfs2_cancel_delete_work(struct gfs2_glock *gl)
2150{
2151 clear_bit(GLF_TRY_TO_EVICT, &gl->gl_flags);
2152 clear_bit(GLF_VERIFY_EVICT, &gl->gl_flags);
2153 if (cancel_delayed_work(&gl->gl_delete))
2154 gfs2_glock_put(gl);
2155}
2156
2157static void flush_delete_work(struct gfs2_glock *gl)
2158{
2159 if (gl->gl_name.ln_type == LM_TYPE_IOPEN) {
2160 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2161
2162 if (cancel_delayed_work(&gl->gl_delete)) {
2163 queue_delayed_work(sdp->sd_delete_wq,
2164 &gl->gl_delete, 0);
2165 }
2166 }
2167}
2168
2169void gfs2_flush_delete_work(struct gfs2_sbd *sdp)
2170{
2171 glock_hash_walk(flush_delete_work, sdp);
2172 flush_workqueue(sdp->sd_delete_wq);
2173}
2174
2175/**
2176 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
2177 * @gl: The glock to thaw
2178 *
2179 */
2180
2181static void thaw_glock(struct gfs2_glock *gl)
2182{
2183 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
2184 return;
2185 if (!lockref_get_not_dead(&gl->gl_lockref))
2186 return;
2187
2188 spin_lock(&gl->gl_lockref.lock);
2189 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
2190 __gfs2_glock_queue_work(gl, 0);
2191 spin_unlock(&gl->gl_lockref.lock);
2192}
2193
2194/**
2195 * clear_glock - look at a glock and see if we can free it from glock cache
2196 * @gl: the glock to look at
2197 *
2198 */
2199
2200static void clear_glock(struct gfs2_glock *gl)
2201{
2202 gfs2_glock_remove_from_lru(gl);
2203
2204 spin_lock(&gl->gl_lockref.lock);
2205 if (!__lockref_is_dead(&gl->gl_lockref)) {
2206 gl->gl_lockref.count++;
2207 if (gl->gl_state != LM_ST_UNLOCKED)
2208 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2209 __gfs2_glock_queue_work(gl, 0);
2210 }
2211 spin_unlock(&gl->gl_lockref.lock);
2212}
2213
2214/**
2215 * gfs2_glock_thaw - Thaw any frozen glocks
2216 * @sdp: The super block
2217 *
2218 */
2219
2220void gfs2_glock_thaw(struct gfs2_sbd *sdp)
2221{
2222 glock_hash_walk(thaw_glock, sdp);
2223}
2224
2225static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2226{
2227 spin_lock(&gl->gl_lockref.lock);
2228 gfs2_dump_glock(seq, gl, fsid);
2229 spin_unlock(&gl->gl_lockref.lock);
2230}
2231
2232static void dump_glock_func(struct gfs2_glock *gl)
2233{
2234 dump_glock(NULL, gl, true);
2235}
2236
2237static void withdraw_dq(struct gfs2_glock *gl)
2238{
2239 spin_lock(&gl->gl_lockref.lock);
2240 if (!__lockref_is_dead(&gl->gl_lockref) &&
2241 glock_blocked_by_withdraw(gl))
2242 do_error(gl, LM_OUT_ERROR); /* remove pending waiters */
2243 spin_unlock(&gl->gl_lockref.lock);
2244}
2245
2246void gfs2_gl_dq_holders(struct gfs2_sbd *sdp)
2247{
2248 glock_hash_walk(withdraw_dq, sdp);
2249}
2250
2251/**
2252 * gfs2_gl_hash_clear - Empty out the glock hash table
2253 * @sdp: the filesystem
2254 *
2255 * Called when unmounting the filesystem.
