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/jhash.h>
19 #include <linux/kallsyms.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <linux/list.h>
22 #include <linux/wait.h>
23 #include <linux/module.h>
24 #include <asm/uaccess.h>
25 #include <linux/seq_file.h>
26 #include <linux/debugfs.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/workqueue.h>
30 #include <linux/jiffies.h>
31 #include <linux/rcupdate.h>
32 #include <linux/rculist_bl.h>
33 #include <linux/bit_spinlock.h>
34 #include <linux/percpu.h>
35 #include <linux/list_sort.h>
36 #include <linux/lockref.h>
37 #include <linux/rhashtable.h>
50 #define CREATE_TRACE_POINTS
51 #include "trace_gfs2.h"
53 struct gfs2_glock_iter {
54 struct gfs2_sbd *sdp; /* incore superblock */
55 struct rhashtable_iter hti; /* rhashtable iterator */
56 struct gfs2_glock *gl; /* current glock struct */
57 loff_t last_pos; /* last position */
60 typedef void (*glock_examiner) (struct gfs2_glock * gl);
62 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
64 static struct dentry *gfs2_root;
65 static struct workqueue_struct *glock_workqueue;
66 struct workqueue_struct *gfs2_delete_workqueue;
67 static LIST_HEAD(lru_list);
68 static atomic_t lru_count = ATOMIC_INIT(0);
69 static DEFINE_SPINLOCK(lru_lock);
71 #define GFS2_GL_HASH_SHIFT 15
72 #define GFS2_GL_HASH_SIZE (1 << GFS2_GL_HASH_SHIFT)
74 static struct rhashtable_params ht_parms = {
75 .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4,
76 .key_len = sizeof(struct lm_lockname),
77 .key_offset = offsetof(struct gfs2_glock, gl_name),
78 .head_offset = offsetof(struct gfs2_glock, gl_node),
81 static struct rhashtable gl_hash_table;
83 static void gfs2_glock_dealloc(struct rcu_head *rcu)
85 struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
87 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
88 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
90 kfree(gl->gl_lksb.sb_lvbptr);
91 kmem_cache_free(gfs2_glock_cachep, gl);
95 void gfs2_glock_free(struct gfs2_glock *gl)
97 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
99 call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
100 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
101 wake_up(&sdp->sd_glock_wait);
105 * gfs2_glock_hold() - increment reference count on glock
106 * @gl: The glock to hold
110 static void gfs2_glock_hold(struct gfs2_glock *gl)
112 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
113 lockref_get(&gl->gl_lockref);
117 * demote_ok - Check to see if it's ok to unlock a glock
120 * Returns: 1 if it's ok
123 static int demote_ok(const struct gfs2_glock *gl)
125 const struct gfs2_glock_operations *glops = gl->gl_ops;
127 if (gl->gl_state == LM_ST_UNLOCKED)
129 if (!list_empty(&gl->gl_holders))
131 if (glops->go_demote_ok)
132 return glops->go_demote_ok(gl);
137 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
139 spin_lock(&lru_lock);
141 if (!list_empty(&gl->gl_lru))
142 list_del_init(&gl->gl_lru);
144 atomic_inc(&lru_count);
146 list_add_tail(&gl->gl_lru, &lru_list);
147 set_bit(GLF_LRU, &gl->gl_flags);
148 spin_unlock(&lru_lock);
151 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
153 spin_lock(&lru_lock);
154 if (!list_empty(&gl->gl_lru)) {
155 list_del_init(&gl->gl_lru);
156 atomic_dec(&lru_count);
157 clear_bit(GLF_LRU, &gl->gl_flags);
159 spin_unlock(&lru_lock);
163 * gfs2_glock_put() - Decrement reference count on glock
164 * @gl: The glock to put
168 void gfs2_glock_put(struct gfs2_glock *gl)
170 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
171 struct address_space *mapping = gfs2_glock2aspace(gl);
173 if (lockref_put_or_lock(&gl->gl_lockref))
176 lockref_mark_dead(&gl->gl_lockref);
178 gfs2_glock_remove_from_lru(gl);
179 spin_unlock(&gl->gl_lockref.lock);
180 rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
181 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
182 GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
183 trace_gfs2_glock_put(gl);
184 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
188 * may_grant - check if its ok to grant a new lock
190 * @gh: The lock request which we wish to grant
192 * Returns: true if its ok to grant the lock
195 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
197 const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
198 if ((gh->gh_state == LM_ST_EXCLUSIVE ||
199 gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
201 if (gl->gl_state == gh->gh_state)
203 if (gh->gh_flags & GL_EXACT)
205 if (gl->gl_state == LM_ST_EXCLUSIVE) {
206 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
208 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
211 if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
216 static void gfs2_holder_wake(struct gfs2_holder *gh)
218 clear_bit(HIF_WAIT, &gh->gh_iflags);
219 smp_mb__after_atomic();
220 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
224 * do_error - Something unexpected has happened during a lock request
228 static inline void do_error(struct gfs2_glock *gl, const int ret)
230 struct gfs2_holder *gh, *tmp;
232 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
233 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
235 if (ret & LM_OUT_ERROR)
237 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
238 gh->gh_error = GLR_TRYFAILED;
241 list_del_init(&gh->gh_list);
242 trace_gfs2_glock_queue(gh, 0);
243 gfs2_holder_wake(gh);
248 * do_promote - promote as many requests as possible on the current queue
251 * Returns: 1 if there is a blocked holder at the head of the list, or 2
252 * if a type specific operation is underway.
