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 (list_is_first(&gh->gh_list, &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 && (ret & LM_OUT_CANCELED))
673 gfs2_holder_wake(gh);
674 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
675 /* move to back of queue and try next entry */
676 if (ret & LM_OUT_CANCELED) {
677 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
678 list_move_tail(&gh->gh_list, &gl->gl_holders);
679 gh = find_first_waiter(gl);
680 gl->gl_target = gh->gh_state;
683 /* Some error or failed "try lock" - report it */
684 if ((ret & LM_OUT_ERROR) ||
685 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
686 gl->gl_target = gl->gl_state;
692 /* Unlocked due to conversion deadlock, try again */
695 do_xmote(gl, gh, gl->gl_target);
697 /* Conversion fails, unlock and try again */
700 do_xmote(gl, gh, LM_ST_UNLOCKED);
702 default: /* Everything else */
703 fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n",
704 gl->gl_target, state);
707 spin_unlock(&gl->gl_lockref.lock);
711 /* Fast path - we got what we asked for */
712 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
713 gfs2_demote_wake(gl);
714 if (state != LM_ST_UNLOCKED) {
715 if (glops->go_xmote_bh) {
716 spin_unlock(&gl->gl_lockref.lock);
717 rv = glops->go_xmote_bh(gl);
718 spin_lock(&gl->gl_lockref.lock);
729 clear_bit(GLF_LOCK, &gl->gl_flags);
731 spin_unlock(&gl->gl_lockref.lock);
734 static bool is_system_glock(struct gfs2_glock *gl)
736 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
737 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
739 if (gl == m_ip->i_gl)
745 * do_xmote - Calls the DLM to change the state of a lock
746 * @gl: The lock state
747 * @gh: The holder (only for promotes)
748 * @target: The target lock state
752 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
753 __releases(&gl->gl_lockref.lock)
754 __acquires(&gl->gl_lockref.lock)
756 const struct gfs2_glock_operations *glops = gl->gl_ops;
757 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
758 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
761 if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) &&
762 gh && !(gh->gh_flags & LM_FLAG_NOEXP))
764 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
766 GLOCK_BUG_ON(gl, gl->gl_state == target);
767 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
768 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
771 * If another process is already doing the invalidate, let that
772 * finish first. The glock state machine will get back to this
773 * holder again later.
775 if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
778 do_error(gl, 0); /* Fail queued try locks */
781 set_bit(GLF_BLOCKING, &gl->gl_flags);
782 if ((gl->gl_req == LM_ST_UNLOCKED) ||
783 (gl->gl_state == LM_ST_EXCLUSIVE) ||
784 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
785 clear_bit(GLF_BLOCKING, &gl->gl_flags);
786 spin_unlock(&gl->gl_lockref.lock);
787 if (glops->go_sync) {
788 ret = glops->go_sync(gl);
789 /* If we had a problem syncing (due to io errors or whatever,
790 * we should not invalidate the metadata or tell dlm to
791 * release the glock to other nodes.
794 if (cmpxchg(&sdp->sd_log_error, 0, ret)) {
795 fs_err(sdp, "Error %d syncing glock \n", ret);
796 gfs2_dump_glock(NULL, gl, true);
801 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) {
803 * The call to go_sync should have cleared out the ail list.
804 * If there are still items, we have a problem. We ought to
805 * withdraw, but we can't because the withdraw code also uses
806 * glocks. Warn about the error, dump the glock, then fall
807 * through and wait for logd to do the withdraw for us.
809 if ((atomic_read(&gl->gl_ail_count) != 0) &&
810 (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) {
811 gfs2_glock_assert_warn(gl,
812 !atomic_read(&gl->gl_ail_count));
813 gfs2_dump_glock(NULL, gl, true);
815 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
816 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
822 * Check for an error encountered since we called go_sync and go_inval.
823 * If so, we can't withdraw from the glock code because the withdraw
824 * code itself uses glocks (see function signal_our_withdraw) to
825 * change the mount to read-only. Most importantly, we must not call
826 * dlm to unlock the glock until the journal is in a known good state
827 * (after journal replay) otherwise other nodes may use the object
828 * (rgrp or dinode) and then later, journal replay will corrupt the
829 * file system. The best we can do here is wait for the logd daemon
830 * to see sd_log_error and withdraw, and in the meantime, requeue the
833 * We make a special exception for some system glocks, such as the
834 * system statfs inode glock, which needs to be granted before the
835 * gfs2_quotad daemon can exit, and that exit needs to finish before
836 * we can unmount the withdrawn file system.
838 * However, if we're just unlocking the lock (say, for unmount, when
839 * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
840 * then it's okay to tell dlm to unlock it.
842 if (unlikely(sdp->sd_log_error && !gfs2_withdrawn(sdp)))
843 gfs2_withdraw_delayed(sdp);
844 if (glock_blocked_by_withdraw(gl) &&
845 (target != LM_ST_UNLOCKED ||
846 test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags))) {
847 if (!is_system_glock(gl)) {
848 gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD);
851 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
855 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
857 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
858 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
859 target == LM_ST_UNLOCKED &&
860 test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
861 finish_xmote(gl, target);
862 gfs2_glock_queue_work(gl, 0);
864 fs_err(sdp, "lm_lock ret %d\n", ret);
865 GLOCK_BUG_ON(gl, !gfs2_withdrawn(sdp));
867 } else { /* lock_nolock */
868 finish_xmote(gl, target);
869 gfs2_glock_queue_work(gl, 0);
872 spin_lock(&gl->gl_lockref.lock);
876 * run_queue - do all outstanding tasks related to a glock
877 * @gl: The glock in question
878 * @nonblock: True if we must not block in run_queue
882 static void run_queue(struct gfs2_glock *gl, const int nonblock)
883 __releases(&gl->gl_lockref.lock)
884 __acquires(&gl->gl_lockref.lock)
886 struct gfs2_holder *gh = NULL;
889 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
892 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
894 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
895 gl->gl_demote_state != gl->gl_state) {
896 if (find_first_holder(gl))
900 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
901 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
902 gl->gl_target = gl->gl_demote_state;
904 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
905 gfs2_demote_wake(gl);
906 ret = do_promote(gl);
911 gh = find_first_waiter(gl);
912 gl->gl_target = gh->gh_state;
913 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
914 do_error(gl, 0); /* Fail queued try locks */
916 do_xmote(gl, gh, gl->gl_target);
921 clear_bit(GLF_LOCK, &gl->gl_flags);
922 smp_mb__after_atomic();
923 gl->gl_lockref.count++;
924 __gfs2_glock_queue_work(gl, 0);
928 clear_bit(GLF_LOCK, &gl->gl_flags);
929 smp_mb__after_atomic();
933 void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation)
935 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
937 if (ri->ri_magic == 0)
938 ri->ri_magic = cpu_to_be32(GFS2_MAGIC);
939 if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC))
940 ri->ri_generation_deleted = cpu_to_be64(generation);
943 bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation)
945 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
947 if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC))
949 return generation <= be64_to_cpu(ri->ri_generation_deleted);
952 static void gfs2_glock_poke(struct gfs2_glock *gl)
954 int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP;
955 struct gfs2_holder gh;
958 __gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh, _RET_IP_);
959 error = gfs2_glock_nq(&gh);
962 gfs2_holder_uninit(&gh);
965 static bool gfs2_try_evict(struct gfs2_glock *gl)
967 struct gfs2_inode *ip;
968 bool evicted = false;
971 * If there is contention on the iopen glock and we have an inode, try
972 * to grab and release the inode so that it can be evicted. This will
973 * allow the remote node to go ahead and delete the inode without us
974 * having to do it, which will avoid rgrp glock thrashing.
