blob: aea1ed0aebd0f6a615b8118775583e7e8e21ad43 [file] [log] [blame]
rjw1f884582022-01-06 17:20:42 +08001/*
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#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11
12#include <linux/sched.h>
13#include <linux/slab.h>
14#include <linux/spinlock.h>
15#include <linux/buffer_head.h>
16#include <linux/delay.h>
17#include <linux/sort.h>
18#include <linux/hash.h>
19#include <linux/jhash.h>
20#include <linux/kallsyms.h>
21#include <linux/gfs2_ondisk.h>
22#include <linux/list.h>
23#include <linux/wait.h>
24#include <linux/module.h>
25#include <linux/uaccess.h>
26#include <linux/seq_file.h>
27#include <linux/debugfs.h>
28#include <linux/kthread.h>
29#include <linux/freezer.h>
30#include <linux/workqueue.h>
31#include <linux/jiffies.h>
32#include <linux/rcupdate.h>
33#include <linux/rculist_bl.h>
34#include <linux/bit_spinlock.h>
35#include <linux/percpu.h>
36#include <linux/list_sort.h>
37#include <linux/lockref.h>
38#include <linux/rhashtable.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);
64
65static struct dentry *gfs2_root;
66static struct workqueue_struct *glock_workqueue;
67struct workqueue_struct *gfs2_delete_workqueue;
68static LIST_HEAD(lru_list);
69static atomic_t lru_count = ATOMIC_INIT(0);
70static DEFINE_SPINLOCK(lru_lock);
71
72#define GFS2_GL_HASH_SHIFT 15
73#define GFS2_GL_HASH_SIZE BIT(GFS2_GL_HASH_SHIFT)
74
75static const struct rhashtable_params ht_parms = {
76 .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4,
77 .key_len = offsetofend(struct lm_lockname, ln_type),
78 .key_offset = offsetof(struct gfs2_glock, gl_name),
79 .head_offset = offsetof(struct gfs2_glock, gl_node),
80};
81
82static struct rhashtable gl_hash_table;
83
84#define GLOCK_WAIT_TABLE_BITS 12
85#define GLOCK_WAIT_TABLE_SIZE (1 << GLOCK_WAIT_TABLE_BITS)
86static wait_queue_head_t glock_wait_table[GLOCK_WAIT_TABLE_SIZE] __cacheline_aligned;
87
88struct wait_glock_queue {
89 struct lm_lockname *name;
90 wait_queue_entry_t wait;
91};
92
93static int glock_wake_function(wait_queue_entry_t *wait, unsigned int mode,
94 int sync, void *key)
95{
96 struct wait_glock_queue *wait_glock =
97 container_of(wait, struct wait_glock_queue, wait);
98 struct lm_lockname *wait_name = wait_glock->name;
99 struct lm_lockname *wake_name = key;
100
101 if (wake_name->ln_sbd != wait_name->ln_sbd ||
102 wake_name->ln_number != wait_name->ln_number ||
103 wake_name->ln_type != wait_name->ln_type)
104 return 0;
105 return autoremove_wake_function(wait, mode, sync, key);
106}
107
108static wait_queue_head_t *glock_waitqueue(struct lm_lockname *name)
109{
110 u32 hash = jhash2((u32 *)name, ht_parms.key_len / 4, 0);
111
112 return glock_wait_table + hash_32(hash, GLOCK_WAIT_TABLE_BITS);
113}
114
115/**
116 * wake_up_glock - Wake up waiters on a glock
117 * @gl: the glock
118 */
119static void wake_up_glock(struct gfs2_glock *gl)
120{
121 wait_queue_head_t *wq = glock_waitqueue(&gl->gl_name);
122
123 if (waitqueue_active(wq))
124 __wake_up(wq, TASK_NORMAL, 1, &gl->gl_name);
125}
126
127static void gfs2_glock_dealloc(struct rcu_head *rcu)
128{
129 struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
130
131 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
132 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
133 } else {
134 kfree(gl->gl_lksb.sb_lvbptr);
135 kmem_cache_free(gfs2_glock_cachep, gl);
136 }
137}
138
139void gfs2_glock_free(struct gfs2_glock *gl)
140{
141 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
142
143 BUG_ON(atomic_read(&gl->gl_revokes));
144 rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
145 smp_mb();
146 wake_up_glock(gl);
147 call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
148 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
149 wake_up(&sdp->sd_glock_wait);
150}
151
152/**
153 * gfs2_glock_hold() - increment reference count on glock
154 * @gl: The glock to hold
155 *
156 */
157
158void gfs2_glock_hold(struct gfs2_glock *gl)
159{
160 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
161 lockref_get(&gl->gl_lockref);
162}
163
164/**
165 * demote_ok - Check to see if it's ok to unlock a glock
166 * @gl: the glock
167 *
168 * Returns: 1 if it's ok
169 */
170
171static int demote_ok(const struct gfs2_glock *gl)
172{
173 const struct gfs2_glock_operations *glops = gl->gl_ops;
174
175 if (gl->gl_state == LM_ST_UNLOCKED)
176 return 0;
177 if (!list_empty(&gl->gl_holders))
178 return 0;
179 if (glops->go_demote_ok)
180 return glops->go_demote_ok(gl);
181 return 1;
182}
183
184
185void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
186{
187 if (!(gl->gl_ops->go_flags & GLOF_LRU))
188 return;
189
190 spin_lock(&lru_lock);
191
192 list_del(&gl->gl_lru);
193 list_add_tail(&gl->gl_lru, &lru_list);
194
195 if (!test_bit(GLF_LRU, &gl->gl_flags)) {
196 set_bit(GLF_LRU, &gl->gl_flags);
197 atomic_inc(&lru_count);
198 }
199
200 spin_unlock(&lru_lock);
201}
202
203static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
204{
205 if (!(gl->gl_ops->go_flags & GLOF_LRU))
206 return;
207
208 spin_lock(&lru_lock);
209 if (test_bit(GLF_LRU, &gl->gl_flags)) {
210 list_del_init(&gl->gl_lru);
211 atomic_dec(&lru_count);
212 clear_bit(GLF_LRU, &gl->gl_flags);
213 }
214 spin_unlock(&lru_lock);
215}
216
217/*
218 * Enqueue the glock on the work queue. Passes one glock reference on to the
219 * work queue.
220 */
221static void __gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
222 if (!queue_delayed_work(glock_workqueue, &gl->gl_work, delay)) {
223 /*
224 * We are holding the lockref spinlock, and the work was still
225 * queued above. The queued work (glock_work_func) takes that
226 * spinlock before dropping its glock reference(s), so it
227 * cannot have dropped them in the meantime.
228 */
229 GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2);
230 gl->gl_lockref.count--;
231 }
232}
233
234static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
235 spin_lock(&gl->gl_lockref.lock);
236 __gfs2_glock_queue_work(gl, delay);
237 spin_unlock(&gl->gl_lockref.lock);
238}
239
240static void __gfs2_glock_put(struct gfs2_glock *gl)
241{
242 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
243 struct address_space *mapping = gfs2_glock2aspace(gl);
244
245 lockref_mark_dead(&gl->gl_lockref);
246
247 gfs2_glock_remove_from_lru(gl);
248 spin_unlock(&gl->gl_lockref.lock);
249 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
250 GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
251 trace_gfs2_glock_put(gl);
252 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
253}
254
255/*
256 * Cause the glock to be put in work queue context.