255 static int do_promote(struct gfs2_glock *gl)
256 __releases(&gl->gl_lockref.lock)
257 __acquires(&gl->gl_lockref.lock)
259 const struct gfs2_glock_operations *glops = gl->gl_ops;
260 struct gfs2_holder *gh, *tmp;
264 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
265 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
267 if (may_grant(gl, gh)) {
268 if (gh->gh_list.prev == &gl->gl_holders &&
270 spin_unlock(&gl->gl_lockref.lock);
271 /* FIXME: eliminate this eventually */
272 ret = glops->go_lock(gh);
273 spin_lock(&gl->gl_lockref.lock);
278 list_del_init(&gh->gh_list);
279 trace_gfs2_glock_queue(gh, 0);
280 gfs2_holder_wake(gh);
283 set_bit(HIF_HOLDER, &gh->gh_iflags);
284 trace_gfs2_promote(gh, 1);
285 gfs2_holder_wake(gh);
288 set_bit(HIF_HOLDER, &gh->gh_iflags);
289 trace_gfs2_promote(gh, 0);
290 gfs2_holder_wake(gh);
293 if (gh->gh_list.prev == &gl->gl_holders)
302 * find_first_waiter - find the first gh that's waiting for the glock
306 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
308 struct gfs2_holder *gh;
310 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
311 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
318 * state_change - record that the glock is now in a different state
320 * @new_state the new state
324 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
328 held1 = (gl->gl_state != LM_ST_UNLOCKED);
329 held2 = (new_state != LM_ST_UNLOCKED);
331 if (held1 != held2) {
332 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
334 gl->gl_lockref.count++;
336 gl->gl_lockref.count--;
338 if (held1 && held2 && list_empty(&gl->gl_holders))
339 clear_bit(GLF_QUEUED, &gl->gl_flags);
341 if (new_state != gl->gl_target)
342 /* shorten our minimum hold time */
343 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
345 gl->gl_state = new_state;
346 gl->gl_tchange = jiffies;
349 static void gfs2_demote_wake(struct gfs2_glock *gl)
351 gl->gl_demote_state = LM_ST_EXCLUSIVE;
352 clear_bit(GLF_DEMOTE, &gl->gl_flags);
353 smp_mb__after_atomic();
354 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
358 * finish_xmote - The DLM has replied to one of our lock requests
360 * @ret: The status from the DLM
364 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
366 const struct gfs2_glock_operations *glops = gl->gl_ops;
367 struct gfs2_holder *gh;
368 unsigned state = ret & LM_OUT_ST_MASK;
371 spin_lock(&gl->gl_lockref.lock);
372 trace_gfs2_glock_state_change(gl, state);
373 state_change(gl, state);
374 gh = find_first_waiter(gl);
376 /* Demote to UN request arrived during demote to SH or DF */
377 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
378 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
379 gl->gl_target = LM_ST_UNLOCKED;
381 /* Check for state != intended state */
382 if (unlikely(state != gl->gl_target)) {
383 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
384 /* move to back of queue and try next entry */
385 if (ret & LM_OUT_CANCELED) {
386 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
387 list_move_tail(&gh->gh_list, &gl->gl_holders);
388 gh = find_first_waiter(gl);
389 gl->gl_target = gh->gh_state;
392 /* Some error or failed "try lock" - report it */
393 if ((ret & LM_OUT_ERROR) ||
394 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
395 gl->gl_target = gl->gl_state;
401 /* Unlocked due to conversion deadlock, try again */
404 do_xmote(gl, gh, gl->gl_target);
406 /* Conversion fails, unlock and try again */
409 do_xmote(gl, gh, LM_ST_UNLOCKED);
411 default: /* Everything else */
412 pr_err("wanted %u got %u\n", gl->gl_target, state);
415 spin_unlock(&gl->gl_lockref.lock);
419 /* Fast path - we got what we asked for */
420 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
421 gfs2_demote_wake(gl);
422 if (state != LM_ST_UNLOCKED) {
423 if (glops->go_xmote_bh) {
424 spin_unlock(&gl->gl_lockref.lock);
425 rv = glops->go_xmote_bh(gl, gh);
426 spin_lock(&gl->gl_lockref.lock);
437 clear_bit(GLF_LOCK, &gl->gl_flags);
439 spin_unlock(&gl->gl_lockref.lock);
443 * do_xmote - Calls the DLM to change the state of a lock
444 * @gl: The lock state
445 * @gh: The holder (only for promotes)
446 * @target: The target lock state
450 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
451 __releases(&gl->gl_lockref.lock)
452 __acquires(&gl->gl_lockref.lock)
454 const struct gfs2_glock_operations *glops = gl->gl_ops;
455 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
456 unsigned int lck_flags = gh ? gh->gh_flags : 0;
459 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
461 GLOCK_BUG_ON(gl, gl->gl_state == target);
462 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
463 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
465 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
466 do_error(gl, 0); /* Fail queued try locks */
469 set_bit(GLF_BLOCKING, &gl->gl_flags);
470 if ((gl->gl_req == LM_ST_UNLOCKED) ||
471 (gl->gl_state == LM_ST_EXCLUSIVE) ||
472 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
473 clear_bit(GLF_BLOCKING, &gl->gl_flags);
474 spin_unlock(&gl->gl_lockref.lock);
477 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
478 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
479 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
482 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
484 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
486 pr_err("lm_lock ret %d\n", ret);
489 } else { /* lock_nolock */
490 finish_xmote(gl, target);
491 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
495 spin_lock(&gl->gl_lockref.lock);
499 * find_first_holder - find the first "holder" gh
503 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
505 struct gfs2_holder *gh;
507 if (!list_empty(&gl->gl_holders)) {
508 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
