1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 */
6
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
9 #include <linux/sched.h>
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/buffer_head.h>
13 #include <linux/delay.h>
14 #include <linux/sort.h>
15 #include <linux/hash.h>
16 #include <linux/jhash.h>
17 #include <linux/kallsyms.h>
18 #include <linux/gfs2_ondisk.h>
19 #include <linux/list.h>
20 #include <linux/wait.h>
21 #include <linux/module.h>
22 #include <linux/uaccess.h>
23 #include <linux/seq_file.h>
24 #include <linux/debugfs.h>
25 #include <linux/kthread.h>
26 #include <linux/freezer.h>
27 #include <linux/workqueue.h>
28 #include <linux/jiffies.h>
29 #include <linux/rcupdate.h>
30 #include <linux/rculist_bl.h>
31 #include <linux/bit_spinlock.h>
32 #include <linux/percpu.h>
33 #include <linux/list_sort.h>
34 #include <linux/lockref.h>
35 #include <linux/rhashtable.h>
36
37 #include "gfs2.h"
38 #include "incore.h"
39 #include "glock.h"
40 #include "glops.h"
41 #include "inode.h"
42 #include "lops.h"
43 #include "meta_io.h"
44 #include "quota.h"
45 #include "super.h"
46 #include "util.h"
47 #include "bmap.h"
48 #define CREATE_TRACE_POINTS
49 #include "trace_gfs2.h"
50
51 struct gfs2_glock_iter {
52 struct gfs2_sbd *sdp; /* incore superblock */
53 struct rhashtable_iter hti; /* rhashtable iterator */
54 struct gfs2_glock *gl; /* current glock struct */
55 loff_t last_pos; /* last position */
56 };
57
58 typedef void (*glock_examiner) (struct gfs2_glock * gl);
59
60 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
61 static void __gfs2_glock_dq(struct gfs2_holder *gh);
62
63 static struct dentry *gfs2_root;
64 static struct workqueue_struct *glock_workqueue;
65 struct workqueue_struct *gfs2_delete_workqueue;
66 static LIST_HEAD(lru_list);
67 static atomic_t lru_count = ATOMIC_INIT(0);
68 static DEFINE_SPINLOCK(lru_lock);
69
70 #define GFS2_GL_HASH_SHIFT 15
71 #define GFS2_GL_HASH_SIZE BIT(GFS2_GL_HASH_SHIFT)
72
73 static const struct rhashtable_params ht_parms = {
74 .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4,
75 .key_len = offsetofend(struct lm_lockname, ln_type),
76 .key_offset = offsetof(struct gfs2_glock, gl_name),
77 .head_offset = offsetof(struct gfs2_glock, gl_node),
78 };
79
80 static struct rhashtable gl_hash_table;
81
82 #define GLOCK_WAIT_TABLE_BITS 12
83 #define GLOCK_WAIT_TABLE_SIZE (1 << GLOCK_WAIT_TABLE_BITS)
84 static wait_queue_head_t glock_wait_table[GLOCK_WAIT_TABLE_SIZE] __cacheline_aligned;
85
86 struct wait_glock_queue {
87 struct lm_lockname *name;
88 wait_queue_entry_t wait;
89 };
90
glock_wake_function(wait_queue_entry_t * wait,unsigned int mode,int sync,void * key)91 static int glock_wake_function(wait_queue_entry_t *wait, unsigned int mode,
92 int sync, void *key)
93 {
94 struct wait_glock_queue *wait_glock =
95 container_of(wait, struct wait_glock_queue, wait);
96 struct lm_lockname *wait_name = wait_glock->name;
97 struct lm_lockname *wake_name = key;
98
99 if (wake_name->ln_sbd != wait_name->ln_sbd ||
100 wake_name->ln_number != wait_name->ln_number ||
101 wake_name->ln_type != wait_name->ln_type)
102 return 0;
103 return autoremove_wake_function(wait, mode, sync, key);
104 }
105
glock_waitqueue(struct lm_lockname * name)106 static wait_queue_head_t *glock_waitqueue(struct lm_lockname *name)
107 {
108 u32 hash = jhash2((u32 *)name, ht_parms.key_len / 4, 0);
109
110 return glock_wait_table + hash_32(hash, GLOCK_WAIT_TABLE_BITS);
111 }
112
113 /**
114 * wake_up_glock - Wake up waiters on a glock
115 * @gl: the glock
116 */
wake_up_glock(struct gfs2_glock * gl)117 static void wake_up_glock(struct gfs2_glock *gl)
118 {
119 wait_queue_head_t *wq = glock_waitqueue(&gl->gl_name);
120
121 if (waitqueue_active(wq))
122 __wake_up(wq, TASK_NORMAL, 1, &gl->gl_name);
123 }
124
gfs2_glock_dealloc(struct rcu_head * rcu)125 static void gfs2_glock_dealloc(struct rcu_head *rcu)
126 {
127 struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
128
129 kfree(gl->gl_lksb.sb_lvbptr);
130 if (gl->gl_ops->go_flags & GLOF_ASPACE)
131 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
132 else
133 kmem_cache_free(gfs2_glock_cachep, gl);
134 }
135
136 /**
137 * glock_blocked_by_withdraw - determine if we can still use a glock
138 * @gl: the glock
139 *
140 * We need to allow some glocks to be enqueued, dequeued, promoted, and demoted
141 * when we're withdrawn. For example, to maintain metadata integrity, we should
142 * disallow the use of inode and rgrp glocks when withdrawn. Other glocks, like
143 * iopen or the transaction glocks may be safely used because none of their
144 * metadata goes through the journal. So in general, we should disallow all
145 * glocks that are journaled, and allow all the others. One exception is:
146 * we need to allow our active journal to be promoted and demoted so others
147 * may recover it and we can reacquire it when they're done.
148 */
glock_blocked_by_withdraw(struct gfs2_glock * gl)149 static bool glock_blocked_by_withdraw(struct gfs2_glock *gl)
150 {
151 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
152
153 if (likely(!gfs2_withdrawn(sdp)))
154 return false;
155 if (gl->gl_ops->go_flags & GLOF_NONDISK)
156 return false;
157 if (!sdp->sd_jdesc ||
158 gl->gl_name.ln_number == sdp->sd_jdesc->jd_no_addr)
159 return false;
160 return true;
161 }
162
gfs2_glock_free(struct gfs2_glock * gl)163 void gfs2_glock_free(struct gfs2_glock *gl)
164 {
165 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
166
167 gfs2_glock_assert_withdraw(gl, atomic_read(&gl->gl_revokes) == 0);
168 rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
169 smp_mb();
170 wake_up_glock(gl);
171 call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
172 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
173 wake_up(&sdp->sd_glock_wait);
174 }
175
176 /**
177 * gfs2_glock_hold() - increment reference count on glock
178 * @gl: The glock to hold
179 *
180 */
181
gfs2_glock_hold(struct gfs2_glock * gl)182 void gfs2_glock_hold(struct gfs2_glock *gl)
183 {
184 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
185 lockref_get(&gl->gl_lockref);
186 }
187
188 /**
189 * demote_ok - Check to see if it's ok to unlock a glock
190 * @gl: the glock
191 *
192 * Returns: 1 if it's ok
193 */
194
demote_ok(const struct gfs2_glock * gl)195 static int demote_ok(const struct gfs2_glock *gl)
196 {
197 const struct gfs2_glock_operations *glops = gl->gl_ops;
198
199 if (gl->gl_state == LM_ST_UNLOCKED)
200 return 0;
201 /*
202 * Note that demote_ok is used for the lru process of disposing of
203 * glocks. For this purpose, we don't care if the glock's holders
204 * have the HIF_MAY_DEMOTE flag set or not. If someone is using
205 * them, don't demote.
206 */
207 if (!list_empty(&gl->gl_holders))
208 return 0;
209 if (glops->go_demote_ok)
210 return glops->go_demote_ok(gl);
211 return 1;
212 }
213
214
gfs2_glock_add_to_lru(struct gfs2_glock * gl)215 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
216 {
217 if (!(gl->gl_ops->go_flags & GLOF_LRU))
218 return;
219
220 spin_lock(&lru_lock);
221
222 list_move_tail(&gl->gl_lru, &lru_list);
223
224 if (!test_bit(GLF_LRU, &gl->gl_flags)) {
225 set_bit(GLF_LRU, &gl->gl_flags);
226 atomic_inc(&lru_count);
227 }
228
229 spin_unlock(&lru_lock);
230 }
231
gfs2_glock_remove_from_lru(struct gfs2_glock * gl)232 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
233 {
234 if (!(gl->gl_ops->go_flags & GLOF_LRU))
235 return;
236
237 spin_lock(&lru_lock);
238 if (test_bit(GLF_LRU, &gl->gl_flags)) {
239 list_del_init(&gl->gl_lru);
240 atomic_dec(&lru_count);
241 clear_bit(GLF_LRU, &gl->gl_flags);
242 }
243 spin_unlock(&lru_lock);
244 }
245
246 /*
247 * Enqueue the glock on the work queue. Passes one glock reference on to the
248 * work queue.
249 */
__gfs2_glock_queue_work(struct gfs2_glock * gl,unsigned long delay)250 static void __gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
251 if (!queue_delayed_work(glock_workqueue, &gl->gl_work, delay)) {
252 /*
253 * We are holding the lockref spinlock, and the work was still
254 * queued above. The queued work (glock_work_func) takes that
255 * spinlock before dropping its glock reference(s), so it
256 * cannot have dropped them in the meantime.
257 */
258 GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2);
259 gl->gl_lockref.count--;
260 }
261 }
262
gfs2_glock_queue_work(struct gfs2_glock * gl,unsigned long delay)263 static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
264 spin_lock(&gl->gl_lockref.lock);
265 __gfs2_glock_queue_work(gl, delay);
266 spin_unlock(&gl->gl_lockref.lock);
267 }
268
__gfs2_glock_put(struct gfs2_glock * gl)269 static void __gfs2_glock_put(struct gfs2_glock *gl)
270 {
271 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
272 struct address_space *mapping = gfs2_glock2aspace(gl);
273
274 lockref_mark_dead(&gl->gl_lockref);
275
276 gfs2_glock_remove_from_lru(gl);
277 spin_unlock(&gl->gl_lockref.lock);
278 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
279 if (mapping) {
280 truncate_inode_pages_final(mapping);
281 if (!gfs2_withdrawn(sdp))
282 GLOCK_BUG_ON(gl, !mapping_empty(mapping));
283 }
284 trace_gfs2_glock_put(gl);
285 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
286 }
287
288 /*
289 * Cause the glock to be put in work queue context.
290 */
gfs2_glock_queue_put(struct gfs2_glock * gl)291 void gfs2_glock_queue_put(struct gfs2_glock *gl)
292 {
293 gfs2_glock_queue_work(gl, 0);
294 }
295
296 /**
297 * gfs2_glock_put() - Decrement reference count on glock
298 * @gl: The glock to put
299 *
300 */
301
gfs2_glock_put(struct gfs2_glock * gl)302 void gfs2_glock_put(struct gfs2_glock *gl)
303 {
304 /* last put could call sleepable dlm api */
305 might_sleep();
306
307 if (lockref_put_or_lock(&gl->gl_lockref))
308 return;
309
310 __gfs2_glock_put(gl);
311 }
312
313 /**
314 * may_grant - check if it's ok to grant a new lock
315 * @gl: The glock
316 * @current_gh: One of the current holders of @gl
317 * @gh: The lock request which we wish to grant
318 *
319 * With our current compatibility rules, if a glock has one or more active
320 * holders (HIF_HOLDER flag set), any of those holders can be passed in as
321 * @current_gh; they are all the same as far as compatibility with the new @gh
322 * goes.
323 *
324 * Returns true if it's ok to grant the lock.
