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);
61 static void __gfs2_glock_dq(struct gfs2_holder *gh);
63 static struct dentry *gfs2_root;
64 static struct workqueue_struct *glock_workqueue;
65 struct workqueue_struct *gfs2_delete_workqueue;
66 static LIST_HEAD(lru_list);
67 static atomic_t lru_count = ATOMIC_INIT(0);
68 static DEFINE_SPINLOCK(lru_lock);
70 #define GFS2_GL_HASH_SHIFT 15
71 #define GFS2_GL_HASH_SIZE BIT(GFS2_GL_HASH_SHIFT)
73 static const struct rhashtable_params ht_parms = {
74 .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4,
75 .key_len = offsetofend(struct lm_lockname, ln_type),
76 .key_offset = offsetof(struct gfs2_glock, gl_name),
77 .head_offset = offsetof(struct gfs2_glock, gl_node),
80 static struct rhashtable gl_hash_table;
82 #define GLOCK_WAIT_TABLE_BITS 12
83 #define GLOCK_WAIT_TABLE_SIZE (1 << GLOCK_WAIT_TABLE_BITS)
84 static wait_queue_head_t glock_wait_table[GLOCK_WAIT_TABLE_SIZE] __cacheline_aligned;
86 struct wait_glock_queue {
87 struct lm_lockname *name;
88 wait_queue_entry_t wait;
91 static int glock_wake_function(wait_queue_entry_t *wait, unsigned int mode,
94 struct wait_glock_queue *wait_glock =
95 container_of(wait, struct wait_glock_queue, wait);
96 struct lm_lockname *wait_name = wait_glock->name;
97 struct lm_lockname *wake_name = key;
99 if (wake_name->ln_sbd != wait_name->ln_sbd ||
100 wake_name->ln_number != wait_name->ln_number ||
101 wake_name->ln_type != wait_name->ln_type)
103 return autoremove_wake_function(wait, mode, sync, key);
106 static wait_queue_head_t *glock_waitqueue(struct lm_lockname *name)
108 u32 hash = jhash2((u32 *)name, ht_parms.key_len / 4, 0);
110 return glock_wait_table + hash_32(hash, GLOCK_WAIT_TABLE_BITS);
114 * wake_up_glock - Wake up waiters on a glock
117 static void wake_up_glock(struct gfs2_glock *gl)
119 wait_queue_head_t *wq = glock_waitqueue(&gl->gl_name);
121 if (waitqueue_active(wq))
122 __wake_up(wq, TASK_NORMAL, 1, &gl->gl_name);
125 static void gfs2_glock_dealloc(struct rcu_head *rcu)
127 struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
129 kfree(gl->gl_lksb.sb_lvbptr);
130 if (gl->gl_ops->go_flags & GLOF_ASPACE)
131 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
133 kmem_cache_free(gfs2_glock_cachep, gl);
137 * glock_blocked_by_withdraw - determine if we can still use a glock
140 * We need to allow some glocks to be enqueued, dequeued, promoted, and demoted
141 * when we're withdrawn. For example, to maintain metadata integrity, we should
142 * disallow the use of inode and rgrp glocks when withdrawn. Other glocks, like
143 * iopen or the transaction glocks may be safely used because none of their
144 * metadata goes through the journal. So in general, we should disallow all
145 * glocks that are journaled, and allow all the others. One exception is:
146 * we need to allow our active journal to be promoted and demoted so others
147 * may recover it and we can reacquire it when they're done.
149 static bool glock_blocked_by_withdraw(struct gfs2_glock *gl)
151 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
153 if (likely(!gfs2_withdrawn(sdp)))
155 if (gl->gl_ops->go_flags & GLOF_NONDISK)
157 if (!sdp->sd_jdesc ||
158 gl->gl_name.ln_number == sdp->sd_jdesc->jd_no_addr)
163 void gfs2_glock_free(struct gfs2_glock *gl)
165 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
167 gfs2_glock_assert_withdraw(gl, atomic_read(&gl->gl_revokes) == 0);
168 rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
171 call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
172 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
173 wake_up(&sdp->sd_glock_wait);
177 * gfs2_glock_hold() - increment reference count on glock
178 * @gl: The glock to hold
182 void gfs2_glock_hold(struct gfs2_glock *gl)
184 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
185 lockref_get(&gl->gl_lockref);
189 * demote_ok - Check to see if it's ok to unlock a glock
192 * Returns: 1 if it's ok
195 static int demote_ok(const struct gfs2_glock *gl)
197 const struct gfs2_glock_operations *glops = gl->gl_ops;
199 if (gl->gl_state == LM_ST_UNLOCKED)
202 * Note that demote_ok is used for the lru process of disposing of
203 * glocks. For this purpose, we don't care if the glock's holders
204 * have the HIF_MAY_DEMOTE flag set or not. If someone is using
205 * them, don't demote.
207 if (!list_empty(&gl->gl_holders))
209 if (glops->go_demote_ok)
210 return glops->go_demote_ok(gl);
215 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
217 if (!(gl->gl_ops->go_flags & GLOF_LRU))
220 spin_lock(&lru_lock);
222 list_move_tail(&gl->gl_lru, &lru_list);
224 if (!test_bit(GLF_LRU, &gl->gl_flags)) {
225 set_bit(GLF_LRU, &gl->gl_flags);
226 atomic_inc(&lru_count);
229 spin_unlock(&lru_lock);
232 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
234 if (!(gl->gl_ops->go_flags & GLOF_LRU))
237 spin_lock(&lru_lock);
238 if (test_bit(GLF_LRU, &gl->gl_flags)) {
239 list_del_init(&gl->gl_lru);
240 atomic_dec(&lru_count);
241 clear_bit(GLF_LRU, &gl->gl_flags);
243 spin_unlock(&lru_lock);
247 * Enqueue the glock on the work queue. Passes one glock reference on to the
250 static void __gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
251 if (!queue_delayed_work(glock_workqueue, &gl->gl_work, delay)) {
253 * We are holding the lockref spinlock, and the work was still
254 * queued above. The queued work (glock_work_func) takes that
255 * spinlock before dropping its glock reference(s), so it
256 * cannot have dropped them in the meantime.
258 GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2);
259 gl->gl_lockref.count--;
263 static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
264 spin_lock(&gl->gl_lockref.lock);
265 __gfs2_glock_queue_work(gl, delay);
266 spin_unlock(&gl->gl_lockref.lock);
269 static void __gfs2_glock_put(struct gfs2_glock *gl)
271 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
272 struct address_space *mapping = gfs2_glock2aspace(gl);
274 lockref_mark_dead(&gl->gl_lockref);
276 gfs2_glock_remove_from_lru(gl);
277 spin_unlock(&gl->gl_lockref.lock);
278 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
280 truncate_inode_pages_final(mapping);
281 if (!gfs2_withdrawn(sdp))
282 GLOCK_BUG_ON(gl, !mapping_empty(mapping));
284 trace_gfs2_glock_put(gl);
285 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
289 * Cause the glock to be put in work queue context.
291 void gfs2_glock_queue_put(struct gfs2_glock *gl)
293 gfs2_glock_queue_work(gl, 0);
297 * gfs2_glock_put() - Decrement reference count on glock
298 * @gl: The glock to put
302 void gfs2_glock_put(struct gfs2_glock *gl)
304 if (lockref_put_or_lock(&gl->gl_lockref))
307 __gfs2_glock_put(gl);
311 * may_grant - check if it's ok to grant a new lock
313 * @current_gh: One of the current holders of @gl
314 * @gh: The lock request which we wish to grant
316 * With our current compatibility rules, if a glock has one or more active
317 * holders (HIF_HOLDER flag set), any of those holders can be passed in as
318 * @current_gh; they are all the same as far as compatibility with the new @gh
321 * Returns true if it's ok to grant the lock.
324 static inline bool may_grant(struct gfs2_glock *gl,
325 struct gfs2_holder *current_gh,
326 struct gfs2_holder *gh)
329 GLOCK_BUG_ON(gl, !test_bit(HIF_HOLDER, ¤t_gh->gh_iflags));
331 switch(current_gh->gh_state) {
332 case LM_ST_EXCLUSIVE:
334 * Here we make a special exception to grant holders
335 * who agree to share the EX lock with other holders
336 * who also have the bit set. If the original holder
337 * has the LM_FLAG_NODE_SCOPE bit set, we grant more
338 * holders with the bit set.
