1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 #include <linux/sched.h>
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/buffer_head.h>
13 #include <linux/delay.h>
14 #include <linux/sort.h>
15 #include <linux/hash.h>
16 #include <linux/jhash.h>
17 #include <linux/kallsyms.h>
18 #include <linux/gfs2_ondisk.h>
19 #include <linux/list.h>
20 #include <linux/wait.h>
21 #include <linux/module.h>
22 #include <linux/uaccess.h>
23 #include <linux/seq_file.h>
24 #include <linux/debugfs.h>
25 #include <linux/kthread.h>
26 #include <linux/freezer.h>
27 #include <linux/workqueue.h>
28 #include <linux/jiffies.h>
29 #include <linux/rcupdate.h>
30 #include <linux/rculist_bl.h>
31 #include <linux/bit_spinlock.h>
32 #include <linux/percpu.h>
33 #include <linux/list_sort.h>
34 #include <linux/lockref.h>
35 #include <linux/rhashtable.h>
48 #define CREATE_TRACE_POINTS
49 #include "trace_gfs2.h"
51 struct gfs2_glock_iter {
52 struct gfs2_sbd *sdp; /* incore superblock */
53 struct rhashtable_iter hti; /* rhashtable iterator */
54 struct gfs2_glock *gl; /* current glock struct */
55 loff_t last_pos; /* last position */
58 typedef void (*glock_examiner) (struct gfs2_glock * gl);
60 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
62 static struct dentry *gfs2_root;
63 static struct workqueue_struct *glock_workqueue;
64 struct workqueue_struct *gfs2_delete_workqueue;
65 static LIST_HEAD(lru_list);
66 static atomic_t lru_count = ATOMIC_INIT(0);
67 static DEFINE_SPINLOCK(lru_lock);
69 #define GFS2_GL_HASH_SHIFT 15
70 #define GFS2_GL_HASH_SIZE BIT(GFS2_GL_HASH_SHIFT)
72 static const struct rhashtable_params ht_parms = {
73 .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4,
74 .key_len = offsetofend(struct lm_lockname, ln_type),
75 .key_offset = offsetof(struct gfs2_glock, gl_name),
76 .head_offset = offsetof(struct gfs2_glock, gl_node),
79 static struct rhashtable gl_hash_table;
81 #define GLOCK_WAIT_TABLE_BITS 12
82 #define GLOCK_WAIT_TABLE_SIZE (1 << GLOCK_WAIT_TABLE_BITS)
83 static wait_queue_head_t glock_wait_table[GLOCK_WAIT_TABLE_SIZE] __cacheline_aligned;
85 struct wait_glock_queue {
86 struct lm_lockname *name;
87 wait_queue_entry_t wait;
90 static int glock_wake_function(wait_queue_entry_t *wait, unsigned int mode,
93 struct wait_glock_queue *wait_glock =
94 container_of(wait, struct wait_glock_queue, wait);
95 struct lm_lockname *wait_name = wait_glock->name;
96 struct lm_lockname *wake_name = key;
98 if (wake_name->ln_sbd != wait_name->ln_sbd ||
99 wake_name->ln_number != wait_name->ln_number ||
100 wake_name->ln_type != wait_name->ln_type)
102 return autoremove_wake_function(wait, mode, sync, key);
105 static wait_queue_head_t *glock_waitqueue(struct lm_lockname *name)
107 u32 hash = jhash2((u32 *)name, ht_parms.key_len / 4, 0);
109 return glock_wait_table + hash_32(hash, GLOCK_WAIT_TABLE_BITS);
113 * wake_up_glock - Wake up waiters on a glock
116 static void wake_up_glock(struct gfs2_glock *gl)
118 wait_queue_head_t *wq = glock_waitqueue(&gl->gl_name);
120 if (waitqueue_active(wq))
121 __wake_up(wq, TASK_NORMAL, 1, &gl->gl_name);
124 static void gfs2_glock_dealloc(struct rcu_head *rcu)
126 struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
128 kfree(gl->gl_lksb.sb_lvbptr);
129 if (gl->gl_ops->go_flags & GLOF_ASPACE)
130 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
132 kmem_cache_free(gfs2_glock_cachep, gl);
136 * glock_blocked_by_withdraw - determine if we can still use a glock
139 * We need to allow some glocks to be enqueued, dequeued, promoted, and demoted
140 * when we're withdrawn. For example, to maintain metadata integrity, we should
141 * disallow the use of inode and rgrp glocks when withdrawn. Other glocks, like
142 * iopen or the transaction glocks may be safely used because none of their
143 * metadata goes through the journal. So in general, we should disallow all
144 * glocks that are journaled, and allow all the others. One exception is:
145 * we need to allow our active journal to be promoted and demoted so others
146 * may recover it and we can reacquire it when they're done.
148 static bool glock_blocked_by_withdraw(struct gfs2_glock *gl)
150 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
152 if (likely(!gfs2_withdrawn(sdp)))
154 if (gl->gl_ops->go_flags & GLOF_NONDISK)
156 if (!sdp->sd_jdesc ||
157 gl->gl_name.ln_number == sdp->sd_jdesc->jd_no_addr)
162 void gfs2_glock_free(struct gfs2_glock *gl)
164 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
166 gfs2_glock_assert_withdraw(gl, atomic_read(&gl->gl_revokes) == 0);
167 rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
170 call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
171 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
172 wake_up(&sdp->sd_glock_wait);
176 * gfs2_glock_hold() - increment reference count on glock
177 * @gl: The glock to hold
181 void gfs2_glock_hold(struct gfs2_glock *gl)
183 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
184 lockref_get(&gl->gl_lockref);
188 * demote_ok - Check to see if it's ok to unlock a glock
191 * Returns: 1 if it's ok
194 static int demote_ok(const struct gfs2_glock *gl)
196 const struct gfs2_glock_operations *glops = gl->gl_ops;
198 if (gl->gl_state == LM_ST_UNLOCKED)
200 if (!list_empty(&gl->gl_holders))
202 if (glops->go_demote_ok)
203 return glops->go_demote_ok(gl);
208 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
210 if (!(gl->gl_ops->go_flags & GLOF_LRU))
213 spin_lock(&lru_lock);
215 list_del(&gl->gl_lru);
216 list_add_tail(&gl->gl_lru, &lru_list);
218 if (!test_bit(GLF_LRU, &gl->gl_flags)) {
219 set_bit(GLF_LRU, &gl->gl_flags);
220 atomic_inc(&lru_count);
223 spin_unlock(&lru_lock);
226 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
228 if (!(gl->gl_ops->go_flags & GLOF_LRU))
231 spin_lock(&lru_lock);
232 if (test_bit(GLF_LRU, &gl->gl_flags)) {
233 list_del_init(&gl->gl_lru);
234 atomic_dec(&lru_count);
235 clear_bit(GLF_LRU, &gl->gl_flags);
237 spin_unlock(&lru_lock);
241 * Enqueue the glock on the work queue. Passes one glock reference on to the
244 static void __gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
245 if (!queue_delayed_work(glock_workqueue, &gl->gl_work, delay)) {
247 * We are holding the lockref spinlock, and the work was still
248 * queued above. The queued work (glock_work_func) takes that
249 * spinlock before dropping its glock reference(s), so it
250 * cannot have dropped them in the meantime.
