2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
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.
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
51 #define CREATE_TRACE_POINTS
52 #include "trace_gfs2.h"
54 struct 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 */
61 typedef void (*glock_examiner) (struct gfs2_glock * gl);
63 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
65 static struct dentry *gfs2_root;
66 static struct workqueue_struct *glock_workqueue;
67 struct workqueue_struct *gfs2_delete_workqueue;
68 static LIST_HEAD(lru_list);
69 static atomic_t lru_count = ATOMIC_INIT(0);
70 static DEFINE_SPINLOCK(lru_lock);
72 #define GFS2_GL_HASH_SHIFT 15
73 #define GFS2_GL_HASH_SIZE BIT(GFS2_GL_HASH_SHIFT)
75 static 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),
82 static struct rhashtable gl_hash_table;
84 #define GLOCK_WAIT_TABLE_BITS 12
85 #define GLOCK_WAIT_TABLE_SIZE (1 << GLOCK_WAIT_TABLE_BITS)
86 static wait_queue_head_t glock_wait_table[GLOCK_WAIT_TABLE_SIZE] __cacheline_aligned;
88 struct wait_glock_queue {
89 struct lm_lockname *name;
90 wait_queue_entry_t wait;
93 static int glock_wake_function(wait_queue_entry_t *wait, unsigned int mode,
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;
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)
105 return autoremove_wake_function(wait, mode, sync, key);
108 static wait_queue_head_t *glock_waitqueue(struct lm_lockname *name)
110 u32 hash = jhash2((u32 *)name, sizeof(*name) / 4, 0);
112 return glock_wait_table + hash_32(hash, GLOCK_WAIT_TABLE_BITS);
116 * wake_up_glock - Wake up waiters on a glock
119 static void wake_up_glock(struct gfs2_glock *gl)
121 wait_queue_head_t *wq = glock_waitqueue(&gl->gl_name);
123 if (waitqueue_active(wq))
124 __wake_up(wq, TASK_NORMAL, 1, &gl->gl_name);
127 static void gfs2_glock_dealloc(struct rcu_head *rcu)
129 struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
131 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
132 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
134 kfree(gl->gl_lksb.sb_lvbptr);
135 kmem_cache_free(gfs2_glock_cachep, gl);
139 void gfs2_glock_free(struct gfs2_glock *gl)
141 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
143 rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
146 call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
147 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
148 wake_up(&sdp->sd_glock_wait);
152 * gfs2_glock_hold() - increment reference count on glock
153 * @gl: The glock to hold
157 void gfs2_glock_hold(struct gfs2_glock *gl)
159 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
160 lockref_get(&gl->gl_lockref);
164 * demote_ok - Check to see if it's ok to unlock a glock
167 * Returns: 1 if it's ok
170 static int demote_ok(const struct gfs2_glock *gl)
172 const struct gfs2_glock_operations *glops = gl->gl_ops;
174 if (gl->gl_state == LM_ST_UNLOCKED)
176 if (!list_empty(&gl->gl_holders))
178 if (glops->go_demote_ok)
179 return glops->go_demote_ok(gl);
184 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
186 spin_lock(&lru_lock);
188 if (!list_empty(&gl->gl_lru))
189 list_del_init(&gl->gl_lru);
191 atomic_inc(&lru_count);
193 list_add_tail(&gl->gl_lru, &lru_list);
194 set_bit(GLF_LRU, &gl->gl_flags);
195 spin_unlock(&lru_lock);
198 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
200 if (!(gl->gl_ops->go_flags & GLOF_LRU))
203 spin_lock(&lru_lock);
204 if (!list_empty(&gl->gl_lru)) {
205 list_del_init(&gl->gl_lru);
206 atomic_dec(&lru_count);
207 clear_bit(GLF_LRU, &gl->gl_flags);
209 spin_unlock(&lru_lock);
213 * Enqueue the glock on the work queue. Passes one glock reference on to the
216 static void __gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
217 if (!queue_delayed_work(glock_workqueue, &gl->gl_work, delay)) {
219 * We are holding the lockref spinlock, and the work was still
220 * queued above. The queued work (glock_work_func) takes that
221 * spinlock before dropping its glock reference(s), so it
222 * cannot have dropped them in the meantime.
224 GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2);
225 gl->gl_lockref.count--;
229 static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
230 spin_lock(&gl->gl_lockref.lock);
231 __gfs2_glock_queue_work(gl, delay);
232 spin_unlock(&gl->gl_lockref.lock);
235 static void __gfs2_glock_put(struct gfs2_glock *gl)
237 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
238 struct address_space *mapping = gfs2_glock2aspace(gl);
240 lockref_mark_dead(&gl->gl_lockref);
242 gfs2_glock_remove_from_lru(gl);
243 spin_unlock(&gl->gl_lockref.lock);
244 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
245 GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
246 trace_gfs2_glock_put(gl);
247 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
251 * Cause the glock to be put in work queue context.
253 void gfs2_glock_queue_put(struct gfs2_glock *gl)
255 gfs2_glock_queue_work(gl, 0);
259 * gfs2_glock_put() - Decrement reference count on glock
260 * @gl: The glock to put
264 void gfs2_glock_put(struct gfs2_glock *gl)
266 if (lockref_put_or_lock(&gl->gl_lockref))
269 __gfs2_glock_put(gl);
273 * may_grant - check if its ok to grant a new lock
275 * @gh: The lock request which we wish to grant
277 * Returns: true if its ok to grant the lock
280 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
282 const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
283 if ((gh->gh_state == LM_ST_EXCLUSIVE ||
284 gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
286 if (gl->gl_state == gh->gh_state)
288 if (gh->gh_flags & GL_EXACT)
290 if (gl->gl_state == LM_ST_EXCLUSIVE) {
291 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
293 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
296 if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
301 static void gfs2_holder_wake(struct gfs2_holder *gh)
303 clear_bit(HIF_WAIT, &gh->gh_iflags);
304 smp_mb__after_atomic();
305 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
309 * do_error - Something unexpected has happened during a lock request
313 static void do_error(struct gfs2_glock *gl, const int ret)
315 struct gfs2_holder *gh, *tmp;
317 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
318 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
320 if (ret & LM_OUT_ERROR)
322 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
323 gh->gh_error = GLR_TRYFAILED;
326 list_del_init(&gh->gh_list);
327 trace_gfs2_glock_queue(gh, 0);
328 gfs2_holder_wake(gh);
333 * do_promote - promote as many requests as possible on the current queue
336 * Returns: 1 if there is a blocked holder at the head of the list, or 2
337 * if a type specific operation is underway.