257 */
258void gfs2_glock_queue_put(struct gfs2_glock *gl)
259{
260 gfs2_glock_queue_work(gl, 0);
261}
262
263/**
264 * gfs2_glock_put() - Decrement reference count on glock
265 * @gl: The glock to put
266 *
267 */
268
269void gfs2_glock_put(struct gfs2_glock *gl)
270{
271 if (lockref_put_or_lock(&gl->gl_lockref))
272 return;
273
274 __gfs2_glock_put(gl);
275}
276
277/**
278 * may_grant - check if its ok to grant a new lock
279 * @gl: The glock
280 * @gh: The lock request which we wish to grant
281 *
282 * Returns: true if its ok to grant the lock
283 */
284
285static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
286{
287 const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
288 if ((gh->gh_state == LM_ST_EXCLUSIVE ||
289 gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
290 return 0;
291 if (gl->gl_state == gh->gh_state)
292 return 1;
293 if (gh->gh_flags & GL_EXACT)
294 return 0;
295 if (gl->gl_state == LM_ST_EXCLUSIVE) {
296 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
297 return 1;
298 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
299 return 1;
300 }
301 if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
302 return 1;
303 return 0;
304}
305
306static void gfs2_holder_wake(struct gfs2_holder *gh)
307{
308 clear_bit(HIF_WAIT, &gh->gh_iflags);
309 smp_mb__after_atomic();
310 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
311}
312
313/**
314 * do_error - Something unexpected has happened during a lock request
315 *
316 */
317
318static void do_error(struct gfs2_glock *gl, const int ret)
319{
320 struct gfs2_holder *gh, *tmp;
321
322 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
323 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
324 continue;
325 if (ret & LM_OUT_ERROR)
326 gh->gh_error = -EIO;
327 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
328 gh->gh_error = GLR_TRYFAILED;
329 else
330 continue;
331 list_del_init(&gh->gh_list);
332 trace_gfs2_glock_queue(gh, 0);
333 gfs2_holder_wake(gh);
334 }
335}
336
337/**
338 * do_promote - promote as many requests as possible on the current queue
339 * @gl: The glock
340 *
341 * Returns: 1 if there is a blocked holder at the head of the list, or 2
342 * if a type specific operation is underway.
343 */
344
345static int do_promote(struct gfs2_glock *gl)
346__releases(&gl->gl_lockref.lock)
347__acquires(&gl->gl_lockref.lock)
348{
349 const struct gfs2_glock_operations *glops = gl->gl_ops;
350 struct gfs2_holder *gh, *tmp;
351 int ret;
352
353restart:
354 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
355 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
356 continue;
357 if (may_grant(gl, gh)) {
358 if (gh->gh_list.prev == &gl->gl_holders &&
359 glops->go_lock) {
360 spin_unlock(&gl->gl_lockref.lock);
361 /* FIXME: eliminate this eventually */
362 ret = glops->go_lock(gh);
363 spin_lock(&gl->gl_lockref.lock);
364 if (ret) {
365 if (ret == 1)
366 return 2;
367 gh->gh_error = ret;
368 list_del_init(&gh->gh_list);
369 trace_gfs2_glock_queue(gh, 0);
370 gfs2_holder_wake(gh);
371 goto restart;
372 }
373 set_bit(HIF_HOLDER, &gh->gh_iflags);
374 trace_gfs2_promote(gh, 1);
375 gfs2_holder_wake(gh);
376 goto restart;
377 }
378 set_bit(HIF_HOLDER, &gh->gh_iflags);
379 trace_gfs2_promote(gh, 0);
380 gfs2_holder_wake(gh);
381 continue;
382 }
383 if (gh->gh_list.prev == &gl->gl_holders)
384 return 1;
385 do_error(gl, 0);
386 break;
387 }
388 return 0;
389}
390
391/**
392 * find_first_waiter - find the first gh that's waiting for the glock
393 * @gl: the glock
394 */
395
396static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
397{
398 struct gfs2_holder *gh;
399
400 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
401 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
402 return gh;
403 }
404 return NULL;
405}
406
407/**
408 * state_change - record that the glock is now in a different state
409 * @gl: the glock
410 * @new_state the new state
411 *
412 */
413
414static void state_change(struct gfs2_glock *gl, unsigned int new_state)
415{
416 int held1, held2;
417
418 held1 = (gl->gl_state != LM_ST_UNLOCKED);
419 held2 = (new_state != LM_ST_UNLOCKED);
420
421 if (held1 != held2) {
422 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
423 if (held2)
424 gl->gl_lockref.count++;
425 else
426 gl->gl_lockref.count--;
427 }
428 if (held1 && held2 && list_empty(&gl->gl_holders))
429 clear_bit(GLF_QUEUED, &gl->gl_flags);
430
431 if (new_state != gl->gl_target)
432 /* shorten our minimum hold time */
433 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
434 GL_GLOCK_MIN_HOLD);
435 gl->gl_state = new_state;
436 gl->gl_tchange = jiffies;
437}
438
439static void gfs2_demote_wake(struct gfs2_glock *gl)
440{
441 gl->gl_demote_state = LM_ST_EXCLUSIVE;
442 clear_bit(GLF_DEMOTE, &gl->gl_flags);
443 smp_mb__after_atomic();
444 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
445}
446
447/**
448 * finish_xmote - The DLM has replied to one of our lock requests
449 * @gl: The glock
450 * @ret: The status from the DLM
451 *
452 */
453
454static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
455{
456 const struct gfs2_glock_operations *glops = gl->gl_ops;
457 struct gfs2_holder *gh;
458 unsigned state = ret & LM_OUT_ST_MASK;
459 int rv;
460
461 spin_lock(&gl->gl_lockref.lock);
462 trace_gfs2_glock_state_change(gl, state);
463 state_change(gl, state);
464 gh = find_first_waiter(gl);
465
466 /* Demote to UN request arrived during demote to SH or DF */
467 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
468 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
469 gl->gl_target = LM_ST_UNLOCKED;
470
471 /* Check for state != intended state */
472 if (unlikely(state != gl->gl_target)) {
473 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
474 /* move to back of queue and try next entry */
475 if (ret & LM_OUT_CANCELED) {
476 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
477 list_move_tail(&gh->gh_list, &gl->gl_holders);
478 gh = find_first_waiter(gl);
479 gl->gl_target = gh->gh_state;
480 goto retry;
481 }
482 /* Some error or failed "try lock" - report it */
483 if ((ret & LM_OUT_ERROR) ||
484 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
485 gl->gl_target = gl->gl_state;
486 do_error(gl, ret);
487 goto out;
488 }
489 }
490 switch(state) {
491 /* Unlocked due to conversion deadlock, try again */
492 case LM_ST_UNLOCKED:
493retry:
494 do_xmote(gl, gh, gl->gl_target);
495 break;
496 /* Conversion fails, unlock and try again */
497 case LM_ST_SHARED:
498 case LM_ST_DEFERRED:
499 do_xmote(gl, gh, LM_ST_UNLOCKED);
500 break;
501 default: /* Everything else */
502 pr_err("wanted %u got %u\n", gl->gl_target, state);
503 GLOCK_BUG_ON(gl, 1);
504 }
505 spin_unlock(&gl->gl_lockref.lock);
506 return;
507 }
508
509 /* Fast path - we got what we asked for */
510 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
511 gfs2_demote_wake(gl);
512 if (state != LM_ST_UNLOCKED) {
513 if (glops->go_xmote_bh) {
514 spin_unlock(&gl->gl_lockref.lock);
515 rv = glops->go_xmote_bh(gl, gh);
516 spin_lock(&gl->gl_lockref.lock);
517 if (rv) {
518 do_error(gl, rv);
519 goto out;
520 }
521 }
522 rv = do_promote(gl);
523 if (rv == 2)
524 goto out_locked;
525 }
526out:
527 clear_bit(GLF_LOCK, &gl->gl_flags);
528out_locked:
529 spin_unlock(&gl->gl_lockref.lock);
530}
531
532/**