509 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
516 * run_queue - do all outstanding tasks related to a glock
517 * @gl: The glock in question
518 * @nonblock: True if we must not block in run_queue
522 static void run_queue(struct gfs2_glock *gl, const int nonblock)
523 __releases(&gl->gl_lockref.lock)
524 __acquires(&gl->gl_lockref.lock)
526 struct gfs2_holder *gh = NULL;
529 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
532 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
534 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
535 gl->gl_demote_state != gl->gl_state) {
536 if (find_first_holder(gl))
540 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
541 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
542 gl->gl_target = gl->gl_demote_state;
544 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
545 gfs2_demote_wake(gl);
546 ret = do_promote(gl);
551 gh = find_first_waiter(gl);
552 gl->gl_target = gh->gh_state;
553 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
554 do_error(gl, 0); /* Fail queued try locks */
556 do_xmote(gl, gh, gl->gl_target);
561 clear_bit(GLF_LOCK, &gl->gl_flags);
562 smp_mb__after_atomic();
563 gl->gl_lockref.count++;
564 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
565 gl->gl_lockref.count--;
569 clear_bit(GLF_LOCK, &gl->gl_flags);
570 smp_mb__after_atomic();
574 static void delete_work_func(struct work_struct *work)
576 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
577 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
578 struct gfs2_inode *ip;
580 u64 no_addr = gl->gl_name.ln_number;
583 /* Note: Unsafe to dereference ip as we don't hold right refs/locks */
586 inode = gfs2_ilookup(sdp->sd_vfs, no_addr, 1);
588 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
589 if (inode && !IS_ERR(inode)) {
590 d_prune_aliases(inode);
596 static void glock_work_func(struct work_struct *work)
598 unsigned long delay = 0;
599 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
602 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
603 finish_xmote(gl, gl->gl_reply);
606 spin_lock(&gl->gl_lockref.lock);
607 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
608 gl->gl_state != LM_ST_UNLOCKED &&
609 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
610 unsigned long holdtime, now = jiffies;
612 holdtime = gl->gl_tchange + gl->gl_hold_time;
613 if (time_before(now, holdtime))
614 delay = holdtime - now;
617 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
618 set_bit(GLF_DEMOTE, &gl->gl_flags);
622 spin_unlock(&gl->gl_lockref.lock);
626 if (gl->gl_name.ln_type != LM_TYPE_INODE)
628 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
636 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
637 * @sdp: The GFS2 superblock
638 * @number: the lock number
639 * @glops: The glock_operations to use
640 * @create: If 0, don't create the glock if it doesn't exist
641 * @glp: the glock is returned here
643 * This does not lock a glock, just finds/creates structures for one.
648 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
649 const struct gfs2_glock_operations *glops, int create,
650 struct gfs2_glock **glp)
652 struct super_block *s = sdp->sd_vfs;
653 struct lm_lockname name = { .ln_number = number,
654 .ln_type = glops->go_type,
656 struct gfs2_glock *gl, *tmp = NULL;
657 struct address_space *mapping;
658 struct kmem_cache *cachep;
662 gl = rhashtable_lookup_fast(&gl_hash_table, &name, ht_parms);
663 if (gl && !lockref_get_not_dead(&gl->gl_lockref))
673 if (glops->go_flags & GLOF_ASPACE)
674 cachep = gfs2_glock_aspace_cachep;
676 cachep = gfs2_glock_cachep;
677 gl = kmem_cache_alloc(cachep, GFP_NOFS);
681 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
683 if (glops->go_flags & GLOF_LVB) {
684 gl->gl_lksb.sb_lvbptr = kzalloc(GFS2_MIN_LVB_SIZE, GFP_NOFS);
685 if (!gl->gl_lksb.sb_lvbptr) {
686 kmem_cache_free(cachep, gl);
691 atomic_inc(&sdp->sd_glock_disposal);
692 gl->gl_node.next = NULL;
695 gl->gl_lockref.count = 1;
696 gl->gl_state = LM_ST_UNLOCKED;
697 gl->gl_target = LM_ST_UNLOCKED;
698 gl->gl_demote_state = LM_ST_EXCLUSIVE;
700 gl->gl_dstamp = ktime_set(0, 0);
702 /* We use the global stats to estimate the initial per-glock stats */
703 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
705 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
706 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
707 gl->gl_tchange = jiffies;
708 gl->gl_object = NULL;
709 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
710 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
711 INIT_WORK(&gl->gl_delete, delete_work_func);
713 mapping = gfs2_glock2aspace(gl);
715 mapping->a_ops = &gfs2_meta_aops;
716 mapping->host = s->s_bdev->bd_inode;
718 mapping_set_gfp_mask(mapping, GFP_NOFS);
719 mapping->private_data = NULL;
720 mapping->writeback_index = 0;
724 ret = rhashtable_lookup_insert_fast(&gl_hash_table, &gl->gl_node,
731 if (ret == -EEXIST) {
734 tmp = rhashtable_lookup_fast(&gl_hash_table, &name, ht_parms);
735 if (tmp == NULL || !lockref_get_not_dead(&tmp->gl_lockref)) {
748 kfree(gl->gl_lksb.sb_lvbptr);
749 kmem_cache_free(cachep, gl);
750 atomic_dec(&sdp->sd_glock_disposal);
757 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
759 * @state: the state we're requesting
760 * @flags: the modifier flags
761 * @gh: the holder structure
765 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
766 struct gfs2_holder *gh)
768 INIT_LIST_HEAD(&gh->gh_list);
770 gh->gh_ip = _RET_IP_;
771 gh->gh_owner_pid = get_pid(task_pid(current));
772 gh->gh_state = state;
773 gh->gh_flags = flags;
780 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
781 * @state: the state we're requesting
782 * @flags: the modifier flags
783 * @gh: the holder structure
785 * Don't mess with the glock.