976 * The remote node is likely still holding the corresponding inode
977 * glock, so it will run before we get to verify that the delete has
980 spin_lock(&gl->gl_lockref.lock);
982 if (ip && !igrab(&ip->i_inode))
984 spin_unlock(&gl->gl_lockref.lock);
986 struct gfs2_glock *inode_gl = NULL;
988 gl->gl_no_formal_ino = ip->i_no_formal_ino;
989 set_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
990 d_prune_aliases(&ip->i_inode);
993 /* If the inode was evicted, gl->gl_object will now be NULL. */
994 spin_lock(&gl->gl_lockref.lock);
998 lockref_get(&inode_gl->gl_lockref);
999 clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
1001 spin_unlock(&gl->gl_lockref.lock);
1003 gfs2_glock_poke(inode_gl);
1004 gfs2_glock_put(inode_gl);
1011 static void delete_work_func(struct work_struct *work)
1013 struct delayed_work *dwork = to_delayed_work(work);
1014 struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete);
1015 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1016 struct inode *inode;
1017 u64 no_addr = gl->gl_name.ln_number;
1019 spin_lock(&gl->gl_lockref.lock);
1020 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1021 spin_unlock(&gl->gl_lockref.lock);
1023 if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
1025 * If we can evict the inode, give the remote node trying to
1026 * delete the inode some time before verifying that the delete
1027 * has happened. Otherwise, if we cause contention on the inode glock
1028 * immediately, the remote node will think that we still have
1029 * the inode in use, and so it will give up waiting.
1031 * If we can't evict the inode, signal to the remote node that
1032 * the inode is still in use. We'll later try to delete the
1033 * inode locally in gfs2_evict_inode.
1035 * FIXME: We only need to verify that the remote node has
1036 * deleted the inode because nodes before this remote delete
1037 * rework won't cooperate. At a later time, when we no longer
1038 * care about compatibility with such nodes, we can skip this
1041 if (gfs2_try_evict(gl)) {
1042 if (gfs2_queue_delete_work(gl, 5 * HZ))
1048 inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino,
1049 GFS2_BLKST_UNLINKED);
1050 if (!IS_ERR_OR_NULL(inode)) {
1051 d_prune_aliases(inode);
1058 static void glock_work_func(struct work_struct *work)
1060 unsigned long delay = 0;
1061 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
1062 unsigned int drop_refs = 1;
1064 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
1065 finish_xmote(gl, gl->gl_reply);
1068 spin_lock(&gl->gl_lockref.lock);
1069 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1070 gl->gl_state != LM_ST_UNLOCKED &&
1071 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
1072 unsigned long holdtime, now = jiffies;
1074 holdtime = gl->gl_tchange + gl->gl_hold_time;
1075 if (time_before(now, holdtime))
1076 delay = holdtime - now;
1079 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1080 gfs2_set_demote(gl);
1085 /* Keep one glock reference for the work we requeue. */
1087 if (gl->gl_name.ln_type != LM_TYPE_INODE)
1089 __gfs2_glock_queue_work(gl, delay);
1093 * Drop the remaining glock references manually here. (Mind that
1094 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
1097 gl->gl_lockref.count -= drop_refs;
1098 if (!gl->gl_lockref.count) {
1099 __gfs2_glock_put(gl);
1102 spin_unlock(&gl->gl_lockref.lock);
1105 static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
1106 struct gfs2_glock *new)
1108 struct wait_glock_queue wait;
1109 wait_queue_head_t *wq = glock_waitqueue(name);
1110 struct gfs2_glock *gl;
1113 init_wait(&wait.wait);
1114 wait.wait.func = glock_wake_function;
1117 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1120 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
1121 &new->gl_node, ht_parms);
1125 gl = rhashtable_lookup_fast(&gl_hash_table,
1128 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
1135 finish_wait(wq, &wait.wait);
1140 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
1141 * @sdp: The GFS2 superblock
1142 * @number: the lock number
1143 * @glops: The glock_operations to use
1144 * @create: If 0, don't create the glock if it doesn't exist
1145 * @glp: the glock is returned here
1147 * This does not lock a glock, just finds/creates structures for one.