533 * do_xmote - Calls the DLM to change the state of a lock
534 * @gl: The lock state
535 * @gh: The holder (only for promotes)
536 * @target: The target lock state
537 *
538 */
539
540static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
541__releases(&gl->gl_lockref.lock)
542__acquires(&gl->gl_lockref.lock)
543{
544 const struct gfs2_glock_operations *glops = gl->gl_ops;
545 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
546 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
547 int ret;
548
549 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
550 target != LM_ST_UNLOCKED)
551 return;
552 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
553 LM_FLAG_PRIORITY);
554 GLOCK_BUG_ON(gl, gl->gl_state == target);
555 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
556 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
557 glops->go_inval) {
558 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
559 do_error(gl, 0); /* Fail queued try locks */
560 }
561 gl->gl_req = target;
562 set_bit(GLF_BLOCKING, &gl->gl_flags);
563 if ((gl->gl_req == LM_ST_UNLOCKED) ||
564 (gl->gl_state == LM_ST_EXCLUSIVE) ||
565 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
566 clear_bit(GLF_BLOCKING, &gl->gl_flags);
567 spin_unlock(&gl->gl_lockref.lock);
568 if (glops->go_sync)
569 glops->go_sync(gl);
570 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
571 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
572 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
573
574 gfs2_glock_hold(gl);
575 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
576 /* lock_dlm */
577 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
578 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
579 target == LM_ST_UNLOCKED &&
580 test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
581 finish_xmote(gl, target);
582 gfs2_glock_queue_work(gl, 0);
583 }
584 else if (ret) {
585 pr_err("lm_lock ret %d\n", ret);
586 GLOCK_BUG_ON(gl, !test_bit(SDF_SHUTDOWN,
587 &sdp->sd_flags));
588 }
589 } else { /* lock_nolock */
590 finish_xmote(gl, target);
591 gfs2_glock_queue_work(gl, 0);
592 }
593
594 spin_lock(&gl->gl_lockref.lock);
595}
596
597/**
598 * find_first_holder - find the first "holder" gh
599 * @gl: the glock
600 */
601
602static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
603{
604 struct gfs2_holder *gh;
605
606 if (!list_empty(&gl->gl_holders)) {
607 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
608 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
609 return gh;
610 }
611 return NULL;
612}
613
614/**
615 * run_queue - do all outstanding tasks related to a glock
616 * @gl: The glock in question
617 * @nonblock: True if we must not block in run_queue
618 *
619 */
620
621static void run_queue(struct gfs2_glock *gl, const int nonblock)
622__releases(&gl->gl_lockref.lock)
623__acquires(&gl->gl_lockref.lock)
624{
625 struct gfs2_holder *gh = NULL;
626 int ret;
627
628 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
629 return;
630
631 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
632
633 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
634 gl->gl_demote_state != gl->gl_state) {
635 if (find_first_holder(gl))
636 goto out_unlock;
637 if (nonblock)
638 goto out_sched;
639 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
640 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
641 gl->gl_target = gl->gl_demote_state;
642 } else {
643 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
644 gfs2_demote_wake(gl);
645 ret = do_promote(gl);
646 if (ret == 0)
647 goto out_unlock;
648 if (ret == 2)
649 goto out;
650 gh = find_first_waiter(gl);
651 gl->gl_target = gh->gh_state;
652 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
653 do_error(gl, 0); /* Fail queued try locks */
654 }
655 do_xmote(gl, gh, gl->gl_target);
656out:
657 return;
658
659out_sched:
660 clear_bit(GLF_LOCK, &gl->gl_flags);
661 smp_mb__after_atomic();
662 gl->gl_lockref.count++;
663 __gfs2_glock_queue_work(gl, 0);
664 return;
665
666out_unlock:
667 clear_bit(GLF_LOCK, &gl->gl_flags);
668 smp_mb__after_atomic();
669 return;
670}
671
672static void delete_work_func(struct work_struct *work)
673{
674 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
675 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
676 struct inode *inode;
677 u64 no_addr = gl->gl_name.ln_number;
678
679 /* If someone's using this glock to create a new dinode, the block must
680 have been freed by another node, then re-used, in which case our
681 iopen callback is too late after the fact. Ignore it. */
682 if (test_bit(GLF_INODE_CREATING, &gl->gl_flags))
683 goto out;
684
685 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
686 if (inode && !IS_ERR(inode)) {
687 d_prune_aliases(inode);
688 iput(inode);
689 }
690out:
691 gfs2_glock_put(gl);
692}
693
694static void glock_work_func(struct work_struct *work)
695{
696 unsigned long delay = 0;
697 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
698 unsigned int drop_refs = 1;
699
700 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
701 finish_xmote(gl, gl->gl_reply);
702 drop_refs++;
703 }
704 spin_lock(&gl->gl_lockref.lock);
705 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
706 gl->gl_state != LM_ST_UNLOCKED &&
707 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
708 unsigned long holdtime, now = jiffies;
709
710 holdtime = gl->gl_tchange + gl->gl_hold_time;
711 if (time_before(now, holdtime))
712 delay = holdtime - now;
713
714 if (!delay) {
715 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
716 set_bit(GLF_DEMOTE, &gl->gl_flags);
717 }
718 }
719 run_queue(gl, 0);
720 if (delay) {
721 /* Keep one glock reference for the work we requeue. */
722 drop_refs--;
723 if (gl->gl_name.ln_type != LM_TYPE_INODE)
724 delay = 0;
725 __gfs2_glock_queue_work(gl, delay);
726 }
727
728 /*
729 * Drop the remaining glock references manually here. (Mind that
730 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
731 * here as well.)
732 */
733 gl->gl_lockref.count -= drop_refs;
734 if (!gl->gl_lockref.count) {
735 __gfs2_glock_put(gl);
736 return;
737 }
738 spin_unlock(&gl->gl_lockref.lock);
739}
740
741static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
742 struct gfs2_glock *new)
743{
744 struct wait_glock_queue wait;
745 wait_queue_head_t *wq = glock_waitqueue(name);
746 struct gfs2_glock *gl;
747
748 wait.name = name;
749 init_wait(&wait.wait);
750 wait.wait.func = glock_wake_function;
751
752again:
753 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
754 rcu_read_lock();
755 if (new) {
756 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
757 &new->gl_node, ht_parms);
758 if (IS_ERR(gl))
759 goto out;
760 } else {
761 gl = rhashtable_lookup_fast(&gl_hash_table,
762 name, ht_parms);
763 }
764 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
765 rcu_read_unlock();
766 schedule();
767 goto again;
768 }
769out:
770 rcu_read_unlock();
771 finish_wait(wq, &wait.wait);
772 return gl;
773}
774
775/**
776 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
777 * @sdp: The GFS2 superblock
778 * @number: the lock number
779 * @glops: The glock_operations to use
780 * @create: If 0, don't create the glock if it doesn't exist
781 * @glp: the glock is returned here
782 *
783 * This does not lock a glock, just finds/creates structures for one.