340 static int do_promote(struct gfs2_glock *gl)
341 __releases(&gl->gl_lockref.lock)
342 __acquires(&gl->gl_lockref.lock)
344 const struct gfs2_glock_operations *glops = gl->gl_ops;
345 struct gfs2_holder *gh, *tmp;
349 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
350 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
352 if (may_grant(gl, gh)) {
353 if (gh->gh_list.prev == &gl->gl_holders &&
355 spin_unlock(&gl->gl_lockref.lock);
356 /* FIXME: eliminate this eventually */
357 ret = glops->go_lock(gh);
358 spin_lock(&gl->gl_lockref.lock);
363 list_del_init(&gh->gh_list);
364 trace_gfs2_glock_queue(gh, 0);
365 gfs2_holder_wake(gh);
368 set_bit(HIF_HOLDER, &gh->gh_iflags);
369 trace_gfs2_promote(gh, 1);
370 gfs2_holder_wake(gh);
373 set_bit(HIF_HOLDER, &gh->gh_iflags);
374 trace_gfs2_promote(gh, 0);
375 gfs2_holder_wake(gh);
378 if (gh->gh_list.prev == &gl->gl_holders)
387 * find_first_waiter - find the first gh that's waiting for the glock
391 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
393 struct gfs2_holder *gh;
395 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
396 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
403 * state_change - record that the glock is now in a different state
405 * @new_state the new state
409 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
413 held1 = (gl->gl_state != LM_ST_UNLOCKED);
414 held2 = (new_state != LM_ST_UNLOCKED);
416 if (held1 != held2) {
417 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
419 gl->gl_lockref.count++;
421 gl->gl_lockref.count--;
423 if (held1 && held2 && list_empty(&gl->gl_holders))
424 clear_bit(GLF_QUEUED, &gl->gl_flags);
426 if (new_state != gl->gl_target)
427 /* shorten our minimum hold time */
428 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
430 gl->gl_state = new_state;
431 gl->gl_tchange = jiffies;
434 static void gfs2_demote_wake(struct gfs2_glock *gl)
436 gl->gl_demote_state = LM_ST_EXCLUSIVE;
437 clear_bit(GLF_DEMOTE, &gl->gl_flags);
438 smp_mb__after_atomic();
439 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
443 * finish_xmote - The DLM has replied to one of our lock requests
445 * @ret: The status from the DLM
449 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
451 const struct gfs2_glock_operations *glops = gl->gl_ops;
452 struct gfs2_holder *gh;
453 unsigned state = ret & LM_OUT_ST_MASK;
456 spin_lock(&gl->gl_lockref.lock);
457 trace_gfs2_glock_state_change(gl, state);
458 state_change(gl, state);
459 gh = find_first_waiter(gl);
461 /* Demote to UN request arrived during demote to SH or DF */
462 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
463 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
464 gl->gl_target = LM_ST_UNLOCKED;
466 /* Check for state != intended state */
467 if (unlikely(state != gl->gl_target)) {
468 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
469 /* move to back of queue and try next entry */
470 if (ret & LM_OUT_CANCELED) {
471 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
472 list_move_tail(&gh->gh_list, &gl->gl_holders);
473 gh = find_first_waiter(gl);
474 gl->gl_target = gh->gh_state;
477 /* Some error or failed "try lock" - report it */
478 if ((ret & LM_OUT_ERROR) ||
479 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
480 gl->gl_target = gl->gl_state;
486 /* Unlocked due to conversion deadlock, try again */
489 do_xmote(gl, gh, gl->gl_target);
491 /* Conversion fails, unlock and try again */
494 do_xmote(gl, gh, LM_ST_UNLOCKED);
496 default: /* Everything else */
497 pr_err("wanted %u got %u\n", gl->gl_target, state);
500 spin_unlock(&gl->gl_lockref.lock);
504 /* Fast path - we got what we asked for */
505 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
506 gfs2_demote_wake(gl);
507 if (state != LM_ST_UNLOCKED) {
508 if (glops->go_xmote_bh) {
509 spin_unlock(&gl->gl_lockref.lock);
510 rv = glops->go_xmote_bh(gl, gh);
511 spin_lock(&gl->gl_lockref.lock);
522 clear_bit(GLF_LOCK, &gl->gl_flags);
524 spin_unlock(&gl->gl_lockref.lock);
528 * do_xmote - Calls the DLM to change the state of a lock
529 * @gl: The lock state
530 * @gh: The holder (only for promotes)
531 * @target: The target lock state
535 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
536 __releases(&gl->gl_lockref.lock)
537 __acquires(&gl->gl_lockref.lock)
539 const struct gfs2_glock_operations *glops = gl->gl_ops;
540 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
541 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
544 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
545 target != LM_ST_UNLOCKED)
547 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
549 GLOCK_BUG_ON(gl, gl->gl_state == target);
550 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
551 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
553 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
554 do_error(gl, 0); /* Fail queued try locks */
557 set_bit(GLF_BLOCKING, &gl->gl_flags);
558 if ((gl->gl_req == LM_ST_UNLOCKED) ||
559 (gl->gl_state == LM_ST_EXCLUSIVE) ||
560 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
561 clear_bit(GLF_BLOCKING, &gl->gl_flags);
562 spin_unlock(&gl->gl_lockref.lock);
565 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
566 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
567 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
570 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
572 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
573 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
574 target == LM_ST_UNLOCKED &&
575 test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
576 finish_xmote(gl, target);
577 gfs2_glock_queue_work(gl, 0);
580 pr_err("lm_lock ret %d\n", ret);
581 GLOCK_BUG_ON(gl, !test_bit(SDF_SHUTDOWN,
584 } else { /* lock_nolock */
585 finish_xmote(gl, target);
586 gfs2_glock_queue_work(gl, 0);
589 spin_lock(&gl->gl_lockref.lock);
593 * find_first_holder - find the first "holder" gh
597 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
599 struct gfs2_holder *gh;
601 if (!list_empty(&gl->gl_holders)) {
602 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
603 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
610 * run_queue - do all outstanding tasks related to a glock