325 */
326
may_grant(struct gfs2_glock * gl,struct gfs2_holder * current_gh,struct gfs2_holder * gh)327 static inline bool may_grant(struct gfs2_glock *gl,
328 struct gfs2_holder *current_gh,
329 struct gfs2_holder *gh)
330 {
331 if (current_gh) {
332 GLOCK_BUG_ON(gl, !test_bit(HIF_HOLDER, ¤t_gh->gh_iflags));
333
334 switch(current_gh->gh_state) {
335 case LM_ST_EXCLUSIVE:
336 /*
337 * Here we make a special exception to grant holders
338 * who agree to share the EX lock with other holders
339 * who also have the bit set. If the original holder
340 * has the LM_FLAG_NODE_SCOPE bit set, we grant more
341 * holders with the bit set.
342 */
343 return gh->gh_state == LM_ST_EXCLUSIVE &&
344 (current_gh->gh_flags & LM_FLAG_NODE_SCOPE) &&
345 (gh->gh_flags & LM_FLAG_NODE_SCOPE);
346
347 case LM_ST_SHARED:
348 case LM_ST_DEFERRED:
349 return gh->gh_state == current_gh->gh_state;
350
351 default:
352 return false;
353 }
354 }
355
356 if (gl->gl_state == gh->gh_state)
357 return true;
358 if (gh->gh_flags & GL_EXACT)
359 return false;
360 if (gl->gl_state == LM_ST_EXCLUSIVE) {
361 return gh->gh_state == LM_ST_SHARED ||
362 gh->gh_state == LM_ST_DEFERRED;
363 }
364 if (gh->gh_flags & LM_FLAG_ANY)
365 return gl->gl_state != LM_ST_UNLOCKED;
366 return false;
367 }
368
gfs2_holder_wake(struct gfs2_holder * gh)369 static void gfs2_holder_wake(struct gfs2_holder *gh)
370 {
371 clear_bit(HIF_WAIT, &gh->gh_iflags);
372 smp_mb__after_atomic();
373 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
374 if (gh->gh_flags & GL_ASYNC) {
375 struct gfs2_sbd *sdp = gh->gh_gl->gl_name.ln_sbd;
376
377 wake_up(&sdp->sd_async_glock_wait);
378 }
379 }
380
381 /**
382 * do_error - Something unexpected has happened during a lock request
383 * @gl: The glock
384 * @ret: The status from the DLM
385 */
386
do_error(struct gfs2_glock * gl,const int ret)387 static void do_error(struct gfs2_glock *gl, const int ret)
388 {
389 struct gfs2_holder *gh, *tmp;
390
391 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
392 if (!test_bit(HIF_WAIT, &gh->gh_iflags))
393 continue;
394 if (ret & LM_OUT_ERROR)
395 gh->gh_error = -EIO;
396 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
397 gh->gh_error = GLR_TRYFAILED;
398 else
399 continue;
400 list_del_init(&gh->gh_list);
401 trace_gfs2_glock_queue(gh, 0);
402 gfs2_holder_wake(gh);
403 }
404 }
405
406 /**
407 * demote_incompat_holders - demote incompatible demoteable holders
408 * @gl: the glock we want to promote
409 * @new_gh: the new holder to be promoted
410 */
demote_incompat_holders(struct gfs2_glock * gl,struct gfs2_holder * new_gh)411 static void demote_incompat_holders(struct gfs2_glock *gl,
412 struct gfs2_holder *new_gh)
413 {
414 struct gfs2_holder *gh, *tmp;
415
416 /*
417 * Demote incompatible holders before we make ourselves eligible.
418 * (This holder may or may not allow auto-demoting, but we don't want
419 * to demote the new holder before it's even granted.)
420 */
421 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
422 /*
423 * Since holders are at the front of the list, we stop when we
424 * find the first non-holder.
425 */
426 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
427 return;
428 if (test_bit(HIF_MAY_DEMOTE, &gh->gh_iflags) &&
429 !may_grant(gl, new_gh, gh)) {
430 /*
431 * We should not recurse into do_promote because
432 * __gfs2_glock_dq only calls handle_callback,
433 * gfs2_glock_add_to_lru and __gfs2_glock_queue_work.
434 */
435 __gfs2_glock_dq(gh);
436 }
437 }
438 }
439
440 /**
441 * find_first_holder - find the first "holder" gh
442 * @gl: the glock
443 */
444
find_first_holder(const struct gfs2_glock * gl)445 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
446 {
447 struct gfs2_holder *gh;
448
449 if (!list_empty(&gl->gl_holders)) {
450 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder,
451 gh_list);
452 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
453 return gh;
454 }
455 return NULL;
456 }
457
458 /**
459 * find_first_strong_holder - find the first non-demoteable holder
460 * @gl: the glock
461 *
462 * Find the first holder that doesn't have the HIF_MAY_DEMOTE flag set.
463 */
464 static inline struct gfs2_holder *
find_first_strong_holder(struct gfs2_glock * gl)465 find_first_strong_holder(struct gfs2_glock *gl)
466 {
467 struct gfs2_holder *gh;
468
469 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
470 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
471 return NULL;
472 if (!test_bit(HIF_MAY_DEMOTE, &gh->gh_iflags))
473 return gh;
474 }
475 return NULL;
476 }
477
478 /*
479 * gfs2_instantiate - Call the glops instantiate function
480 * @gl: The glock
481 *
482 * Returns: 0 if instantiate was successful, 2 if type specific operation is
483 * underway, or error.
484 */
gfs2_instantiate(struct gfs2_holder * gh)485 int gfs2_instantiate(struct gfs2_holder *gh)
486 {
487 struct gfs2_glock *gl = gh->gh_gl;
488 const struct gfs2_glock_operations *glops = gl->gl_ops;
489 int ret;
490
491 again:
492 if (!test_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags))
493 return 0;
494
495 /*
496 * Since we unlock the lockref lock, we set a flag to indicate
497 * instantiate is in progress.
498 */
499 if (test_and_set_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags)) {
500 wait_on_bit(&gl->gl_flags, GLF_INSTANTIATE_IN_PROG,
501 TASK_UNINTERRUPTIBLE);
502 /*
503 * Here we just waited for a different instantiate to finish.
504 * But that may not have been successful, as when a process
505 * locks an inode glock _before_ it has an actual inode to
506 * instantiate into. So we check again. This process might
507 * have an inode to instantiate, so might be successful.
508 */
509 goto again;
510 }
511
512 ret = glops->go_instantiate(gh);
513 if (!ret)
514 clear_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags);
515 clear_and_wake_up_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags);
516 return ret;
517 }
518
519 /**
520 * do_promote - promote as many requests as possible on the current queue
521 * @gl: The glock
522 *
523 * Returns: 1 if there is a blocked holder at the head of the list, or 2
524 * if a type specific operation is underway.
525 */
526
do_promote(struct gfs2_glock * gl)527 static int do_promote(struct gfs2_glock *gl)
528 __releases(&gl->gl_lockref.lock)
529 __acquires(&gl->gl_lockref.lock)
530 {
531 struct gfs2_holder *gh, *tmp, *first_gh;
532 bool incompat_holders_demoted = false;
533 bool lock_released;
534 int ret;
535
536 restart:
537 first_gh = find_first_strong_holder(gl);
538 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
539 lock_released = false;
540 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
541 continue;
542 if (!may_grant(gl, first_gh, gh)) {
543 /*
544 * If we get here, it means we may not grant this holder for
545 * some reason. If this holder is the head of the list, it
546 * means we have a blocked holder at the head, so return 1.
547 */
548 if (gh->gh_list.prev == &gl->gl_holders)
549 return 1;
550 do_error(gl, 0);
551 break;
552 }
553 if (!incompat_holders_demoted) {
554 demote_incompat_holders(gl, first_gh);
555 incompat_holders_demoted = true;
556 first_gh = gh;
557 }
558 if (test_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags) &&
559 !(gh->gh_flags & GL_SKIP) && gl->gl_ops->go_instantiate) {
560 lock_released = true;
561 spin_unlock(&gl->gl_lockref.lock);
562 ret = gfs2_instantiate(gh);
563 spin_lock(&gl->gl_lockref.lock);
564 if (ret) {
565 if (ret == 1)
566 return 2;
567 gh->gh_error = ret;
568 list_del_init(&gh->gh_list);
569 trace_gfs2_glock_queue(gh, 0);
570 gfs2_holder_wake(gh);
571 goto restart;
572 }
573 }
574 set_bit(HIF_HOLDER, &gh->gh_iflags);
575 trace_gfs2_promote(gh);
576 gfs2_holder_wake(gh);
577 /*
578 * If we released the gl_lockref.lock the holders list may have
579 * changed. For that reason, we start again at the start of
580 * the holders queue.
581 */
582 if (lock_released)
583 goto restart;
584 }
585 return 0;
586 }
587
588 /**
589 * find_first_waiter - find the first gh that's waiting for the glock
590 * @gl: the glock
591 */
592
find_first_waiter(const struct gfs2_glock * gl)593 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
594 {
595 struct gfs2_holder *gh;
596
597 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
598 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
599 return gh;
600 }
601 return NULL;
602 }
603
604 /**
605 * state_change - record that the glock is now in a different state
606 * @gl: the glock
607 * @new_state: the new state
608 */
609
state_change(struct gfs2_glock * gl,unsigned int new_state)610 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
611 {
612 int held1, held2;
613
614 held1 = (gl->gl_state != LM_ST_UNLOCKED);
615 held2 = (new_state != LM_ST_UNLOCKED);
616
617 if (held1 != held2) {
618 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
619 if (held2)
620 gl->gl_lockref.count++;
621 else
622 gl->gl_lockref.count--;
623 }
624 if (new_state != gl->gl_target)
625 /* shorten our minimum hold time */
626 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
627 GL_GLOCK_MIN_HOLD);
628 gl->gl_state = new_state;
629 gl->gl_tchange = jiffies;
630 }
631
gfs2_set_demote(struct gfs2_glock * gl)632 static void gfs2_set_demote(struct gfs2_glock *gl)
633 {
634 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
635
636 set_bit(GLF_DEMOTE, &gl->gl_flags);
637 smp_mb();
638 wake_up(&sdp->sd_async_glock_wait);
639 }
640
gfs2_demote_wake(struct gfs2_glock * gl)641 static void gfs2_demote_wake(struct gfs2_glock *gl)
642 {
643 gl->gl_demote_state = LM_ST_EXCLUSIVE;
644 clear_bit(GLF_DEMOTE, &gl->gl_flags);
645 smp_mb__after_atomic();
646 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
647 }
648
649 /**
650 * finish_xmote - The DLM has replied to one of our lock requests
651 * @gl: The glock
652 * @ret: The status from the DLM
653 *
654 */
655
finish_xmote(struct gfs2_glock * gl,unsigned int ret)656 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
657 {
658 const struct gfs2_glock_operations *glops = gl->gl_ops;
659 struct gfs2_holder *gh;
660 unsigned state = ret & LM_OUT_ST_MASK;
661 int rv;
662
663 spin_lock(&gl->gl_lockref.lock);
664 trace_gfs2_glock_state_change(gl, state);
665 state_change(gl, state);
666 gh = find_first_waiter(gl);
667
668 /* Demote to UN request arrived during demote to SH or DF */
669 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
670 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
671 gl->gl_target = LM_ST_UNLOCKED;
672
673 /* Check for state != intended state */
674 if (unlikely(state != gl->gl_target)) {
675 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
676 /* move to back of queue and try next entry */
677 if (ret & LM_OUT_CANCELED) {
678 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
679 list_move_tail(&gh->gh_list, &gl->gl_holders);
680 gh = find_first_waiter(gl);
681 gl->gl_target = gh->gh_state;
682 goto retry;
683 }
684 /* Some error or failed "try lock" - report it */
685 if ((ret & LM_OUT_ERROR) ||
686 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
687 gl->gl_target = gl->gl_state;
688 do_error(gl, ret);
689 goto out;
690 }
691 }
692 switch(state) {
693 /* Unlocked due to conversion deadlock, try again */
694 case LM_ST_UNLOCKED:
695 retry:
696 do_xmote(gl, gh, gl->gl_target);
697 break;
698 /* Conversion fails, unlock and try again */
699 case LM_ST_SHARED:
700 case LM_ST_DEFERRED:
701 do_xmote(gl, gh, LM_ST_UNLOCKED);
702 break;
703 default: /* Everything else */
704 fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n",
705 gl->gl_target, state);
706 GLOCK_BUG_ON(gl, 1);
707 }
708 spin_unlock(&gl->gl_lockref.lock);
709 return;
710 }
711
712 /* Fast path - we got what we asked for */
713 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
714 gfs2_demote_wake(gl);
715 if (state != LM_ST_UNLOCKED) {
716 if (glops->go_xmote_bh) {
717 spin_unlock(&gl->gl_lockref.lock);
718 rv = glops->go_xmote_bh(gl);
719 spin_lock(&gl->gl_lockref.lock);
720 if (rv) {
721 do_error(gl, rv);
722 goto out;
723 }
724 }
725 rv = do_promote(gl);
726 if (rv == 2)
727 goto out_locked;
728 }
729 out:
730 clear_bit(GLF_LOCK, &gl->gl_flags);
731 out_locked:
732 spin_unlock(&gl->gl_lockref.lock);
733 }
734
is_system_glock(struct gfs2_glock * gl)735 static bool is_system_glock(struct gfs2_glock *gl)
736 {
737 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
738 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
739
740 if (gl == m_ip->i_gl)
741 return true;
742 return false;
743 }
744
745 /**
746 * do_xmote - Calls the DLM to change the state of a lock
747 * @gl: The lock state
748 * @gh: The holder (only for promotes)
749 * @target: The target lock state
750 *
751 */
752
do_xmote(struct gfs2_glock * gl,struct gfs2_holder * gh,unsigned int target)753 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
754 __releases(&gl->gl_lockref.lock)
755 __acquires(&gl->gl_lockref.lock)
756 {
757 const struct gfs2_glock_operations *glops = gl->gl_ops;
758 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
759 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
760 int ret;
761
762 if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) &&
763 gh && !(gh->gh_flags & LM_FLAG_NOEXP))
764 return;
765 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
766 LM_FLAG_PRIORITY);
767 GLOCK_BUG_ON(gl, gl->gl_state == target);
768 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
769 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
770 glops->go_inval) {
771 /*
772 * If another process is already doing the invalidate, let that
773 * finish first. The glock state machine will get back to this
774 * holder again later.