1152 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
1153 const struct gfs2_glock_operations *glops, int create,
1154 struct gfs2_glock **glp)
1156 struct super_block *s = sdp->sd_vfs;
1157 struct lm_lockname name = { .ln_number = number,
1158 .ln_type = glops->go_type,
1160 struct gfs2_glock *gl, *tmp;
1161 struct address_space *mapping;
1162 struct kmem_cache *cachep;
1165 gl = find_insert_glock(&name, NULL);
1173 if (glops->go_flags & GLOF_ASPACE)
1174 cachep = gfs2_glock_aspace_cachep;
1176 cachep = gfs2_glock_cachep;
1177 gl = kmem_cache_alloc(cachep, GFP_NOFS);
1181 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
1183 if (glops->go_flags & GLOF_LVB) {
1184 gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1185 if (!gl->gl_lksb.sb_lvbptr) {
1186 kmem_cache_free(cachep, gl);
1191 atomic_inc(&sdp->sd_glock_disposal);
1192 gl->gl_node.next = NULL;
1193 gl->gl_flags = glops->go_instantiate ? BIT(GLF_INSTANTIATE_NEEDED) : 0;
1195 lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass);
1196 gl->gl_lockref.count = 1;
1197 gl->gl_state = LM_ST_UNLOCKED;
1198 gl->gl_target = LM_ST_UNLOCKED;
1199 gl->gl_demote_state = LM_ST_EXCLUSIVE;
1203 /* We use the global stats to estimate the initial per-glock stats */
1204 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
1206 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
1207 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
1208 gl->gl_tchange = jiffies;
1209 gl->gl_object = NULL;
1210 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
1211 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
1212 if (gl->gl_name.ln_type == LM_TYPE_IOPEN)
1213 INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func);
1215 mapping = gfs2_glock2aspace(gl);
1217 mapping->a_ops = &gfs2_meta_aops;
1218 mapping->host = s->s_bdev->bd_inode;
1220 mapping_set_gfp_mask(mapping, GFP_NOFS);
1221 mapping->private_data = NULL;
1222 mapping->writeback_index = 0;
1225 tmp = find_insert_glock(&name, gl);
1237 kfree(gl->gl_lksb.sb_lvbptr);
1238 kmem_cache_free(cachep, gl);
1239 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
1240 wake_up(&sdp->sd_glock_wait);
1247 * __gfs2_holder_init - initialize a struct gfs2_holder in the default way
1249 * @state: the state we're requesting
1250 * @flags: the modifier flags
1251 * @gh: the holder structure
1255 void __gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
1256 struct gfs2_holder *gh, unsigned long ip)
1258 INIT_LIST_HEAD(&gh->gh_list);
1261 gh->gh_owner_pid = get_pid(task_pid(current));
1262 gh->gh_state = state;
1263 gh->gh_flags = flags;
1265 gfs2_glock_hold(gl);
1269 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1270 * @state: the state we're requesting
1271 * @flags: the modifier flags
1272 * @gh: the holder structure
1274 * Don't mess with the glock.
1278 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
1280 gh->gh_state = state;
1281 gh->gh_flags = flags;
1283 gh->gh_ip = _RET_IP_;
1284 put_pid(gh->gh_owner_pid);
1285 gh->gh_owner_pid = get_pid(task_pid(current));
1289 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1290 * @gh: the holder structure
1294 void gfs2_holder_uninit(struct gfs2_holder *gh)
1296 put_pid(gh->gh_owner_pid);
1297 gfs2_glock_put(gh->gh_gl);
1298 gfs2_holder_mark_uninitialized(gh);
1302 static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
1303 unsigned long start_time)
1305 /* Have we waited longer that a second? */
1306 if (time_after(jiffies, start_time + HZ)) {
1307 /* Lengthen the minimum hold time. */
1308 gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
1314 * gfs2_glock_wait - wait on a glock acquisition
1315 * @gh: the glock holder
1317 * Returns: 0 on success
1320 int gfs2_glock_wait(struct gfs2_holder *gh)
1322 unsigned long start_time = jiffies;
1325 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1326 gfs2_glock_update_hold_time(gh->gh_gl, start_time);
1327 return gh->gh_error;
1330 static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
1334 for (i = 0; i < num_gh; i++)
1335 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
1341 * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1342 * @num_gh: the number of holders in the array
1343 * @ghs: the glock holder array
1345 * Returns: 0 on success, meaning all glocks have been granted and are held.
1346 * -ESTALE if the request timed out, meaning all glocks were released,
1347 * and the caller should retry the operation.
1350 int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
1352 struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
1353 int i, ret = 0, timeout = 0;
1354 unsigned long start_time = jiffies;
1359 * Total up the (minimum hold time * 2) of all glocks and use that to
1360 * determine the max amount of time we should wait.
1362 for (i = 0; i < num_gh; i++)
1363 timeout += ghs[i].gh_gl->gl_hold_time << 1;
1366 if (!wait_event_timeout(sdp->sd_async_glock_wait,
1367 !glocks_pending(num_gh, ghs), timeout))
1368 ret = -ESTALE; /* request timed out. */
1371 * If dlm granted all our requests, we need to adjust the glock
1372 * minimum hold time values according to how long we waited.
1374 * If our request timed out, we need to repeatedly release any held
1375 * glocks we acquired thus far to allow dlm to acquire the remaining
1376 * glocks without deadlocking. We cannot currently cancel outstanding
1377 * glock acquisitions.
1379 * The HIF_WAIT bit tells us which requests still need a response from
1382 * If dlm sent us any errors, we return the first error we find.
1384 keep_waiting = false;
1385 for (i = 0; i < num_gh; i++) {
1386 /* Skip holders we have already dequeued below. */
1387 if (!gfs2_holder_queued(&ghs[i]))
1389 /* Skip holders with a pending DLM response. */
1390 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags)) {
1391 keep_waiting = true;
1395 if (test_bit(HIF_HOLDER, &ghs[i].gh_iflags)) {
1397 gfs2_glock_dq(&ghs[i]);
1399 gfs2_glock_update_hold_time(ghs[i].gh_gl,
1403 ret = ghs[i].gh_error;
1410 * At this point, we've either acquired all locks or released them all.
1416 * handle_callback - process a demote request
1418 * @state: the state the caller wants us to change to
1419 * @delay: zero to demote immediately; otherwise pending demote
1420 * @remote: true if this came from a different cluster node
1422 * There are only two requests that we are going to see in actual
1423 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1426 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
1427 unsigned long delay, bool remote)
1430 set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1432 gfs2_set_demote(gl);
1433 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
1434 gl->gl_demote_state = state;
1435 gl->gl_demote_time = jiffies;
1436 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
1437 gl->gl_demote_state != state) {
1438 gl->gl_demote_state = LM_ST_UNLOCKED;
1440 if (gl->gl_ops->go_callback)
1441 gl->gl_ops->go_callback(gl, remote);
1442 trace_gfs2_demote_rq(gl, remote);
1445 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
1447 struct va_format vaf;
1450 va_start(args, fmt);
1453 seq_vprintf(seq, fmt, args);
1458 pr_err("%pV", &vaf);
1465 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1466 * @gh: the holder structure to add
1468 * Eventually we should move the recursive locking trap to a
1469 * debugging option or something like that. This is the fast
1470 * path and needs to have the minimum number of distractions.