252 GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2);
253 gl->gl_lockref.count--;
257 static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
258 spin_lock(&gl->gl_lockref.lock);
259 __gfs2_glock_queue_work(gl, delay);
260 spin_unlock(&gl->gl_lockref.lock);
263 static void __gfs2_glock_put(struct gfs2_glock *gl)
265 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
266 struct address_space *mapping = gfs2_glock2aspace(gl);
268 lockref_mark_dead(&gl->gl_lockref);
270 gfs2_glock_remove_from_lru(gl);
271 spin_unlock(&gl->gl_lockref.lock);
272 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
274 truncate_inode_pages_final(mapping);
275 if (!gfs2_withdrawn(sdp))
276 GLOCK_BUG_ON(gl, mapping->nrpages ||
277 mapping->nrexceptional);
279 trace_gfs2_glock_put(gl);
280 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
284 * Cause the glock to be put in work queue context.
286 void gfs2_glock_queue_put(struct gfs2_glock *gl)
288 gfs2_glock_queue_work(gl, 0);
292 * gfs2_glock_put() - Decrement reference count on glock
293 * @gl: The glock to put
297 void gfs2_glock_put(struct gfs2_glock *gl)
299 if (lockref_put_or_lock(&gl->gl_lockref))
302 __gfs2_glock_put(gl);
306 * may_grant - check if its ok to grant a new lock
308 * @gh: The lock request which we wish to grant
310 * Returns: true if its ok to grant the lock
313 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
315 const struct gfs2_holder *gh_head = list_first_entry(&gl->gl_holders, const struct gfs2_holder, gh_list);
319 * Here we make a special exception to grant holders who agree
320 * to share the EX lock with other holders who also have the
321 * bit set. If the original holder has the LM_FLAG_NODE_SCOPE bit
322 * is set, we grant more holders with the bit set.
324 if (gh_head->gh_state == LM_ST_EXCLUSIVE &&
325 (gh_head->gh_flags & LM_FLAG_NODE_SCOPE) &&
326 gh->gh_state == LM_ST_EXCLUSIVE &&
327 (gh->gh_flags & LM_FLAG_NODE_SCOPE))
329 if ((gh->gh_state == LM_ST_EXCLUSIVE ||
330 gh_head->gh_state == LM_ST_EXCLUSIVE))
333 if (gl->gl_state == gh->gh_state)
335 if (gh->gh_flags & GL_EXACT)
337 if (gl->gl_state == LM_ST_EXCLUSIVE) {
338 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
340 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
343 if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
348 static void gfs2_holder_wake(struct gfs2_holder *gh)
350 clear_bit(HIF_WAIT, &gh->gh_iflags);
351 smp_mb__after_atomic();
352 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
353 if (gh->gh_flags & GL_ASYNC) {
354 struct gfs2_sbd *sdp = gh->gh_gl->gl_name.ln_sbd;
356 wake_up(&sdp->sd_async_glock_wait);
361 * do_error - Something unexpected has happened during a lock request
365 static void do_error(struct gfs2_glock *gl, const int ret)
367 struct gfs2_holder *gh, *tmp;
369 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
370 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
372 if (ret & LM_OUT_ERROR)
374 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
375 gh->gh_error = GLR_TRYFAILED;
378 list_del_init(&gh->gh_list);
379 trace_gfs2_glock_queue(gh, 0);
380 gfs2_holder_wake(gh);
385 * do_promote - promote as many requests as possible on the current queue
388 * Returns: 1 if there is a blocked holder at the head of the list, or 2
389 * if a type specific operation is underway.
392 static int do_promote(struct gfs2_glock *gl)
393 __releases(&gl->gl_lockref.lock)
394 __acquires(&gl->gl_lockref.lock)
396 const struct gfs2_glock_operations *glops = gl->gl_ops;
397 struct gfs2_holder *gh, *tmp;
401 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
402 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
404 if (may_grant(gl, gh)) {
405 if (gh->gh_list.prev == &gl->gl_holders &&
407 spin_unlock(&gl->gl_lockref.lock);
408 /* FIXME: eliminate this eventually */
409 ret = glops->go_lock(gh);
410 spin_lock(&gl->gl_lockref.lock);
415 list_del_init(&gh->gh_list);
416 trace_gfs2_glock_queue(gh, 0);
417 gfs2_holder_wake(gh);
420 set_bit(HIF_HOLDER, &gh->gh_iflags);
421 trace_gfs2_promote(gh, 1);
422 gfs2_holder_wake(gh);
425 set_bit(HIF_HOLDER, &gh->gh_iflags);
426 trace_gfs2_promote(gh, 0);
427 gfs2_holder_wake(gh);
430 if (gh->gh_list.prev == &gl->gl_holders)
439 * find_first_waiter - find the first gh that's waiting for the glock
443 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
445 struct gfs2_holder *gh;
447 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
448 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
455 * state_change - record that the glock is now in a different state
457 * @new_state the new state
461 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
465 held1 = (gl->gl_state != LM_ST_UNLOCKED);
466 held2 = (new_state != LM_ST_UNLOCKED);
468 if (held1 != held2) {
469 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
471 gl->gl_lockref.count++;
473 gl->gl_lockref.count--;
475 if (new_state != gl->gl_target)
476 /* shorten our minimum hold time */
477 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
479 gl->gl_state = new_state;
480 gl->gl_tchange = jiffies;
483 static void gfs2_set_demote(struct gfs2_glock *gl)
485 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
487 set_bit(GLF_DEMOTE, &gl->gl_flags);
489 wake_up(&sdp->sd_async_glock_wait);
492 static void gfs2_demote_wake(struct gfs2_glock *gl)
494 gl->gl_demote_state = LM_ST_EXCLUSIVE;
495 clear_bit(GLF_DEMOTE, &gl->gl_flags);
496 smp_mb__after_atomic();
497 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
501 * finish_xmote - The DLM has replied to one of our lock requests
503 * @ret: The status from the DLM
507 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
509 const struct gfs2_glock_operations *glops = gl->gl_ops;
510 struct gfs2_holder *gh;
511 unsigned state = ret & LM_OUT_ST_MASK;
514 spin_lock(&gl->gl_lockref.lock);
515 trace_gfs2_glock_state_change(gl, state);
516 state_change(gl, state);
517 gh = find_first_waiter(gl);
519 /* Demote to UN request arrived during demote to SH or DF */
520 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
521 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
522 gl->gl_target = LM_ST_UNLOCKED;
524 /* Check for state != intended state */
525 if (unlikely(state != gl->gl_target)) {
526 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
527 /* move to back of queue and try next entry */
528 if (ret & LM_OUT_CANCELED) {
529 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
530 list_move_tail(&gh->gh_list, &gl->gl_holders);
531 gh = find_first_waiter(gl);
532 gl->gl_target = gh->gh_state;
535 /* Some error or failed "try lock" - report it */
536 if ((ret & LM_OUT_ERROR) ||
537 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
538 gl->gl_target = gl->gl_state;
544 /* Unlocked due to conversion deadlock, try again */
547 do_xmote(gl, gh, gl->gl_target);
549 /* Conversion fails, unlock and try again */
552 do_xmote(gl, gh, LM_ST_UNLOCKED);
554 default: /* Everything else */
555 fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n",
556 gl->gl_target, state);
559 spin_unlock(&gl->gl_lockref.lock);
563 /* Fast path - we got what we asked for */
564 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
565 gfs2_demote_wake(gl);
566 if (state != LM_ST_UNLOCKED) {
567 if (glops->go_xmote_bh) {
568 spin_unlock(&gl->gl_lockref.lock);
569 rv = glops->go_xmote_bh(gl, gh);
570 spin_lock(&gl->gl_lockref.lock);
581 clear_bit(GLF_LOCK, &gl->gl_flags);
583 spin_unlock(&gl->gl_lockref.lock);
587 * do_xmote - Calls the DLM to change the state of a lock
588 * @gl: The lock state
589 * @gh: The holder (only for promotes)
590 * @target: The target lock state
594 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
595 __releases(&gl->gl_lockref.lock)
596 __acquires(&gl->gl_lockref.lock)
598 const struct gfs2_glock_operations *glops = gl->gl_ops;
599 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
600 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
603 if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) &&
604 gh && !(gh->gh_flags & LM_FLAG_NOEXP))
606 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
608 GLOCK_BUG_ON(gl, gl->gl_state == target);
609 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
610 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
613 * If another process is already doing the invalidate, let that
614 * finish first. The glock state machine will get back to this
615 * holder again later.