775 */
776 if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
777 &gl->gl_flags))
778 return;
779 do_error(gl, 0); /* Fail queued try locks */
780 }
781 gl->gl_req = target;
782 set_bit(GLF_BLOCKING, &gl->gl_flags);
783 if ((gl->gl_req == LM_ST_UNLOCKED) ||
784 (gl->gl_state == LM_ST_EXCLUSIVE) ||
785 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
786 clear_bit(GLF_BLOCKING, &gl->gl_flags);
787 spin_unlock(&gl->gl_lockref.lock);
788 if (glops->go_sync) {
789 ret = glops->go_sync(gl);
790 /* If we had a problem syncing (due to io errors or whatever,
791 * we should not invalidate the metadata or tell dlm to
792 * release the glock to other nodes.
793 */
794 if (ret) {
795 if (cmpxchg(&sdp->sd_log_error, 0, ret)) {
796 fs_err(sdp, "Error %d syncing glock \n", ret);
797 gfs2_dump_glock(NULL, gl, true);
798 }
799 goto skip_inval;
800 }
801 }
802 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) {
803 /*
804 * The call to go_sync should have cleared out the ail list.
805 * If there are still items, we have a problem. We ought to
806 * withdraw, but we can't because the withdraw code also uses
807 * glocks. Warn about the error, dump the glock, then fall
808 * through and wait for logd to do the withdraw for us.
809 */
810 if ((atomic_read(&gl->gl_ail_count) != 0) &&
811 (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) {
812 gfs2_glock_assert_warn(gl,
813 !atomic_read(&gl->gl_ail_count));
814 gfs2_dump_glock(NULL, gl, true);
815 }
816 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
817 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
818 }
819
820 skip_inval:
821 gfs2_glock_hold(gl);
822 /*
823 * Check for an error encountered since we called go_sync and go_inval.
824 * If so, we can't withdraw from the glock code because the withdraw
825 * code itself uses glocks (see function signal_our_withdraw) to
826 * change the mount to read-only. Most importantly, we must not call
827 * dlm to unlock the glock until the journal is in a known good state
828 * (after journal replay) otherwise other nodes may use the object
829 * (rgrp or dinode) and then later, journal replay will corrupt the
830 * file system. The best we can do here is wait for the logd daemon
831 * to see sd_log_error and withdraw, and in the meantime, requeue the
832 * work for later.
833 *
834 * We make a special exception for some system glocks, such as the
835 * system statfs inode glock, which needs to be granted before the
836 * gfs2_quotad daemon can exit, and that exit needs to finish before
837 * we can unmount the withdrawn file system.
838 *
839 * However, if we're just unlocking the lock (say, for unmount, when
840 * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
841 * then it's okay to tell dlm to unlock it.
842 */
843 if (unlikely(sdp->sd_log_error && !gfs2_withdrawn(sdp)))
844 gfs2_withdraw_delayed(sdp);
845 if (glock_blocked_by_withdraw(gl) &&
846 (target != LM_ST_UNLOCKED ||
847 test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags))) {
848 if (!is_system_glock(gl)) {
849 gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD);
850 goto out;
851 } else {
852 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
853 }
854 }
855
856 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
857 /* lock_dlm */
858 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
859 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
860 target == LM_ST_UNLOCKED &&
861 test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
862 finish_xmote(gl, target);
863 gfs2_glock_queue_work(gl, 0);
864 } else if (ret) {
865 fs_err(sdp, "lm_lock ret %d\n", ret);
866 GLOCK_BUG_ON(gl, !gfs2_withdrawn(sdp));
867 }
868 } else { /* lock_nolock */
869 finish_xmote(gl, target);
870 gfs2_glock_queue_work(gl, 0);
871 }
872 out:
873 spin_lock(&gl->gl_lockref.lock);
874 }
875
876 /**
877 * run_queue - do all outstanding tasks related to a glock
878 * @gl: The glock in question
879 * @nonblock: True if we must not block in run_queue
880 *
881 */
882
run_queue(struct gfs2_glock * gl,const int nonblock)883 static void run_queue(struct gfs2_glock *gl, const int nonblock)
884 __releases(&gl->gl_lockref.lock)
885 __acquires(&gl->gl_lockref.lock)
886 {
887 struct gfs2_holder *gh = NULL;
888 int ret;
889
890 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
891 return;
892
893 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
894
895 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
896 gl->gl_demote_state != gl->gl_state) {
897 if (find_first_holder(gl))
898 goto out_unlock;
899 if (nonblock)
900 goto out_sched;
901 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
902 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
903 gl->gl_target = gl->gl_demote_state;
904 } else {
905 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
906 gfs2_demote_wake(gl);
907 ret = do_promote(gl);
908 if (ret == 0)
909 goto out_unlock;
910 if (ret == 2)
911 goto out;
912 gh = find_first_waiter(gl);
913 gl->gl_target = gh->gh_state;
914 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
915 do_error(gl, 0); /* Fail queued try locks */
916 }
917 do_xmote(gl, gh, gl->gl_target);
918 out:
919 return;
920
921 out_sched:
922 clear_bit(GLF_LOCK, &gl->gl_flags);
923 smp_mb__after_atomic();
924 gl->gl_lockref.count++;
925 __gfs2_glock_queue_work(gl, 0);
926 return;
927
928 out_unlock:
929 clear_bit(GLF_LOCK, &gl->gl_flags);
930 smp_mb__after_atomic();
931 return;
932 }
933
gfs2_inode_remember_delete(struct gfs2_glock * gl,u64 generation)934 void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation)
935 {
936 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
937
938 if (ri->ri_magic == 0)
939 ri->ri_magic = cpu_to_be32(GFS2_MAGIC);
940 if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC))
941 ri->ri_generation_deleted = cpu_to_be64(generation);
942 }
943
gfs2_inode_already_deleted(struct gfs2_glock * gl,u64 generation)944 bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation)
945 {
946 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
947
948 if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC))
949 return false;
950 return generation <= be64_to_cpu(ri->ri_generation_deleted);
951 }
952
gfs2_glock_poke(struct gfs2_glock * gl)953 static void gfs2_glock_poke(struct gfs2_glock *gl)
954 {
955 int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP;
956 struct gfs2_holder gh;
957 int error;
958
959 __gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh, _RET_IP_);
960 error = gfs2_glock_nq(&gh);
961 if (!error)
962 gfs2_glock_dq(&gh);
963 gfs2_holder_uninit(&gh);
964 }
965
gfs2_try_evict(struct gfs2_glock * gl)966 static bool gfs2_try_evict(struct gfs2_glock *gl)
967 {
968 struct gfs2_inode *ip;
969 bool evicted = false;
970
971 /*
972 * If there is contention on the iopen glock and we have an inode, try
973 * to grab and release the inode so that it can be evicted. This will
974 * allow the remote node to go ahead and delete the inode without us
975 * having to do it, which will avoid rgrp glock thrashing.
976 *
977 * The remote node is likely still holding the corresponding inode
978 * glock, so it will run before we get to verify that the delete has
979 * happened below.
980 */
981 spin_lock(&gl->gl_lockref.lock);
982 ip = gl->gl_object;
983 if (ip && !igrab(&ip->i_inode))
984 ip = NULL;
985 spin_unlock(&gl->gl_lockref.lock);
986 if (ip) {
987 struct gfs2_glock *inode_gl = NULL;
988
989 gl->gl_no_formal_ino = ip->i_no_formal_ino;
990 set_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
991 d_prune_aliases(&ip->i_inode);
992 iput(&ip->i_inode);
993
994 /* If the inode was evicted, gl->gl_object will now be NULL. */
995 spin_lock(&gl->gl_lockref.lock);
996 ip = gl->gl_object;
997 if (ip) {
998 inode_gl = ip->i_gl;
999 lockref_get(&inode_gl->gl_lockref);
1000 clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
1001 }
1002 spin_unlock(&gl->gl_lockref.lock);
1003 if (inode_gl) {
1004 gfs2_glock_poke(inode_gl);
1005 gfs2_glock_put(inode_gl);
1006 }
1007 evicted = !ip;
1008 }
1009 return evicted;
1010 }
1011
delete_work_func(struct work_struct * work)1012 static void delete_work_func(struct work_struct *work)
1013 {
1014 struct delayed_work *dwork = to_delayed_work(work);
1015 struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete);
1016 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1017 struct inode *inode;
1018 u64 no_addr = gl->gl_name.ln_number;
1019
1020 spin_lock(&gl->gl_lockref.lock);
1021 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1022 spin_unlock(&gl->gl_lockref.lock);
1023
1024 if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
1025 /*
1026 * If we can evict the inode, give the remote node trying to
1027 * delete the inode some time before verifying that the delete
1028 * has happened. Otherwise, if we cause contention on the inode glock
1029 * immediately, the remote node will think that we still have
1030 * the inode in use, and so it will give up waiting.
1031 *
1032 * If we can't evict the inode, signal to the remote node that
1033 * the inode is still in use. We'll later try to delete the
1034 * inode locally in gfs2_evict_inode.