340 return gh->gh_state == LM_ST_EXCLUSIVE &&
341 (current_gh->gh_flags & LM_FLAG_NODE_SCOPE) &&
342 (gh->gh_flags & LM_FLAG_NODE_SCOPE);
346 return gh->gh_state == current_gh->gh_state;
353 if (gl->gl_state == gh->gh_state)
355 if (gh->gh_flags & GL_EXACT)
357 if (gl->gl_state == LM_ST_EXCLUSIVE) {
358 return gh->gh_state == LM_ST_SHARED ||
359 gh->gh_state == LM_ST_DEFERRED;
361 if (gh->gh_flags & LM_FLAG_ANY)
362 return gl->gl_state != LM_ST_UNLOCKED;
366 static void gfs2_holder_wake(struct gfs2_holder *gh)
368 clear_bit(HIF_WAIT, &gh->gh_iflags);
369 smp_mb__after_atomic();
370 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
371 if (gh->gh_flags & GL_ASYNC) {
372 struct gfs2_sbd *sdp = gh->gh_gl->gl_name.ln_sbd;
374 wake_up(&sdp->sd_async_glock_wait);
379 * do_error - Something unexpected has happened during a lock request
381 * @ret: The status from the DLM
384 static void do_error(struct gfs2_glock *gl, const int ret)
386 struct gfs2_holder *gh, *tmp;
388 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
389 if (!test_bit(HIF_WAIT, &gh->gh_iflags))
391 if (ret & LM_OUT_ERROR)
393 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
394 gh->gh_error = GLR_TRYFAILED;
397 list_del_init(&gh->gh_list);
398 trace_gfs2_glock_queue(gh, 0);
399 gfs2_holder_wake(gh);
404 * demote_incompat_holders - demote incompatible demoteable holders
405 * @gl: the glock we want to promote
406 * @new_gh: the new holder to be promoted
408 static void demote_incompat_holders(struct gfs2_glock *gl,
409 struct gfs2_holder *new_gh)
411 struct gfs2_holder *gh, *tmp;
414 * Demote incompatible holders before we make ourselves eligible.
415 * (This holder may or may not allow auto-demoting, but we don't want
416 * to demote the new holder before it's even granted.)
418 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
420 * Since holders are at the front of the list, we stop when we
421 * find the first non-holder.
423 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
425 if (test_bit(HIF_MAY_DEMOTE, &gh->gh_iflags) &&
426 !may_grant(gl, new_gh, gh)) {
428 * We should not recurse into do_promote because
429 * __gfs2_glock_dq only calls handle_callback,
430 * gfs2_glock_add_to_lru and __gfs2_glock_queue_work.
438 * find_first_holder - find the first "holder" gh
442 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
444 struct gfs2_holder *gh;
446 if (!list_empty(&gl->gl_holders)) {
447 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder,
449 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
456 * find_first_strong_holder - find the first non-demoteable holder
459 * Find the first holder that doesn't have the HIF_MAY_DEMOTE flag set.
461 static inline struct gfs2_holder *
462 find_first_strong_holder(struct gfs2_glock *gl)
464 struct gfs2_holder *gh;
466 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
467 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
469 if (!test_bit(HIF_MAY_DEMOTE, &gh->gh_iflags))
476 * gfs2_instantiate - Call the glops instantiate function
477 * @gh: The glock holder
479 * Returns: 0 if instantiate was successful, 2 if type specific operation is
480 * underway, or error.
482 int gfs2_instantiate(struct gfs2_holder *gh)
484 struct gfs2_glock *gl = gh->gh_gl;
485 const struct gfs2_glock_operations *glops = gl->gl_ops;
489 if (!test_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags))
493 * Since we unlock the lockref lock, we set a flag to indicate
494 * instantiate is in progress.
496 if (test_and_set_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags)) {
497 wait_on_bit(&gl->gl_flags, GLF_INSTANTIATE_IN_PROG,
498 TASK_UNINTERRUPTIBLE);
500 * Here we just waited for a different instantiate to finish.
501 * But that may not have been successful, as when a process
502 * locks an inode glock _before_ it has an actual inode to
503 * instantiate into. So we check again. This process might
504 * have an inode to instantiate, so might be successful.
509 ret = glops->go_instantiate(gh);
511 clear_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags);
512 clear_and_wake_up_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags);
517 * do_promote - promote as many requests as possible on the current queue
520 * Returns: 1 if there is a blocked holder at the head of the list, or 2
521 * if a type specific operation is underway.
524 static int do_promote(struct gfs2_glock *gl)
525 __releases(&gl->gl_lockref.lock)
526 __acquires(&gl->gl_lockref.lock)
528 struct gfs2_holder *gh, *tmp, *first_gh;
529 bool incompat_holders_demoted = false;
534 first_gh = find_first_strong_holder(gl);
535 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
536 lock_released = false;
537 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
539 if (!may_grant(gl, first_gh, gh)) {
541 * If we get here, it means we may not grant this holder for
542 * some reason. If this holder is the head of the list, it
543 * means we have a blocked holder at the head, so return 1.
545 if (gh->gh_list.prev == &gl->gl_holders)
550 if (!incompat_holders_demoted) {
551 demote_incompat_holders(gl, first_gh);
552 incompat_holders_demoted = true;
555 if (test_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags) &&
556 !(gh->gh_flags & GL_SKIP) && gl->gl_ops->go_instantiate) {
557 lock_released = true;
558 spin_unlock(&gl->gl_lockref.lock);
559 ret = gfs2_instantiate(gh);
560 spin_lock(&gl->gl_lockref.lock);
565 list_del_init(&gh->gh_list);
566 trace_gfs2_glock_queue(gh, 0);
567 gfs2_holder_wake(gh);
571 set_bit(HIF_HOLDER, &gh->gh_iflags);
572 trace_gfs2_promote(gh);
573 gfs2_holder_wake(gh);
575 * If we released the gl_lockref.lock the holders list may have
576 * changed. For that reason, we start again at the start of
586 * find_first_waiter - find the first gh that's waiting for the glock
590 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
592 struct gfs2_holder *gh;
594 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
595 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
602 * state_change - record that the glock is now in a different state
604 * @new_state: the new state
607 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
611 held1 = (gl->gl_state != LM_ST_UNLOCKED);
612 held2 = (new_state != LM_ST_UNLOCKED);
614 if (held1 != held2) {
615 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
617 gl->gl_lockref.count++;
619 gl->gl_lockref.count--;
621 if (new_state != gl->gl_target)
622 /* shorten our minimum hold time */
623 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
625 gl->gl_state = new_state;
626 gl->gl_tchange = jiffies;
629 static void gfs2_set_demote(struct gfs2_glock *gl)
631 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
633 set_bit(GLF_DEMOTE, &gl->gl_flags);
635 wake_up(&sdp->sd_async_glock_wait);
638 static void gfs2_demote_wake(struct gfs2_glock *gl)
640 gl->gl_demote_state = LM_ST_EXCLUSIVE;
641 clear_bit(GLF_DEMOTE, &gl->gl_flags);
642 smp_mb__after_atomic();
643 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
647 * finish_xmote - The DLM has replied to one of our lock requests
649 * @ret: The status from the DLM
653 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
655 const struct gfs2_glock_operations *glops = gl->gl_ops;
656 struct gfs2_holder *gh;
657 unsigned state = ret & LM_OUT_ST_MASK;
660 spin_lock(&gl->gl_lockref.lock);
661 trace_gfs2_glock_state_change(gl, state);
662 state_change(gl, state);
663 gh = find_first_waiter(gl);
665 /* Demote to UN request arrived during demote to SH or DF */
666 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
667 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
668 gl->gl_target = LM_ST_UNLOCKED;
670 /* Check for state != intended state */
671 if (unlikely(state != gl->gl_target)) {
672 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
673 /* move to back of queue and try next entry */
674 if (ret & LM_OUT_CANCELED) {
675 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
676 list_move_tail(&gh->gh_list, &gl->gl_holders);
677 gh = find_first_waiter(gl);
678 gl->gl_target = gh->gh_state;
681 /* Some error or failed "try lock" - report it */
682 if ((ret & LM_OUT_ERROR) ||
683 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
684 gl->gl_target = gl->gl_state;
690 /* Unlocked due to conversion deadlock, try again */
693 do_xmote(gl, gh, gl->gl_target);
695 /* Conversion fails, unlock and try again */
698 do_xmote(gl, gh, LM_ST_UNLOCKED);
700 default: /* Everything else */
701 fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n",
702 gl->gl_target, state);
705 spin_unlock(&gl->gl_lockref.lock);
709 /* Fast path - we got what we asked for */
710 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
711 gfs2_demote_wake(gl);
712 if (state != LM_ST_UNLOCKED) {
713 if (glops->go_xmote_bh) {
714 spin_unlock(&gl->gl_lockref.lock);
715 rv = glops->go_xmote_bh(gl);
716 spin_lock(&gl->gl_lockref.lock);
727 clear_bit(GLF_LOCK, &gl->gl_flags);
729 spin_unlock(&gl->gl_lockref.lock);
732 static bool is_system_glock(struct gfs2_glock *gl)
734 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
735 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
737 if (gl == m_ip->i_gl)
743 * do_xmote - Calls the DLM to change the state of a lock
744 * @gl: The lock state
745 * @gh: The holder (only for promotes)
746 * @target: The target lock state
750 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
751 __releases(&gl->gl_lockref.lock)
752 __acquires(&gl->gl_lockref.lock)
754 const struct gfs2_glock_operations *glops = gl->gl_ops;
755 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
756 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
759 if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) &&
760 gh && !(gh->gh_flags & LM_FLAG_NOEXP))
762 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
764 GLOCK_BUG_ON(gl, gl->gl_state == target);
765 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
766 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
769 * If another process is already doing the invalidate, let that
770 * finish first. The glock state machine will get back to this
771 * holder again later.