611 * @gl: The glock in question
612 * @nonblock: True if we must not block in run_queue
616 static void run_queue(struct gfs2_glock *gl, const int nonblock)
617 __releases(&gl->gl_lockref.lock)
618 __acquires(&gl->gl_lockref.lock)
620 struct gfs2_holder *gh = NULL;
623 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
626 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
628 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
629 gl->gl_demote_state != gl->gl_state) {
630 if (find_first_holder(gl))
634 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
635 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
636 gl->gl_target = gl->gl_demote_state;
638 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
639 gfs2_demote_wake(gl);
640 ret = do_promote(gl);
645 gh = find_first_waiter(gl);
646 gl->gl_target = gh->gh_state;
647 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
648 do_error(gl, 0); /* Fail queued try locks */
650 do_xmote(gl, gh, gl->gl_target);
655 clear_bit(GLF_LOCK, &gl->gl_flags);
656 smp_mb__after_atomic();
657 gl->gl_lockref.count++;
658 __gfs2_glock_queue_work(gl, 0);
662 clear_bit(GLF_LOCK, &gl->gl_flags);
663 smp_mb__after_atomic();
667 static void delete_work_func(struct work_struct *work)
669 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
670 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
672 u64 no_addr = gl->gl_name.ln_number;
674 /* If someone's using this glock to create a new dinode, the block must
675 have been freed by another node, then re-used, in which case our
676 iopen callback is too late after the fact. Ignore it. */
677 if (test_bit(GLF_INODE_CREATING, &gl->gl_flags))
680 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
681 if (inode && !IS_ERR(inode)) {
682 d_prune_aliases(inode);
689 static void glock_work_func(struct work_struct *work)
691 unsigned long delay = 0;
692 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
693 unsigned int drop_refs = 1;
695 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
696 finish_xmote(gl, gl->gl_reply);
699 spin_lock(&gl->gl_lockref.lock);
700 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
701 gl->gl_state != LM_ST_UNLOCKED &&
702 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
703 unsigned long holdtime, now = jiffies;
705 holdtime = gl->gl_tchange + gl->gl_hold_time;
706 if (time_before(now, holdtime))
707 delay = holdtime - now;
710 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
711 set_bit(GLF_DEMOTE, &gl->gl_flags);
716 /* Keep one glock reference for the work we requeue. */
718 if (gl->gl_name.ln_type != LM_TYPE_INODE)
720 __gfs2_glock_queue_work(gl, delay);
724 * Drop the remaining glock references manually here. (Mind that
725 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
728 gl->gl_lockref.count -= drop_refs;
729 if (!gl->gl_lockref.count) {
730 __gfs2_glock_put(gl);
733 spin_unlock(&gl->gl_lockref.lock);
736 static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
737 struct gfs2_glock *new)
739 struct wait_glock_queue wait;
740 wait_queue_head_t *wq = glock_waitqueue(name);
741 struct gfs2_glock *gl;
744 init_wait(&wait.wait);
745 wait.wait.func = glock_wake_function;
748 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
751 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
752 &new->gl_node, ht_parms);
756 gl = rhashtable_lookup_fast(&gl_hash_table,
759 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
766 finish_wait(wq, &wait.wait);
771 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
772 * @sdp: The GFS2 superblock
773 * @number: the lock number
774 * @glops: The glock_operations to use
775 * @create: If 0, don't create the glock if it doesn't exist
776 * @glp: the glock is returned here
778 * This does not lock a glock, just finds/creates structures for one.
783 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
784 const struct gfs2_glock_operations *glops, int create,
785 struct gfs2_glock **glp)
787 struct super_block *s = sdp->sd_vfs;
788 struct lm_lockname name = { .ln_number = number,
789 .ln_type = glops->go_type,
791 struct gfs2_glock *gl, *tmp;
792 struct address_space *mapping;
793 struct kmem_cache *cachep;
796 gl = find_insert_glock(&name, NULL);
804 if (glops->go_flags & GLOF_ASPACE)
805 cachep = gfs2_glock_aspace_cachep;
807 cachep = gfs2_glock_cachep;
808 gl = kmem_cache_alloc(cachep, GFP_NOFS);
812 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
814 if (glops->go_flags & GLOF_LVB) {
815 gl->gl_lksb.sb_lvbptr = kzalloc(GFS2_MIN_LVB_SIZE, GFP_NOFS);
816 if (!gl->gl_lksb.sb_lvbptr) {
817 kmem_cache_free(cachep, gl);
822 atomic_inc(&sdp->sd_glock_disposal);
823 gl->gl_node.next = NULL;
826 gl->gl_lockref.count = 1;
827 gl->gl_state = LM_ST_UNLOCKED;
828 gl->gl_target = LM_ST_UNLOCKED;
829 gl->gl_demote_state = LM_ST_EXCLUSIVE;
833 /* We use the global stats to estimate the initial per-glock stats */
834 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
836 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
837 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
838 gl->gl_tchange = jiffies;
839 gl->gl_object = NULL;
840 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
841 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
842 INIT_WORK(&gl->gl_delete, delete_work_func);
844 mapping = gfs2_glock2aspace(gl);
846 mapping->a_ops = &gfs2_meta_aops;
847 mapping->host = s->s_bdev->bd_inode;
849 mapping_set_gfp_mask(mapping, GFP_NOFS);
850 mapping->private_data = NULL;
851 mapping->writeback_index = 0;
854 tmp = find_insert_glock(&name, gl);
866 kfree(gl->gl_lksb.sb_lvbptr);
867 kmem_cache_free(cachep, gl);
868 atomic_dec(&sdp->sd_glock_disposal);
875 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
877 * @state: the state we're requesting
878 * @flags: the modifier flags
879 * @gh: the holder structure
883 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
884 struct gfs2_holder *gh)
886 INIT_LIST_HEAD(&gh->gh_list);
888 gh->gh_ip = _RET_IP_;
889 gh->gh_owner_pid = get_pid(task_pid(current));
890 gh->gh_state = state;
891 gh->gh_flags = flags;
898 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
899 * @state: the state we're requesting
900 * @flags: the modifier flags
901 * @gh: the holder structure
903 * Don't mess with the glock.