1035 *
1036 * FIXME: We only need to verify that the remote node has
1037 * deleted the inode because nodes before this remote delete
1038 * rework won't cooperate. At a later time, when we no longer
1039 * care about compatibility with such nodes, we can skip this
1040 * step entirely.
1041 */
1042 if (gfs2_try_evict(gl)) {
1043 if (gfs2_queue_delete_work(gl, 5 * HZ))
1044 return;
1045 }
1046 goto out;
1047 }
1048
1049 inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino,
1050 GFS2_BLKST_UNLINKED);
1051 if (!IS_ERR_OR_NULL(inode)) {
1052 d_prune_aliases(inode);
1053 iput(inode);
1054 }
1055 out:
1056 gfs2_glock_put(gl);
1057 }
1058
glock_work_func(struct work_struct * work)1059 static void glock_work_func(struct work_struct *work)
1060 {
1061 unsigned long delay = 0;
1062 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
1063 unsigned int drop_refs = 1;
1064
1065 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
1066 finish_xmote(gl, gl->gl_reply);
1067 drop_refs++;
1068 }
1069 spin_lock(&gl->gl_lockref.lock);
1070 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1071 gl->gl_state != LM_ST_UNLOCKED &&
1072 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
1073 unsigned long holdtime, now = jiffies;
1074
1075 holdtime = gl->gl_tchange + gl->gl_hold_time;
1076 if (time_before(now, holdtime))
1077 delay = holdtime - now;
1078
1079 if (!delay) {
1080 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1081 gfs2_set_demote(gl);
1082 }
1083 }
1084 run_queue(gl, 0);
1085 if (delay) {
1086 /* Keep one glock reference for the work we requeue. */
1087 drop_refs--;
1088 if (gl->gl_name.ln_type != LM_TYPE_INODE)
1089 delay = 0;
1090 __gfs2_glock_queue_work(gl, delay);
1091 }
1092
1093 /*
1094 * Drop the remaining glock references manually here. (Mind that
1095 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
1096 * here as well.)
1097 */
1098 gl->gl_lockref.count -= drop_refs;
1099 if (!gl->gl_lockref.count) {
1100 __gfs2_glock_put(gl);
1101 return;
1102 }
1103 spin_unlock(&gl->gl_lockref.lock);
1104 }
1105
find_insert_glock(struct lm_lockname * name,struct gfs2_glock * new)1106 static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
1107 struct gfs2_glock *new)
1108 {
1109 struct wait_glock_queue wait;
1110 wait_queue_head_t *wq = glock_waitqueue(name);
1111 struct gfs2_glock *gl;
1112
1113 wait.name = name;
1114 init_wait(&wait.wait);
1115 wait.wait.func = glock_wake_function;
1116
1117 again:
1118 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1119 rcu_read_lock();
1120 if (new) {
1121 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
1122 &new->gl_node, ht_parms);
1123 if (IS_ERR(gl))
1124 goto out;
1125 } else {
1126 gl = rhashtable_lookup_fast(&gl_hash_table,
1127 name, ht_parms);
1128 }
1129 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
1130 rcu_read_unlock();
1131 schedule();
1132 goto again;
1133 }
1134 out:
1135 rcu_read_unlock();
1136 finish_wait(wq, &wait.wait);
1137 return gl;
1138 }
1139
1140 /**
1141 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
1142 * @sdp: The GFS2 superblock
1143 * @number: the lock number
1144 * @glops: The glock_operations to use
1145 * @create: If 0, don't create the glock if it doesn't exist
1146 * @glp: the glock is returned here
1147 *
1148 * This does not lock a glock, just finds/creates structures for one.
1149 *
1150 * Returns: errno
1151 */
1152
gfs2_glock_get(struct gfs2_sbd * sdp,u64 number,const struct gfs2_glock_operations * glops,int create,struct gfs2_glock ** glp)1153 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
1154 const struct gfs2_glock_operations *glops, int create,
1155 struct gfs2_glock **glp)
1156 {
1157 struct super_block *s = sdp->sd_vfs;
1158 struct lm_lockname name = { .ln_number = number,
1159 .ln_type = glops->go_type,
1160 .ln_sbd = sdp };
1161 struct gfs2_glock *gl, *tmp;
1162 struct address_space *mapping;
1163 struct kmem_cache *cachep;
1164 int ret = 0;
1165
1166 gl = find_insert_glock(&name, NULL);
1167 if (gl) {
1168 *glp = gl;
1169 return 0;
1170 }
1171 if (!create)
1172 return -ENOENT;
1173
1174 if (glops->go_flags & GLOF_ASPACE)
1175 cachep = gfs2_glock_aspace_cachep;
1176 else
1177 cachep = gfs2_glock_cachep;
1178 gl = kmem_cache_alloc(cachep, GFP_NOFS);
1179 if (!gl)
1180 return -ENOMEM;
1181
1182 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
1183
1184 if (glops->go_flags & GLOF_LVB) {
1185 gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1186 if (!gl->gl_lksb.sb_lvbptr) {
1187 kmem_cache_free(cachep, gl);
1188 return -ENOMEM;
1189 }
1190 }
1191
1192 atomic_inc(&sdp->sd_glock_disposal);
1193 gl->gl_node.next = NULL;
1194 gl->gl_flags = glops->go_instantiate ? BIT(GLF_INSTANTIATE_NEEDED) : 0;
1195 gl->gl_name = name;
1196 lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass);
1197 gl->gl_lockref.count = 1;
1198 gl->gl_state = LM_ST_UNLOCKED;
1199 gl->gl_target = LM_ST_UNLOCKED;
1200 gl->gl_demote_state = LM_ST_EXCLUSIVE;
1201 gl->gl_ops = glops;
1202 gl->gl_dstamp = 0;
1203 preempt_disable();
1204 /* We use the global stats to estimate the initial per-glock stats */
1205 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
1206 preempt_enable();
1207 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
1208 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
1209 gl->gl_tchange = jiffies;
1210 gl->gl_object = NULL;
1211 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
1212 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
1213 if (gl->gl_name.ln_type == LM_TYPE_IOPEN)
1214 INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func);
1215
1216 mapping = gfs2_glock2aspace(gl);
1217 if (mapping) {
1218 mapping->a_ops = &gfs2_meta_aops;
1219 mapping->host = s->s_bdev->bd_inode;
1220 mapping->flags = 0;
1221 mapping_set_gfp_mask(mapping, GFP_NOFS);
1222 mapping->private_data = NULL;
1223 mapping->writeback_index = 0;
1224 }
1225
1226 tmp = find_insert_glock(&name, gl);
1227 if (!tmp) {
1228 *glp = gl;
1229 goto out;
1230 }
1231 if (IS_ERR(tmp)) {
1232 ret = PTR_ERR(tmp);
1233 goto out_free;
1234 }
1235 *glp = tmp;
1236
1237 out_free:
1238 kfree(gl->gl_lksb.sb_lvbptr);
1239 kmem_cache_free(cachep, gl);
1240 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
1241 wake_up(&sdp->sd_glock_wait);
1242
1243 out:
1244 return ret;
1245 }
1246
1247 /**
1248 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
1249 * @gl: the glock
1250 * @state: the state we're requesting
1251 * @flags: the modifier flags
1252 * @gh: the holder structure
1253 *
1254 */
1255
__gfs2_holder_init(struct gfs2_glock * gl,unsigned int state,u16 flags,struct gfs2_holder * gh,unsigned long ip)1256 void __gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
1257 struct gfs2_holder *gh, unsigned long ip)
1258 {
1259 INIT_LIST_HEAD(&gh->gh_list);
1260 gh->gh_gl = gl;
1261 gh->gh_ip = ip;
1262 gh->gh_owner_pid = get_pid(task_pid(current));
1263 gh->gh_state = state;
1264 gh->gh_flags = flags;
1265 gh->gh_error = 0;
1266 gh->gh_iflags = 0;
1267 gfs2_glock_hold(gl);
1268 }
1269
1270 /**
1271 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1272 * @state: the state we're requesting
1273 * @flags: the modifier flags
1274 * @gh: the holder structure
1275 *
1276 * Don't mess with the glock.
1277 *
1278 */
1279
gfs2_holder_reinit(unsigned int state,u16 flags,struct gfs2_holder * gh)1280 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
1281 {
1282 gh->gh_state = state;
1283 gh->gh_flags = flags;
1284 gh->gh_iflags = 0;
1285 gh->gh_ip = _RET_IP_;
1286 put_pid(gh->gh_owner_pid);
1287 gh->gh_owner_pid = get_pid(task_pid(current));
1288 }
1289
1290 /**
1291 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1292 * @gh: the holder structure
1293 *
1294 */
1295
gfs2_holder_uninit(struct gfs2_holder * gh)1296 void gfs2_holder_uninit(struct gfs2_holder *gh)
1297 {
1298 put_pid(gh->gh_owner_pid);
1299 gfs2_glock_put(gh->gh_gl);
1300 gfs2_holder_mark_uninitialized(gh);
1301 gh->gh_ip = 0;
1302 }
1303
gfs2_glock_update_hold_time(struct gfs2_glock * gl,unsigned long start_time)1304 static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
1305 unsigned long start_time)
1306 {
1307 /* Have we waited longer that a second? */
1308 if (time_after(jiffies, start_time + HZ)) {
1309 /* Lengthen the minimum hold time. */
1310 gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
1311 GL_GLOCK_MAX_HOLD);
1312 }
1313 }
1314
1315 /**
1316 * gfs2_glock_wait - wait on a glock acquisition
1317 * @gh: the glock holder
1318 *
1319 * Returns: 0 on success
1320 */
1321
gfs2_glock_wait(struct gfs2_holder * gh)1322 int gfs2_glock_wait(struct gfs2_holder *gh)
1323 {
1324 unsigned long start_time = jiffies;
1325
1326 might_sleep();
1327 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1328 gfs2_glock_update_hold_time(gh->gh_gl, start_time);
1329 return gh->gh_error;
1330 }
1331
glocks_pending(unsigned int num_gh,struct gfs2_holder * ghs)1332 static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
1333 {
1334 int i;
1335
1336 for (i = 0; i < num_gh; i++)
1337 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
1338 return 1;
1339 return 0;
1340 }
1341
1342 /**
1343 * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1344 * @num_gh: the number of holders in the array
1345 * @ghs: the glock holder array
1346 *
1347 * Returns: 0 on success, meaning all glocks have been granted and are held.
1348 * -ESTALE if the request timed out, meaning all glocks were released,
1349 * and the caller should retry the operation.
1350 */
1351
gfs2_glock_async_wait(unsigned int num_gh,struct gfs2_holder * ghs)1352 int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
1353 {
1354 struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
1355 int i, ret = 0, timeout = 0;
1356 unsigned long start_time = jiffies;
1357 bool keep_waiting;
1358
1359 might_sleep();
1360 /*
1361 * Total up the (minimum hold time * 2) of all glocks and use that to
1362 * determine the max amount of time we should wait.
1363 */
1364 for (i = 0; i < num_gh; i++)
1365 timeout += ghs[i].gh_gl->gl_hold_time << 1;
1366
1367 wait_for_dlm:
1368 if (!wait_event_timeout(sdp->sd_async_glock_wait,
1369 !glocks_pending(num_gh, ghs), timeout))
1370 ret = -ESTALE; /* request timed out. */
1371
1372 /*
1373 * If dlm granted all our requests, we need to adjust the glock
1374 * minimum hold time values according to how long we waited.
1375 *
1376 * If our request timed out, we need to repeatedly release any held
1377 * glocks we acquired thus far to allow dlm to acquire the remaining
1378 * glocks without deadlocking. We cannot currently cancel outstanding
1379 * glock acquisitions.