773 if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
776 do_error(gl, 0); /* Fail queued try locks */
779 set_bit(GLF_BLOCKING, &gl->gl_flags);
780 if ((gl->gl_req == LM_ST_UNLOCKED) ||
781 (gl->gl_state == LM_ST_EXCLUSIVE) ||
782 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
783 clear_bit(GLF_BLOCKING, &gl->gl_flags);
784 spin_unlock(&gl->gl_lockref.lock);
785 if (glops->go_sync) {
786 ret = glops->go_sync(gl);
787 /* If we had a problem syncing (due to io errors or whatever,
788 * we should not invalidate the metadata or tell dlm to
789 * release the glock to other nodes.
792 if (cmpxchg(&sdp->sd_log_error, 0, ret)) {
793 fs_err(sdp, "Error %d syncing glock \n", ret);
794 gfs2_dump_glock(NULL, gl, true);
799 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) {
801 * The call to go_sync should have cleared out the ail list.
802 * If there are still items, we have a problem. We ought to
803 * withdraw, but we can't because the withdraw code also uses
804 * glocks. Warn about the error, dump the glock, then fall
805 * through and wait for logd to do the withdraw for us.
807 if ((atomic_read(&gl->gl_ail_count) != 0) &&
808 (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) {
809 gfs2_glock_assert_warn(gl,
810 !atomic_read(&gl->gl_ail_count));
811 gfs2_dump_glock(NULL, gl, true);
813 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
814 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
820 * Check for an error encountered since we called go_sync and go_inval.
821 * If so, we can't withdraw from the glock code because the withdraw
822 * code itself uses glocks (see function signal_our_withdraw) to
823 * change the mount to read-only. Most importantly, we must not call
824 * dlm to unlock the glock until the journal is in a known good state
825 * (after journal replay) otherwise other nodes may use the object
826 * (rgrp or dinode) and then later, journal replay will corrupt the
827 * file system. The best we can do here is wait for the logd daemon
828 * to see sd_log_error and withdraw, and in the meantime, requeue the
831 * We make a special exception for some system glocks, such as the
832 * system statfs inode glock, which needs to be granted before the
833 * gfs2_quotad daemon can exit, and that exit needs to finish before
834 * we can unmount the withdrawn file system.
836 * However, if we're just unlocking the lock (say, for unmount, when
837 * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
838 * then it's okay to tell dlm to unlock it.
840 if (unlikely(sdp->sd_log_error && !gfs2_withdrawn(sdp)))
841 gfs2_withdraw_delayed(sdp);
842 if (glock_blocked_by_withdraw(gl) &&
843 (target != LM_ST_UNLOCKED ||
844 test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags))) {
845 if (!is_system_glock(gl)) {
846 gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD);
849 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
853 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
855 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
856 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
857 target == LM_ST_UNLOCKED &&
858 test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
859 finish_xmote(gl, target);
860 gfs2_glock_queue_work(gl, 0);
862 fs_err(sdp, "lm_lock ret %d\n", ret);
863 GLOCK_BUG_ON(gl, !gfs2_withdrawn(sdp));
865 } else { /* lock_nolock */
866 finish_xmote(gl, target);
867 gfs2_glock_queue_work(gl, 0);
870 spin_lock(&gl->gl_lockref.lock);
874 * run_queue - do all outstanding tasks related to a glock
875 * @gl: The glock in question
876 * @nonblock: True if we must not block in run_queue
880 static void run_queue(struct gfs2_glock *gl, const int nonblock)
881 __releases(&gl->gl_lockref.lock)
882 __acquires(&gl->gl_lockref.lock)
884 struct gfs2_holder *gh = NULL;
887 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
890 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
892 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
893 gl->gl_demote_state != gl->gl_state) {
894 if (find_first_holder(gl))
898 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
899 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
900 gl->gl_target = gl->gl_demote_state;
902 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
903 gfs2_demote_wake(gl);
904 ret = do_promote(gl);
909 gh = find_first_waiter(gl);
910 gl->gl_target = gh->gh_state;
911 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
912 do_error(gl, 0); /* Fail queued try locks */
914 do_xmote(gl, gh, gl->gl_target);
919 clear_bit(GLF_LOCK, &gl->gl_flags);
920 smp_mb__after_atomic();
921 gl->gl_lockref.count++;
922 __gfs2_glock_queue_work(gl, 0);
926 clear_bit(GLF_LOCK, &gl->gl_flags);
927 smp_mb__after_atomic();
931 void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation)
933 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
935 if (ri->ri_magic == 0)
936 ri->ri_magic = cpu_to_be32(GFS2_MAGIC);
937 if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC))
938 ri->ri_generation_deleted = cpu_to_be64(generation);
941 bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation)
943 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
945 if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC))
947 return generation <= be64_to_cpu(ri->ri_generation_deleted);
950 static void gfs2_glock_poke(struct gfs2_glock *gl)
952 int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP;
953 struct gfs2_holder gh;
956 __gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh, _RET_IP_);
957 error = gfs2_glock_nq(&gh);
960 gfs2_holder_uninit(&gh);
963 static bool gfs2_try_evict(struct gfs2_glock *gl)
965 struct gfs2_inode *ip;
966 bool evicted = false;
969 * If there is contention on the iopen glock and we have an inode, try
970 * to grab and release the inode so that it can be evicted. This will
971 * allow the remote node to go ahead and delete the inode without us
972 * having to do it, which will avoid rgrp glock thrashing.