617 if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
620 do_error(gl, 0); /* Fail queued try locks */
623 set_bit(GLF_BLOCKING, &gl->gl_flags);
624 if ((gl->gl_req == LM_ST_UNLOCKED) ||
625 (gl->gl_state == LM_ST_EXCLUSIVE) ||
626 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
627 clear_bit(GLF_BLOCKING, &gl->gl_flags);
628 spin_unlock(&gl->gl_lockref.lock);
629 if (glops->go_sync) {
630 ret = glops->go_sync(gl);
631 /* If we had a problem syncing (due to io errors or whatever,
632 * we should not invalidate the metadata or tell dlm to
633 * release the glock to other nodes.
636 if (cmpxchg(&sdp->sd_log_error, 0, ret)) {
637 fs_err(sdp, "Error %d syncing glock \n", ret);
638 gfs2_dump_glock(NULL, gl, true);
643 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) {
645 * The call to go_sync should have cleared out the ail list.
646 * If there are still items, we have a problem. We ought to
647 * withdraw, but we can't because the withdraw code also uses
648 * glocks. Warn about the error, dump the glock, then fall
649 * through and wait for logd to do the withdraw for us.
651 if ((atomic_read(&gl->gl_ail_count) != 0) &&
652 (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) {
653 gfs2_glock_assert_warn(gl,
654 !atomic_read(&gl->gl_ail_count));
655 gfs2_dump_glock(NULL, gl, true);
657 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
658 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
664 * Check for an error encountered since we called go_sync and go_inval.
665 * If so, we can't withdraw from the glock code because the withdraw
666 * code itself uses glocks (see function signal_our_withdraw) to
667 * change the mount to read-only. Most importantly, we must not call
668 * dlm to unlock the glock until the journal is in a known good state
669 * (after journal replay) otherwise other nodes may use the object
670 * (rgrp or dinode) and then later, journal replay will corrupt the
671 * file system. The best we can do here is wait for the logd daemon
672 * to see sd_log_error and withdraw, and in the meantime, requeue the
675 * However, if we're just unlocking the lock (say, for unmount, when
676 * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
677 * then it's okay to tell dlm to unlock it.
679 if (unlikely(sdp->sd_log_error && !gfs2_withdrawn(sdp)))
680 gfs2_withdraw_delayed(sdp);
681 if (glock_blocked_by_withdraw(gl)) {
682 if (target != LM_ST_UNLOCKED ||
683 test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags)) {
684 gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD);
689 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
691 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
692 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
693 target == LM_ST_UNLOCKED &&
694 test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
695 finish_xmote(gl, target);
696 gfs2_glock_queue_work(gl, 0);
698 fs_err(sdp, "lm_lock ret %d\n", ret);
699 GLOCK_BUG_ON(gl, !gfs2_withdrawn(sdp));
701 } else { /* lock_nolock */
702 finish_xmote(gl, target);
703 gfs2_glock_queue_work(gl, 0);
706 spin_lock(&gl->gl_lockref.lock);
710 * find_first_holder - find the first "holder" gh
714 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
716 struct gfs2_holder *gh;
718 if (!list_empty(&gl->gl_holders)) {
719 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
720 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
727 * run_queue - do all outstanding tasks related to a glock
728 * @gl: The glock in question
729 * @nonblock: True if we must not block in run_queue
733 static void run_queue(struct gfs2_glock *gl, const int nonblock)
734 __releases(&gl->gl_lockref.lock)
735 __acquires(&gl->gl_lockref.lock)
737 struct gfs2_holder *gh = NULL;
740 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
743 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
745 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
746 gl->gl_demote_state != gl->gl_state) {
747 if (find_first_holder(gl))
751 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
752 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
753 gl->gl_target = gl->gl_demote_state;
755 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
756 gfs2_demote_wake(gl);
757 ret = do_promote(gl);
762 gh = find_first_waiter(gl);
763 gl->gl_target = gh->gh_state;
764 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
765 do_error(gl, 0); /* Fail queued try locks */
767 do_xmote(gl, gh, gl->gl_target);
772 clear_bit(GLF_LOCK, &gl->gl_flags);
773 smp_mb__after_atomic();
774 gl->gl_lockref.count++;
775 __gfs2_glock_queue_work(gl, 0);
779 clear_bit(GLF_LOCK, &gl->gl_flags);
780 smp_mb__after_atomic();
784 void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation)
786 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
788 if (ri->ri_magic == 0)
789 ri->ri_magic = cpu_to_be32(GFS2_MAGIC);
790 if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC))
791 ri->ri_generation_deleted = cpu_to_be64(generation);
794 bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation)
796 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
798 if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC))
800 return generation <= be64_to_cpu(ri->ri_generation_deleted);
803 static void gfs2_glock_poke(struct gfs2_glock *gl)
805 int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP;
806 struct gfs2_holder gh;
809 gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh);
810 error = gfs2_glock_nq(&gh);
813 gfs2_holder_uninit(&gh);
816 static bool gfs2_try_evict(struct gfs2_glock *gl)
818 struct gfs2_inode *ip;
819 bool evicted = false;
822 * If there is contention on the iopen glock and we have an inode, try
823 * to grab and release the inode so that it can be evicted. This will
824 * allow the remote node to go ahead and delete the inode without us
825 * having to do it, which will avoid rgrp glock thrashing.
827 * The remote node is likely still holding the corresponding inode
828 * glock, so it will run before we get to verify that the delete has
831 spin_lock(&gl->gl_lockref.lock);
833 if (ip && !igrab(&ip->i_inode))
835 spin_unlock(&gl->gl_lockref.lock);
837 struct gfs2_glock *inode_gl = NULL;
839 gl->gl_no_formal_ino = ip->i_no_formal_ino;
840 set_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
841 d_prune_aliases(&ip->i_inode);
844 /* If the inode was evicted, gl->gl_object will now be NULL. */
845 spin_lock(&gl->gl_lockref.lock);
849 lockref_get(&inode_gl->gl_lockref);
850 clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
852 spin_unlock(&gl->gl_lockref.lock);
854 gfs2_glock_poke(inode_gl);
855 gfs2_glock_put(inode_gl);
862 static void delete_work_func(struct work_struct *work)
864 struct delayed_work *dwork = to_delayed_work(work);
865 struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete);
866 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
868 u64 no_addr = gl->gl_name.ln_number;
870 spin_lock(&gl->gl_lockref.lock);
871 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
872 spin_unlock(&gl->gl_lockref.lock);
874 if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
876 * If we can evict the inode, give the remote node trying to
877 * delete the inode some time before verifying that the delete
878 * has happened. Otherwise, if we cause contention on the inode glock
879 * immediately, the remote node will think that we still have
880 * the inode in use, and so it will give up waiting.