789 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
791 gh->gh_state = state;
792 gh->gh_flags = flags;
794 gh->gh_ip = _RET_IP_;
795 put_pid(gh->gh_owner_pid);
796 gh->gh_owner_pid = get_pid(task_pid(current));
800 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
801 * @gh: the holder structure
805 void gfs2_holder_uninit(struct gfs2_holder *gh)
807 put_pid(gh->gh_owner_pid);
808 gfs2_glock_put(gh->gh_gl);
814 * gfs2_glock_wait - wait on a glock acquisition
815 * @gh: the glock holder
817 * Returns: 0 on success
820 int gfs2_glock_wait(struct gfs2_holder *gh)
822 unsigned long time1 = jiffies;
825 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
826 if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
827 /* Lengthen the minimum hold time. */
828 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
835 * handle_callback - process a demote request
837 * @state: the state the caller wants us to change to
839 * There are only two requests that we are going to see in actual
840 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
843 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
844 unsigned long delay, bool remote)
846 int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
848 set_bit(bit, &gl->gl_flags);
849 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
850 gl->gl_demote_state = state;
851 gl->gl_demote_time = jiffies;
852 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
853 gl->gl_demote_state != state) {
854 gl->gl_demote_state = LM_ST_UNLOCKED;
856 if (gl->gl_ops->go_callback)
857 gl->gl_ops->go_callback(gl, remote);
858 trace_gfs2_demote_rq(gl, remote);
861 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
863 struct va_format vaf;
869 seq_vprintf(seq, fmt, args);
881 * add_to_queue - Add a holder to the wait queue (but look for recursion)
882 * @gh: the holder structure to add
884 * Eventually we should move the recursive locking trap to a
885 * debugging option or something like that. This is the fast
886 * path and needs to have the minimum number of distractions.
890 static inline void add_to_queue(struct gfs2_holder *gh)
891 __releases(&gl->gl_lockref.lock)
892 __acquires(&gl->gl_lockref.lock)
894 struct gfs2_glock *gl = gh->gh_gl;
895 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
896 struct list_head *insert_pt = NULL;
897 struct gfs2_holder *gh2;
900 BUG_ON(gh->gh_owner_pid == NULL);
901 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
904 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
905 if (test_bit(GLF_LOCK, &gl->gl_flags))
906 try_futile = !may_grant(gl, gh);
907 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
911 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
912 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
913 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
916 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
918 gh->gh_error = GLR_TRYFAILED;
919 gfs2_holder_wake(gh);
922 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
924 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
925 insert_pt = &gh2->gh_list;
927 set_bit(GLF_QUEUED, &gl->gl_flags);
928 trace_gfs2_glock_queue(gh, 1);
929 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
930 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
931 if (likely(insert_pt == NULL)) {
932 list_add_tail(&gh->gh_list, &gl->gl_holders);
933 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
937 list_add_tail(&gh->gh_list, insert_pt);
939 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
940 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
941 spin_unlock(&gl->gl_lockref.lock);
942 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
943 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
944 spin_lock(&gl->gl_lockref.lock);
949 pr_err("original: %pSR\n", (void *)gh2->gh_ip);
950 pr_err("pid: %d\n", pid_nr(gh2->gh_owner_pid));
951 pr_err("lock type: %d req lock state : %d\n",
952 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
953 pr_err("new: %pSR\n", (void *)gh->gh_ip);
954 pr_err("pid: %d\n", pid_nr(gh->gh_owner_pid));
955 pr_err("lock type: %d req lock state : %d\n",
956 gh->gh_gl->gl_name.ln_type, gh->gh_state);
957 gfs2_dump_glock(NULL, gl);
962 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
963 * @gh: the holder structure
965 * if (gh->gh_flags & GL_ASYNC), this never returns an error
967 * Returns: 0, GLR_TRYFAILED, or errno on failure
970 int gfs2_glock_nq(struct gfs2_holder *gh)
972 struct gfs2_glock *gl = gh->gh_gl;
973 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
976 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
979 if (test_bit(GLF_LRU, &gl->gl_flags))
980 gfs2_glock_remove_from_lru(gl);
982 spin_lock(&gl->gl_lockref.lock);
984 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
985 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
986 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
987 gl->gl_lockref.count++;
988 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
989 gl->gl_lockref.count--;
992 spin_unlock(&gl->gl_lockref.lock);
994 if (!(gh->gh_flags & GL_ASYNC))
995 error = gfs2_glock_wait(gh);
1001 * gfs2_glock_poll - poll to see if an async request has been completed
1004 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1007 int gfs2_glock_poll(struct gfs2_holder *gh)
1009 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1013 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1014 * @gh: the glock holder
1018 void gfs2_glock_dq(struct gfs2_holder *gh)
1020 struct gfs2_glock *gl = gh->gh_gl;
1021 const struct gfs2_glock_operations *glops = gl->gl_ops;
1025 spin_lock(&gl->gl_lockref.lock);
1026 if (gh->gh_flags & GL_NOCACHE)
1027 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1029 list_del_init(&gh->gh_list);
1030 if (find_first_holder(gl) == NULL) {
1031 if (glops->go_unlock) {
1032 GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1033 spin_unlock(&gl->gl_lockref.lock);
1034 glops->go_unlock(gh);
1035 spin_lock(&gl->gl_lockref.lock);
1036 clear_bit(GLF_LOCK, &gl->gl_flags);
1038 if (list_empty(&gl->gl_holders) &&
1039 !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1040 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1043 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl) &&