1380 *
1381 * The HIF_WAIT bit tells us which requests still need a response from
1382 * dlm.
1383 *
1384 * If dlm sent us any errors, we return the first error we find.
1385 */
1386 keep_waiting = false;
1387 for (i = 0; i < num_gh; i++) {
1388 /* Skip holders we have already dequeued below. */
1389 if (!gfs2_holder_queued(&ghs[i]))
1390 continue;
1391 /* Skip holders with a pending DLM response. */
1392 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags)) {
1393 keep_waiting = true;
1394 continue;
1395 }
1396
1397 if (test_bit(HIF_HOLDER, &ghs[i].gh_iflags)) {
1398 if (ret == -ESTALE)
1399 gfs2_glock_dq(&ghs[i]);
1400 else
1401 gfs2_glock_update_hold_time(ghs[i].gh_gl,
1402 start_time);
1403 }
1404 if (!ret)
1405 ret = ghs[i].gh_error;
1406 }
1407
1408 if (keep_waiting)
1409 goto wait_for_dlm;
1410
1411 /*
1412 * At this point, we've either acquired all locks or released them all.
1413 */
1414 return ret;
1415 }
1416
1417 /**
1418 * handle_callback - process a demote request
1419 * @gl: the glock
1420 * @state: the state the caller wants us to change to
1421 * @delay: zero to demote immediately; otherwise pending demote
1422 * @remote: true if this came from a different cluster node
1423 *
1424 * There are only two requests that we are going to see in actual
1425 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1426 */
1427
handle_callback(struct gfs2_glock * gl,unsigned int state,unsigned long delay,bool remote)1428 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
1429 unsigned long delay, bool remote)
1430 {
1431 if (delay)
1432 set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1433 else
1434 gfs2_set_demote(gl);
1435 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
1436 gl->gl_demote_state = state;
1437 gl->gl_demote_time = jiffies;
1438 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
1439 gl->gl_demote_state != state) {
1440 gl->gl_demote_state = LM_ST_UNLOCKED;
1441 }
1442 if (gl->gl_ops->go_callback)
1443 gl->gl_ops->go_callback(gl, remote);
1444 trace_gfs2_demote_rq(gl, remote);
1445 }
1446
gfs2_print_dbg(struct seq_file * seq,const char * fmt,...)1447 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
1448 {
1449 struct va_format vaf;
1450 va_list args;
1451
1452 va_start(args, fmt);
1453
1454 if (seq) {
1455 seq_vprintf(seq, fmt, args);
1456 } else {
1457 vaf.fmt = fmt;
1458 vaf.va = &args;
1459
1460 pr_err("%pV", &vaf);
1461 }
1462
1463 va_end(args);
1464 }
1465
1466 /**
1467 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1468 * @gh: the holder structure to add
1469 *
1470 * Eventually we should move the recursive locking trap to a
1471 * debugging option or something like that. This is the fast
1472 * path and needs to have the minimum number of distractions.
1473 *
1474 */
1475
add_to_queue(struct gfs2_holder * gh)1476 static inline void add_to_queue(struct gfs2_holder *gh)
1477 __releases(&gl->gl_lockref.lock)
1478 __acquires(&gl->gl_lockref.lock)
1479 {
1480 struct gfs2_glock *gl = gh->gh_gl;
1481 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1482 struct list_head *insert_pt = NULL;
1483 struct gfs2_holder *gh2;
1484 int try_futile = 0;
1485
1486 GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1487 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1488 GLOCK_BUG_ON(gl, true);
1489
1490 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1491 if (test_bit(GLF_LOCK, &gl->gl_flags)) {
1492 struct gfs2_holder *first_gh;
1493
1494 first_gh = find_first_strong_holder(gl);
1495 try_futile = !may_grant(gl, first_gh, gh);
1496 }
1497 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1498 goto fail;
1499 }
1500
1501 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1502 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
1503 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK) &&
1504 !test_bit(HIF_MAY_DEMOTE, &gh2->gh_iflags)))
1505 goto trap_recursive;
1506 if (try_futile &&
1507 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1508 fail:
1509 gh->gh_error = GLR_TRYFAILED;
1510 gfs2_holder_wake(gh);
1511 return;
1512 }
1513 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1514 continue;
1515 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1516 insert_pt = &gh2->gh_list;
1517 }
1518 trace_gfs2_glock_queue(gh, 1);
1519 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1520 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1521 if (likely(insert_pt == NULL)) {
1522 list_add_tail(&gh->gh_list, &gl->gl_holders);
1523 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1524 goto do_cancel;
1525 return;
1526 }
1527 list_add_tail(&gh->gh_list, insert_pt);
1528 do_cancel:
1529 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
1530 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1531 spin_unlock(&gl->gl_lockref.lock);
1532 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1533 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1534 spin_lock(&gl->gl_lockref.lock);
1535 }
1536 return;
1537
1538 trap_recursive:
1539 fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1540 fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1541 fs_err(sdp, "lock type: %d req lock state : %d\n",
1542 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1543 fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1544 fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1545 fs_err(sdp, "lock type: %d req lock state : %d\n",
1546 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1547 gfs2_dump_glock(NULL, gl, true);
1548 BUG();
1549 }
1550
1551 /**
1552 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1553 * @gh: the holder structure
1554 *
1555 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1556 *
1557 * Returns: 0, GLR_TRYFAILED, or errno on failure
1558 */
1559
gfs2_glock_nq(struct gfs2_holder * gh)1560 int gfs2_glock_nq(struct gfs2_holder *gh)
1561 {
1562 struct gfs2_glock *gl = gh->gh_gl;
1563 int error = 0;
1564
1565 if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP))
1566 return -EIO;
1567
1568 if (test_bit(GLF_LRU, &gl->gl_flags))
1569 gfs2_glock_remove_from_lru(gl);
1570
1571 spin_lock(&gl->gl_lockref.lock);
1572 add_to_queue(gh);
1573 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1574 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1575 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1576 gl->gl_lockref.count++;
1577 __gfs2_glock_queue_work(gl, 0);
1578 }
1579 run_queue(gl, 1);
1580 spin_unlock(&gl->gl_lockref.lock);
1581
1582 if (!(gh->gh_flags & GL_ASYNC))
1583 error = gfs2_glock_wait(gh);
1584
1585 return error;
1586 }
1587
1588 /**
1589 * gfs2_glock_poll - poll to see if an async request has been completed
1590 * @gh: the holder
1591 *
1592 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1593 */
1594
gfs2_glock_poll(struct gfs2_holder * gh)1595 int gfs2_glock_poll(struct gfs2_holder *gh)
1596 {
1597 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1598 }
1599
needs_demote(struct gfs2_glock * gl)1600 static inline bool needs_demote(struct gfs2_glock *gl)
1601 {
1602 return (test_bit(GLF_DEMOTE, &gl->gl_flags) ||
1603 test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags));
1604 }
1605
__gfs2_glock_dq(struct gfs2_holder * gh)1606 static void __gfs2_glock_dq(struct gfs2_holder *gh)
1607 {
1608 struct gfs2_glock *gl = gh->gh_gl;
1609 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1610 unsigned delay = 0;
1611 int fast_path = 0;
1612
1613 /*
1614 * This while loop is similar to function demote_incompat_holders:
1615 * If the glock is due to be demoted (which may be from another node
1616 * or even if this holder is GL_NOCACHE), the weak holders are
1617 * demoted as well, allowing the glock to be demoted.
1618 */
1619 while (gh) {
1620 /*
1621 * If we're in the process of file system withdraw, we cannot
1622 * just dequeue any glocks until our journal is recovered, lest
1623 * we introduce file system corruption. We need two exceptions
1624 * to this rule: We need to allow unlocking of nondisk glocks
1625 * and the glock for our own journal that needs recovery.
1626 */
1627 if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) &&
1628 glock_blocked_by_withdraw(gl) &&
1629 gh->gh_gl != sdp->sd_jinode_gl) {
1630 sdp->sd_glock_dqs_held++;
1631 spin_unlock(&gl->gl_lockref.lock);
1632 might_sleep();
1633 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
1634 TASK_UNINTERRUPTIBLE);
1635 spin_lock(&gl->gl_lockref.lock);
1636 }
1637
1638 /*
1639 * This holder should not be cached, so mark it for demote.
1640 * Note: this should be done before the check for needs_demote
1641 * below.
1642 */
1643 if (gh->gh_flags & GL_NOCACHE)
1644 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1645
1646 list_del_init(&gh->gh_list);
1647 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1648 trace_gfs2_glock_queue(gh, 0);
1649
1650 /*
1651 * If there hasn't been a demote request we are done.
1652 * (Let the remaining holders, if any, keep holding it.)
1653 */
1654 if (!needs_demote(gl)) {
1655 if (list_empty(&gl->gl_holders))
1656 fast_path = 1;
1657 break;
1658 }
1659 /*
1660 * If we have another strong holder (we cannot auto-demote)
1661 * we are done. It keeps holding it until it is done.
1662 */
1663 if (find_first_strong_holder(gl))
1664 break;
1665
1666 /*
1667 * If we have a weak holder at the head of the list, it
1668 * (and all others like it) must be auto-demoted. If there
1669 * are no more weak holders, we exit the while loop.