2256 */
2257
2258void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
2259{
2260 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
2261 flush_workqueue(glock_workqueue);
2262 glock_hash_walk(clear_glock, sdp);
2263 flush_workqueue(glock_workqueue);
2264 wait_event_timeout(sdp->sd_kill_wait,
2265 atomic_read(&sdp->sd_glock_disposal) == 0,
2266 HZ * 600);
2267 gfs2_lm_unmount(sdp);
2268 gfs2_free_dead_glocks(sdp);
2269 glock_hash_walk(dump_glock_func, sdp);
2270}
2271
2272static const char *state2str(unsigned state)
2273{
2274 switch(state) {
2275 case LM_ST_UNLOCKED:
2276 return "UN";
2277 case LM_ST_SHARED:
2278 return "SH";
2279 case LM_ST_DEFERRED:
2280 return "DF";
2281 case LM_ST_EXCLUSIVE:
2282 return "EX";
2283 }
2284 return "??";
2285}
2286
2287static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
2288{
2289 char *p = buf;
2290 if (flags & LM_FLAG_TRY)
2291 *p++ = 't';
2292 if (flags & LM_FLAG_TRY_1CB)
2293 *p++ = 'T';
2294 if (flags & LM_FLAG_NOEXP)
2295 *p++ = 'e';
2296 if (flags & LM_FLAG_ANY)
2297 *p++ = 'A';
2298 if (flags & LM_FLAG_NODE_SCOPE)
2299 *p++ = 'n';
2300 if (flags & GL_ASYNC)
2301 *p++ = 'a';
2302 if (flags & GL_EXACT)
2303 *p++ = 'E';
2304 if (flags & GL_NOCACHE)
2305 *p++ = 'c';
2306 if (test_bit(HIF_HOLDER, &iflags))
2307 *p++ = 'H';
2308 if (test_bit(HIF_WAIT, &iflags))
2309 *p++ = 'W';
2310 if (flags & GL_SKIP)
2311 *p++ = 's';
2312 *p = 0;
2313 return buf;
2314}
2315
2316/**
2317 * dump_holder - print information about a glock holder
2318 * @seq: the seq_file struct
2319 * @gh: the glock holder
2320 * @fs_id_buf: pointer to file system id (if requested)
2321 *
2322 */
2323
2324static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
2325 const char *fs_id_buf)
2326{
2327 const char *comm = "(none)";
2328 pid_t owner_pid = 0;
2329 char flags_buf[32];
2330
2331 rcu_read_lock();
2332 if (pid_is_meaningful(gh)) {
2333 struct task_struct *gh_owner;
2334
2335 comm = "(ended)";
2336 owner_pid = pid_nr(gh->gh_owner_pid);
2337 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
2338 if (gh_owner)
2339 comm = gh_owner->comm;
2340 }
2341 gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2342 fs_id_buf, state2str(gh->gh_state),
2343 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
2344 gh->gh_error, (long)owner_pid, comm, (void *)gh->gh_ip);
2345 rcu_read_unlock();
2346}
2347
2348static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
2349{
2350 const unsigned long *gflags = &gl->gl_flags;
2351 char *p = buf;
2352
2353 if (test_bit(GLF_LOCK, gflags))
2354 *p++ = 'l';
2355 if (test_bit(GLF_DEMOTE, gflags))
2356 *p++ = 'D';
2357 if (test_bit(GLF_PENDING_DEMOTE, gflags))
2358 *p++ = 'd';
2359 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
2360 *p++ = 'p';
2361 if (test_bit(GLF_DIRTY, gflags))
2362 *p++ = 'y';
2363 if (test_bit(GLF_LFLUSH, gflags))
2364 *p++ = 'f';
2365 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
2366 *p++ = 'i';
2367 if (test_bit(GLF_REPLY_PENDING, gflags))
2368 *p++ = 'r';
2369 if (test_bit(GLF_INITIAL, gflags))
2370 *p++ = 'I';
2371 if (test_bit(GLF_FROZEN, gflags))
2372 *p++ = 'F';
2373 if (!list_empty(&gl->gl_holders))
2374 *p++ = 'q';
2375 if (test_bit(GLF_LRU, gflags))
2376 *p++ = 'L';
2377 if (gl->gl_object)
2378 *p++ = 'o';
2379 if (test_bit(GLF_BLOCKING, gflags))
2380 *p++ = 'b';
2381 if (test_bit(GLF_FREEING, gflags))
2382 *p++ = 'x';
2383 if (test_bit(GLF_INSTANTIATE_NEEDED, gflags))
2384 *p++ = 'n';
2385 if (test_bit(GLF_INSTANTIATE_IN_PROG, gflags))
2386 *p++ = 'N';
2387 if (test_bit(GLF_TRY_TO_EVICT, gflags))
2388 *p++ = 'e';
2389 if (test_bit(GLF_VERIFY_EVICT, gflags))
2390 *p++ = 'E';
2391 *p = 0;
2392 return buf;
2393}
2394
2395/**
2396 * gfs2_dump_glock - print information about a glock
2397 * @seq: The seq_file struct
2398 * @gl: the glock
2399 * @fsid: If true, also dump the file system id
2400 *
2401 * The file format is as follows:
2402 * One line per object, capital letters are used to indicate objects
2403 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2404 * other objects are indented by a single space and follow the glock to
2405 * which they are related. Fields are indicated by lower case letters
2406 * followed by a colon and the field value, except for strings which are in
2407 * [] so that its possible to see if they are composed of spaces for
2408 * example. The field's are n = number (id of the object), f = flags,
2409 * t = type, s = state, r = refcount, e = error, p = pid.