1044 (glops->go_flags & GLOF_LRU))
1045 gfs2_glock_add_to_lru(gl);
1047 trace_gfs2_glock_queue(gh, 0);
1048 spin_unlock(&gl->gl_lockref.lock);
1049 if (likely(fast_path))
1052 gfs2_glock_hold(gl);
1053 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1054 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1055 gl->gl_name.ln_type == LM_TYPE_INODE)
1056 delay = gl->gl_hold_time;
1057 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1061 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1063 struct gfs2_glock *gl = gh->gh_gl;
1066 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1070 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1071 * @gh: the holder structure
1075 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1078 gfs2_holder_uninit(gh);
1082 * gfs2_glock_nq_num - acquire a glock based on lock number
1083 * @sdp: the filesystem
1084 * @number: the lock number
1085 * @glops: the glock operations for the type of glock
1086 * @state: the state to acquire the glock in
1087 * @flags: modifier flags for the acquisition
1088 * @gh: the struct gfs2_holder
1093 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1094 const struct gfs2_glock_operations *glops,
1095 unsigned int state, int flags, struct gfs2_holder *gh)
1097 struct gfs2_glock *gl;
1100 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1102 error = gfs2_glock_nq_init(gl, state, flags, gh);
1110 * glock_compare - Compare two struct gfs2_glock structures for sorting
1111 * @arg_a: the first structure
1112 * @arg_b: the second structure
1116 static int glock_compare(const void *arg_a, const void *arg_b)
1118 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1119 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1120 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1121 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1123 if (a->ln_number > b->ln_number)
1125 if (a->ln_number < b->ln_number)
1127 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1132 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1133 * @num_gh: the number of structures
1134 * @ghs: an array of struct gfs2_holder structures
1136 * Returns: 0 on success (all glocks acquired),
1137 * errno on failure (no glocks acquired)
1140 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1141 struct gfs2_holder **p)
1146 for (x = 0; x < num_gh; x++)
1149 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1151 for (x = 0; x < num_gh; x++) {
1152 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1154 error = gfs2_glock_nq(p[x]);
1157 gfs2_glock_dq(p[x]);
1166 * gfs2_glock_nq_m - acquire multiple glocks
1167 * @num_gh: the number of structures
1168 * @ghs: an array of struct gfs2_holder structures
1171 * Returns: 0 on success (all glocks acquired),
1172 * errno on failure (no glocks acquired)
1175 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1177 struct gfs2_holder *tmp[4];
1178 struct gfs2_holder **pph = tmp;
1185 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1186 return gfs2_glock_nq(ghs);
1190 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1195 error = nq_m_sync(num_gh, ghs, pph);
1204 * gfs2_glock_dq_m - release multiple glocks
1205 * @num_gh: the number of structures
1206 * @ghs: an array of struct gfs2_holder structures
1210 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1213 gfs2_glock_dq(&ghs[num_gh]);
1216 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1218 unsigned long delay = 0;
1219 unsigned long holdtime;
1220 unsigned long now = jiffies;
1222 gfs2_glock_hold(gl);
1223 holdtime = gl->gl_tchange + gl->gl_hold_time;
1224 if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1225 gl->gl_name.ln_type == LM_TYPE_INODE) {
1226 if (time_before(now, holdtime))
1227 delay = holdtime - now;
1228 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1229 delay = gl->gl_hold_time;
1232 spin_lock(&gl->gl_lockref.lock);
1233 handle_callback(gl, state, delay, true);
1234 spin_unlock(&gl->gl_lockref.lock);
1235 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1240 * gfs2_should_freeze - Figure out if glock should be frozen
1241 * @gl: The glock in question
1243 * Glocks are not frozen if (a) the result of the dlm operation is
1244 * an error, (b) the locking operation was an unlock operation or
1245 * (c) if there is a "noexp" flagged request anywhere in the queue
1247 * Returns: 1 if freezing should occur, 0 otherwise
1250 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1252 const struct gfs2_holder *gh;
1254 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1256 if (gl->gl_target == LM_ST_UNLOCKED)
1259 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1260 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1262 if (LM_FLAG_NOEXP & gh->gh_flags)
1270 * gfs2_glock_complete - Callback used by locking
1271 * @gl: Pointer to the glock
1272 * @ret: The return value from the dlm
1274 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1275 * to use a bitfield shared with other glock state fields.
1278 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1280 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1282 spin_lock(&gl->gl_lockref.lock);
1285 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1286 if (gfs2_should_freeze(gl)) {
1287 set_bit(GLF_FROZEN, &gl->gl_flags);
1288 spin_unlock(&gl->gl_lockref.lock);
1293 gl->gl_lockref.count++;
1294 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1295 spin_unlock(&gl->gl_lockref.lock);
1297 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1301 static int glock_cmp(void *priv, struct list_head *a, struct list_head *b)
1303 struct gfs2_glock *gla, *glb;
1305 gla = list_entry(a, struct gfs2_glock, gl_lru);
1306 glb = list_entry(b, struct gfs2_glock, gl_lru);
1308 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1310 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1317 * gfs2_dispose_glock_lru - Demote a list of glocks
1318 * @list: The list to dispose of
1320 * Disposing of glocks may involve disk accesses, so that here we sort
1321 * the glocks by number (i.e. disk location of the inodes) so that if
1322 * there are any such accesses, they'll be sent in order (mostly).