1670 */
1671 gh = find_first_holder(gl);
1672 }
1673
1674 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1675 gfs2_glock_add_to_lru(gl);
1676
1677 if (unlikely(!fast_path)) {
1678 gl->gl_lockref.count++;
1679 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1680 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1681 gl->gl_name.ln_type == LM_TYPE_INODE)
1682 delay = gl->gl_hold_time;
1683 __gfs2_glock_queue_work(gl, delay);
1684 }
1685 }
1686
1687 /**
1688 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1689 * @gh: the glock holder
1690 *
1691 */
gfs2_glock_dq(struct gfs2_holder * gh)1692 void gfs2_glock_dq(struct gfs2_holder *gh)
1693 {
1694 struct gfs2_glock *gl = gh->gh_gl;
1695
1696 spin_lock(&gl->gl_lockref.lock);
1697 __gfs2_glock_dq(gh);
1698 spin_unlock(&gl->gl_lockref.lock);
1699 }
1700
gfs2_glock_dq_wait(struct gfs2_holder * gh)1701 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1702 {
1703 struct gfs2_glock *gl = gh->gh_gl;
1704 gfs2_glock_dq(gh);
1705 might_sleep();
1706 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1707 }
1708
1709 /**
1710 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1711 * @gh: the holder structure
1712 *
1713 */
1714
gfs2_glock_dq_uninit(struct gfs2_holder * gh)1715 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1716 {
1717 gfs2_glock_dq(gh);
1718 gfs2_holder_uninit(gh);
1719 }
1720
1721 /**
1722 * gfs2_glock_nq_num - acquire a glock based on lock number
1723 * @sdp: the filesystem
1724 * @number: the lock number
1725 * @glops: the glock operations for the type of glock
1726 * @state: the state to acquire the glock in
1727 * @flags: modifier flags for the acquisition
1728 * @gh: the struct gfs2_holder
1729 *
1730 * Returns: errno
1731 */
1732
gfs2_glock_nq_num(struct gfs2_sbd * sdp,u64 number,const struct gfs2_glock_operations * glops,unsigned int state,u16 flags,struct gfs2_holder * gh)1733 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1734 const struct gfs2_glock_operations *glops,
1735 unsigned int state, u16 flags, struct gfs2_holder *gh)
1736 {
1737 struct gfs2_glock *gl;
1738 int error;
1739
1740 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1741 if (!error) {
1742 error = gfs2_glock_nq_init(gl, state, flags, gh);
1743 gfs2_glock_put(gl);
1744 }
1745
1746 return error;
1747 }
1748
1749 /**
1750 * glock_compare - Compare two struct gfs2_glock structures for sorting
1751 * @arg_a: the first structure
1752 * @arg_b: the second structure
1753 *
1754 */
1755
glock_compare(const void * arg_a,const void * arg_b)1756 static int glock_compare(const void *arg_a, const void *arg_b)
1757 {
1758 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1759 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1760 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1761 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1762
1763 if (a->ln_number > b->ln_number)
1764 return 1;
1765 if (a->ln_number < b->ln_number)
1766 return -1;
1767 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1768 return 0;
1769 }
1770
1771 /**
1772 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1773 * @num_gh: the number of structures
1774 * @ghs: an array of struct gfs2_holder structures
1775 * @p: placeholder for the holder structure to pass back
1776 *
1777 * Returns: 0 on success (all glocks acquired),
1778 * errno on failure (no glocks acquired)
1779 */
1780
nq_m_sync(unsigned int num_gh,struct gfs2_holder * ghs,struct gfs2_holder ** p)1781 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1782 struct gfs2_holder **p)
1783 {
1784 unsigned int x;
1785 int error = 0;
1786
1787 for (x = 0; x < num_gh; x++)
1788 p[x] = &ghs[x];
1789
1790 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1791
1792 for (x = 0; x < num_gh; x++) {
1793 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1794
1795 error = gfs2_glock_nq(p[x]);
1796 if (error) {
1797 while (x--)
1798 gfs2_glock_dq(p[x]);
1799 break;
1800 }
1801 }
1802
1803 return error;
1804 }
1805
1806 /**
1807 * gfs2_glock_nq_m - acquire multiple glocks
1808 * @num_gh: the number of structures
1809 * @ghs: an array of struct gfs2_holder structures
1810 *
1811 *
1812 * Returns: 0 on success (all glocks acquired),
1813 * errno on failure (no glocks acquired)
1814 */
1815
gfs2_glock_nq_m(unsigned int num_gh,struct gfs2_holder * ghs)1816 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1817 {
1818 struct gfs2_holder *tmp[4];
1819 struct gfs2_holder **pph = tmp;
1820 int error = 0;
1821
1822 switch(num_gh) {
1823 case 0:
1824 return 0;
1825 case 1:
1826 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1827 return gfs2_glock_nq(ghs);
1828 default:
1829 if (num_gh <= 4)
1830 break;
1831 pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *),
1832 GFP_NOFS);
1833 if (!pph)
1834 return -ENOMEM;
1835 }
1836
1837 error = nq_m_sync(num_gh, ghs, pph);
1838
1839 if (pph != tmp)
1840 kfree(pph);
1841
1842 return error;
1843 }
1844
1845 /**
1846 * gfs2_glock_dq_m - release multiple glocks
1847 * @num_gh: the number of structures
1848 * @ghs: an array of struct gfs2_holder structures
1849 *
1850 */
1851
gfs2_glock_dq_m(unsigned int num_gh,struct gfs2_holder * ghs)1852 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1853 {
1854 while (num_gh--)
1855 gfs2_glock_dq(&ghs[num_gh]);
1856 }
1857
gfs2_glock_cb(struct gfs2_glock * gl,unsigned int state)1858 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1859 {
1860 unsigned long delay = 0;
1861 unsigned long holdtime;
1862 unsigned long now = jiffies;
1863
1864 gfs2_glock_hold(gl);
1865 spin_lock(&gl->gl_lockref.lock);
1866 holdtime = gl->gl_tchange + gl->gl_hold_time;
1867 if (!list_empty(&gl->gl_holders) &&
1868 gl->gl_name.ln_type == LM_TYPE_INODE) {
1869 if (time_before(now, holdtime))
1870 delay = holdtime - now;
1871 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1872 delay = gl->gl_hold_time;
1873 }
1874 /*
1875 * Note 1: We cannot call demote_incompat_holders from handle_callback
1876 * or gfs2_set_demote due to recursion problems like: gfs2_glock_dq ->
1877 * handle_callback -> demote_incompat_holders -> gfs2_glock_dq
1878 * Plus, we only want to demote the holders if the request comes from
1879 * a remote cluster node because local holder conflicts are resolved
1880 * elsewhere.
1881 *
1882 * Note 2: if a remote node wants this glock in EX mode, lock_dlm will
1883 * request that we set our state to UNLOCKED. Here we mock up a holder
1884 * to make it look like someone wants the lock EX locally. Any SH
1885 * and DF requests should be able to share the lock without demoting.
1886 *
1887 * Note 3: We only want to demote the demoteable holders when there
1888 * are no more strong holders. The demoteable holders might as well
1889 * keep the glock until the last strong holder is done with it.
1890 */
1891 if (!find_first_strong_holder(gl)) {
1892 struct gfs2_holder mock_gh = {
1893 .gh_gl = gl,
1894 .gh_state = (state == LM_ST_UNLOCKED) ?
1895 LM_ST_EXCLUSIVE : state,
1896 .gh_iflags = BIT(HIF_HOLDER)
1897 };
1898
1899 demote_incompat_holders(gl, &mock_gh);
1900 }
1901 handle_callback(gl, state, delay, true);
1902 __gfs2_glock_queue_work(gl, delay);
1903 spin_unlock(&gl->gl_lockref.lock);
1904 }
1905
1906 /**
1907 * gfs2_should_freeze - Figure out if glock should be frozen
1908 * @gl: The glock in question
1909 *
1910 * Glocks are not frozen if (a) the result of the dlm operation is
1911 * an error, (b) the locking operation was an unlock operation or
1912 * (c) if there is a "noexp" flagged request anywhere in the queue
1913 *
1914 * Returns: 1 if freezing should occur, 0 otherwise
1915 */
1916
gfs2_should_freeze(const struct gfs2_glock * gl)1917 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1918 {
1919 const struct gfs2_holder *gh;
1920
1921 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1922 return 0;
1923 if (gl->gl_target == LM_ST_UNLOCKED)
1924 return 0;
1925
1926 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1927 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1928 continue;
1929 if (LM_FLAG_NOEXP & gh->gh_flags)
1930 return 0;
1931 }
1932
1933 return 1;
1934 }
1935
1936 /**
1937 * gfs2_glock_complete - Callback used by locking
1938 * @gl: Pointer to the glock
1939 * @ret: The return value from the dlm
1940 *
1941 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1942 * to use a bitfield shared with other glock state fields.
1943 */
1944
gfs2_glock_complete(struct gfs2_glock * gl,int ret)1945 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1946 {
1947 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1948
1949 spin_lock(&gl->gl_lockref.lock);
1950 gl->gl_reply = ret;
1951
1952 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1953 if (gfs2_should_freeze(gl)) {
1954 set_bit(GLF_FROZEN, &gl->gl_flags);
1955 spin_unlock(&gl->gl_lockref.lock);
1956 return;
1957 }
1958 }
1959
1960 gl->gl_lockref.count++;
1961 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1962 __gfs2_glock_queue_work(gl, 0);
1963 spin_unlock(&gl->gl_lockref.lock);
1964 }
1965
glock_cmp(void * priv,const struct list_head * a,const struct list_head * b)1966 static int glock_cmp(void *priv, const struct list_head *a,
1967 const struct list_head *b)
1968 {
1969 struct gfs2_glock *gla, *glb;
1970
1971 gla = list_entry(a, struct gfs2_glock, gl_lru);
1972 glb = list_entry(b, struct gfs2_glock, gl_lru);
1973
1974 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1975 return 1;
1976 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1977 return -1;
1978
1979 return 0;
1980 }
1981
1982 /**
1983 * gfs2_dispose_glock_lru - Demote a list of glocks
1984 * @list: The list to dispose of
1985 *
1986 * Disposing of glocks may involve disk accesses, so that here we sort
1987 * the glocks by number (i.e. disk location of the inodes) so that if
1988 * there are any such accesses, they'll be sent in order (mostly).
1989 *
1990 * Must be called under the lru_lock, but may drop and retake this
1991 * lock. While the lru_lock is dropped, entries may vanish from the
1992 * list, but no new entries will appear on the list (since it is
1993 * private)
1994 */
1995
gfs2_dispose_glock_lru(struct list_head * list)1996 static void gfs2_dispose_glock_lru(struct list_head *list)
1997 __releases(&lru_lock)
1998 __acquires(&lru_lock)
1999 {
2000 struct gfs2_glock *gl;
2001
2002 list_sort(NULL, list, glock_cmp);
2003
2004 while(!list_empty(list)) {
2005 gl = list_first_entry(list, struct gfs2_glock, gl_lru);
2006 list_del_init(&gl->gl_lru);
2007 clear_bit(GLF_LRU, &gl->gl_flags);
2008 if (!spin_trylock(&gl->gl_lockref.lock)) {
2009 add_back_to_lru:
2010 list_add(&gl->gl_lru, &lru_list);
2011 set_bit(GLF_LRU, &gl->gl_flags);
2012 atomic_inc(&lru_count);
2013 continue;
2014 }
2015 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
2016 spin_unlock(&gl->gl_lockref.lock);
2017 goto add_back_to_lru;
2018 }
2019 gl->gl_lockref.count++;
2020 if (demote_ok(gl))
2021 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2022 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
2023 __gfs2_glock_queue_work(gl, 0);
2024 spin_unlock(&gl->gl_lockref.lock);
2025 cond_resched_lock(&lru_lock);
2026 }
2027 }
2028
2029 /**
2030 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
2031 * @nr: The number of entries to scan
2032 *
2033 * This function selects the entries on the LRU which are able to
2034 * be demoted, and then kicks off the process by calling
2035 * gfs2_dispose_glock_lru() above.
2036 */
2037
gfs2_scan_glock_lru(int nr)2038 static long gfs2_scan_glock_lru(int nr)
2039 {
2040 struct gfs2_glock *gl;
2041 LIST_HEAD(skipped);
2042 LIST_HEAD(dispose);
2043 long freed = 0;
2044
2045 spin_lock(&lru_lock);
2046 while ((nr-- >= 0) && !list_empty(&lru_list)) {
2047 gl = list_first_entry(&lru_list, struct gfs2_glock, gl_lru);
2048
2049 /* Test for being demotable */
2050 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
2051 list_move(&gl->gl_lru, &dispose);
2052 atomic_dec(&lru_count);
2053 freed++;
2054 continue;
2055 }
2056
2057 list_move(&gl->gl_lru, &skipped);
2058 }
2059 list_splice(&skipped, &lru_list);
2060 if (!list_empty(&dispose))
2061 gfs2_dispose_glock_lru(&dispose);
2062 spin_unlock(&lru_lock);
2063
2064 return freed;
2065 }
2066
gfs2_glock_shrink_scan(struct shrinker * shrink,struct shrink_control * sc)2067 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
2068 struct shrink_control *sc)
2069 {
2070 if (!(sc->gfp_mask & __GFP_FS))
2071 return SHRINK_STOP;
2072 return gfs2_scan_glock_lru(sc->nr_to_scan);
2073 }
2074
gfs2_glock_shrink_count(struct shrinker * shrink,struct shrink_control * sc)2075 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
2076 struct shrink_control *sc)
2077 {
2078 return vfs_pressure_ratio(atomic_read(&lru_count));
2079 }
2080
2081 static struct shrinker glock_shrinker = {
2082 .seeks = DEFAULT_SEEKS,
2083 .count_objects = gfs2_glock_shrink_count,
2084 .scan_objects = gfs2_glock_shrink_scan,
2085 };
2086
2087 /**
2088 * glock_hash_walk - Call a function for glock in a hash bucket
2089 * @examiner: the function
2090 * @sdp: the filesystem
2091 *
2092 * Note that the function can be called multiple times on the same
2093 * object. So the user must ensure that the function can cope with
2094 * that.