784 *
785 * Returns: errno
786 */
787
788int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
789 const struct gfs2_glock_operations *glops, int create,
790 struct gfs2_glock **glp)
791{
792 struct super_block *s = sdp->sd_vfs;
793 struct lm_lockname name = { .ln_number = number,
794 .ln_type = glops->go_type,
795 .ln_sbd = sdp };
796 struct gfs2_glock *gl, *tmp;
797 struct address_space *mapping;
798 struct kmem_cache *cachep;
799 int ret = 0;
800
801 gl = find_insert_glock(&name, NULL);
802 if (gl) {
803 *glp = gl;
804 return 0;
805 }
806 if (!create)
807 return -ENOENT;
808
809 if (glops->go_flags & GLOF_ASPACE)
810 cachep = gfs2_glock_aspace_cachep;
811 else
812 cachep = gfs2_glock_cachep;
813 gl = kmem_cache_alloc(cachep, GFP_NOFS);
814 if (!gl)
815 return -ENOMEM;
816
817 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
818
819 if (glops->go_flags & GLOF_LVB) {
820 gl->gl_lksb.sb_lvbptr = kzalloc(GFS2_MIN_LVB_SIZE, GFP_NOFS);
821 if (!gl->gl_lksb.sb_lvbptr) {
822 kmem_cache_free(cachep, gl);
823 return -ENOMEM;
824 }
825 }
826
827 atomic_inc(&sdp->sd_glock_disposal);
828 gl->gl_node.next = NULL;
829 gl->gl_flags = 0;
830 gl->gl_name = name;
831 gl->gl_lockref.count = 1;
832 gl->gl_state = LM_ST_UNLOCKED;
833 gl->gl_target = LM_ST_UNLOCKED;
834 gl->gl_demote_state = LM_ST_EXCLUSIVE;
835 gl->gl_ops = glops;
836 gl->gl_dstamp = 0;
837 preempt_disable();
838 /* We use the global stats to estimate the initial per-glock stats */
839 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
840 preempt_enable();
841 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
842 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
843 gl->gl_tchange = jiffies;
844 gl->gl_object = NULL;
845 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
846 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
847 INIT_WORK(&gl->gl_delete, delete_work_func);
848
849 mapping = gfs2_glock2aspace(gl);
850 if (mapping) {
851 mapping->a_ops = &gfs2_meta_aops;
852 mapping->host = s->s_bdev->bd_inode;
853 mapping->flags = 0;
854 mapping_set_gfp_mask(mapping, GFP_NOFS);
855 mapping->private_data = NULL;
856 mapping->writeback_index = 0;
857 }
858
859 tmp = find_insert_glock(&name, gl);
860 if (!tmp) {
861 *glp = gl;
862 goto out;
863 }
864 if (IS_ERR(tmp)) {
865 ret = PTR_ERR(tmp);
866 goto out_free;
867 }
868 *glp = tmp;
869
870out_free:
871 kfree(gl->gl_lksb.sb_lvbptr);
872 kmem_cache_free(cachep, gl);
873 atomic_dec(&sdp->sd_glock_disposal);
874
875out:
876 return ret;
877}
878
879/**
880 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
881 * @gl: the glock
882 * @state: the state we're requesting
883 * @flags: the modifier flags
884 * @gh: the holder structure
885 *
886 */
887
888void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
889 struct gfs2_holder *gh)
890{
891 INIT_LIST_HEAD(&gh->gh_list);
892 gh->gh_gl = gl;
893 gh->gh_ip = _RET_IP_;
894 gh->gh_owner_pid = get_pid(task_pid(current));
895 gh->gh_state = state;
896 gh->gh_flags = flags;
897 gh->gh_error = 0;
898 gh->gh_iflags = 0;
899 gfs2_glock_hold(gl);
900}
901
902/**
903 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
904 * @state: the state we're requesting
905 * @flags: the modifier flags
906 * @gh: the holder structure
907 *
908 * Don't mess with the glock.
909 *
910 */
911
912void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
913{
914 gh->gh_state = state;
915 gh->gh_flags = flags;
916 gh->gh_iflags = 0;
917 gh->gh_ip = _RET_IP_;
918 put_pid(gh->gh_owner_pid);
919 gh->gh_owner_pid = get_pid(task_pid(current));
920}
921
922/**
923 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
924 * @gh: the holder structure
925 *
926 */
927
928void gfs2_holder_uninit(struct gfs2_holder *gh)
929{
930 put_pid(gh->gh_owner_pid);
931 gfs2_glock_put(gh->gh_gl);
932 gfs2_holder_mark_uninitialized(gh);
933 gh->gh_ip = 0;
934}
935
936/**
937 * gfs2_glock_wait - wait on a glock acquisition
938 * @gh: the glock holder
939 *
940 * Returns: 0 on success
941 */
942
943int gfs2_glock_wait(struct gfs2_holder *gh)
944{
945 unsigned long time1 = jiffies;
946
947 might_sleep();
948 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
949 if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
950 /* Lengthen the minimum hold time. */
951 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
952 GL_GLOCK_HOLD_INCR,
953 GL_GLOCK_MAX_HOLD);
954 return gh->gh_error;
955}
956
957/**
958 * handle_callback - process a demote request
959 * @gl: the glock
960 * @state: the state the caller wants us to change to
961 *
962 * There are only two requests that we are going to see in actual
963 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
964 */
965
966static void handle_callback(struct gfs2_glock *gl, unsigned int state,
967 unsigned long delay, bool remote)
968{
969 int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
970
971 set_bit(bit, &gl->gl_flags);
972 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
973 gl->gl_demote_state = state;
974 gl->gl_demote_time = jiffies;
975 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
976 gl->gl_demote_state != state) {
977 gl->gl_demote_state = LM_ST_UNLOCKED;
978 }
979 if (gl->gl_ops->go_callback)
980 gl->gl_ops->go_callback(gl, remote);
981 trace_gfs2_demote_rq(gl, remote);
982}
983
984void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
985{
986 struct va_format vaf;
987 va_list args;
988
989 va_start(args, fmt);
990
991 if (seq) {
992 seq_vprintf(seq, fmt, args);
993 } else {
994 vaf.fmt = fmt;
995 vaf.va = &args;
996
997 pr_err("%pV", &vaf);
998 }
999
1000 va_end(args);
1001}
1002
1003/**
1004 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1005 * @gh: the holder structure to add
1006 *
1007 * Eventually we should move the recursive locking trap to a
1008 * debugging option or something like that. This is the fast
1009 * path and needs to have the minimum number of distractions.
1010 *
1011 */
1012
1013static inline void add_to_queue(struct gfs2_holder *gh)
1014__releases(&gl->gl_lockref.lock)
1015__acquires(&gl->gl_lockref.lock)
1016{
1017 struct gfs2_glock *gl = gh->gh_gl;
1018 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1019 struct list_head *insert_pt = NULL;
1020 struct gfs2_holder *gh2;
1021 int try_futile = 0;
1022
1023 BUG_ON(gh->gh_owner_pid == NULL);
1024 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1025 BUG();
1026
1027 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1028 if (test_bit(GLF_LOCK, &gl->gl_flags))
1029 try_futile = !may_grant(gl, gh);
1030 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1031 goto fail;
1032 }
1033
1034 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1035 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
1036 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
1037 goto trap_recursive;
1038 if (try_futile &&
1039 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1040fail:
1041 gh->gh_error = GLR_TRYFAILED;
1042 gfs2_holder_wake(gh);
1043 return;
1044 }
1045 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1046 continue;
1047 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1048 insert_pt = &gh2->gh_list;
1049 }
1050 set_bit(GLF_QUEUED, &gl->gl_flags);
1051 trace_gfs2_glock_queue(gh, 1);
1052 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1053 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1054 if (likely(insert_pt == NULL)) {
1055 list_add_tail(&gh->gh_list, &gl->gl_holders);
1056 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1057 goto do_cancel;
1058 return;
1059 }
1060 list_add_tail(&gh->gh_list, insert_pt);
1061do_cancel:
1062 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
1063 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1064 spin_unlock(&gl->gl_lockref.lock);
1065 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1066 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1067 spin_lock(&gl->gl_lockref.lock);
1068 }
1069 return;
1070
1071trap_recursive:
1072 pr_err("original: %pSR\n", (void *)gh2->gh_ip);
1073 pr_err("pid: %d\n", pid_nr(gh2->gh_owner_pid));
1074 pr_err("lock type: %d req lock state : %d\n",
1075 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1076 pr_err("new: %pSR\n", (void *)gh->gh_ip);
1077 pr_err("pid: %d\n", pid_nr(gh->gh_owner_pid));
1078 pr_err("lock type: %d req lock state : %d\n",
1079 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1080 gfs2_dump_glock(NULL, gl);
1081 BUG();
1082}
1083
1084/**
1085 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1086 * @gh: the holder structure