974 * The remote node is likely still holding the corresponding inode
975 * glock, so it will run before we get to verify that the delete has
978 spin_lock(&gl->gl_lockref.lock);
980 if (ip && !igrab(&ip->i_inode))
982 spin_unlock(&gl->gl_lockref.lock);
984 struct gfs2_glock *inode_gl = NULL;
986 gl->gl_no_formal_ino = ip->i_no_formal_ino;
987 set_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
988 d_prune_aliases(&ip->i_inode);
991 /* If the inode was evicted, gl->gl_object will now be NULL. */
992 spin_lock(&gl->gl_lockref.lock);
996 lockref_get(&inode_gl->gl_lockref);
997 clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
999 spin_unlock(&gl->gl_lockref.lock);
1001 gfs2_glock_poke(inode_gl);
1002 gfs2_glock_put(inode_gl);
1009 static void delete_work_func(struct work_struct *work)
1011 struct delayed_work *dwork = to_delayed_work(work);
1012 struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete);
1013 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1014 struct inode *inode;
1015 u64 no_addr = gl->gl_name.ln_number;
1017 spin_lock(&gl->gl_lockref.lock);
1018 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1019 spin_unlock(&gl->gl_lockref.lock);
1021 if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
1023 * If we can evict the inode, give the remote node trying to
1024 * delete the inode some time before verifying that the delete
1025 * has happened. Otherwise, if we cause contention on the inode glock
1026 * immediately, the remote node will think that we still have
1027 * the inode in use, and so it will give up waiting.
1029 * If we can't evict the inode, signal to the remote node that
1030 * the inode is still in use. We'll later try to delete the
1031 * inode locally in gfs2_evict_inode.
1033 * FIXME: We only need to verify that the remote node has
1034 * deleted the inode because nodes before this remote delete
1035 * rework won't cooperate. At a later time, when we no longer
1036 * care about compatibility with such nodes, we can skip this
1039 if (gfs2_try_evict(gl)) {
1040 if (gfs2_queue_delete_work(gl, 5 * HZ))
1046 inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino,
1047 GFS2_BLKST_UNLINKED);
1048 if (!IS_ERR_OR_NULL(inode)) {
1049 d_prune_aliases(inode);
1056 static void glock_work_func(struct work_struct *work)
1058 unsigned long delay = 0;
1059 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
1060 unsigned int drop_refs = 1;
1062 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
1063 finish_xmote(gl, gl->gl_reply);
1066 spin_lock(&gl->gl_lockref.lock);
1067 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1068 gl->gl_state != LM_ST_UNLOCKED &&
1069 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
1070 unsigned long holdtime, now = jiffies;
1072 holdtime = gl->gl_tchange + gl->gl_hold_time;
1073 if (time_before(now, holdtime))
1074 delay = holdtime - now;
1077 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1078 gfs2_set_demote(gl);
1083 /* Keep one glock reference for the work we requeue. */
1085 if (gl->gl_name.ln_type != LM_TYPE_INODE)
1087 __gfs2_glock_queue_work(gl, delay);
1091 * Drop the remaining glock references manually here. (Mind that
1092 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
1095 gl->gl_lockref.count -= drop_refs;
1096 if (!gl->gl_lockref.count) {
1097 __gfs2_glock_put(gl);
1100 spin_unlock(&gl->gl_lockref.lock);
1103 static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
1104 struct gfs2_glock *new)
1106 struct wait_glock_queue wait;
1107 wait_queue_head_t *wq = glock_waitqueue(name);
1108 struct gfs2_glock *gl;
1111 init_wait(&wait.wait);
1112 wait.wait.func = glock_wake_function;
1115 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1118 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
1119 &new->gl_node, ht_parms);
1123 gl = rhashtable_lookup_fast(&gl_hash_table,
1126 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
1133 finish_wait(wq, &wait.wait);
1138 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
1139 * @sdp: The GFS2 superblock
1140 * @number: the lock number
1141 * @glops: The glock_operations to use
1142 * @create: If 0, don't create the glock if it doesn't exist
1143 * @glp: the glock is returned here
1145 * This does not lock a glock, just finds/creates structures for one.
1150 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
1151 const struct gfs2_glock_operations *glops, int create,
1152 struct gfs2_glock **glp)
1154 struct super_block *s = sdp->sd_vfs;
1155 struct lm_lockname name = { .ln_number = number,
1156 .ln_type = glops->go_type,
1158 struct gfs2_glock *gl, *tmp;
1159 struct address_space *mapping;
1160 struct kmem_cache *cachep;
1163 gl = find_insert_glock(&name, NULL);
1171 if (glops->go_flags & GLOF_ASPACE)
1172 cachep = gfs2_glock_aspace_cachep;
1174 cachep = gfs2_glock_cachep;
1175 gl = kmem_cache_alloc(cachep, GFP_NOFS);
1179 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
1181 if (glops->go_flags & GLOF_LVB) {
1182 gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1183 if (!gl->gl_lksb.sb_lvbptr) {
1184 kmem_cache_free(cachep, gl);
1189 atomic_inc(&sdp->sd_glock_disposal);
1190 gl->gl_node.next = NULL;
1191 gl->gl_flags = glops->go_instantiate ? BIT(GLF_INSTANTIATE_NEEDED) : 0;
1193 lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass);
1194 gl->gl_lockref.count = 1;
1195 gl->gl_state = LM_ST_UNLOCKED;
1196 gl->gl_target = LM_ST_UNLOCKED;
1197 gl->gl_demote_state = LM_ST_EXCLUSIVE;
1201 /* We use the global stats to estimate the initial per-glock stats */
1202 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
1204 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
1205 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
1206 gl->gl_tchange = jiffies;
1207 gl->gl_object = NULL;
1208 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
1209 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
1210 if (gl->gl_name.ln_type == LM_TYPE_IOPEN)
1211 INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func);
1213 mapping = gfs2_glock2aspace(gl);
1215 mapping->a_ops = &gfs2_meta_aops;
1216 mapping->host = s->s_bdev->bd_inode;
1218 mapping_set_gfp_mask(mapping, GFP_NOFS);
1219 mapping->private_data = NULL;
1220 mapping->writeback_index = 0;
1223 tmp = find_insert_glock(&name, gl);
1235 kfree(gl->gl_lksb.sb_lvbptr);
1236 kmem_cache_free(cachep, gl);
1237 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
1238 wake_up(&sdp->sd_glock_wait);
1245 * __gfs2_holder_init - initialize a struct gfs2_holder in the default way
1247 * @state: the state we're requesting
1248 * @flags: the modifier flags
1249 * @gh: the holder structure
1253 void __gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
1254 struct gfs2_holder *gh, unsigned long ip)
1256 INIT_LIST_HEAD(&gh->gh_list);
1259 gh->gh_owner_pid = get_pid(task_pid(current));
1260 gh->gh_state = state;
1261 gh->gh_flags = flags;
1264 gfs2_glock_hold(gl);
1268 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1269 * @state: the state we're requesting
1270 * @flags: the modifier flags
1271 * @gh: the holder structure
1273 * Don't mess with the glock.
1277 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
1279 gh->gh_state = state;
1280 gh->gh_flags = flags;
1282 gh->gh_ip = _RET_IP_;
1283 put_pid(gh->gh_owner_pid);
1284 gh->gh_owner_pid = get_pid(task_pid(current));
1288 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1289 * @gh: the holder structure
1293 void gfs2_holder_uninit(struct gfs2_holder *gh)
1295 put_pid(gh->gh_owner_pid);
1296 gfs2_glock_put(gh->gh_gl);
1297 gfs2_holder_mark_uninitialized(gh);
1301 static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
1302 unsigned long start_time)
1304 /* Have we waited longer that a second? */
1305 if (time_after(jiffies, start_time + HZ)) {
1306 /* Lengthen the minimum hold time. */
1307 gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
1313 * gfs2_glock_wait - wait on a glock acquisition
1314 * @gh: the glock holder
1316 * Returns: 0 on success
1319 int gfs2_glock_wait(struct gfs2_holder *gh)
1321 unsigned long start_time = jiffies;
1324 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1325 gfs2_glock_update_hold_time(gh->gh_gl, start_time);
1326 return gh->gh_error;
1329 static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
1333 for (i = 0; i < num_gh; i++)
1334 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
1340 * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1341 * @num_gh: the number of holders in the array
1342 * @ghs: the glock holder array
1344 * Returns: 0 on success, meaning all glocks have been granted and are held.
1345 * -ESTALE if the request timed out, meaning all glocks were released,
1346 * and the caller should retry the operation.
1349 int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
1351 struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
1352 int i, ret = 0, timeout = 0;
1353 unsigned long start_time = jiffies;
1358 * Total up the (minimum hold time * 2) of all glocks and use that to
1359 * determine the max amount of time we should wait.