2095 */
2096
glock_hash_walk(glock_examiner examiner,const struct gfs2_sbd * sdp)2097 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
2098 {
2099 struct gfs2_glock *gl;
2100 struct rhashtable_iter iter;
2101
2102 rhashtable_walk_enter(&gl_hash_table, &iter);
2103
2104 do {
2105 rhashtable_walk_start(&iter);
2106
2107 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl)) {
2108 if (gl->gl_name.ln_sbd == sdp)
2109 examiner(gl);
2110 }
2111
2112 rhashtable_walk_stop(&iter);
2113 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
2114
2115 rhashtable_walk_exit(&iter);
2116 }
2117
gfs2_queue_delete_work(struct gfs2_glock * gl,unsigned long delay)2118 bool gfs2_queue_delete_work(struct gfs2_glock *gl, unsigned long delay)
2119 {
2120 bool queued;
2121
2122 spin_lock(&gl->gl_lockref.lock);
2123 queued = queue_delayed_work(gfs2_delete_workqueue,
2124 &gl->gl_delete, delay);
2125 if (queued)
2126 set_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2127 spin_unlock(&gl->gl_lockref.lock);
2128 return queued;
2129 }
2130
gfs2_cancel_delete_work(struct gfs2_glock * gl)2131 void gfs2_cancel_delete_work(struct gfs2_glock *gl)
2132 {
2133 if (cancel_delayed_work(&gl->gl_delete)) {
2134 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2135 gfs2_glock_put(gl);
2136 }
2137 }
2138
gfs2_delete_work_queued(const struct gfs2_glock * gl)2139 bool gfs2_delete_work_queued(const struct gfs2_glock *gl)
2140 {
2141 return test_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2142 }
2143
flush_delete_work(struct gfs2_glock * gl)2144 static void flush_delete_work(struct gfs2_glock *gl)
2145 {
2146 if (gl->gl_name.ln_type == LM_TYPE_IOPEN) {
2147 if (cancel_delayed_work(&gl->gl_delete)) {
2148 queue_delayed_work(gfs2_delete_workqueue,
2149 &gl->gl_delete, 0);
2150 }
2151 }
2152 }
2153
gfs2_flush_delete_work(struct gfs2_sbd * sdp)2154 void gfs2_flush_delete_work(struct gfs2_sbd *sdp)
2155 {
2156 glock_hash_walk(flush_delete_work, sdp);
2157 flush_workqueue(gfs2_delete_workqueue);
2158 }
2159
2160 /**
2161 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
2162 * @gl: The glock to thaw
2163 *
2164 */
2165
thaw_glock(struct gfs2_glock * gl)2166 static void thaw_glock(struct gfs2_glock *gl)
2167 {
2168 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
2169 return;
2170 if (!lockref_get_not_dead(&gl->gl_lockref))
2171 return;
2172 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
2173 gfs2_glock_queue_work(gl, 0);
2174 }
2175
2176 /**
2177 * clear_glock - look at a glock and see if we can free it from glock cache
2178 * @gl: the glock to look at
2179 *
2180 */
2181
clear_glock(struct gfs2_glock * gl)2182 static void clear_glock(struct gfs2_glock *gl)
2183 {
2184 gfs2_glock_remove_from_lru(gl);
2185
2186 spin_lock(&gl->gl_lockref.lock);
2187 if (!__lockref_is_dead(&gl->gl_lockref)) {
2188 gl->gl_lockref.count++;
2189 if (gl->gl_state != LM_ST_UNLOCKED)
2190 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2191 __gfs2_glock_queue_work(gl, 0);
2192 }
2193 spin_unlock(&gl->gl_lockref.lock);
2194 }
2195
2196 /**
2197 * gfs2_glock_thaw - Thaw any frozen glocks
2198 * @sdp: The super block
2199 *
2200 */
2201
gfs2_glock_thaw(struct gfs2_sbd * sdp)2202 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
2203 {
2204 glock_hash_walk(thaw_glock, sdp);
2205 }
2206
dump_glock(struct seq_file * seq,struct gfs2_glock * gl,bool fsid)2207 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2208 {
2209 spin_lock(&gl->gl_lockref.lock);
2210 gfs2_dump_glock(seq, gl, fsid);
2211 spin_unlock(&gl->gl_lockref.lock);
2212 }
2213
dump_glock_func(struct gfs2_glock * gl)2214 static void dump_glock_func(struct gfs2_glock *gl)
2215 {
2216 dump_glock(NULL, gl, true);
2217 }
2218
2219 /**
2220 * gfs2_gl_hash_clear - Empty out the glock hash table
2221 * @sdp: the filesystem
2222 *
2223 * Called when unmounting the filesystem.
2224 */
2225
gfs2_gl_hash_clear(struct gfs2_sbd * sdp)2226 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
2227 {
2228 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
2229 flush_workqueue(glock_workqueue);
2230 glock_hash_walk(clear_glock, sdp);
2231 flush_workqueue(glock_workqueue);
2232 wait_event_timeout(sdp->sd_glock_wait,
2233 atomic_read(&sdp->sd_glock_disposal) == 0,
2234 HZ * 600);
2235 glock_hash_walk(dump_glock_func, sdp);
2236 }
2237
gfs2_glock_finish_truncate(struct gfs2_inode * ip)2238 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
2239 {
2240 struct gfs2_glock *gl = ip->i_gl;
2241 int ret;
2242
2243 ret = gfs2_truncatei_resume(ip);
2244 gfs2_glock_assert_withdraw(gl, ret == 0);
2245
2246 spin_lock(&gl->gl_lockref.lock);
2247 clear_bit(GLF_LOCK, &gl->gl_flags);
2248 run_queue(gl, 1);
2249 spin_unlock(&gl->gl_lockref.lock);
2250 }
2251
state2str(unsigned state)2252 static const char *state2str(unsigned state)
2253 {
2254 switch(state) {
2255 case LM_ST_UNLOCKED:
2256 return "UN";
2257 case LM_ST_SHARED:
2258 return "SH";
2259 case LM_ST_DEFERRED:
2260 return "DF";
2261 case LM_ST_EXCLUSIVE:
2262 return "EX";
2263 }
2264 return "??";
2265 }
2266
hflags2str(char * buf,u16 flags,unsigned long iflags)2267 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
2268 {
2269 char *p = buf;
2270 if (flags & LM_FLAG_TRY)
2271 *p++ = 't';
2272 if (flags & LM_FLAG_TRY_1CB)
2273 *p++ = 'T';
2274 if (flags & LM_FLAG_NOEXP)
2275 *p++ = 'e';
2276 if (flags & LM_FLAG_ANY)
2277 *p++ = 'A';
2278 if (flags & LM_FLAG_PRIORITY)
2279 *p++ = 'p';
2280 if (flags & LM_FLAG_NODE_SCOPE)
2281 *p++ = 'n';
2282 if (flags & GL_ASYNC)
2283 *p++ = 'a';
2284 if (flags & GL_EXACT)
2285 *p++ = 'E';
2286 if (flags & GL_NOCACHE)
2287 *p++ = 'c';
2288 if (test_bit(HIF_HOLDER, &iflags))
2289 *p++ = 'H';
2290 if (test_bit(HIF_WAIT, &iflags))
2291 *p++ = 'W';
2292 if (test_bit(HIF_MAY_DEMOTE, &iflags))
2293 *p++ = 'D';
2294 if (flags & GL_SKIP)
2295 *p++ = 's';
2296 *p = 0;
2297 return buf;
2298 }
2299
2300 /**
2301 * dump_holder - print information about a glock holder
2302 * @seq: the seq_file struct
2303 * @gh: the glock holder
2304 * @fs_id_buf: pointer to file system id (if requested)
2305 *
2306 */
2307
dump_holder(struct seq_file * seq,const struct gfs2_holder * gh,const char * fs_id_buf)2308 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
2309 const char *fs_id_buf)
2310 {
2311 struct task_struct *gh_owner = NULL;
2312 char flags_buf[32];
2313
2314 rcu_read_lock();
2315 if (gh->gh_owner_pid)
2316 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
2317 gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2318 fs_id_buf, state2str(gh->gh_state),
2319 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
2320 gh->gh_error,
2321 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
2322 gh_owner ? gh_owner->comm : "(ended)",
2323 (void *)gh->gh_ip);
2324 rcu_read_unlock();
2325 }
2326
gflags2str(char * buf,const struct gfs2_glock * gl)2327 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
2328 {
2329 const unsigned long *gflags = &gl->gl_flags;
2330 char *p = buf;
2331
2332 if (test_bit(GLF_LOCK, gflags))
2333 *p++ = 'l';
2334 if (test_bit(GLF_DEMOTE, gflags))
2335 *p++ = 'D';
2336 if (test_bit(GLF_PENDING_DEMOTE, gflags))
2337 *p++ = 'd';
2338 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
2339 *p++ = 'p';
2340 if (test_bit(GLF_DIRTY, gflags))
2341 *p++ = 'y';
2342 if (test_bit(GLF_LFLUSH, gflags))
2343 *p++ = 'f';
2344 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
2345 *p++ = 'i';
2346 if (test_bit(GLF_REPLY_PENDING, gflags))
2347 *p++ = 'r';
2348 if (test_bit(GLF_INITIAL, gflags))
2349 *p++ = 'I';
2350 if (test_bit(GLF_FROZEN, gflags))
2351 *p++ = 'F';
2352 if (!list_empty(&gl->gl_holders))
2353 *p++ = 'q';
2354 if (test_bit(GLF_LRU, gflags))
2355 *p++ = 'L';
2356 if (gl->gl_object)
2357 *p++ = 'o';
2358 if (test_bit(GLF_BLOCKING, gflags))
2359 *p++ = 'b';
2360 if (test_bit(GLF_PENDING_DELETE, gflags))
2361 *p++ = 'P';
2362 if (test_bit(GLF_FREEING, gflags))
2363 *p++ = 'x';
2364 if (test_bit(GLF_INSTANTIATE_NEEDED, gflags))
2365 *p++ = 'n';
2366 if (test_bit(GLF_INSTANTIATE_IN_PROG, gflags))
2367 *p++ = 'N';
2368 *p = 0;
2369 return buf;
2370 }
2371
2372 /**
2373 * gfs2_dump_glock - print information about a glock
2374 * @seq: The seq_file struct
2375 * @gl: the glock
2376 * @fsid: If true, also dump the file system id
2377 *
2378 * The file format is as follows:
2379 * One line per object, capital letters are used to indicate objects
2380 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2381 * other objects are indented by a single space and follow the glock to
2382 * which they are related. Fields are indicated by lower case letters
2383 * followed by a colon and the field value, except for strings which are in
2384 * [] so that its possible to see if they are composed of spaces for
2385 * example. The field's are n = number (id of the object), f = flags,
2386 * t = type, s = state, r = refcount, e = error, p = pid.