882 * If we can't evict the inode, signal to the remote node that
883 * the inode is still in use. We'll later try to delete the
884 * inode locally in gfs2_evict_inode.
886 * FIXME: We only need to verify that the remote node has
887 * deleted the inode because nodes before this remote delete
888 * rework won't cooperate. At a later time, when we no longer
889 * care about compatibility with such nodes, we can skip this
892 if (gfs2_try_evict(gl)) {
893 if (gfs2_queue_delete_work(gl, 5 * HZ))
899 inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino,
900 GFS2_BLKST_UNLINKED);
901 if (!IS_ERR_OR_NULL(inode)) {
902 d_prune_aliases(inode);
909 static void glock_work_func(struct work_struct *work)
911 unsigned long delay = 0;
912 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
913 unsigned int drop_refs = 1;
915 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
916 finish_xmote(gl, gl->gl_reply);
919 spin_lock(&gl->gl_lockref.lock);
920 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
921 gl->gl_state != LM_ST_UNLOCKED &&
922 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
923 unsigned long holdtime, now = jiffies;
925 holdtime = gl->gl_tchange + gl->gl_hold_time;
926 if (time_before(now, holdtime))
927 delay = holdtime - now;
930 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
936 /* Keep one glock reference for the work we requeue. */
938 if (gl->gl_name.ln_type != LM_TYPE_INODE)
940 __gfs2_glock_queue_work(gl, delay);
944 * Drop the remaining glock references manually here. (Mind that
945 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
948 gl->gl_lockref.count -= drop_refs;
949 if (!gl->gl_lockref.count) {
950 __gfs2_glock_put(gl);
953 spin_unlock(&gl->gl_lockref.lock);
956 static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
957 struct gfs2_glock *new)
959 struct wait_glock_queue wait;
960 wait_queue_head_t *wq = glock_waitqueue(name);
961 struct gfs2_glock *gl;
964 init_wait(&wait.wait);
965 wait.wait.func = glock_wake_function;
968 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
971 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
972 &new->gl_node, ht_parms);
976 gl = rhashtable_lookup_fast(&gl_hash_table,
979 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
986 finish_wait(wq, &wait.wait);
991 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
992 * @sdp: The GFS2 superblock
993 * @number: the lock number
994 * @glops: The glock_operations to use
995 * @create: If 0, don't create the glock if it doesn't exist
996 * @glp: the glock is returned here
998 * This does not lock a glock, just finds/creates structures for one.
1003 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
1004 const struct gfs2_glock_operations *glops, int create,
1005 struct gfs2_glock **glp)
1007 struct super_block *s = sdp->sd_vfs;
1008 struct lm_lockname name = { .ln_number = number,
1009 .ln_type = glops->go_type,
1011 struct gfs2_glock *gl, *tmp;
1012 struct address_space *mapping;
1013 struct kmem_cache *cachep;
1016 gl = find_insert_glock(&name, NULL);
1024 if (glops->go_flags & GLOF_ASPACE)
1025 cachep = gfs2_glock_aspace_cachep;
1027 cachep = gfs2_glock_cachep;
1028 gl = kmem_cache_alloc(cachep, GFP_NOFS);
1032 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
1034 if (glops->go_flags & GLOF_LVB) {
1035 gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1036 if (!gl->gl_lksb.sb_lvbptr) {
1037 kmem_cache_free(cachep, gl);
1042 atomic_inc(&sdp->sd_glock_disposal);
1043 gl->gl_node.next = NULL;
1046 lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass);
1047 gl->gl_lockref.count = 1;
1048 gl->gl_state = LM_ST_UNLOCKED;
1049 gl->gl_target = LM_ST_UNLOCKED;
1050 gl->gl_demote_state = LM_ST_EXCLUSIVE;
1054 /* We use the global stats to estimate the initial per-glock stats */
1055 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
1057 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
1058 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
1059 gl->gl_tchange = jiffies;
1060 gl->gl_object = NULL;
1061 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
1062 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
1063 if (gl->gl_name.ln_type == LM_TYPE_IOPEN)
1064 INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func);
1066 mapping = gfs2_glock2aspace(gl);
1068 mapping->a_ops = &gfs2_meta_aops;
1069 mapping->host = s->s_bdev->bd_inode;
1071 mapping_set_gfp_mask(mapping, GFP_NOFS);
1072 mapping->private_data = NULL;
1073 mapping->writeback_index = 0;
1076 tmp = find_insert_glock(&name, gl);
1088 kfree(gl->gl_lksb.sb_lvbptr);
1089 kmem_cache_free(cachep, gl);
1090 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
1091 wake_up(&sdp->sd_glock_wait);
1098 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
1100 * @state: the state we're requesting
1101 * @flags: the modifier flags
1102 * @gh: the holder structure
1106 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
1107 struct gfs2_holder *gh)
1109 INIT_LIST_HEAD(&gh->gh_list);
1111 gh->gh_ip = _RET_IP_;
1112 gh->gh_owner_pid = get_pid(task_pid(current));
1113 gh->gh_state = state;
1114 gh->gh_flags = flags;
1117 gfs2_glock_hold(gl);
1121 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1122 * @state: the state we're requesting
1123 * @flags: the modifier flags
1124 * @gh: the holder structure
1126 * Don't mess with the glock.
1130 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
1132 gh->gh_state = state;
1133 gh->gh_flags = flags;
1135 gh->gh_ip = _RET_IP_;
1136 put_pid(gh->gh_owner_pid);
1137 gh->gh_owner_pid = get_pid(task_pid(current));
1141 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1142 * @gh: the holder structure
1146 void gfs2_holder_uninit(struct gfs2_holder *gh)
1148 put_pid(gh->gh_owner_pid);
1149 gfs2_glock_put(gh->gh_gl);
1150 gfs2_holder_mark_uninitialized(gh);
1154 static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
1155 unsigned long start_time)
1157 /* Have we waited longer that a second? */
1158 if (time_after(jiffies, start_time + HZ)) {
1159 /* Lengthen the minimum hold time. */
1160 gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
1166 * gfs2_glock_wait - wait on a glock acquisition
1167 * @gh: the glock holder
1169 * Returns: 0 on success
1172 int gfs2_glock_wait(struct gfs2_holder *gh)
1174 unsigned long start_time = jiffies;
1177 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1178 gfs2_glock_update_hold_time(gh->gh_gl, start_time);
1179 return gh->gh_error;
1182 static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
1186 for (i = 0; i < num_gh; i++)
1187 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
1193 * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1194 * @num_gh: the number of holders in the array
1195 * @ghs: the glock holder array
1197 * Returns: 0 on success, meaning all glocks have been granted and are held.
1198 * -ESTALE if the request timed out, meaning all glocks were released,
1199 * and the caller should retry the operation.
1202 int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
1204 struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
1205 int i, ret = 0, timeout = 0;
1206 unsigned long start_time = jiffies;
1211 * Total up the (minimum hold time * 2) of all glocks and use that to
1212 * determine the max amount of time we should wait.