1361 for (i = 0; i < num_gh; i++)
1362 timeout += ghs[i].gh_gl->gl_hold_time << 1;
1365 if (!wait_event_timeout(sdp->sd_async_glock_wait,
1366 !glocks_pending(num_gh, ghs), timeout))
1367 ret = -ESTALE; /* request timed out. */
1370 * If dlm granted all our requests, we need to adjust the glock
1371 * minimum hold time values according to how long we waited.
1373 * If our request timed out, we need to repeatedly release any held
1374 * glocks we acquired thus far to allow dlm to acquire the remaining
1375 * glocks without deadlocking. We cannot currently cancel outstanding
1376 * glock acquisitions.
1378 * The HIF_WAIT bit tells us which requests still need a response from
1381 * If dlm sent us any errors, we return the first error we find.
1383 keep_waiting = false;
1384 for (i = 0; i < num_gh; i++) {
1385 /* Skip holders we have already dequeued below. */
1386 if (!gfs2_holder_queued(&ghs[i]))
1388 /* Skip holders with a pending DLM response. */
1389 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags)) {
1390 keep_waiting = true;
1394 if (test_bit(HIF_HOLDER, &ghs[i].gh_iflags)) {
1396 gfs2_glock_dq(&ghs[i]);
1398 gfs2_glock_update_hold_time(ghs[i].gh_gl,
1402 ret = ghs[i].gh_error;
1409 * At this point, we've either acquired all locks or released them all.
1415 * handle_callback - process a demote request
1417 * @state: the state the caller wants us to change to
1418 * @delay: zero to demote immediately; otherwise pending demote
1419 * @remote: true if this came from a different cluster node
1421 * There are only two requests that we are going to see in actual
1422 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1425 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
1426 unsigned long delay, bool remote)
1429 set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1431 gfs2_set_demote(gl);
1432 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
1433 gl->gl_demote_state = state;
1434 gl->gl_demote_time = jiffies;
1435 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
1436 gl->gl_demote_state != state) {
1437 gl->gl_demote_state = LM_ST_UNLOCKED;
1439 if (gl->gl_ops->go_callback)
1440 gl->gl_ops->go_callback(gl, remote);
1441 trace_gfs2_demote_rq(gl, remote);
1444 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
1446 struct va_format vaf;
1449 va_start(args, fmt);
1452 seq_vprintf(seq, fmt, args);
1457 pr_err("%pV", &vaf);
1464 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1465 * @gh: the holder structure to add
1467 * Eventually we should move the recursive locking trap to a
1468 * debugging option or something like that. This is the fast
1469 * path and needs to have the minimum number of distractions.
1473 static inline void add_to_queue(struct gfs2_holder *gh)
1474 __releases(&gl->gl_lockref.lock)
1475 __acquires(&gl->gl_lockref.lock)
1477 struct gfs2_glock *gl = gh->gh_gl;
1478 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1479 struct list_head *insert_pt = NULL;
1480 struct gfs2_holder *gh2;
1483 GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1484 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1485 GLOCK_BUG_ON(gl, true);
1487 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1488 if (test_bit(GLF_LOCK, &gl->gl_flags)) {
1489 struct gfs2_holder *first_gh;
1491 first_gh = find_first_strong_holder(gl);
1492 try_futile = !may_grant(gl, first_gh, gh);
1494 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1498 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1499 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
1500 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK) &&
1501 !test_bit(HIF_MAY_DEMOTE, &gh2->gh_iflags)))
1502 goto trap_recursive;
1504 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1506 gh->gh_error = GLR_TRYFAILED;
1507 gfs2_holder_wake(gh);
1510 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1512 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1513 insert_pt = &gh2->gh_list;
1515 trace_gfs2_glock_queue(gh, 1);
1516 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1517 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1518 if (likely(insert_pt == NULL)) {
1519 list_add_tail(&gh->gh_list, &gl->gl_holders);
1520 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1524 list_add_tail(&gh->gh_list, insert_pt);
1526 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
1527 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1528 spin_unlock(&gl->gl_lockref.lock);
1529 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1530 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1531 spin_lock(&gl->gl_lockref.lock);
1536 fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1537 fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1538 fs_err(sdp, "lock type: %d req lock state : %d\n",
1539 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1540 fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1541 fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1542 fs_err(sdp, "lock type: %d req lock state : %d\n",
1543 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1544 gfs2_dump_glock(NULL, gl, true);
1549 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1550 * @gh: the holder structure
1552 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1554 * Returns: 0, GLR_TRYFAILED, or errno on failure
1557 int gfs2_glock_nq(struct gfs2_holder *gh)
1559 struct gfs2_glock *gl = gh->gh_gl;
1562 if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP))
1565 if (test_bit(GLF_LRU, &gl->gl_flags))
1566 gfs2_glock_remove_from_lru(gl);
1568 spin_lock(&gl->gl_lockref.lock);
1570 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1571 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1572 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1573 gl->gl_lockref.count++;
1574 __gfs2_glock_queue_work(gl, 0);
1577 spin_unlock(&gl->gl_lockref.lock);
1579 if (!(gh->gh_flags & GL_ASYNC))
1580 error = gfs2_glock_wait(gh);
1586 * gfs2_glock_poll - poll to see if an async request has been completed
1589 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1592 int gfs2_glock_poll(struct gfs2_holder *gh)
1594 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1597 static inline bool needs_demote(struct gfs2_glock *gl)
1599 return (test_bit(GLF_DEMOTE, &gl->gl_flags) ||
1600 test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags));
1603 static void __gfs2_glock_dq(struct gfs2_holder *gh)
1605 struct gfs2_glock *gl = gh->gh_gl;
1606 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1611 * This while loop is similar to function demote_incompat_holders:
1612 * If the glock is due to be demoted (which may be from another node
1613 * or even if this holder is GL_NOCACHE), the weak holders are
1614 * demoted as well, allowing the glock to be demoted.
1618 * If we're in the process of file system withdraw, we cannot
1619 * just dequeue any glocks until our journal is recovered, lest
1620 * we introduce file system corruption. We need two exceptions
1621 * to this rule: We need to allow unlocking of nondisk glocks
1622 * and the glock for our own journal that needs recovery.
1624 if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) &&
1625 glock_blocked_by_withdraw(gl) &&
1626 gh->gh_gl != sdp->sd_jinode_gl) {
1627 sdp->sd_glock_dqs_held++;
1628 spin_unlock(&gl->gl_lockref.lock);
1630 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
1631 TASK_UNINTERRUPTIBLE);
1632 spin_lock(&gl->gl_lockref.lock);
1636 * This holder should not be cached, so mark it for demote.
1637 * Note: this should be done before the check for needs_demote
1640 if (gh->gh_flags & GL_NOCACHE)
1641 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1643 list_del_init(&gh->gh_list);
1644 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1645 trace_gfs2_glock_queue(gh, 0);
1648 * If there hasn't been a demote request we are done.
1649 * (Let the remaining holders, if any, keep holding it.)
1651 if (!needs_demote(gl)) {
1652 if (list_empty(&gl->gl_holders))
1657 * If we have another strong holder (we cannot auto-demote)
1658 * we are done. It keeps holding it until it is done.
1660 if (find_first_strong_holder(gl))
1664 * If we have a weak holder at the head of the list, it
1665 * (and all others like it) must be auto-demoted. If there
1666 * are no more weak holders, we exit the while loop.