1474 static inline void add_to_queue(struct gfs2_holder *gh)
1475 __releases(&gl->gl_lockref.lock)
1476 __acquires(&gl->gl_lockref.lock)
1478 struct gfs2_glock *gl = gh->gh_gl;
1479 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1480 struct list_head *insert_pt = NULL;
1481 struct gfs2_holder *gh2;
1484 GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1485 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1486 GLOCK_BUG_ON(gl, true);
1488 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1489 if (test_bit(GLF_LOCK, &gl->gl_flags)) {
1490 struct gfs2_holder *first_gh;
1492 first_gh = find_first_strong_holder(gl);
1493 try_futile = !may_grant(gl, first_gh, gh);
1495 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1499 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1500 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
1501 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK) &&
1502 !test_bit(HIF_MAY_DEMOTE, &gh2->gh_iflags)))
1503 goto trap_recursive;
1505 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1507 gh->gh_error = GLR_TRYFAILED;
1508 gfs2_holder_wake(gh);
1511 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1513 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1514 insert_pt = &gh2->gh_list;
1516 trace_gfs2_glock_queue(gh, 1);
1517 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1518 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1519 if (likely(insert_pt == NULL)) {
1520 list_add_tail(&gh->gh_list, &gl->gl_holders);
1521 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1525 list_add_tail(&gh->gh_list, insert_pt);
1527 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
1528 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1529 spin_unlock(&gl->gl_lockref.lock);
1530 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1531 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1532 spin_lock(&gl->gl_lockref.lock);
1537 fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1538 fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1539 fs_err(sdp, "lock type: %d req lock state : %d\n",
1540 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1541 fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1542 fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1543 fs_err(sdp, "lock type: %d req lock state : %d\n",
1544 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1545 gfs2_dump_glock(NULL, gl, true);
1550 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1551 * @gh: the holder structure
1553 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1555 * Returns: 0, GLR_TRYFAILED, or errno on failure
1558 int gfs2_glock_nq(struct gfs2_holder *gh)
1560 struct gfs2_glock *gl = gh->gh_gl;
1563 if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP))
1566 if (test_bit(GLF_LRU, &gl->gl_flags))
1567 gfs2_glock_remove_from_lru(gl);
1570 spin_lock(&gl->gl_lockref.lock);
1572 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1573 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1574 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1575 gl->gl_lockref.count++;
1576 __gfs2_glock_queue_work(gl, 0);
1579 spin_unlock(&gl->gl_lockref.lock);
1581 if (!(gh->gh_flags & GL_ASYNC))
1582 error = gfs2_glock_wait(gh);
1588 * gfs2_glock_poll - poll to see if an async request has been completed
1591 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1594 int gfs2_glock_poll(struct gfs2_holder *gh)
1596 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1599 static inline bool needs_demote(struct gfs2_glock *gl)
1601 return (test_bit(GLF_DEMOTE, &gl->gl_flags) ||
1602 test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags));
1605 static void __gfs2_glock_dq(struct gfs2_holder *gh)
1607 struct gfs2_glock *gl = gh->gh_gl;
1608 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1613 * This while loop is similar to function demote_incompat_holders:
1614 * If the glock is due to be demoted (which may be from another node
1615 * or even if this holder is GL_NOCACHE), the weak holders are
1616 * demoted as well, allowing the glock to be demoted.
1620 * If we're in the process of file system withdraw, we cannot
1621 * just dequeue any glocks until our journal is recovered, lest
1622 * we introduce file system corruption. We need two exceptions
1623 * to this rule: We need to allow unlocking of nondisk glocks
1624 * and the glock for our own journal that needs recovery.
1626 if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) &&
1627 glock_blocked_by_withdraw(gl) &&
1628 gh->gh_gl != sdp->sd_jinode_gl) {
1629 sdp->sd_glock_dqs_held++;
1630 spin_unlock(&gl->gl_lockref.lock);
1632 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
1633 TASK_UNINTERRUPTIBLE);
1634 spin_lock(&gl->gl_lockref.lock);
1638 * This holder should not be cached, so mark it for demote.
1639 * Note: this should be done before the check for needs_demote
1642 if (gh->gh_flags & GL_NOCACHE)
1643 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1645 list_del_init(&gh->gh_list);
1646 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1647 trace_gfs2_glock_queue(gh, 0);
1650 * If there hasn't been a demote request we are done.
1651 * (Let the remaining holders, if any, keep holding it.)
1653 if (!needs_demote(gl)) {
1654 if (list_empty(&gl->gl_holders))
1659 * If we have another strong holder (we cannot auto-demote)
1660 * we are done. It keeps holding it until it is done.
1662 if (find_first_strong_holder(gl))
1666 * If we have a weak holder at the head of the list, it
1667 * (and all others like it) must be auto-demoted. If there
1668 * are no more weak holders, we exit the while loop.