2410 *
2411 */
2412
2413void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2414{
2415 const struct gfs2_glock_operations *glops = gl->gl_ops;
2416 unsigned long long dtime;
2417 const struct gfs2_holder *gh;
2418 char gflags_buf[32];
2419 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2420 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2421 unsigned long nrpages = 0;
2422
2423 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
2424 struct address_space *mapping = gfs2_glock2aspace(gl);
2425
2426 nrpages = mapping->nrpages;
2427 }
2428 memset(fs_id_buf, 0, sizeof(fs_id_buf));
2429 if (fsid && sdp) /* safety precaution */
2430 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2431 dtime = jiffies - gl->gl_demote_time;
2432 dtime *= 1000000/HZ; /* demote time in uSec */
2433 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
2434 dtime = 0;
2435 gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2436 "v:%d r:%d m:%ld p:%lu\n",
2437 fs_id_buf, state2str(gl->gl_state),
2438 gl->gl_name.ln_type,
2439 (unsigned long long)gl->gl_name.ln_number,
2440 gflags2str(gflags_buf, gl),
2441 state2str(gl->gl_target),
2442 state2str(gl->gl_demote_state), dtime,
2443 atomic_read(&gl->gl_ail_count),
2444 atomic_read(&gl->gl_revokes),
2445 (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages);
2446
2447 list_for_each_entry(gh, &gl->gl_holders, gh_list)
2448 dump_holder(seq, gh, fs_id_buf);
2449
2450 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
2451 glops->go_dump(seq, gl, fs_id_buf);
2452}
2453
2454static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
2455{
2456 struct gfs2_glock *gl = iter_ptr;
2457
2458 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2459 gl->gl_name.ln_type,
2460 (unsigned long long)gl->gl_name.ln_number,
2461 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
2462 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
2463 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
2464 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
2465 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
2466 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
2467 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
2468 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
2469 return 0;
2470}
2471
2472static const char *gfs2_gltype[] = {
2473 "type",
2474 "reserved",
2475 "nondisk",
2476 "inode",
2477 "rgrp",
2478 "meta",
2479 "iopen",
2480 "flock",
2481 "plock",
2482 "quota",
2483 "journal",
2484};
2485
2486static const char *gfs2_stype[] = {
2487 [GFS2_LKS_SRTT] = "srtt",
2488 [GFS2_LKS_SRTTVAR] = "srttvar",
2489 [GFS2_LKS_SRTTB] = "srttb",
2490 [GFS2_LKS_SRTTVARB] = "srttvarb",
2491 [GFS2_LKS_SIRT] = "sirt",
2492 [GFS2_LKS_SIRTVAR] = "sirtvar",
2493 [GFS2_LKS_DCOUNT] = "dlm",
2494 [GFS2_LKS_QCOUNT] = "queue",
2495};
2496
2497#define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2498
2499static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
2500{
2501 struct gfs2_sbd *sdp = seq->private;
2502 loff_t pos = *(loff_t *)iter_ptr;
2503 unsigned index = pos >> 3;
2504 unsigned subindex = pos & 0x07;
2505 int i;
2506
2507 if (index == 0 && subindex != 0)
2508 return 0;
2509
2510 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
2511 (index == 0) ? "cpu": gfs2_stype[subindex]);
2512
2513 for_each_possible_cpu(i) {
2514 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
2515
2516 if (index == 0)
2517 seq_printf(seq, " %15u", i);
2518 else
2519 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
2520 lkstats[index - 1].stats[subindex]);
2521 }
2522 seq_putc(seq, '\n');
2523 return 0;
2524}
2525
2526int __init gfs2_glock_init(void)
2527{
2528 int i, ret;
2529
2530 ret = rhashtable_init(&gl_hash_table, &ht_parms);
2531 if (ret < 0)
2532 return ret;
2533
2534 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
2535 WQ_HIGHPRI | WQ_FREEZABLE, 0);
2536 if (!glock_workqueue) {
2537 rhashtable_destroy(&gl_hash_table);
2538 return -ENOMEM;
2539 }
2540
2541 glock_shrinker = shrinker_alloc(0, "gfs2-glock");
2542 if (!glock_shrinker) {
2543 destroy_workqueue(glock_workqueue);
2544 rhashtable_destroy(&gl_hash_table);