1668 gh = find_first_holder(gl);
1671 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1672 gfs2_glock_add_to_lru(gl);
1674 if (unlikely(!fast_path)) {
1675 gl->gl_lockref.count++;
1676 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1677 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1678 gl->gl_name.ln_type == LM_TYPE_INODE)
1679 delay = gl->gl_hold_time;
1680 __gfs2_glock_queue_work(gl, delay);
1685 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1686 * @gh: the glock holder
1689 void gfs2_glock_dq(struct gfs2_holder *gh)
1691 struct gfs2_glock *gl = gh->gh_gl;
1693 spin_lock(&gl->gl_lockref.lock);
1694 __gfs2_glock_dq(gh);
1695 spin_unlock(&gl->gl_lockref.lock);
1698 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1700 struct gfs2_glock *gl = gh->gh_gl;
1703 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1707 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1708 * @gh: the holder structure
1712 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1715 gfs2_holder_uninit(gh);
1719 * gfs2_glock_nq_num - acquire a glock based on lock number
1720 * @sdp: the filesystem
1721 * @number: the lock number
1722 * @glops: the glock operations for the type of glock
1723 * @state: the state to acquire the glock in
1724 * @flags: modifier flags for the acquisition
1725 * @gh: the struct gfs2_holder
1730 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1731 const struct gfs2_glock_operations *glops,
1732 unsigned int state, u16 flags, struct gfs2_holder *gh)
1734 struct gfs2_glock *gl;
1737 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1739 error = gfs2_glock_nq_init(gl, state, flags, gh);
1747 * glock_compare - Compare two struct gfs2_glock structures for sorting
1748 * @arg_a: the first structure
1749 * @arg_b: the second structure
1753 static int glock_compare(const void *arg_a, const void *arg_b)
1755 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1756 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1757 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1758 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1760 if (a->ln_number > b->ln_number)
1762 if (a->ln_number < b->ln_number)
1764 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1769 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1770 * @num_gh: the number of structures
1771 * @ghs: an array of struct gfs2_holder structures
1772 * @p: placeholder for the holder structure to pass back
1774 * Returns: 0 on success (all glocks acquired),
1775 * errno on failure (no glocks acquired)
1778 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1779 struct gfs2_holder **p)
1784 for (x = 0; x < num_gh; x++)
1787 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1789 for (x = 0; x < num_gh; x++) {
1790 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1792 error = gfs2_glock_nq(p[x]);
1795 gfs2_glock_dq(p[x]);
1804 * gfs2_glock_nq_m - acquire multiple glocks
1805 * @num_gh: the number of structures
1806 * @ghs: an array of struct gfs2_holder structures
1809 * Returns: 0 on success (all glocks acquired),
1810 * errno on failure (no glocks acquired)
1813 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1815 struct gfs2_holder *tmp[4];
1816 struct gfs2_holder **pph = tmp;
1823 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1824 return gfs2_glock_nq(ghs);
1828 pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *),
1834 error = nq_m_sync(num_gh, ghs, pph);
1843 * gfs2_glock_dq_m - release multiple glocks
1844 * @num_gh: the number of structures
1845 * @ghs: an array of struct gfs2_holder structures
1849 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1852 gfs2_glock_dq(&ghs[num_gh]);
1855 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1857 unsigned long delay = 0;
1858 unsigned long holdtime;
1859 unsigned long now = jiffies;
1861 gfs2_glock_hold(gl);
1862 spin_lock(&gl->gl_lockref.lock);
1863 holdtime = gl->gl_tchange + gl->gl_hold_time;
1864 if (!list_empty(&gl->gl_holders) &&
1865 gl->gl_name.ln_type == LM_TYPE_INODE) {
1866 if (time_before(now, holdtime))
1867 delay = holdtime - now;
1868 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1869 delay = gl->gl_hold_time;
1872 * Note 1: We cannot call demote_incompat_holders from handle_callback
1873 * or gfs2_set_demote due to recursion problems like: gfs2_glock_dq ->
1874 * handle_callback -> demote_incompat_holders -> gfs2_glock_dq
1875 * Plus, we only want to demote the holders if the request comes from
1876 * a remote cluster node because local holder conflicts are resolved
1879 * Note 2: if a remote node wants this glock in EX mode, lock_dlm will
1880 * request that we set our state to UNLOCKED. Here we mock up a holder
1881 * to make it look like someone wants the lock EX locally. Any SH
1882 * and DF requests should be able to share the lock without demoting.
1884 * Note 3: We only want to demote the demoteable holders when there
1885 * are no more strong holders. The demoteable holders might as well
1886 * keep the glock until the last strong holder is done with it.
1888 if (!find_first_strong_holder(gl)) {
1889 struct gfs2_holder mock_gh = {
1891 .gh_state = (state == LM_ST_UNLOCKED) ?
1892 LM_ST_EXCLUSIVE : state,
1893 .gh_iflags = BIT(HIF_HOLDER)
1896 demote_incompat_holders(gl, &mock_gh);
1898 handle_callback(gl, state, delay, true);
1899 __gfs2_glock_queue_work(gl, delay);
1900 spin_unlock(&gl->gl_lockref.lock);
1904 * gfs2_should_freeze - Figure out if glock should be frozen
1905 * @gl: The glock in question
1907 * Glocks are not frozen if (a) the result of the dlm operation is
1908 * an error, (b) the locking operation was an unlock operation or
1909 * (c) if there is a "noexp" flagged request anywhere in the queue
1911 * Returns: 1 if freezing should occur, 0 otherwise
1914 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1916 const struct gfs2_holder *gh;
1918 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1920 if (gl->gl_target == LM_ST_UNLOCKED)
1923 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1924 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1926 if (LM_FLAG_NOEXP & gh->gh_flags)
1934 * gfs2_glock_complete - Callback used by locking
1935 * @gl: Pointer to the glock
1936 * @ret: The return value from the dlm
1938 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1939 * to use a bitfield shared with other glock state fields.
1942 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1944 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1946 spin_lock(&gl->gl_lockref.lock);
1949 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1950 if (gfs2_should_freeze(gl)) {
1951 set_bit(GLF_FROZEN, &gl->gl_flags);
1952 spin_unlock(&gl->gl_lockref.lock);
1957 gl->gl_lockref.count++;
1958 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1959 __gfs2_glock_queue_work(gl, 0);
1960 spin_unlock(&gl->gl_lockref.lock);
1963 static int glock_cmp(void *priv, const struct list_head *a,
1964 const struct list_head *b)
1966 struct gfs2_glock *gla, *glb;
1968 gla = list_entry(a, struct gfs2_glock, gl_lru);
1969 glb = list_entry(b, struct gfs2_glock, gl_lru);
1971 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1973 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1980 * gfs2_dispose_glock_lru - Demote a list of glocks
1981 * @list: The list to dispose of
1983 * Disposing of glocks may involve disk accesses, so that here we sort
1984 * the glocks by number (i.e. disk location of the inodes) so that if
1985 * there are any such accesses, they'll be sent in order (mostly).
1987 * Must be called under the lru_lock, but may drop and retake this
1988 * lock. While the lru_lock is dropped, entries may vanish from the
1989 * list, but no new entries will appear on the list (since it is
1993 static void gfs2_dispose_glock_lru(struct list_head *list)
1994 __releases(&lru_lock)
1995 __acquires(&lru_lock)
1997 struct gfs2_glock *gl;
1999 list_sort(NULL, list, glock_cmp);
2001 while(!list_empty(list)) {
2002 gl = list_first_entry(list, struct gfs2_glock, gl_lru);
2003 list_del_init(&gl->gl_lru);
2004 clear_bit(GLF_LRU, &gl->gl_flags);
2005 if (!spin_trylock(&gl->gl_lockref.lock)) {
2007 list_add(&gl->gl_lru, &lru_list);
2008 set_bit(GLF_LRU, &gl->gl_flags);
2009 atomic_inc(&lru_count);
2012 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
2013 spin_unlock(&gl->gl_lockref.lock);
2014 goto add_back_to_lru;
2016 gl->gl_lockref.count++;
2018 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2019 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
2020 __gfs2_glock_queue_work(gl, 0);
2021 spin_unlock(&gl->gl_lockref.lock);
2022 cond_resched_lock(&lru_lock);
2027 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
2028 * @nr: The number of entries to scan
2030 * This function selects the entries on the LRU which are able to
2031 * be demoted, and then kicks off the process by calling
2032 * gfs2_dispose_glock_lru() above.