1670 gh = find_first_holder(gl);
1673 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1674 gfs2_glock_add_to_lru(gl);
1676 if (unlikely(!fast_path)) {
1677 gl->gl_lockref.count++;
1678 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1679 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1680 gl->gl_name.ln_type == LM_TYPE_INODE)
1681 delay = gl->gl_hold_time;
1682 __gfs2_glock_queue_work(gl, delay);
1687 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1688 * @gh: the glock holder
1691 void gfs2_glock_dq(struct gfs2_holder *gh)
1693 struct gfs2_glock *gl = gh->gh_gl;
1695 spin_lock(&gl->gl_lockref.lock);
1696 if (list_is_first(&gh->gh_list, &gl->gl_holders) &&
1697 !test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1698 spin_unlock(&gl->gl_lockref.lock);
1699 gl->gl_name.ln_sbd->sd_lockstruct.ls_ops->lm_cancel(gl);
1700 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1701 spin_lock(&gl->gl_lockref.lock);
1704 __gfs2_glock_dq(gh);
1705 spin_unlock(&gl->gl_lockref.lock);
1708 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1710 struct gfs2_glock *gl = gh->gh_gl;
1713 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1717 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1718 * @gh: the holder structure
1722 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1725 gfs2_holder_uninit(gh);
1729 * gfs2_glock_nq_num - acquire a glock based on lock number
1730 * @sdp: the filesystem
1731 * @number: the lock number
1732 * @glops: the glock operations for the type of glock
1733 * @state: the state to acquire the glock in
1734 * @flags: modifier flags for the acquisition
1735 * @gh: the struct gfs2_holder
1740 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1741 const struct gfs2_glock_operations *glops,
1742 unsigned int state, u16 flags, struct gfs2_holder *gh)
1744 struct gfs2_glock *gl;
1747 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1749 error = gfs2_glock_nq_init(gl, state, flags, gh);
1757 * glock_compare - Compare two struct gfs2_glock structures for sorting
1758 * @arg_a: the first structure
1759 * @arg_b: the second structure
1763 static int glock_compare(const void *arg_a, const void *arg_b)
1765 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1766 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1767 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1768 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1770 if (a->ln_number > b->ln_number)
1772 if (a->ln_number < b->ln_number)
1774 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1779 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1780 * @num_gh: the number of structures
1781 * @ghs: an array of struct gfs2_holder structures
1782 * @p: placeholder for the holder structure to pass back
1784 * Returns: 0 on success (all glocks acquired),
1785 * errno on failure (no glocks acquired)
1788 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1789 struct gfs2_holder **p)
1794 for (x = 0; x < num_gh; x++)
1797 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1799 for (x = 0; x < num_gh; x++) {
1800 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1802 error = gfs2_glock_nq(p[x]);
1805 gfs2_glock_dq(p[x]);
1814 * gfs2_glock_nq_m - acquire multiple glocks
1815 * @num_gh: the number of structures
1816 * @ghs: an array of struct gfs2_holder structures
1819 * Returns: 0 on success (all glocks acquired),
1820 * errno on failure (no glocks acquired)
1823 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1825 struct gfs2_holder *tmp[4];
1826 struct gfs2_holder **pph = tmp;
1833 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1834 return gfs2_glock_nq(ghs);
1838 pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *),
1844 error = nq_m_sync(num_gh, ghs, pph);
1853 * gfs2_glock_dq_m - release multiple glocks
1854 * @num_gh: the number of structures
1855 * @ghs: an array of struct gfs2_holder structures
1859 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1862 gfs2_glock_dq(&ghs[num_gh]);
1865 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1867 unsigned long delay = 0;
1868 unsigned long holdtime;
1869 unsigned long now = jiffies;
1871 gfs2_glock_hold(gl);
1872 spin_lock(&gl->gl_lockref.lock);
1873 holdtime = gl->gl_tchange + gl->gl_hold_time;
1874 if (!list_empty(&gl->gl_holders) &&
1875 gl->gl_name.ln_type == LM_TYPE_INODE) {
1876 if (time_before(now, holdtime))
1877 delay = holdtime - now;
1878 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1879 delay = gl->gl_hold_time;
1882 * Note 1: We cannot call demote_incompat_holders from handle_callback
1883 * or gfs2_set_demote due to recursion problems like: gfs2_glock_dq ->
1884 * handle_callback -> demote_incompat_holders -> gfs2_glock_dq
1885 * Plus, we only want to demote the holders if the request comes from
1886 * a remote cluster node because local holder conflicts are resolved
1889 * Note 2: if a remote node wants this glock in EX mode, lock_dlm will
1890 * request that we set our state to UNLOCKED. Here we mock up a holder
1891 * to make it look like someone wants the lock EX locally. Any SH
1892 * and DF requests should be able to share the lock without demoting.
1894 * Note 3: We only want to demote the demoteable holders when there
1895 * are no more strong holders. The demoteable holders might as well
1896 * keep the glock until the last strong holder is done with it.
1898 if (!find_first_strong_holder(gl)) {
1899 struct gfs2_holder mock_gh = {
1901 .gh_state = (state == LM_ST_UNLOCKED) ?
1902 LM_ST_EXCLUSIVE : state,
1903 .gh_iflags = BIT(HIF_HOLDER)
1906 demote_incompat_holders(gl, &mock_gh);
1908 handle_callback(gl, state, delay, true);
1909 __gfs2_glock_queue_work(gl, delay);
1910 spin_unlock(&gl->gl_lockref.lock);
1914 * gfs2_should_freeze - Figure out if glock should be frozen
1915 * @gl: The glock in question
1917 * Glocks are not frozen if (a) the result of the dlm operation is
1918 * an error, (b) the locking operation was an unlock operation or
1919 * (c) if there is a "noexp" flagged request anywhere in the queue
1921 * Returns: 1 if freezing should occur, 0 otherwise
1924 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1926 const struct gfs2_holder *gh;
1928 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1930 if (gl->gl_target == LM_ST_UNLOCKED)
1933 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1934 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1936 if (LM_FLAG_NOEXP & gh->gh_flags)
1944 * gfs2_glock_complete - Callback used by locking
1945 * @gl: Pointer to the glock
1946 * @ret: The return value from the dlm
1948 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1949 * to use a bitfield shared with other glock state fields.
1952 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1954 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1956 spin_lock(&gl->gl_lockref.lock);
1959 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1960 if (gfs2_should_freeze(gl)) {
1961 set_bit(GLF_FROZEN, &gl->gl_flags);
1962 spin_unlock(&gl->gl_lockref.lock);
1967 gl->gl_lockref.count++;
1968 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1969 __gfs2_glock_queue_work(gl, 0);
1970 spin_unlock(&gl->gl_lockref.lock);
1973 static int glock_cmp(void *priv, const struct list_head *a,
1974 const struct list_head *b)
1976 struct gfs2_glock *gla, *glb;
1978 gla = list_entry(a, struct gfs2_glock, gl_lru);
1979 glb = list_entry(b, struct gfs2_glock, gl_lru);
1981 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1983 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1990 * gfs2_dispose_glock_lru - Demote a list of glocks
1991 * @list: The list to dispose of
1993 * Disposing of glocks may involve disk accesses, so that here we sort
1994 * the glocks by number (i.e. disk location of the inodes) so that if
1995 * there are any such accesses, they'll be sent in order (mostly).