1214 for (i = 0; i < num_gh; i++)
1215 timeout += ghs[i].gh_gl->gl_hold_time << 1;
1218 if (!wait_event_timeout(sdp->sd_async_glock_wait,
1219 !glocks_pending(num_gh, ghs), timeout))
1220 ret = -ESTALE; /* request timed out. */
1223 * If dlm granted all our requests, we need to adjust the glock
1224 * minimum hold time values according to how long we waited.
1226 * If our request timed out, we need to repeatedly release any held
1227 * glocks we acquired thus far to allow dlm to acquire the remaining
1228 * glocks without deadlocking. We cannot currently cancel outstanding
1229 * glock acquisitions.
1231 * The HIF_WAIT bit tells us which requests still need a response from
1234 * If dlm sent us any errors, we return the first error we find.
1236 keep_waiting = false;
1237 for (i = 0; i < num_gh; i++) {
1238 /* Skip holders we have already dequeued below. */
1239 if (!gfs2_holder_queued(&ghs[i]))
1241 /* Skip holders with a pending DLM response. */
1242 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags)) {
1243 keep_waiting = true;
1247 if (test_bit(HIF_HOLDER, &ghs[i].gh_iflags)) {
1249 gfs2_glock_dq(&ghs[i]);
1251 gfs2_glock_update_hold_time(ghs[i].gh_gl,
1255 ret = ghs[i].gh_error;
1262 * At this point, we've either acquired all locks or released them all.
1268 * handle_callback - process a demote request
1270 * @state: the state the caller wants us to change to
1272 * There are only two requests that we are going to see in actual
1273 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1276 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
1277 unsigned long delay, bool remote)
1280 set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1282 gfs2_set_demote(gl);
1283 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
1284 gl->gl_demote_state = state;
1285 gl->gl_demote_time = jiffies;
1286 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
1287 gl->gl_demote_state != state) {
1288 gl->gl_demote_state = LM_ST_UNLOCKED;
1290 if (gl->gl_ops->go_callback)
1291 gl->gl_ops->go_callback(gl, remote);
1292 trace_gfs2_demote_rq(gl, remote);
1295 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
1297 struct va_format vaf;
1300 va_start(args, fmt);
1303 seq_vprintf(seq, fmt, args);
1308 pr_err("%pV", &vaf);
1315 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1316 * @gh: the holder structure to add
1318 * Eventually we should move the recursive locking trap to a
1319 * debugging option or something like that. This is the fast
1320 * path and needs to have the minimum number of distractions.
1324 static inline void add_to_queue(struct gfs2_holder *gh)
1325 __releases(&gl->gl_lockref.lock)
1326 __acquires(&gl->gl_lockref.lock)
1328 struct gfs2_glock *gl = gh->gh_gl;
1329 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1330 struct list_head *insert_pt = NULL;
1331 struct gfs2_holder *gh2;
1334 GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1335 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1336 GLOCK_BUG_ON(gl, true);
1338 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1339 if (test_bit(GLF_LOCK, &gl->gl_flags))
1340 try_futile = !may_grant(gl, gh);
1341 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1345 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1346 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
1347 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
1348 goto trap_recursive;
1350 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1352 gh->gh_error = GLR_TRYFAILED;
1353 gfs2_holder_wake(gh);
1356 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1358 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1359 insert_pt = &gh2->gh_list;
1361 trace_gfs2_glock_queue(gh, 1);
1362 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1363 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1364 if (likely(insert_pt == NULL)) {
1365 list_add_tail(&gh->gh_list, &gl->gl_holders);
1366 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1370 list_add_tail(&gh->gh_list, insert_pt);
1372 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
1373 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1374 spin_unlock(&gl->gl_lockref.lock);
1375 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1376 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1377 spin_lock(&gl->gl_lockref.lock);
1382 fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1383 fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1384 fs_err(sdp, "lock type: %d req lock state : %d\n",
1385 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1386 fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1387 fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1388 fs_err(sdp, "lock type: %d req lock state : %d\n",
1389 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1390 gfs2_dump_glock(NULL, gl, true);
1395 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1396 * @gh: the holder structure
1398 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1400 * Returns: 0, GLR_TRYFAILED, or errno on failure
1403 int gfs2_glock_nq(struct gfs2_holder *gh)
1405 struct gfs2_glock *gl = gh->gh_gl;
1408 if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP))
1411 if (test_bit(GLF_LRU, &gl->gl_flags))
1412 gfs2_glock_remove_from_lru(gl);
1414 spin_lock(&gl->gl_lockref.lock);
1416 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1417 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1418 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1419 gl->gl_lockref.count++;
1420 __gfs2_glock_queue_work(gl, 0);
1423 spin_unlock(&gl->gl_lockref.lock);
1425 if (!(gh->gh_flags & GL_ASYNC))
1426 error = gfs2_glock_wait(gh);
1432 * gfs2_glock_poll - poll to see if an async request has been completed
1435 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1438 int gfs2_glock_poll(struct gfs2_holder *gh)
1440 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1444 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1445 * @gh: the glock holder
1449 void gfs2_glock_dq(struct gfs2_holder *gh)
1451 struct gfs2_glock *gl = gh->gh_gl;
1452 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1456 spin_lock(&gl->gl_lockref.lock);
1458 * If we're in the process of file system withdraw, we cannot just
1459 * dequeue any glocks until our journal is recovered, lest we
1460 * introduce file system corruption. We need two exceptions to this
1461 * rule: We need to allow unlocking of nondisk glocks and the glock
1462 * for our own journal that needs recovery.