1087 *
1088 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1089 *
1090 * Returns: 0, GLR_TRYFAILED, or errno on failure
1091 */
1092
1093int gfs2_glock_nq(struct gfs2_holder *gh)
1094{
1095 struct gfs2_glock *gl = gh->gh_gl;
1096 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1097 int error = 0;
1098
1099 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1100 return -EIO;
1101
1102 if (test_bit(GLF_LRU, &gl->gl_flags))
1103 gfs2_glock_remove_from_lru(gl);
1104
1105 spin_lock(&gl->gl_lockref.lock);
1106 add_to_queue(gh);
1107 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1108 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1109 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1110 gl->gl_lockref.count++;
1111 __gfs2_glock_queue_work(gl, 0);
1112 }
1113 run_queue(gl, 1);
1114 spin_unlock(&gl->gl_lockref.lock);
1115
1116 if (!(gh->gh_flags & GL_ASYNC))
1117 error = gfs2_glock_wait(gh);
1118
1119 return error;
1120}
1121
1122/**
1123 * gfs2_glock_poll - poll to see if an async request has been completed
1124 * @gh: the holder
1125 *
1126 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1127 */
1128
1129int gfs2_glock_poll(struct gfs2_holder *gh)
1130{
1131 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1132}
1133
1134/**
1135 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1136 * @gh: the glock holder
1137 *
1138 */
1139
1140void gfs2_glock_dq(struct gfs2_holder *gh)
1141{
1142 struct gfs2_glock *gl = gh->gh_gl;
1143 const struct gfs2_glock_operations *glops = gl->gl_ops;
1144 unsigned delay = 0;
1145 int fast_path = 0;
1146
1147 spin_lock(&gl->gl_lockref.lock);
1148 if (gh->gh_flags & GL_NOCACHE)
1149 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1150
1151 list_del_init(&gh->gh_list);
1152 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1153 if (find_first_holder(gl) == NULL) {
1154 if (glops->go_unlock) {
1155 GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1156 spin_unlock(&gl->gl_lockref.lock);
1157 glops->go_unlock(gh);
1158 spin_lock(&gl->gl_lockref.lock);
1159 clear_bit(GLF_LOCK, &gl->gl_flags);
1160 }
1161 if (list_empty(&gl->gl_holders) &&
1162 !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1163 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1164 fast_path = 1;
1165 }
1166 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1167 gfs2_glock_add_to_lru(gl);
1168
1169 trace_gfs2_glock_queue(gh, 0);
1170 if (unlikely(!fast_path)) {
1171 gl->gl_lockref.count++;
1172 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1173 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1174 gl->gl_name.ln_type == LM_TYPE_INODE)
1175 delay = gl->gl_hold_time;
1176 __gfs2_glock_queue_work(gl, delay);
1177 }
1178 spin_unlock(&gl->gl_lockref.lock);
1179}
1180
1181void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1182{
1183 struct gfs2_glock *gl = gh->gh_gl;
1184 gfs2_glock_dq(gh);
1185 might_sleep();
1186 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1187}
1188
1189/**
1190 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1191 * @gh: the holder structure
1192 *
1193 */
1194
1195void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1196{
1197 gfs2_glock_dq(gh);
1198 gfs2_holder_uninit(gh);
1199}
1200
1201/**
1202 * gfs2_glock_nq_num - acquire a glock based on lock number
1203 * @sdp: the filesystem
1204 * @number: the lock number
1205 * @glops: the glock operations for the type of glock
1206 * @state: the state to acquire the glock in
1207 * @flags: modifier flags for the acquisition
1208 * @gh: the struct gfs2_holder
1209 *
1210 * Returns: errno
1211 */
1212
1213int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1214 const struct gfs2_glock_operations *glops,
1215 unsigned int state, u16 flags, struct gfs2_holder *gh)
1216{
1217 struct gfs2_glock *gl;
1218 int error;
1219
1220 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1221 if (!error) {
1222 error = gfs2_glock_nq_init(gl, state, flags, gh);
1223 gfs2_glock_put(gl);
1224 }
1225
1226 return error;
1227}
1228
1229/**
1230 * glock_compare - Compare two struct gfs2_glock structures for sorting
1231 * @arg_a: the first structure
1232 * @arg_b: the second structure
1233 *
1234 */
1235
1236static int glock_compare(const void *arg_a, const void *arg_b)
1237{
1238 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1239 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1240 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1241 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1242
1243 if (a->ln_number > b->ln_number)
1244 return 1;
1245 if (a->ln_number < b->ln_number)
1246 return -1;
1247 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1248 return 0;
1249}
1250
1251/**
1252 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1253 * @num_gh: the number of structures
1254 * @ghs: an array of struct gfs2_holder structures
1255 *
1256 * Returns: 0 on success (all glocks acquired),
1257 * errno on failure (no glocks acquired)
1258 */
1259
1260static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1261 struct gfs2_holder **p)
1262{
1263 unsigned int x;
1264 int error = 0;
1265
1266 for (x = 0; x < num_gh; x++)
1267 p[x] = &ghs[x];
1268
1269 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1270
1271 for (x = 0; x < num_gh; x++) {
1272 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1273
1274 error = gfs2_glock_nq(p[x]);
1275 if (error) {
1276 while (x--)
1277 gfs2_glock_dq(p[x]);
1278 break;
1279 }
1280 }
1281
1282 return error;
1283}
1284
1285/**
1286 * gfs2_glock_nq_m - acquire multiple glocks
1287 * @num_gh: the number of structures
1288 * @ghs: an array of struct gfs2_holder structures
1289 *
1290 *
1291 * Returns: 0 on success (all glocks acquired),
1292 * errno on failure (no glocks acquired)
1293 */
1294
1295int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1296{
1297 struct gfs2_holder *tmp[4];
1298 struct gfs2_holder **pph = tmp;
1299 int error = 0;
1300
1301 switch(num_gh) {
1302 case 0:
1303 return 0;
1304 case 1:
1305 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1306 return gfs2_glock_nq(ghs);
1307 default:
1308 if (num_gh <= 4)
1309 break;
1310 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1311 if (!pph)
1312 return -ENOMEM;
1313 }
1314
1315 error = nq_m_sync(num_gh, ghs, pph);
1316
1317 if (pph != tmp)
1318 kfree(pph);
1319
1320 return error;
1321}
1322
1323/**
1324 * gfs2_glock_dq_m - release multiple glocks
1325 * @num_gh: the number of structures
1326 * @ghs: an array of struct gfs2_holder structures
1327 *
1328 */
1329
1330void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1331{
1332 while (num_gh--)
1333 gfs2_glock_dq(&ghs[num_gh]);
1334}
1335
1336void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1337{
1338 unsigned long delay = 0;
1339 unsigned long holdtime;
1340 unsigned long now = jiffies;
1341
1342 gfs2_glock_hold(gl);
1343 holdtime = gl->gl_tchange + gl->gl_hold_time;
1344 if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1345 gl->gl_name.ln_type == LM_TYPE_INODE) {
1346 if (time_before(now, holdtime))
1347 delay = holdtime - now;
1348 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1349 delay = gl->gl_hold_time;
1350 }
1351
1352 spin_lock(&gl->gl_lockref.lock);
1353 handle_callback(gl, state, delay, true);
1354 __gfs2_glock_queue_work(gl, delay);
1355 spin_unlock(&gl->gl_lockref.lock);
1356}
1357
1358/**
1359 * gfs2_should_freeze - Figure out if glock should be frozen
1360 * @gl: The glock in question
1361 *
1362 * Glocks are not frozen if (a) the result of the dlm operation is
1363 * an error, (b) the locking operation was an unlock operation or
1364 * (c) if there is a "noexp" flagged request anywhere in the queue
1365 *
1366 * Returns: 1 if freezing should occur, 0 otherwise
1367 */
1368
1369static int gfs2_should_freeze(const struct gfs2_glock *gl)
1370{
1371 const struct gfs2_holder *gh;
1372
1373 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1374 return 0;
1375 if (gl->gl_target == LM_ST_UNLOCKED)
1376 return 0;
1377
1378 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1379 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1380 continue;
1381 if (LM_FLAG_NOEXP & gh->gh_flags)
1382 return 0;
1383 }
1384
1385 return 1;
1386}
1387
1388/**
1389 * gfs2_glock_complete - Callback used by locking
1390 * @gl: Pointer to the glock
1391 * @ret: The return value from the dlm
1392 *
1393 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1394 * to use a bitfield shared with other glock state fields.