907 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
909 gh->gh_state = state;
910 gh->gh_flags = flags;
912 gh->gh_ip = _RET_IP_;
913 put_pid(gh->gh_owner_pid);
914 gh->gh_owner_pid = get_pid(task_pid(current));
918 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
919 * @gh: the holder structure
923 void gfs2_holder_uninit(struct gfs2_holder *gh)
925 put_pid(gh->gh_owner_pid);
926 gfs2_glock_put(gh->gh_gl);
927 gfs2_holder_mark_uninitialized(gh);
932 * gfs2_glock_wait - wait on a glock acquisition
933 * @gh: the glock holder
935 * Returns: 0 on success
938 int gfs2_glock_wait(struct gfs2_holder *gh)
940 unsigned long time1 = jiffies;
943 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
944 if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
945 /* Lengthen the minimum hold time. */
946 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
953 * handle_callback - process a demote request
955 * @state: the state the caller wants us to change to
957 * There are only two requests that we are going to see in actual
958 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
961 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
962 unsigned long delay, bool remote)
964 int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
966 set_bit(bit, &gl->gl_flags);
967 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
968 gl->gl_demote_state = state;
969 gl->gl_demote_time = jiffies;
970 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
971 gl->gl_demote_state != state) {
972 gl->gl_demote_state = LM_ST_UNLOCKED;
974 if (gl->gl_ops->go_callback)
975 gl->gl_ops->go_callback(gl, remote);
976 trace_gfs2_demote_rq(gl, remote);
979 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
981 struct va_format vaf;
987 seq_vprintf(seq, fmt, args);
999 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1000 * @gh: the holder structure to add
1002 * Eventually we should move the recursive locking trap to a
1003 * debugging option or something like that. This is the fast
1004 * path and needs to have the minimum number of distractions.
1008 static inline void add_to_queue(struct gfs2_holder *gh)
1009 __releases(&gl->gl_lockref.lock)
1010 __acquires(&gl->gl_lockref.lock)
1012 struct gfs2_glock *gl = gh->gh_gl;
1013 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1014 struct list_head *insert_pt = NULL;
1015 struct gfs2_holder *gh2;
1018 BUG_ON(gh->gh_owner_pid == NULL);
1019 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1022 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1023 if (test_bit(GLF_LOCK, &gl->gl_flags))
1024 try_futile = !may_grant(gl, gh);
1025 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1029 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1030 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
1031 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
1032 goto trap_recursive;
1034 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1036 gh->gh_error = GLR_TRYFAILED;
1037 gfs2_holder_wake(gh);
1040 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1042 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1043 insert_pt = &gh2->gh_list;
1045 set_bit(GLF_QUEUED, &gl->gl_flags);
1046 trace_gfs2_glock_queue(gh, 1);
1047 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1048 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1049 if (likely(insert_pt == NULL)) {
1050 list_add_tail(&gh->gh_list, &gl->gl_holders);
1051 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1055 list_add_tail(&gh->gh_list, insert_pt);
1057 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
1058 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1059 spin_unlock(&gl->gl_lockref.lock);
1060 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1061 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1062 spin_lock(&gl->gl_lockref.lock);
1067 pr_err("original: %pSR\n", (void *)gh2->gh_ip);
1068 pr_err("pid: %d\n", pid_nr(gh2->gh_owner_pid));
1069 pr_err("lock type: %d req lock state : %d\n",
1070 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1071 pr_err("new: %pSR\n", (void *)gh->gh_ip);
1072 pr_err("pid: %d\n", pid_nr(gh->gh_owner_pid));
1073 pr_err("lock type: %d req lock state : %d\n",
1074 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1075 gfs2_dump_glock(NULL, gl);
1080 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1081 * @gh: the holder structure
1083 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1085 * Returns: 0, GLR_TRYFAILED, or errno on failure
1088 int gfs2_glock_nq(struct gfs2_holder *gh)
1090 struct gfs2_glock *gl = gh->gh_gl;
1091 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1094 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1097 if (test_bit(GLF_LRU, &gl->gl_flags))
1098 gfs2_glock_remove_from_lru(gl);
1100 spin_lock(&gl->gl_lockref.lock);
1102 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1103 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1104 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1105 gl->gl_lockref.count++;
1106 __gfs2_glock_queue_work(gl, 0);
1109 spin_unlock(&gl->gl_lockref.lock);
1111 if (!(gh->gh_flags & GL_ASYNC))
1112 error = gfs2_glock_wait(gh);
1118 * gfs2_glock_poll - poll to see if an async request has been completed
1121 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1124 int gfs2_glock_poll(struct gfs2_holder *gh)
1126 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1130 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1131 * @gh: the glock holder
1135 void gfs2_glock_dq(struct gfs2_holder *gh)
1137 struct gfs2_glock *gl = gh->gh_gl;
1138 const struct gfs2_glock_operations *glops = gl->gl_ops;
1142 spin_lock(&gl->gl_lockref.lock);
1143 if (gh->gh_flags & GL_NOCACHE)
1144 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1146 list_del_init(&gh->gh_list);
1147 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1148 if (find_first_holder(gl) == NULL) {
1149 if (glops->go_unlock) {
1150 GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1151 spin_unlock(&gl->gl_lockref.lock);
1152 glops->go_unlock(gh);
1153 spin_lock(&gl->gl_lockref.lock);
1154 clear_bit(GLF_LOCK, &gl->gl_flags);
1156 if (list_empty(&gl->gl_holders) &&