1464 if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) &&
1465 glock_blocked_by_withdraw(gl) &&
1466 gh->gh_gl != sdp->sd_jinode_gl) {
1467 sdp->sd_glock_dqs_held++;
1469 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
1470 TASK_UNINTERRUPTIBLE);
1472 if (gh->gh_flags & GL_NOCACHE)
1473 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1475 list_del_init(&gh->gh_list);
1476 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1477 if (find_first_holder(gl) == NULL) {
1478 if (list_empty(&gl->gl_holders) &&
1479 !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1480 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1483 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1484 gfs2_glock_add_to_lru(gl);
1486 trace_gfs2_glock_queue(gh, 0);
1487 if (unlikely(!fast_path)) {
1488 gl->gl_lockref.count++;
1489 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1490 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1491 gl->gl_name.ln_type == LM_TYPE_INODE)
1492 delay = gl->gl_hold_time;
1493 __gfs2_glock_queue_work(gl, delay);
1495 spin_unlock(&gl->gl_lockref.lock);
1498 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1500 struct gfs2_glock *gl = gh->gh_gl;
1503 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1507 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1508 * @gh: the holder structure
1512 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1515 gfs2_holder_uninit(gh);
1519 * gfs2_glock_nq_num - acquire a glock based on lock number
1520 * @sdp: the filesystem
1521 * @number: the lock number
1522 * @glops: the glock operations for the type of glock
1523 * @state: the state to acquire the glock in
1524 * @flags: modifier flags for the acquisition
1525 * @gh: the struct gfs2_holder
1530 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1531 const struct gfs2_glock_operations *glops,
1532 unsigned int state, u16 flags, struct gfs2_holder *gh)
1534 struct gfs2_glock *gl;
1537 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1539 error = gfs2_glock_nq_init(gl, state, flags, gh);
1547 * glock_compare - Compare two struct gfs2_glock structures for sorting
1548 * @arg_a: the first structure
1549 * @arg_b: the second structure
1553 static int glock_compare(const void *arg_a, const void *arg_b)
1555 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1556 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1557 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1558 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1560 if (a->ln_number > b->ln_number)
1562 if (a->ln_number < b->ln_number)
1564 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1569 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1570 * @num_gh: the number of structures
1571 * @ghs: an array of struct gfs2_holder structures
1573 * Returns: 0 on success (all glocks acquired),
1574 * errno on failure (no glocks acquired)
1577 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1578 struct gfs2_holder **p)
1583 for (x = 0; x < num_gh; x++)
1586 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1588 for (x = 0; x < num_gh; x++) {
1589 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1591 error = gfs2_glock_nq(p[x]);
1594 gfs2_glock_dq(p[x]);
1603 * gfs2_glock_nq_m - acquire multiple glocks
1604 * @num_gh: the number of structures
1605 * @ghs: an array of struct gfs2_holder structures
1608 * Returns: 0 on success (all glocks acquired),
1609 * errno on failure (no glocks acquired)
1612 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1614 struct gfs2_holder *tmp[4];
1615 struct gfs2_holder **pph = tmp;
1622 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1623 return gfs2_glock_nq(ghs);
1627 pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *),
1633 error = nq_m_sync(num_gh, ghs, pph);
1642 * gfs2_glock_dq_m - release multiple glocks
1643 * @num_gh: the number of structures
1644 * @ghs: an array of struct gfs2_holder structures
1648 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1651 gfs2_glock_dq(&ghs[num_gh]);
1654 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1656 unsigned long delay = 0;
1657 unsigned long holdtime;
1658 unsigned long now = jiffies;
1660 gfs2_glock_hold(gl);
1661 spin_lock(&gl->gl_lockref.lock);
1662 holdtime = gl->gl_tchange + gl->gl_hold_time;
1663 if (!list_empty(&gl->gl_holders) &&
1664 gl->gl_name.ln_type == LM_TYPE_INODE) {
1665 if (time_before(now, holdtime))
1666 delay = holdtime - now;
1667 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1668 delay = gl->gl_hold_time;
1670 handle_callback(gl, state, delay, true);
1671 __gfs2_glock_queue_work(gl, delay);
1672 spin_unlock(&gl->gl_lockref.lock);
1676 * gfs2_should_freeze - Figure out if glock should be frozen
1677 * @gl: The glock in question
1679 * Glocks are not frozen if (a) the result of the dlm operation is
1680 * an error, (b) the locking operation was an unlock operation or
1681 * (c) if there is a "noexp" flagged request anywhere in the queue
1683 * Returns: 1 if freezing should occur, 0 otherwise
1686 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1688 const struct gfs2_holder *gh;
1690 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1692 if (gl->gl_target == LM_ST_UNLOCKED)
1695 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1696 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1698 if (LM_FLAG_NOEXP & gh->gh_flags)
1706 * gfs2_glock_complete - Callback used by locking
1707 * @gl: Pointer to the glock
1708 * @ret: The return value from the dlm
1710 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1711 * to use a bitfield shared with other glock state fields.
1714 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1716 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1718 spin_lock(&gl->gl_lockref.lock);
1721 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1722 if (gfs2_should_freeze(gl)) {
1723 set_bit(GLF_FROZEN, &gl->gl_flags);
1724 spin_unlock(&gl->gl_lockref.lock);
1729 gl->gl_lockref.count++;
1730 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1731 __gfs2_glock_queue_work(gl, 0);
1732 spin_unlock(&gl->gl_lockref.lock);
1735 static int glock_cmp(void *priv, const struct list_head *a,
1736 const struct list_head *b)
1738 struct gfs2_glock *gla, *glb;
1740 gla = list_entry(a, struct gfs2_glock, gl_lru);
1741 glb = list_entry(b, struct gfs2_glock, gl_lru);
1743 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1745 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1752 * gfs2_dispose_glock_lru - Demote a list of glocks
1753 * @list: The list to dispose of
1755 * Disposing of glocks may involve disk accesses, so that here we sort
1756 * the glocks by number (i.e. disk location of the inodes) so that if
1757 * there are any such accesses, they'll be sent in order (mostly).
1759 * Must be called under the lru_lock, but may drop and retake this
1760 * lock. While the lru_lock is dropped, entries may vanish from the
1761 * list, but no new entries will appear on the list (since it is
1765 static void gfs2_dispose_glock_lru(struct list_head *list)
1766 __releases(&lru_lock)
1767 __acquires(&lru_lock)
1769 struct gfs2_glock *gl;
1771 list_sort(NULL, list, glock_cmp);
1773 while(!list_empty(list)) {
1774 gl = list_first_entry(list, struct gfs2_glock, gl_lru);
1775 list_del_init(&gl->gl_lru);
1776 if (!spin_trylock(&gl->gl_lockref.lock)) {
1778 list_add(&gl->gl_lru, &lru_list);
1779 set_bit(GLF_LRU, &gl->gl_flags);
1780 atomic_inc(&lru_count);
1783 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1784 spin_unlock(&gl->gl_lockref.lock);
1785 goto add_back_to_lru;
1787 gl->gl_lockref.count++;
1789 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1790 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1791 __gfs2_glock_queue_work(gl, 0);
1792 spin_unlock(&gl->gl_lockref.lock);
1793 cond_resched_lock(&lru_lock);
1798 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1799 * @nr: The number of entries to scan
1801 * This function selects the entries on the LRU which are able to
1802 * be demoted, and then kicks off the process by calling
1803 * gfs2_dispose_glock_lru() above.