1395 */
1396
1397void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1398{
1399 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1400
1401 spin_lock(&gl->gl_lockref.lock);
1402 gl->gl_reply = ret;
1403
1404 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1405 if (gfs2_should_freeze(gl)) {
1406 set_bit(GLF_FROZEN, &gl->gl_flags);
1407 spin_unlock(&gl->gl_lockref.lock);
1408 return;
1409 }
1410 }
1411
1412 gl->gl_lockref.count++;
1413 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1414 __gfs2_glock_queue_work(gl, 0);
1415 spin_unlock(&gl->gl_lockref.lock);
1416}
1417
1418static int glock_cmp(void *priv, struct list_head *a, struct list_head *b)
1419{
1420 struct gfs2_glock *gla, *glb;
1421
1422 gla = list_entry(a, struct gfs2_glock, gl_lru);
1423 glb = list_entry(b, struct gfs2_glock, gl_lru);
1424
1425 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1426 return 1;
1427 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1428 return -1;
1429
1430 return 0;
1431}
1432
1433/**
1434 * gfs2_dispose_glock_lru - Demote a list of glocks
1435 * @list: The list to dispose of
1436 *
1437 * Disposing of glocks may involve disk accesses, so that here we sort
1438 * the glocks by number (i.e. disk location of the inodes) so that if
1439 * there are any such accesses, they'll be sent in order (mostly).
1440 *
1441 * Must be called under the lru_lock, but may drop and retake this
1442 * lock. While the lru_lock is dropped, entries may vanish from the
1443 * list, but no new entries will appear on the list (since it is
1444 * private)
1445 */
1446
1447static void gfs2_dispose_glock_lru(struct list_head *list)
1448__releases(&lru_lock)
1449__acquires(&lru_lock)
1450{
1451 struct gfs2_glock *gl;
1452
1453 list_sort(NULL, list, glock_cmp);
1454
1455 while(!list_empty(list)) {
1456 gl = list_entry(list->next, struct gfs2_glock, gl_lru);
1457 list_del_init(&gl->gl_lru);
1458 if (!spin_trylock(&gl->gl_lockref.lock)) {
1459add_back_to_lru:
1460 list_add(&gl->gl_lru, &lru_list);
1461 set_bit(GLF_LRU, &gl->gl_flags);
1462 atomic_inc(&lru_count);
1463 continue;
1464 }
1465 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1466 spin_unlock(&gl->gl_lockref.lock);
1467 goto add_back_to_lru;
1468 }
1469 gl->gl_lockref.count++;
1470 if (demote_ok(gl))
1471 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1472 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1473 __gfs2_glock_queue_work(gl, 0);
1474 spin_unlock(&gl->gl_lockref.lock);
1475 cond_resched_lock(&lru_lock);
1476 }
1477}
1478
1479/**
1480 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1481 * @nr: The number of entries to scan
1482 *
1483 * This function selects the entries on the LRU which are able to
1484 * be demoted, and then kicks off the process by calling
1485 * gfs2_dispose_glock_lru() above.
1486 */
1487
1488static long gfs2_scan_glock_lru(int nr)
1489{
1490 struct gfs2_glock *gl;
1491 LIST_HEAD(skipped);
1492 LIST_HEAD(dispose);
1493 long freed = 0;
1494
1495 spin_lock(&lru_lock);
1496 while ((nr-- >= 0) && !list_empty(&lru_list)) {
1497 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1498
1499 /* Test for being demotable */
1500 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1501 list_move(&gl->gl_lru, &dispose);
1502 atomic_dec(&lru_count);
1503 clear_bit(GLF_LRU, &gl->gl_flags);
1504 freed++;
1505 continue;
1506 }
1507
1508 list_move(&gl->gl_lru, &skipped);
1509 }
1510 list_splice(&skipped, &lru_list);
1511 if (!list_empty(&dispose))
1512 gfs2_dispose_glock_lru(&dispose);
1513 spin_unlock(&lru_lock);
1514
1515 return freed;
1516}
1517
1518static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1519 struct shrink_control *sc)
1520{
1521 if (!(sc->gfp_mask & __GFP_FS))
1522 return SHRINK_STOP;
1523 return gfs2_scan_glock_lru(sc->nr_to_scan);
1524}
1525
1526static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1527 struct shrink_control *sc)
1528{
1529 return vfs_pressure_ratio(atomic_read(&lru_count));
1530}
1531
1532static struct shrinker glock_shrinker = {
1533 .seeks = DEFAULT_SEEKS,
1534 .count_objects = gfs2_glock_shrink_count,
1535 .scan_objects = gfs2_glock_shrink_scan,
1536};
1537
1538/**
1539 * examine_bucket - Call a function for glock in a hash bucket
1540 * @examiner: the function
1541 * @sdp: the filesystem
1542 * @bucket: the bucket
1543 *
1544 * Note that the function can be called multiple times on the same
1545 * object. So the user must ensure that the function can cope with
1546 * that.
1547 */
1548
1549static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1550{
1551 struct gfs2_glock *gl;
1552 struct rhashtable_iter iter;
1553
1554 rhashtable_walk_enter(&gl_hash_table, &iter);
1555
1556 do {
1557 gl = ERR_PTR(rhashtable_walk_start(&iter));
1558 if (IS_ERR(gl))
1559 goto walk_stop;
1560
1561 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl))
1562 if (gl->gl_name.ln_sbd == sdp &&
1563 lockref_get_not_dead(&gl->gl_lockref))
1564 examiner(gl);
1565
1566walk_stop:
1567 rhashtable_walk_stop(&iter);
1568 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
1569
1570 rhashtable_walk_exit(&iter);
1571}
1572
1573/**
1574 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1575 * @gl: The glock to thaw
1576 *
1577 */
1578
1579static void thaw_glock(struct gfs2_glock *gl)
1580{
1581 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags)) {
1582 gfs2_glock_put(gl);
1583 return;
1584 }
1585 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1586 gfs2_glock_queue_work(gl, 0);
1587}
1588
1589/**
1590 * clear_glock - look at a glock and see if we can free it from glock cache
1591 * @gl: the glock to look at
1592 *
1593 */
1594
1595static void clear_glock(struct gfs2_glock *gl)
1596{
1597 gfs2_glock_remove_from_lru(gl);
1598
1599 spin_lock(&gl->gl_lockref.lock);
1600 if (gl->gl_state != LM_ST_UNLOCKED)
1601 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1602 __gfs2_glock_queue_work(gl, 0);
1603 spin_unlock(&gl->gl_lockref.lock);
1604}
1605
1606/**
1607 * gfs2_glock_thaw - Thaw any frozen glocks
1608 * @sdp: The super block
1609 *
1610 */
1611
1612void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1613{
1614 glock_hash_walk(thaw_glock, sdp);
1615}
1616
1617static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1618{
1619 spin_lock(&gl->gl_lockref.lock);
1620 gfs2_dump_glock(seq, gl);
1621 spin_unlock(&gl->gl_lockref.lock);
1622}
1623
1624static void dump_glock_func(struct gfs2_glock *gl)
1625{
1626 dump_glock(NULL, gl);
1627}
1628
1629/**
1630 * gfs2_gl_hash_clear - Empty out the glock hash table
1631 * @sdp: the filesystem
1632 * @wait: wait until it's all gone
1633 *
1634 * Called when unmounting the filesystem.