1157 !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1158 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1161 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl) &&
1162 (glops->go_flags & GLOF_LRU))
1163 gfs2_glock_add_to_lru(gl);
1165 trace_gfs2_glock_queue(gh, 0);
1166 if (unlikely(!fast_path)) {
1167 gl->gl_lockref.count++;
1168 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1169 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1170 gl->gl_name.ln_type == LM_TYPE_INODE)
1171 delay = gl->gl_hold_time;
1172 __gfs2_glock_queue_work(gl, delay);
1174 spin_unlock(&gl->gl_lockref.lock);
1177 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1179 struct gfs2_glock *gl = gh->gh_gl;
1182 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1186 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1187 * @gh: the holder structure
1191 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1194 gfs2_holder_uninit(gh);
1198 * gfs2_glock_nq_num - acquire a glock based on lock number
1199 * @sdp: the filesystem
1200 * @number: the lock number
1201 * @glops: the glock operations for the type of glock
1202 * @state: the state to acquire the glock in
1203 * @flags: modifier flags for the acquisition
1204 * @gh: the struct gfs2_holder
1209 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1210 const struct gfs2_glock_operations *glops,
1211 unsigned int state, u16 flags, struct gfs2_holder *gh)
1213 struct gfs2_glock *gl;
1216 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1218 error = gfs2_glock_nq_init(gl, state, flags, gh);
1226 * glock_compare - Compare two struct gfs2_glock structures for sorting
1227 * @arg_a: the first structure
1228 * @arg_b: the second structure
1232 static int glock_compare(const void *arg_a, const void *arg_b)
1234 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1235 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1236 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1237 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1239 if (a->ln_number > b->ln_number)
1241 if (a->ln_number < b->ln_number)
1243 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1248 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1249 * @num_gh: the number of structures
1250 * @ghs: an array of struct gfs2_holder structures
1252 * Returns: 0 on success (all glocks acquired),
1253 * errno on failure (no glocks acquired)
1256 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1257 struct gfs2_holder **p)
1262 for (x = 0; x < num_gh; x++)
1265 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1267 for (x = 0; x < num_gh; x++) {
1268 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1270 error = gfs2_glock_nq(p[x]);
1273 gfs2_glock_dq(p[x]);
1282 * gfs2_glock_nq_m - acquire multiple glocks
1283 * @num_gh: the number of structures
1284 * @ghs: an array of struct gfs2_holder structures
1287 * Returns: 0 on success (all glocks acquired),
1288 * errno on failure (no glocks acquired)
1291 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1293 struct gfs2_holder *tmp[4];
1294 struct gfs2_holder **pph = tmp;
1301 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1302 return gfs2_glock_nq(ghs);
1306 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1311 error = nq_m_sync(num_gh, ghs, pph);
1320 * gfs2_glock_dq_m - release multiple glocks
1321 * @num_gh: the number of structures
1322 * @ghs: an array of struct gfs2_holder structures
1326 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1329 gfs2_glock_dq(&ghs[num_gh]);
1332 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1334 unsigned long delay = 0;
1335 unsigned long holdtime;
1336 unsigned long now = jiffies;
1338 gfs2_glock_hold(gl);
1339 holdtime = gl->gl_tchange + gl->gl_hold_time;
1340 if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1341 gl->gl_name.ln_type == LM_TYPE_INODE) {
1342 if (time_before(now, holdtime))
1343 delay = holdtime - now;
1344 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1345 delay = gl->gl_hold_time;
1348 spin_lock(&gl->gl_lockref.lock);
1349 handle_callback(gl, state, delay, true);
1350 __gfs2_glock_queue_work(gl, delay);
1351 spin_unlock(&gl->gl_lockref.lock);
1355 * gfs2_should_freeze - Figure out if glock should be frozen
1356 * @gl: The glock in question
1358 * Glocks are not frozen if (a) the result of the dlm operation is
1359 * an error, (b) the locking operation was an unlock operation or
1360 * (c) if there is a "noexp" flagged request anywhere in the queue
1362 * Returns: 1 if freezing should occur, 0 otherwise
1365 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1367 const struct gfs2_holder *gh;
1369 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1371 if (gl->gl_target == LM_ST_UNLOCKED)
1374 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1375 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1377 if (LM_FLAG_NOEXP & gh->gh_flags)
1385 * gfs2_glock_complete - Callback used by locking
1386 * @gl: Pointer to the glock
1387 * @ret: The return value from the dlm
1389 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1390 * to use a bitfield shared with other glock state fields.
1393 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1395 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1397 spin_lock(&gl->gl_lockref.lock);
1400 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1401 if (gfs2_should_freeze(gl)) {
1402 set_bit(GLF_FROZEN, &gl->gl_flags);
1403 spin_unlock(&gl->gl_lockref.lock);
1408 gl->gl_lockref.count++;
1409 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1410 __gfs2_glock_queue_work(gl, 0);
1411 spin_unlock(&gl->gl_lockref.lock);
1414 static int glock_cmp(void *priv, struct list_head *a, struct list_head *b)
1416 struct gfs2_glock *gla, *glb;
1418 gla = list_entry(a, struct gfs2_glock, gl_lru);
1419 glb = list_entry(b, struct gfs2_glock, gl_lru);
1421 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1423 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1430 * gfs2_dispose_glock_lru - Demote a list of glocks
1431 * @list: The list to dispose of
1433 * Disposing of glocks may involve disk accesses, so that here we sort
1434 * the glocks by number (i.e. disk location of the inodes) so that if
1435 * there are any such accesses, they'll be sent in order (mostly).