1997 * Must be called under the lru_lock, but may drop and retake this
1998 * lock. While the lru_lock is dropped, entries may vanish from the
1999 * list, but no new entries will appear on the list (since it is
2003 static void gfs2_dispose_glock_lru(struct list_head *list)
2004 __releases(&lru_lock)
2005 __acquires(&lru_lock)
2007 struct gfs2_glock *gl;
2009 list_sort(NULL, list, glock_cmp);
2011 while(!list_empty(list)) {
2012 gl = list_first_entry(list, struct gfs2_glock, gl_lru);
2013 list_del_init(&gl->gl_lru);
2014 clear_bit(GLF_LRU, &gl->gl_flags);
2015 if (!spin_trylock(&gl->gl_lockref.lock)) {
2017 list_add(&gl->gl_lru, &lru_list);
2018 set_bit(GLF_LRU, &gl->gl_flags);
2019 atomic_inc(&lru_count);
2022 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
2023 spin_unlock(&gl->gl_lockref.lock);
2024 goto add_back_to_lru;
2026 gl->gl_lockref.count++;
2028 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2029 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
2030 __gfs2_glock_queue_work(gl, 0);
2031 spin_unlock(&gl->gl_lockref.lock);
2032 cond_resched_lock(&lru_lock);
2037 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
2038 * @nr: The number of entries to scan
2040 * This function selects the entries on the LRU which are able to
2041 * be demoted, and then kicks off the process by calling
2042 * gfs2_dispose_glock_lru() above.
2045 static long gfs2_scan_glock_lru(int nr)
2047 struct gfs2_glock *gl;
2052 spin_lock(&lru_lock);
2053 while ((nr-- >= 0) && !list_empty(&lru_list)) {
2054 gl = list_first_entry(&lru_list, struct gfs2_glock, gl_lru);
2056 /* Test for being demotable */
2057 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
2058 list_move(&gl->gl_lru, &dispose);
2059 atomic_dec(&lru_count);
2064 list_move(&gl->gl_lru, &skipped);
2066 list_splice(&skipped, &lru_list);
2067 if (!list_empty(&dispose))
2068 gfs2_dispose_glock_lru(&dispose);
2069 spin_unlock(&lru_lock);
2074 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
2075 struct shrink_control *sc)
2077 if (!(sc->gfp_mask & __GFP_FS))
2079 return gfs2_scan_glock_lru(sc->nr_to_scan);
2082 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
2083 struct shrink_control *sc)
2085 return vfs_pressure_ratio(atomic_read(&lru_count));
2088 static struct shrinker glock_shrinker = {
2089 .seeks = DEFAULT_SEEKS,
2090 .count_objects = gfs2_glock_shrink_count,
2091 .scan_objects = gfs2_glock_shrink_scan,
2095 * glock_hash_walk - Call a function for glock in a hash bucket
2096 * @examiner: the function
2097 * @sdp: the filesystem
2099 * Note that the function can be called multiple times on the same
2100 * object. So the user must ensure that the function can cope with
2104 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
2106 struct gfs2_glock *gl;
2107 struct rhashtable_iter iter;
2109 rhashtable_walk_enter(&gl_hash_table, &iter);
2112 rhashtable_walk_start(&iter);
2114 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl)) {
2115 if (gl->gl_name.ln_sbd == sdp)
2119 rhashtable_walk_stop(&iter);
2120 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
2122 rhashtable_walk_exit(&iter);
2125 bool gfs2_queue_delete_work(struct gfs2_glock *gl, unsigned long delay)
2129 spin_lock(&gl->gl_lockref.lock);
2130 queued = queue_delayed_work(gfs2_delete_workqueue,
2131 &gl->gl_delete, delay);
2133 set_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2134 spin_unlock(&gl->gl_lockref.lock);
2138 void gfs2_cancel_delete_work(struct gfs2_glock *gl)
2140 if (cancel_delayed_work(&gl->gl_delete)) {
2141 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2146 bool gfs2_delete_work_queued(const struct gfs2_glock *gl)
2148 return test_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2151 static void flush_delete_work(struct gfs2_glock *gl)
2153 if (gl->gl_name.ln_type == LM_TYPE_IOPEN) {
2154 if (cancel_delayed_work(&gl->gl_delete)) {
2155 queue_delayed_work(gfs2_delete_workqueue,
2161 void gfs2_flush_delete_work(struct gfs2_sbd *sdp)
2163 glock_hash_walk(flush_delete_work, sdp);
2164 flush_workqueue(gfs2_delete_workqueue);
2168 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
2169 * @gl: The glock to thaw
2173 static void thaw_glock(struct gfs2_glock *gl)
2175 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
2177 if (!lockref_get_not_dead(&gl->gl_lockref))
2179 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
2180 gfs2_glock_queue_work(gl, 0);
2184 * clear_glock - look at a glock and see if we can free it from glock cache
2185 * @gl: the glock to look at
2189 static void clear_glock(struct gfs2_glock *gl)
2191 gfs2_glock_remove_from_lru(gl);
2193 spin_lock(&gl->gl_lockref.lock);
2194 if (!__lockref_is_dead(&gl->gl_lockref)) {
2195 gl->gl_lockref.count++;
2196 if (gl->gl_state != LM_ST_UNLOCKED)
2197 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2198 __gfs2_glock_queue_work(gl, 0);
2200 spin_unlock(&gl->gl_lockref.lock);
2204 * gfs2_glock_thaw - Thaw any frozen glocks
2205 * @sdp: The super block
2209 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
2211 glock_hash_walk(thaw_glock, sdp);
2214 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2216 spin_lock(&gl->gl_lockref.lock);
2217 gfs2_dump_glock(seq, gl, fsid);
2218 spin_unlock(&gl->gl_lockref.lock);
2221 static void dump_glock_func(struct gfs2_glock *gl)
2223 dump_glock(NULL, gl, true);
2227 * gfs2_gl_hash_clear - Empty out the glock hash table
2228 * @sdp: the filesystem
2230 * Called when unmounting the filesystem.