2035 static long gfs2_scan_glock_lru(int nr)
2037 struct gfs2_glock *gl;
2042 spin_lock(&lru_lock);
2043 while ((nr-- >= 0) && !list_empty(&lru_list)) {
2044 gl = list_first_entry(&lru_list, struct gfs2_glock, gl_lru);
2046 /* Test for being demotable */
2047 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
2048 list_move(&gl->gl_lru, &dispose);
2049 atomic_dec(&lru_count);
2054 list_move(&gl->gl_lru, &skipped);
2056 list_splice(&skipped, &lru_list);
2057 if (!list_empty(&dispose))
2058 gfs2_dispose_glock_lru(&dispose);
2059 spin_unlock(&lru_lock);
2064 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
2065 struct shrink_control *sc)
2067 if (!(sc->gfp_mask & __GFP_FS))
2069 return gfs2_scan_glock_lru(sc->nr_to_scan);
2072 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
2073 struct shrink_control *sc)
2075 return vfs_pressure_ratio(atomic_read(&lru_count));
2078 static struct shrinker glock_shrinker = {
2079 .seeks = DEFAULT_SEEKS,
2080 .count_objects = gfs2_glock_shrink_count,
2081 .scan_objects = gfs2_glock_shrink_scan,
2085 * glock_hash_walk - Call a function for glock in a hash bucket
2086 * @examiner: the function
2087 * @sdp: the filesystem
2089 * Note that the function can be called multiple times on the same
2090 * object. So the user must ensure that the function can cope with
2094 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
2096 struct gfs2_glock *gl;
2097 struct rhashtable_iter iter;
2099 rhashtable_walk_enter(&gl_hash_table, &iter);
2102 rhashtable_walk_start(&iter);
2104 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl)) {
2105 if (gl->gl_name.ln_sbd == sdp)
2109 rhashtable_walk_stop(&iter);
2110 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
2112 rhashtable_walk_exit(&iter);
2115 bool gfs2_queue_delete_work(struct gfs2_glock *gl, unsigned long delay)
2119 spin_lock(&gl->gl_lockref.lock);
2120 queued = queue_delayed_work(gfs2_delete_workqueue,
2121 &gl->gl_delete, delay);
2123 set_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2124 spin_unlock(&gl->gl_lockref.lock);
2128 void gfs2_cancel_delete_work(struct gfs2_glock *gl)
2130 if (cancel_delayed_work(&gl->gl_delete)) {
2131 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2136 bool gfs2_delete_work_queued(const struct gfs2_glock *gl)
2138 return test_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2141 static void flush_delete_work(struct gfs2_glock *gl)
2143 if (gl->gl_name.ln_type == LM_TYPE_IOPEN) {
2144 if (cancel_delayed_work(&gl->gl_delete)) {
2145 queue_delayed_work(gfs2_delete_workqueue,
2151 void gfs2_flush_delete_work(struct gfs2_sbd *sdp)
2153 glock_hash_walk(flush_delete_work, sdp);
2154 flush_workqueue(gfs2_delete_workqueue);
2158 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
2159 * @gl: The glock to thaw
2163 static void thaw_glock(struct gfs2_glock *gl)
2165 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
2167 if (!lockref_get_not_dead(&gl->gl_lockref))
2169 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
2170 gfs2_glock_queue_work(gl, 0);
2174 * clear_glock - look at a glock and see if we can free it from glock cache
2175 * @gl: the glock to look at
2179 static void clear_glock(struct gfs2_glock *gl)
2181 gfs2_glock_remove_from_lru(gl);
2183 spin_lock(&gl->gl_lockref.lock);
2184 if (!__lockref_is_dead(&gl->gl_lockref)) {
2185 gl->gl_lockref.count++;
2186 if (gl->gl_state != LM_ST_UNLOCKED)
2187 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2188 __gfs2_glock_queue_work(gl, 0);
2190 spin_unlock(&gl->gl_lockref.lock);
2194 * gfs2_glock_thaw - Thaw any frozen glocks
2195 * @sdp: The super block
2199 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
2201 glock_hash_walk(thaw_glock, sdp);
2204 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2206 spin_lock(&gl->gl_lockref.lock);
2207 gfs2_dump_glock(seq, gl, fsid);
2208 spin_unlock(&gl->gl_lockref.lock);
2211 static void dump_glock_func(struct gfs2_glock *gl)
2213 dump_glock(NULL, gl, true);
2217 * gfs2_gl_hash_clear - Empty out the glock hash table
2218 * @sdp: the filesystem
2220 * Called when unmounting the filesystem.
2223 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
2225 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
2226 flush_workqueue(glock_workqueue);
2227 glock_hash_walk(clear_glock, sdp);
2228 flush_workqueue(glock_workqueue);
2229 wait_event_timeout(sdp->sd_glock_wait,
2230 atomic_read(&sdp->sd_glock_disposal) == 0,
2232 glock_hash_walk(dump_glock_func, sdp);
2235 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
2237 struct gfs2_glock *gl = ip->i_gl;
2240 ret = gfs2_truncatei_resume(ip);
2241 gfs2_glock_assert_withdraw(gl, ret == 0);
2243 spin_lock(&gl->gl_lockref.lock);
2244 clear_bit(GLF_LOCK, &gl->gl_flags);
2246 spin_unlock(&gl->gl_lockref.lock);
2249 static const char *state2str(unsigned state)
2252 case LM_ST_UNLOCKED:
2256 case LM_ST_DEFERRED:
2258 case LM_ST_EXCLUSIVE:
2264 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
2267 if (flags & LM_FLAG_TRY)
2269 if (flags & LM_FLAG_TRY_1CB)
2271 if (flags & LM_FLAG_NOEXP)
2273 if (flags & LM_FLAG_ANY)
2275 if (flags & LM_FLAG_PRIORITY)
2277 if (flags & LM_FLAG_NODE_SCOPE)
2279 if (flags & GL_ASYNC)
2281 if (flags & GL_EXACT)
2283 if (flags & GL_NOCACHE)
2285 if (test_bit(HIF_HOLDER, &iflags))
2287 if (test_bit(HIF_WAIT, &iflags))
2289 if (test_bit(HIF_MAY_DEMOTE, &iflags))
2291 if (flags & GL_SKIP)
2298 * dump_holder - print information about a glock holder
2299 * @seq: the seq_file struct
2300 * @gh: the glock holder
2301 * @fs_id_buf: pointer to file system id (if requested)
2305 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
2306 const char *fs_id_buf)
2308 struct task_struct *gh_owner = NULL;
2312 if (gh->gh_owner_pid)
2313 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
2314 gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2315 fs_id_buf, state2str(gh->gh_state),
2316 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
2318 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
2319 gh_owner ? gh_owner->comm : "(ended)",
2324 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
2326 const unsigned long *gflags = &gl->gl_flags;
2329 if (test_bit(GLF_LOCK, gflags))
2331 if (test_bit(GLF_DEMOTE, gflags))
2333 if (test_bit(GLF_PENDING_DEMOTE, gflags))
2335 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
2337 if (test_bit(GLF_DIRTY, gflags))
2339 if (test_bit(GLF_LFLUSH, gflags))
2341 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
2343 if (test_bit(GLF_REPLY_PENDING, gflags))
2345 if (test_bit(GLF_INITIAL, gflags))
2347 if (test_bit(GLF_FROZEN, gflags))
2349 if (!list_empty(&gl->gl_holders))
2351 if (test_bit(GLF_LRU, gflags))
2355 if (test_bit(GLF_BLOCKING, gflags))
2357 if (test_bit(GLF_PENDING_DELETE, gflags))
2359 if (test_bit(GLF_FREEING, gflags))
2361 if (test_bit(GLF_INSTANTIATE_NEEDED, gflags))
2363 if (test_bit(GLF_INSTANTIATE_IN_PROG, gflags))
2370 * gfs2_dump_glock - print information about a glock
2371 * @seq: The seq_file struct
2373 * @fsid: If true, also dump the file system id
2375 * The file format is as follows:
2376 * One line per object, capital letters are used to indicate objects
2377 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2378 * other objects are indented by a single space and follow the glock to
2379 * which they are related. Fields are indicated by lower case letters
2380 * followed by a colon and the field value, except for strings which are in
2381 * [] so that its possible to see if they are composed of spaces for
2382 * example. The field's are n = number (id of the object), f = flags,
2383 * t = type, s = state, r = refcount, e = error, p = pid.