1437 * Must be called under the lru_lock, but may drop and retake this
1438 * lock. While the lru_lock is dropped, entries may vanish from the
1439 * list, but no new entries will appear on the list (since it is
1443 static void gfs2_dispose_glock_lru(struct list_head *list)
1444 __releases(&lru_lock)
1445 __acquires(&lru_lock)
1447 struct gfs2_glock *gl;
1449 list_sort(NULL, list, glock_cmp);
1451 while(!list_empty(list)) {
1452 gl = list_entry(list->next, struct gfs2_glock, gl_lru);
1453 list_del_init(&gl->gl_lru);
1454 if (!spin_trylock(&gl->gl_lockref.lock)) {
1456 list_add(&gl->gl_lru, &lru_list);
1457 atomic_inc(&lru_count);
1460 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1461 spin_unlock(&gl->gl_lockref.lock);
1462 goto add_back_to_lru;
1464 clear_bit(GLF_LRU, &gl->gl_flags);
1465 gl->gl_lockref.count++;
1467 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1468 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1469 __gfs2_glock_queue_work(gl, 0);
1470 spin_unlock(&gl->gl_lockref.lock);
1471 cond_resched_lock(&lru_lock);
1476 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1477 * @nr: The number of entries to scan
1479 * This function selects the entries on the LRU which are able to
1480 * be demoted, and then kicks off the process by calling
1481 * gfs2_dispose_glock_lru() above.
1484 static long gfs2_scan_glock_lru(int nr)
1486 struct gfs2_glock *gl;
1491 spin_lock(&lru_lock);
1492 while ((nr-- >= 0) && !list_empty(&lru_list)) {
1493 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1495 /* Test for being demotable */
1496 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1497 list_move(&gl->gl_lru, &dispose);
1498 atomic_dec(&lru_count);
1503 list_move(&gl->gl_lru, &skipped);
1505 list_splice(&skipped, &lru_list);
1506 if (!list_empty(&dispose))
1507 gfs2_dispose_glock_lru(&dispose);
1508 spin_unlock(&lru_lock);
1513 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1514 struct shrink_control *sc)
1516 if (!(sc->gfp_mask & __GFP_FS))
1518 return gfs2_scan_glock_lru(sc->nr_to_scan);
1521 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1522 struct shrink_control *sc)
1524 return vfs_pressure_ratio(atomic_read(&lru_count));
1527 static struct shrinker glock_shrinker = {
1528 .seeks = DEFAULT_SEEKS,
1529 .count_objects = gfs2_glock_shrink_count,
1530 .scan_objects = gfs2_glock_shrink_scan,
1534 * examine_bucket - Call a function for glock in a hash bucket
1535 * @examiner: the function
1536 * @sdp: the filesystem
1537 * @bucket: the bucket
1539 * Note that the function can be called multiple times on the same
1540 * object. So the user must ensure that the function can cope with
1544 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1546 struct gfs2_glock *gl;
1547 struct rhashtable_iter iter;
1549 rhashtable_walk_enter(&gl_hash_table, &iter);
1552 gl = ERR_PTR(rhashtable_walk_start(&iter));
1556 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl))
1557 if (gl->gl_name.ln_sbd == sdp &&
1558 lockref_get_not_dead(&gl->gl_lockref))
1562 rhashtable_walk_stop(&iter);
1563 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
1565 rhashtable_walk_exit(&iter);
1569 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1570 * @gl: The glock to thaw
1574 static void thaw_glock(struct gfs2_glock *gl)
1576 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags)) {
1580 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1581 gfs2_glock_queue_work(gl, 0);
1585 * clear_glock - look at a glock and see if we can free it from glock cache
1586 * @gl: the glock to look at
1590 static void clear_glock(struct gfs2_glock *gl)
1592 gfs2_glock_remove_from_lru(gl);
1594 spin_lock(&gl->gl_lockref.lock);
1595 if (gl->gl_state != LM_ST_UNLOCKED)
1596 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1597 __gfs2_glock_queue_work(gl, 0);
1598 spin_unlock(&gl->gl_lockref.lock);
1602 * gfs2_glock_thaw - Thaw any frozen glocks
1603 * @sdp: The super block
1607 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1609 glock_hash_walk(thaw_glock, sdp);
1612 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1614 spin_lock(&gl->gl_lockref.lock);
1615 gfs2_dump_glock(seq, gl);
1616 spin_unlock(&gl->gl_lockref.lock);
1619 static void dump_glock_func(struct gfs2_glock *gl)
1621 dump_glock(NULL, gl);
1625 * gfs2_gl_hash_clear - Empty out the glock hash table
1626 * @sdp: the filesystem
1627 * @wait: wait until it's all gone
1629 * Called when unmounting the filesystem.