2233 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
2235 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
2236 flush_workqueue(glock_workqueue);
2237 glock_hash_walk(clear_glock, sdp);
2238 flush_workqueue(glock_workqueue);
2239 wait_event_timeout(sdp->sd_glock_wait,
2240 atomic_read(&sdp->sd_glock_disposal) == 0,
2242 glock_hash_walk(dump_glock_func, sdp);
2245 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
2247 struct gfs2_glock *gl = ip->i_gl;
2250 ret = gfs2_truncatei_resume(ip);
2251 gfs2_glock_assert_withdraw(gl, ret == 0);
2253 spin_lock(&gl->gl_lockref.lock);
2254 clear_bit(GLF_LOCK, &gl->gl_flags);
2256 spin_unlock(&gl->gl_lockref.lock);
2259 static const char *state2str(unsigned state)
2262 case LM_ST_UNLOCKED:
2266 case LM_ST_DEFERRED:
2268 case LM_ST_EXCLUSIVE:
2274 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
2277 if (flags & LM_FLAG_TRY)
2279 if (flags & LM_FLAG_TRY_1CB)
2281 if (flags & LM_FLAG_NOEXP)
2283 if (flags & LM_FLAG_ANY)
2285 if (flags & LM_FLAG_PRIORITY)
2287 if (flags & LM_FLAG_NODE_SCOPE)
2289 if (flags & GL_ASYNC)
2291 if (flags & GL_EXACT)
2293 if (flags & GL_NOCACHE)
2295 if (test_bit(HIF_HOLDER, &iflags))
2297 if (test_bit(HIF_WAIT, &iflags))
2299 if (test_bit(HIF_MAY_DEMOTE, &iflags))
2301 if (flags & GL_SKIP)
2308 * dump_holder - print information about a glock holder
2309 * @seq: the seq_file struct
2310 * @gh: the glock holder
2311 * @fs_id_buf: pointer to file system id (if requested)
2315 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
2316 const char *fs_id_buf)
2318 struct task_struct *gh_owner = NULL;
2322 if (gh->gh_owner_pid)
2323 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
2324 gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2325 fs_id_buf, state2str(gh->gh_state),
2326 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
2328 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
2329 gh_owner ? gh_owner->comm : "(ended)",
2334 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
2336 const unsigned long *gflags = &gl->gl_flags;
2339 if (test_bit(GLF_LOCK, gflags))
2341 if (test_bit(GLF_DEMOTE, gflags))
2343 if (test_bit(GLF_PENDING_DEMOTE, gflags))
2345 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
2347 if (test_bit(GLF_DIRTY, gflags))
2349 if (test_bit(GLF_LFLUSH, gflags))
2351 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
2353 if (test_bit(GLF_REPLY_PENDING, gflags))
2355 if (test_bit(GLF_INITIAL, gflags))
2357 if (test_bit(GLF_FROZEN, gflags))
2359 if (!list_empty(&gl->gl_holders))
2361 if (test_bit(GLF_LRU, gflags))
2365 if (test_bit(GLF_BLOCKING, gflags))
2367 if (test_bit(GLF_PENDING_DELETE, gflags))
2369 if (test_bit(GLF_FREEING, gflags))
2371 if (test_bit(GLF_INSTANTIATE_NEEDED, gflags))
2373 if (test_bit(GLF_INSTANTIATE_IN_PROG, gflags))
2380 * gfs2_dump_glock - print information about a glock
2381 * @seq: The seq_file struct
2383 * @fsid: If true, also dump the file system id
2385 * The file format is as follows:
2386 * One line per object, capital letters are used to indicate objects
2387 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2388 * other objects are indented by a single space and follow the glock to
2389 * which they are related. Fields are indicated by lower case letters
2390 * followed by a colon and the field value, except for strings which are in
2391 * [] so that its possible to see if they are composed of spaces for
2392 * example. The field's are n = number (id of the object), f = flags,
2393 * t = type, s = state, r = refcount, e = error, p = pid.
2397 void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2399 const struct gfs2_glock_operations *glops = gl->gl_ops;
2400 unsigned long long dtime;
2401 const struct gfs2_holder *gh;
2402 char gflags_buf[32];
2403 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2404 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2405 unsigned long nrpages = 0;
2407 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
2408 struct address_space *mapping = gfs2_glock2aspace(gl);
2410 nrpages = mapping->nrpages;
2412 memset(fs_id_buf, 0, sizeof(fs_id_buf));
2413 if (fsid && sdp) /* safety precaution */
2414 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2415 dtime = jiffies - gl->gl_demote_time;
2416 dtime *= 1000000/HZ; /* demote time in uSec */
2417 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
2419 gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2420 "v:%d r:%d m:%ld p:%lu\n",
2421 fs_id_buf, state2str(gl->gl_state),
2422 gl->gl_name.ln_type,
2423 (unsigned long long)gl->gl_name.ln_number,
2424 gflags2str(gflags_buf, gl),
2425 state2str(gl->gl_target),
2426 state2str(gl->gl_demote_state), dtime,
2427 atomic_read(&gl->gl_ail_count),
2428 atomic_read(&gl->gl_revokes),
2429 (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages);
2431 list_for_each_entry(gh, &gl->gl_holders, gh_list)
2432 dump_holder(seq, gh, fs_id_buf);
2434 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
2435 glops->go_dump(seq, gl, fs_id_buf);
2438 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
2440 struct gfs2_glock *gl = iter_ptr;
2442 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2443 gl->gl_name.ln_type,
2444 (unsigned long long)gl->gl_name.ln_number,
2445 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
2446 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
2447 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
2448 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
2449 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
2450 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
2451 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
2452 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
2456 static const char *gfs2_gltype[] = {
2470 static const char *gfs2_stype[] = {
2471 [GFS2_LKS_SRTT] = "srtt",
2472 [GFS2_LKS_SRTTVAR] = "srttvar",
2473 [GFS2_LKS_SRTTB] = "srttb",
2474 [GFS2_LKS_SRTTVARB] = "srttvarb",
2475 [GFS2_LKS_SIRT] = "sirt",
2476 [GFS2_LKS_SIRTVAR] = "sirtvar",
2477 [GFS2_LKS_DCOUNT] = "dlm",