1635 */
1636
1637void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1638{
1639 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1640 flush_workqueue(glock_workqueue);
1641 glock_hash_walk(clear_glock, sdp);
1642 flush_workqueue(glock_workqueue);
1643 wait_event_timeout(sdp->sd_glock_wait,
1644 atomic_read(&sdp->sd_glock_disposal) == 0,
1645 HZ * 600);
1646 glock_hash_walk(dump_glock_func, sdp);
1647}
1648
1649void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1650{
1651 struct gfs2_glock *gl = ip->i_gl;
1652 int ret;
1653
1654 ret = gfs2_truncatei_resume(ip);
1655 gfs2_assert_withdraw(gl->gl_name.ln_sbd, ret == 0);
1656
1657 spin_lock(&gl->gl_lockref.lock);
1658 clear_bit(GLF_LOCK, &gl->gl_flags);
1659 run_queue(gl, 1);
1660 spin_unlock(&gl->gl_lockref.lock);
1661}
1662
1663static const char *state2str(unsigned state)
1664{
1665 switch(state) {
1666 case LM_ST_UNLOCKED:
1667 return "UN";
1668 case LM_ST_SHARED:
1669 return "SH";
1670 case LM_ST_DEFERRED:
1671 return "DF";
1672 case LM_ST_EXCLUSIVE:
1673 return "EX";
1674 }
1675 return "??";
1676}
1677
1678static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
1679{
1680 char *p = buf;
1681 if (flags & LM_FLAG_TRY)
1682 *p++ = 't';
1683 if (flags & LM_FLAG_TRY_1CB)
1684 *p++ = 'T';
1685 if (flags & LM_FLAG_NOEXP)
1686 *p++ = 'e';
1687 if (flags & LM_FLAG_ANY)
1688 *p++ = 'A';
1689 if (flags & LM_FLAG_PRIORITY)
1690 *p++ = 'p';
1691 if (flags & GL_ASYNC)
1692 *p++ = 'a';
1693 if (flags & GL_EXACT)
1694 *p++ = 'E';
1695 if (flags & GL_NOCACHE)
1696 *p++ = 'c';
1697 if (test_bit(HIF_HOLDER, &iflags))
1698 *p++ = 'H';
1699 if (test_bit(HIF_WAIT, &iflags))
1700 *p++ = 'W';
1701 if (test_bit(HIF_FIRST, &iflags))
1702 *p++ = 'F';
1703 *p = 0;
1704 return buf;
1705}
1706
1707/**
1708 * dump_holder - print information about a glock holder
1709 * @seq: the seq_file struct
1710 * @gh: the glock holder
1711 *
1712 */
1713
1714static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1715{
1716 struct task_struct *gh_owner = NULL;
1717 char flags_buf[32];
1718
1719 rcu_read_lock();
1720 if (gh->gh_owner_pid)
1721 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1722 gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1723 state2str(gh->gh_state),
1724 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1725 gh->gh_error,
1726 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1727 gh_owner ? gh_owner->comm : "(ended)",
1728 (void *)gh->gh_ip);
1729 rcu_read_unlock();
1730}
1731
1732static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1733{
1734 const unsigned long *gflags = &gl->gl_flags;
1735 char *p = buf;
1736
1737 if (test_bit(GLF_LOCK, gflags))
1738 *p++ = 'l';
1739 if (test_bit(GLF_DEMOTE, gflags))
1740 *p++ = 'D';
1741 if (test_bit(GLF_PENDING_DEMOTE, gflags))
1742 *p++ = 'd';
1743 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1744 *p++ = 'p';
1745 if (test_bit(GLF_DIRTY, gflags))
1746 *p++ = 'y';
1747 if (test_bit(GLF_LFLUSH, gflags))
1748 *p++ = 'f';
1749 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1750 *p++ = 'i';
1751 if (test_bit(GLF_REPLY_PENDING, gflags))
1752 *p++ = 'r';
1753 if (test_bit(GLF_INITIAL, gflags))
1754 *p++ = 'I';
1755 if (test_bit(GLF_FROZEN, gflags))
1756 *p++ = 'F';
1757 if (test_bit(GLF_QUEUED, gflags))
1758 *p++ = 'q';
1759 if (test_bit(GLF_LRU, gflags))
1760 *p++ = 'L';
1761 if (gl->gl_object)
1762 *p++ = 'o';
1763 if (test_bit(GLF_BLOCKING, gflags))
1764 *p++ = 'b';
1765 *p = 0;
1766 return buf;
1767}
1768
1769/**
1770 * gfs2_dump_glock - print information about a glock
1771 * @seq: The seq_file struct
1772 * @gl: the glock
1773 *
1774 * The file format is as follows:
1775 * One line per object, capital letters are used to indicate objects
1776 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1777 * other objects are indented by a single space and follow the glock to
1778 * which they are related. Fields are indicated by lower case letters
1779 * followed by a colon and the field value, except for strings which are in
1780 * [] so that its possible to see if they are composed of spaces for
1781 * example. The field's are n = number (id of the object), f = flags,
1782 * t = type, s = state, r = refcount, e = error, p = pid.
1783 *
1784 */
1785
1786void gfs2_dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1787{
1788 const struct gfs2_glock_operations *glops = gl->gl_ops;
1789 unsigned long long dtime;
1790 const struct gfs2_holder *gh;
1791 char gflags_buf[32];
1792
1793 dtime = jiffies - gl->gl_demote_time;
1794 dtime *= 1000000/HZ; /* demote time in uSec */
1795 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1796 dtime = 0;
1797 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",
1798 state2str(gl->gl_state),
1799 gl->gl_name.ln_type,
1800 (unsigned long long)gl->gl_name.ln_number,
1801 gflags2str(gflags_buf, gl),
1802 state2str(gl->gl_target),
1803 state2str(gl->gl_demote_state), dtime,
1804 atomic_read(&gl->gl_ail_count),
1805 atomic_read(&gl->gl_revokes),
1806 (int)gl->gl_lockref.count, gl->gl_hold_time);
1807
1808 list_for_each_entry(gh, &gl->gl_holders, gh_list)
1809 dump_holder(seq, gh);
1810
1811 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1812 glops->go_dump(seq, gl);
1813}
1814
1815static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
1816{
1817 struct gfs2_glock *gl = iter_ptr;
1818
1819 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
1820 gl->gl_name.ln_type,
1821 (unsigned long long)gl->gl_name.ln_number,
1822 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
1823 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
1824 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
1825 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
1826 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
1827 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
1828 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
1829 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
1830 return 0;
1831}
1832
1833static const char *gfs2_gltype[] = {
1834 "type",
1835 "reserved",
1836 "nondisk",
1837 "inode",
1838 "rgrp",
1839 "meta",
1840 "iopen",
1841 "flock",
1842 "plock",
1843 "quota",
1844 "journal",
1845};
1846
1847static const char *gfs2_stype[] = {
1848 [GFS2_LKS_SRTT] = "srtt",
1849 [GFS2_LKS_SRTTVAR] = "srttvar",
1850 [GFS2_LKS_SRTTB] = "srttb",
1851 [GFS2_LKS_SRTTVARB] = "srttvarb",
1852 [GFS2_LKS_SIRT] = "sirt",
1853 [GFS2_LKS_SIRTVAR] = "sirtvar",
1854 [GFS2_LKS_DCOUNT] = "dlm",
1855 [GFS2_LKS_QCOUNT] = "queue",
1856};
1857
1858#define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1859
1860static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
1861{
1862 struct gfs2_sbd *sdp = seq->private;
1863 loff_t pos = *(loff_t *)iter_ptr;
1864 unsigned index = pos >> 3;
1865 unsigned subindex = pos & 0x07;
1866 int i;
1867
1868 if (index == 0 && subindex != 0)
1869 return 0;
1870
1871 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
1872 (index == 0) ? "cpu": gfs2_stype[subindex]);
1873
1874 for_each_possible_cpu(i) {
1875 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
1876
1877 if (index == 0)
1878 seq_printf(seq, " %15u", i);
1879 else
1880 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
1881 lkstats[index - 1].stats[subindex]);
1882 }
1883 seq_putc(seq, '\n');
1884 return 0;
1885}
1886
1887int __init gfs2_glock_init(void)
1888{
1889 int i, ret;
1890
1891 ret = rhashtable_init(&gl_hash_table, &ht_parms);
1892 if (ret < 0)
1893 return ret;
1894
1895 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1896 WQ_HIGHPRI | WQ_FREEZABLE, 0);