1324 * Must be called under the lru_lock, but may drop and retake this
1325 * lock. While the lru_lock is dropped, entries may vanish from the
1326 * list, but no new entries will appear on the list (since it is
1330 static void gfs2_dispose_glock_lru(struct list_head *list)
1331 __releases(&lru_lock)
1332 __acquires(&lru_lock)
1334 struct gfs2_glock *gl;
1336 list_sort(NULL, list, glock_cmp);
1338 while(!list_empty(list)) {
1339 gl = list_entry(list->next, struct gfs2_glock, gl_lru);
1340 list_del_init(&gl->gl_lru);
1341 if (!spin_trylock(&gl->gl_lockref.lock)) {
1343 list_add(&gl->gl_lru, &lru_list);
1344 atomic_inc(&lru_count);
1347 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1348 spin_unlock(&gl->gl_lockref.lock);
1349 goto add_back_to_lru;
1351 clear_bit(GLF_LRU, &gl->gl_flags);
1352 gl->gl_lockref.count++;
1354 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1355 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1356 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1357 gl->gl_lockref.count--;
1358 spin_unlock(&gl->gl_lockref.lock);
1359 cond_resched_lock(&lru_lock);
1364 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1365 * @nr: The number of entries to scan
1367 * This function selects the entries on the LRU which are able to
1368 * be demoted, and then kicks off the process by calling
1369 * gfs2_dispose_glock_lru() above.
1372 static long gfs2_scan_glock_lru(int nr)
1374 struct gfs2_glock *gl;
1379 spin_lock(&lru_lock);
1380 while ((nr-- >= 0) && !list_empty(&lru_list)) {
1381 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1383 /* Test for being demotable */
1384 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1385 list_move(&gl->gl_lru, &dispose);
1386 atomic_dec(&lru_count);
1391 list_move(&gl->gl_lru, &skipped);
1393 list_splice(&skipped, &lru_list);
1394 if (!list_empty(&dispose))
1395 gfs2_dispose_glock_lru(&dispose);
1396 spin_unlock(&lru_lock);
1401 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1402 struct shrink_control *sc)
1404 if (!(sc->gfp_mask & __GFP_FS))
1406 return gfs2_scan_glock_lru(sc->nr_to_scan);
1409 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1410 struct shrink_control *sc)
1412 return vfs_pressure_ratio(atomic_read(&lru_count));
1415 static struct shrinker glock_shrinker = {
1416 .seeks = DEFAULT_SEEKS,
1417 .count_objects = gfs2_glock_shrink_count,
1418 .scan_objects = gfs2_glock_shrink_scan,
1422 * examine_bucket - Call a function for glock in a hash bucket
1423 * @examiner: the function
1424 * @sdp: the filesystem
1425 * @bucket: the bucket
1429 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1431 struct gfs2_glock *gl;
1432 struct rhash_head *pos, *next;
1433 const struct bucket_table *tbl;
1437 tbl = rht_dereference_rcu(gl_hash_table.tbl, &gl_hash_table);
1438 for (i = 0; i < tbl->size; i++) {
1439 rht_for_each_entry_safe(gl, pos, next, tbl, i, gl_node) {
1440 if ((gl->gl_name.ln_sbd == sdp) &&
1441 lockref_get_not_dead(&gl->gl_lockref))
1450 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1451 * @gl: The glock to thaw
1455 static void thaw_glock(struct gfs2_glock *gl)
1457 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1459 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1460 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0) {
1467 * clear_glock - look at a glock and see if we can free it from glock cache
1468 * @gl: the glock to look at
1472 static void clear_glock(struct gfs2_glock *gl)
1474 gfs2_glock_remove_from_lru(gl);
1476 spin_lock(&gl->gl_lockref.lock);
1477 if (gl->gl_state != LM_ST_UNLOCKED)
1478 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1479 spin_unlock(&gl->gl_lockref.lock);
1480 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1485 * gfs2_glock_thaw - Thaw any frozen glocks
1486 * @sdp: The super block
1490 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1492 glock_hash_walk(thaw_glock, sdp);
1495 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1497 spin_lock(&gl->gl_lockref.lock);
1498 gfs2_dump_glock(seq, gl);
1499 spin_unlock(&gl->gl_lockref.lock);
1502 static void dump_glock_func(struct gfs2_glock *gl)
1504 dump_glock(NULL, gl);
1508 * gfs2_gl_hash_clear - Empty out the glock hash table
1509 * @sdp: the filesystem
1510 * @wait: wait until it's all gone
1512 * Called when unmounting the filesystem.
1515 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1517 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1518 flush_workqueue(glock_workqueue);
1519 glock_hash_walk(clear_glock, sdp);
1520 flush_workqueue(glock_workqueue);
1521 wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
1522 glock_hash_walk(dump_glock_func, sdp);
1525 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1527 struct gfs2_glock *gl = ip->i_gl;
1530 ret = gfs2_truncatei_resume(ip);
1531 gfs2_assert_withdraw(gl->gl_name.ln_sbd, ret == 0);
1533 spin_lock(&gl->gl_lockref.lock);
1534 clear_bit(GLF_LOCK, &gl->gl_flags);
1536 spin_unlock(&gl->gl_lockref.lock);
1539 static const char *state2str(unsigned state)
1542 case LM_ST_UNLOCKED:
1546 case LM_ST_DEFERRED:
1548 case LM_ST_EXCLUSIVE:
1554 static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1557 if (flags & LM_FLAG_TRY)
1559 if (flags & LM_FLAG_TRY_1CB)
1561 if (flags & LM_FLAG_NOEXP)
1563 if (flags & LM_FLAG_ANY)
1565 if (flags & LM_FLAG_PRIORITY)
1567 if (flags & GL_ASYNC)
1569 if (flags & GL_EXACT)
1571 if (flags & GL_NOCACHE)
1573 if (test_bit(HIF_HOLDER, &iflags))
1575 if (test_bit(HIF_WAIT, &iflags))
1577 if (test_bit(HIF_FIRST, &iflags))
1584 * dump_holder - print information about a glock holder
1585 * @seq: the seq_file struct
1586 * @gh: the glock holder
1590 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1592 struct task_struct *gh_owner = NULL;
1596 if (gh->gh_owner_pid)
1597 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1598 gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1599 state2str(gh->gh_state),
1600 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1602 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1603 gh_owner ? gh_owner->comm : "(ended)",
1608 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1610 const unsigned long *gflags = &gl->gl_flags;
1613 if (test_bit(GLF_LOCK, gflags))
1615 if (test_bit(GLF_DEMOTE, gflags))
1617 if (test_bit(GLF_PENDING_DEMOTE, gflags))
1619 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1621 if (test_bit(GLF_DIRTY, gflags))
1623 if (test_bit(GLF_LFLUSH, gflags))
1625 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1627 if (test_bit(GLF_REPLY_PENDING, gflags))
1629 if (test_bit(GLF_INITIAL, gflags))
1631 if (test_bit(GLF_FROZEN, gflags))
1633 if (test_bit(GLF_QUEUED, gflags))
1635 if (test_bit(GLF_LRU, gflags))
1639 if (test_bit(GLF_BLOCKING, gflags))
1646 * gfs2_dump_glock - print information about a glock
1647 * @seq: The seq_file struct
1650 * The file format is as follows:
1651 * One line per object, capital letters are used to indicate objects
1652 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1653 * other objects are indented by a single space and follow the glock to
1654 * which they are related. Fields are indicated by lower case letters
1655 * followed by a colon and the field value, except for strings which are in
1656 * [] so that its possible to see if they are composed of spaces for
1657 * example. The field's are n = number (id of the object), f = flags,
1658 * t = type, s = state, r = refcount, e = error, p = pid.