2545 return -ENOMEM;
2546 }
2547
2548 glock_shrinker->count_objects = gfs2_glock_shrink_count;
2549 glock_shrinker->scan_objects = gfs2_glock_shrink_scan;
2550
2551 shrinker_register(glock_shrinker);
2552
2553 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2554 init_waitqueue_head(glock_wait_table + i);
2555
2556 return 0;
2557}
2558
2559void gfs2_glock_exit(void)
2560{
2561 shrinker_free(glock_shrinker);
2562 rhashtable_destroy(&gl_hash_table);
2563 destroy_workqueue(glock_workqueue);
2564}
2565
2566static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2567{
2568 struct gfs2_glock *gl = gi->gl;
2569
2570 if (gl) {
2571 if (n == 0)
2572 return;
2573 if (!lockref_put_not_zero(&gl->gl_lockref))
2574 gfs2_glock_queue_put(gl);
2575 }
2576 for (;;) {
2577 gl = rhashtable_walk_next(&gi->hti);
2578 if (IS_ERR_OR_NULL(gl)) {
2579 if (gl == ERR_PTR(-EAGAIN)) {
2580 n = 1;
2581 continue;
2582 }
2583 gl = NULL;
2584 break;
2585 }
2586 if (gl->gl_name.ln_sbd != gi->sdp)
2587 continue;
2588 if (n <= 1) {
2589 if (!lockref_get_not_dead(&gl->gl_lockref))
2590 continue;
2591 break;
2592 } else {
2593 if (__lockref_is_dead(&gl->gl_lockref))
2594 continue;
2595 n--;
2596 }
2597 }
2598 gi->gl = gl;
2599}
2600
2601static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2602 __acquires(RCU)
2603{
2604 struct gfs2_glock_iter *gi = seq->private;
2605 loff_t n;
2606
2607 /*
2608 * We can either stay where we are, skip to the next hash table
2609 * entry, or start from the beginning.
2610 */
2611 if (*pos < gi->last_pos) {
2612 rhashtable_walk_exit(&gi->hti);
2613 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2614 n = *pos + 1;
2615 } else {
2616 n = *pos - gi->last_pos;
2617 }
2618
2619 rhashtable_walk_start(&gi->hti);
2620
2621 gfs2_glock_iter_next(gi, n);
2622 gi->last_pos = *pos;
2623 return gi->gl;
2624}
2625
2626static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2627 loff_t *pos)
2628{
2629 struct gfs2_glock_iter *gi = seq->private;
2630
2631 (*pos)++;
2632 gi->last_pos = *pos;
2633 gfs2_glock_iter_next(gi, 1);
2634 return gi->gl;
2635}
2636
2637static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2638 __releases(RCU)
2639{
2640 struct gfs2_glock_iter *gi = seq->private;
2641
2642 rhashtable_walk_stop(&gi->hti);
2643}
2644
2645static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2646{
2647 dump_glock(seq, iter_ptr, false);
2648 return 0;
2649}
2650
2651static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2652{
2653 preempt_disable();
2654 if (*pos >= GFS2_NR_SBSTATS)
2655 return NULL;
2656 return pos;
2657}
2658
2659static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2660 loff_t *pos)
2661{
2662 (*pos)++;
2663 if (*pos >= GFS2_NR_SBSTATS)
2664 return NULL;
2665 return pos;
2666}
2667
2668static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2669{
2670 preempt_enable();
2671}
2672
2673static const struct seq_operations gfs2_glock_seq_ops = {
2674 .start = gfs2_glock_seq_start,
2675 .next = gfs2_glock_seq_next,
2676 .stop = gfs2_glock_seq_stop,
2677 .show = gfs2_glock_seq_show,
2678};
2679
2680static const struct seq_operations gfs2_glstats_seq_ops = {
2681 .start = gfs2_glock_seq_start,
2682 .next = gfs2_glock_seq_next,
2683 .stop = gfs2_glock_seq_stop,
2684 .show = gfs2_glstats_seq_show,
2685};
2686
2687static const struct seq_operations gfs2_sbstats_sops = {
2688 .start = gfs2_sbstats_seq_start,
2689 .next = gfs2_sbstats_seq_next,
2690 .stop = gfs2_sbstats_seq_stop,
2691 .show = gfs2_sbstats_seq_show,
2692};
2693
2694#define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2695
2696static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2697 const struct seq_operations *ops)
2698{
2699 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2700 if (ret == 0) {
2701 struct seq_file *seq = file->private_data;
2702 struct gfs2_glock_iter *gi = seq->private;
2703
2704 gi->sdp = inode->i_private;
2705 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2706 if (seq->buf)
2707 seq->size = GFS2_SEQ_GOODSIZE;
2708 /*
2709 * Initially, we are "before" the first hash table entry; the
2710 * first call to rhashtable_walk_next gets us the first entry.