2387 void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2389 const struct gfs2_glock_operations *glops = gl->gl_ops;
2390 unsigned long long dtime;
2391 const struct gfs2_holder *gh;
2392 char gflags_buf[32];
2393 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2394 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2395 unsigned long nrpages = 0;
2397 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
2398 struct address_space *mapping = gfs2_glock2aspace(gl);
2400 nrpages = mapping->nrpages;
2402 memset(fs_id_buf, 0, sizeof(fs_id_buf));
2403 if (fsid && sdp) /* safety precaution */
2404 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2405 dtime = jiffies - gl->gl_demote_time;
2406 dtime *= 1000000/HZ; /* demote time in uSec */
2407 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
2409 gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2410 "v:%d r:%d m:%ld p:%lu\n",
2411 fs_id_buf, state2str(gl->gl_state),
2412 gl->gl_name.ln_type,
2413 (unsigned long long)gl->gl_name.ln_number,
2414 gflags2str(gflags_buf, gl),
2415 state2str(gl->gl_target),
2416 state2str(gl->gl_demote_state), dtime,
2417 atomic_read(&gl->gl_ail_count),
2418 atomic_read(&gl->gl_revokes),
2419 (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages);
2421 list_for_each_entry(gh, &gl->gl_holders, gh_list)
2422 dump_holder(seq, gh, fs_id_buf);
2424 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
2425 glops->go_dump(seq, gl, fs_id_buf);
2428 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
2430 struct gfs2_glock *gl = iter_ptr;
2432 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2433 gl->gl_name.ln_type,
2434 (unsigned long long)gl->gl_name.ln_number,
2435 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
2436 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
2437 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
2438 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
2439 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
2440 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
2441 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
2442 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
2446 static const char *gfs2_gltype[] = {
2460 static const char *gfs2_stype[] = {
2461 [GFS2_LKS_SRTT] = "srtt",
2462 [GFS2_LKS_SRTTVAR] = "srttvar",
2463 [GFS2_LKS_SRTTB] = "srttb",
2464 [GFS2_LKS_SRTTVARB] = "srttvarb",
2465 [GFS2_LKS_SIRT] = "sirt",
2466 [GFS2_LKS_SIRTVAR] = "sirtvar",
2467 [GFS2_LKS_DCOUNT] = "dlm",
2468 [GFS2_LKS_QCOUNT] = "queue",
2471 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2473 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
2475 struct gfs2_sbd *sdp = seq->private;
2476 loff_t pos = *(loff_t *)iter_ptr;
2477 unsigned index = pos >> 3;
2478 unsigned subindex = pos & 0x07;
2481 if (index == 0 && subindex != 0)
2484 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
2485 (index == 0) ? "cpu": gfs2_stype[subindex]);
2487 for_each_possible_cpu(i) {
2488 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
2491 seq_printf(seq, " %15u", i);
2493 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
2494 lkstats[index - 1].stats[subindex]);
2496 seq_putc(seq, '\n');
2500 int __init gfs2_glock_init(void)
2504 ret = rhashtable_init(&gl_hash_table, &ht_parms);
2508 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
2509 WQ_HIGHPRI | WQ_FREEZABLE, 0);
2510 if (!glock_workqueue) {
2511 rhashtable_destroy(&gl_hash_table);
2514 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
2515 WQ_MEM_RECLAIM | WQ_FREEZABLE,
2517 if (!gfs2_delete_workqueue) {
2518 destroy_workqueue(glock_workqueue);
2519 rhashtable_destroy(&gl_hash_table);
2523 ret = register_shrinker(&glock_shrinker);
2525 destroy_workqueue(gfs2_delete_workqueue);
2526 destroy_workqueue(glock_workqueue);
2527 rhashtable_destroy(&gl_hash_table);
2531 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2532 init_waitqueue_head(glock_wait_table + i);
2537 void gfs2_glock_exit(void)
2539 unregister_shrinker(&glock_shrinker);
2540 rhashtable_destroy(&gl_hash_table);
2541 destroy_workqueue(glock_workqueue);
2542 destroy_workqueue(gfs2_delete_workqueue);
2545 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2547 struct gfs2_glock *gl = gi->gl;
2552 if (!lockref_put_not_zero(&gl->gl_lockref))
2553 gfs2_glock_queue_put(gl);
2556 gl = rhashtable_walk_next(&gi->hti);
2557 if (IS_ERR_OR_NULL(gl)) {
2558 if (gl == ERR_PTR(-EAGAIN)) {
2565 if (gl->gl_name.ln_sbd != gi->sdp)
2568 if (!lockref_get_not_dead(&gl->gl_lockref))
2572 if (__lockref_is_dead(&gl->gl_lockref))
2580 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2583 struct gfs2_glock_iter *gi = seq->private;
2587 * We can either stay where we are, skip to the next hash table
2588 * entry, or start from the beginning.
2590 if (*pos < gi->last_pos) {
2591 rhashtable_walk_exit(&gi->hti);
2592 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2595 n = *pos - gi->last_pos;
2598 rhashtable_walk_start(&gi->hti);
2600 gfs2_glock_iter_next(gi, n);
2601 gi->last_pos = *pos;
2605 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2608 struct gfs2_glock_iter *gi = seq->private;
2611 gi->last_pos = *pos;
2612 gfs2_glock_iter_next(gi, 1);
2616 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2619 struct gfs2_glock_iter *gi = seq->private;
2621 rhashtable_walk_stop(&gi->hti);
2624 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2626 dump_glock(seq, iter_ptr, false);
2630 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2633 if (*pos >= GFS2_NR_SBSTATS)
2638 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2642 if (*pos >= GFS2_NR_SBSTATS)
2647 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2652 static const struct seq_operations gfs2_glock_seq_ops = {
2653 .start = gfs2_glock_seq_start,
2654 .next = gfs2_glock_seq_next,
2655 .stop = gfs2_glock_seq_stop,
2656 .show = gfs2_glock_seq_show,
2659 static const struct seq_operations gfs2_glstats_seq_ops = {
2660 .start = gfs2_glock_seq_start,
2661 .next = gfs2_glock_seq_next,
2662 .stop = gfs2_glock_seq_stop,
2663 .show = gfs2_glstats_seq_show,
2666 static const struct seq_operations gfs2_sbstats_sops = {
2667 .start = gfs2_sbstats_seq_start,
2668 .next = gfs2_sbstats_seq_next,
2669 .stop = gfs2_sbstats_seq_stop,
2670 .show = gfs2_sbstats_seq_show,
2673 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2675 static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2676 const struct seq_operations *ops)
2678 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2680 struct seq_file *seq = file->private_data;
2681 struct gfs2_glock_iter *gi = seq->private;
2683 gi->sdp = inode->i_private;
2684 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2686 seq->size = GFS2_SEQ_GOODSIZE;
2688 * Initially, we are "before" the first hash table entry; the
2689 * first call to rhashtable_walk_next gets us the first entry.
2693 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2698 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2700 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2703 static int gfs2_glocks_release(struct inode *inode, struct file *file)
2705 struct seq_file *seq = file->private_data;
2706 struct gfs2_glock_iter *gi = seq->private;
2709 gfs2_glock_put(gi->gl);
2710 rhashtable_walk_exit(&gi->hti);
2711 return seq_release_private(inode, file);
2714 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2716 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2719 static const struct file_operations gfs2_glocks_fops = {
2720 .owner = THIS_MODULE,
2721 .open = gfs2_glocks_open,
2723 .llseek = seq_lseek,
2724 .release = gfs2_glocks_release,
2727 static const struct file_operations gfs2_glstats_fops = {
2728 .owner = THIS_MODULE,
2729 .open = gfs2_glstats_open,
2731 .llseek = seq_lseek,
2732 .release = gfs2_glocks_release,
2735 DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats);
2737 void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2739 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2741 debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2744 debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2745 &gfs2_glstats_fops);
2747 debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2748 &gfs2_sbstats_fops);
2751 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2753 debugfs_remove_recursive(sdp->debugfs_dir);
2754 sdp->debugfs_dir = NULL;
2757 void gfs2_register_debugfs(void)
2759 gfs2_root = debugfs_create_dir("gfs2", NULL);
2762 void gfs2_unregister_debugfs(void)
2764 debugfs_remove(gfs2_root);