1662 void gfs2_dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1664 const struct gfs2_glock_operations *glops = gl->gl_ops;
1665 unsigned long long dtime;
1666 const struct gfs2_holder *gh;
1667 char gflags_buf[32];
1669 dtime = jiffies - gl->gl_demote_time;
1670 dtime *= 1000000/HZ; /* demote time in uSec */
1671 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1673 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",
1674 state2str(gl->gl_state),
1675 gl->gl_name.ln_type,
1676 (unsigned long long)gl->gl_name.ln_number,
1677 gflags2str(gflags_buf, gl),
1678 state2str(gl->gl_target),
1679 state2str(gl->gl_demote_state), dtime,
1680 atomic_read(&gl->gl_ail_count),
1681 atomic_read(&gl->gl_revokes),
1682 (int)gl->gl_lockref.count, gl->gl_hold_time);
1684 list_for_each_entry(gh, &gl->gl_holders, gh_list)
1685 dump_holder(seq, gh);
1687 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1688 glops->go_dump(seq, gl);
1691 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
1693 struct gfs2_glock *gl = iter_ptr;
1695 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
1696 gl->gl_name.ln_type,
1697 (unsigned long long)gl->gl_name.ln_number,
1698 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
1699 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
1700 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
1701 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
1702 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
1703 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
1704 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
1705 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
1709 static const char *gfs2_gltype[] = {
1723 static const char *gfs2_stype[] = {
1724 [GFS2_LKS_SRTT] = "srtt",
1725 [GFS2_LKS_SRTTVAR] = "srttvar",
1726 [GFS2_LKS_SRTTB] = "srttb",
1727 [GFS2_LKS_SRTTVARB] = "srttvarb",
1728 [GFS2_LKS_SIRT] = "sirt",
1729 [GFS2_LKS_SIRTVAR] = "sirtvar",
1730 [GFS2_LKS_DCOUNT] = "dlm",
1731 [GFS2_LKS_QCOUNT] = "queue",
1734 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1736 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
1738 struct gfs2_sbd *sdp = seq->private;
1739 loff_t pos = *(loff_t *)iter_ptr;
1740 unsigned index = pos >> 3;
1741 unsigned subindex = pos & 0x07;
1744 if (index == 0 && subindex != 0)
1747 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
1748 (index == 0) ? "cpu": gfs2_stype[subindex]);
1750 for_each_possible_cpu(i) {
1751 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
1754 seq_printf(seq, " %15u", i);
1756 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
1757 lkstats[index - 1].stats[subindex]);
1759 seq_putc(seq, '\n');
1763 int __init gfs2_glock_init(void)
1767 ret = rhashtable_init(&gl_hash_table, &ht_parms);
1771 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1772 WQ_HIGHPRI | WQ_FREEZABLE, 0);
1773 if (!glock_workqueue) {
1774 rhashtable_destroy(&gl_hash_table);
1777 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1778 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1780 if (!gfs2_delete_workqueue) {
1781 destroy_workqueue(glock_workqueue);
1782 rhashtable_destroy(&gl_hash_table);
1786 register_shrinker(&glock_shrinker);
1791 void gfs2_glock_exit(void)
1793 unregister_shrinker(&glock_shrinker);
1794 rhashtable_destroy(&gl_hash_table);
1795 destroy_workqueue(glock_workqueue);
1796 destroy_workqueue(gfs2_delete_workqueue);
1799 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1802 gi->gl = rhashtable_walk_next(&gi->hti);
1803 if (IS_ERR(gi->gl)) {
1804 if (PTR_ERR(gi->gl) == -EAGAIN)
1808 /* Skip entries for other sb and dead entries */
1809 } while ((gi->gl) && ((gi->sdp != gi->gl->gl_name.ln_sbd) ||
1810 __lockref_is_dead(&gi->gl->gl_lockref)));
1813 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1815 struct gfs2_glock_iter *gi = seq->private;
1819 if (gi->last_pos <= *pos)
1820 n = (*pos - gi->last_pos);
1822 ret = rhashtable_walk_start(&gi->hti);
1827 gfs2_glock_iter_next(gi);
1828 } while (gi->gl && n--);
1830 gi->last_pos = *pos;
1834 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1837 struct gfs2_glock_iter *gi = seq->private;
1840 gi->last_pos = *pos;
1841 gfs2_glock_iter_next(gi);
1845 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1847 struct gfs2_glock_iter *gi = seq->private;
1850 rhashtable_walk_stop(&gi->hti);
1853 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1855 dump_glock(seq, iter_ptr);