1806 static long gfs2_scan_glock_lru(int nr)
1808 struct gfs2_glock *gl;
1813 spin_lock(&lru_lock);
1814 while ((nr-- >= 0) && !list_empty(&lru_list)) {
1815 gl = list_first_entry(&lru_list, struct gfs2_glock, gl_lru);
1817 /* Test for being demotable */
1818 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1819 list_move(&gl->gl_lru, &dispose);
1820 atomic_dec(&lru_count);
1821 clear_bit(GLF_LRU, &gl->gl_flags);
1826 list_move(&gl->gl_lru, &skipped);
1828 list_splice(&skipped, &lru_list);
1829 if (!list_empty(&dispose))
1830 gfs2_dispose_glock_lru(&dispose);
1831 spin_unlock(&lru_lock);
1836 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1837 struct shrink_control *sc)
1839 if (!(sc->gfp_mask & __GFP_FS))
1841 return gfs2_scan_glock_lru(sc->nr_to_scan);
1844 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1845 struct shrink_control *sc)
1847 return vfs_pressure_ratio(atomic_read(&lru_count));
1850 static struct shrinker glock_shrinker = {
1851 .seeks = DEFAULT_SEEKS,
1852 .count_objects = gfs2_glock_shrink_count,
1853 .scan_objects = gfs2_glock_shrink_scan,
1857 * glock_hash_walk - Call a function for glock in a hash bucket
1858 * @examiner: the function
1859 * @sdp: the filesystem
1861 * Note that the function can be called multiple times on the same
1862 * object. So the user must ensure that the function can cope with
1866 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1868 struct gfs2_glock *gl;
1869 struct rhashtable_iter iter;
1871 rhashtable_walk_enter(&gl_hash_table, &iter);
1874 rhashtable_walk_start(&iter);
1876 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl))
1877 if (gl->gl_name.ln_sbd == sdp &&
1878 lockref_get_not_dead(&gl->gl_lockref))
1881 rhashtable_walk_stop(&iter);
1882 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
1884 rhashtable_walk_exit(&iter);
1887 bool gfs2_queue_delete_work(struct gfs2_glock *gl, unsigned long delay)
1891 spin_lock(&gl->gl_lockref.lock);
1892 queued = queue_delayed_work(gfs2_delete_workqueue,
1893 &gl->gl_delete, delay);
1895 set_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1896 spin_unlock(&gl->gl_lockref.lock);
1900 void gfs2_cancel_delete_work(struct gfs2_glock *gl)
1902 if (cancel_delayed_work_sync(&gl->gl_delete)) {
1903 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1908 bool gfs2_delete_work_queued(const struct gfs2_glock *gl)
1910 return test_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1913 static void flush_delete_work(struct gfs2_glock *gl)
1915 if (gl->gl_name.ln_type == LM_TYPE_IOPEN) {
1916 if (cancel_delayed_work(&gl->gl_delete)) {
1917 queue_delayed_work(gfs2_delete_workqueue,
1921 gfs2_glock_queue_work(gl, 0);
1924 void gfs2_flush_delete_work(struct gfs2_sbd *sdp)
1926 glock_hash_walk(flush_delete_work, sdp);
1927 flush_workqueue(gfs2_delete_workqueue);
1931 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1932 * @gl: The glock to thaw
1936 static void thaw_glock(struct gfs2_glock *gl)
1938 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags)) {
1942 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1943 gfs2_glock_queue_work(gl, 0);
1947 * clear_glock - look at a glock and see if we can free it from glock cache
1948 * @gl: the glock to look at
1952 static void clear_glock(struct gfs2_glock *gl)
1954 gfs2_glock_remove_from_lru(gl);
1956 spin_lock(&gl->gl_lockref.lock);
1957 if (gl->gl_state != LM_ST_UNLOCKED)
1958 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1959 __gfs2_glock_queue_work(gl, 0);
1960 spin_unlock(&gl->gl_lockref.lock);
1964 * gfs2_glock_thaw - Thaw any frozen glocks
1965 * @sdp: The super block
1969 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1971 glock_hash_walk(thaw_glock, sdp);
1974 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
1976 spin_lock(&gl->gl_lockref.lock);
1977 gfs2_dump_glock(seq, gl, fsid);
1978 spin_unlock(&gl->gl_lockref.lock);
1981 static void dump_glock_func(struct gfs2_glock *gl)
1983 dump_glock(NULL, gl, true);
1987 * gfs2_gl_hash_clear - Empty out the glock hash table
1988 * @sdp: the filesystem
1989 * @wait: wait until it's all gone
1991 * Called when unmounting the filesystem.
1994 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1996 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1997 flush_workqueue(glock_workqueue);
1998 glock_hash_walk(clear_glock, sdp);
1999 flush_workqueue(glock_workqueue);
2000 wait_event_timeout(sdp->sd_glock_wait,
2001 atomic_read(&sdp->sd_glock_disposal) == 0,
2003 glock_hash_walk(dump_glock_func, sdp);
2006 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
2008 struct gfs2_glock *gl = ip->i_gl;
2011 ret = gfs2_truncatei_resume(ip);
2012 gfs2_glock_assert_withdraw(gl, ret == 0);
2014 spin_lock(&gl->gl_lockref.lock);
2015 clear_bit(GLF_LOCK, &gl->gl_flags);
2017 spin_unlock(&gl->gl_lockref.lock);
2020 static const char *state2str(unsigned state)
2023 case LM_ST_UNLOCKED:
2027 case LM_ST_DEFERRED:
2029 case LM_ST_EXCLUSIVE:
2035 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
2038 if (flags & LM_FLAG_TRY)
2040 if (flags & LM_FLAG_TRY_1CB)
2042 if (flags & LM_FLAG_NOEXP)
2044 if (flags & LM_FLAG_ANY)
2046 if (flags & LM_FLAG_PRIORITY)
2048 if (flags & LM_FLAG_NODE_SCOPE)
2050 if (flags & GL_ASYNC)
2052 if (flags & GL_EXACT)
2054 if (flags & GL_NOCACHE)
2056 if (test_bit(HIF_HOLDER, &iflags))
2058 if (test_bit(HIF_WAIT, &iflags))
2060 if (test_bit(HIF_FIRST, &iflags))
2067 * dump_holder - print information about a glock holder
2068 * @seq: the seq_file struct
2069 * @gh: the glock holder
2070 * @fs_id_buf: pointer to file system id (if requested)
2074 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
2075 const char *fs_id_buf)
2077 struct task_struct *gh_owner = NULL;
2081 if (gh->gh_owner_pid)
2082 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
2083 gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2084 fs_id_buf, state2str(gh->gh_state),
2085 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
2087 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
2088 gh_owner ? gh_owner->comm : "(ended)",
2093 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
2095 const unsigned long *gflags = &gl->gl_flags;
2098 if (test_bit(GLF_LOCK, gflags))
2100 if (test_bit(GLF_DEMOTE, gflags))
2102 if (test_bit(GLF_PENDING_DEMOTE, gflags))
2104 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
2106 if (test_bit(GLF_DIRTY, gflags))
2108 if (test_bit(GLF_LFLUSH, gflags))
2110 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
2112 if (test_bit(GLF_REPLY_PENDING, gflags))
2114 if (test_bit(GLF_INITIAL, gflags))
2116 if (test_bit(GLF_FROZEN, gflags))
2118 if (!list_empty(&gl->gl_holders))
2120 if (test_bit(GLF_LRU, gflags))
2124 if (test_bit(GLF_BLOCKING, gflags))
2126 if (test_bit(GLF_PENDING_DELETE, gflags))
2128 if (test_bit(GLF_FREEING, gflags))
2135 * gfs2_dump_glock - print information about a glock
2136 * @seq: The seq_file struct
2138 * @fsid: If true, also dump the file system id
2140 * The file format is as follows:
2141 * One line per object, capital letters are used to indicate objects
2142 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2143 * other objects are indented by a single space and follow the glock to
2144 * which they are related. Fields are indicated by lower case letters
2145 * followed by a colon and the field value, except for strings which are in
2146 * [] so that its possible to see if they are composed of spaces for
2147 * example. The field's are n = number (id of the object), f = flags,
2148 * t = type, s = state, r = refcount, e = error, p = pid.