1632 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1634 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1635 flush_workqueue(glock_workqueue);
1636 glock_hash_walk(clear_glock, sdp);
1637 flush_workqueue(glock_workqueue);
1638 wait_event_timeout(sdp->sd_glock_wait,
1639 atomic_read(&sdp->sd_glock_disposal) == 0,
1641 glock_hash_walk(dump_glock_func, sdp);
1644 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1646 struct gfs2_glock *gl = ip->i_gl;
1649 ret = gfs2_truncatei_resume(ip);
1650 gfs2_assert_withdraw(gl->gl_name.ln_sbd, ret == 0);
1652 spin_lock(&gl->gl_lockref.lock);
1653 clear_bit(GLF_LOCK, &gl->gl_flags);
1655 spin_unlock(&gl->gl_lockref.lock);
1658 static const char *state2str(unsigned state)
1661 case LM_ST_UNLOCKED:
1665 case LM_ST_DEFERRED:
1667 case LM_ST_EXCLUSIVE:
1673 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
1676 if (flags & LM_FLAG_TRY)
1678 if (flags & LM_FLAG_TRY_1CB)
1680 if (flags & LM_FLAG_NOEXP)
1682 if (flags & LM_FLAG_ANY)
1684 if (flags & LM_FLAG_PRIORITY)
1686 if (flags & GL_ASYNC)
1688 if (flags & GL_EXACT)
1690 if (flags & GL_NOCACHE)
1692 if (test_bit(HIF_HOLDER, &iflags))
1694 if (test_bit(HIF_WAIT, &iflags))
1696 if (test_bit(HIF_FIRST, &iflags))
1703 * dump_holder - print information about a glock holder
1704 * @seq: the seq_file struct
1705 * @gh: the glock holder
1709 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1711 struct task_struct *gh_owner = NULL;
1715 if (gh->gh_owner_pid)
1716 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1717 gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1718 state2str(gh->gh_state),
1719 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1721 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1722 gh_owner ? gh_owner->comm : "(ended)",
1727 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1729 const unsigned long *gflags = &gl->gl_flags;
1732 if (test_bit(GLF_LOCK, gflags))
1734 if (test_bit(GLF_DEMOTE, gflags))
1736 if (test_bit(GLF_PENDING_DEMOTE, gflags))
1738 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1740 if (test_bit(GLF_DIRTY, gflags))
1742 if (test_bit(GLF_LFLUSH, gflags))
1744 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1746 if (test_bit(GLF_REPLY_PENDING, gflags))
1748 if (test_bit(GLF_INITIAL, gflags))
1750 if (test_bit(GLF_FROZEN, gflags))
1752 if (test_bit(GLF_QUEUED, gflags))
1754 if (test_bit(GLF_LRU, gflags))
1758 if (test_bit(GLF_BLOCKING, gflags))
1765 * gfs2_dump_glock - print information about a glock
1766 * @seq: The seq_file struct
1769 * The file format is as follows:
1770 * One line per object, capital letters are used to indicate objects
1771 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1772 * other objects are indented by a single space and follow the glock to
1773 * which they are related. Fields are indicated by lower case letters
1774 * followed by a colon and the field value, except for strings which are in
1775 * [] so that its possible to see if they are composed of spaces for
1776 * example. The field's are n = number (id of the object), f = flags,
1777 * t = type, s = state, r = refcount, e = error, p = pid.
1781 void gfs2_dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1783 const struct gfs2_glock_operations *glops = gl->gl_ops;
1784 unsigned long long dtime;
1785 const struct gfs2_holder *gh;
1786 char gflags_buf[32];
1788 dtime = jiffies - gl->gl_demote_time;
1789 dtime *= 1000000/HZ; /* demote time in uSec */
1790 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1792 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",
1793 state2str(gl->gl_state),
1794 gl->gl_name.ln_type,
1795 (unsigned long long)gl->gl_name.ln_number,
1796 gflags2str(gflags_buf, gl),
1797 state2str(gl->gl_target),
1798 state2str(gl->gl_demote_state), dtime,
1799 atomic_read(&gl->gl_ail_count),
1800 atomic_read(&gl->gl_revokes),
1801 (int)gl->gl_lockref.count, gl->gl_hold_time);
1803 list_for_each_entry(gh, &gl->gl_holders, gh_list)
1804 dump_holder(seq, gh);
1806 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1807 glops->go_dump(seq, gl);
1810 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
1812 struct gfs2_glock *gl = iter_ptr;
1814 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
1815 gl->gl_name.ln_type,
1816 (unsigned long long)gl->gl_name.ln_number,
1817 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
1818 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
1819 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
1820 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
1821 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
1822 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
1823 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
1824 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
1828 static const char *gfs2_gltype[] = {
1842 static const char *gfs2_stype[] = {
1843 [GFS2_LKS_SRTT] = "srtt",
1844 [GFS2_LKS_SRTTVAR] = "srttvar",
1845 [GFS2_LKS_SRTTB] = "srttb",
1846 [GFS2_LKS_SRTTVARB] = "srttvarb",
1847 [GFS2_LKS_SIRT] = "sirt",
1848 [GFS2_LKS_SIRTVAR] = "sirtvar",
1849 [GFS2_LKS_DCOUNT] = "dlm",
1850 [GFS2_LKS_QCOUNT] = "queue",
1853 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1855 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
1857 struct gfs2_sbd *sdp = seq->private;
1858 loff_t pos = *(loff_t *)iter_ptr;
1859 unsigned index = pos >> 3;
1860 unsigned subindex = pos & 0x07;
1863 if (index == 0 && subindex != 0)
1866 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
1867 (index == 0) ? "cpu": gfs2_stype[subindex]);
1869 for_each_possible_cpu(i) {
1870 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
1873 seq_printf(seq, " %15u", i);
1875 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
1876 lkstats[index - 1].stats[subindex]);
1878 seq_putc(seq, '\n');
1882 int __init gfs2_glock_init(void)
1886 ret = rhashtable_init(&gl_hash_table, &ht_parms);
1890 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1891 WQ_HIGHPRI | WQ_FREEZABLE, 0);