1897 if (!glock_workqueue) {
1898 rhashtable_destroy(&gl_hash_table);
1899 return -ENOMEM;
1900 }
1901 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1902 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1903 0);
1904 if (!gfs2_delete_workqueue) {
1905 destroy_workqueue(glock_workqueue);
1906 rhashtable_destroy(&gl_hash_table);
1907 return -ENOMEM;
1908 }
1909
1910 ret = register_shrinker(&glock_shrinker);
1911 if (ret) {
1912 destroy_workqueue(gfs2_delete_workqueue);
1913 destroy_workqueue(glock_workqueue);
1914 rhashtable_destroy(&gl_hash_table);
1915 return ret;
1916 }
1917
1918 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
1919 init_waitqueue_head(glock_wait_table + i);
1920
1921 return 0;
1922}
1923
1924void gfs2_glock_exit(void)
1925{
1926 unregister_shrinker(&glock_shrinker);
1927 rhashtable_destroy(&gl_hash_table);
1928 destroy_workqueue(glock_workqueue);
1929 destroy_workqueue(gfs2_delete_workqueue);
1930}
1931
1932static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1933{
1934 while ((gi->gl = rhashtable_walk_next(&gi->hti))) {
1935 if (IS_ERR(gi->gl)) {
1936 if (PTR_ERR(gi->gl) == -EAGAIN)
1937 continue;
1938 gi->gl = NULL;
1939 return;
1940 }
1941 /* Skip entries for other sb and dead entries */
1942 if (gi->sdp == gi->gl->gl_name.ln_sbd &&
1943 !__lockref_is_dead(&gi->gl->gl_lockref))
1944 return;
1945 }
1946}
1947
1948static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1949 __acquires(RCU)
1950{
1951 struct gfs2_glock_iter *gi = seq->private;
1952 loff_t n = *pos;
1953
1954 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
1955 if (rhashtable_walk_start(&gi->hti) != 0)
1956 return NULL;
1957
1958 do {
1959 gfs2_glock_iter_next(gi);
1960 } while (gi->gl && n--);
1961
1962 gi->last_pos = *pos;
1963
1964 return gi->gl;
1965}
1966
1967static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1968 loff_t *pos)
1969{
1970 struct gfs2_glock_iter *gi = seq->private;
1971
1972 (*pos)++;
1973 gi->last_pos = *pos;
1974 gfs2_glock_iter_next(gi);
1975
1976 return gi->gl;
1977}
1978
1979static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1980 __releases(RCU)
1981{
1982 struct gfs2_glock_iter *gi = seq->private;
1983
1984 gi->gl = NULL;
1985 rhashtable_walk_stop(&gi->hti);
1986 rhashtable_walk_exit(&gi->hti);
1987}
1988
1989static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1990{
1991 dump_glock(seq, iter_ptr);
1992 return 0;
1993}
1994
1995static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
1996{
1997 preempt_disable();
1998 if (*pos >= GFS2_NR_SBSTATS)
1999 return NULL;
2000 return pos;
2001}
2002
2003static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2004 loff_t *pos)
2005{
2006 (*pos)++;
2007 if (*pos >= GFS2_NR_SBSTATS)
2008 return NULL;
2009 return pos;
2010}
2011
2012static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2013{
2014 preempt_enable();
2015}
2016
2017static const struct seq_operations gfs2_glock_seq_ops = {
2018 .start = gfs2_glock_seq_start,
2019 .next = gfs2_glock_seq_next,
2020 .stop = gfs2_glock_seq_stop,
2021 .show = gfs2_glock_seq_show,
2022};
2023
2024static const struct seq_operations gfs2_glstats_seq_ops = {
2025 .start = gfs2_glock_seq_start,
2026 .next = gfs2_glock_seq_next,
2027 .stop = gfs2_glock_seq_stop,
2028 .show = gfs2_glstats_seq_show,
2029};
2030
2031static const struct seq_operations gfs2_sbstats_seq_ops = {
2032 .start = gfs2_sbstats_seq_start,
2033 .next = gfs2_sbstats_seq_next,
2034 .stop = gfs2_sbstats_seq_stop,
2035 .show = gfs2_sbstats_seq_show,
2036};
2037
2038#define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2039
2040static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2041 const struct seq_operations *ops)
2042{
2043 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2044 if (ret == 0) {
2045 struct seq_file *seq = file->private_data;
2046 struct gfs2_glock_iter *gi = seq->private;
2047
2048 gi->sdp = inode->i_private;
2049 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2050 if (seq->buf)
2051 seq->size = GFS2_SEQ_GOODSIZE;
2052 gi->gl = NULL;
2053 }
2054 return ret;
2055}
2056
2057static int gfs2_glocks_open(struct inode *inode, struct file *file)
2058{
2059 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2060}
2061
2062static int gfs2_glocks_release(struct inode *inode, struct file *file)
2063{
2064 struct seq_file *seq = file->private_data;
2065 struct gfs2_glock_iter *gi = seq->private;
2066
2067 gi->gl = NULL;
2068 return seq_release_private(inode, file);
2069}
2070
2071static int gfs2_glstats_open(struct inode *inode, struct file *file)
2072{
2073 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2074}
2075
2076static int gfs2_sbstats_open(struct inode *inode, struct file *file)
2077{
2078 int ret = seq_open(file, &gfs2_sbstats_seq_ops);
2079 if (ret == 0) {
2080 struct seq_file *seq = file->private_data;
2081 seq->private = inode->i_private; /* sdp */
2082 }
2083 return ret;
2084}
2085
2086static const struct file_operations gfs2_glocks_fops = {
2087 .owner = THIS_MODULE,
2088 .open = gfs2_glocks_open,
2089 .read = seq_read,
2090 .llseek = seq_lseek,
2091 .release = gfs2_glocks_release,
2092};
2093
2094static const struct file_operations gfs2_glstats_fops = {
2095 .owner = THIS_MODULE,
2096 .open = gfs2_glstats_open,
2097 .read = seq_read,
2098 .llseek = seq_lseek,
2099 .release = gfs2_glocks_release,
2100};
2101
2102static const struct file_operations gfs2_sbstats_fops = {
2103 .owner = THIS_MODULE,
2104 .open = gfs2_sbstats_open,
2105 .read = seq_read,
2106 .llseek = seq_lseek,
2107 .release = seq_release,
2108};
2109
2110int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2111{
2112 struct dentry *dent;
2113
2114 dent = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2115 if (IS_ERR_OR_NULL(dent))
2116 goto fail;
2117 sdp->debugfs_dir = dent;
2118
2119 dent = debugfs_create_file("glocks",
2120 S_IFREG | S_IRUGO,
2121 sdp->debugfs_dir, sdp,
2122 &gfs2_glocks_fops);
2123 if (IS_ERR_OR_NULL(dent))
2124 goto fail;
2125 sdp->debugfs_dentry_glocks = dent;
2126
2127 dent = debugfs_create_file("glstats",
2128 S_IFREG | S_IRUGO,
2129 sdp->debugfs_dir, sdp,
2130 &gfs2_glstats_fops);
2131 if (IS_ERR_OR_NULL(dent))
2132 goto fail;
2133 sdp->debugfs_dentry_glstats = dent;
2134
2135 dent = debugfs_create_file("sbstats",
2136 S_IFREG | S_IRUGO,
2137 sdp->debugfs_dir, sdp,
2138 &gfs2_sbstats_fops);
2139 if (IS_ERR_OR_NULL(dent))
2140 goto fail;
2141 sdp->debugfs_dentry_sbstats = dent;
2142
2143 return 0;
2144fail:
2145 gfs2_delete_debugfs_file(sdp);
2146 return dent ? PTR_ERR(dent) : -ENOMEM;
2147}
2148
2149void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2150{
2151 if (sdp->debugfs_dir) {
2152 if (sdp->debugfs_dentry_glocks) {
2153 debugfs_remove(sdp->debugfs_dentry_glocks);
2154 sdp->debugfs_dentry_glocks = NULL;
2155 }
2156 if (sdp->debugfs_dentry_glstats) {
2157 debugfs_remove(sdp->debugfs_dentry_glstats);
2158 sdp->debugfs_dentry_glstats = NULL;
2159 }
2160 if (sdp->debugfs_dentry_sbstats) {
2161 debugfs_remove(sdp->debugfs_dentry_sbstats);
2162 sdp->debugfs_dentry_sbstats = NULL;
2163 }
2164 debugfs_remove(sdp->debugfs_dir);
2165 sdp->debugfs_dir = NULL;
2166 }
2167}
2168
2169int gfs2_register_debugfs(void)
2170{
2171 gfs2_root = debugfs_create_dir("gfs2", NULL);
2172 if (IS_ERR(gfs2_root))
2173 return PTR_ERR(gfs2_root);
2174 return gfs2_root ? 0 : -ENOMEM;
2175}
2176
2177void gfs2_unregister_debugfs(void)
2178{
2179 debugfs_remove(gfs2_root);
2180 gfs2_root = NULL;
2181}