2387 *
2388 */
2389
gfs2_dump_glock(struct seq_file * seq,struct gfs2_glock * gl,bool fsid)2390 void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2391 {
2392 const struct gfs2_glock_operations *glops = gl->gl_ops;
2393 unsigned long long dtime;
2394 const struct gfs2_holder *gh;
2395 char gflags_buf[32];
2396 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2397 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2398 unsigned long nrpages = 0;
2399
2400 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
2401 struct address_space *mapping = gfs2_glock2aspace(gl);
2402
2403 nrpages = mapping->nrpages;
2404 }
2405 memset(fs_id_buf, 0, sizeof(fs_id_buf));
2406 if (fsid && sdp) /* safety precaution */
2407 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2408 dtime = jiffies - gl->gl_demote_time;
2409 dtime *= 1000000/HZ; /* demote time in uSec */
2410 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
2411 dtime = 0;
2412 gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2413 "v:%d r:%d m:%ld p:%lu\n",
2414 fs_id_buf, state2str(gl->gl_state),
2415 gl->gl_name.ln_type,
2416 (unsigned long long)gl->gl_name.ln_number,
2417 gflags2str(gflags_buf, gl),
2418 state2str(gl->gl_target),
2419 state2str(gl->gl_demote_state), dtime,
2420 atomic_read(&gl->gl_ail_count),
2421 atomic_read(&gl->gl_revokes),
2422 (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages);
2423
2424 list_for_each_entry(gh, &gl->gl_holders, gh_list)
2425 dump_holder(seq, gh, fs_id_buf);
2426
2427 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
2428 glops->go_dump(seq, gl, fs_id_buf);
2429 }
2430
gfs2_glstats_seq_show(struct seq_file * seq,void * iter_ptr)2431 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
2432 {
2433 struct gfs2_glock *gl = iter_ptr;
2434
2435 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2436 gl->gl_name.ln_type,
2437 (unsigned long long)gl->gl_name.ln_number,
2438 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
2439 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
2440 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
2441 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
2442 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
2443 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
2444 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
2445 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
2446 return 0;
2447 }
2448
2449 static const char *gfs2_gltype[] = {
2450 "type",
2451 "reserved",
2452 "nondisk",
2453 "inode",
2454 "rgrp",
2455 "meta",
2456 "iopen",
2457 "flock",
2458 "plock",
2459 "quota",
2460 "journal",
2461 };
2462
2463 static const char *gfs2_stype[] = {
2464 [GFS2_LKS_SRTT] = "srtt",
2465 [GFS2_LKS_SRTTVAR] = "srttvar",
2466 [GFS2_LKS_SRTTB] = "srttb",
2467 [GFS2_LKS_SRTTVARB] = "srttvarb",
2468 [GFS2_LKS_SIRT] = "sirt",
2469 [GFS2_LKS_SIRTVAR] = "sirtvar",
2470 [GFS2_LKS_DCOUNT] = "dlm",
2471 [GFS2_LKS_QCOUNT] = "queue",
2472 };
2473
2474 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2475
gfs2_sbstats_seq_show(struct seq_file * seq,void * iter_ptr)2476 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
2477 {
2478 struct gfs2_sbd *sdp = seq->private;
2479 loff_t pos = *(loff_t *)iter_ptr;
2480 unsigned index = pos >> 3;
2481 unsigned subindex = pos & 0x07;
2482 int i;
2483
2484 if (index == 0 && subindex != 0)
2485 return 0;
2486
2487 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
2488 (index == 0) ? "cpu": gfs2_stype[subindex]);
2489
2490 for_each_possible_cpu(i) {
2491 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
2492
2493 if (index == 0)
2494 seq_printf(seq, " %15u", i);
2495 else
2496 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
2497 lkstats[index - 1].stats[subindex]);
2498 }
2499 seq_putc(seq, '\n');
2500 return 0;
2501 }
2502
gfs2_glock_init(void)2503 int __init gfs2_glock_init(void)
2504 {
2505 int i, ret;
2506
2507 ret = rhashtable_init(&gl_hash_table, &ht_parms);
2508 if (ret < 0)
2509 return ret;
2510
2511 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
2512 WQ_HIGHPRI | WQ_FREEZABLE, 0);
2513 if (!glock_workqueue) {
2514 rhashtable_destroy(&gl_hash_table);
2515 return -ENOMEM;
2516 }
2517 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
2518 WQ_MEM_RECLAIM | WQ_FREEZABLE,
2519 0);
2520 if (!gfs2_delete_workqueue) {
2521 destroy_workqueue(glock_workqueue);
2522 rhashtable_destroy(&gl_hash_table);
2523 return -ENOMEM;
2524 }
2525
2526 ret = register_shrinker(&glock_shrinker);
2527 if (ret) {
2528 destroy_workqueue(gfs2_delete_workqueue);
2529 destroy_workqueue(glock_workqueue);
2530 rhashtable_destroy(&gl_hash_table);
2531 return ret;
2532 }
2533
2534 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2535 init_waitqueue_head(glock_wait_table + i);
2536
2537 return 0;
2538 }
2539
gfs2_glock_exit(void)2540 void gfs2_glock_exit(void)
2541 {
2542 unregister_shrinker(&glock_shrinker);
2543 rhashtable_destroy(&gl_hash_table);
2544 destroy_workqueue(glock_workqueue);
2545 destroy_workqueue(gfs2_delete_workqueue);
2546 }
2547
gfs2_glock_iter_next(struct gfs2_glock_iter * gi,loff_t n)2548 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2549 {
2550 struct gfs2_glock *gl = gi->gl;
2551
2552 if (gl) {
2553 if (n == 0)
2554 return;
2555 if (!lockref_put_not_zero(&gl->gl_lockref))
2556 gfs2_glock_queue_put(gl);
2557 }
2558 for (;;) {
2559 gl = rhashtable_walk_next(&gi->hti);
2560 if (IS_ERR_OR_NULL(gl)) {
2561 if (gl == ERR_PTR(-EAGAIN)) {
2562 n = 1;
2563 continue;
2564 }
2565 gl = NULL;
2566 break;
2567 }
2568 if (gl->gl_name.ln_sbd != gi->sdp)
2569 continue;
2570 if (n <= 1) {
2571 if (!lockref_get_not_dead(&gl->gl_lockref))
2572 continue;
2573 break;
2574 } else {
2575 if (__lockref_is_dead(&gl->gl_lockref))
2576 continue;
2577 n--;
2578 }
2579 }
2580 gi->gl = gl;
2581 }
2582
gfs2_glock_seq_start(struct seq_file * seq,loff_t * pos)2583 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2584 __acquires(RCU)
2585 {
2586 struct gfs2_glock_iter *gi = seq->private;
2587 loff_t n;
2588
2589 /*
2590 * We can either stay where we are, skip to the next hash table
2591 * entry, or start from the beginning.
2592 */
2593 if (*pos < gi->last_pos) {
2594 rhashtable_walk_exit(&gi->hti);
2595 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2596 n = *pos + 1;
2597 } else {
2598 n = *pos - gi->last_pos;
2599 }
2600
2601 rhashtable_walk_start(&gi->hti);
2602
2603 gfs2_glock_iter_next(gi, n);
2604 gi->last_pos = *pos;
2605 return gi->gl;
2606 }
2607
gfs2_glock_seq_next(struct seq_file * seq,void * iter_ptr,loff_t * pos)2608 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2609 loff_t *pos)
2610 {
2611 struct gfs2_glock_iter *gi = seq->private;
2612
2613 (*pos)++;
2614 gi->last_pos = *pos;
2615 gfs2_glock_iter_next(gi, 1);
2616 return gi->gl;
2617 }
2618
gfs2_glock_seq_stop(struct seq_file * seq,void * iter_ptr)2619 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2620 __releases(RCU)
2621 {
2622 struct gfs2_glock_iter *gi = seq->private;
2623
2624 rhashtable_walk_stop(&gi->hti);
2625 }
2626
gfs2_glock_seq_show(struct seq_file * seq,void * iter_ptr)2627 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2628 {
2629 dump_glock(seq, iter_ptr, false);
2630 return 0;
2631 }
2632
gfs2_sbstats_seq_start(struct seq_file * seq,loff_t * pos)2633 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2634 {
2635 preempt_disable();
2636 if (*pos >= GFS2_NR_SBSTATS)
2637 return NULL;
2638 return pos;
2639 }
2640
gfs2_sbstats_seq_next(struct seq_file * seq,void * iter_ptr,loff_t * pos)2641 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2642 loff_t *pos)
2643 {
2644 (*pos)++;
2645 if (*pos >= GFS2_NR_SBSTATS)
2646 return NULL;
2647 return pos;
2648 }
2649
gfs2_sbstats_seq_stop(struct seq_file * seq,void * iter_ptr)2650 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2651 {
2652 preempt_enable();
2653 }
2654
2655 static const struct seq_operations gfs2_glock_seq_ops = {
2656 .start = gfs2_glock_seq_start,
2657 .next = gfs2_glock_seq_next,
2658 .stop = gfs2_glock_seq_stop,
2659 .show = gfs2_glock_seq_show,
2660 };
2661
2662 static const struct seq_operations gfs2_glstats_seq_ops = {
2663 .start = gfs2_glock_seq_start,
2664 .next = gfs2_glock_seq_next,
2665 .stop = gfs2_glock_seq_stop,
2666 .show = gfs2_glstats_seq_show,
2667 };
2668
2669 static const struct seq_operations gfs2_sbstats_sops = {
2670 .start = gfs2_sbstats_seq_start,
2671 .next = gfs2_sbstats_seq_next,
2672 .stop = gfs2_sbstats_seq_stop,
2673 .show = gfs2_sbstats_seq_show,
2674 };
2675
2676 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2677
__gfs2_glocks_open(struct inode * inode,struct file * file,const struct seq_operations * ops)2678 static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2679 const struct seq_operations *ops)
2680 {
2681 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2682 if (ret == 0) {
2683 struct seq_file *seq = file->private_data;
2684 struct gfs2_glock_iter *gi = seq->private;
2685
2686 gi->sdp = inode->i_private;
2687 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2688 if (seq->buf)
2689 seq->size = GFS2_SEQ_GOODSIZE;
2690 /*
2691 * Initially, we are "before" the first hash table entry; the
2692 * first call to rhashtable_walk_next gets us the first entry.
2693 */
2694 gi->last_pos = -1;
2695 gi->gl = NULL;
2696 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2697 }
2698 return ret;
2699 }
2700
gfs2_glocks_open(struct inode * inode,struct file * file)2701 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2702 {
2703 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2704 }
2705
gfs2_glocks_release(struct inode * inode,struct file * file)2706 static int gfs2_glocks_release(struct inode *inode, struct file *file)
2707 {
2708 struct seq_file *seq = file->private_data;
2709 struct gfs2_glock_iter *gi = seq->private;
2710
2711 if (gi->gl)
2712 gfs2_glock_put(gi->gl);
2713 rhashtable_walk_exit(&gi->hti);
2714 return seq_release_private(inode, file);
2715 }
2716
gfs2_glstats_open(struct inode * inode,struct file * file)2717 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2718 {
2719 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2720 }
2721
2722 static const struct file_operations gfs2_glocks_fops = {
2723 .owner = THIS_MODULE,
2724 .open = gfs2_glocks_open,
2725 .read = seq_read,
2726 .llseek = seq_lseek,
2727 .release = gfs2_glocks_release,
2728 };
2729
2730 static const struct file_operations gfs2_glstats_fops = {
2731 .owner = THIS_MODULE,
2732 .open = gfs2_glstats_open,
2733 .read = seq_read,
2734 .llseek = seq_lseek,
2735 .release = gfs2_glocks_release,
2736 };
2737
2738 DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats);
2739
gfs2_create_debugfs_file(struct gfs2_sbd * sdp)2740 void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2741 {
2742 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2743
2744 debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2745 &gfs2_glocks_fops);
2746
2747 debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2748 &gfs2_glstats_fops);
2749
2750 debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2751 &gfs2_sbstats_fops);
2752 }
2753
gfs2_delete_debugfs_file(struct gfs2_sbd * sdp)2754 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2755 {
2756 debugfs_remove_recursive(sdp->debugfs_dir);
2757 sdp->debugfs_dir = NULL;
2758 }
2759
gfs2_register_debugfs(void)2760 void gfs2_register_debugfs(void)
2761 {
2762 gfs2_root = debugfs_create_dir("gfs2", NULL);
2763 }
2764
gfs2_unregister_debugfs(void)2765 void gfs2_unregister_debugfs(void)
2766 {
2767 debugfs_remove(gfs2_root);
2768 gfs2_root = NULL;
2769 }
2770