1892 if (!glock_workqueue) {
1893 rhashtable_destroy(&gl_hash_table);
1896 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1897 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1899 if (!gfs2_delete_workqueue) {
1900 destroy_workqueue(glock_workqueue);
1901 rhashtable_destroy(&gl_hash_table);
1905 ret = register_shrinker(&glock_shrinker);
1907 destroy_workqueue(gfs2_delete_workqueue);
1908 destroy_workqueue(glock_workqueue);
1909 rhashtable_destroy(&gl_hash_table);
1913 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
1914 init_waitqueue_head(glock_wait_table + i);
1919 void gfs2_glock_exit(void)
1921 unregister_shrinker(&glock_shrinker);
1922 rhashtable_destroy(&gl_hash_table);
1923 destroy_workqueue(glock_workqueue);
1924 destroy_workqueue(gfs2_delete_workqueue);
1927 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1929 while ((gi->gl = rhashtable_walk_next(&gi->hti))) {
1930 if (IS_ERR(gi->gl)) {
1931 if (PTR_ERR(gi->gl) == -EAGAIN)
1936 /* Skip entries for other sb and dead entries */
1937 if (gi->sdp == gi->gl->gl_name.ln_sbd &&
1938 !__lockref_is_dead(&gi->gl->gl_lockref))
1943 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1946 struct gfs2_glock_iter *gi = seq->private;
1949 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
1950 if (rhashtable_walk_start(&gi->hti) != 0)
1954 gfs2_glock_iter_next(gi);
1955 } while (gi->gl && n--);
1957 gi->last_pos = *pos;
1962 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1965 struct gfs2_glock_iter *gi = seq->private;
1968 gi->last_pos = *pos;
1969 gfs2_glock_iter_next(gi);
1974 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1977 struct gfs2_glock_iter *gi = seq->private;
1980 rhashtable_walk_stop(&gi->hti);
1981 rhashtable_walk_exit(&gi->hti);
1984 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1986 dump_glock(seq, iter_ptr);
1990 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
1993 if (*pos >= GFS2_NR_SBSTATS)
1998 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2002 if (*pos >= GFS2_NR_SBSTATS)
2007 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2012 static const struct seq_operations gfs2_glock_seq_ops = {
2013 .start = gfs2_glock_seq_start,
2014 .next = gfs2_glock_seq_next,
2015 .stop = gfs2_glock_seq_stop,
2016 .show = gfs2_glock_seq_show,
2019 static const struct seq_operations gfs2_glstats_seq_ops = {
2020 .start = gfs2_glock_seq_start,
2021 .next = gfs2_glock_seq_next,
2022 .stop = gfs2_glock_seq_stop,
2023 .show = gfs2_glstats_seq_show,
2026 static const struct seq_operations gfs2_sbstats_seq_ops = {
2027 .start = gfs2_sbstats_seq_start,
2028 .next = gfs2_sbstats_seq_next,
2029 .stop = gfs2_sbstats_seq_stop,
2030 .show = gfs2_sbstats_seq_show,
2033 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2035 static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2036 const struct seq_operations *ops)
2038 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2040 struct seq_file *seq = file->private_data;
2041 struct gfs2_glock_iter *gi = seq->private;
2043 gi->sdp = inode->i_private;
2044 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2046 seq->size = GFS2_SEQ_GOODSIZE;
2052 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2054 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2057 static int gfs2_glocks_release(struct inode *inode, struct file *file)
2059 struct seq_file *seq = file->private_data;
2060 struct gfs2_glock_iter *gi = seq->private;
2063 return seq_release_private(inode, file);
2066 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2068 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2071 static int gfs2_sbstats_open(struct inode *inode, struct file *file)
2073 int ret = seq_open(file, &gfs2_sbstats_seq_ops);
2075 struct seq_file *seq = file->private_data;
2076 seq->private = inode->i_private; /* sdp */
2081 static const struct file_operations gfs2_glocks_fops = {
2082 .owner = THIS_MODULE,
2083 .open = gfs2_glocks_open,
2085 .llseek = seq_lseek,
2086 .release = gfs2_glocks_release,
2089 static const struct file_operations gfs2_glstats_fops = {
2090 .owner = THIS_MODULE,
2091 .open = gfs2_glstats_open,
2093 .llseek = seq_lseek,
2094 .release = gfs2_glocks_release,
2097 static const struct file_operations gfs2_sbstats_fops = {
2098 .owner = THIS_MODULE,
2099 .open = gfs2_sbstats_open,
2101 .llseek = seq_lseek,
2102 .release = seq_release,
2105 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2107 struct dentry *dent;
2109 dent = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2110 if (IS_ERR_OR_NULL(dent))
2112 sdp->debugfs_dir = dent;
2114 dent = debugfs_create_file("glocks",
2116 sdp->debugfs_dir, sdp,
2118 if (IS_ERR_OR_NULL(dent))
2120 sdp->debugfs_dentry_glocks = dent;
2122 dent = debugfs_create_file("glstats",
2124 sdp->debugfs_dir, sdp,
2125 &gfs2_glstats_fops);
2126 if (IS_ERR_OR_NULL(dent))
2128 sdp->debugfs_dentry_glstats = dent;
2130 dent = debugfs_create_file("sbstats",
2132 sdp->debugfs_dir, sdp,
2133 &gfs2_sbstats_fops);
2134 if (IS_ERR_OR_NULL(dent))
2136 sdp->debugfs_dentry_sbstats = dent;
2140 gfs2_delete_debugfs_file(sdp);
2141 return dent ? PTR_ERR(dent) : -ENOMEM;
2144 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2146 if (sdp->debugfs_dir) {
2147 if (sdp->debugfs_dentry_glocks) {
2148 debugfs_remove(sdp->debugfs_dentry_glocks);
2149 sdp->debugfs_dentry_glocks = NULL;
2151 if (sdp->debugfs_dentry_glstats) {
2152 debugfs_remove(sdp->debugfs_dentry_glstats);
2153 sdp->debugfs_dentry_glstats = NULL;
2155 if (sdp->debugfs_dentry_sbstats) {
2156 debugfs_remove(sdp->debugfs_dentry_sbstats);
2157 sdp->debugfs_dentry_sbstats = NULL;
2159 debugfs_remove(sdp->debugfs_dir);
2160 sdp->debugfs_dir = NULL;
2164 int gfs2_register_debugfs(void)
2166 gfs2_root = debugfs_create_dir("gfs2", NULL);
2167 if (IS_ERR(gfs2_root))
2168 return PTR_ERR(gfs2_root);
2169 return gfs2_root ? 0 : -ENOMEM;
2172 void gfs2_unregister_debugfs(void)
2174 debugfs_remove(gfs2_root);