2152 void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2154 const struct gfs2_glock_operations *glops = gl->gl_ops;
2155 unsigned long long dtime;
2156 const struct gfs2_holder *gh;
2157 char gflags_buf[32];
2158 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2159 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2160 unsigned long nrpages = 0;
2162 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
2163 struct address_space *mapping = gfs2_glock2aspace(gl);
2165 nrpages = mapping->nrpages;
2167 memset(fs_id_buf, 0, sizeof(fs_id_buf));
2168 if (fsid && sdp) /* safety precaution */
2169 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2170 dtime = jiffies - gl->gl_demote_time;
2171 dtime *= 1000000/HZ; /* demote time in uSec */
2172 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
2174 gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2175 "v:%d r:%d m:%ld p:%lu\n",
2176 fs_id_buf, state2str(gl->gl_state),
2177 gl->gl_name.ln_type,
2178 (unsigned long long)gl->gl_name.ln_number,
2179 gflags2str(gflags_buf, gl),
2180 state2str(gl->gl_target),
2181 state2str(gl->gl_demote_state), dtime,
2182 atomic_read(&gl->gl_ail_count),
2183 atomic_read(&gl->gl_revokes),
2184 (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages);
2186 list_for_each_entry(gh, &gl->gl_holders, gh_list)
2187 dump_holder(seq, gh, fs_id_buf);
2189 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
2190 glops->go_dump(seq, gl, fs_id_buf);
2193 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
2195 struct gfs2_glock *gl = iter_ptr;
2197 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2198 gl->gl_name.ln_type,
2199 (unsigned long long)gl->gl_name.ln_number,
2200 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
2201 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
2202 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
2203 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
2204 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
2205 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
2206 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
2207 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
2211 static const char *gfs2_gltype[] = {
2225 static const char *gfs2_stype[] = {
2226 [GFS2_LKS_SRTT] = "srtt",
2227 [GFS2_LKS_SRTTVAR] = "srttvar",
2228 [GFS2_LKS_SRTTB] = "srttb",
2229 [GFS2_LKS_SRTTVARB] = "srttvarb",
2230 [GFS2_LKS_SIRT] = "sirt",
2231 [GFS2_LKS_SIRTVAR] = "sirtvar",
2232 [GFS2_LKS_DCOUNT] = "dlm",
2233 [GFS2_LKS_QCOUNT] = "queue",
2236 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2238 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
2240 struct gfs2_sbd *sdp = seq->private;
2241 loff_t pos = *(loff_t *)iter_ptr;
2242 unsigned index = pos >> 3;
2243 unsigned subindex = pos & 0x07;
2246 if (index == 0 && subindex != 0)
2249 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
2250 (index == 0) ? "cpu": gfs2_stype[subindex]);
2252 for_each_possible_cpu(i) {
2253 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
2256 seq_printf(seq, " %15u", i);
2258 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
2259 lkstats[index - 1].stats[subindex]);
2261 seq_putc(seq, '\n');
2265 int __init gfs2_glock_init(void)
2269 ret = rhashtable_init(&gl_hash_table, &ht_parms);
2273 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
2274 WQ_HIGHPRI | WQ_FREEZABLE, 0);
2275 if (!glock_workqueue) {
2276 rhashtable_destroy(&gl_hash_table);
2279 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
2280 WQ_MEM_RECLAIM | WQ_FREEZABLE,
2282 if (!gfs2_delete_workqueue) {
2283 destroy_workqueue(glock_workqueue);
2284 rhashtable_destroy(&gl_hash_table);
2288 ret = register_shrinker(&glock_shrinker);
2290 destroy_workqueue(gfs2_delete_workqueue);
2291 destroy_workqueue(glock_workqueue);
2292 rhashtable_destroy(&gl_hash_table);
2296 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2297 init_waitqueue_head(glock_wait_table + i);
2302 void gfs2_glock_exit(void)
2304 unregister_shrinker(&glock_shrinker);
2305 rhashtable_destroy(&gl_hash_table);
2306 destroy_workqueue(glock_workqueue);
2307 destroy_workqueue(gfs2_delete_workqueue);
2310 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2312 struct gfs2_glock *gl = gi->gl;
2317 if (!lockref_put_not_zero(&gl->gl_lockref))
2318 gfs2_glock_queue_put(gl);
2321 gl = rhashtable_walk_next(&gi->hti);
2322 if (IS_ERR_OR_NULL(gl)) {
2323 if (gl == ERR_PTR(-EAGAIN)) {
2330 if (gl->gl_name.ln_sbd != gi->sdp)
2333 if (!lockref_get_not_dead(&gl->gl_lockref))
2337 if (__lockref_is_dead(&gl->gl_lockref))
2345 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2348 struct gfs2_glock_iter *gi = seq->private;
2352 * We can either stay where we are, skip to the next hash table
2353 * entry, or start from the beginning.
2355 if (*pos < gi->last_pos) {
2356 rhashtable_walk_exit(&gi->hti);
2357 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2360 n = *pos - gi->last_pos;
2363 rhashtable_walk_start(&gi->hti);
2365 gfs2_glock_iter_next(gi, n);
2366 gi->last_pos = *pos;
2370 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2373 struct gfs2_glock_iter *gi = seq->private;
2376 gi->last_pos = *pos;
2377 gfs2_glock_iter_next(gi, 1);
2381 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2384 struct gfs2_glock_iter *gi = seq->private;
2386 rhashtable_walk_stop(&gi->hti);
2389 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2391 dump_glock(seq, iter_ptr, false);
2395 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2398 if (*pos >= GFS2_NR_SBSTATS)
2403 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2407 if (*pos >= GFS2_NR_SBSTATS)
2412 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2417 static const struct seq_operations gfs2_glock_seq_ops = {
2418 .start = gfs2_glock_seq_start,
2419 .next = gfs2_glock_seq_next,
2420 .stop = gfs2_glock_seq_stop,
2421 .show = gfs2_glock_seq_show,
2424 static const struct seq_operations gfs2_glstats_seq_ops = {
2425 .start = gfs2_glock_seq_start,
2426 .next = gfs2_glock_seq_next,
2427 .stop = gfs2_glock_seq_stop,
2428 .show = gfs2_glstats_seq_show,
2431 static const struct seq_operations gfs2_sbstats_sops = {
2432 .start = gfs2_sbstats_seq_start,
2433 .next = gfs2_sbstats_seq_next,
2434 .stop = gfs2_sbstats_seq_stop,
2435 .show = gfs2_sbstats_seq_show,
2438 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2440 static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2441 const struct seq_operations *ops)
2443 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2445 struct seq_file *seq = file->private_data;
2446 struct gfs2_glock_iter *gi = seq->private;
2448 gi->sdp = inode->i_private;
2449 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2451 seq->size = GFS2_SEQ_GOODSIZE;
2453 * Initially, we are "before" the first hash table entry; the
2454 * first call to rhashtable_walk_next gets us the first entry.
2458 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2463 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2465 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2468 static int gfs2_glocks_release(struct inode *inode, struct file *file)
2470 struct seq_file *seq = file->private_data;
2471 struct gfs2_glock_iter *gi = seq->private;
2474 gfs2_glock_put(gi->gl);
2475 rhashtable_walk_exit(&gi->hti);
2476 return seq_release_private(inode, file);
2479 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2481 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2484 static const struct file_operations gfs2_glocks_fops = {
2485 .owner = THIS_MODULE,
2486 .open = gfs2_glocks_open,
2488 .llseek = seq_lseek,
2489 .release = gfs2_glocks_release,
2492 static const struct file_operations gfs2_glstats_fops = {
2493 .owner = THIS_MODULE,
2494 .open = gfs2_glstats_open,
2496 .llseek = seq_lseek,
2497 .release = gfs2_glocks_release,
2500 DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats);
2502 void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2504 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2506 debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2509 debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2510 &gfs2_glstats_fops);
2512 debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2513 &gfs2_sbstats_fops);
2516 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2518 debugfs_remove_recursive(sdp->debugfs_dir);
2519 sdp->debugfs_dir = NULL;
2522 void gfs2_register_debugfs(void)
2524 gfs2_root = debugfs_create_dir("gfs2", NULL);
2527 void gfs2_unregister_debugfs(void)
2529 debugfs_remove(gfs2_root);