2478 [GFS2_LKS_QCOUNT] = "queue",
2481 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2483 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
2485 struct gfs2_sbd *sdp = seq->private;
2486 loff_t pos = *(loff_t *)iter_ptr;
2487 unsigned index = pos >> 3;
2488 unsigned subindex = pos & 0x07;
2491 if (index == 0 && subindex != 0)
2494 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
2495 (index == 0) ? "cpu": gfs2_stype[subindex]);
2497 for_each_possible_cpu(i) {
2498 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
2501 seq_printf(seq, " %15u", i);
2503 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
2504 lkstats[index - 1].stats[subindex]);
2506 seq_putc(seq, '\n');
2510 int __init gfs2_glock_init(void)
2514 ret = rhashtable_init(&gl_hash_table, &ht_parms);
2518 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
2519 WQ_HIGHPRI | WQ_FREEZABLE, 0);
2520 if (!glock_workqueue) {
2521 rhashtable_destroy(&gl_hash_table);
2524 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
2525 WQ_MEM_RECLAIM | WQ_FREEZABLE,
2527 if (!gfs2_delete_workqueue) {
2528 destroy_workqueue(glock_workqueue);
2529 rhashtable_destroy(&gl_hash_table);
2533 ret = register_shrinker(&glock_shrinker);
2535 destroy_workqueue(gfs2_delete_workqueue);
2536 destroy_workqueue(glock_workqueue);
2537 rhashtable_destroy(&gl_hash_table);
2541 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2542 init_waitqueue_head(glock_wait_table + i);
2547 void gfs2_glock_exit(void)
2549 unregister_shrinker(&glock_shrinker);
2550 rhashtable_destroy(&gl_hash_table);
2551 destroy_workqueue(glock_workqueue);
2552 destroy_workqueue(gfs2_delete_workqueue);
2555 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2557 struct gfs2_glock *gl = gi->gl;
2562 if (!lockref_put_not_zero(&gl->gl_lockref))
2563 gfs2_glock_queue_put(gl);
2566 gl = rhashtable_walk_next(&gi->hti);
2567 if (IS_ERR_OR_NULL(gl)) {
2568 if (gl == ERR_PTR(-EAGAIN)) {
2575 if (gl->gl_name.ln_sbd != gi->sdp)
2578 if (!lockref_get_not_dead(&gl->gl_lockref))
2582 if (__lockref_is_dead(&gl->gl_lockref))
2590 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2593 struct gfs2_glock_iter *gi = seq->private;
2597 * We can either stay where we are, skip to the next hash table
2598 * entry, or start from the beginning.
2600 if (*pos < gi->last_pos) {
2601 rhashtable_walk_exit(&gi->hti);
2602 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2605 n = *pos - gi->last_pos;
2608 rhashtable_walk_start(&gi->hti);
2610 gfs2_glock_iter_next(gi, n);
2611 gi->last_pos = *pos;
2615 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2618 struct gfs2_glock_iter *gi = seq->private;
2621 gi->last_pos = *pos;
2622 gfs2_glock_iter_next(gi, 1);
2626 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2629 struct gfs2_glock_iter *gi = seq->private;
2631 rhashtable_walk_stop(&gi->hti);
2634 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2636 dump_glock(seq, iter_ptr, false);
2640 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2643 if (*pos >= GFS2_NR_SBSTATS)
2648 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2652 if (*pos >= GFS2_NR_SBSTATS)
2657 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2662 static const struct seq_operations gfs2_glock_seq_ops = {
2663 .start = gfs2_glock_seq_start,
2664 .next = gfs2_glock_seq_next,
2665 .stop = gfs2_glock_seq_stop,
2666 .show = gfs2_glock_seq_show,
2669 static const struct seq_operations gfs2_glstats_seq_ops = {
2670 .start = gfs2_glock_seq_start,
2671 .next = gfs2_glock_seq_next,
2672 .stop = gfs2_glock_seq_stop,
2673 .show = gfs2_glstats_seq_show,
2676 static const struct seq_operations gfs2_sbstats_sops = {
2677 .start = gfs2_sbstats_seq_start,
2678 .next = gfs2_sbstats_seq_next,
2679 .stop = gfs2_sbstats_seq_stop,
2680 .show = gfs2_sbstats_seq_show,
2683 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2685 static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2686 const struct seq_operations *ops)
2688 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2690 struct seq_file *seq = file->private_data;
2691 struct gfs2_glock_iter *gi = seq->private;
2693 gi->sdp = inode->i_private;
2694 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2696 seq->size = GFS2_SEQ_GOODSIZE;
2698 * Initially, we are "before" the first hash table entry; the
2699 * first call to rhashtable_walk_next gets us the first entry.
2703 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2708 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2710 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2713 static int gfs2_glocks_release(struct inode *inode, struct file *file)
2715 struct seq_file *seq = file->private_data;
2716 struct gfs2_glock_iter *gi = seq->private;
2719 gfs2_glock_put(gi->gl);
2720 rhashtable_walk_exit(&gi->hti);
2721 return seq_release_private(inode, file);
2724 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2726 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2729 static const struct file_operations gfs2_glocks_fops = {
2730 .owner = THIS_MODULE,
2731 .open = gfs2_glocks_open,
2733 .llseek = seq_lseek,
2734 .release = gfs2_glocks_release,
2737 static const struct file_operations gfs2_glstats_fops = {
2738 .owner = THIS_MODULE,
2739 .open = gfs2_glstats_open,
2741 .llseek = seq_lseek,
2742 .release = gfs2_glocks_release,
2745 DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats);
2747 void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2749 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2751 debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2754 debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2755 &gfs2_glstats_fops);
2757 debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2758 &gfs2_sbstats_fops);
2761 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2763 debugfs_remove_recursive(sdp->debugfs_dir);
2764 sdp->debugfs_dir = NULL;
2767 void gfs2_register_debugfs(void)
2769 gfs2_root = debugfs_create_dir("gfs2", NULL);
2772 void gfs2_unregister_debugfs(void)
2774 debugfs_remove(gfs2_root);