2711 */
2712 gi->last_pos = -1;
2713 gi->gl = NULL;
2714 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2715 }
2716 return ret;
2717}
2718
2719static int gfs2_glocks_open(struct inode *inode, struct file *file)
2720{
2721 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2722}
2723
2724static int gfs2_glocks_release(struct inode *inode, struct file *file)
2725{
2726 struct seq_file *seq = file->private_data;
2727 struct gfs2_glock_iter *gi = seq->private;
2728
2729 if (gi->gl)
2730 gfs2_glock_put(gi->gl);
2731 rhashtable_walk_exit(&gi->hti);
2732 return seq_release_private(inode, file);
2733}
2734
2735static int gfs2_glstats_open(struct inode *inode, struct file *file)
2736{
2737 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2738}
2739
2740static const struct file_operations gfs2_glocks_fops = {
2741 .owner = THIS_MODULE,
2742 .open = gfs2_glocks_open,
2743 .read = seq_read,
2744 .llseek = seq_lseek,
2745 .release = gfs2_glocks_release,
2746};
2747
2748static const struct file_operations gfs2_glstats_fops = {
2749 .owner = THIS_MODULE,
2750 .open = gfs2_glstats_open,
2751 .read = seq_read,
2752 .llseek = seq_lseek,
2753 .release = gfs2_glocks_release,
2754};
2755
2756struct gfs2_glockfd_iter {
2757 struct super_block *sb;
2758 unsigned int tgid;
2759 struct task_struct *task;
2760 unsigned int fd;
2761 struct file *file;
2762};
2763
2764static struct task_struct *gfs2_glockfd_next_task(struct gfs2_glockfd_iter *i)
2765{
2766 struct pid_namespace *ns = task_active_pid_ns(current);
2767 struct pid *pid;
2768
2769 if (i->task)
2770 put_task_struct(i->task);
2771
2772 rcu_read_lock();
2773retry:
2774 i->task = NULL;
2775 pid = find_ge_pid(i->tgid, ns);
2776 if (pid) {
2777 i->tgid = pid_nr_ns(pid, ns);
2778 i->task = pid_task(pid, PIDTYPE_TGID);
2779 if (!i->task) {
2780 i->tgid++;
2781 goto retry;
2782 }
2783 get_task_struct(i->task);
2784 }
2785 rcu_read_unlock();
2786 return i->task;
2787}
2788
2789static struct file *gfs2_glockfd_next_file(struct gfs2_glockfd_iter *i)
2790{
2791 if (i->file) {
2792 fput(i->file);
2793 i->file = NULL;
2794 }
2795
2796 rcu_read_lock();
2797 for(;; i->fd++) {
2798 struct inode *inode;
2799
2800 i->file = task_lookup_next_fdget_rcu(i->task, &i->fd);
2801 if (!i->file) {
2802 i->fd = 0;
2803 break;
2804 }
2805
2806 inode = file_inode(i->file);
2807 if (inode->i_sb == i->sb)
2808 break;
2809
2810 rcu_read_unlock();
2811 fput(i->file);
2812 rcu_read_lock();
2813 }
2814 rcu_read_unlock();
2815 return i->file;
2816}
2817
2818static void *gfs2_glockfd_seq_start(struct seq_file *seq, loff_t *pos)
2819{
2820 struct gfs2_glockfd_iter *i = seq->private;
2821
2822 if (*pos)
2823 return NULL;
2824 while (gfs2_glockfd_next_task(i)) {
2825 if (gfs2_glockfd_next_file(i))
2826 return i;
2827 i->tgid++;
2828 }
2829 return NULL;
2830}
2831
2832static void *gfs2_glockfd_seq_next(struct seq_file *seq, void *iter_ptr,
2833 loff_t *pos)
2834{
2835 struct gfs2_glockfd_iter *i = seq->private;
2836
2837 (*pos)++;
2838 i->fd++;
2839 do {
2840 if (gfs2_glockfd_next_file(i))
2841 return i;
2842 i->tgid++;
2843 } while (gfs2_glockfd_next_task(i));
2844 return NULL;
2845}
2846