1859 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
1862 if (*pos >= GFS2_NR_SBSTATS)
1867 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
1871 if (*pos >= GFS2_NR_SBSTATS)
1876 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
1881 static const struct seq_operations gfs2_glock_seq_ops = {
1882 .start = gfs2_glock_seq_start,
1883 .next = gfs2_glock_seq_next,
1884 .stop = gfs2_glock_seq_stop,
1885 .show = gfs2_glock_seq_show,
1888 static const struct seq_operations gfs2_glstats_seq_ops = {
1889 .start = gfs2_glock_seq_start,
1890 .next = gfs2_glock_seq_next,
1891 .stop = gfs2_glock_seq_stop,
1892 .show = gfs2_glstats_seq_show,
1895 static const struct seq_operations gfs2_sbstats_seq_ops = {
1896 .start = gfs2_sbstats_seq_start,
1897 .next = gfs2_sbstats_seq_next,
1898 .stop = gfs2_sbstats_seq_stop,
1899 .show = gfs2_sbstats_seq_show,
1902 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
1904 static int gfs2_glocks_open(struct inode *inode, struct file *file)
1906 int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1907 sizeof(struct gfs2_glock_iter));
1909 struct seq_file *seq = file->private_data;
1910 struct gfs2_glock_iter *gi = seq->private;
1912 gi->sdp = inode->i_private;
1914 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
1916 seq->size = GFS2_SEQ_GOODSIZE;
1918 ret = rhashtable_walk_init(&gl_hash_table, &gi->hti);
1923 static int gfs2_glocks_release(struct inode *inode, struct file *file)
1925 struct seq_file *seq = file->private_data;
1926 struct gfs2_glock_iter *gi = seq->private;
1929 rhashtable_walk_exit(&gi->hti);
1930 return seq_release_private(inode, file);
1933 static int gfs2_glstats_open(struct inode *inode, struct file *file)
1935 int ret = seq_open_private(file, &gfs2_glstats_seq_ops,
1936 sizeof(struct gfs2_glock_iter));
1938 struct seq_file *seq = file->private_data;
1939 struct gfs2_glock_iter *gi = seq->private;
1940 gi->sdp = inode->i_private;
1942 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
1944 seq->size = GFS2_SEQ_GOODSIZE;
1946 ret = rhashtable_walk_init(&gl_hash_table, &gi->hti);
1951 static int gfs2_sbstats_open(struct inode *inode, struct file *file)
1953 int ret = seq_open(file, &gfs2_sbstats_seq_ops);
1955 struct seq_file *seq = file->private_data;
1956 seq->private = inode->i_private; /* sdp */
1961 static const struct file_operations gfs2_glocks_fops = {
1962 .owner = THIS_MODULE,
1963 .open = gfs2_glocks_open,
1965 .llseek = seq_lseek,
1966 .release = gfs2_glocks_release,
1969 static const struct file_operations gfs2_glstats_fops = {
1970 .owner = THIS_MODULE,
1971 .open = gfs2_glstats_open,
1973 .llseek = seq_lseek,
1974 .release = gfs2_glocks_release,
1977 static const struct file_operations gfs2_sbstats_fops = {
1978 .owner = THIS_MODULE,
1979 .open = gfs2_sbstats_open,
1981 .llseek = seq_lseek,
1982 .release = seq_release,
1985 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
1987 struct dentry *dent;
1989 dent = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
1990 if (IS_ERR_OR_NULL(dent))
1992 sdp->debugfs_dir = dent;
1994 dent = debugfs_create_file("glocks",
1996 sdp->debugfs_dir, sdp,
1998 if (IS_ERR_OR_NULL(dent))
2000 sdp->debugfs_dentry_glocks = dent;
2002 dent = debugfs_create_file("glstats",
2004 sdp->debugfs_dir, sdp,
2005 &gfs2_glstats_fops);
2006 if (IS_ERR_OR_NULL(dent))
2008 sdp->debugfs_dentry_glstats = dent;
2010 dent = debugfs_create_file("sbstats",
2012 sdp->debugfs_dir, sdp,
2013 &gfs2_sbstats_fops);
2014 if (IS_ERR_OR_NULL(dent))
2016 sdp->debugfs_dentry_sbstats = dent;
2020 gfs2_delete_debugfs_file(sdp);
2021 return dent ? PTR_ERR(dent) : -ENOMEM;
2024 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2026 if (sdp->debugfs_dir) {
2027 if (sdp->debugfs_dentry_glocks) {
2028 debugfs_remove(sdp->debugfs_dentry_glocks);
2029 sdp->debugfs_dentry_glocks = NULL;
2031 if (sdp->debugfs_dentry_glstats) {
2032 debugfs_remove(sdp->debugfs_dentry_glstats);
2033 sdp->debugfs_dentry_glstats = NULL;
2035 if (sdp->debugfs_dentry_sbstats) {
2036 debugfs_remove(sdp->debugfs_dentry_sbstats);
2037 sdp->debugfs_dentry_sbstats = NULL;
2039 debugfs_remove(sdp->debugfs_dir);
2040 sdp->debugfs_dir = NULL;
2044 int gfs2_register_debugfs(void)
2046 gfs2_root = debugfs_create_dir("gfs2", NULL);
2047 if (IS_ERR(gfs2_root))
2048 return PTR_ERR(gfs2_root);
2049 return gfs2_root ? 0 : -ENOMEM;
2052 void gfs2_unregister_debugfs(void)
2054 debugfs_remove(gfs2_root);