2847static void gfs2_glockfd_seq_stop(struct seq_file *seq, void *iter_ptr)
2848{
2849 struct gfs2_glockfd_iter *i = seq->private;
2850
2851 if (i->file)
2852 fput(i->file);
2853 if (i->task)
2854 put_task_struct(i->task);
2855}
2856
2857static void gfs2_glockfd_seq_show_flock(struct seq_file *seq,
2858 struct gfs2_glockfd_iter *i)
2859{
2860 struct gfs2_file *fp = i->file->private_data;
2861 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
2862 struct lm_lockname gl_name = { .ln_type = LM_TYPE_RESERVED };
2863
2864 if (!READ_ONCE(fl_gh->gh_gl))
2865 return;
2866
2867 spin_lock(&i->file->f_lock);
2868 if (gfs2_holder_initialized(fl_gh))
2869 gl_name = fl_gh->gh_gl->gl_name;
2870 spin_unlock(&i->file->f_lock);
2871
2872 if (gl_name.ln_type != LM_TYPE_RESERVED) {
2873 seq_printf(seq, "%d %u %u/%llx\n",
2874 i->tgid, i->fd, gl_name.ln_type,
2875 (unsigned long long)gl_name.ln_number);
2876 }
2877}
2878
2879static int gfs2_glockfd_seq_show(struct seq_file *seq, void *iter_ptr)
2880{
2881 struct gfs2_glockfd_iter *i = seq->private;
2882 struct inode *inode = file_inode(i->file);
2883 struct gfs2_glock *gl;
2884
2885 inode_lock_shared(inode);
2886 gl = GFS2_I(inode)->i_iopen_gh.gh_gl;
2887 if (gl) {
2888 seq_printf(seq, "%d %u %u/%llx\n",
2889 i->tgid, i->fd, gl->gl_name.ln_type,
2890 (unsigned long long)gl->gl_name.ln_number);
2891 }
2892 gfs2_glockfd_seq_show_flock(seq, i);
2893 inode_unlock_shared(inode);
2894 return 0;
2895}
2896
2897static const struct seq_operations gfs2_glockfd_seq_ops = {
2898 .start = gfs2_glockfd_seq_start,
2899 .next = gfs2_glockfd_seq_next,
2900 .stop = gfs2_glockfd_seq_stop,
2901 .show = gfs2_glockfd_seq_show,
2902};
2903
2904static int gfs2_glockfd_open(struct inode *inode, struct file *file)
2905{
2906 struct gfs2_glockfd_iter *i;
2907 struct gfs2_sbd *sdp = inode->i_private;
2908
2909 i = __seq_open_private(file, &gfs2_glockfd_seq_ops,
2910 sizeof(struct gfs2_glockfd_iter));
2911 if (!i)
2912 return -ENOMEM;
2913 i->sb = sdp->sd_vfs;
2914 return 0;
2915}
2916
2917static const struct file_operations gfs2_glockfd_fops = {
2918 .owner = THIS_MODULE,
2919 .open = gfs2_glockfd_open,
2920 .read = seq_read,
2921 .llseek = seq_lseek,
2922 .release = seq_release_private,
2923};
2924
2925DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats);
2926
2927void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2928{
2929 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2930
2931 debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2932 &gfs2_glocks_fops);
2933
2934 debugfs_create_file("glockfd", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2935 &gfs2_glockfd_fops);
2936
2937 debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2938 &gfs2_glstats_fops);
2939
2940 debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2941 &gfs2_sbstats_fops);
2942}
2943
2944void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2945{
2946 debugfs_remove_recursive(sdp->debugfs_dir);
2947 sdp->debugfs_dir = NULL;
2948}
2949
2950void gfs2_register_debugfs(void)
2951{
2952 gfs2_root = debugfs_create_dir("gfs2", NULL);
2953}
2954
2955void gfs2_unregister_debugfs(void)
2956{
2957 debugfs_remove(gfs2_root);
2958 gfs2_root = NULL;
2959}